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
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
PMC11277541_p19
|
PMC11277541
|
sec[2]/p[0]
|
3. Results
| 1.96582 |
biomedical
|
Study
|
[
0.9931640625,
0.003017425537109375,
0.0035762786865234375
] |
[
0.9931640625,
0.005859375,
0.0006127357482910156,
0.000347137451171875
] |
Data about age, parity and geographical origin (Italian/not Italian) of the women enrolled were registered. We report the clinical and demographic characteristics of the sample in Table 1 and the clinical and demographic characteristics of the Italian and non-Italian populations in Table 2 . Data are reported expressing the numerical value and the percentage on the sample, while the age is reported as the mean age expressed in years.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11277541_p20
|
PMC11277541
|
sec[2]/p[1]
|
3. Results
| 3.505859 |
biomedical
|
Study
|
[
0.99853515625,
0.0009784698486328125,
0.00042176246643066406
] |
[
0.99853515625,
0.0012311935424804688,
0.0001901388168334961,
0.00016415119171142578
] |
We collected 7937 non-consecutive cases. The age distribution of the sample examined varies from a minimum of 14 to a maximum of 56 years old, with a mean age of 33.8 years old . Anti-rubella IgG antibodies were found in 7224 (91%) women while 713 (9%) were susceptible to rubella (IgG negative). No IgM-positive cases were documented.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11277541_p21
|
PMC11277541
|
sec[2]/p[2]
|
3. Results
| 3.404297 |
biomedical
|
Study
|
[
0.998046875,
0.00043320655822753906,
0.0014743804931640625
] |
[
0.9990234375,
0.0005412101745605469,
0.00018584728240966797,
0.0000514984130859375
] |
The age analysis showed a statistically significant older age of immune women than receptive women (mean age 33.96 ± 5.6 versus 32.69 ± 5.4 years old, p < 0.05, Table 3 ).
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11277541_p22
|
PMC11277541
|
sec[2]/p[3]
|
3. Results
| 3.736328 |
biomedical
|
Study
|
[
0.9990234375,
0.000331878662109375,
0.0007066726684570312
] |
[
0.99951171875,
0.00031566619873046875,
0.00017392635345458984,
0.000044405460357666016
] |
A subgroup analysis for the 7224 immune women was performed and showed a statistically significant older age of Italian immune women than non-Italian immune women (mean age 34.36 ± 5.4 versus 32.08 ± 5.9 years old, p < 0.05, Table 4 ).
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11277541_p23
|
PMC11277541
|
sec[2]/p[4]
|
3. Results
| 3.972656 |
biomedical
|
Study
|
[
0.9990234375,
0.0005040168762207031,
0.0005407333374023438
] |
[
0.99951171875,
0.0003345012664794922,
0.0002570152282714844,
0.00005143880844116211
] |
Considering rubella immune women, 3822 (52.9%) were nulliparous, 2673 (37%) had 1 child and 729 (10.1%) had at least 2 children; the mean age was 34 years old (14–56). Considering women who were found to be receptive to rubella, 420 (58.9%) were nulliparous, 216 (30.3%) had 1 child and 77 (10.8%) had at least 2 children; the mean age was 32.7 years old (17–46). Considering nationality, 6512 (82%) were Italian and 1425 (18%) were non-Italian. Among Italian population, 5938 (91.2%) were immune and 574 (8.8%) were susceptible to rubella, while among the non-Italian population, 1286 (90.2%) were immune and 139 (9.8%) were found to be susceptible to rubella. The mean age of the Italian immune population was 34.36 years old (14–56); 3325 (56%) were nulliparous, 2169 (36.5%) had 1 child and 444 (7.5%) had at least 2 children. The mean age of the Italian receptive population was 33.2 years old (20–46); 364 (63.4%) were nulliparous, 165 (28.8%) had 1 child and 45 (7.8%) had at least 2 children. The mean age of the non-Italian immune population was 32.08 years old (15–55); 487 (37.9%) were nulliparous, 508 (39.5%) had 1 child and 291 (22.6%) had at least 2 children. The mean age of the non-Italian receptive population was 30.5 years old (17–38); 56 (40.3%) were nulliparous, 51 (36.7%) had 1 child and 32 (23%) had at least 2 children.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11277541_p24
|
PMC11277541
|
sec[3]/p[0]
|
4. Discussion
| 4.054688 |
biomedical
|
Study
|
[
0.99951171875,
0.00021183490753173828,
0.0002429485321044922
] |
[
0.9990234375,
0.00019216537475585938,
0.0005354881286621094,
0.0000527501106262207
] |
The data observed in our examined sample showed a percentage of pregnant women with immunity in line with national data for the general population, indicating immunity rates of 91.2% in Italian women and 90.2% in non-Italian women. No differences in terms of immunity rate were observed between Italian and non-Italian women, in contrast with previous data . Approximately 9% of pregnant women in both groups were therefore susceptible to rubella infection during pregnancy, these data are in line with other previously reported in the literature. In the metanalysis of Pandolfi et al. a pooled estimate of seronegativity of 9.4% for pregnant women was underlined .
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11277541_p25
|
PMC11277541
|
sec[3]/p[1]
|
4. Discussion
| 2.539063 |
biomedical
|
Other
|
[
0.984375,
0.0020389556884765625,
0.013580322265625
] |
[
0.060302734375,
0.93603515625,
0.0033416748046875,
0.0005741119384765625
] |
The objective of the National Plan for the Elimination of Measles and Congenital Rubella for the period 2010–2015 was the achievement of a percentage of women of childbearing age susceptible to infection that was below 5% . Our data indicate that additional efforts are needed to reach this goal. Until the disease is eliminated in all countries of the world, it will be necessary to maintain high vaccination coverage, and further strengthen the surveillance and investigation of reported cases, ensuring a rapid response to any imported cases.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11277541_p26
|
PMC11277541
|
sec[3]/p[2]
|
4. Discussion
| 4.035156 |
biomedical
|
Study
|
[
0.998046875,
0.0016222000122070312,
0.0005135536193847656
] |
[
0.51708984375,
0.020477294921875,
0.4609375,
0.0014057159423828125
] |
The new guidelines for physiological pregnancy published in Italy in December 2023 no longer recommend screening for this infection during pregnancy, as it has been declared eliminated . The elimination of the endemic transmission of the rubella virus is described as a major public health success, and the result of tenacious work to achieve high vaccination coverage in the population . However, this claim could lead to a risky underestimation of the problems associated with susceptibility to rubella and the implications related to infection in pregnancy. The collected information from the Italian Behavioral Risk Factor Surveillance System (PASSI), in the period 2017–2020, highlights a significant lack of awareness among women of childbearing age regarding the risks associated with rubella infection during pregnancy. Indeed, a rather high number, nearly 4 out of 10 women (37%), is unaware of their immune status regarding rubella . Pregnancy represents one of the few moments in which women undergo serologic screening, and very often the importance of a preconception visit, aimed at identifying correctable risk factors, before beginning to seek pregnancy is underestimated. Not infrequently, the patient presents to her gynecologist while already pregnant having disregarded the possibility of performing proper counseling and remediating any susceptibility to infection.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11277541_p27
|
PMC11277541
|
sec[3]/p[3]
|
4. Discussion
| 3.931641 |
biomedical
|
Study
|
[
0.9990234375,
0.00018358230590820312,
0.0008134841918945312
] |
[
0.9814453125,
0.0090484619140625,
0.0092926025390625,
0.00013756752014160156
] |
It is important to note that the percentage of Italian women susceptible to infection tends to decrease based on parity, probably thanks to a significant strategy to catch up with post-partum women; these data are confirmed by the evidence that women immune to rubella are older than women receptive to rubella. This phenomenon occurs to a lesser extent in non-Italian women, in which 59% of the non-immune group have at least 1 child. These data suggest that compliance to the post-partum vaccination recommendations among these women should have a positive impact. In fact, although Italian women are older than non-Italian women when seropositive for rubella, these data could reflect the demographical characteristic of the pregnant Italian population, which tends to have a first pregnancy at an older age than non-Italian women .
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11277541_p28
|
PMC11277541
|
sec[3]/p[4]
|
4. Discussion
| 2.060547 |
biomedical
|
Other
|
[
0.982421875,
0.0011663436889648438,
0.0164337158203125
] |
[
0.338623046875,
0.6591796875,
0.0012216567993164062,
0.00107574462890625
] |
A proportion of 63.4% of Italian pregnant women susceptible to rubella in our sample are nulliparous. According to data from the Institute of National Statistics, the fertility rate per woman in Italy is 1.24 , so it is plausible that many of these women may not have a second pregnancy.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11277541_p29
|
PMC11277541
|
sec[3]/p[5]
|
4. Discussion
| 3.964844 |
biomedical
|
Study
|
[
0.9990234375,
0.00038886070251464844,
0.0005917549133300781
] |
[
0.87255859375,
0.006450653076171875,
0.12042236328125,
0.0003123283386230469
] |
Both the measles, mumps, and rubella (MMR) and varicella vaccines contain live viruses and are therefore not recommended during pregnancy. As a result, it is important to plan for these vaccinations ahead of time, through preconception counseling if possible. In a study conducted in the United States in 2021, Foley et al. showed how women who did not want to perform rubella vaccination in the preconception phase, despite recommendations, asserted as their motivation the desire not to delay the search for pregnancy . Indeed, opinions from the American Society for Reproductive Medicine and the American College of Obstetricians and Gynecologists uphold that pregnancy is contraindicated for four weeks after vaccination with a live attenuated virus . Shifting the maternal age in the search for first pregnancy and the idea that delaying this search may affect the chances of becoming pregnant may be limiting factors in reaching a threshold of susceptible women that is less than 5%.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11277541_p30
|
PMC11277541
|
sec[3]/p[6]
|
4. Discussion
| 2.304688 |
biomedical
|
Other
|
[
0.98681640625,
0.005420684814453125,
0.00771331787109375
] |
[
0.018280029296875,
0.97607421875,
0.004596710205078125,
0.0009150505065917969
] |
It is crucial, therefore, to focus efforts on patients of this population before they experience their first pregnancy. Increasing awareness among healthcare professionals who encounter women in the preconception phase, such as gynecologists or general practitioners, is very important. Rubella testing is offered free of charge to women seeking to conceive, but, evidently, it is still not prescribed frequently enough. This aligns with the fact that the vaccination rate among women of childbearing age is relatively low in Italy, with an average of 44.9% .
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11277541_p31
|
PMC11277541
|
sec[3]/p[7]
|
4. Discussion
| 2.148438 |
biomedical
|
Study
|
[
0.99658203125,
0.0008983612060546875,
0.0024662017822265625
] |
[
0.9306640625,
0.0667724609375,
0.001461029052734375,
0.0009379386901855469
] |
In this study the rate of vaccinated women is lacking, so it is unknown if the current rubella immunity of the subjects was derived from vaccination or from infections that occurred when rubella was still endemic.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11277541_p32
|
PMC11277541
|
sec[3]/p[8]
|
4. Discussion
| 1.366211 |
biomedical
|
Other
|
[
0.56640625,
0.005268096923828125,
0.428466796875
] |
[
0.00942230224609375,
0.9873046875,
0.0024166107177734375,
0.0007433891296386719
] |
The mandatory vaccination established in 2017 may help future generations but the outcomes of this decision are not yet observable.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11277541_p33
|
PMC11277541
|
sec[3]/p[9]
|
4. Discussion
| 4.0625 |
biomedical
|
Study
|
[
0.99951171875,
0.0002372264862060547,
0.00039267539978027344
] |
[
0.99951171875,
0.00024962425231933594,
0.0003459453582763672,
0.000041961669921875
] |
Similar strategies to that initiated in Italy in 2017 have been implemented in other parts of the world, yielding significant results that have bought the percentage of women of childbearing age that is susceptible to infection below 5%. In a Korean study conducted from 2004 to 2018, the IgG serum of 72,114 women was analyzed. Women born between 1977 and 1993 benefited from a universal vaccination program initially using only the RCV vaccine and then, later, the MMR vaccine. The overall proportion of seronegative women decreased significantly, from 6.1% in 2004 to 2.5% in 2018. The rate of seronegativity was highest among women who were not targeted for national immunization , while it was lowest among candidates receiving routine and catch-up vaccinations .
