id
stringlengths 11
17
| article_id
stringlengths 8
11
| path
stringlengths 11
60
| section_title
stringlengths 1
1.33k
| educational_score
float64 0
5.16
| domain
stringclasses 4
values | document_type
stringclasses 5
values | domain_scores
listlengths 0
3
| document_type_scores
listlengths 0
4
| text
stringlengths 1
110k
| authors
listlengths 0
8.02k
| article_url
stringlengths 3
63
| license_type
stringclasses 1
value | license_url
stringclasses 15
values | language
stringclasses 45
values | language_score
float64 0
1
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
39057383_p10
|
39057383
|
sec[2]/p[4]
|
3. Results
| 4.175781 |
biomedical
|
Study
|
[
0.99853515625,
0.00133514404296875,
0.00021708011627197266
] |
[
0.99853515625,
0.00032401084899902344,
0.000919342041015625,
0.0001747608184814453
] |
The association of clinical and laboratory parameters with the IFD occurrence is shown in Table 1 . The development of an IFD was associated with an ICU hospitalization for more than seven days (31% vs. 4%; p = 0.017), the administration of more than seven different antimicrobial courses during the last three months (50% vs. 16%; p = 0.01), the history of severe neutropenia for more than 44 days (38% vs. 17%; = 0.011), or the colonization by fungi in sites other than the oral cavity, and specifically the throat, the urinary bladder, and indwelling catheters (31% vs. 6%; p = 0.007). The administration of prednisolone equivalent ≥ 0.3 mg/kg for ≥3 weeks also seems to confer a significantly increased risk for an IFD. The absolute number of lymphocytes or monocytes did not seem to be an independent risk factor for IFD. An axillary body temperature greater than 38.8 °C for more than 12 days and a C-reactive protein (CRP) greater than 10 mg/dL were associated with IFDs (50% vs. 29% with p = 0.019; 44% vs. 6% with p < 0.001; and 38% vs. 22% with p = 0.011, respectively). According to the statistical analysis, the prophylactic administration of antifungals and granulocyte colony-stimulating factor (GCSF) were not associated with the development of an IFD. Conversely, the administration of IVIG seems to be significantly associated with the development of an IFD, but this finding can be attributed to the increased rates of IVIG administration when an IFD is suspected. HSCT and therapy with immunomodulatory agents, such as monoclonal antibodies or small molecule inhibitors, have not been proven as independent risk factors for IFD development. The prevalence of comorbidities, such as endocrine disorders, genetic syndromes, kidney disease, thrombosis, and cardiovascular disorders, did not differ significantly between cases with IFD and the rest of the cohort. In addition, the rate of relapsed or refractory (r/r) underlying disease in this retrospective cohort study was calculated at 13%, and even though r/r cases were more frequent among patients with IFD, no significant effect was noted.
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057383_p11
|
39057383
|
sec[2]/p[5]
|
3. Results
| 4.066406 |
biomedical
|
Study
|
[
0.998046875,
0.001911163330078125,
0.00019228458404541016
] |
[
0.9990234375,
0.0004074573516845703,
0.00038504600524902344,
0.0002605915069580078
] |
Investigations among IFD cases revealed that CRP was significantly higher in patients who succumbed to the IFD compared to those who recovered (21 ± 3.7 vs. 9.7 ± 6.1; p = 0.017). No other parameter was correlated significantly with IFD-related deceased cases. Of note, both patients with IFD who experienced a relapse of the underlying disease survived the invasive infection. A positive outcome was also recorded for the single patient with IFD who had undergone allo-HSCT. Table 2 describes all IFD cases in this cohort, along with their clinical characteristics in detail.
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057383_p12
|
39057383
|
sec[2]/p[6]
|
3. Results
| 4.078125 |
biomedical
|
Study
|
[
0.86083984375,
0.138427734375,
0.0005645751953125
] |
[
0.93310546875,
0.02142333984375,
0.00470733642578125,
0.040802001953125
] |
Regarding IPA, the most prevalent clinical trait was fever, with an average temperature of 39.1 °C and an average duration of 17 days, despite broad-spectrum antimicrobials. IPA manifested with a dry cough in 62% of the patients. Regarding radiologic findings, the halo sign was evident in the chest’s high-resolution computed tomography (HRCT) in most cases (62%), while 25% of the patients presented with the air-crescent sign. Initial IPA treatment corresponded to a combination of intravenous LAMB and voriconazole in 7/8 patients (for an average of 19 days; range 11–31), succeeded by oral administration of voriconazole alone for an entire three-month course. Two IPA patients did not survive (2/7; 28.6%), but their death was attributed to refractory AML M5. One patient was treated successfully with isavuconazole for 48 days.
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057383_p13
|
39057383
|
sec[2]/p[7]
|
3. Results
| 4.058594 |
biomedical
|
Study
|
[
0.98828125,
0.0116424560546875,
0.000217437744140625
] |
[
0.99267578125,
0.0036869049072265625,
0.0008573532104492188,
0.0025539398193359375
] |
In fungemia cases, the only common clinical feature was fever (mean 38.8 °C; 38.1–39.6 °C), lasting 10 days (2–25). The central venous catheter (CVC) was removed in four out of the six respective cases. Candida parapsilosis was the most prevalent species to cause fungemia in this cohort (67%). C. albicans was identified in one case, while the case of Exophiala dermatitidis fungemia has been previously reported by our Department .
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057383_p14
|
39057383
|
sec[2]/p[8]
|
3. Results
| 4.09375 |
biomedical
|
Study
|
[
0.99609375,
0.003765106201171875,
0.00017881393432617188
] |
[
0.9970703125,
0.0013599395751953125,
0.0007042884826660156,
0.0006313323974609375
] |
Non-invasive fungal infections affected 5.8% of the study population (9/154) and regarded 12 infections (half identified with C. albicans and half with C. parapsilosis ). Most of these cases corresponded to nail infections (4/12), followed by rash located in the inner thighs, inguinal regions, and genitalia (two cases each). One case of Candida esophagitis was documented, while C. parapsilosis was isolated along with Pseudomonas aeruginosa in a patient with pus discharge from his gastrostomy. Most of the cases mentioned above were under prophylaxis (10/12).
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057383_p15
|
39057383
|
sec[2]/p[9]
|
3. Results
| 4.085938 |
biomedical
|
Study
|
[
0.9482421875,
0.0513916015625,
0.0005517005920410156
] |
[
0.97900390625,
0.015411376953125,
0.001651763916015625,
0.00370025634765625
] |
As far as adverse events due to antifungals are concerned, 16% (4/25) of the patients administered with LAMB experienced allergic reactions, most commonly manifested with a skin rash developing within the first 15 min of the drug’s infusion (3/4) plus one patient who developed facial angioedema. Side effects were documented in 25% (2/8) of the patients who received voriconazole, which comprised visual hallucinations in one case and vomiting after co-administration with Sopa-K in the other.
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057383_p16
|
39057383
|
sec[3]/p[0]
|
4. Discussion
| 4.207031 |
biomedical
|
Study
|
[
0.9990234375,
0.0009889602661132812,
0.0001989603042602539
] |
[
0.9990234375,
0.0002562999725341797,
0.0007419586181640625,
0.000156402587890625
] |
The available information for the incidence and outcome of IFDs in children is derived mainly from single-center retrospective studies . In the present retrospective study, data were retrieved from the medical records of children and adolescents with neoplasia treated in the only Pediatric Hematology–Oncology Department in the island region of Crete during the last decade. Of interest, 8% of the treated patients suffered from one or more IFDs, and among them, most cases corresponded to possible/probable pulmonary aspergillosis (50%) followed by fungemia due to Candida parapsilosis . The mortality rate was as low as 12.5% (two fatal IFDs out of 16 episodes). These findings are in agreement with the published literature about prevalence (5–10%; range 2–15%, depending on geographic location, patient characteristics, and specific hospital practices) and mortality rates (ranging from 20 to 50%, depending on the IFD, the underlying neoplasia, and the prompt diagnosis and treatment) . According to the literature, probable/proven IFDs’ prevalence among the pediatric AML population ranges between 5% and 15% (referring mostly to IPA followed by non- albicans invasive candidiasis) and is associated indirectly with fatal outcomes in up to 18% of patients . AML diagnosis was strongly associated with IFD occurrence (OR 76.7; 95% CI: 16.6 to 353.1; p < 0.001) in this retrospective cohort study. Of note, 70% of AML patients developed an IFD, while only 2.4% of ALL cases (1/42; with ETV6 :: RUNX1 fusion and under the high-risk protocol due to high measurable residual disease during induction) manifested an invasive fungal infection. The mean age of the children with IFD was 9.8 years, which agrees with other studies which associate the higher age of children with manifesting an IFD .
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057383_p17
|
39057383
|
sec[3]/p[1]
|
4. Discussion
| 4.101563 |
biomedical
|
Study
|
[
0.9990234375,
0.0005884170532226562,
0.00015676021575927734
] |
[
0.9990234375,
0.00037097930908203125,
0.0004661083221435547,
0.00011336803436279297
] |
C. parapsilosis was the primary colonizer of mucus membranes and medical devices such as drainage catheters and CVCs, despite antifungal prophylaxis with micafungin in most cases. C. parapsilosis was also the predominant confirmed cause of an IFD under the same prophylaxis. As a result, antifungal prophylaxis did not have a statistically significant effect on IFD prevention. However, the population with IFD was small, comprising 12/154 children with different types of neoplasia. Hence, more studies are necessary to establish the definite role of antifungal prophylaxis in preventing IFDs. Considering the 6-year-old boy with B-ALL and IPA in our cohort, twice-a-week micafungin prophylaxis displayed a significantly lower occurrence of Aspergillus infections during the early phase of childhood ALL treatment, according to a recent study .
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057383_p18
|
39057383
|
sec[3]/p[2]
|
4. Discussion
| 4.257813 |
biomedical
|
Study
|
[
0.9990234375,
0.001018524169921875,
0.0001608133316040039
] |
[
0.99853515625,
0.0004208087921142578,
0.0009756088256835938,
0.00023651123046875
] |
In agreement with the findings of the current study, the incidence of candidemia due to C. parapsilosis has increased in recent years, bearing clinical implications due to its decreased susceptibility to echinocandins . However, no deaths were recorded due to fungemia or cellulitis caused by this yeast or combined with C. paratropicalis (as observed in this cohort’s case of cellulitis). All cases were successfully treated with a combination of LAMB and second-generation triazoles, followed by a step-down treatment with triazole only. Numerous studies are focusing on a personalized risk assessment for the decision on prophylactic or empirical antifungal administration . The present study confirmed several risk factors that should trigger the investigations for an IFD: AML diagnosis, ICU setting for more than a week, administration of more than seven antimicrobials during the last trimester, history of severe neutropenia for 1.5 months, fungal colonization in sites other than the oral cavity, fever > 38.8 °C for >12 days, CRP levels above 10 mg/dL, treatment with prednisolone equivalent ≥ 0.3 mg/kg for more than three weeks, and IVIG administration. Intriguingly, IVIG administration often leads to false positive BDG levels that may remain detectable for more than two weeks. Therefore, BDG should not be used to diagnose an IFD within three weeks after IVIG administration . Of course, medical devices colonized by fungi need immediate removal to prevent fungemia. The literature supports the equivalent prevalence of yeast and mold infections (50% and 50%), with this ratio remaining stable for both the 2018–2022 and 2013–2017 periods and despite the increased IFD incidence during the last 5 years . An increase in bronchopulmonary mold infections and breakthrough IFDs has been noted by several studies in the field .
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057383_p19
|
39057383
|
sec[3]/p[3]
|
4. Discussion
| 4.214844 |
biomedical
|
Study
|
[
0.99755859375,
0.0022754669189453125,
0.00014507770538330078
] |
[
0.9951171875,
0.0011358261108398438,
0.003204345703125,
0.00045800209045410156
] |
Broad-spectrum antibiotics, which are sometimes required but are primarily non-specific, are a highly relevant risk factor for developing systemic fungal infections . A variety of antibiotic and antiviral regimens—more than seven in total number according to the present study—should be avoided. Moreover, all Pediatric Hematology–Oncology clinics should be regularly visited by an infectious disease specialist for more efficient use of antimicrobial substances. Emphasis should be placed on strict hand hygiene, safe catheter management, and catheter care in general . Disinfection of the ward and medical equipment or objects that enter the patients’ wards with chlorine-based disinfectants can also eliminate Candida species that survive long on surfaces. The use of special air filters creates appropriate particle-free air flow and is crucial in the prevention of pulmonary IFDs . Regarding antifungal therapy, seven out of eight cases of IPA received combined antifungal treatment with LAMB and voriconazole, although current guidelines based on randomized clinical trials promote monotherapy with voriconazole or isavuconazole. The combination of antifungals seems to represent a common practice in cases of severe and life-threatening diseases when management reasoning eclipses evidence-based medicine . Of interest, the lower mortality rates of invasive mold infections reported in our study could be attributed to the early diagnosis and prompt administration of antifungal therapy. This is in line with previous work on the 2-fold increase in mortality following a ≥6 days delay in the administration of effective antifungal therapy .
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057383_p20
|
39057383
|
sec[3]/p[4]
|
4. Discussion
| 3.876953 |
biomedical
|
Study
|
[
0.99951171875,
0.00040984153747558594,
0.00025343894958496094
] |
[
0.9990234375,
0.000331878662109375,
0.0005540847778320312,
0.00008308887481689453
] |
The current study has several limitations that should be considered when interpreting the results, with its retrospective nature being the major one. The fact that all IPA cases were not proven or probable, the possibility of undiagnosed IFDs, and the presence of potential confounding variables limit the generalizability of the findings. In addition, the observational nature of the study precludes establishing causality between the identified factors and the outcomes.
