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
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
PMC11276350_p32
|
PMC11276350
|
sec[3]/p[2]
|
4. Discussion
| 1.824219 |
biomedical
|
Study
|
[
0.47509765625,
0.3955078125,
0.1295166015625
] |
[
0.58740234375,
0.39892578125,
0.002964019775390625,
0.01073455810546875
] |
A total of 55.4% of the CYP were referred to the psychology service either during or after their admission to hospital and received an average of five sessions with a psychologist. The reason for referral to psychology from MDT colleagues rarely mentioned trauma as a presenting concern, yet a significant proportion of the CYP were experiencing trauma-related symptoms. In addition, only 8.6% of the referrals were for psychological support for the parent, yet the data suggested a significant proportion were experiencing distress.
|
[
"Anita Freeman",
"Emily Golding",
"Jennifer Gardner",
"Zoe Berger"
] |
https://doi.org/10.3390/children11070858
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11276350_p33
|
PMC11276350
|
sec[3]/p[3]
|
4. Discussion
| 1.835938 |
biomedical
|
Study
|
[
0.681640625,
0.0156707763671875,
0.302734375
] |
[
0.8359375,
0.1602783203125,
0.0024852752685546875,
0.0013532638549804688
] |
When exploring the relationship between the referrals to psychology and the demographic of the population, the data indicates that a significantly greater proportion of CYP from a Global Majority ethnic background were referred to the service. It is unclear why this difference was observed. It could be hypothesised that this is related to the identified referral themes of isolation, stigma and shame acting as increased psychological risk factors for those from these backgrounds compared to those from UK and other White backgrounds.
|
[
"Anita Freeman",
"Emily Golding",
"Jennifer Gardner",
"Zoe Berger"
] |
https://doi.org/10.3390/children11070858
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276350_p34
|
PMC11276350
|
sec[3]/p[4]
|
4. Discussion
| 3.738281 |
biomedical
|
Review
|
[
0.96533203125,
0.033172607421875,
0.0015087127685546875
] |
[
0.285400390625,
0.03533935546875,
0.6689453125,
0.0100860595703125
] |
There are several limitations to the data, including that this was a single centre audit of a tertiary children’s hospital based in London. There is a possibility of referral bias, with referral rates observed to be higher for the most unwell patients with PIMS-TS and those from particularly disadvantaged backgrounds due to regional demographics. Despite these limitations, there are several important clinical implications from this retrospective review. It once again highlights the importance of proactive screening for both trauma and emotional distress in CYP and their parents/caregivers following sudden and unexpected hospital admission, especially to intensive care units. It also highlights the need to think about a CYP as part of a system of care and to ensure that clinicians pay attention to parental wellbeing and mental health when understanding the psychological impact on a child.
|
[
"Anita Freeman",
"Emily Golding",
"Jennifer Gardner",
"Zoe Berger"
] |
https://doi.org/10.3390/children11070858
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
PMC11276350_p35
|
PMC11276350
|
sec[4]/p[0]
|
5. Conclusions
| 4.085938 |
biomedical
|
Study
|
[
0.9716796875,
0.0274505615234375,
0.0010356903076171875
] |
[
0.99365234375,
0.0025882720947265625,
0.002460479736328125,
0.0014476776123046875
] |
These results shine a spotlight on the potentially important role medical teams have in containing and ameliorating unnecessary distress in CYP and their families following a physical health diagnosis. At the point at which the CYP in this sample were being admitted with this novel disease and during the height of the pandemic, there was limited knowledge about the trajectory, prognosis or potential recovery from the illness. The medical team were, therefore, unable to provide necessary information and/or reassurance to CYP and their families both to navigate the current admission or to plan for the future. Preventative interventions which might otherwise have been available in the form of leaflets, psychoeducation and psychologically informed resources were not yet available. This understandably might have accounted for higher levels of anxiety and distress in this population. In addition, there were a number of CYP and parents who reported feelings of isolation, fear and uncertainty as well as shame and stigma. This related to feeling as if they could have avoided contracting the illness, that they would be perceived to be ‘to blame’ for becoming unwell and/or had concerns about contamination and how they would be experienced by others once discharged. Clinically, this felt reminiscent of the clinical themes that emerge when working with CYP with an HIV diagnosis . We recognise that there were no opportunities for CYP and their parents to mix or meet with other families on the ward when admitted, whereby peer-to-peer support would usually be fostered. Lastly, we think it is important to note that a large proportion of the CYP with PIMS-TS were from a UK Global Majority background. We wondered if this meant that a number of the CYP had already had experiences of discrimination, marginalisation and structural racism that heightened their experience of shame and stigma. As already discussed in a previous paper , there were considerable benefits from the CYP being able to meet with other CYP who had PIMS-TS and talk about their experiences. This had positive benefits on wellbeing by reducing isolation and helping CYP to connect with stories of recovery and growth.
|
[
"Anita Freeman",
"Emily Golding",
"Jennifer Gardner",
"Zoe Berger"
] |
https://doi.org/10.3390/children11070858
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p0
|
39056900
|
sec[0]/p[0]
|
1. Introduction
| 1.071289 |
other
|
Other
|
[
0.00969696044921875,
0.00034332275390625,
0.98974609375
] |
[
0.064697265625,
0.93212890625,
0.002269744873046875,
0.0009336471557617188
] |
The term elephant flows refers to the most substantial data transfers across the Internet, which, despite their limited numbers, usually carry the majority of the traffic. In contrast, the numerous mouse flows constitute a large portion of the flow count but account for only a minor fraction of total traffic. This imbalance surpasses the conventional 80/20 ratio defined by the Pareto principle. Recent studies, including references such as , have revealed that a mere 0.2–0.4% of all flows might be responsible for as much as 80% of the entirety of Internet traffic, showcasing an extreme concentration of data within a small fraction of flows.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39056900_p1
|
39056900
|
sec[0]/p[1]
|
1. Introduction
| 1.234375 |
other
|
Other
|
[
0.0028743743896484375,
0.00047516822814941406,
0.99658203125
] |
[
0.00818634033203125,
0.9892578125,
0.0016431808471679688,
0.0006837844848632812
] |
The approach of flow-based traffic engineering has recently emerged as a powerful technique for addressing the challenges of escalating network demands while preserving the quality of service (QoS) . This strategy involves assigning a unique forwarding entry to each flow in the switch’s memory, with each entry detailing the subsequent hop along the flow’s path. This arrangement allows for the use of varied paths for flows that share the same source and destination, thereby facilitating multipath routing. Moreover, paths for incoming flows can be chosen based on present or expected network congestion, enabling an adaptive routing that effectively avoids congested links. Furthermore, this method of flow-based adaptive routing is known to offer higher stability compared to conventional dynamic load-balancing techniques.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p2
|
39056900
|
sec[0]/p[2]
|
1. Introduction
| 1.164063 |
other
|
Other
|
[
0.00420379638671875,
0.0004050731658935547,
0.99560546875
] |
[
0.2298583984375,
0.765625,
0.0025348663330078125,
0.0019016265869140625
] |
The primary issue in flow-based traffic engineering arises from the fact that the number of concurrent flows in a network often exceeds the capacity of the flow tables within switches . Furthermore, in centralized software-defined networks (SDNs), there is a bottleneck concerning the controller’s ability to handle new flow setups due to its throughput limitations. Beyond the issue of capacity, having fewer entries in the flow tables can lead to faster table lookups, thereby enhancing the packet switching speed. One viable approach to mitigate these challenges involves dedicating entries exclusively to the largest flows. Consequently, the majority of smaller flows could be directed along default, shortest-path routes. This strategy significantly reduces the number of flow table entries while still effectively managing a substantial volume of traffic through specialized, flow-specific entries. It is crucial to predict these values as early as possible to quickly establish individual entries and route most of their packets through specific paths. Ideally, flows should be classified with their initial packet to prevent mid-connection rerouting, which can disrupt transport protocols’ path state estimations. Additionally, we find that first packet classification is mostly an unexplored field, especially within the context of SDN traffic engineering. Therefore, we decided to focus solely on the first packet classification to fill this gap.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39056900_p3
|
39056900
|
sec[0]/p[3]
|
1. Introduction
| 1.692383 |
other
|
Study
|
[
0.052734375,
0.0007615089416503906,
0.9462890625
] |
[
0.98291015625,
0.016265869140625,
0.0006184577941894531,
0.00040078163146972656
] |
Our research evaluates the efficacy of the TabNet deep tabular data learning architecture, with a particular focus on metrics vital for traffic engineering within SDNs. The key contributions of our work are as follows: Traffic Coverage: We examine the volume of traffic managed by flows classified as elephants following their identification, termed traffic coverage . Flow Table Reduction: We analyze the reduction in the necessity for individual flow entries in the tables, denoted as flow table operation reduction . Entropy Analysis: We provide an analysis of the average information entropy contained in each 5-tuple field in packet headers. Feature Significance: We identify which 5-tuple fields were the most significant for predictions.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
39056900_p4
|
39056900
|
sec[1]/p[0]
|
2. Related Work
| 1.106445 |
other
|
Other
|
[
0.005664825439453125,
0.0006299018859863281,
0.99365234375
] |
[
0.0062103271484375,
0.99072265625,
0.00252532958984375,
0.0007648468017578125
] |
The strategy of selectively managing elephant flows dates back to 1999 when the idea of adaptively routing substantial data flows was introduced . Initially, due to the hardware constraints of the era, the concept was largely theoretical and confined to academic discussions. However, the rise of SDNs has revitalized interest in this approach. In the contemporary networking landscape, a controller with comprehensive insight into network dynamics is well positioned to efficiently oversee large flows, leveraging the advanced capabilities of SDNs.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p5
|
39056900
|
sec[1]/p[1]
|
2. Related Work
| 1.412109 |
other
|
Other
|
[
0.0091552734375,
0.0003190040588378906,
0.99072265625
] |
[
0.40380859375,
0.58984375,
0.0046844482421875,
0.0016307830810546875
] |
The Hedera traffic engineering system, unveiled in , was created to dynamically reroute flows through an embedded controller once they surpassed a predefined threshold, guiding these flows along paths selected in real time. It presupposes that edge devices are responsible for collecting comprehensive flow statistics via OpenFlow counters, with a focus on optimizing the performance of non-edge devices. DevoFlow, introduced in , emphasizes the management of elephant flows by implementing sampling techniques and utilizing threshold values for their identification. Nevertheless, the evaluation of DevoFlow’s effectiveness is conducted based on the network’s aggregated performance, not on flow table characteristics. A similar approach to DevoFlow is explored in , where elephant flows are identified at edge devices using an adapted Bloom filter. This method’s underlying traffic model, assuming a disproportionate contribution of 20% of the flows to 80% of the traffic, diverges significantly from real-world data distributions, as recent research, such as , suggests a much more tail-skewed distribution.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p6
|
39056900
|
sec[1]/p[2]
|
2. Related Work
| 3.521484 |
other
|
Study
|
[
0.484619140625,
0.00099945068359375,
0.51416015625
] |
[
0.97021484375,
0.00487518310546875,
0.0247650146484375,
0.00023162364959716797
] |
The referenced studies primarily employ rudimentary techniques such as sampling, counters, and threshold settings for the detection of large flows. However, there has been a shift towards more sophisticated, machine-learning-based approaches in recent years. For instance, a decision tree model dedicated to identifying elephant flows was introduced and assessed in , with a particular emphasis on the accuracy of detection. In another study, ref. by Poupart et al. explored the capabilities of three different machine learning (ML) strategies for estimating flow sizes and categorizing them as elephant flows. Their analysis was based on a comprehensive dataset of three million flows, covering both TCP (Transmission Control Protocol) and UDP (User Datagram Protocol). Their evaluation focused on two principal metrics: the success rate in correctly identifying large flows ( true positive rate ) and the success rate in accurately identifying smaller flows ( true negative rate ).