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11277541_p34
|
PMC11277541
|
sec[4]/p[0]
|
5. Conclusions
| 2.037109 |
biomedical
|
Other
|
[
0.9853515625,
0.00920867919921875,
0.005588531494140625
] |
[
0.0019016265869140625,
0.99609375,
0.0010662078857421875,
0.0011358261108398438
] |
Though rubella endemic infection has been declared eliminated in Europe, it is essential for women of childbearing age to be fully aware of their immune status before embarking on pregnancy and, in case of susceptibility, to take appropriate measures.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11277541_p35
|
PMC11277541
|
sec[4]/p[1]
|
5. Conclusions
| 2.806641 |
biomedical
|
Other
|
[
0.98046875,
0.013946533203125,
0.005413055419921875
] |
[
0.0015382766723632812,
0.99267578125,
0.0048675537109375,
0.0010499954223632812
] |
The MMR vaccine must be still recommended to all non-immune women and offered to them during preconception and in the post-partum period, as an important element for the prevention of communicable diseases. Vaccination hesitancy can be minimized through a consistent recommendation to all women, one that is offered by medical staff during routine care. Vaccination campaigns and the education of couples about the risks of CRS remain crucial.
|
[
"Anna Franca Cavaliere",
"Marco Parasiliti",
"Rita Franco",
"Vitalba Gallitelli",
"Federica Perelli",
"Amelia Spanò",
"Barbara Pallone",
"Maria Grazia Serafini",
"Fabrizio Signore",
"Georgios Eleftheriou",
"Giovanni Scambia",
"Antonio Lanzone",
"Annalisa Vidiri"
] |
https://doi.org/10.3390/ijerph21070957
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057609_p0
|
39057609
|
sec[0]/p[0]
|
1. Introduction
| 3.933594 |
biomedical
|
Review
|
[
0.9970703125,
0.0013456344604492188,
0.001800537109375
] |
[
0.0266571044921875,
0.01215362548828125,
0.9609375,
0.000461578369140625
] |
In recent decades, pharmaceutical research and development (R&D) has been highly successful in the treatment of cardiovascular diseases (CVDs). As a consequence, significant therapeutic experience has been accumulated with cardiovascular (CV) medicines, accompanied by well-known safety profiles and affordable price tags. In comparison with other disease areas (e.g., oncology), the unmet need for general CV patient populations is relatively small . Due to off-patent standard therapies and regulators’ expectations for large-scale clinical trials with over ten thousand patients, the return on investment for de novo revolutionary innovation in CVDs has become limited , and so innovators have been focusing only on narrow patient populations in niche areas, such as amyloid cardiomyopathy .
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057609_p1
|
39057609
|
sec[0]/p[1]
|
1. Introduction
| 4.414063 |
biomedical
|
Review
|
[
0.9912109375,
0.005451202392578125,
0.0031871795654296875
] |
[
0.018157958984375,
0.0024566650390625,
0.978515625,
0.0006556510925292969
] |
Despite the wide treatment armamentarium, CVDs still represent the largest global disease burden for adults . CVD is the leading cause of death globally , and due to the rising prevalence of CVDs, the health burden will grow in the future. Many CVD risk factors remain poorly controlled. The silent course of the CVD, complexity of treatment sequencing, increasing rates of comorbidities (e.g., diabetes, obesity, and autoimmune diseases), and suboptimal adherence are important factors limiting the translation of clinical trial efficacy into real-world effectiveness at the population level. In the absence of de novo innovation for the general CV population, there are two potential ways forward: First, the synergistic effect of co-administering medicines with different mechanisms of action has improved therapeutic outcomes, e.g., in hypertension, the combination of diuretics, calcium channel blockers, beta-blockers, and renin–angiotensin–aldosterone system blockers has been proved to be effective in reducing major adverse cardiac events (MACEs) . The second way forward is the incremental or evolutionary innovation of existing medicines to create value-added or repurposed medicines. There are three drug repurposing models: (i) drug repositioning refers to finding new indications or applying a medicine in a new patient population; (ii) drug reformulation is a process in which the pharmaceutical formulation of a product is modified to gain new attributes; finally, (iii) the combination of established medicines into a single pill. Fixed dose combinations (FDCs) of monocomponents in the same indication, or polypills of medicines from different indications, have the potential to deliver additional value to patients, healthcare professionals, and healthcare payers . The selection of active product ingredients (APIs) for combination should be based on the following considerations: (1) the APIs in the combination should have different and complementary mechanisms of action, (2) their pharmacokinetics should not be widely different, and (3) they should not have supra-additive toxicity compared to monotherapy . For example, better blood pressure control and a reduction in low-density lipoprotein concentration have been demonstrated with the use of FDCs in secondary prevention of CVDs . Importantly, the WHO has recently incorporated FDCs in its Essential Medicines List . Due to the high global prevalence of CVDs, incremental innovation of marketed medicines has more potential to reduce the global burden of CVDs than de novo innovation in niche CV patient populations.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p2
|
39057609
|
sec[0]/p[2]
|
1. Introduction
| 3.875 |
biomedical
|
Review
|
[
0.99853515625,
0.0012063980102539062,
0.0004742145538330078
] |
[
0.14208984375,
0.0219879150390625,
0.8349609375,
0.0008783340454101562
] |
Despite the availability of numerous effective and well-tolerated pharmacotherapies in CVDs, their potential benefits are often not fully realized due to medication non-adherence . Moreover, the necessity to take multiple medications for CVD combination therapies may negatively impact adherence. Approximately half of the patients on chronic pharmaceutical therapy fail to follow their long-term medication regimens , which is amplified by the rising prevalence of multimorbidity within aging populations . As a potential solution, simplified treatment regimens and reduced pill burden through FDCs can improve medication adherence among CVD patients .
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057609_p3
|
39057609
|
sec[0]/p[3]
|
1. Introduction
| 4.121094 |
biomedical
|
Study
|
[
0.9990234375,
0.0007433891296386719,
0.0004696846008300781
] |
[
0.759765625,
0.0010194778442382812,
0.239013671875,
0.00042510032653808594
] |
The impact of non-adherence is especially critical for those patients for whom limited disease control is associated with negative clinical events and increased mortality . A recent meta-analysis––which included observational (cohort or case–control) studies with risk estimates for cardiovascular events, strokes, or all-cause mortality related to lipid-lowering, antihypertensive, antidiabetics, and antithrombotic agents––highlighted that a 20% increase in adherence to CV medication is associated with a 9% reduction in the risk of CV events, a 16% reduction in the risk of stroke, and a 10% reduction in the risk of all-cause mortality . Conversely, in Europe, 9% of all CV events can be associated with poor medication adherence . Real-world studies on CVDs have demonstrated that FDCs, by improving adherence, are associated with a significant reduction in all-cause mortality (RR: 0.90; 95% CI: 0.81–1.00) and a non-significant reduction in the risk of major cardiac events (MACEs) (RR: 0.85, 95% CI: 0.70–1.02) . Obviously, these indicators may vary for CV diseases (such as coronary heart diseases, chronic heart failure, or stroke) and subpopulations with different baseline risks.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p4
|
39057609
|
sec[0]/p[4]
|
1. Introduction
| 4.019531 |
biomedical
|
Study
|
[
0.9990234375,
0.0006508827209472656,
0.0003979206085205078
] |
[
0.94775390625,
0.0027256011962890625,
0.049224853515625,
0.0002810955047607422
] |
Medication non-adherence in CVDs also leads to substantial healthcare costs due to increased utilization of preventable healthcare services, such as medication use, emergency department visits, outpatient visits, and hospital admissions . Such cost increases become particularly critical when considering that expenditures related to CVD represent up to 7.6% of total healthcare spending in the European Union, partly driven by medication non-adherence . This suggests that, among other potential adherence interventions, FDCs offer a promising strategy to mitigate the healthcare burden associated with CVDs . The annual savings associated with changing combination antihypertensive therapy to FDCs in Poland was estimated to be EUR 12.3 million from the public payer’s perspective and an additional EUR 5.0 million from the patients’ perspective. Apart from the immediate savings in pharmaceutical costs, the broadened use of FDCs was also expected to improve the patients’ adherence and non-drug costs (e.g., avoided complications) .
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p5
|
39057609
|
sec[0]/p[5]
|
1. Introduction
| 2.875 |
biomedical
|
Other
|
[
0.994140625,
0.001163482666015625,
0.00478363037109375
] |
[
0.358154296875,
0.61865234375,
0.0217742919921875,
0.0013246536254882812
] |
Finally, the applicability of lower doses due to the improved therapeutic effect of FDCs may be translated to potentially lower collateral effects and reduced utilization and cost of the API.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057609_p6
|
39057609
|
sec[0]/p[6]
|
1. Introduction
| 3.371094 |
biomedical
|
Other
|
[
0.9921875,
0.006320953369140625,
0.0017299652099609375
] |
[
0.0300445556640625,
0.88818359375,
0.08026123046875,
0.0015554428100585938
] |
Potential disadvantages of FDCs include limited flexibility in dosing (both during the titration period and during the maintenance treatment period), hindering personalized medicines, and, compared to administering only one monocomponent, potentially increased chances of adverse drug effects and interactions .