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057383_p21
|
39057383
|
sec[4]/p[0]
|
5. Conclusions
| 3.951172 |
biomedical
|
Review
|
[
0.9990234375,
0.0006079673767089844,
0.00035881996154785156
] |
[
0.376708984375,
0.01039886474609375,
0.6123046875,
0.0007338523864746094
] |
The limited antifungal arsenal, combined with the well-documented resistance to the available agents, requires the development of alternative, effective, and safe strategies for IFD treatment. Moreover, prompt IFD diagnosis is crucial to prevent the observed high mortality rates, as late diagnosis equals a poor prognosis. Further studies, standardizing the existing technologies, and mastering novel tools (PCR and rapid point-of-care assays, T2Candida platform, etc.) are needed. IFDs are a significant burden among children with cancer, constituting an independent risk factor for both event-free and overall survival. Determining the relevant host factors and high-risk clinical traits could alleviate the issue of IFDs and breakthrough fungal infections.
|
[
"Eleni Moraitaki",
"Ioannis Kyriakidis",
"Iordanis Pelagiadis",
"Nikolaos Katzilakis",
"Maria Stratigaki",
"Georgios Chamilos",
"Athanasios Tragiannidis",
"Eftichia Stiakaki"
] |
https://doi.org/10.3390/jof10070498
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p0
|
PMC11278118
|
sec[0]/p[0]
|
1. Introduction
| 4.222656 |
biomedical
|
Study
|
[
0.98583984375,
0.00054931640625,
0.01385498046875
] |
[
0.99462890625,
0.0006699562072753906,
0.0045623779296875,
0.00007241964340209961
] |
The most suitable plant pollinators are honeybees due to their biology and adaptability to plants . Some plant species are selectively pollinated by a single insect species, and this has a significant effect on plant seed productivity and survival . The flowers of red clover have a long, approximately 10 mm, corolla tube and nectarines are located at nectary base of it, making the nectar collection difficult for short-tongued bees . The most important pollinators of red clover belong to different species and subspecies of wild long-tongued bumblebees ( Bombus ssp.), such as B. pascuorum ssp., B. ruderatus , and B. hortorum ssp., and some races or hybrids of honeybees . While long-tongued bumblebees collect nectar by reaching it down in the corolla tube, short-tongued bumblebees, such as B. terrestris and B. lucorum , bite holes in the lower part of the corolla to access nectar without pollinating, thus reducing pollination efficiency and seed yield . Bees can also bite holes in corolla tubes or use the holes previously bitten by bumblebees to collect nectar from red clover, resulting in no pollen transfer to nectar . The study of diploid and tetraploid clover seed production revealed that tetraploid varieties produce more nectar per floret than diploid ones; however, that differences in corolla tube dimensions due to ploidy level did not affect seed production, suggesting that insects visited and pollinated the flowers . Therefore, it is necessary to maintain medium- and long-tongued honeybee populations and to protect the variety of wild pollinators . The studies of Balžekas et al. reveal that Caucasian x European dark bee ( Apis mellifera mellifera L. have been widespread in Lithuania since ancient times, and are listed as a Lithuanian native bee) hybrids and Caucasian × Carniolan bee hybrids collected 74.8% and 65.6% more honey per colony compared to pure Caucasian bees, respectively.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p1
|
PMC11278118
|
sec[0]/p[1]
|
1. Introduction
| 4.394531 |
biomedical
|
Study
|
[
0.9990234375,
0.0003285408020019531,
0.000659942626953125
] |
[
0.99462890625,
0.00030541419982910156,
0.005153656005859375,
0.0000959634780883789
] |
The earlier studies on the protein composition of manually collected clover pollen from flowers of red clover cvs. ‘Kiršiniai’ and ‘Vyčiai’, berseem clover ( Trifolium alexandrinum L.) cv. ‘Faraon’, and white clover ( Trifolium repens L.) cv. ‘Medūnai’ identified and described over 200 protein spots from which quantitative levels were most divergent in 30 investigated clover pollen proteome maps . The berseem clover honey had no sulfur-containing amino acids methionine and cysteine, but contained a sufficiently high amount of lysine and was the most acidic (pH 3.26) compared to the other types of honey studied . It is known that legumes possess antioxidant enzymes, such as superoxide dismutase, catalase, and various peroxidases as well as non-enzymatic antioxidants such as ascorbate and glutathione, which protect them from reactive oxygen species (ROS) and have a symbiotic relationship with nitrogen-fixing soil bacteria called rhizobia . Legumes have been shown to have various proteins, such as natural resistance-associated macrophage proteins/duodenal metal transporter (NRAMP/DMT) homologs involved in metal ion transport across membranes within the legume nodule , transferrin-mediating iron transport , and comprise about 24% mass of leghaemoglobin present in legume nodules . Another protein identified in legumes that contains iron is ferritin, whose accumulation (around 24 dpi) is correlated with the highest level of leghaemoglobin . Ferritin constitutes almost half of the of total seed iron in soybeans, common beans, and peas and, depending on the plant species, can range from 18 to 42% and is known for participating in cell detoxification as well as indicating stress-induced responses . Therefore, the identification of legume proteins and the determination of their role in various biological processes are of great importance.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p2
|
PMC11278118
|
sec[0]/p[2]
|
1. Introduction
| 4.25 |
biomedical
|
Study
|
[
0.99609375,
0.0003120899200439453,
0.0038013458251953125
] |
[
0.99267578125,
0.0005087852478027344,
0.006580352783203125,
0.00006979703903198242
] |
The composition of sugars and proteins in nectar honey differs from that of honeydew honey in terms of color, organoleptic properties, and the values for electrical conductivity, pH, optical rotation, ash content, sugar profile, and mineral content; however, physicochemical indicators do not necessarily reflect its authenticity . For better identification chromatography, spectroscopy, and molecular biology approaches are used along with melissopalynological analysis when the botanical origin of honey pollen and the amount of honeydew elements are visually assessed . Honeydew honey is a specific type of honey, obtained from the secretions of plants or aphids and, in some cases, insects’ excretions, especially when natural sources and climatic conditions favor the harvest of honeydew over nectar. Red and white clovers are preferred by several aphid species, such as Aphis coronillae Ferrari, Therioaphis trifolii , and Acyrthosiphon pisum Harris . Honeydew honey is mainly studied for the variety of sugars it contains, such as α,α-trehalose, melezitose, theanderose, nystose, or maltotetraose in honeydew, as well as chemical indicators, enabling the differentiation of its types and the distinction from nectar honey . Studies of honeydew collected from field beans infected with aphids ( Acyrthosiphon pisum ) were conducted in Belgium, revealing that the protein diversity of aphid honeydew originates from the host aphid and its microbiota, including endosymbiotic bacteria and gut flora .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p3
|
PMC11278118
|
sec[0]/p[3]
|
1. Introduction
| 3.824219 |
biomedical
|
Study
|
[
0.99609375,
0.00042748451232910156,
0.003414154052734375
] |
[
0.99951171875,
0.0003802776336669922,
0.00012433528900146484,
0.000058710575103759766
] |
This study aimed to determine monofloral red clover and rapeseed honey samples of Lithuanian origin at the protein level by describing and comparing their protein composition and aimed to evaluate proteins associated with aphids and lactic acid bacteria in honey samples.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p4
|
PMC11278118
|
sec[1]/sec[0]/p[0]
|
2.1. Collection of Honey and Determination of Its Botanical Origin
| 4.148438 |
biomedical
|
Study
|
[
0.99853515625,
0.00026607513427734375,
0.0009713172912597656
] |
[
0.99951171875,
0.0002560615539550781,
0.00022864341735839844,
0.000040531158447265625
] |
Honey samples were collected in different districts of Lithuania. All honey samples were collected from Apis mellifera bees bred in Lithuania. Monofloral clover honey (C3) was collected from a private beekeeper farm in Rokiškis district, while monofloral rapeseed honey samples (S5 and S15) were collected in Kėdainiai, as well as polyfloral samples from Ukmergė (S22) and Rokiškis (S23) districts. Honey samples were tested after 6 months of storage in dark glass bottles in the refrigerator at 5 °C until used in further analysis. Honey sample preparation for botanical composition analysis was performed using the melissopalynology technique as described in . In brief, a 10 g honey sample was weighed and dissolved in distilled water and centrifuged. The sediment was washed with 20 mL of distilled water and again centrifuged. Sediment was collected and spread on a slide over an area of approximately 20 mm × 20 mm, dried and covered with glycerine jelly. Pollen photos taken under a Nikon Eclipse E600 microscope (Nikon Corporation, Tokyo, Japan) at two positions: polar and equatorial view, at 400× magnification, focusing on pollen wall and surface sculpture. The botanical composition of honey was assessed by calculating the frequency of pollen in honey samples and expressed as a percentage of total pollen sum and considered as monofloral if the species was predominant and accounted for 45%; secondary pollen—16–45%; important pollen 3–15; minor pollen <3%. The botanical origin of the pollen was determined by taking photos and compared with the known plant pollen photos presented in the pollen catalogue .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p5
|
PMC11278118
|
sec[1]/sec[1]/p[0]
|
2.2. Protein Isolation and Preparation for LC–MS
| 4.0625 |
biomedical
|
Study
|
[
0.99951171875,
0.00013899803161621094,
0.00029540061950683594
] |
[
0.9990234375,
0.0008697509765625,
0.0002105236053466797,
0.00005644559860229492
] |
Proteins have been identified in the pollen separated from the honey. Pollen proteins were extracted as described in our previous study . Briefly, pollens were homogenized in buffer, then lysed by boiling for 5 min at 95 °C and centrifuged for 30 min. The pellets containing proteins were precipitated using 5 vol of ice-cold 97.6% acetone, stored at −20 °C overnight, and afterwards the pellet was washed twice with 96.6% ethanol by centrifugation. The protein pellet was dissolved in 8 M urea solution and supplied for mass spectrometry analysis.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278118_p6
|
PMC11278118
|
sec[1]/sec[1]/p[1]
|
2.2. Protein Isolation and Preparation for LC–MS
| 4.140625 |
biomedical
|
Study
|
[
0.99951171875,
0.0001914501190185547,
0.00022602081298828125
] |
[
0.99462890625,
0.004619598388671875,
0.0007414817810058594,
0.00016200542449951172
] |
Whole proteome samples were digested with trypsin according to FASP protocol as described by Wiśniewski et al. . Briefly, proteins were diluted in urea, alkylated and digested overnight with TPCK Trypsin 20233 (ThermoFisher Scientific, Vilnius, Lithuania), then centrifuged and additionally eluted using 20% CH 3 CN. The solution was acidified with 10% CF 3 COOH and lyophilized in a vacuum centrifuge. The lyophilized peptides were redissolved in 0.1% formic acid.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p7
|
PMC11278118
|
sec[1]/sec[2]/p[0]
|
2.3. LC–MS E (DIA)-Based Protein Identification
| 4.316406 |
biomedical
|
Study
|
[
0.99951171875,
0.00033664703369140625,
0.0001804828643798828
] |
[
0.99853515625,
0.0008807182312011719,
0.0005025863647460938,
0.0001399517059326172
] |
Liquid chromatography (LC) was performed using a Waters Acquity Ultra-Performance LC system (Waters Corporation, Wilmslow, UK) with an analytical column of ACQUITY UPLC HSS T3 250 mm. Data were acquired using the Synapt G2 mass spectrometer and Masslynx 4.1 software (Waters Corporation) in positive ion mode, using data-independent acquisition (DIA) coupled with ion mobility separation (IMS, UDMS E ) . For the survey scan, the mass range was set at 50–2000 Da with a scan time of 0.8 s. Raw data were lock mass-corrected using the doubly charged ion of [Glu1]-fibrinopeptide B and a 0.25 Da tolerance window and processed with the ProteinLynx Global SERVER (PLGS) version 3.0.1 (Waters Corporation, Manchester, UK) Apex3D and Pep3D algorithms to generate precursor mass lists and associated product ion mass lists for subsequent protein identification and quantification. Peak lists were generated using the following parameters: (i) low energy threshold was set to 150 counts, (ii) elevated energy threshold was set to 50 counts, (iii) intensity threshold was set to 750 counts. Database searching was performed with the PLGS search engine using automatic peptide tolerance and fragment tolerance, minimum fragment ion matches of 1 per peptide and 3 per protein, and false discovery rate (FDR < 4%). Trypsin as the cleavage protease was used for data analysis, one missed cleavage was allowed, and the fixed modification was set to carbamidomethylation of cysteines, the variable modification was set to oxidation of methionine. UniProtKB/SwissProt databases were used for protein identification.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p8
|
PMC11278118
|
sec[1]/sec[3]/p[0]
|
2.4. Statistical Analysis
| 4.089844 |
biomedical
|
Study
|
[
0.99951171875,
0.00017368793487548828,
0.00016570091247558594
] |
[
0.99853515625,
0.0007753372192382812,
0.0003886222839355469,
0.00008052587509155273
] |
Label-free quantification using the TOP3 approach was used for the quantification of proteins. TOP3 intensity was calculated as the average intensity of the three best ionizing peptides using ISOQuant . The maximum FDR of protein identification was set to 1%. Identified proteins were submitted to AgBase (Version 2.0) for annotation of Gene Ontology (GO) functions.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p9
|
PMC11278118
|
sec[2]/sec[0]/p[0]
|
3.1. Comparison of the Protein Number of Monofloral Red Clover Honey with Other Honeys of Different Origins
| 4.007813 |
biomedical
|
Study
|
[
0.99658203125,
0.0002980232238769531,
0.003116607666015625
] |
[
0.99951171875,
0.00022208690643310547,
0.00019216537475585938,
0.000038504600524902344
] |
The total number of proteins identified in the studied honey samples ranged from 606 to 558, where C3 sample contained the highest number of proteins and S5 the lowest . The total number of identified red clover proteins was 240, where monofloral red clover (C3) and monofloral rapeseed (S23) honey samples showed the highest number of red clover proteins, 39 and 40, respectively, although no red clover pollen was found in sample S23. Though the majority of the identified red clover proteins were repetitive throughout all five honey samples, about 25% of red clover proteins were non-repetitive . A heat map was created for the comparison of the abundance of red clover proteins in different honey samples .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p10
|
PMC11278118
|
sec[2]/sec[0]/p[1]
|
3.1. Comparison of the Protein Number of Monofloral Red Clover Honey with Other Honeys of Different Origins
| 4.144531 |
biomedical
|
Study
|
[
0.9990234375,
0.00028395652770996094,
0.0006742477416992188
] |
[
0.99951171875,
0.0001285076141357422,
0.00020503997802734375,
0.000041484832763671875
] |
The highest frequency of occurrence was found for 19 red clover proteins in all honey samples studied. It was found that not every red clover protein was repeated in the individual honey tested. Proteins that are present only in honey samples S22, C3, and S23 are NADH-dependent glutamate synthase, plasma membrane ATPase, zinc finger C3HC4 type protein (RING finger) protein, and bifunctional polymyxin resistance protein ArnA, indicated by the yellow color in the heatmap. The relative composition of these non-repetitive proteins was the lowest compared to the other proteins present in the five honey samples studied, at 20.0%. Among six identified ribosomal proteins, three were identified in all samples, while a 40S s16-like ribosomal protein was present in all sample except S5. The other ribosomal proteins, 40S ribosomal protein s9-2-like (Fragment) and 40S ribosomal protein sa-like (Fragment), were not present in two honey samples, specifically S5 and S15, and the sequence coverage for the latter ribosomal proteins was low—7.41 and 12.50% ( Table 1 ).