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p7
|
39056900
|
sec[1]/p[3]
|
2. Related Work
| 2.568359 |
other
|
Study
|
[
0.20947265625,
0.0006232261657714844,
0.7900390625
] |
[
0.974609375,
0.023162841796875,
0.001922607421875,
0.00028514862060546875
] |
In , Liu et al. recommend the application of a random forest decision tree for pinpointing eight essential features crucial for developing a classification model. They introduce a dual-layered architectural framework that involves an initial pre-classification phase at the edge devices within an SDN setup, followed by a more detailed classification at the network’s central controller. This classification scheme distinguishes between four distinct types of flows: elephant, cheetah, tortoise, and porcupine, each representing different characteristics and behaviors within the network traffic. The research primarily evaluates the effectiveness of this system based on two metrics: the precision of the flow classification and the time delay associated with the classification procedure.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p8
|
39056900
|
sec[1]/p[4]
|
2. Related Work
| 1.506836 |
other
|
Study
|
[
0.0249786376953125,
0.000400543212890625,
0.974609375
] |
[
0.69140625,
0.302490234375,
0.004604339599609375,
0.0013217926025390625
] |
In their research, Hamdan et al. introduce a two-level classification framework for network traffic, initially sorting flows at switches and finalizing classifications at the central controller. This method uses the count-min sketch algorithm at the switch level to separate mice from potential elephant flows, with a decision tree at the controller for final decisions. The system’s algorithms are periodically refreshed with data from the controller, emphasizing classifier accuracy, and validated with real traffic models.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39056900_p9
|
39056900
|
sec[1]/p[5]
|
2. Related Work
| 1.482422 |
other
|
Study
|
[
0.037261962890625,
0.0006456375122070312,
0.9619140625
] |
[
0.751953125,
0.1885986328125,
0.0572509765625,
0.0020198822021484375
] |
He et al. and Qian et al. , in 2022, each proposed sketch-based solutions for flow table optimization. He et al. developed a single-level, lightweight scheme, while Qian et al. introduced TCAM-based storage for elephant flow labels to balance accuracy between elephant and mouse flow identification. Both studies utilized real Internet Service Provider (ISP) packet traces for evaluation, indicating the practical effectiveness of their approaches in traffic management.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p10
|
39056900
|
sec[1]/p[6]
|
2. Related Work
| 2.832031 |
other
|
Study
|
[
0.1820068359375,
0.0006999969482421875,
0.8173828125
] |
[
0.89306640625,
0.059844970703125,
0.046661376953125,
0.0006880760192871094
] |
In the study , da Silva et al. introduced a predictive model using the Locally Weighted Regression (LWR) algorithm to estimate the size and duration of new network flows by examining patterns from previous flows and their immediate correlations. Following up, in 2022, employing a hashing mechanism inspired by the Cuckoo Search meta-heuristic for enhanced flow management was proposed by the same authors. Pekar et al. presented a novel threshold-agnostic heavy-hitter classification system , which utilizes template matching to identify elephant flows based on the packet size distribution observed in the initial packets, offering a nuanced method for flow classification without predetermined thresholds.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39056900_p11
|
39056900
|
sec[1]/p[7]
|
2. Related Work
| 2.134766 |
other
|
Study
|
[
0.1192626953125,
0.0005054473876953125,
0.88037109375
] |
[
0.7392578125,
0.25634765625,
0.00372314453125,
0.0007877349853515625
] |
The CrossBal system, detailed in , is a hybrid load-balancing solution that employs Deep Reinforcement Learning (DRL) to specifically address elephant flows through a three-level detection mechanism, including threshold-based filtering, followed by rerouting for efficient load distribution. In a related study, Wassie et al. introduced a deep learning approach utilizing deep autoencoders, gradient boosting, and autoML predictive algorithms like eXtreme gradient boosting (XGBoost) and the gradient boosting machine (GBM) , aimed at enhancing flow management.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39056900_p12
|
39056900
|
sec[1]/p[8]
|
2. Related Work
| 1.481445 |
other
|
Study
|
[
0.0261077880859375,
0.0003781318664550781,
0.9736328125
] |
[
0.69970703125,
0.29345703125,
0.00567626953125,
0.0012226104736328125
] |
All the mentioned studies focus on classifying flows after observing several initial packets. However, our goal is to identify a flow as quickly as possible, ideally based on the information carried in the first packet. Durner et al. achieved flow classification using just the first packet’s 5-tuple data and its size. Hardegen et al. proposed using multiclass prediction instead of binary classification (elephant/mouse) with a deep neural network to predict flow characteristics from the first packet’s 5-tuple. This approach follows a similar methodology to their earlier work on predicting a flow’s bit rate from the first packet’s 5-tuple.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p13
|
39056900
|
sec[1]/p[9]
|
2. Related Work
| 1.495117 |
other
|
Study
|
[
0.0384521484375,
0.0006022453308105469,
0.9609375
] |
[
0.79736328125,
0.1756591796875,
0.0255126953125,
0.0014657974243164062
] |
Regarding the most recent works, in 2023 Gomez et al. evaluated several machine learning algorithms for classifying flows from the first packet. Similar to other studies, it focused on classification accuracy and not on flow table impact or traffic coverage. Xie et al.’s 2024 paper proposed a two-stage decision tree system for elephant flow classification. The first stage is utilizing information contained in first packet headers. The system was developed in P4, but tested only in an emulator, lacking real-device validation.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p14
|
39056900
|
sec[1]/p[10]
|
2. Related Work
| 1.71582 |
other
|
Study
|
[
0.018646240234375,
0.0003795623779296875,
0.98095703125
] |
[
0.7421875,
0.23681640625,
0.01934814453125,
0.001537322998046875
] |
Recent studies have also applied neural networks for network flow classification with a focus on QoS rather than traffic engineering. Alkhalidi et al. introduced a one-dimensional convolutional neural network for classifying flows into various classes using packet header information. A notable innovation is the automatic selection of specific packet header bits, reducing feature count, processing time, and energy consumption while maintaining satisfactory accuracy. Yaseen et al. employed a similar approach to classify traffic and assign the Differentiated Services Code Point (DSCP) field, implementing their system within an SDN controller and testing it in the Mininet emulator for emergency traffic prioritization scenarios.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999993 |
39056900_p15
|
39056900
|
sec[1]/p[11]
|
2. Related Work
| 1.367188 |
other
|
Study
|
[
0.01300048828125,
0.000499725341796875,
0.986328125
] |
[
0.81982421875,
0.16064453125,
0.0177154541015625,
0.0017938613891601562
] |
As can be seen, the referenced studies focus on flow classification accuracy or true positive/true negative rates. However, they neglect the practical implications of these algorithms for traffic engineering goals. For instance, misclassifying the largest flow in a network can substantially impact traffic coverage, far more than the misclassification of smaller flows. The metrics employed in these studies do not account for such nuances. Specifically, there has been a lack of focus on metrics essential for traffic engineering, such as the reduction in flow table entries or the volume of traffic managed after classification. These aspects are critical for assessing the load on switches and controllers, as well as for understanding the broader effects on traffic engineering strategies and overall system performance. Moreover, an analysis of the significance and entropy of the information contained in the first packet’s 5-tuple—specifically, identifying which fields are crucial for detecting an elephant flow at its inception—has not yet been addressed.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p16
|
39056900
|
sec[2]/p[0]
|
3. Methodology
| 1.28418 |
other
|
Other
|
[
0.033355712890625,
0.0007176399230957031,
0.9658203125
] |
[
0.0171661376953125,
0.9814453125,
0.0006589889526367188,
0.0005021095275878906
] |
Predicting the size of a flow based on its initial packet is achievable with a type of machine learning known as regression. Regression, a principal method of supervised learning, requires labeled input data to train the model for predictive tasks. The TabNet model utilized in this research is available in the GitHub repository: https://github.com/dreamquark-ai/tabnet .
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p17
|
39056900
|
sec[2]/sec[0]/p[0]
|
3.1. Training Environment
| 1.381836 |
other
|
Other
|
[
0.1422119140625,
0.0016021728515625,
0.85595703125
] |
[
0.0161590576171875,
0.98291015625,
0.0003390312194824219,
0.0004267692565917969
] |
Training was conducted on a high-performance machine equipped with the following specifications: Memory: 128 GB RAM Graphics Processing Unit (GPU): NVIDIA GeForce RTX 4090 with 24 GB of VRAM Central Processing Unit (CPU): Intel Core i9-13900KF with 24 cores
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39056900_p18
|
39056900
|
sec[2]/sec[0]/p[1]
|
3.1. Training Environment
| 2.550781 |
biomedical
|
Study
|
[
0.62255859375,
0.0010690689086914062,
0.37646484375
] |
[
0.89453125,
0.1041259765625,
0.0009517669677734375,
0.0005173683166503906
] |
These resources were more than sufficient to conduct the training and validation processes. In fact, the computational resources were not consumed beyond 20%, even when using the most demanding hyperparameter combinations. Training times for a single epoch ranged from as short as 10 s to as long as 2 min, depending on the complexity of the hyperparameter configurations. This ample capacity ensured efficient handling of the large dataset and complex computations involved in training the TabNet model, facilitating timely convergence and optimal performance.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p19
|
39056900
|
sec[2]/sec[0]/p[2]
|
3.1. Training Environment
| 1.513672 |
other
|
Other
|
[
0.061859130859375,
0.0009441375732421875,
0.93701171875
] |
[
0.1463623046875,
0.85107421875,
0.0012378692626953125,
0.0012044906616210938
] |
Inference, which involves using the trained model to make predictions on a validation dataset, was also highly efficient. Inference latency varied from 10 to 23 s, depending on the model complexity and the size of the validation dataset (maximum 1,303,496; minimum 130,349). This capability is crucial for practical deployment in high-speed network environments where timely decision-making is essential.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39056900_p20
|
39056900
|
sec[2]/sec[1]/p[0]
|
3.2. Dataset
| 1.655273 |
other
|
Study
|
[
0.3818359375,
0.001567840576171875,
0.61669921875
] |
[
0.91015625,
0.08843994140625,
0.0008044242858886719,
0.0008459091186523438
] |
The effectiveness of an ML algorithm is significantly influenced by the dataset it is trained on. In our study, we base our evaluation on data that includes length and size distributions of flows, collected from a large campus network over 30 days . For processing these data, we employed the package described in .