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057609_p7
|
39057609
|
sec[0]/p[7]
|
1. Introduction
| 2.394531 |
biomedical
|
Other
|
[
0.97705078125,
0.004261016845703125,
0.0185089111328125
] |
[
0.03564453125,
0.6845703125,
0.278076171875,
0.0018024444580078125
] |
Several publications describe how medication adherence at the patient level can be facilitated by healthcare professionals (e.g., physicians, pharmacists, and nurses) with the support of clinical societies and patient advocacy groups . However, there is limited information about how policymakers and healthcare payers at the health system level can advocate incremental innovation of health technologies—such as FDCs or digital health solutions—with the potential to improve therapeutic adherence.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057609_p8
|
39057609
|
sec[0]/p[8]
|
1. Introduction
| 3.003906 |
biomedical
|
Study
|
[
0.95068359375,
0.0018930435180664062,
0.047454833984375
] |
[
0.68408203125,
0.07940673828125,
0.2353515625,
0.0011014938354492188
] |
This perspective paper aims to describe the barriers preventing the increased utilization of FDCs in CVDs and, based on the identified barriers, to provide system-level policy solutions and recommendations from a multi-stakeholder perspective.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p9
|
39057609
|
sec[1]/p[0]
|
2. System Level Barriers to Advocate FDCs
| 2.148438 |
biomedical
|
Other
|
[
0.9306640625,
0.00579833984375,
0.06365966796875
] |
[
0.024993896484375,
0.96826171875,
0.00638580322265625,
0.0005426406860351562
] |
In order to explore system-level barriers towards the development and use of FDCs to improve medication adherence, it is important to understand what differentiates incremental innovation of widely used medicines from other types of pharmaceutical innovations from the perspective of regulators and healthcare payers.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057609_p10
|
39057609
|
sec[1]/p[1]
|
2. System Level Barriers to Advocate FDCs
| 3.46875 |
biomedical
|
Other
|
[
0.9775390625,
0.003261566162109375,
0.019012451171875
] |
[
0.102294921875,
0.54345703125,
0.353271484375,
0.0008783340454101562
] |
As described in Table 1 , the patentability of FDCs is fairly limited compared to other types of pharmaceutical R&D. HTA bodies and healthcare payers treat FDCs similarly to generic or biosimilar medicines. They do not expect health gain compared with the combination of monocomponents, which has to be rewarded with a price premium. In theory, healthcare payers are not against paying a fair price premium for a reduction in mortality or an improvement in quality of life if manufacturers are able to prove these benefits at product launch from randomized clinical trials (RCTs). The current approach of macro-level decisions is associated with several barriers to the extended development and use of FDCs in CVDs.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057609_p11
|
39057609
|
sec[1]/sec[0]/p[0]
|
2.1. Barrier #1: Hurdles in Evidence Generation for Innovators of FDCs
| 3.173828 |
biomedical
|
Other
|
[
0.97265625,
0.00385284423828125,
0.023529052734375
] |
[
0.031585693359375,
0.93798828125,
0.0297088623046875,
0.0005116462707519531
] |
When comparing FDCs to combination therapies in separate pills, it is difficult to separate the improved therapeutic effect coming (1) from the synergistic effect of co-administering single medicines with different mechanisms of action, or (2) from the benefit of combining established medicines into a single pill. Investment in evidence generation related to the added value of the single pill formulation is a dilemma for the first movers due to the limited patentability, as once the evidence is in the public domain, other manufacturers will have the opportunity to free-ride on the first mover’s investment and develop their own FDCs with the same components.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057609_p12
|
39057609
|
sec[1]/sec[0]/p[1]
|
2.1. Barrier #1: Hurdles in Evidence Generation for Innovators of FDCs
| 3.998047 |
biomedical
|
Study
|
[
0.9990234375,
0.0006136894226074219,
0.0005846023559570312
] |
[
0.638671875,
0.01348876953125,
0.34716796875,
0.0005397796630859375
] |
As a consequence of the limited patentability, pharmaceutical companies have to limit the R&D budget to test the benefits of FDCs of off-patent medicines in prospective randomized controlled trials (RCTs). Even if they are willing to invest a development budget comparable to de novo investigational medicines, it is not possible to detect adherence improvement of FDCs in blinded RCTs. Therefore, clinical differentiation of FDCs in RCTs has to be based on hard clinical endpoints, which necessitates the inclusion of thousands of patients with long-term follow-up. Improved medication adherence (and related improvement in hard clinical endpoints) can potentially be measured in observational studies. Recently, there has been increasing evidence that biochemical urine testing is a promising tool for measuring adherence in multiple chronic diseases, including diabetes and mental disorders . However, the generation of real-world evidence is possible only after the wide-scale use of FDCs.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057609_p13
|
39057609
|
sec[1]/sec[0]/p[2]
|
2.1. Barrier #1: Hurdles in Evidence Generation for Innovators of FDCs
| 2.480469 |
biomedical
|
Other
|
[
0.97900390625,
0.003917694091796875,
0.016845703125
] |
[
0.0731201171875,
0.8310546875,
0.09423828125,
0.0014896392822265625
] |
In conclusion, the benefits of FDCs can be proven only in highly expensive, large-scale RCTs and real-world evidence is available only after the reimbursement of FDCs by healthcare payers.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057609_p14
|
39057609
|
sec[1]/sec[1]/p[0]
|
2.2. Barrier #2: Limited Acceptance of Adherence as an Endpoint by Clinical Guideline Developers and Policymakers
| 3.853516 |
biomedical
|
Review
|
[
0.96337890625,
0.0308074951171875,
0.006011962890625
] |
[
0.003025054931640625,
0.1790771484375,
0.814453125,
0.003467559814453125
] |
Certain clinical guidelines already recognize the importance of FDCs in reducing pill burden and improving adherence in the pharmacotherapy of CVDs. Specifically, the ESH (European Society of Hypertension) guidelines emphasize treatment initiation with FDCs rather than the administration of separate monocomponents for most patients with hypertension. Similarly, the ACC/AHA (American College of Cardiology/American Heart Association) guidelines recommend the same approach for patients with stage 2 hypertension, those with blood pressure more than 20/10 mm Hg above their target blood pressure, and Black patients.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057609_p15
|
39057609
|
sec[1]/sec[1]/p[1]
|
2.2. Barrier #2: Limited Acceptance of Adherence as an Endpoint by Clinical Guideline Developers and Policymakers
| 4.078125 |
biomedical
|
Review
|
[
0.9921875,
0.005435943603515625,
0.002346038818359375
] |
[
0.0178070068359375,
0.0025463104248046875,
0.97900390625,
0.0005145072937011719
] |
Nevertheless, the integration of FDCs into CV clinical guidelines is still in its early stages and not yet comprehensive. Despite the significant potential of FDCs in enhancing medication adherence, they face four main challenges in gaining extensive recognition as a valid strategy for clinical improvement within clinical guidelines. First, the development of evidence-based guidelines heavily focuses on hard clinical endpoints, such as mortality and MACEs, thus typically not considering the impact of FDCs on adherence as a valid clinical endpoint. Second, the assertion that FDCs improve medication adherence lacks robust support from RCTs due to the aforementioned reasons, hindering the acceptance of adherence benefits as a legitimate clinical improvement . Third, the lack of regulatory recommendations on standardized definitions, measurement methods, and reporting protocols of medication adherence, both in RCTs and in real-world data analysis, complicates the assessment of adherence interventions, including FDCs . Finally, various other factors beyond the formulation of medications can significantly influence medication adherence . These factors underscore the complexity of enhancing adherence and suggest that a multifaceted approach is necessary to comprehensively address this issue.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057609_p16
|
39057609
|
sec[1]/sec[2]/p[0]
|
2.3. Barrier #3: Limited Options for Price Premium for Incremental Innovation by Healthcare Payers
| 3.236328 |
biomedical
|
Other
|
[
0.568359375,
0.01540374755859375,
0.416259765625
] |
[
0.0172119140625,
0.970703125,
0.01169586181640625,
0.00039577484130859375
] |
In the financing protocol for treatments of CVDs, the first-line therapies are off-patent medicines. After the patent expiry, the primary expectation of healthcare payers is the generation of savings from generic price erosion, and they devote limited attention to health outcomes. Payers routinely apply internal reference pricing (IRP) for medicines in the retail sector or single-criterion tenders with a ‘lowest-price-wins’ decision rule in the hospital sector . Pharmaceutical price regulators, healthcare payers, and procurement bodies set up similarly conservative pricing rules for FDCs, which permit discount prices for FDCs compared with the aggregated price level of monocomponents; in other words, they expect that ‘1 + 1 < 2’. Some payers even maximize the price of FDCs according to the higher-priced monocomponent (’1 + 1 = 1’ or even ‘1 + 1 + 1 = 1’) and hope that manufacturers of FDCs accept such an unfair pricing rule for a potentially higher market share compared with monocomponents.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p17
|
39057609
|
sec[1]/sec[2]/p[1]
|
2.3. Barrier #3: Limited Options for Price Premium for Incremental Innovation by Healthcare Payers
| 3.955078 |
biomedical
|
Study
|
[
0.966796875,
0.00566864013671875,
0.0276336669921875
] |
[
0.97314453125,
0.019622802734375,
0.00719451904296875,
0.0002779960632324219
] |
Exemptions from IRP or a modest price premium are more realistic in areas with high unmet needs, such as oncology or pediatric indications. Based on a series of structured interviews with health policy experts from Ireland, the price potential of FDCs is still not sufficient to incentivize investment in R&D . A good example of that is a polypill called TRINOMIA (containing acetylsalicylic acid, atorvastatin, and ramipril) which is used for the secondary prevention of cardiovascular accidents. Our interviewees highlighted that, despite the substantial added therapeutic value of the polypill, they expected the medicines not to receive a price premium . Based on the reimbursement price of TRINOMIA in Ireland, it was priced slightly higher than the most expensive monocomponent, atorvastatin (EUR 13.11 and EUR 12, respectively), for a 28-day treatment period . Additionally, based on the Irish Pharmaceutical Healthcare Association’s 2021–2025 Framework Agreement, a yearly price reduction resulted in an even lower price for TRINOMIA (EUR 12.83) as of the 1st of March 2022 . The negative incentivizing effect of such pricing policies is well demonstrated by the fact that the pivotal trial for TRINOMIA (SECURE trial) was funded by the EU . With more adaptive pricing policies and a better price potential, a local pharmaceutical company might have been incentivized to run a trial for the FDC and supply the medicine in Ireland. Our interviews revealed that a similar pricing structure is true for Poland, where the price is usually set to the price level of the highest-priced monocomponent (‘1 + 1 = 1’) . Our experts highlighted that even after the expected future revision of the Polish legislation, the full price may not reach the sum of the two monocomponents’ prices (i.e., ‘1 + 1 < 2’), and they see that this discourages investment in FDCs. Therefore, price regulators, healthcare payers, and procurement bodies do not financially incentivize the added value of improved therapeutic adherence generated by incremental innovation.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057609_p18
|
39057609
|
sec[1]/sec[2]/p[2]
|
2.3. Barrier #3: Limited Options for Price Premium for Incremental Innovation by Healthcare Payers
| 3.320313 |
biomedical
|
Other
|
[
0.818359375,
0.01557159423828125,
0.166015625
] |
[
0.059967041015625,
0.89990234375,
0.0396728515625,
0.000682830810546875
] |
Another potential reason why payers are reluctant to pay a price premium for FDCs (with their main benefit being improved adherence) is the multifactorial nature of the issue of non-adherence. As often argued, in addition to therapy-related factors, non-adherence is also influenced by social and economic, disease-related, patient-related, and healthcare-system-related factors. Hence, according to the argument of some payers, isolating and attributing improved adherence solely to the pharmaceutical product itself may be misleading. As an example, one might argue that promotional activities of sales representatives of pharmaceutical companies may have an impact on medication adherence, therefore, payers, similarly to developers of clinical guidelines, also prefer the hard clinical endpoints over medication adherence.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057609_p19
|
39057609
|
sec[1]/sec[3]/p[0]
|
2.4. Barrier #4: Limited Availability of Real-World Evidence
| 2.697266 |
biomedical
|
Study
|
[
0.99267578125,
0.0020599365234375,
0.005092620849609375
] |
[
0.69140625,
0.30029296875,
0.007678985595703125,
0.0009350776672363281
] |
Pathways of CVD patients and their long-term outcomes (i.e., mortality and MACEs) can potentially be followed in payer’s databases in countries with a single healthcare payer. However, the accessibility to individualized patient records is often limited to researchers, even in an anonymized format, and missing or poor-quality data also limits the potential of capturing the benefits of FDCs in payer’s databases.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057609_p20
|
39057609
|
sec[1]/sec[3]/p[1]
|
2.4. Barrier #4: Limited Availability of Real-World Evidence
| 1.608398 |
other
|
Other
|
[
0.2159423828125,
0.04461669921875,
0.7392578125
] |
[
0.004150390625,
0.994140625,
0.0006923675537109375,
0.0008287429809570312
] |
In countries with multiple payers, where healthcare financing is fragmented, the patient pathways may not be reliably traceable.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p21
|
39057609
|
sec[1]/sec[4]/p[0]
|
2.5. Barrier #5: Methodological Issues to Measure Improved Adherence
| 3.960938 |
biomedical
|
Study
|
[
0.9990234375,
0.0007491111755371094,
0.0004661083221435547
] |
[
0.97607421875,
0.01837158203125,
0.0052642822265625,
0.0002391338348388672
] |
The lack of consensus on how to calculate medication adherence may also complicate the analysis of prescription-refill databases. In the case of polypharmacy, complexity arises from the need to integrate the dosing and timing schedules of each drug as their effects merge into a single treatment regimen . This may pose a challenge for retrospective database analyses attempting to assess real-world adherence improvements with FDCs compared with the combination of monocomponents.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p22
|
39057609
|
sec[2]/p[0]
|
3. Recommended Solutions
| 1.831055 |
biomedical
|
Other
|
[
0.9306640625,
0.01528167724609375,
0.053802490234375
] |
[
0.0018472671508789062,
0.99658203125,
0.0010204315185546875,
0.0004057884216308594
] |
While the extended use of FDCs at the level of individual CV patients has to be supported by healthcare professionals, it is of utmost importance to change the entire healthcare ecosystem with the involvement of different stakeholders at the macro level.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p23
|
39057609
|
sec[2]/p[1]
|
3. Recommended Solutions
| 1.78125 |
biomedical
|
Other
|
[
0.82275390625,
0.01433563232421875,
0.1630859375
] |
[
0.005634307861328125,
0.9873046875,
0.006328582763671875,
0.0006337165832519531
] |
Dissemination of evidence about the clinical and economic benefits of FDCs has to be facilitated to different stakeholders, including macro-level decision makers, such as regulators and healthcare payers, clinical societies and patient organizations.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057609_p24