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11278118_p11
|
PMC11278118
|
sec[2]/sec[0]/p[2]
|
3.1. Comparison of the Protein Number of Monofloral Red Clover Honey with Other Honeys of Different Origins
| 4.09375 |
biomedical
|
Study
|
[
0.9990234375,
0.00024008750915527344,
0.0005211830139160156
] |
[
0.99951171875,
0.00017631053924560547,
0.0001780986785888672,
0.00004494190216064453
] |
The mitochondrial-like UDP-arabinopyranose mutase, actin 3, and ATP synthase subunit proteins had highest sequence coverage, 38.81%, 37.67%, and 33.33%, compared to all identified proteins in this study. A total of 19 red clover proteins were common to all tested honey samples of different botanical origin and comprised 39.6% of the total red clover proteins identified . The relative abundance of these 19 proteins ranged from 6730 to 62,775.1, with HECT-type E3 ubiquitin transferase being the lowest and S-adenosylmethionine synthase the highest.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p12
|
PMC11278118
|
sec[2]/sec[0]/p[3]
|
3.1. Comparison of the Protein Number of Monofloral Red Clover Honey with Other Honeys of Different Origins
| 3.380859 |
biomedical
|
Study
|
[
0.9990234375,
0.0002422332763671875,
0.0009431838989257812
] |
[
0.9853515625,
0.01390838623046875,
0.0003666877746582031,
0.00026345252990722656
] |
The bifunctional polymyxin resistance protein ArnA was identified only in monofloral rapeseed honey (S23). NADH-dependent glutamate synthase was detected only in the honey sample S22.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p13
|
PMC11278118
|
sec[2]/sec[1]/p[0]
|
3.2. Pollen Composition of Honey Samples from Different Regions of Lithuania
| 4.191406 |
biomedical
|
Study
|
[
0.98779296875,
0.0004181861877441406,
0.0116119384765625
] |
[
0.9990234375,
0.000461578369140625,
0.00046253204345703125,
0.00004178285598754883
] |
The monofloral red clover honey (sample C3) contained 48% of red clover pollen, while rapeseed pollen comprised 35% of total pollen content . The amount of pollen in monofloral rapeseed honey samples S5, S15, and S23 ranged from 47.0 to 54.4% and polyfloral honey sample S22 contained 29.9% of pollen of this species. The secondary pollen of faba bean ( Vicia faba ) and thistle ( Cirsium vulgare ) was found in samples S5 and S22 comprising 35.0% and 21.1% of total pollen, respectively, while, in addition, sample S22 contained 16.8% of buckwheat ( Fagopyrum esculentum ) pollen. The important minor pollen, in particular willow ( Salix caprea ), meadowsweet ( Filipendula ulmaria ), and faba bean ( Vicia faba ), were found in the sample S15 and accounted for 15.6%, 13.1%, and 10.4%, respectively. The samples S22 and S23 contained other important minor pollen, such as fruit tree ( Malus domestica ) and caraway ( Carum carvi ), ranging from 10.6% to 10.3%, respectively. The lowest content of pollen from the latter group, namely, raspberry ( Rubus idaeus ), willow ( Salix caprea ), and maple ( Acer platanoides ), was found in sample C3 and accounted for 6.0%, 5.4%, and 4.0%, respectively. Abundant concentrations of anemophilous pollen (63.0%) were detected in polyfloral honey sample S22. Among them were sweet wormwood ( Artemisia annua ), accounting for 46.0%, and tansy ( Tanacetum vulgare ), accounting for 17.0%. The concentration of honeydew elements was higher in monofloral rapeseed honey samples S5 and S15, accounting for 14.4% and 7.3%, while only 5.0% of it was found in monofloral clover honey.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
PMC11278118_p14
|
PMC11278118
|
sec[2]/sec[1]/p[1]
|
3.2. Pollen Composition of Honey Samples from Different Regions of Lithuania
| 4.035156 |
biomedical
|
Study
|
[
0.99951171875,
0.00013816356658935547,
0.0002148151397705078
] |
[
0.9990234375,
0.000553131103515625,
0.0003247261047363281,
0.00005263090133666992
] |
The principal component analysis (PCA) and the contribution of variables were calculated in R with the factoextra package (v.1.0.7) using transformed Log2 data of protein abundance . Top 30 proteins contributed to the principal components (total contribution of a given protein, on explaining the variations retained by two principal components). Proteins correlated with PC1 and PC2 are the most important in explaining variability in the data set.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p15
|
PMC11278118
|
sec[2]/sec[2]/p[0]
|
3.3. Comparison of the Diversity of Plant Proteins Found in Honey Samples
| 3.974609 |
biomedical
|
Study
|
[
0.998046875,
0.00021529197692871094,
0.001705169677734375
] |
[
0.99951171875,
0.00028061866760253906,
0.00019407272338867188,
0.00003546476364135742
] |
Proteins related to 13 plants were detected in the honey samples ( Table 2 ) and associated with pollen of 8 nectariferous and 5 anemophilous plants. Among them, there were mainly proteins associated with rapeseed ( Brassica napus ), the number of which varied within limits from 59 to 82 and consisted of 12.1%. Lower, but nevertheless relative, amounts of protein are associated with red clover ( Trifolium pratense ) and apple tree ( Malus domestica ) plants—6.2%, willow ( Salix viminalis )—5.0%, and cherry ( Prunus avium )—4.2%. The properties of the latter proteins have been reported in our previous studies .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p16
|
PMC11278118
|
sec[2]/sec[2]/p[1]
|
3.3. Comparison of the Diversity of Plant Proteins Found in Honey Samples
| 3.972656 |
biomedical
|
Study
|
[
0.998046875,
0.00015878677368164062,
0.001621246337890625
] |
[
0.99951171875,
0.00042700767517089844,
0.00024008750915527344,
0.0000388026237487793
] |
There were only three proteins related to faba bean ( Vicia faba ), and their number was the same in all honey samples tested. Honey samples contain a large group of specific proteins for bees ( Apis mellifera ). The number of honeybee-specific proteins obtained in current study ranged from 59 to 61 and accounted for 10.3% of the total amount. Other plant proteins detected in this assay have been associated with pollen from anemophilous plants, e.g., arabidopsis ( Arabidopsis thaliana ), annual mugwort ( Artemisia annua ) and mugwort ( Artemisia keiskeana ) , carrot ( Daucus carota subsp. Sativus), and potato ( Solanum tuberosum ), among which the Artemisia annua pollen was the highest at 7.4%.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p17
|
PMC11278118
|
sec[2]/sec[3]/p[0]
|
3.4. The Identified Proteins of Aphids and Their Endosymbionts
| 4.128906 |
biomedical
|
Study
|
[
0.99853515625,
0.00027441978454589844,
0.0013933181762695312
] |
[
0.99951171875,
0.0001550912857055664,
0.00014972686767578125,
0.000038683414459228516
] |
Honeydew is a sweet and sticky liquid excreted by certain insects, usually aphids, and is collected by bees for honey production. The amount of honeydew found in the honey samples ranged from 2.4% to 14.4%. Detected aphid proteins associated with Acyrthosiphon pisum represent the largest group in comparison to Aphis craccivora and Aphis glycines , and varied within limits of 17–25, 10–16, and 4–11, respectively, in tested honey samples. Considering the relative amount of these protein numbers, we can say that they constitute small amounts of 3.5%, 2.2%, and 1.3% ( Table 2 ). Only four single proteins related to the endosymbiont black bean aphid Buchnera aphidicola ( Aphis fabae ), the soybean aphid Buchnera aphidicola ( Aphis glycines ) and cotton aphid Buchnera aphidicola ( Aphis gossypii ) were found in the studied honey samples ( Table 2 ). Facultative symbionts Serratia symbiotica is a species of bacteria endosymbiont of the black bean aphid Aphis fabae . Slightly more (2–3) facultative endosymbionts proteins of Serratia symbiotica and the Arsenophonus endosymbiont of Aphis craccivor were found in our honey samples than those of Buchnera aphidicola Aphis fabae. The content of aphid endosymbionts in our honey samples was very low, between 0.07 and 0.5%.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p18
|
PMC11278118
|
sec[2]/sec[4]/p[0]
|
3.5. Lactic Acid Bacteria in Honey
| 4.148438 |
biomedical
|
Study
|
[
0.99951171875,
0.0002913475036621094,
0.00040435791015625
] |
[
0.99951171875,
0.00012981891632080078,
0.0002779960632324219,
0.00004851818084716797
] |
We identified proteins associated with Apilactobacillus kunkeei and Apilactobacillus apinorum in five honey samples studied; the number of proteins related to these bacteria prevailed in the range of 87–105 and 23–36, respectively ( Table 3 ). The total numbers of these proteins present in all honey samples are 146 and 471, representing 16.1% and 5.0% of all proteins found in the samples. Proteins specific for bees ( Apis mellifera ) account for 10.3%, while other different LABs account for less than 1.0%. The total percentage of proteins associated with plants in honey samples is 58.84%. The same honey samples contained microbiota including aphis ( Acyrthosiphon pisum ), Aphis craccivora , and Aphis glycines , all LAB taken together, as well endosymbionts, and their numbers accounted for 41.07%. These data indicate that the number of plant proteins exceeds the microbiota present in the honey samples.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11278118_p19
|
PMC11278118
|
sec[2]/sec[4]/p[1]
|
3.5. Lactic Acid Bacteria in Honey
| 4.113281 |
biomedical
|
Study
|
[
0.9990234375,
0.00021386146545410156,
0.0005335807800292969
] |
[
0.99951171875,
0.00018227100372314453,
0.00020778179168701172,
0.00003820657730102539
] |
The data on protein content in honey show a slightly different trend compared to the differentiation in the protein number obtained from microbiota, plants, and bee-specific proteins. The following sequence of these data is obtained after evaluating the significant protein amounts: mean proteins content for Apilactobacillus kunkeei was 161.78 µg, plant proteins, Apilactobacillus apinorum , and Apis mellifera 130.78 µg, 48.56 µg, and 57.8 µg, respectively ( Table 4 ). The relative amount for Apilactobacillus kunkeei is 40.55%, plant proteins 32.78%, while Apilactobacillus apinorum and Apis mellifera accounted for 21.17% and 14.49%, respectively.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278118_p20
|
PMC11278118
|
sec[2]/sec[4]/p[2]
|
3.5. Lactic Acid Bacteria in Honey
| 3.884766 |
biomedical
|
Study
|
[
0.9951171875,
0.000263214111328125,
0.00438690185546875
] |
[
0.99951171875,
0.0003688335418701172,
0.0003211498260498047,
0.00003504753112792969
] |
Statistical analysis reveals significant differences between the protein content of rapeseed and all proteins related to the proteins of anemophilous plants at p < 0.05. The same trend was found between proteins associated with red clover and proteins characteristic of anemophilous plants, such as Arabidopsis thaliana , Solanum tuberosum , and Daucus carota subsp. sativus . However, there were no significant differences between the proteins associated with rapeseed oil and red clover proteins, as well as apple tree proteins.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p21
|
PMC11278118
|
sec[2]/sec[4]/p[3]
|
3.5. Lactic Acid Bacteria in Honey
| 4.113281 |
biomedical
|
Study
|
[
0.9990234375,
0.00026297569274902344,
0.0005612373352050781
] |
[
0.99951171875,
0.0001285076141357422,
0.0002799034118652344,
0.00003522634506225586
] |
Statistically significant differences in protein content are observed between Apilactobacillus kunkeei , Apilactobacillus apinorum , and Apis mellifera , at p < 0.05. Significant differences in protein content are also obtained between plant proteins compared to Apilactobacillus apinorum and Apis mellifera at p < 0.05, while differences were insignificant between the Apilactobacillus kunkeei and plant proteins groups. A significant linear relationship was determined between the content of Apilactobacillus kunkeei and Apilactobacillus apinorum , correlation coefficient ( r = 0.97). Assuming that the identified proteins belong to aphids that live in leguminous plants, the relationship between the amount of faba bean ( Vicia faba ) pollen found in honey and the proteins associated with aphids and lactic acid bacteria was calculated ( Table 2 and Table 4 ).