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39056900_p21
|
39056900
|
sec[2]/sec[1]/p[1]
|
3.2. Dataset
| 1.853516 |
other
|
Study
|
[
0.1744384765625,
0.0008420944213867188,
0.82470703125
] |
[
0.94677734375,
0.05194091796875,
0.0007443428039550781,
0.0004200935363769531
] |
The dataset in question comprises over 4 billion flows, with its complete flow records amounting to approximately 278 GB in binary format. Given this immense size, we used an anonymized subset of the data for training and evaluating our models, as published in . This subset represents one hour of traffic, encompassing 6,517,484 flows and 547 GB of data transmission. This specific time frame was chosen to ensure it was free of anomalies and that the theoretical reduction rate curve of a perfect elephant classifier for this hour closely mirrors that of the complete 30-day dataset. In the published open-source dataset, IP addresses were anonymized using the prefix-preserving Crypto-PAn algorithm . As demonstrated in , this anonymization process does not affect the performance of the ML models.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p22
|
39056900
|
sec[2]/sec[2]/p[0]
|
3.3. Input Features
| 1.399414 |
other
|
Other
|
[
0.020233154296875,
0.0004596710205078125,
0.9794921875
] |
[
0.272705078125,
0.72509765625,
0.0010442733764648438,
0.0011320114135742188
] |
The input data, sourced from the flow 5-tuple, encompasses the source IP address, destination IP address, transport layer source port, transport layer destination port, and transport layer protocol, cumulatively contributing to 104 bits. Our investigation focuses on two distinct representations of this input data: Bits : Each header field is segmented into separate bits, yielding 104 unique features, which are denoted as binary values (0 or 1). Octets : Headers that exceed 8 bits in length, such as IP addresses or ports, are segmented into distinct octets. This approach produces 13 features, with each feature represented as an 8-bit integer.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p23
|
39056900
|
sec[2]/sec[3]/p[0]
|
3.4. Balancing the Dataset
| 1.888672 |
other
|
Study
|
[
0.1787109375,
0.0006327629089355469,
0.82080078125
] |
[
0.70654296875,
0.291748046875,
0.0009140968322753906,
0.000667572021484375
] |
Achieving a balanced training dataset was key to the effectiveness of the model. In our initial training dataset of 5,213,988 flows, mouse flows greatly outnumbered elephant flows, necessitating measures to balance this disparity for optimal accuracy. The results discussed in this paper stem from the model trained on a balanced dataset, achieved through various ratios, following these steps: Define the ratio , e.g., 10%. Calculate the balanced dataset size as the size of the initial training dataset multiplied by the ratio (5,213,988 × 10% = 521,398 flows). Organize the initial training dataset in descending order, with the largest flows positioned at the start. Extract the top half of the balanced dataset size number of flows from the start of this sorted list. Randomly select the remaining half of the balanced dataset size number of flows from the rest of the initial dataset.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39056900_p24
|
39056900
|
sec[2]/sec[4]/p[0]
|
3.5. Training
| 2.193359 |
biomedical
|
Other
|
[
0.91357421875,
0.0017642974853515625,
0.08441162109375
] |
[
0.482421875,
0.5107421875,
0.005615234375,
0.00140380859375
] |
This phase encompasses the selection of hyperparameters, normalization of labels, and the model training process. The workflow of the training phase is depicted in Figure 1 .
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39056900_p25
|
39056900
|
sec[2]/sec[4]/p[1]
|
3.5. Training
| 2.310547 |
biomedical
|
Study
|
[
0.7958984375,
0.0012836456298828125,
0.20263671875
] |
[
0.97412109375,
0.0246124267578125,
0.0008935928344726562,
0.0004162788391113281
] |
The model underwent training on a shuffled, balanced training dataset before its performance was assessed using the validation dataset. Training and validation were carried out with several combinations of hyperparameters. The hyperparameters that were varied are listed in Table 1 , while those that remained unchanged throughout all training sessions are detailed in Table 2 . We also present the table with the parameters that varied and are coupled directly to the TabNet model in Table 3 .
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39056900_p26
|
39056900
|
sec[2]/sec[4]/p[2]
|
3.5. Training
| 3.726563 |
biomedical
|
Study
|
[
0.6298828125,
0.0011806488037109375,
0.368896484375
] |
[
0.9951171875,
0.004467010498046875,
0.0004754066467285156,
0.00012862682342529297
] |
The batch size was varied to observe its effect on model convergence and generalization. The selected range of 2560 to 10,240 was chosen based on several considerations: Computational Efficiency: Batch sizes in the range of 2560 to 10,240 were selected to balance between memory usage and computational efficiency. Very small batch sizes might lead to inefficient Graphics Processing Unit (GPU) utilization, while very large batch sizes could exceed the memory limits of the hardware, leading to slower training times due to increased paging or the need to reduce model complexity. Empirical Performance: Preliminary experiments indicated that this range of batch sizes yielded good performance across various metrics. A batch size of 2560 provided a good trade-off between frequent weight updates and manageable noise in gradient estimates. Increasing the batch size to 5120 and 10,240 allowed for more stable training with slightly slower convergence, which was beneficial in achieving better generalization on the validation set. Model and Data Characteristics: The nature of the dataset and the model architecture also influenced the choice of batch size. Given the large dataset (5,213,988 flows) and the complexity of the TabNet architecture, batch sizes within this range were found to be effective in leveraging the computational capabilities of modern GPUs while ensuring efficient training dynamics.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p27
|
39056900
|
sec[2]/sec[4]/p[3]
|
3.5. Training
| 3.822266 |
biomedical
|
Study
|
[
0.7041015625,
0.0008273124694824219,
0.295166015625
] |
[
0.9970703125,
0.0025119781494140625,
0.0005002021789550781,
0.0000807046890258789
] |
Different learning rates were tested to find the optimal balance between convergence speed and stability. A lower learning rate (1 × 10 − 3 ) allows for finer weight adjustments, potentially reducing the risk of overshooting minima. Higher learning rates (6 × 10 − 3 ) can accelerate convergence but may require careful tuning to avoid instability. In general, a larger batch size can lead to more stable training by decreasing the likelihood of overfitting the model. In tandem with increasing the batch size, we also scaled the learning rate. This approach enabled the model to more quickly locate local minima and maxima without necessitating a proportional adjustment in the number of epochs. The use of different loss functions, MAE and MSE, allows us to assess their impact on regression performance. MAE is less sensitive to outliers compared to MSE, which penalizes larger errors more heavily. Specific balancing dataset ratios were chosen to ensure a sufficient number of elephant flows were included without overwhelming computational resources. A 10% ratio provides a conservative balance, while a 20% ratio allows for a more comprehensive inclusion of elephant flows. The 100% ratio indicates no rebalancing was performed, serving as a control to compare against the balanced scenarios. These steps ensure that minority classes are adequately represented, improving the model’s ability to generalize across different traffic types while keeping the computational expense manageable.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p28
|
39056900
|
sec[2]/sec[4]/p[4]
|
3.5. Training
| 2.478516 |
biomedical
|
Study
|
[
0.6376953125,
0.0012578964233398438,
0.361083984375
] |
[
0.84326171875,
0.1551513671875,
0.0011034011840820312,
0.0006785392761230469
] |
The model was trained for up to 200 epochs to ensure sufficient learning time for convergence. This duration was selected based on preliminary experiments indicating that 200 epochs allow the model to adequately learn from the data without overfitting. However, training did not always take the full 200 epochs thanks to the early stopping feature, which halted training when no significant improvement in performance was observed over a set number of epochs. The Adam optimizer was chosen for its adaptive learning rate capabilities, which can improve convergence speed and stability.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p29
|
39056900
|
sec[2]/sec[4]/p[5]
|
3.5. Training
| 2.417969 |
other
|
Study
|
[
0.3125,
0.0009236335754394531,
0.6865234375
] |
[
0.64599609375,
0.349365234375,
0.00360870361328125,
0.0009636878967285156
] |
Varying the width of this layer (8, 16, 32) allows us to investigate the impact of model capacity on performance. A wider layer can capture more complex patterns but may also increase the risk of overfitting. Similar to the decision layer, varying the width of the attention embedding (8, 16, 32) helps us understand how the model’s attention mechanism scales with complexity. Wider embeddings can capture more detailed feature interactions. The number of steps (3, 6, 9) determines how many sequential decision and attention layers the data pass through. More steps can improve model performance by allowing more complex transformations but at the cost of increased computational requirements.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p30
|
39056900
|
sec[2]/sec[4]/p[6]
|
3.5. Training
| 2.826172 |
biomedical
|
Study
|
[
0.6923828125,
0.0007266998291015625,
0.306640625
] |
[
0.94873046875,
0.05059814453125,
0.0005931854248046875,
0.0003066062927246094
] |
Training and validation were conducted with various normalization techniques. We explored two distinct approaches to label normalization , designated as NONE, and MINMAX: NONE refers to the absence of label normalization. Models are trained and assessed using the unaltered labels, which vary from 64 bytes (minimum flow size) to 3,218,210,994 bytes (maximum flow size). MINMAX is a transformation where its minimum value becomes 0, its maximum value becomes 1, and all other values are scaled proportionally to fall within the range of 0 to 1. The procedure is detailed in Equation ( 1 ).
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39056900_p31
|
39056900
|
sec[2]/sec[4]/p[7]
|
3.5. Training
| 2.472656 |
biomedical
|
Other
|
[
0.51953125,
0.0010976791381835938,
0.479248046875
] |
[
0.22998046875,
0.76806640625,
0.0013341903686523438,
0.0006389617919921875
] |
Let labels = { l 1 , l 2 , … , l n } , then for each label l i in labels, the normalized value T ( l i ) is defined by: (1) T ( l i ) = l i − l m i n l m a x − l m i n
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
ca
| 0.999999 |
39056900_p32
|
39056900
|
sec[2]/sec[5]/p[0]
|
3.6. Model
| 1.623047 |
other
|
Other
|
[
0.0202178955078125,
0.0004570484161376953,
0.9794921875
] |
[
0.01306915283203125,
0.984375,
0.0020313262939453125,
0.0004291534423828125
] |
The TabNet model is a type of neural network architecture designed specifically for tabular data. Developed by researchers at Google Cloud AI, TabNet uses sequential attention mechanisms to selectively choose which features to process at each decision step, effectively enabling the model to make decisions based on important, learned features from the data. This selective feature processing allows TabNet to interpret and learn from the data in a way similar to how decision trees isolate important features but with the added flexibility and power of a neural network. TabNet’s design also promotes interpretable decision-making, which is a valuable attribute for applications requiring transparency in how input features affect predictions. This model has been shown to perform competitively on various benchmark datasets, outperforming traditional ensemble models like random forests and gradient-boosting machines in some cases.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
39056900_p33
|
39056900
|
sec[2]/sec[6]/p[0]
|
3.7. Model Decision
| 1.954102 |
other
|
Study
|
[
0.3076171875,
0.001010894775390625,
0.69140625
] |
[
0.7880859375,
0.2103271484375,
0.000858306884765625,
0.0007200241088867188
] |
In regression analysis, the algorithm predicts a continuous outcome, which, for our study, corresponds to the anticipated flow size in bytes. To illustrate the relationship between flow table reduction and traffic coverage, retraining and refitting the model repeatedly is unnecessary. We can simulate decision-making adjustments by modifying the threshold for classifying a flow as an elephant based on its predicted size. Here, the term label denotes the true flow size as extracted from the dataset.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p34
|
39056900
|
sec[2]/sec[7]/p[0]
|
3.8. Evaluation
| 1.222656 |
other
|
Study
|
[
0.006061553955078125,
0.00037479400634765625,
0.99365234375
] |
[
0.61474609375,
0.344482421875,
0.03839111328125,
0.0025119781494140625
] |
Current research in the field largely neglects metrics essential for assessing the effectiveness of flow-based traffic engineering. Many studies emphasize the accuracy of flow classification, measuring success through parameters such as the true positive rate, true negative rate, and precision in predicting flow size and duration. Yet, these metrics offer limited insights into the practical implementation of algorithms in this research area. Crucially, the misclassification of a network’s largest flow disproportionately affects overall traffic coverage compared to the misclassification of smaller flows. The metrics commonly used in existing literature fail to capture this significant disparity. Apart from the metrics, it is also unknown which information is the most important for predicting which flow belongs to which class .