|
39057609
|
sec[2]/p[2]
|
3. Recommended Solutions
| 1.791016 |
biomedical
|
Other
|
[
0.5703125,
0.01300048828125,
0.416748046875
] |
[
0.0022735595703125,
0.99560546875,
0.0016002655029296875,
0.0003211498260498047
] |
Policymakers have to support the extended use of FDCs in CVDs by different measures. FDCs should be included in clinical guidelines and financing protocols at a preferred first-line position. Minimum quotas for the percentage of FDCs in selected CVDs and financial and other incentives to physicians for the prescription of FDCs need to be considered.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
39057609_p25
|
39057609
|
sec[2]/p[3]
|
3. Recommended Solutions
| 3.972656 |
biomedical
|
Review
|
[
0.99658203125,
0.002864837646484375,
0.0006246566772460938
] |
[
0.2841796875,
0.1278076171875,
0.58642578125,
0.0017986297607421875
] |
Clinical societies and regulators should create consensus about commonly accepted adherence measurement and calculation methods and reporting guidelines . If all adherence studies apply the same measures, meta-analyses can be conducted to strengthen the evidence base of adherence-improving technologies. With the increasing evidence about the benefits of FDCs in improving adherence and reducing mortality and MACEs, additional secondary research is needed to explore the generalizability of risk-reduction estimates to future FDCs and also to explore the effect size in different CV diseases and subpopulations with different baseline risks.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057609_p26
|
39057609
|
sec[2]/p[4]
|
3. Recommended Solutions
| 3.830078 |
biomedical
|
Study
|
[
0.9990234375,
0.00046443939208984375,
0.0003345012664794922
] |
[
0.8984375,
0.038116455078125,
0.0628662109375,
0.0005402565002441406
] |
Digitalization of medical records and recent initiatives to both standardize and link big data in European countries , accompanied by current advances in artificial intelligence methodologies, create a unique opportunity to generate real-world evidence (RWE) about the benefits of FDCs in reducing MACEs and mortality, as well as on reduced collateral effects and lower overall API utilization due to potentially lower doses.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057609_p27
|
39057609
|
sec[2]/p[5]
|
3. Recommended Solutions
| 1.705078 |
biomedical
|
Other
|
[
0.6357421875,
0.00926971435546875,
0.354736328125
] |
[
0.00899505615234375,
0.98779296875,
0.002582550048828125,
0.0004239082336425781
] |
The accessibility of payers’ databases to researchers has to be improved, partly because the RWE about the benefits of FDCs would be more acceptable for payers and HTA bodies if the evidence is generated in public ‘non-biased’ databases governed by healthcare payers.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057609_p28
|
39057609
|
sec[2]/p[6]
|
3. Recommended Solutions
| 3.4375 |
biomedical
|
Other
|
[
0.8115234375,
0.0301666259765625,
0.1583251953125
] |
[
0.01499176025390625,
0.94091796875,
0.0433349609375,
0.0006403923034667969
] |
Finally, the added value of FDCs should be acknowledged and incentivized by healthcare payers through new payment models, revising the current conservative pricing models, primarily based on internal reference pricing, into a formula that allows a fair premium price for incremental health gain. However, HTA bodies and healthcare payers cannot expect RCT evidence about the health gain of FDCs before pricing and reimbursement decisions, but they should rely on generalized evidence about the potential reduction in mortality and MACEs by FDCs, and/or they should facilitate ex-post evidence-generation methods based on real-world data, such as coverage with evidence-generation schemes.
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057609_p29
|
39057609
|
sec[3]/p[0]
|
4. Conclusions
| 3.851563 |
biomedical
|
Review
|
[
0.99267578125,
0.00292205810546875,
0.0045623779296875
] |
[
0.0243988037109375,
0.00634765625,
0.96875,
0.0003113746643066406
] |
From the societal and industry policy perspectives, the cost of incremental innovation is significantly less than the innovation of de novo medicines. The magnitude of the real-world benefits of FDCs is similar to new medicines, with less uncertainty about safety. FDCs have the potential to improve patient centricity in healthcare, which is an important attribute of extended value assessment. In conclusion, incremental innovation of FDCs in CVDs may be more cost-effective than the development of original medicines . The confirmation of the positive impact of FDCs on adherence has been reinforced recently by the inclusion of these treatments (FDCs and polypills) in the WHO Essential Medicines List .
|
[
"András Inotai",
"Zoltán Kaló",
"Zsuzsanna Petykó",
"Kristóf Gyöngyösi",
"Derek T. O’Keeffe",
"Marcin Czech",
"Tamás Ágh"
] |
https://doi.org/10.3390/jcdd11070186
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p0
|
39057742
|
sec[0]/p[0]
|
1. Introduction
| 3.378906 |
other
|
Other
|
[
0.1650390625,
0.0007505416870117188,
0.833984375
] |
[
0.272216796875,
0.55908203125,
0.1676025390625,
0.00091552734375
] |
Vision-based multiple object tracking (MOT) is a longstanding research problem with broad applications in computer vision such as intelligent surveillance systems, robotics, human–computer interaction, medical image processing, and autonomous driving. The MOT algorithm provides a robust framework for real-time monitoring and analysis of multiple moving objects, enabling accurate tracking and prediction of their movements in various dynamic scenarios. The tracking-by-detection paradigm is widely recognized as the most effective approach to multiple object tracking (MOT). It involves utilizing an efficient object-detection algorithm to identify objects within each frame of a video sequence. Subsequently, a data association algorithm is employed to establish associations between detections across frames, thereby creating object trajectories . Although various approaches have been presented to handle the problem, MOT is still a challenging research area due to factors like object occlusions, the varying number of objects per frame, abrupt appearance changes, etc.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p1
|
39057742
|
sec[0]/p[1]
|
1. Introduction
| 3.693359 |
other
|
Study
|
[
0.421875,
0.0008425712585449219,
0.5771484375
] |
[
0.99462890625,
0.004489898681640625,
0.0007576942443847656,
0.00012767314910888672
] |
The state-of-the-art MOT concepts have become more potent with recent advances in deep learning. MOT treats object detection and data association as two independent tasks. Modern advancements in deep learning have led to the development of highly effective off-the-shelf object detectors capable of accurately detecting various objects in complex scenes . Once object detections are obtained in each frame, the subsequent task of data association focuses on linking these detections across consecutive frames to establish object trajectories over time. Data association remains a challenging task in its own right and has not fully leveraged the advancements in deep learning. The standard process of data association typically involves extracting representative features from individual detections and then matching them with existing object trajectories using specified similarity metrics. Deep learning networks offer enhanced capability for learning robust feature representations of objects. In our study, we employed a modified VGGNet deep network for hierarchical feature learning and extraction from all detected objects. This deep feature extractor allowed us to capture distinctions and variations in object appearances, improving the accuracy and reliability of object associations over time.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057742_p2
|
39057742
|
sec[0]/p[2]
|
1. Introduction
| 3.53125 |
other
|
Study
|
[
0.330322265625,
0.0007205009460449219,
0.6689453125
] |
[
0.9697265625,
0.02899169921875,
0.0009918212890625,
0.0002675056457519531
] |
After extracting features for detections in each frame, the next task is to associate these detections with previously tracked objects. This association involves comparing the extracted features of detections with those of existing object trajectories to find the most suitable matches. Detections with the highest similarity scores are linked to their corresponding trajectories. Our research mainly focuses on enhancing the data association task within multiple object tracking (MOT) frameworks. We propose an efficient association framework that integrates a deep feature association network (deepFAN) and the Structural Similarity Index Metric (SSIM) . This framework jointly calculates association scores for object detection–target pairs. The deepFAN learns the complex feature association function to encode the association score between detections and tracked targets, while the SSIM handles uncertainties in association by comparing its feature similarities. By combining the deep learning capabilities of the deep feature association network with SSIM, our approach aims to improve the accuracy and reliability of object associations across frames.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p3
|
39057742
|
sec[0]/p[3]
|
1. Introduction
| 3.972656 |
other
|
Study
|
[
0.47119140625,
0.0008955001831054688,
0.52783203125
] |
[
0.970703125,
0.027099609375,
0.0018415451049804688,
0.00022518634796142578
] |
Traditionally, affinities in multiple object tracking (MOT) are calculated by exhaustively evaluating all possible pairs of detection and target features. In contrast, our proposed MOT framework integrates a neighborhood detection estimation (NDE) module to refine this process, selecting a more reliable subset of detection–target pairs. The NDE module enhances efficiency by focusing on nearby or contextually relevant detections rather than evaluating every possible permutation. This filtering step improves the quality of associations by prioritizing those with a higher likelihood of accuracy. In our framework, the deep feature association network (deepFAN) and the Structural Similarity Index Metric (SSIM) jointly determine the association score for these refined pairs. Furthermore, the training method we employed for the deep feature association network (deepFAN) enables efficient object association across multiple frames in a video sequence, ensuring reliable trajectory tracking. During training, the network is exposed to input frame pairs that are not necessarily consecutive. This strategy proves beneficial by allowing the framework to link objects across non-adjacent frames. This capability reduces instances of identity switches and fragmented object trajectories. By integrating the NDE module and optimizing deepFAN training, our MOT framework enhances tracking accuracy while maintaining computational efficiency.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p4
|
39057742
|
sec[0]/p[4]
|
1. Introduction
| 4.113281 |
biomedical
|
Study
|
[
0.61181640625,
0.00168609619140625,
0.38623046875
] |
[
0.99609375,
0.0025196075439453125,
0.001190185546875,
0.00013399124145507812
] |
This study presents a systematic approach to estimating an efficient association matrix for multiple object tracking (MOT), which effectively summarizes the correspondence between current frame detections and previously estimated target trajectories. Leveraging the capabilities of deep learning architectures, our proposed framework integrates innovative components aimed at enhancing MOT performance. Key components of our approach include the following: The proposed data association framework employs the deep feature association network (deepFAN) along with the Structural Similarity Index Metric (SSIM) to estimate an efficient association matrix. This combination improves the robustness of object associations by leveraging deep learning for feature extraction and similarity evaluation. In the proposed data association framework, a neighborhood-detection-estimation (NDE) scheme is introduced to select a reliable subset of detection–target pairs. This neighborhood detection estimation, along with post-processing steps within the deep feature association network, contributes to enhancing the computational efficiency. Experimental evaluations highlight that the proposed approach minimizes incorrect associations, thereby improving overall tracking performance. A specialized training strategy is developed for the deep feature association network (deepFAN), allowing the network to utilize non-consecutive frame pairs for the effective learning of the data association function. This method improves the overall ability of the network to link objects across frames, thereby reducing identity switches and fragmented trajectories.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p5
|
39057742
|
sec[0]/p[5]
|
1. Introduction
| 2.558594 |
other
|
Study
|
[
0.470703125,
0.0016069412231445312,
0.52783203125
] |
[
0.98681640625,
0.01224517822265625,
0.000823974609375,
0.0003447532653808594
] |
We validated the effectiveness of each component through ablative experiments on the MOT validation dataset. Additionally, comprehensive analyses on the benchmark datasets, including MOT15, MOT17, MOT20, and UA-DETRAC, demonstrated that our method achieved competitive and state-of-the-art results across various MOT evaluation metrics. The MOT metric scores for identity switches, fragmentation, and false negatives were reduced, indicating the reduction in the wrong association among detection target pairs.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p6
|
39057742
|
sec[0]/p[6]
|
1. Introduction
| 1.025391 |
other
|
Other
|
[
0.0090789794921875,
0.0005402565002441406,
0.990234375
] |
[
0.038299560546875,
0.9541015625,
0.00616455078125,
0.0015058517456054688
] |
The rest of the article is organized as follows: Section 2 reviews the existing literature on multiple object tracking (MOT). Section 3 details the methodology employed in the online multiple object tracking framework. In Section 4 , we present the experimental findings and comparative results and discuss them in depth. Finally, Section 5 concludes the study with a summary of the findings and suggestions for future research directions.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p7
|
39057742
|
sec[1]/p[0]
|
2. Related Works
| 1.643555 |
other
|
Other
|
[
0.0293426513671875,
0.0008521080017089844,
0.9697265625
] |
[
0.119873046875,
0.73291015625,
0.1448974609375,
0.0020961761474609375
] |
To obtain a comprehensive overview of the multiple object tracking (MOT) problem, we refer to foundational studies . Within MOT frameworks, the tracking-by-detection approach is the most commonly utilized method . The effectiveness of this approach relies heavily on the quality of object detections and the accuracy of trajectory estimation. The recent advancements in deep learning have significantly improved object detectors , leading to better object detection performance and, consequently, enhancing the overall efficiency of the MOT framework.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057742_p8
|
39057742
|
sec[1]/p[1]
|
2. Related Works
| 3.294922 |
other
|
Study
|
[
0.27490234375,
0.0005064010620117188,
0.724609375
] |
[
0.92333984375,
0.06866455078125,
0.0077056884765625,
0.0003657341003417969
] |
This discussion will focus specifically on data association approaches used for trajectory estimation in MOT. An essential step for any data association method is computing representative features of the detections in each frame. Several approaches exist for determining representation models, including appearance-based , motion-based , and composite models . For MOT frameworks, deep learning-based feature extraction methods provide robust and discriminative representation models for object detections, which significantly boost tracking performance. Typically, pre-trained classification or object detection models are employed for feature extraction in tracking tasks . In particular, ShiJie et al. proposed a deep affinity network that jointly learns representational features and their affinities with targets. Our proposed MOT framework adopts the feature extraction model utilized in .