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278118_p22
|
PMC11278118
|
sec[2]/sec[4]/p[4]
|
3.5. Lactic Acid Bacteria in Honey
| 3.326172 |
biomedical
|
Study
|
[
0.99365234375,
0.0003807544708251953,
0.005985260009765625
] |
[
0.99609375,
0.0029239654541015625,
0.0006818771362304688,
0.00006979703903198242
] |
The strongest correlation coefficients were observed between FBP and LABN as well FBP and LABC, at r = 0.943 and 0.935, respectively. Moderate correlation ( r = 0.764) was found between the LABC and LABN.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p23
|
PMC11278118
|
sec[2]/sec[5]/sec[0]/p[0]
|
3.6.1. Evaluation of Red Clover Proteins According to Biological Processes
| 4.15625 |
biomedical
|
Study
|
[
0.9990234375,
0.00021386146545410156,
0.0005702972412109375
] |
[
0.99951171875,
0.0002276897430419922,
0.0002789497375488281,
0.00004297494888305664
] |
The red clover proteins were submitted to AgBase (version 2.0) and evaluated according to the Gene Ontology resource (GO), which consists of ternary parts . The highest number of proteins related to red clover found in the tested honey samples was associated with the biological processes of metabolic process (eight), biosynthetic process (six), and translation (six). The other biological processes (four) include the metabolic process of cellular amino acids, the metabolic process of sulfur compounds, transport, and cofactor metabolic process ( Tables S1 and S2 ). There is also a small-molecule metabolic process, glycolytic process, glycine biosynthetic process from serine, a cellular amino acid biosynthetic process, and five other processes. The biosynthetic process is associated with nucleoside monophosphate phosphorylation, methionine biosynthetic process nucleoside, S-adenosylmethionine biosynthetic process, and three other processes.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p24
|
PMC11278118
|
sec[2]/sec[5]/sec[0]/p[1]
|
3.6.1. Evaluation of Red Clover Proteins According to Biological Processes
| 4.15625 |
biomedical
|
Study
|
[
0.99951171875,
0.000202178955078125,
0.00029754638671875
] |
[
0.99951171875,
0.00029754638671875,
0.0003058910369873047,
0.000055789947509765625
] |
In total, five proteins were determined as involved in oxidation–reduction process ( Tables S1 and S2 ), namely, phosphorylation—four; a carbon metabolic process—three; each of those processes included two proteins: (i) methylation; (ii) intracellular protein transport; (iii) clathrin coat assembly; a protein was in process of proton transmembrane transport process, and many other processes were also involved. The four proteins group involved in transport, sulfur compound metabolic process, cellular amino acid metabolic process, and cofactor metabolic process accounted for 9.5%. The three smallest protein groups involved in chromosome organization signal transduction and protein folding composed an equal part of 2.4%.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p25
|
PMC11278118
|
sec[2]/sec[5]/sec[1]/p[0]
|
3.6.2. Characteristics of Red Clover Proteins Annotated in the Biological Process and Results of Experimental Data
| 4.121094 |
biomedical
|
Study
|
[
0.99951171875,
0.0002453327178955078,
0.0002532005310058594
] |
[
0.99951171875,
0.0001519918441772461,
0.0002789497375488281,
0.00005453824996948242
] |
In this investigation, we present detailed protein characterization involved in the 125 different biological processes based on protein annotation. Proteins involved in oxidation–reduction included various proteins ( Table S1 ). During the study of honey extracts, the following enzymes involved in the oxidation–reduction process were identified: L-ascorbate oxidase; pyruvate dehydrogenase E1 component subunit beta; oxoglutarate dehydrogenase (succinyl-transferring); NADH-dependent glutamate synthase (Fragment); and formate dehydrogenase (Fragment). The molecular weight and isoelectric point (IEP) of those enzymes were 62.1; 39.1; 117.2; 143.1; and 39.0 kDa and 8.91; 13.61; 1.96; 5.34; and 4.52, accordingly ( Table 1 ).
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p26
|
PMC11278118
|
sec[2]/sec[5]/sec[1]/p[1]
|
3.6.2. Characteristics of Red Clover Proteins Annotated in the Biological Process and Results of Experimental Data
| 4.429688 |
biomedical
|
Study
|
[
0.99951171875,
0.0003895759582519531,
0.00027251243591308594
] |
[
0.9990234375,
0.00028777122497558594,
0.0005383491516113281,
0.0001169443130493164
] |
The metabolic process included eleven different processes, and the proteins were identified in eight processes ( Table S1 ). The enzyme identified in these processes, the beta subunit of the beta component of pyruvate dehydrogenase E1, is involved in the glycolytic process. Enzymes adenosylhomocysteinase, S-adenosylmethionine synthase, and serine hydroxymethyltransferase are involved in the metabolic process. The sequence coverage for proteins from a group of small molecule metabolic processes is higher compared to proteins involved in the oxidation–reduction process, and the coverage varies in the range of 5.34 to 22.98%. Our data reveal the process of tetrahydrofolate interconversion in the metabolic processes of the cellular nitrogen compound as well in small molecule metabolic processes. UTP--glucose-1-phosphate uridylyltransferase (Fragment) participates in the UDP-glucose metabolic process, and ATP:AMP phosphotransferase (Fragment) participates in the nucleoside monophosphate phosphorylation process. NADH-dependent glutamate synthase (Fragment) participates in the glutamate biosynthetic process. The isoelectric point of those enzymes varied from 5.67 to 7.17 and the molecular mass ranged from 42.0 to 143.1 kDa. The lowest molecular mass was determined for UTP--glucose-1-phosphate uridylyltransferase (Fragment) and highest for NADH-dependent glutamate synthase (Fragment) ( Table 1 ). We establish the biosynthesis process that is included in the above-mentioned glycine biosynthetic process mentioned above from serine (A0A2K3NMF1), cellular amino acid biosynthetic process (A0A2K3P3K7), and others, such as methionine biosynthetic process (A0A2K3P3K7), glutamate biosynthetic process (A0A2K3PCD9), nucleoside monophosphate phosphorylation (A0A2K3MUW7), and acetyl-CoA biosynthetic process from pyruvate (A0A2K3NQ11).
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p27
|
PMC11278118
|
sec[2]/sec[5]/sec[1]/p[2]
|
3.6.2. Characteristics of Red Clover Proteins Annotated in the Biological Process and Results of Experimental Data
| 4.289063 |
biomedical
|
Study
|
[
0.99951171875,
0.0002486705780029297,
0.00022339820861816406
] |
[
0.9990234375,
0.00029349327087402344,
0.00044226646423339844,
0.00007867813110351562
] |
The protein translation process is related to ribosomal proteins. Data reveal six ribosomal proteins: 40S ribosomal protein s9-2-like (Fragment), 40S ribosomal protein s16-like, 40S ribosomal protein s13-like (Fragment), 60S ribosomal protein l10-like (Fragment), 40S ribosomal protein sa-like (Fragment), and 60S ribosomal protein l4-like (Fragment). The data show that these proteins are characterized by high isoelectric points 10.63–11.09, except 40S ribosomal protein sa-like (Fragment), which was extracted at isoelectric point of 5.07. This indicates that the latter proteins are exclusively basic. The molecular weight of the identified ribosomal proteins ranged from 13.6 to 28.6 kDa and the number of reported peptides ranged from 2 to 6. The sequence coverage (31.09%) was highest for the 40S ribosomal protein s13-like (Fragment); lowest (7.41%) for the 40S ribosomal protein s9-2-like (Fragment), and for others from 17.86% to 21.05%.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p28
|
PMC11278118
|
sec[2]/sec[5]/sec[2]/p[0]
|
3.6.3. Evaluation of Red Clover Proteins According to Molecular Functions
| 4.027344 |
biomedical
|
Study
|
[
0.99951171875,
0.0001537799835205078,
0.0005011558532714844
] |
[
0.99951171875,
0.00038552284240722656,
0.000152587890625,
0.00004363059997558594
] |
The molecular function usually is a single-step reaction. We present red clover proteins according to the annotations determined for 128 molecular functions that were detected from experimental study of honey samples ( Table 1 and Table S1 ). Among the proteins whose molecular functions were determined are the activity of phosphotransferase and an alcohol group as acceptor ( Table S1 ). This protein catalyzes the transfer of a phosphorus-containing group from a compound (donor) to an alcohol group (acceptor).
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278118_p29
|
PMC11278118
|
sec[2]/sec[5]/sec[2]/p[1]
|
3.6.3. Evaluation of Red Clover Proteins According to Molecular Functions
| 4.179688 |
biomedical
|
Study
|
[
0.99951171875,
0.00022864341735839844,
0.00041961669921875
] |
[
0.99951171875,
0.00013840198516845703,
0.00031065940856933594,
0.0000502467155456543
] |
The other proteins with molecular function annotated in the GO description and related to red clover proteins involve two different areas ( Tables S1 and S3 ). Data corresponding to molecular functions of proteins include protein serine/threonine kinase activity; hydrolase activity; methionine adenosyltransferase activity; adenosylhomocysteinase activity; adenylate kinase activity; transferase activity; catalytic activity; peptidase activity; protein serine/threonine kinase activity; kinase activity; adenylate kinase activity; and others. The molecular function is associated with proteins and the metal ions binding process. Our data reveal a molecular binding function for those proteins: NAD binding; pyridoxal phosphate binding; nucleotide binding; coenzyme binding; thiamine pyrophosphate binding; ADP binding; protein domain-specific binding; clathrin light chain binding; ATP binding; RNA binding; and rRNA binding, as well as others ( Table S1 ). In the studied honey samples, we identified zinc ion binding; copper ion binding; and metal ion binding as well as some clusters with iron: 4 iron, 4 sulfur cluster binding; and iron–sulfur cluster binding. The study determined the number of proteins involved in various molecular functions ( Table S1 ).
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278118_p30
|
PMC11278118
|
sec[2]/sec[5]/sec[3]/p[0]
|
3.6.4. Evaluation of Red Clover Proteins According to the Cellular Components or Macromolecular Complexes
| 4.1875 |
biomedical
|
Study
|
[
0.99951171875,
0.00020062923431396484,
0.0003199577331542969
] |
[
0.99951171875,
0.00022459030151367188,
0.00033855438232421875,
0.000046253204345703125
] |
Gene Ontology analysis of the cellular components reveals 78 terms of the cell components based on the data of the annotation results ( Table S1 ). The major cell components include the nucleus; nucleoplasm; cytoplasm; and Golgi apparatus. According to the GO analysis, the cellular components of red clover include small ribosomal subunit; mitochondrial matrix; clathrin complex; proton-transporting V-type ATPase, V1 domain; clathrin coat of the trans-Golgi network vesicle; and clathrin coat of the coated pit ( Table S1 ). Various membranes were identified, including cytoplasmic vesicle membrane; integral component of the membrane; and plasma membrane ( Tables S1 and S4 ). Protein-containing complexes were found as assemblies that contains small ribosomal subunit; clathrin complex; clathrin coat of trans-Golgi network vesicle; clathrin coat of coated pit and histone acetyltransferase complex; and proton-transporting V-type ATPase, V1 domain ( Tables S1 and S4 ).
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278118_p31
|
PMC11278118
|
sec[3]/p[0]
|
4. Discussion
| 4.191406 |
biomedical
|
Study
|
[
0.9990234375,
0.0002789497375488281,
0.0007109642028808594
] |
[
0.99951171875,
0.00012314319610595703,
0.0002918243408203125,
0.00004589557647705078
] |
The protein composition of five honey samples was investigated, among which one sample was monofloral red clover honey. However, all samples of honey contained proteins related to red clover proteins. Seventeen red clover proteins whose peptides were sequenced had protein coverage greater than 20.0%, and five proteins ranged from 17.86% to 10.34%. A total of 26 proteins out of 48 studied had a sequence coverage of 9.97 to 0.59%, among which the lowest was HECT-type E3 ubiquitin transferase ( Table 1 ). A total of 331 functions of red clover proteins were annotated, among which 125 are involved in biological processes, 128 have a molecular function, and 78 are related to the components of cell structure ( Table S1 ). The molecular function of the proteins studied was associated with binding, catalytic, transporter, structural, and nucleic acid binding transcription factor activities, as well as with components of membranes, organelle, and protein-containing complexes. The red-clover-related proteins have been identified in various cellular components are as follows: histone acetyltransferase complex, proton-transporting V-type ATPase, V1 domain, and clathrin complex ( Table S1 ).