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p35
|
39056900
|
sec[2]/sec[7]/p[1]
|
3.8. Evaluation
| 1.205078 |
other
|
Study
|
[
0.01568603515625,
0.0005769729614257812,
0.98388671875
] |
[
0.5810546875,
0.41455078125,
0.002361297607421875,
0.0019273757934570312
] |
To bridge these gaps, we introduce new metrics specifically designed to evaluate ML models in the context of detecting elephant flows for traffic engineering purposes. We employ two particular metrics for this evaluation: the reduction in the number of flow table entries created and the percentage of traffic covered . These metrics are intended to provide a more relevant assessment of how well the models perform in practical traffic management scenarios, focusing on optimizing network efficiency and capacity utilization.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p36
|
39056900
|
sec[2]/sec[7]/p[2]
|
3.8. Evaluation
| 1.132813 |
other
|
Other
|
[
0.00940704345703125,
0.0006933212280273438,
0.98974609375
] |
[
0.009796142578125,
0.98876953125,
0.0008177757263183594,
0.0004892349243164062
] |
It is important to understand the inherent trade-off between these metrics. Increasing the threshold for elephant flow detection results in a larger reduction in the number of flow table entries, but it also diminishes the percentage of traffic that is covered. Striking the right balance between these factors is crucial for optimizing network efficiency and maintaining high QoS.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39056900_p37
|
39056900
|
sec[2]/sec[7]/p[3]
|
3.8. Evaluation
| 2.320313 |
biomedical
|
Study
|
[
0.865234375,
0.0017957687377929688,
0.1328125
] |
[
0.9951171875,
0.00429534912109375,
0.0005140304565429688,
0.00023031234741210938
] |
Additionally, we assess the information contained in the input 5-tuple (entropy) and its significance in influencing the output predictions’ (feature importance) . Our analysis explores the impact of information across the two proposed input data approaches. This study aims to provide a more in-depth understanding of which elements of the input data are more relevant than others.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39056900_p38
|
39056900
|
sec[3]/p[0]
|
4. Results
| 1.46582 |
other
|
Study
|
[
0.07623291015625,
0.0008220672607421875,
0.9228515625
] |
[
0.705078125,
0.2919921875,
0.001667022705078125,
0.0013608932495117188
] |
Out of 504 distinct results (two input data types, three dataset ratios, seven TabNet hyperparameter combinations, three batch size and learning rate combinations, two loss functions, and two normalization types) we selected the best result per input data type and dataset ratio. In this research, the best means the largest flow table reduction at 80% traffic coverage.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
39056900_p39
|
39056900
|
sec[3]/sec[0]/p[0]
|
4.1. Flow Table Reduction vs. Traffic Coverage
| 1.117188 |
other
|
Other
|
[
0.01273345947265625,
0.0007562637329101562,
0.986328125
] |
[
0.021697998046875,
0.9755859375,
0.0016689300537109375,
0.0008778572082519531
] |
The visual representations illustrate the reduction in flow table operations and achieved traffic coverage. Remarkably, the y-axis exhibits a logarithmic scale. On the y-axis, each unit corresponds to a multiplier . The goal is to minimize creation of individual flow entries while preserving optimal traffic coverage. A model is deemed more effective as its curve approaches the top-right corner of the graph.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39056900_p40
|
39056900
|
sec[3]/sec[0]/p[1]
|
4.1. Flow Table Reduction vs. Traffic Coverage
| 1.438477 |
other
|
Other
|
[
0.050445556640625,
0.0006842613220214844,
0.94873046875
] |
[
0.294677734375,
0.70263671875,
0.0013751983642578125,
0.0011091232299804688
] |
The black line, identified as Data , illustrates the projected performance derived from the validation dataset, comprising 1,303,496 flows. This projection is predicated on the assumption of perfect prediction of each flow’s size on its initial packet. This methodology, described in as the first method, involves selecting the smallest subset of the largest flows, arranged by size in descending order, which collectively represent a predetermined percentage of the total network traffic.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p41
|
39056900
|
sec[3]/sec[0]/p[2]
|
4.1. Flow Table Reduction vs. Traffic Coverage
| 1.542969 |
other
|
Study
|
[
0.088623046875,
0.0006394386291503906,
0.91064453125
] |
[
0.8701171875,
0.1279296875,
0.0013341903686523438,
0.00077056884765625
] |
Figure 2 , Figure 3 and Figure 4 present results for bit vector input data representation and the balanced dataset with ratios of 10%, 20%, and 100%, whereas Figure 5 , Figure 6 and Figure 7 present results for octet input data representation and the balanced dataset with ratios of 10%, 20%, and 100%. Additionally, as seen in Figure 4 and Figure 7 we were unable to draw the reduction vs. coverage result for the MAE with MINMAX normalization type, due to the fact that obtained results did not fit in the traffic coverage area of interest (50–100%). It seems that the model in these configurations was extremely underfitted, and it was not able to sufficiently recognize trends and patterns based on the input data.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p42
|
39056900
|
sec[3]/sec[0]/p[3]
|
4.1. Flow Table Reduction vs. Traffic Coverage
| 1.137695 |
other
|
Other
|
[
0.0196685791015625,
0.00054931640625,
0.97998046875
] |
[
0.360595703125,
0.63525390625,
0.00254058837890625,
0.0017766952514648438
] |
In Table 4 , we presented the five top-performing configurations. The table illustrates how varying training configurations can impact the effectiveness of TabNet models in reducing created flow entries number while maintaining constant 80% traffic coverage.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
39056900_p43
|
39056900
|
sec[3]/sec[1]/p[0]
|
4.2. Feature Entropy and Importance
| 2.255859 |
biomedical
|
Study
|
[
0.90185546875,
0.0009598731994628906,
0.0970458984375
] |
[
0.9892578125,
0.009979248046875,
0.0006461143493652344,
0.0001977682113647461
] |
To provide additional insight into which features are most essential for providing an accurate prediction, we performed an analysis of the information amount contained in the input 5-tuple (feature entropy) and its significance in influencing the model (feature importance). The results of the analysis are presented in Figure 8 and Figure 9 .
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999994 |
39056900_p44
|
39056900
|
sec[3]/sec[1]/p[1]
|
4.2. Feature Entropy and Importance
| 3.222656 |
biomedical
|
Study
|
[
0.86962890625,
0.0005087852478027344,
0.1298828125
] |
[
0.955078125,
0.0438232421875,
0.0007419586181640625,
0.00019979476928710938
] |
Feature importance analysis was performed for the all input data variations and all dataset balancing ratios. In entropy analysis, we calculated the entropy for both input data representations. In this context, entropy measures the average amount of information contained in a feature (byte or bit, depending on the input data representation). Higher entropy indicates greater randomness, while lower entropy indicates less varied values. We express the entropy in bits. This tells us how many bits on average are needed to encode the information contained in a particular feature.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p45
|
39056900
|
sec[4]/p[0]
|
5. Discussion
| 3.574219 |
other
|
Study
|
[
0.409912109375,
0.000835418701171875,
0.58935546875
] |
[
0.99267578125,
0.006465911865234375,
0.000789642333984375,
0.00014495849609375
] |
The flow table reduction results show the superiority of MSE over the MAE loss functions. MSE employs an error amplification mechanism. For larger errors, the squared term magnifies their impact, which accelerates the minimization process during training. MSE amplifies the influence of outliers, which seems to fit our scenario much better than the MAE, which, on the other hand, treats all errors equally, minimizing the impact of outliers on the loss function. Additionally, as can be seen in the reduction results, TabNet worked much better on unnormalized labels rather than normalized labels. Regarding the TabNet parameters, the best results were obtained with the width (both the decision prediction layer and attention embedding for each mask) set to 8. The best reduction rate achieved for the 80% traffic coverage was 20.14. As shown in Figure 10 this is a 25% higher reduction rate than achieved previously with neural networks comprising solely linear layers, which provided only 15-fold reduction for the best parameter combination .
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
39056900_p46
|
39056900
|
sec[4]/p[1]
|
5. Discussion
| 1.177734 |
other
|
Other
|
[
0.0162506103515625,
0.0005903244018554688,
0.9833984375
] |
[
0.1285400390625,
0.86865234375,
0.0014238357543945312,
0.0011682510375976562
] |
Feature entropy analysis shows that the most predictable fields are related to the transport protocol, and source port in both input data representations. This is expected, as the transport protocol field contains mostly one of the two values: 6 for TCP and 17 for UDP. The least predictable (most random) fields are the addresses (both source and destination) and destination port.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
39056900_p47
|
39056900
|
sec[4]/p[2]
|
5. Discussion
| 1.325195 |
other
|
Study
|
[
0.0269317626953125,
0.00055694580078125,
0.97265625
] |
[
0.68017578125,
0.31689453125,
0.0019063949584960938,
0.0013256072998046875
] |
As the results show, only a fraction of initial input data is significant for the model in predicting the flow size. Features are also not equally important across dataset balancing ratios. In the octets data, different features like transport protocol, ports, and addresses dominate depending on the ratio. For the 100% ratio, the transport protocol has significantly higher weight than the other features. For the 20% ratio, the source and destination ports are the most important, while for the 10% ratio, the source and destination addresses are the most important. Conversely, in the bit vector data, the transport protocol and destination port are consistently important, while the source and destination addresses are not, and the source port’s importance declines at lower ratios.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p48
|
39056900
|
sec[4]/p[3]
|
5. Discussion
| 1.583984 |
other
|
Study
|
[
0.07293701171875,
0.0007238388061523438,
0.92626953125
] |
[
0.61669921875,
0.3798828125,
0.002437591552734375,
0.0011949539184570312
] |
The variation in feature importance across different dataset ratios can be attributed to the nature of the balancing process itself. At higher ratios, where the dataset is more imbalanced, the model may rely heavily on more generalized features such as transport protocols that are universally present in all flows. However, at lower ratios, where the dataset is more balanced, the model can discern more nuanced patterns and dependencies, leading to a higher significance of specific features like source and destination addresses. This deeper exploration reveals that feature importance is inherently tied to the composition and characteristics of the training data, impacting the model’s predictive behavior depending on the dataset’s balance.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
39056900_p49
|
39056900
|
sec[5]/p[0]
|
6. Conclusions
| 1.989258 |
other
|
Study
|
[
0.0748291015625,
0.0007886886596679688,
0.92431640625
] |
[
0.978515625,
0.02056884765625,
0.0007004737854003906,
0.0003590583801269531
] |
As demonstrated in this study, employing a TabNet model to identify elephant flows from the initial packets enables a reduction in the number of flow table entries by approximately 20-fold while still encompassing 80% of the traffic. The reduction in number of required flow table entries can not only enable flow-based traffic engineering on switches with limited capacities but also positively influence flow table lookup, consequently enhancing the switching rate. We also evaluated the significance of the information carried by the initial packet 5-tuple. It was determined that only a subset of all features is truly important for the model in providing accurate results. Utilizing this subset of the input data, one can achieve faster training and inference time, which can result in quicker elephant flow classification and minimization of the additional latency.