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p9
|
39057742
|
sec[1]/p[2]
|
2. Related Works
| 3.855469 |
biomedical
|
Study
|
[
0.6142578125,
0.000606536865234375,
0.38525390625
] |
[
0.974609375,
0.00829315185546875,
0.016693115234375,
0.00019073486328125
] |
The study by Emami et al. views data association as a multidimensional assignment problem and consolidates many popular learning algorithms employed for MOT data association. Researchers have explored various methodologies, including non-probabilistic algorithms, probabilistic graphical models, Markov Chain Monte Carlo (MCMC), and deep learning techniques to solve the data association problem. Among non-probabilistic approaches, the Greedy Randomized Adaptive Search Procedure (GRASP) is frequently used for multi-sensor multi-object tracking . In probabilistic graphical models, common techniques include network optimization , conditional random fields , and belief propagation . Additionally, MCMC has been a valuable tool for data association in multiple object tracking .
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p10
|
39057742
|
sec[1]/p[3]
|
2. Related Works
| 3.681641 |
other
|
Review
|
[
0.27392578125,
0.001129150390625,
0.72509765625
] |
[
0.46142578125,
0.0521240234375,
0.485595703125,
0.0007724761962890625
] |
In recent years, there have been numerous successful attempts to formulate data association in MOT using deep learning methods. The Deep Affinity Network (DAN) proposed by ShiJie et al. represents an end-to-end trainable deep network that jointly learns feature modeling and association estimation. Similarly, FAMNet leverages deep networks for both feature extraction and association estimation. Yihong et al. introduced the Deep Hungarian Network (DHN) , which predicts associations from a cost matrix derived from detections and targets. The Dual-Matching Attention Network (DMAN) employs spatial and temporal attention mechanisms to predict and refine association assignments. The integration of deep models such as Recurrent Neural Networks (RNNs) , autoencoders , and Generative Adversarial Networks (GANs) into the data association problem has led to significant improvements in MOT performance.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p11
|
39057742
|
sec[1]/p[4]
|
2. Related Works
| 1.441406 |
other
|
Other
|
[
0.031494140625,
0.0005059242248535156,
0.9677734375
] |
[
0.27587890625,
0.72021484375,
0.00223541259765625,
0.00167083740234375
] |
This work presents a systematic approach to data association within the MOT framework that harnesses the power of deep learning models. The proposed MOT algorithm for track association enhances both computational efficiency and tracking accuracy. By leveraging the potential of deep learning models, our method aims to address the complexities and challenges associated with data association in MOT, ensuring more reliable and effective tracking outcomes.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057742_p12
|
39057742
|
sec[2]/p[0]
|
3. Methodology: Online Multiple Object Tracking Framework
| 1.619141 |
other
|
Study
|
[
0.039276123046875,
0.000568389892578125,
0.9599609375
] |
[
0.6318359375,
0.364990234375,
0.00182342529296875,
0.0013742446899414062
] |
In the tracking-by-detection paradigm of MOT, the process involves two distinct modules: the object detector and the object tracker. The object detector initially identifies target objects by generating bounding boxes in each video frame. From these bounding boxes, we compute the center locations of objects, C D f for frame I f . Our proposed MOT framework is designed to seamlessly integrate with existing multi-object detection methods. We evaluated our approach across various online challenges in multiple object tracking, where state-of-the-art object detectors provide the initial object detections. Specifically, we utilized detections from prominent MOT challenges such as MOT15, MOT17, MOT20 , and UA-DETRAC . Each challenge provides video sequences annotated with detections generated by specific detectors designated for the challenge.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p13
|
39057742
|
sec[2]/p[1]
|
3. Methodology: Online Multiple Object Tracking Framework
| 2.308594 |
other
|
Study
|
[
0.1571044921875,
0.0007343292236328125,
0.84228515625
] |
[
0.66943359375,
0.32763671875,
0.0019140243530273438,
0.0009927749633789062
] |
The block diagram representation of the proposed MOT framework is shown in Figure 1 . One of the significant components of the proposed MOT framework is a deep feature extractor using a modified VGGNet architecture. The architecture of the feature extractor employed in the proposed framework is based on the state-of-the-art MOT framework described in Reference . The system is expertly developed to efficiently extract comprehensive and compact features from the input object detections. The pretrained VGGNet architecture is fine-tuned within the context of multiple object tracking using training sequences of the MOT benchmark. As shown in Figure 1 , the representative feature of each object is obtained by passing the current video frame I f and object centers C D f through the deep feature extractor. For each object detection, a 520-dimensional feature vector is obtained. We refer to for the architectural details of the modified VGGNet feature extractor.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p14
|
39057742
|
sec[2]/p[2]
|
3. Methodology: Online Multiple Object Tracking Framework
| 2.707031 |
other
|
Study
|
[
0.4287109375,
0.000762939453125,
0.57080078125
] |
[
0.60693359375,
0.39111328125,
0.0012998580932617188,
0.0005998611450195312
] |
Let D f = { d i f } i = 1 N d represent the set of detections given in frame I f , where N d is the number of available detections. We acquire a detection feature matrix, F D f ∈ R 520 × N d , by accumulating the 520-dimensional feature vector for N d detections for each input frame I f . This detection feature matrix F D f is then made available for the data association task.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p15
|
39057742
|
sec[2]/sec[0]/p[0]
|
3.1. Data Association Methodology
| 3.625 |
other
|
Study
|
[
0.406005859375,
0.0005908012390136719,
0.59326171875
] |
[
0.978515625,
0.02032470703125,
0.0009479522705078125,
0.00017845630645751953
] |
In this section, we extend our discussion on the proposed data association framework that incorporates a deep feature association network and the Structural Similarity Index Metric along with neighborhood detection estimation to tackle the problem. The association algorithm identifies a correspondence between the object detections in the current frame and existing trajectories from the previous frames. This involves comparing the extracted features of detections with those of existing object trajectories to find the most suitable matches. Detections with the highest similarity scores are linked to their corresponding trajectories. Here, we employed a deep feature association network (deepFAN) that consists of a pre-trained CNN-based compression network and an image similarity metric, the Structural Similarity Index Metric (SSIM), to estimate the data association efficiently. This part of the MOT framework computes a feature association matrix A , which encodes the pairwise similarities of the detections and the pre-existing targets.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p16
|
39057742
|
sec[2]/sec[0]/sec[0]/p[0]
|
3.1.1. Neighborhood Detection Estimation
| 1.849609 |
other
|
Other
|
[
0.1497802734375,
0.0007944107055664062,
0.849609375
] |
[
0.345947265625,
0.6513671875,
0.0013284683227539062,
0.0010881423950195312
] |
Generally, the data association matrix models a global relationship between all the detections in the current frame and the tracked targets from the previous frames. In the proposed method, instead of considering all the combinations, only the reliable detection–target pairs are chosen for the association task. The neighborhood detection estimation methods are employed to identify those detection–target pairs. This method is based on the assumption that the objects are in a smooth motion, i.e., the location of the objects did not drift drastically in subsequent video frames. Therefore, we have to consider only the detections in the neighborhood area of the targets for the data association.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057742_p17
|
39057742
|
sec[2]/sec[0]/sec[0]/p[1]
|
3.1.1. Neighborhood Detection Estimation
| 2.8125 |
other
|
Study
|
[
0.369140625,
0.0007042884826660156,
0.63037109375
] |
[
0.732421875,
0.265869140625,
0.0010595321655273438,
0.00046634674072265625
] |
Let F T L f − 1 ∈ R 520 × N T L represent the set of target feature vectors in the previous frame I f − 1 , including the feature vectors of tracked and lost targets. (1) F T L f − 1 = { F T f − 1 , F L f − 1 } , where F T f − 1 ∈ R 520 × N T , F L f − 1 ∈ R 520 × N L , N T L = N T + N L . In Equation ( 1 ), F T f − 1 and F L f − 1 are the feature matrices that consist of the feature vectors of the tracked and lost targets from the previous frame I f − 1 and N T and N L are the number of active tracked and lost targets.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057742_p18
|
39057742
|
sec[2]/sec[0]/sec[0]/p[2]
|
3.1.1. Neighborhood Detection Estimation
| 2.960938 |
other
|
Study
|
[
0.336181640625,
0.0006799697875976562,
0.6630859375
] |
[
0.599609375,
0.3984375,
0.0015211105346679688,
0.0005865097045898438
] |
The neighborhood-detection-estimation algorithm simply relies on the distance between the centers of the detection and target feature vectors. In order to find the distance, we need to define a distance metric. Here, we are adopting the Euclidean distance with an additional scaling factor. Let C D = { C D x , C D y } and C T L = { C T L x , C T L y } be the centers of the detections and targets. The scaled Euclidean distance E s between a detection and a target with centers ( c D x , c D y ) and ( c t x , c t y ) is defined as (2) E s = ( c D x − c t x ) 2 + ( c D y − c t y ) 2 I x 2 + I y 2 , where ( I x , I y ) represents the size of the video frame.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p19
|
39057742
|
sec[2]/sec[0]/sec[0]/p[3]
|
3.1.1. Neighborhood Detection Estimation
| 3.619141 |
other
|
Study
|
[
0.35986328125,
0.0007848739624023438,
0.6396484375
] |
[
0.83837890625,
0.159423828125,
0.00162506103515625,
0.00040221214294433594
] |
Optical flow-based motion prediction: From the object detection bounding boxes, we can determine the center locations of all the detections in the image frame, C D f . Further, we have the locations of the tracked and lost targets C T L f − 1 in the frame I f − 1 as feedback information from the previous target trajectories. The possible locations of these targets in the present frame I f , C ^ T L f , are estimated using the optical flow motion model. Specifically, knowing the target center in I f − 1 , c t f − 1 = { c t x f − 1 , c t y f − 1 } , we compute its corresponding location c ^ t f in the following frame ( I f ) using the Lucas–Kanade optical flow method with pyramids . (3) c ^ t f = c t f − 1 + v = ( c t x f − 1 + v x , c t y f − 1 + v y ) , where v = ( v x , v y ) is the optical flow at c t f − 1 . Using optical-flow-based motion prediction, the location of a lost target is continuously updated. Consequently, if the target is occluded in one frame and reappears at a different location in subsequent frames, this motion prediction aids in estimating the likely location of the lost target. This approach improves the efficiency of reidentifying the lost target, leading to more accurate and reliable tracking performance.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057742_p20
|
39057742
|
sec[2]/sec[0]/sec[0]/p[4]
|
3.1.1. Neighborhood Detection Estimation
| 4.074219 |
biomedical
|
Study
|
[
0.9189453125,
0.0006852149963378906,
0.08050537109375
] |
[
0.9970703125,
0.0023136138916015625,
0.00037550926208496094,
0.00007522106170654297
] |