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p32
|
PMC11278118
|
sec[3]/p[1]
|
4. Discussion
| 4.09375 |
biomedical
|
Study
|
[
0.99951171875,
0.00015616416931152344,
0.00039505958557128906
] |
[
0.9990234375,
0.0003230571746826172,
0.00042700767517089844,
0.00004661083221435547
] |
The protein sequence-based prediction method was applied for the annotation of protein function of leguminous plant species such as barrel clover ( Medicago truncatula ) and soybean ( Glycine max ), as well as other plants among which were rice ( Oryza sativa ), poplar ( Populus trichocarpa ), and tomato ( Solanum lycopersicum ) . GO analysis of the above-mentioned and the additional five species of Trifolium subterraneum , Medicago truncatula , Cicer arietinum , Trifolium pratense , and Glycine max confirms the genes involved in biological processes assist in metabolic, cellular, single-organism, localization, biological regulation, and signal processes, etc., and this has been reported in earlier studies .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p33
|
PMC11278118
|
sec[3]/p[2]
|
4. Discussion
| 4.199219 |
biomedical
|
Study
|
[
0.99951171875,
0.00023448467254638672,
0.00033593177795410156
] |
[
0.99951171875,
0.00019788742065429688,
0.00023090839385986328,
0.00005412101745605469
] |
The two proteins identified in honey samples extract are V-ATPase 69 kDa subunit and clathrin heavy chain ( Table 1 ). We have identified L-ascorbate oxidase belonging to the family of multi-copper oxidases and function in plants and fungi by oxidizing ascorbic acid to dehydro-L-ascorbic acid, a potentially toxic product damaging the digestive system of the herbivore Helicoverpa zea . Oxidized ascorbate loses its antioxidant properties, as well as its nutritional and antioxidant functions in phytophagous insects and, thus, has a potentially important role as a plant defense protein against insects . The other enzyme of the oxidase family, namely, glucose oxidase (GOD), with a molecular weight (Mw) of approximately 170 kDa, was identified in buckwheat honey . In this origin of honey, GOD had the highest enzymatic activity of 1.13 µmol min −1 /g −1 , compared to rapeseed (0.55 µmol min −1 /g −1 ) or willow honey (0.1–0.51 µmol min −1 /g −1 ) .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278118_p34
|
PMC11278118
|
sec[3]/p[3]
|
4. Discussion
| 4.386719 |
biomedical
|
Study
|
[
0.99951171875,
0.0002689361572265625,
0.0003597736358642578
] |
[
0.9990234375,
0.00029587745666503906,
0.00042128562927246094,
0.00008690357208251953
] |
Auxin is a plant hormone playing an important role in plant growth and signaling, and it regulates photosynthetic rates and chlorophyll content in many plant species . The BIG auxin transport protein is involved in auxin efflux and polar auxin transport (PAT) . Auxin transport big-like protein, with a molecular weight of 573.9 kDa and isoelectric point pI of 5.78, was found in four of the honey samples studied, among which was monofloral clover honey ( Table 1 ). It was annotated as the protein is involved in the binding of zinc ions ( Table S1 ). It has a high molecular weight of 498.9 kDa and was annotated as zinc finger protein (ring finger) protein (Fragment), and was identified in monofloral clover honey. The study by Mohanta and colleagues on the plant proteome reveals that higher eukaryotic plants contain five proteins with higher molecular mass compared to others, among which is auxin transport protein BIG 568.4 kDa; the auxin protein (auxin TP BI) was identified in the molecular weight of Trifolium pratense of 566.6 kDa compared to other proteins in this plant. The study reveals that proteins with acidic pI predominate over the proteins with an alkaline pI, suggesting that it depends on differential composition of amino acids in different species, and this might be associated with environmental and ecological pressure. The enzyme S-adenosylmethionine synthase, also known as methionine adenosyltransferase (MAT), was determined during our analysis of honey samples. According to research data, this enzyme catalyzes the formation of S-adenosylmethionine (AdoMet, SAM, or SAMe) from methionine and ATP . The SAM component is a precursor for the plant hormone ethylene .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278118_p35
|
PMC11278118
|
sec[3]/p[4]
|
4. Discussion
| 4.175781 |
biomedical
|
Study
|
[
0.9990234375,
0.00011181831359863281,
0.0006227493286132812
] |
[
0.99755859375,
0.0015172958374023438,
0.0006670951843261719,
0.00006216764450073242
] |
UTP-glucose-1-phosphate uridylyltransferase, also known as glucose-1-phosphate uridylyltransferase (UDP–glucose pyrophosphorylase) or UGPase, is an enzyme involved in carbohydrate metabolism and has been used to improve the quality and increase the production of agricultural plants . This enzyme was identified in the honey samples we studied, supporting the earlier findings of Treigytė and coauthors , who found protein uridine diphosphate (UDP)-glucose 6-dehydrogenase 4 and UDP-glucose 6-dehydrogenase 5 with a molecular weight of 56 kDa and 55 kDa, respectively, in red clover pollen. Glucose-1-phosphate uridylyltransferase catalyzes the formation of uridine diphosphate-glucose (UDP-glucose) from glucose-1-phosphate and uridine-5′-triphosphate (UTP) .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p36
|
PMC11278118
|
sec[3]/p[5]
|
4. Discussion
| 4.613281 |
biomedical
|
Study
|
[
0.9990234375,
0.0004162788391113281,
0.00046181678771972656
] |
[
0.99853515625,
0.0005450248718261719,
0.0007300376892089844,
0.00015223026275634766
] |
The sucrose–phosphate synthase enzyme catalyzes the transfer of a uridine diphosphate glucose (UDP-glucose) to D-fructose 6-phosphate to form UDP and D-sucrose-6-phosphate, and in the following step, hydrolyzes D-sucrose-6-phosphate to sucrose . Sucrose phosphate synthase is an essential enzyme in sucrose synthesis, suggesting that the expression of UTP-glucose-1-phosphate uridylyltransferase in plant nectar affects the formation of UDP-glucose, which is an essential component in carbohydrate metabolism. Uridine diphosphate glucose (UDP-glucose) is a nucleotide sugar involved in the synthesis of cellulose, hemicellulose, and pectins for the production of plant cell walls . Nucleotide sugars such as UDP-glucose, GDP-fucose, UDP-xylose, and UDP-N-acetyl galactosamine are transported from one UDP-sugar to another for the exchange of various nucleotides . The transport of UDP-sugars in cells takes place through specific membrane-bound protein transporters. The synthesized compounds are used for the production of glycoproteins and glycolipids . Nitrogen fixation chloroplastic NifU-like protein 3, known for its functions in protecting plants against abiotic and biotic stresses , was identified in spring honey collected by bees mainly from orchards or manually collected directly from orchard blooms , showing its higher expression in manually collected pollen of apple tree ( Malus sylvestris ) and plum ( Prunus ) compared to other orchard tree pollen. The concentration range of the bee-collected pollen of the Prunus and willow ( Salix spp.) mixture was close to that of monofloral Prunus pollen. Purified NifU is a red protein that contains iron in the form of a redox-active [2Fe-2S] 2+,+ cluster. The primary structure consists of the three conserved cysteine residues that are involved in the assembly of a transient FeS cluster . It was stated that transient [2Fe-2S] 2+,+ cluster units are formed on NifU and subsequently released to provide the inorganic iron and sulfur required for the maturation of the nitrogenase component proteins . NifU-type proteins forming iron–sulfur clusters can also be found in organisms that do not fix nitrogen . Although we did not observe NifU-like protein in the honey samples studied, other enzymes with iron–sulfur clusters, in particular iron–sulfur protein aconitate hydratase containing [4Fe-4S] cluster, was determined.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p37
|
PMC11278118
|
sec[3]/p[6]
|
4. Discussion
| 4.230469 |
biomedical
|
Study
|
[
0.99951171875,
0.00017976760864257812,
0.00041794776916503906
] |
[
0.9990234375,
0.0005598068237304688,
0.00022149085998535156,
0.00005692243576049805
] |
Aconitate hydratase, later named aconitase, belongs to the aconitase/isopropylmalate (IPM) isomerase family, comprised of three classes of hydrolyase enzymes: aconitases, homoaconitases, and IPM isomerases. These enzymes have the same Fe-S cluster-binding site in their structure . According to research data, tricarboxylic acid cycle (TCA), also known as the Kreb’s/Citric Acid cycle, is activated in bacteroids during nitrogen fixation. The components of TCA are the main carbon source for bacteroids . The aconitase (AcnA), as well as other seven proteins from TCA, were identified by the proteomics analysis in nitrogen fixation bacteroid Rhizobium Etli . These data were obtained during the experiment when the root nodules of the bean plant ( Phaseolus vulgaris ) were infected by R. etli ( Phaseolus vulgaris ) .
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278118_p38
|
PMC11278118
|
sec[3]/p[7]
|
4. Discussion
| 4.046875 |
biomedical
|
Study
|
[
0.99072265625,
0.0002665519714355469,
0.0088043212890625
] |
[
0.99853515625,
0.0008897781372070312,
0.0003209114074707031,
0.00005137920379638672
] |
Aphids, such as the genera Cinara and Physokermes , which have spread in forests in coniferous trees, including spruce and fir, feed on the sap of the phloem of these plants and produce honeydew honey with a varying composition of sugar, amino acids, and several inorganic ions, among which potassium ions (K+) and phosphate (PO4 3− ) are the most abundant anion in honeydew honey . The pea aphid, Acyrthosiphon pisum (Harris), infests and can cause huge damage to leguminous plants, such as faba bean ( Vicia faba L.), lupin ( Lupinus albus L.), alfalfa ( Medicago sativa L.), lentil ( Lens culinaris Medik.), chickpea ( Cicer arietinum L.), grass pea ( Lathyrus sativus L.), pea ( Pisum sativum L.), and clover ( Trifolium subterraneum ) . Acyrthosiphon pisum was reported to prefer faba bean and clover and can transmit phytoviruses and cause more than 20 species of plant virus diseases . Though red clover aphids ( Aphis coronilla Ferrari (Hemiptera: Aphididae)) are widely distributed in Europe in pulses , mostly on red clover and lucern, we did not identify any of their proteins in the studied honey, suggesting that these pests are not widespread in the Lithuanian legume fields.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278118_p39
|
PMC11278118
|
sec[3]/p[8]
|
4. Discussion
| 4.507813 |
biomedical
|
Study
|
[
0.9990234375,
0.0003261566162109375,
0.000530242919921875
] |
[
0.998046875,
0.0003647804260253906,
0.0013761520385742188,
0.0001017451286315918
] |
The evolution of symbionts along with their hosts existed million years ago as a long-term mutualism. Symbionts are divided into obligate or primary and other facultative or secondary, as well-named, S-symbionts . The primary endosymbiont of pea aphid, Buchnera aphidicola , synthesizes essential amino acids for its aphid host. About 9% of the Buchnera genome produces essential amino acids for the aphid . Some new members of the Lactobacillus genus currently belong to the genus Apilactobacillus , which includes different species. Apilactobacillus kunkeei (basonym Lactobacillus kunkeei ) and Apilactobacillus apinorum (basonym Lactobacillus apinorum ) are incorporated into the genus Apilactobacillus . These LABs are associated with honeybees ( Apis mellifera ) and flowers. Lactic acid bacteria (LAB) are common inhabitants of the honeybee intestinal tract and are found in fresh honey . The authors report that the bee gastric microbiota is dominated by lactobacilli and bifidobacteria. These bacteria are also found in flowers, nectar, fruits, and fermented foods. We identified Lactiplantibacillus plantarum , Lactiplantibacillus acidophilus , Lactiplantibacillus amylovorus , and Lactiplantibacillus delbrueckii subsp. bulgaricus . Most of these proteins were single or absent in the honey samples tested. The strains L. delbrueckii were mainly isolated from dairy products, including cheeses, yogurts, and fermented milk. The data reveal nine species of L. delbrueckii , which are probiotics. The probiotic strains of these bacteria have genes involved in various metabolic processes that affect the organoleptic properties of dairy . In the honey we studied, these bacteria were identified in four samples out of five examined, and their number was very low (1–2); 2–4 proteins were found from Lactiplantibacillus plantarum and 2–3 proteins from Lactiplantibacillus acidophilus . Fructophilic LABs have been found in the digestive tracts of pollinators such as bees, bumblebees, and insects that consume significant amounts of fructose . The prevalence of Apilactobacillus kunkeei , Lactiplantibacillus plantarum , and Fructobacillus fructosus was found in beebread, while low interspecies biodiversity of LAB Apilactobacillus kunkeei was found in the midgut of Apis mellifera ligustica honeybee . A. kunkeei belongs to fructophilic (FLAB), a lactic acid bacteria of a subgroup of LAB, and grows on fructose, unlike other LAB, that grow on glucose ; thus, it often is found in environments associated with bees as high-fructose-consuming insects, on flowers, fruits, and also in fermented food that is produced from fruits . The fructophilic strain of L. plantarum was isolated from honeydew honey collected in Poland and since Poland and Lithuania are neighboring countries with the same climatic conditions, it suggests that the honeydew composition may be similar in terms of fructophilic lactic acid bacteria.