|
[
"Bartosz Kądziołka",
"Piotr Jurkiewicz",
"Robert Wójcik",
"Jerzy Domżał"
] |
https://doi.org/10.3390/e26070537
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276374_p0
|
PMC11276374
|
sec[0]/p[0]
|
1. Introduction
| 3.931641 |
biomedical
|
Study
|
[
0.99951171875,
0.00023174285888671875,
0.0002868175506591797
] |
[
0.8837890625,
0.0273895263671875,
0.08837890625,
0.0005602836608886719
] |
Extremely-low-birth-weight (ELBW) infants refers to newborns whose birth weights are less than 1000 g. ELBW preterm infants have less well-developed systems than other low-birth-weight (LBW) preterm infants, and their poor immune systems make them more susceptible to infections and other preterm complications, often involving the nervous system, which increase the risk of cerebral palsy, intellectual disability, mission, and deafness . The mortality rate of premature infants with ELBW is high. It was reported that the probability of ELBW premature infants dying from clinical complications (such as necrotizing enterocolitis and sepsis) is as high as 23% .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p1
|
PMC11276374
|
sec[0]/p[1]
|
1. Introduction
| 4.695313 |
biomedical
|
Study
|
[
0.99853515625,
0.0008821487426757812,
0.0008206367492675781
] |
[
0.79296875,
0.0015344619750976562,
0.2047119140625,
0.000652313232421875
] |
The gut microbiota begins to colonize the gastrointestinal tract at birth and plays an important role in the growth and development of newborns in the early stages of life and beyond. However, the diversity of gut microbiota is low in early neonatal life, and the structure of the gut microbiota is influenced by a variety of factors, including the mode of delivery, gestational age, birth weight, feeding method, and the environment . The mode of delivery is one of the most important determinants of gut microbiota composition . In vaginally delivered newborns, the abundance of Bacteroidetes is higher, while in cesarean-delivered newborns, Klebsiella and Haemophilus are the dominant species 6. Studies have shown that gestational age and birth weight are the most important factors influencing differences in intestinal microecology. Preterm infants have a unique gut microbiota in the early postnatal period , which is dominated by conditionally pathogenic bacteria, such as Staphylococci , Enterococci , and Enterobacteria , and beneficial bacteria such as Bifidobacteria do not exist as dominant species . Most LBW preterm infants are transferred to a neonatal intensive care unit (NICU) after birth to be maintained on respiratory support equipment because of respiratory distress or other reasons. The gut microbiome colonization in LBW preterm infants can also be influenced by the NICU’s ambient settings and the usage of appropriate equipment. Extended respiratory support in preterm infants can lead to an increase in intestinal aerobic and facultative anaerobic bacteria . Gut microbiota genera in LBW preterm infants in the NICU are dominated by Klebsiella , Enterobacter , and Enterococci , and differences among the gut microbiota decrease with an increase in hospitalization time . An other significant element influencing the gut microbiota makeup in preterm newborns is feeding method. Breast-fed and non-breast-fed infants have different gut microbiota . However, breastfeeding can help premature infants’ immune systems mature and encourage the colonization of intestinal bacteria Bifidobacterium . The maternal diet can also affect the composition of the infant’s gut microbiota . For instance, if the mother consumes plant-based protein or a high-fat diet, it can lead to a significant reduction in the presence of Bacteroides bacteria in the newborn’s gut, and the decrease in Bacteroides may affect the early-immune and metabolic development of newborns . In addition, the use of antibiotics also has a certain impact on the composition of gut microbiota in premature infants. Antibiotics can reduce the diversity of gut microbiota and delay the colonization of Bifidobacterium . The community state type (CST) is based on the gut microbiota abundance obtained from sequencing analysis and classified into different CSTs by clustering . There are also variations in the types of gut microbiota-community states among infants of different age groups. In healthy infants under 6 months old, the gut microbiota CSTs are mainly characterized by a higher abundance of Bifidobacterium , while in infants aged 12 to 36 months typical adult bacterial genera such as Bacteroides and Faecalibacterium predominate . It can be seen that, as the newborn grows and develops, the composition of gut microbiota in the body also undergoes dynamic changes.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276374_p2
|
PMC11276374
|
sec[0]/p[2]
|
1. Introduction
| 4.324219 |
biomedical
|
Study
|
[
0.99951171875,
0.00043392181396484375,
0.0001653432846069336
] |
[
0.99609375,
0.0004191398620605469,
0.0032806396484375,
0.00015544891357421875
] |
Current research has found that preterm infants, because of their prolonged exposure to the NICU environment and the relatively frequent clinical interventions such as respiratory support and antibiotic use they experience, undergo changes in their gut microbiota composition, making them more susceptible to conditions like NEC and late-onset sepsis (LOS) . Supplementing the food of early-stage newborns with probiotics such as Bifidobacterium can promote the colonization of the intestine by beneficial bacteria, thereby preventing or reducing the occurrence of NEC, LOS, and feeding intolerance . Probiotic supplementation improves gut microbial composition, making it closer to that of full-term infants, which is beneficial for promoting immunity and metabolism . Probiotic-supplemented ELBW preterm newborns had low levels of harmful bacteria and a substantial increase in the gut bacterial Bifidobacterium . The results showed that the abundance of Bifidobacterium in the intestinal bacteria of preterm infants of ELBW who received probiotic supplementation was significantly higher than those who did not receive probiotics, and the abundance of pathogenic bacteria was lower. Simultaneously, preterm infants who received probiotic supplementation had higher levels of acetate and lactate (end products of HMO metabolism), and the abundance of acetate was positively correlated with the abundance of Bifidobacterium . At the same time, the gut microbiota diversity of ELBW preterm infants who received probiotic Lactobacillus supplementation increased, and the abundance of the supplemented probiotics also rose. Compared with the control-group infants, ELBW preterm infants who received probiotic supplementation had reduced abundances of Staphylococcaceae and Enterobacteriaceae in their intestines . It can be seen that probiotic supplementation for preterm infants can facilitate colonization of the intestine by beneficial bacteria and reduce harmful bacteria. Probiotics can also promote the metabolism of HMO in breast milk, enabling the beneficial metabolites in HMO to exert their immune-enhancing effects.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p3
|
PMC11276374
|
sec[0]/p[3]
|
1. Introduction
| 4.164063 |
biomedical
|
Study
|
[
0.99951171875,
0.00046181678771972656,
0.0001876354217529297
] |
[
0.9990234375,
0.00020170211791992188,
0.0008158683776855469,
0.00011116266250610352
] |
Most studies on intestinal microbiota have focused on full-term infants; however, the health outcomes of ELBW and VLBW preterm infants are equally important. Because of their immature systemic physiology and immature intestinal microbiota structure, they may be predisposed to long-term outcomes such as neurodevelopmental disorders . Studies have found that there is a correlation between gut microbiota and brain function. A study established a connection between the gut microbiota, immunology, and neurodevelopment in extremely-preterm infants and discovered that excessive growth of the intestinal microbiota can be a strong predictor of brain injury. Abnormal development of the gut-microbiota–immune-system–brain axis may drive or exacerbate brain injury in extremely-preterm infants . The underlying mechanisms of these effects have not been fully elucidated, and some have not even been considered. Therefore, this study aimed to investigate the gut microbiota structure of preterm infants with LBW using 16S rRNA gene sequencing technology. We analyzed the gut microbiota structure, corresponding microbiota profiles, and the CST of the gut microbiota among preterm infants of different birth weights. Correlation analysis of the CST and clinical indicators of preterm infants was conducted, and the clinical value of the intestinal microbiota in diagnosing extremely-LBW preterm infants was evaluated.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276374_p4
|
PMC11276374
|
sec[1]/sec[0]/p[0]
|
2.1. Participant Enrollment and Sample Collection
| 4.117188 |
biomedical
|
Study
|
[
0.9970703125,
0.0025634765625,
0.000270843505859375
] |
[
0.99853515625,
0.0010051727294921875,
0.00022399425506591797,
0.0002422332763671875
] |
This study included a total of 98 fecal samples from 39 preterm infants with LBW. Inclusion criteria: premature infants hospitalized in the NICU of the neonatology department; gestational age at birth of <37 weeks and a birth weight of <2500 g; hospitalization time > 7 days. Exclusion criteria: neonates with a gestational age at birth of ≥37 weeks and a birth weight of ≥2500 g; hospitalization time < 7 days; premature infants with severe congenital heart disease and severe digestive tract malformation who need surgery; premature infants with Down syndrome, hereditary metabolic diseases and severe asphyxia; stillbirths, induced abortions, combined with severe cardiac and renal dysfunction. We selected the first stool sample of NICU low-birth-weight premature infants who met the inclusion criteria, then planned to collect fecal samples every 2 weeks until discharge or until the 8th week of collection. Finally, the preterm infants were divided into three groups based on their birth weights: ELBW , VLBW , and LBW .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276374_p5
|
PMC11276374
|
sec[1]/sec[0]/p[1]
|
2.1. Participant Enrollment and Sample Collection
| 2.677734 |
biomedical
|
Study
|
[
0.9970703125,
0.0021228790283203125,
0.0008625984191894531
] |
[
0.94775390625,
0.050445556640625,
0.0007882118225097656,
0.0009984970092773438
] |
The guardians of the participants collected fecal samples in sterile containers and transported them overnight on ice to the laboratory. The researchers immediately aliquoted the samples into tubes containing 3–5 g each and stored them in a −80 °C freezer. The research protocol of this study was approved by the hospital’s medical ethics committee, and each neonate’s parents provided written informed consent. The research protocol was designed in compliance with the Helsinki Declaration and approved by the hospital’s medical ethics committee.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p6
|
PMC11276374
|
sec[1]/sec[1]/p[0]
|
2.2. Analysis of Gut Microbiota
| 2.845703 |
biomedical
|
Study
|
[
0.99755859375,
0.00043463706970214844,
0.0020389556884765625
] |
[
0.84912109375,
0.1431884765625,
0.006809234619140625,
0.0009341239929199219
] |
Refer to our published articles for detailed methodology on 16S rRNA gene sequencing and bioinformatic analyses (detailed in the Supplementary Materials ).