Using Equation ( 2 ), we calculate the distance between each existing target and all detections D f and select only those targets within the distance threshold, T e , to prioritize nearby detections. The network then encodes the feature vectors of all the possible pairings between the targets and the respective neighboring detections into a tensor, termed the feature permutation matrix Φ ∈ N × N × ( 520 × 2 ) . For clarity, the dimension of the tensor Φ is described as W i d t h × H e i g h t × D e p t h , where the width represents the targets and the height represents the detections. The feature vector of each target is concatenated with the feature vector of each one of its neighboring detections and arranged in Φ along its depth dimension. For each image frame in the video sequence, the number of targets and detections will vary. To maintain consistency in the tensor dimension, we introduce additional zero vectors into the matrix, ensuring that the size consistently remains at N × N × 1040 . The value chosen for N limits the maximum number of object detections in each frame, and through our analysis, N = 80 was found to be a generous bound for the MOT benchmark datasets.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p21
|
39057742
|
sec[2]/sec[0]/sec[1]/p[0]
|
3.1.2. Deep Feature Association Network
| 3.941406 |
biomedical
|
Study
|
[
0.9208984375,
0.0004982948303222656,
0.07867431640625
] |
[
0.994140625,
0.005535125732421875,
0.00029540061950683594,
0.00008302927017211914
] |
The objective of this component in the proposed MOT framework is to estimate the affinities between the selected detection–target pairs using the extracted feature vectors. This sub-network maps the tensor Φ ∈ R N × N × 1040 into a feature association matrix A F ∈ R N × N . In the association matrix A F , the columns account for the detections in the current frame and rows account for the active targets, both tracked and lost, from the previous trajectory. Besides, the scalar score in the matrix A F ( i , j ) indicates the confidence of the j t h detection and i t h target ( d j f and T L i f − 1 ) associated with the same identity.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p22
|
39057742
|
sec[2]/sec[0]/sec[1]/p[1]
|
3.1.2. Deep Feature Association Network
| 3.916016 |
biomedical
|
Study
|
[
0.8759765625,
0.0005168914794921875,
0.12335205078125
] |
[
0.986328125,
0.012939453125,
0.0005016326904296875,
0.00011301040649414062
] |
We refer to the major component of this module as the deep compression network due to its functionality. The architecture of the deep compression network is inspired by the work presented in . The input to this network is the tensor Φ ∈ R N × N × 1040 , which accumulates the feature vectors of target–detection pairs. The output is an association matrix A F ∈ R N × N that encodes the similarity scores of these pairs. The specifications of the deep compression network architecture are detailed in Table 1 . This network employs a five-layer convolutional neural network with 1 × 1 kernels for the task. As the tensor Φ passes through the network, it undergoes gradual dimension reduction along the depth dimension via the 1 × 1 kernels. These convolutional kernels enable the computation of similarity scores for each object pair without interference from neighboring objects.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p23
|
39057742
|
sec[2]/sec[0]/sec[1]/p[2]
|
3.1.2. Deep Feature Association Network
| 4.078125 |
biomedical
|
Study
|
[
0.6494140625,
0.0012674331665039062,
0.349365234375
] |
[
0.9912109375,
0.0079193115234375,
0.0007152557373046875,
0.0001323223114013672
] |
Training deep compression network: During the training process, the deep compression network learns the association function, which estimates the feature association matrix A F ∈ R N × N from the tensor Φ ∈ R N × N × 1040 for reliable online multiple object tracking. The approach used to train the compression network is illustrated in Figure 2 . When we employ the proposed MOT framework for online tracking, the feature extractor functions as a single-stream model. Additionally, during the tracking process, the input frames are presented in the order of the original video. We develop a specialized training strategy for the deep compression network, which enables the network to effectively learn the data association function by utilizing non-consecutive frame pairs from the video sequence. As a result, the network learns to reliably associate objects in a given frame with those in multiple previous frames, benefiting the framework by reducing identity switches and fragmented target trajectories. As shown in Figure 2 , during training, we configured the network as a two-stream network of modified VGGNet with shared parameters. The feature extractor receives two frames, I f and I f − p , separated by p frames (i.e., not adjacent frames), as well as the centers of object detection, C f and C f − p , of pre-detected objects within those frames. These frame pairs are processed by modified VGGNets, which extract a 520-dimensional feature vector for each object detection in the input frames. We obtain feature matrices, F D f and F D f − p , which accumulate the feature vectors for detections in each input frame I f and I f − p . Since the input frames are non-adjacent, neighborhood detection estimation (NDE) is not applicable and is excluded from the training pipeline. The network arranges the columns of F f and F f − p to concatenate the columns of the two feature matrices along the depth dimension of the tensor Φ ∈ R N × N × 1040 in all possible permutations. To maintain consistency in the tensor dimensions, additional zero vectors are introduced, ensuring that the size remains N × N × 1040 . This tensor is then forward-passed through the compression network, which utilizes five convolutional layers with 1 × 1 kernels to map and estimate the feature association matrix A F ∈ R N × N . For computing the error of the network during the learning process, we define a loss function J with the help of ground truth trajectories. A ground truth target association matrix G ∈ R N × N is constructed as a binary matrix encoding the correspondence between the objects detected in frames I f − p and I f . If the i t h target in I f − p corresponds to the j t h target in I f , then the entry to the matrix G ( i , j ) f − p , f is non-zero; otherwise, it is zero. The ground truth target association matrix G is subsequently compared with the network-predicted feature association matrix A F , for the loss computation. The loss function of our training network is defined as (4) J ( G , A F ) = ∑ i , j = 1 : N G ⊙ − ( log A F ) ∑ i , j = 1 : N G , where the symbol ⊙ represents the Hadamard product. The log operation on A F is performed elementwise, and ∑ i , j = 1 : N finds the sum of all elements in the Hadamard product matrix. In the loss function, instead of computing the distance metric between the predicted association matrix A F and ground truth association matrix G , the probabilities encoded by the relevant coefficients of A F are maximized. During learning, the parameters of the compression network are updated by minimizing the loss over the training samples. The trained compression network is employed in online multiple object tracking.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p24
|
39057742
|
sec[2]/sec[0]/sec[1]/p[3]
|
3.1.2. Deep Feature Association Network
| 4.101563 |
biomedical
|
Study
|
[
0.97900390625,
0.0003108978271484375,
0.0208740234375
] |
[
0.99365234375,
0.00580596923828125,
0.0004124641418457031,
0.00007408857345581055
] |
Referring to Section 3.1.1 , for consistency, additional zero vectors are introduced in the tensor Φ , so that the size will always be N × N × ( 520 × 2 ) . Therefore, in the association matrix A F ∈ R N × N , there are irrelevant values corresponding to the appended zero vectors. To reduce the irrelevant information and to normalize the matrix, we performed the following three post-processing steps over the feature association matrix A F : (i) Truncation : Since we have only N d detections and N T L active targets, the matrix A F ∈ R N × N is resized by truncating the matrix to N T L × N d . (ii) Rowwise Softmax : This operation normalizes the rows of the association matrix by fitting a separate probability distribution. The output row values are between the range , and the total sums up to 1. Thus, each row of the resulting association matrix encodes the association probability between each active target in I f − 1 and all the detections in I f . (iii) Thresholding : The association matrix values indicate the similarity between the detection and target objects. For a reliable data association, the values above the threshold T a are retained, and all other values below the threshold are set to zero.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057742_p25
|
39057742
|
sec[2]/sec[0]/sec[1]/p[4]
|
3.1.2. Deep Feature Association Network
| 2.566406 |
biomedical
|
Study
|
[
0.88330078125,
0.00092315673828125,
0.11602783203125
] |
[
0.798828125,
0.1998291015625,
0.0008096694946289062,
0.0005807876586914062
] |
These post-processing steps obtained for us an updated feature association matrix A F ∈ N T L × N d , which was further passed to the SSIM for the association update.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p26
|
39057742
|
sec[2]/sec[0]/sec[2]/p[0]
|
3.1.3. Structural Similarity Index Metric for Association Update
| 1.894531 |
other
|
Other
|
[
0.10174560546875,
0.0007853507995605469,
0.8974609375
] |
[
0.1297607421875,
0.8681640625,
0.0012903213500976562,
0.0007467269897460938
] |
The ultimate aim of the data association module is to develop a robust association model that delivers the most relevant information for achieving accurate multiple object tracking (MOT) performance. In the association matrix, a non-zero association value indicates a potential match between the corresponding target–detection pair. Traditionally, the detection with the highest association score is linked to the target trajectory. However, when multiple detections have similar or nearly equivalent association scores, uncertainties arise, leading to unreliable associations between detections and targets.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
39057742_p27
|
39057742
|
sec[2]/sec[0]/sec[2]/p[1]
|
3.1.3. Structural Similarity Index Metric for Association Update
| 3.574219 |
biomedical
|
Study
|
[
0.5693359375,
0.0006327629089355469,
0.429931640625
] |
[
0.91552734375,
0.08294677734375,
0.0010395050048828125,
0.0003101825714111328
] |
To address this issue, our proposed method incorporates the Structural Similarity Index Metric (SSIM) . The SSIM is a widely recognized perceptual metric that measures the similarity between two images by leveraging their structural characteristics. By integrating the SSIM, we enhance the decision-making process for target associations. The proposed MOT framework considers the association results derived from the SSIM to make the final decision on the target association. This metric evaluates the effective similarity between the target and detection pairs, thereby improving the accuracy and reliability of the associations. We reduced the chance of wrong associations, which can happen when multiple detections have association scores that are very close to each other by using the SSIM. This makes sure that the detections are more accurately aligned with their targets.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057742_p28
|
39057742
|
sec[2]/sec[0]/sec[2]/p[2]
|
3.1.3. Structural Similarity Index Metric for Association Update
| 3.041016 |
other
|
Study
|
[
0.4794921875,
0.0006670951843261719,
0.52001953125
] |
[
0.78759765625,
0.2110595703125,
0.001003265380859375,
0.0004115104675292969
] |
Let ( T L ) i be the i t h active target and { d k } k = 1 K be the detections corresponding to the non-zero association scores with the i t h target. Also, let d m a x represent the detection with the highest score and A F ( i , m a x ) be the highest score. As stated before, if there are other detections with similar or closer scores to this highest association score A F ( i , m a x ) , uncertainties occur in the target association. For the target ( T L ) i , first, the set of detections with uncertainty D s is estimated as follows. (5) D s i = { ∀ d k with A F ( i , k ) ≥ ( A F ( i , m a x ) − 0.1 ) } , D s = { D s i } i = 1 N T L .