|
[
"Violeta Čeksterytė",
"Algirdas Kaupinis",
"Andrius Aleliūnas",
"Rūta Navakauskienė",
"Kristina Jaškūnė"
] |
https://doi.org/10.3390/life14070862
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278131_p0
|
PMC11278131
|
sec[0]/p[0]
|
1. Introduction
| 4.113281 |
biomedical
|
Review
|
[
0.99462890625,
0.0023345947265625,
0.0028934478759765625
] |
[
0.0244140625,
0.020538330078125,
0.9541015625,
0.0007061958312988281
] |
Artificial intelligence (AI) has revolutionized the medical imaging landscape, offering innovative applications that aid diagnosis and treatment. In diagnostic radiology, deep learning algorithms, such as those developed by Zebra Medical Vision and Aidoc, analyze X-rays and CT scans to detect a range of conditions, providing faster and sometimes more accurate readings than traditional methods . In pathology, companies like PathAI use AI to identify patterns in tissue samples, improving cancer diagnoses . Similarly, in ophthalmology, tools like IDx-DR for diabetic retinopathy screening autonomously assess retinal images to identify early signs of disease . In cardiology, AI-powered software like that from Arterys evaluates cardiac MRI and CT scans to provide detailed insights into heart structure and function, aiding in the diagnosis of cardiovascular diseases . Despite these advancements, AI applications are not without concerns. The ‘black box’ nature of many AI systems, where the decision-making process is not transparent, poses challenges to clinical validation and trust. Data privacy and security are also significant issues, as AI models require large datasets for training, potentially exposing sensitive patient information if data are breached or improperly accessed . Real-world breaches, such as the Anthem Inc. and UCLA Health System breaches, underscore these vulnerabilities. Additionally, algorithmic bias and errors in AI systems necessitate meticulous dataset curation and algorithm training to ensure equitable and accurate medical services.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278131_p1
|
PMC11278131
|
sec[0]/p[1]
|
1. Introduction
| 4.113281 |
biomedical
|
Study
|
[
0.9990234375,
0.00064849853515625,
0.00035262107849121094
] |
[
0.9951171875,
0.001125335693359375,
0.0034427642822265625,
0.00015401840209960938
] |
The adoption of Generative Adversarial Networks (GANs) to generate synthetic data presents a promising solution to these challenges . GANs can create realistic medical images, reducing the need to use them and potentially exposing sensitive patient data . This method of data augmentation enriches the dataset required for robust AI diagnostic tools and serves as a critical buffer for maintaining patient privacy. In the current study, we utilized GANs for synthetic image generation in genitourinary pathology, highlighting their potential in this context. The GANs underwent rigorous quality control processes, including validation by board-certified pathologists and quantification of image fidelity through Relative Inception Scores and Fréchet Inception Distance, demonstrating high-quality synthetic image production. These images were indistinguishable from real data in many instances, enabling their use in AI diagnostics without the risk associated with actual patient data. By incorporating synthetic data generation via GANs, the healthcare industry can safeguard sensitive patient information, addressing one of the most significant cybersecurity concerns of our time. As we continue to navigate the complexities introduced by AI in healthcare, the role of GANs in cybersecurity becomes increasingly pertinent. They represent a promising path forward, integrating AI into medical practice in a secure, ethical, and conducive manner to patient trust and safety.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p2
|
PMC11278131
|
sec[1]/sec[0]/p[0]
|
2.1. Cohorts Used
| 4.121094 |
biomedical
|
Study
|
[
0.99951171875,
0.0002065896987915039,
0.00017261505126953125
] |
[
0.9990234375,
0.00026106834411621094,
0.0004582405090332031,
0.00006258487701416016
] |
We harnessed eight genitourinary tissue types—bladder, cervix, kidney, ovary, prostate, testis, uterus, and vagina—obtained from the Genotype-Tissue Expression (GTEx) database, a comprehensive resource that provides open access to tissue expression data. Additionally, histology images from the cancer genome atlas (TCGA) of 500 individuals representing the adenocarcinoma stage were considered controls. Segmentation was performed using PyHIST, a Python-based histological tool, which processed the images into discrete squares of 64, 128, and 256 pixels. Each segment was curated to contain a minimum of 75% tissue content, a criterion set to minimize regional bias and preserve the representativeness of the histological features.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p3
|
PMC11278131
|
sec[1]/sec[1]/p[0]
|
2.2. Development and Evaluation of a Conditional Generative Adversarial Network
| 3.390625 |
biomedical
|
Study
|
[
0.81787109375,
0.0006737709045410156,
0.1815185546875
] |
[
0.96826171875,
0.0307769775390625,
0.0005192756652832031,
0.000225067138671875
] |
A preliminary conditional Generative Adversarial Network (cGAN) was designed and implemented to assess the performance accuracy of various GAN architectures. The cGAN was developed utilizing Python 3.7.3 and the Tensorflow Keras 2.7.0 package. The generator component of the cGAN comprises three input layers and a single output layer. In parallel, the discriminator component is configured with analogous input, hidden, and output layers. The cGAN’s total parameter count was 7.5 million for each of the evaluated image patterns.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p4
|
PMC11278131
|
sec[1]/sec[2]/p[0]
|
2.3. Implementation and Adaptation of StyleGAN for Tissue Image Analysis
| 4.191406 |
biomedical
|
Study
|
[
0.9990234375,
0.00036454200744628906,
0.0006246566772460938
] |
[
0.9970703125,
0.0021190643310546875,
0.0005202293395996094,
0.0001024007797241211
] |
StyleGAN, a progressive generative adversarial network architecture engineered using Python 3.9 and the TensorFlow framework, leveraging the conditional GAN architecture, was used to guide the image synthesis process. This structure allowed the GAN to generate images conditioned on specific tissue types, facilitating targeted image generation. To automate and streamline the process, we employed a bash script tailored for each tissue type that orchestrated the importation of images, their conversion to an RGB color space, and the compilation of these images into a NumPy array. These arrays were then stored as .npy files, ensuring reproducibility and consistency across GAN runs. During the synthetic image generation phase, the generator component of the GAN introduced random noise variables, which were assessed by the discriminator component. This interplay continued iteratively, with the loss graph monitored meticulously until stabilization was observed—a signal to cease the discriminator’s assessment and crystallize the synthetic image output. On average, the GAN system required 2.5 h per run, yielding a thousand synthetic images per tissue type. The loss functions—mathematical functions quantifying the error between the generated images and the actual images—were pivotal in guiding the GAN’s training. Monitoring these allowed us to fine-tune the GAN’s parameters, with an observed convergence of loss functions around the 182nd epoch. This convergence was deemed the optimal stopping point, indicative of the GAN’s ability to generate images with minimal discrepancy from the target dataset. The term “loss” here referred specifically to the number of images that were not deemed accurate enough by the GAN, thereby being ‘dismissed’ during the iterative training process.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
PMC11278131_p5
|
PMC11278131
|
sec[1]/sec[3]/p[0]
|
2.4. StyleGAN
| 3.121094 |
biomedical
|
Study
|
[
0.9541015625,
0.0006146430969238281,
0.045196533203125
] |
[
0.73974609375,
0.258544921875,
0.00109100341796875,
0.0005965232849121094
] |
The StyleGAN implementation was obtained from the NVIDIA Labs Github and was run with Python 3.9. Tensorflow version 1.12.0 and CUDA version 10.2 were used. The training images were imported into a TFRecords dataset object and stored as a .tfrecords file. Initial training was performed on a V100 Tesla GPU, and it took an average of 2.5 days to complete the first round of training. This trained model was then used as the basis for generating new tissue images. The architecture of StyleGAN was kept exactly as is from the NVIDIA download . The only parameter that was changed to generate sufficient images was the resolution factor. This resolution factor was set to 256 in order to output the images at a quality that could be inspected manually.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p6
|
PMC11278131
|
sec[1]/sec[4]/p[0]
|
2.5. Quantification Model
| 4.230469 |
biomedical
|
Study
|
[
0.99951171875,
0.0002810955047607422,
0.00038051605224609375
] |
[
0.9990234375,
0.00029397010803222656,
0.000530242919921875,
0.00006431341171264648
] |
To characterize technical and structural variations between synthetic and real images, we utilized Spatial Heterogeneous Recurrence Quantification Analysis (SHRQA), a robust technique capable of measuring complex microstructures based on spatial patterns . The SHRQA process, as shown in Supplementary Figure S4 , involves six key steps. It begins with the 2D-Discrete Wavelet Transform (2D-DWT) using the Haar wavelet to reveal patterns not visible in the original image . Then, each image is transformed into an attribute vector via the space-filling curve (SFC), which importantly preserves the spatial proximity between pixels in the image within the vector. This step is crucial for analyzing the image’s geometric recurrence in vector form. A trajectory is formed in state space by projecting this attribute vector, highlighting the image’s geometric structure. Through quadtree segmentation, the state space is divided into unique subregions to discern spatial transition patterns . An Iterated Function System projection is then applied, converting each attribute vector into a fractal plot that represents recurrence within the fractal topology. Finally, these fractal structures are quantified to illuminate the intricate geometric properties of the image, providing a detailed profile.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278131_p7
|
PMC11278131
|
sec[1]/sec[5]/p[0]
|
2.6. Statistical Calculations
| 2.919922 |
biomedical
|
Study
|
[
0.98828125,
0.0007562637329101562,
0.01081085205078125
] |
[
0.80908203125,
0.18896484375,
0.0014781951904296875,
0.0007038116455078125
] |
FID was implemented in custom scripts developed in-house. The FID model was pre-trained using Inception V3 weights for transfer learning. In-house code was centered around the FID model and inserted into the StyleGAN to be run during each iteration. Stats were reported at intervals of 1000 and graphed with in-house Python scripts. Reported FID figures represent an inverse relationship between the images; thus, the lower our FID figure, the more similar the images.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
PMC11278131_p8
|
PMC11278131
|
sec[1]/sec[5]/p[1]
|
2.6. Statistical Calculations
| 3.917969 |
biomedical
|
Study
|
[
0.9990234375,
0.00015103816986083984,
0.0007081031799316406
] |
[
0.99755859375,
0.0018529891967773438,
0.00030541419982910156,
0.00006121397018432617
] |
PCA analysis was performed by first transforming the images into numerical arrays. Images were separated into normal and synthetic batches. The intensity was calculated (using the R package imgpalr and magick) as the average of the color of the entire image while keeping the matrix framework (i.e., positional arguments were retained). PCA was conducted using the general prcomp function in R, and the plotted results were displayed in ggplot2.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11278131_p9
|
PMC11278131
|
sec[1]/sec[6]/p[0]
|
2.7. Data Sharing
| 1.836914 |
clinical
|
Other
|
[
0.1678466796875,
0.74365234375,
0.08837890625
] |
[
0.01324462890625,
0.9833984375,
0.0005121231079101562,
0.00279998779296875
] |
De-identified participant data will be made available when all primary and secondary endpoints have been met. Any requests for trial data and Supporting Material (data dictionary, protocol, and statistical analysis plan) will be reviewed by the trial management group in the first instance. Only requests that have a methodologically sound proposal and whose proposed use of the data has been approved by the independent trial steering committee will be considered. Proposals should be directed to the corresponding author in the first instance; to gain access, data requestors will need to sign a data access agreement.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278131_p10
|
PMC11278131
|
sec[2]/p[0]
|
3. Results
| 4.097656 |
biomedical
|
Study
|
[
0.99951171875,
0.000232696533203125,
0.000225067138671875
] |
[
0.99951171875,
0.00022292137145996094,
0.0004119873046875,
0.000059664249420166016
] |
GAN Model Selection: To evaluate the performance of various GAN architectures and select the most appropriate one, digital histology images were downloaded from the Genotype-Tissue Expression (GTEx) database for the prostate. In 9091, 256 × 256 image patches were extracted from 599 individuals and divided into training cohorts. Each training cohort was subjected to cGAN, StyleGAN, and dcGAN architectures . A total of 200 randomly selected synthetic images generated by each GAN were fed into a generic CNN for classification. The cGAN achieved an accuracy of 36% (72 images were classified correctly), while the StyleGAN and dcGAN demonstrated accuracies of 62.5% (125 correctly classified) and 60 (120 correctly classified), respectively. Although StyleGAN and dcGAN exhibited similar accuracies, the quality of output was more extensive for StyleGAN, which is particularly important considering the less heterogeneity that exists in standard/non-cancer tissue image types.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p11
|
PMC11278131
|
sec[2]/p[1]
|
3. Results
| 4.164063 |
biomedical
|
Study
|
[
0.99951171875,
0.00038504600524902344,
0.00016021728515625
] |
[
0.9990234375,
0.0003027915954589844,
0.0004892349243164062,
0.00011050701141357422
] |
Image synthesis: Once the GAN was selected, the GTEx database was used to extract digital histology images from eight genitourinary tissue types; 129 images were available for the bladder, 81 for the cervix, 599 for the kidney, 252 for the ovary, 599 for the prostate, 588 for the testis, 234 for the uterus, and 272 for the vagina. Several factors, such as staining protocols, tissue quality, section thickness, tissue folding, and the amount of tissue on the slide, could negatively impact the efficiency of the GAN model in generating high-quality data . To account for this, we conducted pre-processing normalization of the images. Specifically, we selected all the images from all tissue types and evaluated their color distribution by calculating the mean value of RGB colors and normalizing them. Images with an RGB mean intensity value two standard deviations away from the total mean value of all samples were identified as outliers and removed from the dataset. In total, 21 images were discarded due to being outliers. Overall, our pre-processing steps helped to reduce the variability in tissue biopsy images and ensure a more consistent training dataset for the StyleGAN model. Post-processing, these images were used to train the StyleGAN model. The network generator created a total of 200 random synthetic images for each of the tissue types. The patch size of each of these images was set at 5000 * 5000 to allow sufficient quality for the pathologist’s evaluation. These image patches were analyzed using the Adam optimization algorithm. This process helped us find the best iteration value for our model, which was 15,000 iterations. Figure 1 A summarizes the steps in the processing and generation of synthetic images, and Figure 1 B and Supplementary Figures S1–S8 showcase the examples of synthetic images generated from eight GU tissues.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p12
|
PMC11278131
|
sec[2]/p[2]
|
3. Results
| 4.085938 |
biomedical
|
Study
|
[
0.99951171875,
0.000274658203125,
0.0003211498260498047
] |
[
0.99951171875,
0.00020694732666015625,
0.00034117698669433594,
0.00004947185516357422
] |
Next, we applied standard machine learning metrics to evaluate the synthetic images. The Relative Inception Score (RIS) was a primary metric, measuring the clarity and variety of the generated images. A high RIS of 17.2 with a remarkably low standard deviation of 0.15 across different tissue types demonstrated the synthetic images’ consistent quality. Furthermore, the Fréchet Inception Distance (FID), a crucial index for GAN performance, was used to compare the distribution of generated images with real images. An FID score that stabilized at 120 indicated that the synthetic images closely mirrored the distribution of the real tissue images, solidifying the efficacy of our GAN model.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p13
|
PMC11278131
|
sec[2]/p[3]
|
3. Results
| 4.109375 |
biomedical
|
Study
|
[
0.99853515625,
0.001190185546875,
0.00019848346710205078
] |
[
0.998046875,
0.0010328292846679688,
0.0005707740783691406,
0.0001806020736694336
] |
Quality Control Through Expert Evaluation: The synthetic images underwent a rigorous review process for quality control. A subset of synthetic prostate images were subjected to detailed visual inspection, focusing on aspects such as sharpness and resolution. This scrutiny was critical to ensuring that the generated images met the high standards required for clinical use. For this, two certified pathologists conducted an independent review of the synthetic image cohorts, where they were provided with a randomized pool of 20 images per tissue type, consisting of a mixture of 15 synthetic and five real images, totaling 160 images. The pathologists were tasked with evaluating the quality of the images and highlighting the concerns they may have for each tissue type. Table 1 summarizes the quality evaluation outcomes of pathology evaluation for all eight tissue types. Supplementary Figures S1–S4 show the 20 images per tissue type shared with the pathologists. Results highlighted an 80% approval rate, signifying a robust endorsement of the synthetic images’ clinical utility.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p14
|
PMC11278131
|
sec[2]/p[4]
|
3. Results
| 4.109375 |
biomedical
|
Study
|
[
0.99951171875,
0.00022077560424804688,
0.00038814544677734375
] |
[
0.9990234375,
0.00033783912658691406,
0.0004982948303222656,
0.000050187110900878906
] |
Geometric Analysis of Image Characteristics: To delve deeper into the geometric properties of the synthetic images, we employed Spatial Heterogeneous Recurrence Quantification Analysis (SHRQA). This involved initial image pre-processing, including grayscale conversion, noise reduction, contrast enhancement, and normalization, to reduce the unrelated noise and amplify underlying patterns within the images. Subsequently, each image was transformed into an attribute vector through the application of a Hilbert space-filling curve, a technique that preserves the spatial proximity relationships of the image pixels in a one-dimensional vector. This vectorization facilitated a detailed analysis of the geometric recurrence and structural intricacies within the images. By applying the Iterated Function System projection, we were able to identify and quantify recurrent fractal structures, thereby providing a robust profile of the images’ geometric fidelity .