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p7
|
PMC11276374
|
sec[1]/sec[2]/p[0]
|
2.3. Analysis of Ecological Diversity Indices
| 2.5625 |
biomedical
|
Study
|
[
0.9853515625,
0.0004837512969970703,
0.0142974853515625
] |
[
0.60107421875,
0.396484375,
0.0017414093017578125,
0.0007758140563964844
] |
The diversity function from the R package Vegan (version 2.6-4) was used to calculate the Shannon and Inverse Simpson indices for the samples. The estimateR function from the R package Vegan was used to calculate the richness index for the samples.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276374_p8
|
PMC11276374
|
sec[1]/sec[3]/p[0]
|
2.4. Stacked Bar Chart, Chord Diagram, Venn Plot, Volcano Plot, Manhattan Plot
| 2.251953 |
biomedical
|
Other
|
[
0.9833984375,
0.0007791519165039062,
0.0158233642578125
] |
[
0.370849609375,
0.625,
0.00299072265625,
0.001132965087890625
] |
The processes used to obtain the stacked bar charts, chord diagrams, Venn plots, volcano plots, and Manhattan plots were completed by referring to the EasyAmplicon protocol .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276374_p9
|
PMC11276374
|
sec[1]/sec[4]/p[0]
|
2.5. Constrained Principal Coordinates Analysis
| 2.716797 |
biomedical
|
Other
|
[
0.88037109375,
0.0008993148803710938,
0.11859130859375
] |
[
0.09429931640625,
0.90380859375,
0.0017004013061523438,
0.0003428459167480469
] |
Constrained Principal Coordinate Analysis (CPCoA) refers to the addition of grouping information to the Principal Coordinate Analysis (PCoA) in order to find a plane that can best explain the differences between groups under self-defined grouping conditions. The process was completed by referring to the EASYAMPLICON protocol .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999995 |
PMC11276374_p10
|
PMC11276374
|
sec[1]/sec[5]/p[0]
|
2.6. Gut-Microbiota Network Analysis
| 0.956543 |
biomedical
|
Other
|
[
0.828125,
0.0029296875,
0.168701171875
] |
[
0.01442718505859375,
0.98095703125,
0.003437042236328125,
0.0010538101196289062
] |
The layout and visualization of the gut microbiota network diagram were completed with reference to the article published by the Zhou Jizhong Team, Li Ji Team, and Shen Qirong Team .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276374_p11
|
PMC11276374
|
sec[1]/sec[6]/p[0]
|
2.7. Analysis of Microbial Community Structure
| 3.734375 |
biomedical
|
Study
|
[
0.9990234375,
0.0001646280288696289,
0.0007643699645996094
] |
[
0.97802734375,
0.0211944580078125,
0.0008540153503417969,
0.00016558170318603516
] |
Gap statistics were used to determine the optimal number of clusters in the microbial community structure. This method identifies the best number of clusters by comparing the distribution of clustered data with that of a random distribution through the calculation of the gap (or “gap statistic”) between them.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11276374_p12
|
PMC11276374
|
sec[1]/sec[7]/p[0]
|
2.8. Non-Metric Multidimensional Scaling
| 4.050781 |
biomedical
|
Study
|
[
0.9990234375,
0.0001608133316040039,
0.0006127357482910156
] |
[
0.9990234375,
0.000518798828125,
0.0003268718719482422,
0.00004297494888305664
] |
Non-metric multidimensional scaling (NMDS) was completed with reference to the authors’ previously published research . Firstly, based on the genus-level data, the metaMDS function in the R package Vegan (version 2.6-4) was used to conduct NMDS ordination analysis and obtain the stress value. Simultaneously, the adonis2 function in the R package Vegan was employed to conduct a permutational multivariate analysis of variance (PERMANOVA) based on Bray–Curtis distance, yielding p -values and R 2 values. The ordisurf function in the R package Vegan was used to passively add environmental variables to the NMDS ordination. Finally, the geom_point function in the R package ggplot2 (version 3.3.2) was employed to visualize the results of the NMDS ordination.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276374_p13
|
PMC11276374
|
sec[1]/sec[8]/p[0]
|
2.9. Random Forest Analysis
| 2.158203 |
biomedical
|
Study
|
[
0.98681640625,
0.0008292198181152344,
0.01218414306640625
] |
[
0.8671875,
0.1234130859375,
0.0083465576171875,
0.0009298324584960938
] |
Refer to previously published articles for detailed methodology on the random forest analysis (detailed in the Supplementary Materials ).
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276374_p14
|
PMC11276374
|
sec[1]/sec[9]/p[0]
|
2.10. Other Analyses
| 4.011719 |
biomedical
|
Study
|
[
0.99951171875,
0.00017261505126953125,
0.0001952648162841797
] |
[
0.9990234375,
0.0006008148193359375,
0.0003986358642578125,
0.00006246566772460938
] |
To evaluate the correlation between the significantly different gut microbiota compositions between groups and clinical manifestations, the lm function in R software (version 4.2.3) was used to construct a logistic regression model. The p -value and coefficient of determination (R-squared) of the logistic regression model were obtained through the summary function. The beeswarm function in the R package beeswarm (version 0.4.0) was used to create boxplots, and the wilcox.test function from the R package stats (version 4.2.3) was used for statistical testing to obtain p -values. The visualization of clinical data and other aspects were completed using customized scripts.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276374_p15
|
PMC11276374
|
sec[2]/p[0]
|
3. Results
| 4.171875 |
biomedical
|
Study
|
[
0.99951171875,
0.0004673004150390625,
0.00023353099822998047
] |
[
0.99951171875,
0.00019407272338867188,
0.000347137451171875,
0.00008213520050048828
] |
This study included 98 fecal samples from 39 preterm infants. We conducted a visual analysis of clinical data on premature infants, and the results are shown in Figure 1 A. To determine the saturation of sequencing data for the 16s rRNA gene, that is, whether the number of sequencing data were sufficient, we performed saturation curve analysis based on species richness, and the results are shown in Figure 1 B. It can be seen that the saturation curves for ELBW, LBW, and VLBW all tended to saturate, indicating that the 16s rRNA gene sequencing data were sufficient. At the same time, the species richness in the LBW group was slightly higher than those of ELBW and VLBW infants. We used ANOSIM, which stands for analysis of similarities, to compare the similarity of the gut microbiota composition data among ELBW, LBW, and VLBW infants. As a non-parametric test method, ANOSIM is often used to test for the similarities among high-dimensional data. We also compared the magnitude of differences in gut microbiota compositions both between and within the groups of ELBW, LBW, and VLBW infants, and the results are presented in Figure 1 C. The R-value of 0.0418 indicated the presence of a certain degree of difference both within and between the groups. The p -value of 0.043 suggested that this difference was restricted. To further understand the shared and unique gut microbiota profiles among ELBW, LBW, and VLBW infants, and to visually demonstrate the overlaps in gut microbiota among the three groups, we conducted an analysis using a Venn plot and found that the number of OTUs shared among the three groups was 118, indicating that the majority of gut microbiota were common to ELBW, LBW, and VLBW infants .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276374_p16
|
PMC11276374
|
sec[2]/p[1]
|
3. Results
| 4.207031 |
biomedical
|
Study
|
[
0.99951171875,
0.0004074573516845703,
0.00022459030151367188
] |
[
0.9990234375,
0.00016236305236816406,
0.0005335807800292969,
0.00008815526962280273
] |
To further understand the gut microbiota composition of preterm infants, we analyzed the gut microbiota at the genus level . The results showed that the gut microbiota of the ELBW group was dominated by Enterococcus , followed by Staphylococcus , Acinetobacter , and Klebsiella . The gut microbiota of the VLBW group was primarily composed of Klebsiella , followed by Enterococcus , Staphylococcus , Streptococcus , Acinetobacter , and Pseudescherichia . In the LBW group, Enterococcus , Staphylococcus , Klebsiella , and Streptococcus were the main gut microbiota genera, followed by Bifidobacterium and Pseudescherichia . Compared with those of the LBW group, the ELBW and VLBW groups’ abundances of Acinetobacter were significantly increased, with a notable increase observed in the ELBW group. Conversely, the abundance of Bifidobacterium was significantly reduced. We employed CPCoA to compare the differences in the gut microbiota composition among the ELBW, LBW, and VLBW groups of infants. The results showed that the grouping could explain 2.65% of the variation, and the separation was relatively distinct, indicating that grouping had a certain influence on the composition of gut microbiota . We further analyzed the gut microbiota of preterm infants in the ELBW, LBW, and VLBW groups by NMDS clustering at the genus level. The group data were calculated via the Bray–Curtis index to generate NMDS to visualize the similarity of the gut microbiota. In Figure 1 H, each point in the graph represents the microbiota characteristics of an individual preterm infant in a low-dimensional space. The results showed that there were distinct clusters of gut microbiota genera among the three groups, indicating significant differences in their distribution (R 2 = 0.041, p = 0.001). Simultaneously, we conducted clustering analysis based on the clinical phenotypes of the three groups of preterm infants. The results showed significant differences in gestational age and birth weight among the three groups .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p17
|
PMC11276374
|
sec[2]/p[2]
|
3. Results
| 4.167969 |
biomedical
|
Study
|
[
0.99951171875,
0.00033926963806152344,
0.0002143383026123047
] |
[
0.99951171875,
0.0002052783966064453,
0.0004000663757324219,
0.00007718801498413086
] |
To further understand whether the gut microbiota components were differentially distributed among the ELBW, LBW, and VLBW groups, we performed an analysis of gut microbiota at the genus level by volcano plots. The results showed that 118 genera with differential abundances were identified between ELBW and LBW at the genus level. Among them, 56 genera were less abundant in ELBW, while 62 genera were more abundant in the ELBW than in the LBW group . Compared with VLBW infants, ELBW infants exhibited a total of 83 differentially abundant genera of gut microbiota at the genus level, with 44 genera showing lower, and 39 genera showing higher, abundances compared with those in the VLBW group . A total of 67 differentially abundant genus-level enterobacteria were identified in VLBW infants compared with the findings in LBW infants, with 36 genera less abundant and 31 genera more abundant than in the LBW group . We further specifically analyzed these differentially abundant gut microbiota through Manhattan plots. The results showed that, compared with the LBW group, the ELBW group exhibited more Enterococcus , Streptococcus , and Acinetobacter , but lower amounts of Klebsiella . Enterococcus , Streptococcus , and Clostridium sensu stricto abundances were predominantly lower, and that of Enterobacter was predominantly higher, in ELBW compared with the findings in VLBW. Compared with LBW infants, VLBW infants showed more Acinetobacter and less Enterococcus and Klebsiella .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p18
|
PMC11276374
|
sec[2]/p[3]
|
3. Results
| 4.28125 |
biomedical
|
Study
|
[
0.9990234375,
0.0005221366882324219,
0.0002448558807373047
] |
[
0.9990234375,
0.00021135807037353516,
0.0006127357482910156,
0.00011539459228515625
] |
To further understand the interrelationships among the intestinal microbiota in each group, we employed the molecular ecological networks (MENs) method and visualization tools based on 16S rRNA high-throughput sequencing. The results showed that the gut microbial interaction network of the VLBW group consisted of 416 nodes (ASVs) and 8856 links (interactions). In the network constructed for the ELBW group, more nodes were observed, but fewer links were present . Compared with non-breastfed preterm infants, breastfed preterm infants exhibited a higher number of nodes but fewer links in their gut microbiota networks. Preterm infants with jaundice had fewer nodes and even fewer links compared with those without jaundice . To further understand whether the differences between all enrolled subjects affected their corresponding gut microbiota and clinical phenotypes, for example, we used gap statistics, a clustering method based on interval statistics, and analyzed the optimal number of clusters based on the total sample size. The study subjects were grouped according to their similarities, resulting in high similarity levels within groups and significant differences between groups. Figure 3 B displays the gap statistic plots based on clustering by sample size. Based on B = 100 iterations for each k, the results showed that k = 5 was the optimal k-value, indicating that the clustering performance was basically optimal. As k continued to increase, the performance improved, relatively slowly. Therefore, the final clustering algorithm was chosen with a k-value of 5, meaning that we grouped the samples into five clusters. We further employed NMDS to analyze the five clusters identified through the clustering analysis. By calculating the Bray–Curtis index, we generated an NMDS plot to visually display the similarities among the samples. To further understand the connection between the gut microbiota and clinical phenotypes in preterm infants with low birth weight, we first conducted an analysis of CSTs based on their gut microbiota. Through multidimensional scaling (MDS), we performed ordination analysis based on the sorting of eigenvalues and visualized the first four eigenvectors using NMDS . Then, five CST samples were visualized using the NMDS method. In Figure 3 E, each point on the plot represents the characteristics of a single sample in the low-dimensional space, and the results indicated that the five CSTs exhibited distinct clustering patterns. To understand the relationships among the gut microbiotic abundances of the five clusters identified through clustering analysis, we used a further clustering heatmap to display the variations in the abundance of key gut microbiota across the five CST samples. This allowed us to compare the compositional similarities and differences in the gut microbiota at the genus level among the different groups. The results indicated that the gut microbiota in the five CSTs was primarily composed of harmful bacteria. The six bacteria species with relatively high abundance in the gut microbiota were Enterococcus , Klebsiella , Staphylococcus , Streptococcus , Pseudescherichia , and Acinetobacter . The abundance of gut bacteria also varied among the different CSTs. Specifically, the abundances of Streptococcus and Pseudescherichia were higher in CST 1; Staphylococcus had a higher abundance in CST 2, Enterococcus was more abundant in CST 4, and Klebsiella was more prevalent in CST 5 .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276374_p19
|
PMC11276374
|
sec[2]/p[4]
|
3. Results
| 4.089844 |
biomedical
|
Study
|
[
0.9990234375,
0.0005397796630859375,
0.00029397010803222656
] |
[
0.99951171875,
0.00014281272888183594,
0.00034165382385253906,
0.00007015466690063477
] |
We further analyzed the relationships among the five CSTs and clinical phenotypes, and the results are presented in Figure 4 A. Overall, there were significant differences ( p < 0.05) between the five CSTs in terms of gestational age, parity, birth weight and length, weight and length at 1 month, weight and length at 3 months, and the percentage of neutrophils. In terms of the gestational age, birth weight, and birth length of the infants, there were significant differences between CST 3 and CST 5. In a comparison of body length at 1 month, there were significant differences between CST 1 and CST 3. In a comparison of body length at 3 months, CST 5 exhibited significant differences compared with CST 1, CST 2, and CST 3. There were also significant differences in the percentage of neutrophils between CST 4 and CST 5. In terms of parity, there was also a significant difference between CST 2 and CST 5. We further analyzed the correlation between each group and the clinical indicators . The results showed a significant positive correlation between body length at 1 month and CST 1, while there was a significant negative correlation between body length at 1 month and CST 5. Additionally, there was a significant negative correlation between CST 1 and platelet count (PLT), as well as a negative correlation between CST 4 and total bile acid (TBA).