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p29
|
39057742
|
sec[2]/sec[0]/sec[2]/p[3]
|
3.1.3. Structural Similarity Index Metric for Association Update
| 3.566406 |
biomedical
|
Study
|
[
0.830078125,
0.0007181167602539062,
0.169189453125
] |
[
0.6689453125,
0.329833984375,
0.00109100341796875,
0.0004076957702636719
] |
If the association matrix A F contains any zero rows, then the corresponding detection set in D s becomes an empty set. The SSIM module calculates the similarity score between the target and each detection in D s . The output of the SSIM module is another SSIM association matrix A S ∈ R N T L × N d in which rows and columns represent the same active targets and detections as in A F , but the entries replace the SSIM score of each valid pair, i.e., (6) A S ( i , j ) = S S I M ( T L i , d j ) , if d j ∈ D s i 0 , otherwise
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39057742_p30
|
39057742
|
sec[2]/sec[0]/sec[2]/p[4]
|
3.1.3. Structural Similarity Index Metric for Association Update
| 1.513672 |
other
|
Other
|
[
0.165771484375,
0.0011587142944335938,
0.8330078125
] |
[
0.036529541015625,
0.9619140625,
0.0008721351623535156,
0.0005993843078613281
] |
The SSIM-based association matrix, A s , is utilized alongside A F to establish the final track association, A . The track association matrix is the result of adding both matrices A F and A s together.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p31
|
39057742
|
sec[2]/sec[1]/p[0]
|
3.2. Track Association
| 1.336914 |
other
|
Other
|
[
0.014251708984375,
0.000766754150390625,
0.98486328125
] |
[
0.006893157958984375,
0.9921875,
0.00040149688720703125,
0.0003421306610107422
] |
In a multiple object tracking scenario, an object detected in a video sequence has to undergo different state transitions. When the object detector detects the object for the first time, a new track is initialized in the trajectory list. Now, the object is in the tracked state and remains in the same state until re-detected in the subsequent frames. When the object gets occluded or goes out of the camera’s field of view, the object is transferred to the lost state. If the lost target re-appears, then the state is updated as tracked, and the tracking process resumes. The trajectory of the lost target is terminated if it stays long in the lost state. The data association algorithm in MOT helps to find the state of each detection in the video sequence. It estimates the correspondence between the object detections in the current frame and existing targets.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p32
|
39057742
|
sec[2]/sec[1]/p[1]
|
3.2. Track Association
| 4.125 |
biomedical
|
Study
|
[
0.787109375,
0.0010900497436523438,
0.2117919921875
] |
[
0.9921875,
0.00699615478515625,
0.0005750656127929688,
0.00011104345321655273
] |
After accomplishing the training of the deep compression network with MOT datasets, we employed the trained network in the proposed MOT framework. Algorithm 1 summarizes the online tracking process in the proposed method. The objective of the MOT problem is to find the trajectory of all the possible targets present in the given input image sequence. Here, the MOT framework expects the present image frame I f and the object detection centers C D f as its inputs. The detection feature matrix F D f computed by the VGGNet feature extractor along with the target feature vector matrix F T L f − 1 are utilized to create the feature permutation tensor Φ by a concatenation operation. We stored the feature vectors of the active targets, both tracked and lost targets, from the previous frame to find the association in the current frames. The tensor Φ forward-passed through the compression network is mapped to the association matrix A F as described in Section 3.1.2 . Along with A F , the SSIM-based association matrix A s is also utilized for finding the final track association, A . The track association method adapted in our framework is performed as follows. Algorithm 1: Online multiple object tracking. Input: Video sequence, V = { I f | f = 1 , 2 , ⋯ , F } and object detections D f Output: Set of object trajectories, T = { τ i } i = 1 N , 1: Initialization : T ← ∅ 2: for Video frame I f in V do 3: Feature extraction 4: Input : I f and C D f 5: Output : F D f ∈ R 520 × N d 6: if ( f = = 1 ) then 7: Initialize trajectory τ i 1 for each detection, 8: state==tracked; 9: else 10: Neighborhood estimation detection 11: Input : F D f and F T L f − 1 12: Output : Tensor, Φ ∈ R N × N × 1040 13: Feature association network 14: Input : Φ 15: Output : A F 16: Structural Similarity Index Metric 17: for each active target, ( T L ) i , do 18: find D s i detections with uncertainty. 19: end for 20: Input : ( T L ) i and D s i , i = 1 : N T L 21: Output : A s 22: Final track association matrix 23: Input : A F and A s 24: Output : Final track association matrix, A = A F + A s 25: Target association 26: Hungarian algorithm assigns detection to active targets. 27: Input : A 28: Output : trajectory, τ f 29: if tracked track τ j ( f − 1 ) not assigned to detection then 30: state == lost; 31: end if 32: if lost track τ j ( f − 1 ) assigned to detection d m then 33: state ==tracked; 34: else 35: state==inactive (if length of lost frames > N i n a c t , terminate track); 36: end if 37: for detections not covered by tracked and lost targets do 38: Initialize trajectory τ i f . 39: state==tracked. 40: end for 41: end if 42: end for . 43: return trajectories of the objects, T .
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p33
|
39057742
|
sec[2]/sec[1]/p[2]
|
3.2. Track Association
| 1.394531 |
other
|
Other
|
[
0.08941650390625,
0.0010919570922851562,
0.90966796875
] |
[
0.101806640625,
0.89599609375,
0.001216888427734375,
0.0011205673217773438
] |
In the first frame I 1 , we initialize the trajectory list T with tracks { τ i } i = 1 N d by considering all the detections present in it as new tracked targets. Here, a track τ i is an ordered set of the states of the i t h target in the video sequence. (7) τ i = { s i f e , ⋯ , s i f t } , s i = ( c x , c y , w , h )
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p34
|
39057742
|
sec[2]/sec[1]/p[3]
|
3.2. Track Association
| 1.863281 |
other
|
Other
|
[
0.078857421875,
0.0007495880126953125,
0.92041015625
] |
[
0.1475830078125,
0.8505859375,
0.0011301040649414062,
0.0008273124694824219
] |
In Equation ( 7 ), f e and f t are the entry and terminate frame for the i t h target, ( c x , c y ) is the center of the target, and ( w , h ) are the width and height of the target. For each new target entry, the track is initialized with τ i = s i f e . The trajectory list is updated after each input frame by employing the Hungarian algorithm on the final association matrix A . In the track association part, the targets under the tracked state get higher priority. In this process, the targets that are associated with the detections are labeled as tracked, and the targets without association are labeled as lost. If the target stays in the lost state for a long time (say N i n a c t as the length of frames; here, we chose the value as 20 frames), it is considered that the object has entered an inactive state, and we terminate the trajectory corresponding to that object. Finally, we initialize new trajectories for the detections that are not associated with the tracked targets.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p35
|
39057742
|
sec[3]/p[0]
|
4. Experiment Results and Discussion
| 2.128906 |
other
|
Study
|
[
0.15283203125,
0.0008039474487304688,
0.84619140625
] |
[
0.982421875,
0.0166015625,
0.0008249282836914062,
0.0003669261932373047
] |
In this section, we experimentally demonstrate the performance of the proposed deep MOT framework on the popular MOT benchmark datasets using the standard metrics. Here, we present the implementation details of our MOT framework, followed by the benchmark datasets and metrics used for performance analysis. We first conducted an ablation study on the validation dataset to understand the behavior of our approach better. Further, to obtain an authoritative reference when addressing MOT problems, the proposed framework was evaluated on the test datasets and the results compared with the state-of-the-art methods.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p36
|
39057742
|
sec[3]/p[1]
|
4. Experiment Results and Discussion
| 1.529297 |
other
|
Study
|
[
0.05401611328125,
0.0005383491516113281,
0.9453125
] |
[
0.8125,
0.185302734375,
0.0011234283447265625,
0.0008730888366699219
] |
MOT benchmark datasets: The three popular MOT datasets, namely MOT15, MOT17, and MOT20 from the MOT Challenge , and UA-DETRAC were employed here to test the performance of the proposed approach. These are the centralized benchmark datasets used to evaluate the tracking techniques in online multiple-object-tracking challenges. The annotated training video sequence, which includes the object detections and the ground truth labels in each frame, is used to train the models. The video sequences in the test datasets provided only object detections, whereas the ground truth labels remained unrevealed. Once the new MOT tracker has been submitted for performance analysis, the online MOT challenge hosting server evaluates the tracking results based on the standard MOT metrics .
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p37
|
39057742
|
sec[3]/sec[0]/p[0]
|
4.1. Implementation Details
| 2.203125 |
other
|
Study
|
[
0.291015625,
0.0008687973022460938,
0.7080078125
] |
[
0.70751953125,
0.290771484375,
0.0010242462158203125,
0.0007691383361816406
] |
The proposed MOT framework was implemented in a Python framework, and the training was conducted on an NVIDIA Geforce Titan Xp 12GB GPU. We performed the training of the deep compression network on the MOT17 training dataset using the SGD optimizer. The hyperparameter values finally used in the training process were as follows: a batch size of 8, momentum of 0.9, an initial learning rate of 0.01, a weight decay of 0.0001, and the number of epochs per model of 120.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p38
|
39057742
|
sec[3]/sec[0]/p[1]
|
4.1. Implementation Details
| 2.835938 |
other
|
Study
|
[
0.2017822265625,
0.0006766319274902344,
0.79736328125
] |
[
0.931640625,
0.06671142578125,
0.0010213851928710938,
0.0004794597625732422
] |
In the proposed MOT framework, the decision for the state transition of a target from lost to inactive is based on the hyperparameter N i n a c t , which is the maximum number of frames the target stays in the lost state before being transferred into an inactive state. In our analysis, we kept the value for N i n a c t at 20. We chose N = 80 as a generous bound for the MOT benchmark datasets, because it limits the maximum number of object detections in each frame. The feature extractor network has an input frame size of 900 × 900 . Therefore, the network first resizes all the training and testing data to these dimensions before passing them through. The two threshold parameters used in this proposed framework are distance threshold, T e and association threshold T a . The optimum value for the evaluation metrics obtained with the value of T e is equal to 0.35. In the thresholding step implemented as a post-processing part of the feature association matrix A F , we used a association threshold T a . For T a equal to 0.4, the proposed MOT framework obtained the optimum performance. The selection of T e and T a is explained in Section 4.2 .
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p39
|
39057742
|
sec[3]/sec[1]/p[0]
|
4.2. Ablation Study
| 2.166016 |
other
|
Study
|
[
0.43212890625,
0.0015535354614257812,
0.56640625
] |
[
0.99365234375,
0.00572967529296875,
0.0004887580871582031,
0.00025963783264160156
] |
In this section, to gain a deeper insight into the proposed MOT framework, we experimentally evaluated the contribution of different tracking components. Since the ground truth annotations are not provided for the MOT test datasets, the ablation study was conducted on the MOT15, MOT17, and MOT20 training datasets. We split the MOT training dataset into training and validation datasets. The splitting of the dataset is presented in Table 2 . The proposed framework was trained on the training sequences, and the performance was evaluated on the validation sequences, as provided in Table 2 .