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278131_p15
|
PMC11278131
|
sec[2]/p[5]
|
3. Results
| 4.046875 |
biomedical
|
Study
|
[
0.99951171875,
0.00019824504852294922,
0.0004172325134277344
] |
[
0.99951171875,
0.00023865699768066406,
0.00024437904357910156,
0.000043451786041259766
] |
The SHRQA method was first applied to examine the spatial recurrence properties of real and synthetic image patches across test tissue data, which was prostate in this specific scenario. Our sample set included an equal number of patches from real and synthetic sources, with a balanced representation of each phenotype. To add an extra layer of validation, we downloaded histology images from the cancer genome atlas (TCGA) from individuals representing the adenocarcinoma stage. These images, representing different stages of cancer progression (represented by Gleason grade), were randomized.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278131_p16
|
PMC11278131
|
sec[2]/p[6]
|
3. Results
| 4.171875 |
biomedical
|
Study
|
[
0.99951171875,
0.0002720355987548828,
0.0002396106719970703
] |
[
0.9990234375,
0.00016987323760986328,
0.0004978179931640625,
0.00006651878356933594
] |
On all the image types (normal original (NO), normal synthetic (NS), and cancer original (CO)), segmentation was performed using PyHIST, a Python-based histological tool that processed the images into discrete squares of 256 pixels. Each segment was curated to contain a minimum of 90% tissue content, a criterion set to minimize regional bias and preserve the representativeness of the histological features. We analyzed 2000 image patches, each 256 × 256 pixels, evenly split between real and synthetic. SHRQA quantitatively outlined each patch’s microstructures. From an initial extraction of 112 spatial recurrence features per patch, LASSO selected 102 features that were significant to the Gleason pattern. Hotelling’s T-squared test, a multivariate extension of the two-sample t -test, compared the spatial recurrence attributes of real versus synthetic patches. The resulting p -values of 0.4039 signified no significant differences in spatial recurrence properties between NO and NS, but a significant difference was observed between NO and CO ( p = 1.353 × 10 −7 ) and NS and CO ( p = 1.759 × 10 −7 ), as confirmed by the T-squared tests’ p -values for each Gleason pattern.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11278131_p17
|
PMC11278131
|
sec[2]/p[7]
|
3. Results
| 4.089844 |
biomedical
|
Study
|
[
0.99853515625,
0.0002777576446533203,
0.0013017654418945312
] |
[
0.99951171875,
0.00016379356384277344,
0.0001621246337890625,
0.000034809112548828125
] |
We also employed PCA on the spatial recurrence properties , visualized using radar charts, revealing that the top five principal components capture 90% of the variability. This allowed us to map the distributions of spatial properties for real and synthetic images across phenotypes, as depicted in Figure 3 . Notably, while distributions aligned closely between real and synthetic images (NO–NS), significant differences were evident between NO–CO and NS–CO. These findings across image sections validate the model’s efficiency in capturing the geometric intricacies consistent with real images.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11278131_p18
|
PMC11278131
|
sec[3]/p[0]
|
4. Discussion and Conclusions
| 4.054688 |
biomedical
|
Study
|
[
0.99951171875,
0.0003325939178466797,
0.0003142356872558594
] |
[
0.9990234375,
0.0005578994750976562,
0.00048089027404785156,
0.0000641942024230957
] |
The application of Generative Adversarial Networks (GANs) in producing synthetic medical images, as demonstrated by our research, has significant implications for healthcare. By generating synthetic images that are virtually indistinguishable from real histological samples, GANs provide a powerful tool for training AI systems without the risk of exposing sensitive patient information. This is a key consideration given the notable cybersecurity incidents in recent years, such as the Anthem Inc. and UCLA Health System breaches, which exposed the data of millions. Our study’s success in generating high-quality synthetic genitourinary images serves as a proof of concept for the broader application of GANs in medical imaging. By employing this technology, healthcare providers can enhance the robustness of AI diagnostic tools while maintaining stringent data security. For instance, rather than relying on vast databases of patient images, which pose a potential risk if compromised, medical AI applications can be trained using synthetic datasets that carry no privacy concerns.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p19
|
PMC11278131
|
sec[3]/p[1]
|
4. Discussion and Conclusions
| 3.677734 |
biomedical
|
Study
|
[
0.998046875,
0.0004787445068359375,
0.0012636184692382812
] |
[
0.61474609375,
0.364501953125,
0.019989013671875,
0.0008139610290527344
] |
The practicality of synthetic images generated by GANs is further supported by their performance in standard machine learning metrics and approval by expert pathologists. This dual validation underscores the potential of GANs not only in generating training data but also in providing a buffer against data breaches. As AI continues to permeate the medical field, the ability to create diverse, high-fidelity datasets through GANs becomes increasingly valuable, offering a safeguard against the risks associated with the collection and storage of large-scale patient data.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11278131_p20
|
PMC11278131
|
sec[3]/p[2]
|
4. Discussion and Conclusions
| 4.171875 |
biomedical
|
Study
|
[
0.99951171875,
0.00024199485778808594,
0.00020134449005126953
] |
[
0.99853515625,
0.0002865791320800781,
0.0012884140014648438,
0.00006884336471557617
] |
Looking ahead, the expansion of this methodology to other tissue types and medical conditions could revolutionize the field of medical diagnostics. For example, AI models trained on GAN-generated images could support the early detection of rare diseases without requiring access to potentially sensitive real-world data. Similarly, the generation of synthetic images for rare pathologies could aid in developing diagnostic models where real data are scarce or difficult to obtain due to privacy concerns. However, there are limitations that need to be addressed for the appropriate application of the generated data. For example, to ease the pathology review process, the synthetic images were generated with a large patch size of 5000*5000 pixels. It sorted the purpose, but on the downside, we had to use the training data with a similar patch size of 5000*5000 pixels, which limited the amount of training data. Secondly, we utilized images representing non-diseased conditions, which had more or less a uniform distribution of features and structures compared to cancer images. This limited the GAN model’s ability to generate a vast number of unique synthetic images. Consequently, to perform quantification, we divided the synthetic images into small patch sizes of 256 × 256 pixels before subjecting them to SHQRA models. This allowed us to perform the feature comparison and quantification successfully. To avoid these issues, an increase in the size of training data and starting with a small patch size, which can be localized within the tissue section, will immensely enhance the efficiency of the model while allowing the evaluation by the pathologists. Another limitation is the time StyleGAN takes to generate the synthetic data, which can limit its widespread application. A potential solution to this is to generate image patches of smaller sizes, which may introduce a reduction in the quality of the data but would significantly increase the model’s efficiency. Third, the quantification models utilized in this study may benefit from assisted learning modules, which will allow feature-specific quantification with respect to each tissue type, unlike its current stage.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11278131_p21
|
PMC11278131
|
sec[3]/p[3]
|
4. Discussion and Conclusions
| 3.429688 |
biomedical
|
Other
|
[
0.9921875,
0.00301361083984375,
0.004638671875
] |
[
0.01239013671875,
0.97119140625,
0.0157928466796875,
0.0006976127624511719
] |
In conclusion, the implementation of GANs in digital pathology represents a promising avenue for enhancing both the effectiveness of AI in medical diagnostics and the security of patient data. As healthcare continues to evolve alongside AI, the development of secure, synthetic datasets through GANs will be crucial in mitigating the risks of data breaches while unlocking the potential for more advanced, personalized treatment options.
|
[
"Derek J. Van Booven",
"Cheng-Bang Chen",
"Sheetal Malpani",
"Yasamin Mirzabeigi",
"Maral Mohammadi",
"Yujie Wang",
"Oleksander N. Kryvenko",
"Sanoj Punnen",
"Himanshu Arora"
] |
https://doi.org/10.3390/jpm14070703
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p0
|
39057732
|
sec[0]/p[0]
|
1. Introduction
| 3.980469 |
biomedical
|
Review
|
[
0.998046875,
0.0009593963623046875,
0.0008273124694824219
] |
[
0.166748046875,
0.03021240234375,
0.80224609375,
0.0009632110595703125
] |
Medical image segmentation is a complex yet crucial process within the realm of image analysis. It serves as the foundation for extracting and isolating specific regions of interest. Segmentation is important for conducting detailed quantitative analyses and providing valuable insights into various medical conditions and anomalies. The emergence of deep learning has revolutionized medical image segmentation by automating and refining this intricate process. These techniques, especially convolutional neural networks (CNNs), have shown remarkable capabilities in segmenting medical images with high accuracy and efficiency. Automation saves time and introduces reproducibility to the image analysis pipeline. However, important challenges regarding deep learning segmentation of medical images need to be addressed, including dataset scarcity and difficulty in segmenting 3D structures.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
39057732_p1
|
39057732
|
sec[0]/p[1]
|
1. Introduction
| 3.888672 |
biomedical
|
Study
|
[
0.998046875,
0.0005831718444824219,
0.001468658447265625
] |
[
0.625,
0.3466796875,
0.02777099609375,
0.0006222724914550781
] |
Manual annotations are inherently time-consuming because they require detailed classification of numerous pixels within each image. In medical imaging, this is even more challenging since the segmentation needs to be validated by an experienced professional. It is not feasible to have a clinician spend time curating these masks in clinical settings. Unlike image classification tasks, where annotating each image with a single class label is relatively straightforward, segmentation tasks require meticulous labeling of pixels to accurately outline regions of interest. These factors result in a shortage of annotated datasets for segmentation, which are typically smaller than datasets used for classification tasks.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057732_p2
|
39057732
|
sec[0]/p[2]
|
1. Introduction
| 4.03125 |
biomedical
|
Study
|
[
0.99755859375,
0.0005736351013183594,
0.0019521713256835938
] |
[
0.8564453125,
0.1290283203125,
0.0142669677734375,
0.0004372596740722656
] |
One way to address the challenges of manual annotation and limited datasets is by using weakly supervised self-training methods . These methods use weak annotations, such as bounding boxes, to start the training process. In the context of medical image segmentation, weak annotations can be seen as providing initial guidance by outlining the region of interest within bounding boxes while also recognizing the presence of background pixels. By focusing on the semantic information conveyed by most pixels within the bounding boxes, weakly supervised segmentation techniques effectively guide the training to prioritize relevant features while minimizing the influence of noise or inaccuracies associated with background pixels. The iterative self-training process enables the network to refine its segmentation predictions progressively, gradually improving segmentation accuracy without requiring extensive manual labeling efforts. The adaptive nature of weakly supervised self-training allows the model to learn from its predictions and iteratively enhance segmentation performance.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057732_p3
|
39057732
|
sec[0]/p[3]
|
1. Introduction
| 4.003906 |
biomedical
|
Study
|
[
0.9990234375,
0.0005903244018554688,
0.0005321502685546875
] |
[
0.853515625,
0.08740234375,
0.058380126953125,
0.0006842613220214844
] |
When segmenting volumetric medical images, an important decision involves the processing of the input. One method is to divide the 3D volume into 2D slices and train 2D models for segmentation based on intra-slice information. Another approach is to use the entire 3D volume as input. While 2D models offer faster computation and higher inference speed, they overlook crucial information between adjacent slices, hindering improvements in segmentation accuracy. Additionally, 2D segmentation results can be affected by discontinuities in 3D space, leading to suboptimal segmentation outcomes.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057732_p4
|
39057732
|
sec[0]/p[4]
|
1. Introduction
| 2.658203 |
biomedical
|
Other
|
[
0.59228515625,
0.000904083251953125,
0.406982421875
] |
[
0.30224609375,
0.68896484375,
0.00841522216796875,
0.0005822181701660156
] |
However, 3D CNNs offer a way to understand volumetric spatial information, but they have limitations. Because of the increased dimensionality, 3D CNNs require more significant computational resources and may be more susceptible to overfitting, especially when dealing with limited datasets. Additionally, the slice information that could have been used as multiple instances for model training is now condensed into a single input, exacerbating the challenge of training with limited data.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057732_p5
|
39057732
|
sec[0]/p[5]
|
1. Introduction
| 4.03125 |
biomedical
|
Study
|
[
0.9990234375,
0.0002810955047607422,
0.0005130767822265625
] |
[
0.95654296875,
0.03564453125,
0.007793426513671875,
0.00026345252990722656
] |
To bridge the gap between 2D and 3D CNNs, 2.5D segmentation methods can be utilized. This approach aims to efficiently segment volumetric medical images by creating new architectures or implementing strategies to integrate volumetric information into 2D models. One way this approach combines the advantages of 2D and 3D methodologies is by focusing on a specific slice of a volumetric image while incorporating information from neighboring slices to generate a pseudo-RGB representation. This pseudo-RGB image effectively preserves 3D spatial relationships, enhancing the model’s ability to segment complex 3D structures accurately. By adopting a 2.5D segmentation approach, the segmentation techniques can leverage the computational efficiency of 2D models while capturing crucial spatial contextual information from 3D models.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057732_p6
|
39057732
|
sec[0]/p[6]
|
1. Introduction
| 2.763672 |
biomedical
|
Other
|
[
0.99658203125,
0.0008192062377929688,
0.00273895263671875
] |
[
0.05474853515625,
0.94189453125,
0.00279998779296875,
0.0006833076477050781
] |
Increasing the size of datasets and developing effective strategies for segmenting 3D structures are important for addressing a specific issue: carotid artery segmentation in brain magnetic resonance (MR) images.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057732_p7
|
39057732
|
sec[0]/p[7]
|
1. Introduction
| 4.019531 |
biomedical
|
Other
|
[
0.99853515625,
0.0012035369873046875,
0.0003323554992675781
] |
[
0.34326171875,
0.63671875,
0.0172576904296875,
0.00289154052734375
] |
The carotid arteries are located on each side of the neck and ascend to supply the brain. In axial medical imaging slices, they appear as circular or oval structures positioned laterally to the cervical vertebrae and medially to the sternocleidomastoid muscles. In T1-weighted MR images, the carotid arteries are surrounded by muscles with moderate signal intensity and fat with high signal intensity. This contrast helps distinguish the arteries, which typically have a lower signal intensity than the surrounding fat’s high signal intensity. However, blood flow within the carotid arteries can have variable signal intensity depending on the flow dynamics and the presence of any contrast agent.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057732_p8
|
39057732
|
sec[0]/p[8]
|
1. Introduction
| 4.082031 |
biomedical
|
Review
|
[
0.998046875,
0.0010461807250976562,
0.0006775856018066406
] |
[
0.1339111328125,
0.2080078125,
0.65576171875,
0.0021533966064453125
] |
Computed tomography (CT) and ultrasound are commonly used for and are important for carotid artery studies. However, MR images provide superior soft tissue contrast, enabling detailed visualization of carotid artery walls and plaque composition. It also allows for three-dimensional (3D) imaging, offering comprehensive volumetric analysis and reducing the operator dependency commonly associated with ultrasound. MR imaging can simultaneously image adjacent brain structures, facilitating integrated neurovascular assessments crucial for understanding vascular health’s impact on brain function. Moreover, unlike CT, MR imaging does not involve ionizing radiation, making it a safer option for repeated imaging and use in vulnerable populations .