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276374_p20
|
PMC11276374
|
sec[2]/p[5]
|
3. Results
| 4.117188 |
biomedical
|
Study
|
[
0.99951171875,
0.00035262107849121094,
0.0002281665802001953
] |
[
0.99951171875,
0.00015592575073242188,
0.00036716461181640625,
0.00006383657455444336
] |
To further explore whether there were differences in the gut microbiota between ELBW infants and other LBW infants, we initially classified the preterm infants into two groups. One was the ELBW group (ELBW+), and the other group comprised infants with VLBW and LBW, collectively known as the non-extremely-low-birth-weight group (ELBW−). According to the classification of gut microbiota under ASV conditions, the gut microbiota of the two groups of children were compared, and the results are shown in Figure 5 A. The four bacteria with significantly increased abundance in the ELBS+ group were Acinetobacter _ASV_46, Acinetobacter _ASV_49, Acinetobacter _ASV_51, and Acinetobacter _ASV_54. The abundance of intestinal bacteria Bifidobacterium _ASV_107 and Klebsiella _ASV_2 were significantly lower in the ELBS+ group of infants. To further evaluate the clinical application value of the gut microbiota, we constructed a classifier based on a random forest model, as shown in Figure 5 B,C. The top-three gut microbiota ( Klebsiella _ASV_2, Enterococcus _ASV_38, Klebsiella _ASV_11) used for preterm infant classification had an AUC value of 0.836. The AUC values for preterm infant classification using the top-5 and top-10 gut microbiota were 0.793 and 0.753, respectively. These results indicated that intestinal bacteria may be potential biomarkers for ELBW preterm infants.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276374_p21
|
PMC11276374
|
sec[3]/p[0]
|
4. Discussion
| 4.125 |
biomedical
|
Study
|
[
0.99951171875,
0.00018477439880371094,
0.0002703666687011719
] |
[
0.99267578125,
0.0003085136413574219,
0.00705718994140625,
0.00008416175842285156
] |
The CSTs can be used to discover the dominant bacterial community composition in different age groups and samples. Currently, most studies have focused on analyzing the CSTs of gut microbiota based on samples from the reproductive tract. One study categorized the female vaginal microbiota into five CSTs by 16S rRNA gene sequencing, of which CSTs I, II, III, and V were all dominated by Lactobacillus 24. A study based on adult gut microorganisms found that adult gut microorganisms can be categorized into three distinct clusters, known as enterotypes, driven by different genera of bacteria, namely Bacteroides (enterotype 1), Prevotella (enterotype 2), and Ruminococcus (enterotype 3) . A study conducted on the gut microbiota of school-age children identified three distinct enterotypes: Bacteroides , Prevotella , and Bifidobacterium .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p22
|
PMC11276374
|
sec[3]/p[1]
|
4. Discussion
| 4.140625 |
biomedical
|
Study
|
[
0.99951171875,
0.0002689361572265625,
0.0002524852752685547
] |
[
0.9990234375,
0.00014960765838623047,
0.0006380081176757812,
0.00005894899368286133
] |
There have been fewer CST analyses conducted on samples from neonates. A study focusing on the gut microbiota of infants found that infants under 6 months of age primarily had five community state types, which were dominated by the genus Bifidobacterium . There were seven main infant community state types (ICSTs) for infants aged 6–36 months; these ICSTs were characterized by typical adult bacterial genera and primarily manifested as decreased Bifidobacterium and increased Bacteroides 24. Grier and his team conducted a longitudinal CST analysis by collecting intestinal samples from preterm and full-term infants. The results revealed the existence of CSTs potentially characterized by Enterobacteriaceae , Veillonella , Ruminococcus , Streptococcus , Prevotella , Bacteroides , and Bifidobacterium . The detection of a large number of ICSTs is believed to reflect the high variability and dynamics of the microbiota during early life . In this study, CST analysis was conducted on the gut microbiota of low-weight preterm infants, and we found diverse gut microbiota compositions among the VLBW, LBW, and WELBW infants. Low-weight preterm infants exhibited five distinct CSTs, primarily characterized by Enterococcus , Staphylococcus , Klebsiella , Streptococcus , Pseudescherichia, and Acinetobacter . The primary intestinal bacteria in CST 1 were Streptococcus and Pseudescherichia . CST 2 was dominated by Staphylococcus . CST 4 was primarily made up of Enterococcus , while CST 3 and CST 5 were mainly Klebsiella . It can be seen that the CSTs of the neonatal intestine were generally dominated by opportunistic pathogens.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p23
|
PMC11276374
|
sec[3]/p[2]
|
4. Discussion
| 4.097656 |
biomedical
|
Study
|
[
0.99951171875,
0.0002720355987548828,
0.00019562244415283203
] |
[
0.9990234375,
0.00017023086547851562,
0.0005359649658203125,
0.00006198883056640625
] |
The gut microbiota of neonates is influenced by various factors, and there is a correlation between the community state types of neonatal microbiota and clinical phenotypes. The community state types of the neonatal gut microbiota also differ based on the mode of delivery. Infants delivered vaginally tend to have CSTs dominated by Bifidobacterium , while those delivered by cesarean section are more likely to have Bacteroides as the primary bacteria 24. In this study, preterm infants exhibited significant differences in gestational age, birth weight, and birth length in terms of CSTs, and there were especially significant differences between CST 3 and CST 5. There was a linear relationship between the CST and the length, PLT, and TBA of preterm infants. However, further verification is needed to determine whether there is a causal relationship between the gut microbiota and these clinical indicators.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276374_p24
|
PMC11276374
|
sec[3]/p[3]
|
4. Discussion
| 4.179688 |
biomedical
|
Study
|
[
0.99951171875,
0.00031495094299316406,
0.00018477439880371094
] |
[
0.9990234375,
0.00020575523376464844,
0.0004963874816894531,
0.00007748603820800781
] |
Particularly in preterm children, the degree of intestinal growth is immature during the neonatal era, and the gut microbiota’s structure and function varies markedly. The gut microbiota of preterm infants is often dominated by facultative anaerobic and opportunistic pathogens such as Enterobacter , Enterococcus , and Staphylococcus . In this study, we analyzed the structure of the intestinal bacteria in different low-birth-weight preterm infants. We discovered that, although the intestinal bacterial composition of preterm infants with different low-birth-weights varied, the main bacterial species were still opportunistic pathogens such as Enterococcus , Staphylococcus , Klebsiella , Streptococcus , and Acinetobacter . Compared with the VLBW and LBW groups, the ELBW group in this study exhibited a significant increase in the potentially harmful intestinal bacterial genus Acinetobacter . Acinetobacter belongs to the category of opportunistic pathogens, is also a major cause of neonatal infections and outbreaks in neonatal intensive care units (NICUs) , and can lead to the occurrence of diseases such as meningitis, bloodstream infections, and respiratory infections . Acinetobacter , one of the major drug-resistance-associated mortality pathogens, is associated with high morbidity and mortality rates, and preterm and VLBW infants are highly susceptible to infection .