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p40
|
39057742
|
sec[3]/sec[1]/sec[0]/p[0]
|
Significance of the Proposed Tracking Components
| 2.494141 |
biomedical
|
Study
|
[
0.57080078125,
0.0010614395141601562,
0.42822265625
] |
[
0.98486328125,
0.013946533203125,
0.0008130073547363281,
0.0002739429473876953
] |
This section follows the detailed analysis and discussion on the results obtained for the analyses of the three main components, (i) neighborhood estimation detection, (ii) feature association network, and (iii) SSIM association update. To investigate the significance of each component, we conducted several experiments by disabling one element at a time and studying the performance for the MOT metrics. Table 3 consolidates the evaluation results of the variants of the proposed method on all MOT evaluation metrics that demonstrate the significance of each module in the framework: (i) Neighborhood detection estimation:
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p41
|
39057742
|
sec[3]/sec[1]/sec[0]/p[1]
|
Significance of the Proposed Tracking Components
| 1.645508 |
other
|
Study
|
[
0.11181640625,
0.0007810592651367188,
0.88720703125
] |
[
0.83154296875,
0.1663818359375,
0.0010089874267578125,
0.0008988380432128906
] |
As we discussed earlier, using NDE, we limited the search space for the association of the particular target into its neighborhood, assuming that the target will not move drastically from its position in a single frame change. The neighborhood of the target object was set to a limit using a distance threshold T e . Figure 3 shows the MOTA and IDF1 with different values for distance threshold T e . The optimum value for the evaluation metrics obtained with the value of T e is equal to 0.35, and we used this value of T e for the further experiments.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p42
|
39057742
|
sec[3]/sec[1]/sec[0]/p[2]
|
Significance of the Proposed Tracking Components
| 3.501953 |
other
|
Study
|
[
0.426513671875,
0.0008692741394042969,
0.57275390625
] |
[
0.99560546875,
0.0035152435302734375,
0.0005254745483398438,
0.00012934207916259766
] |
To demonstrate the significance of the proposed NDE in the MOT framework, we compared the performance of the trackers with and without NDE. Figure 4 shows three essential MOT metrics, MOTA, MOTP, and IDF1, evaluated on both the MOT17 and MOT20 validation datasets. Also, Table 3 tabulates the experimental results on all MOT metrics evaluated on the MOT17 and MOT20 validation datasets. The MOTA metric measures the overall accuracy of the detection and tracking, whereas the IDF1 scores highly depend on the association accuracy. The MOTP deals with the detection output. It is evident from the MOT scores that the scores improved with NDE. The MOTA is a metric derived from three types of detection–association errors: false positives, false negatives (missed targets), and identity switches. Since the NDE employed in the proposed method helps to choose only the reliable pairs for the association, it reduces the chance of wrong associations during the track estimation. It is clear from the results that NDE helps to reduce the wrong association, thereby reducing the identity switches, fragmentation, false negatives, and false positives, which in effect improves the MOTA. Also, the improvement in the IDF1 score also justifies that, with NDE, the association accuracy is improved.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p43
|
39057742
|
sec[3]/sec[1]/sec[0]/p[3]
|
Significance of the Proposed Tracking Components
| 3.480469 |
biomedical
|
Study
|
[
0.93310546875,
0.0004131793975830078,
0.06634521484375
] |
[
0.9931640625,
0.006443023681640625,
0.0002872943878173828,
0.00009399652481079102
] |
The deep feature association network (deepFAN) estimates the association matrix that encodes the association scores of each detection–target pair. The module includes three post-processing steps that remove the irrelevant information from the association matrix, improving the trajectory estimation. In the thresholding step, we used a hyperparameter, threshold T a . Figure 5 plots the MOTA and IDF1 scores of the proposed MOT framework with different values of T a , and an optimum result was obtained for T a equal to 0.4.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p44
|
39057742
|
sec[3]/sec[1]/sec[0]/p[4]
|
Significance of the Proposed Tracking Components
| 1.847656 |
other
|
Study
|
[
0.10699462890625,
0.0007538795471191406,
0.89208984375
] |
[
0.923828125,
0.07470703125,
0.0009212493896484375,
0.0006794929504394531
] |
Figure 6 and Table 3 show the performance of the proposed training strategy on the MOT17 and MOT20 validation sequences in terms of the MOT metrics. The deep network was trained on the MOT dataset with the strategy that the input frames need not be sequential, i.e, non-consecutive input frames. Therefore, the data association model becomes robust to the tracking challenges such as appearance variation, illumination changes, scale changes, etc. It also helps in the re-identification of the lost targets and handles object occlusions, thereby reducing the identity switches and fragmentation issues in MOT. The experimental results showed that it improves the overall MOT performance.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p45
|
39057742
|
sec[3]/sec[1]/sec[0]/p[5]
|
Significance of the Proposed Tracking Components
| 2.220703 |
other
|
Study
|
[
0.1864013671875,
0.0007634162902832031,
0.81298828125
] |
[
0.9658203125,
0.03302001953125,
0.0007491111755371094,
0.0004143714904785156
] |
The SSIM introduced in the proposed model can be considered as a second opinion when an ambiguity in association occurs. Figure 7 and Table 3 show the importance of SSIM association by evaluating the model on MOT metrics with the MOT17 and MOT20 validation data sequences. As the performance of the data association algorithm improves, we will obtain better association estimation, which will enhance the tracker’s tracking performance. It is observed from the results that the SSIM enhances the performance of the data association algorithm. It reduces the false negatives and identity switches and, hence, the MOTA. Also, the high IDF1 score validates the significance of SSIM association in the refinement of the association matrix.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39057742_p46
|
39057742
|
sec[3]/sec[2]/p[0]
|
4.3. MOT Benchmark Evaluation
| 1.379883 |
other
|
Study
|
[
0.034881591796875,
0.0005736351013183594,
0.96435546875
] |
[
0.9287109375,
0.0689697265625,
0.001522064208984375,
0.0008063316345214844
] |
This section shows the experimental evaluation of the proposed method on the benchmark datasets. Table 4 summarizes and compares our results with state-of-the-art algorithms on MOT benchmark datasets and Table 5 on UA-DETRAC. Here, we also show the effects of systematically adding neighborhood detection estimation, non-sequential training, and SSIM association update in the proposed tracker.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39057742_p47
|
39057742
|
sec[3]/sec[2]/p[1]
|
4.3. MOT Benchmark Evaluation
| 2.404297 |
biomedical
|
Study
|
[
0.6220703125,
0.001934051513671875,
0.3759765625
] |
[
0.9560546875,
0.041595458984375,
0.0015516281127929688,
0.0005426406860351562
] |
The benchmark evaluation results show that the proposed MOT framework performs very well in terms of the MOT evaluation metrics. We would like to emphasize that the metric scores for identity switches, fragmentation, and false negatives are reduced, indicating the reduction in the wrong association among detection target pairs. This results in better accuracy (MOTA). Also, the IDF1 score is improved, which is a clear indication of the association accuracy. This shows the robustness and efficiency of the proposed data association method.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p48
|
39057742
|
sec[3]/sec[2]/p[2]
|
4.3. MOT Benchmark Evaluation
| 2.548828 |
other
|
Study
|
[
0.43310546875,
0.0017099380493164062,
0.5654296875
] |
[
0.98095703125,
0.0169219970703125,
0.0016193389892578125,
0.0004940032958984375
] |
We compared our results with recent state-of-the-art methods. The benchmark evaluation result depicts that the proposed data association method outperforms the state-of-the-art DAN model . In particular, the nearest neighborhood estimation employed for detection–target feature pair selection reduces the association mismatch and improves the computational efficiency. The post-processing steps after deepFAN also help enhance the association accuracy and reduce the computational complexity. Here, the employment of the SSIM reduces the ambiguity in the data association.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39057742_p49
|
39057742
|
sec[3]/sec[2]/p[3]
|
4.3. MOT Benchmark Evaluation
| 1.219727 |
other
|
Study
|
[
0.01708984375,
0.00048804283142089844,
0.982421875
] |
[
0.544921875,
0.45166015625,
0.002208709716796875,
0.0015239715576171875
] |
The tracking results of the proposed tracker with the UA-DETRAC dataset are summarized in Table 5 . Here, we opted for the EB detector for a fair comparison. Since the trackers in Table 5 used different detectors, the name of the tracker is given along with the detector used. The proposed method gives better results on the UA-DETRAC evaluation compared with other approaches and can also be effectively used for vehicle tracking.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39057742_p50
|
39057742
|
sec[4]/p[0]
|
5. Conclusions
| 1.830078 |
other
|
Study
|
[
0.026885986328125,
0.00047850608825683594,
0.97265625
] |
[
0.951171875,
0.04705810546875,
0.0011425018310546875,
0.0007348060607910156
] |
Developing a better data association framework is very crucial for robust multiple object tracking. This research work proposes two important contributions to enhance the data association. The first one is by introducing neighborhood detection estimation (NDE) only to retain reliable detection–target pairs. Secondly, the SSIM association component is proposed to reject ambiguous associations with high or near high association scores. A comprehensive evaluation strategy was adopted to understand and study the impacts of our technical contributions on popular multiple object tracking benchmarks. Further, we carried out a systematic ablation study to pinpoint the benefits of each proposal. We compared our proposals with recent multiple object tracking frameworks. Our studies found that the proposed tracker gave very low identity switches, which is one of the crucial factors in ranking various trackers. Further, the proposed tracker also achieved very high overall MOTA and IDF1 scores. Another factor that we wish to highlight here is that the proposed framework rejects ambiguous associations and employs only the neighboring detections for data associations. Ultimately, this leads to achieving higher tracking speed, which is another important factor in multiple object tracking. In the future, we would like to deploy this tracker in real-time tracking scenarios by augmenting a dedicated object-detection module along with the proposed tracker for real-world applications.
|
[
"Aswathy Prasannakumar",
"Deepak Mishra"
] |
https://doi.org/10.3390/jimaging10070171
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11277577_p0
|
PMC11277577
|
sec[0]/p[0]
|
1. Introduction
| 4.410156 |
biomedical
|
Review
|
[
0.998046875,
0.001132965087890625,
0.0005774497985839844
] |
[
0.39208984375,
0.003131866455078125,
0.60400390625,
0.0009183883666992188
] |
Antimicrobial resistance (AMR) is a pressing global issue that requires collaborative efforts from nations and foundations worldwide . Clinical and public health challenges posed by emerging AMR pathogens are particularly pronounced in low-resource settings, where enhanced laboratory capabilities and robust data collection systems are needed to fully address this health threat. Until recently, carbapenems served as last-resort treatments for Gram-negative bacterial infections . However, the global emergence and rapid spread of carbapenem-resistant organisms present a significant risk of high mortality across diverse populations due to limited treatment options . Carbapenem resistance can develop through various mechanisms, including (i) structural modifications of penicillin-binding proteins, (ii) reductions in outer-membrane porins, (iii) activation of efflux pumps, and (iv) production of β-lactamases (carbapenemases) that degrade or hydrolyze carbapenems . Among these, the producibility of carbapenemases is particularly noteworthy in terms of its impact on infection prevention and treatment.
|
[
"Lutfun Nahar",
"Hideharu Hagiya",
"Kazuyoshi Gotoh",
"Md Asaduzzaman",
"Fumio Otsuka"
] |
https://doi.org/10.3390/jcm13144199
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11277577_p1
|
PMC11277577
|
sec[0]/p[1]
|
1. Introduction
| 4.523438 |
biomedical
|
Study
|
[
0.9990234375,
0.0005688667297363281,
0.00028443336486816406
] |
[
0.82421875,
0.002292633056640625,
0.1727294921875,
0.0006885528564453125
] |
A wide range of carbapenemases are classified into Ambler Classes based on their hydrolytic profiles and catalytic substrates . Class B enzymes, also known as metallo-β-lactamases (MBLs), employ zinc as a cofactor at the active site of the β-lactam ring. This class mainly includes New Delhi metallo-beta-lactamase (NDM), Verona Integron-encoded metallo-beta-lactamase (VIM), and imipenemase (IMP). Among these, NDM is the most prominent genotype capable of catalyzing a range of β-lactam antibiotics, including carbapenems, and is resistant to various β-lactamase inhibitors . Since the first detection of the NDM-1 gene in Enterobacterales isolated from a patient traveling from India to Sweden in 2008 , a total of 41 NDM variants have been identified in clinically significant pathogens such as Escherichia coli , Klebsiella pneumoniae , Acinetobacter baumannii complex, and Pseudomonas aeruginosa , of which 40 variants have been deposited in the GenBank database . Due to its high-level and multi-drug resistance nature, only a limited number of treatment options are available for NDM-producing bacterial infections. The endemic regions of these NDM producers have rapidly expanded worldwide, affecting communities, animals, agricultural products, and the environment , exposing an increasing number of people to untreatable infections. In the era of international travel and medical tourism, this unfavorable situation is accelerating globally .
|
[
"Lutfun Nahar",
"Hideharu Hagiya",
"Kazuyoshi Gotoh",
"Md Asaduzzaman",
"Fumio Otsuka"
] |
https://doi.org/10.3390/jcm13144199
|
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.