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p9
|
39057732
|
sec[0]/p[9]
|
1. Introduction
| 3.966797 |
biomedical
|
Study
|
[
0.99951171875,
0.00028514862060546875,
0.0003485679626464844
] |
[
0.5439453125,
0.352783203125,
0.10205078125,
0.001201629638671875
] |
Carotid artery segmentation in brain MR images has several applications, particularly in molecular quantitative imaging. Accurate carotid segmentation allows for the extraction of image-derived input functions for analyzing the biokinetics of positron emission tomography (PET) radiotracers after aligning brain MR images with PET . Additionally, the segmentation of MR images allows for quantitative volumetric analysis of the carotid arteries, enabling detailed assessments of vascular health and potential pathologies .
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057732_p10
|
39057732
|
sec[0]/p[10]
|
1. Introduction
| 3.980469 |
biomedical
|
Study
|
[
0.9990234375,
0.0007262229919433594,
0.00044727325439453125
] |
[
0.74267578125,
0.092041015625,
0.1640625,
0.0012073516845703125
] |
Segmenting carotid arteries in medical imaging is challenging due to several factors, including the small size of arteries, which can vary greatly between patients and under different conditions. It is difficult to create a single segmentation model that fits all cases. Carotid arteries are also located near other important anatomical structures in the head and neck, making it hard for segmentation algorithms to accurately differentiate them from neighboring tissue. Moreover, carotid arteries often have complex branching patterns and curves, making it challenging to track their path through multiple imaging slices and volumes. Therefore, algorithms need to be able to handle intricate and non-linear structures. Even small segmentation errors can have significant clinical implications, highlighting the necessity for highly precise and reliable segmentation methods. All these difficulties are exacerbated when the imaging technique is not optimized for vessel detection, as in non-contrast-enhanced MR images.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057732_p11
|
39057732
|
sec[0]/p[11]
|
1. Introduction
| 4.101563 |
biomedical
|
Study
|
[
0.99951171875,
0.0003514289855957031,
0.00017189979553222656
] |
[
0.9990234375,
0.0002734661102294922,
0.0005993843078613281,
0.0000960230827331543
] |
In this study, we developed a deep learning pipeline that utilizes a 2.5D approach combined with a self-training methodology for segmenting the carotid artery in brain T1-weighted MR images without contrast. The model achieved an Intersection over Union (IoU) score of (0.68 ± 0.08) on an unseen dataset, demonstrating commendable qualitative results. This approach augments the slices of the input instead of employing 2.5D techniques directly within model architectures . We also address the challenge of carotid artery segmentation in brain MR images. Unlike conventional vessel analysis techniques that frequently utilize CT or ultrasound, our approach leverages the soft tissue contrast and three-dimensional imaging capabilities of MR, facilitating integrated neurovascular assessment and offering valuable insights into the interplay between vascular health and brain function.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p12
|
39057732
|
sec[1]/p[0]
|
2. Related Work
| 3.349609 |
biomedical
|
Study
|
[
0.99755859375,
0.00022995471954345703,
0.0022220611572265625
] |
[
0.58984375,
0.390380859375,
0.0191497802734375,
0.0007028579711914062
] |
Methods utilizing convolutional neural networks (CNNs) have shown effectiveness in automated and semi-automated vessel segmentation in MR images .
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p13
|
39057732
|
sec[1]/p[1]
|
2. Related Work
| 4.070313 |
biomedical
|
Study
|
[
0.99951171875,
0.0002846717834472656,
0.0002033710479736328
] |
[
0.9970703125,
0.0002262592315673828,
0.002674102783203125,
0.00008827447891235352
] |
Elsheikh et al. explored the application of CNNs for the automated segmentation of the cerebral vasculature in non-contrast-enhanced black-blood MR imaging (BBMRI) scans. Utilizing a hierarchical, multi-scale 3D CNN model, the researchers achieved a promising Dice similarity coefficient (DSC) of 0.72 on their test dataset. The model employed nested image patches with a U-net-type architecture, allowing for effective segmentation across multiple scales. The study highlighted the advantages of BBMRI over traditional time-of-flight magnetic resonance angiography (TOF-MRA), including reduced flow-related artifacts and better stent-related signal preservation. However, they acknowledged the need for further optimization and expansion of the volume of interest to improve segmentation accuracy, particularly in complex intracranial pathologies.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057732_p14
|
39057732
|
sec[1]/p[2]
|
2. Related Work
| 4.066406 |
biomedical
|
Study
|
[
0.99951171875,
0.0002868175506591797,
0.0001933574676513672
] |
[
0.998046875,
0.0002989768981933594,
0.0014715194702148438,
0.00008684396743774414
] |
Quon et al. developed a deep learning model for real-time segmentation of intracranial vessels in pediatric patients using preoperative T2-weighted MR scans. A modified 2D U-net architecture achieved an overall DSC of 0.75. The model showed higher accuracy for patients with normal vascular anatomy (DSC 0.77) than those with lesions (DSC 0.71). The discrepancy was attributed to vascular deformations caused by tumors. Despite the impressive reduction in segmentation time (from hours to seconds), the small sample size and the model’s lower performance in patients with intracranial lesions were noted as significant limitations.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057732_p15
|
39057732
|
sec[1]/p[3]
|
2. Related Work
| 4.085938 |
biomedical
|
Study
|
[
0.99951171875,
0.0003261566162109375,
0.00015532970428466797
] |
[
0.99755859375,
0.0002830028533935547,
0.002025604248046875,
0.00009930133819580078
] |
Shi et al. developed an automated vessel wall segmentation method using a U-net-like fully convolutional network for quantifying MR vessel wall images in patients with intracranial atherosclerotic disease (ICAD). The method achieved DSC of 0.89 for the lumen and 0.77 for the vessel wall, showing strong agreement with manual segmentation. The study’s clinical application revealed significant differences in the normalized wall index (NWI) between symptomatic and asymptomatic patients, underscoring the clinical relevance of the segmentation method. While the results were promising, they emphasized the need for large-scale quantitative plaque analysis to promote the adoption of MR vessel wall imaging in ICAD management.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057732_p16
|
39057732
|
sec[1]/p[4]
|
2. Related Work
| 4.058594 |
biomedical
|
Study
|
[
0.99951171875,
0.00030231475830078125,
0.00017571449279785156
] |
[
0.99853515625,
0.00025081634521484375,
0.0013418197631835938,
0.00008130073547363281
] |
Samber et al. investigated using CNNs for the automated segmentation of carotid arteries in MR imaging data. Using a dataset of 4422 axial T2-weighted MR images, they trained separate CNNs for segmenting the lumen and vessel wall, achieving DSCs of 0.96 and 0.87, respectively. The CNN-based segmentation showed excellent agreement with expert manual segmentations, evidenced by high Pearson correlation and intraclass correlation coefficients. Despite the need for human supervision to ensure consistency, the study showed the potential for integrating CNN algorithms into software platforms to streamline workflow and reduce the burden on radiologists.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p17
|
39057732
|
sec[1]/p[5]
|
2. Related Work
| 4.050781 |
biomedical
|
Study
|
[
0.9990234375,
0.00017321109771728516,
0.0008983612060546875
] |
[
0.998046875,
0.0005044937133789062,
0.001239776611328125,
0.000050187110900878906
] |
Regarding weakly supervised segmentation and 2.5D approaches, Chen and Hong introduced Scribble2D5, a novel approach that addresses the limitations of existing scribble-based methods by enhancing 3D anisotropic image segmentation. Unlike methods that suffer from poor boundary localization and are primarily designed for 2D segmentation, Scribble2D5 leverages volumetric data. It incorporated a label propagation module and a combination of static and active boundary predictions to improve boundary accuracy and shape regularization of the region of interest. Extensive experiments on public datasets for cardiac, tumor, and abdominal MR images demonstrate that Scribble2D5 significantly outperforms current state-of-the-art scribble-based methods, achieving performance comparable to fully-supervised approaches. However, this method was not tested for segmenting MR vascular imaging.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057732_p18
|
39057732
|
sec[1]/p[6]
|
2. Related Work
| 3.908203 |
biomedical
|
Review
|
[
0.9990234375,
0.0004982948303222656,
0.0007147789001464844
] |
[
0.49755859375,
0.0016984939575195312,
0.50048828125,
0.0003933906555175781
] |
Overall, these studies highlight the potential of CNN-based approaches for vascular segmentation and the use of weakly supervised segmentation in MR images. Automatic segmentation techniques significantly reduce segmentation time, have high accuracy comparable to expert manual segmentations, and are applicable across various vascular conditions and imaging modalities. However, common limitations include the need for larger and more diverse datasets, the variability in performance across different patient subgroups, and the necessity for human supervision in some cases. Future research should address these limitations to enhance automated segmentation techniques’ robustness, generalizability, and clinical applicability in brain vascular MR imaging, especially for sequences that are not optimized for vessel detection.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p19
|
39057732
|
sec[2]/sec[0]/p[0]
|
3.1. Datasets
| 3.511719 |
biomedical
|
Study
|
[
0.9990234375,
0.00021016597747802734,
0.0007686614990234375
] |
[
0.99853515625,
0.0010232925415039062,
0.0003559589385986328,
0.00007724761962890625
] |
We utilized 42 brain T1-weighted MR volumetric scans sourced from four distinct datasets (10 scans from Zareda et al. , 10 scans from Van Schuerbeek, Baeken, and De Mey , 10 scans from Koenders et al. , and 12 scans from OASIS 3 ) to train our model. The first three datasets are defaced; only the last dataset did not go under defacing.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
39057732_p20
|
39057732
|
sec[2]/sec[0]/p[1]
|
3.1. Datasets
| 4.03125 |
biomedical
|
Study
|
[
0.9990234375,
0.00046253204345703125,
0.0004172325134277344
] |
[
0.99853515625,
0.0011529922485351562,
0.00015044212341308594,
0.0001081228256225586
] |
We meticulously delineated bounding boxes around the carotid arteries in each axial slice for each scan in our dataset. We performed an automatic 2.5D image processing for each slice, creating the pseudo-RGB images with the G channel being the target and the other channels being the neighboring surrounding slices (R is the slice below the target and B is the above one). At the end of the bounding box delineation process, we had 1869 pairs of slices and their corresponding masks. Visual representations of 2.5D pseudo-RGB MR slices (on the left) and their bounding boxes (on the right) can be seen in Figure 1 .
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057732_p21
|
39057732
|
sec[2]/sec[0]/p[2]
|
3.1. Datasets
| 4.011719 |
biomedical
|
Study
|
[
0.99951171875,
0.00022912025451660156,
0.00025963783264160156
] |
[
0.9990234375,
0.0008301734924316406,
0.00016987323760986328,
0.0000776052474975586
] |
We created a testing dataset to evaluate the model’s performance against a gold standard. The testing dataset was produced in a multiple sclerosis project. High-resolution structural brain T1-weighted MR images were acquired in a GE Healthcare Signa HDxT equipment of 3.0 T, using BRAVO TM sequence, with a repetition time of 2400 ms, echo time of 16 ms, 220 mm field of view, with 1 mm isotropic voxels. MR images have an array of 240 × 240 × 196 pixels, with 16 bits per pixel.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p22
|
39057732
|
sec[2]/sec[0]/p[3]
|
3.1. Datasets
| 4.097656 |
biomedical
|
Study
|
[
0.9990234375,
0.00066375732421875,
0.0002276897430419922
] |
[
0.99951171875,
0.0002987384796142578,
0.0002186298370361328,
0.0000985264778137207
] |
MR scans corresponding to 35 individuals (age 30 ± 8 years) from the first visit were used to build the gold standard. The carotid arteries were visually identified and manually segmented by an experienced medical physicist. We constructed deformable two-dimensional polygons for all scans containing the left and right carotid slice per slice. An experienced radiologist validated each polygonal region, making corrections and modifications. After reviewing and correcting these polygons, we applied a binary transformation which converts the images into binary masks. We built pairs of images containing the original MR slice and its corresponding segmentation. This process allowed us to obtain 948 original pairs of MR slices and masks. Figure 2 shows examples of pairs of T1-weighted MR slices and corresponding carotid artery masks in the testing dataset.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057732_p23
|
39057732
|
sec[2]/sec[1]/p[0]
|
3.2. Preprocessing and Data Augmentation
| 3.929688 |
biomedical
|
Study
|
[
0.99951171875,
0.0001621246337890625,
0.0004572868347167969
] |
[
0.99755859375,
0.0021114349365234375,
0.00026679039001464844,
0.00007528066635131836
] |
For the image preprocessing steps, we reduced the bit depth from 16 bits to 8 bits per pixel and normalized the pixel values by dividing each pixel by 255. Additionally, we ensured that the voxels in the 3D images were isometric (1 × 1 × 1 mm 3 ). We hypothesize that not adjusting for bias inhomogeneities, spatial localization of brain structures, and parameters from the acquisition and reconstruction processes might aid in generalizing our models.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p24
|
39057732
|
sec[2]/sec[1]/p[1]
|
3.2. Preprocessing and Data Augmentation
| 3.597656 |
biomedical
|
Study
|
[
0.9990234375,
0.0003616809844970703,
0.0008487701416015625
] |
[
0.98193359375,
0.0172882080078125,
0.0004432201385498047,
0.0002651214599609375
] |
The carotid arteries have a distinct shape distribution in the slices: they appear as smaller clusters in slices corresponding to the height where the vessels are classified as C1 or C4. In contrast, they appear as larger, cylindrical-like pixel clusters in slices corresponding to the height where the vessels are classified as C2 or C3 . Figure 3 shows the overlay of bounding boxes to each type of carotid artery shape.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057732_p25
|
39057732
|
sec[2]/sec[1]/p[2]
|
3.2. Preprocessing and Data Augmentation
| 2.369141 |
biomedical
|
Study
|
[
0.99365234375,
0.0011539459228515625,
0.00518798828125
] |
[
0.8544921875,
0.143310546875,
0.001140594482421875,
0.0011768341064453125
] |
Generally, the bounding box areas covering the C2 and C3 regions of the arteries are usually larger, although this does not occur in most slices. Figure 4 shows the distribution of the areas of the bounding boxes in the carotid artery slices.
|
[
"Adriel Silva de Araújo",
"Márcio Sarroglia Pinho",
"Ana Maria Marques da Silva",
"Luis Felipe Fiorentini",
"Jefferson Becker"
] |
https://doi.org/10.3390/jimaging10070161
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.