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p25
|
PMC11276374
|
sec[3]/p[4]
|
4. Discussion
| 4.136719 |
biomedical
|
Study
|
[
0.99951171875,
0.0002582073211669922,
0.0001895427703857422
] |
[
0.9990234375,
0.0001825094223022461,
0.00092315673828125,
0.00008195638656616211
] |
In this study, Klebsiella was identified as a potential biomarker bacteria genus in preterm infants. The random forest analysis also indicated that Klebsiella could be a potential biomarker for diagnosing preterm infants. Klebsiella is a common intestinal microorganism during neonatal development . It can act on macrophages to thereby evade the host immune system and to persist, potentially causing opportunistic infections . Because of their gestational age and low body weight, premature infants do not yet possess fully developed or matured systems such as digestion and absorption and immune systems. Preterm infants are more prone to a series of infections, such as neonatal sepsis and necrotizing enterocolitis (NEC), because of their small gestational age, low body weight, and incomplete development of various systems, such as the digestive, absorption, and immune systems. Relevant studies have shown that Klebsiella is associated with bacterial infections and the occurrence of NEC in neonates , and elevated Klebsiella abundance is also associated with neonatal cerebral white-matter damage 40. However, further research is needed to elucidate the specific mechanisms underlying these associations and their relevance to the health of preterm infants.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11276374_p26
|
PMC11276374
|
sec[3]/p[5]
|
4. Discussion
| 4.125 |
biomedical
|
Study
|
[
0.99951171875,
0.00026679039001464844,
0.00018298625946044922
] |
[
0.9970703125,
0.0002231597900390625,
0.00274658203125,
0.00009733438491821289
] |
In this study, a significant decrease in the abundance of the intestinal probiotic Bifidobacterium was observed in ELBW preterm infants. Bifidobacterium are beneficial bacteria in the human gut with pro-inflammatory, anti-inflammatory, anti-viral, and immunomodulatory functions . Studies have found that a higher abundance of Bifidobacterium in early infancy is associated with a better immune response to vaccination and potentially enhanced immune memory . A low abundance of Bifidobacterium may lead to the development of allergies, eczema, and asthma . A study of gut microbial compositions and functions in very-preterm infants given probiotics found that Bifidobacterium can be used to predict microbial maturation and that Bifidobacterium is an important factor in accelerating gut microbial maturation 35. It showed that probiotic supplementation can promote the maturity of gut microbiota in premature infants, thus reducing differences between microbiota. In addition, related studies found that probiotic supplementation in preterm infants can reduce mortality and improve NEC and feeding intolerance, among other benefits . Evidently, changes in intestinal probiotics may affect the health of preterm infants.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276374_p27
|
PMC11276374
|
sec[3]/p[6]
|
4. Discussion
| 4.078125 |
biomedical
|
Study
|
[
0.99951171875,
0.00031566619873046875,
0.00018513202667236328
] |
[
0.9990234375,
0.00020253658294677734,
0.0004699230194091797,
0.00007390975952148438
] |
Using machine learning methods, we demonstrated the value of the gut microbiota composition in diagnosing extremely-low-birth-weight preterm infants. We assessed the clinical value of gut microbiota in ELBW preterm infants by a machine learning method, and found that the AUC values of the intestinal bacteria Klebsiella _ASV_2, Enterococcus _ASV_38, and Klebsiella _ASV_11 were 0.836. The AUC values for Klebsiella _ASV_2, Enterococcus _ASV_38, Klebsiella _ASV_11, Acinetobacter _ASV_51, and Acinetobacter _ASV_46 were found to be 0.793. The results show that the diagnosis of ELBW preterm infants based on gut bacteria is reliable, to some extent. With machine-learning analysis methods, gut bacteria may play a significant role in ELBW preterm infants, and their ROC values can predict diagnostic outcomes.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p28
|
PMC11276374
|
sec[3]/p[7]
|
4. Discussion
| 4.089844 |
biomedical
|
Study
|
[
0.99951171875,
0.0003180503845214844,
0.00021398067474365234
] |
[
0.99951171875,
0.00014090538024902344,
0.00034546852111816406,
0.00006812810897827148
] |
This study demonstrated a certain level of innovation: CST analysis is commonly used in the structural analysis of genital tract microbiota. In this study, we identified five major CSTs through an analysis of community types in low-birth-weight preterm infants, and CST was related to the clinical phenotype of premature infants. Furthermore, machine learning methods were employed to evaluate the potential of using bacteria composition in diagnosing preterm infants with ELBW. As for limitations, the 16S rRNA gene sequencing method used lacks the ability to analyze the functional composition of the gut microbiota. The study also lacked an independent validation cohort to verify the potential of using the bacteria composition in diagnosing preterm infants with ELBW. The next step will be to further investigate the functional aspects of the gut microbiota and conduct larger-scale validation studies.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276374_p29
|
PMC11276374
|
sec[4]/p[0]
|
5. Conclusions
| 3.837891 |
biomedical
|
Study
|
[
0.99951171875,
0.00012600421905517578,
0.0003407001495361328
] |
[
0.978515625,
0.0175933837890625,
0.0037937164306640625,
0.0002186298370361328
] |
The intestinal bacteria of premature infants are characterized by a community state type primarily driven by harmful bacteria such as Enterococcus , Klebsiella , and Acinetobacter . ELBW preterm infants exhibit an increase in the abundance of potentially harmful bacteria in the gut and a decrease in beneficial bacteria. These potentially harmful bacteria may be potential biomarkers for ELBW premature infants.
|
[
"Wanling Chen",
"Kaiping Guo",
"Xunbin Huang",
"Xueli Zhang",
"Xiaoxia Li",
"Zimiao Chen",
"Yanli Wang",
"Zhangxing Wang",
"Rongtian Liu",
"Huixian Qiu",
"Mingbang Wang",
"Shujuan Zeng"
] |
https://doi.org/10.3390/children11070770
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11276386_p0
|
PMC11276386
|
sec[0]/p[0]
|
1. Introduction
| 3.619141 |
biomedical
|
Other
|
[
0.98974609375,
0.0094757080078125,
0.000843048095703125
] |
[
0.01482391357421875,
0.91455078125,
0.0660400390625,
0.0045166015625
] |
The use of extracorporeal membrane oxygenation (ECMO) during cardiac arrest has become more widespread in recent years for a variety of reasons . ECMO is a highly beneficial tool for the treatment of infants with refractory cardiac arrest, and a favorable neurological outcome can be achieved in the majority of survivors, even after prolonged resuscitation .
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276386_p1
|
PMC11276386
|
sec[0]/p[1]
|
1. Introduction
| 3.615234 |
biomedical
|
Other
|
[
0.994140625,
0.00489044189453125,
0.0007739067077636719
] |
[
0.128662109375,
0.5703125,
0.296875,
0.004241943359375
] |
When compared with continuous conventional CPR, extracorporeal cardiopulmonary resuscitation (E-CPR) after refractory cardiac arrest has been associated with better survival outcomes in pediatric victims .
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276386_p2
|
PMC11276386
|
sec[0]/p[2]
|
1. Introduction
| 2.925781 |
biomedical
|
Other
|
[
0.6123046875,
0.380126953125,
0.007415771484375
] |
[
0.006378173828125,
0.982421875,
0.0008664131164550781,
0.01055145263671875
] |
For this reason, in hospitals with suitably trained staff and adequate resources, early E-CPR should be considered in infants for an apparent reversible cause when conventional advanced life support does not lead to spontaneous recovery .
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276386_p3
|
PMC11276386
|
sec[0]/p[3]
|
1. Introduction
| 2.25 |
biomedical
|
Other
|
[
0.93212890625,
0.0609130859375,
0.0071563720703125
] |
[
0.0044403076171875,
0.9921875,
0.00048804283142089844,
0.0028095245361328125
] |
Most in-hospital pediatric cardiac arrests (CA) occur in the Pediatric Intensive Care Unit (PICU) . However, they can also occur in other places such as the emergency room , hospital wards or even radiology rooms . If available, ECMO could be used in all of these cases and during out-of-hospital cardiac arrests as well .
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999999 |
PMC11276386_p4
|
PMC11276386
|
sec[0]/p[4]
|
1. Introduction
| 3.554688 |
biomedical
|
Other
|
[
0.6474609375,
0.345703125,
0.00701904296875
] |
[
0.009124755859375,
0.98291015625,
0.0015821456909179688,
0.00646209716796875
] |
As usually, ECMO procedures are not available is some hospital areas, such patients must be transported immediately to a suitable room (i.e., the PICU or operating room) in order to perform ECMO while continuous, as effective as possible resuscitation maneuvers are being performed until the start of ECMO . The main reason is that in order for ECMO to have positive results, it is absolutely essential to minimize flow interruption and pauses. High-quality maneuvers must be performed constantly, and especially during transfer .
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276386_p5
|
PMC11276386
|
sec[0]/p[5]
|
1. Introduction
| 2.570313 |
biomedical
|
Other
|
[
0.63818359375,
0.350341796875,
0.01149749755859375
] |
[
0.01328277587890625,
0.9736328125,
0.00667572021484375,
0.006496429443359375
] |
The general recommendation is to perform CPR in situ with the victim lying on a firm surface . However, in such cases, resuscitating the victim “on the way to the PICU/operating room” to perform E-CPR would be another of the major exceptions to standard, on-site CPR. However, the evidence regarding CPR quality during intra-hospital transfers with infant victims is extremely limited .
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276386_p6
|
PMC11276386
|
sec[0]/p[6]
|
1. Introduction
| 3.917969 |
biomedical
|
Study
|
[
0.98388671875,
0.01509857177734375,
0.0011272430419921875
] |
[
0.9951171875,
0.00405120849609375,
0.0003402233123779297,
0.0007205009460449219
] |
The hypothesis is that trained personnel are able to maintain CPR quality during an in-hospital transfer of an infant victim. Therefore, the objective of this study is to evaluate the overall quality of resuscitation maneuvers (chest compressions and ventilations) during manual CPR on an infant manikin during simulated in-hospital transfer.
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276386_p7
|
PMC11276386
|
sec[1]/sec[0]/p[0]
|
2.1. Study Design
| 3.494141 |
biomedical
|
Study
|
[
0.98193359375,
0.0170440673828125,
0.0010967254638671875
] |
[
0.9873046875,
0.0111236572265625,
0.0005168914794921875,
0.0011072158813476562
] |
A simulation controlled study was conducted with the aim of comparing CPR during stretcher transfer and baseline CPR by means of a randomized crossover design .
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276386_p8
|
PMC11276386
|
sec[1]/sec[0]/p[1]
|
2.1. Study Design
| 2.423828 |
biomedical
|
Study
|
[
0.99267578125,
0.005290985107421875,
0.00206756591796875
] |
[
0.9970703125,
0.0022907257080078125,
0.0002110004425048828,
0.0004050731658935547
] |
This study involved a convenience sample of 26 nursing students from the Pontevedra School of Nursing (University of Vigo, Pontevedra, Spain). All participants had undergone previous infant CPR training and had duly demonstrated their ability to perform quality CPR (>70%). In addition, the nursing students were familiar with the hospital in question and the material used.
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276386_p9
|
PMC11276386
|
sec[1]/sec[0]/p[2]
|
2.1. Study Design
| 1.777344 |
biomedical
|
Study
|
[
0.96337890625,
0.0016222000122070312,
0.034881591796875
] |
[
0.9169921875,
0.08154296875,
0.0009527206420898438,
0.0007405281066894531
] |
The exclusion criterion comprised not having completed any of the study tests. All participants completed the study and gave informed consent for the use of their data, which were subsequently pseudonymized. This study followed the ethical guidelines of the Declaration of Helsinki. The ethics committee of the University School of Education and Sport Sciences of the University of Vigo, number 19-2802-18, approved the study protocol. This study was conducted in November 2021.
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276386_p10
|
PMC11276386
|
sec[1]/sec[1]/p[0]
|
2.2. Study Protocol
| 1.540039 |
biomedical
|
Study
|
[
0.8857421875,
0.005413055419921875,
0.108642578125
] |
[
0.51318359375,
0.470703125,
0.0127410888671875,
0.00353240966796875
] |
The details can be seen in Figure 1 .
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.85714 |
PMC11276386_p11
|
PMC11276386
|
sec[1]/sec[1]/p[1]
|
2.2. Study Protocol
| 3.689453 |
biomedical
|
Study
|
[
0.9599609375,
0.0389404296875,
0.0013341903686523438
] |
[
0.9345703125,
0.060791015625,
0.0006995201110839844,
0.00402069091796875
] |
In pairs, the participants randomly performed two 2 min CPR tests on the infant manikin. An emergency scenario was simulated in which an infant victim suffers CA on the hospital ward and is immediately transferred to the operating room in order to perform ECMO. However, in the interim, resuscitation maneuvers are simultaneously being performed (i.e., compressions and ventilations with a self-inflating bag) during the transfer to minimize the risk of fatality.
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276386_p12
|
PMC11276386
|
sec[1]/sec[1]/p[2]
|
2.2. Study Protocol
| 3.912109 |
biomedical
|
Study
|
[
0.673828125,
0.323974609375,
0.0020999908447265625
] |
[
0.79443359375,
0.154052734375,
0.003940582275390625,
0.047332763671875
] |
The rescuers followed the European Resuscitation Council (ERC) 2021 guidelines, performing 5 rescue breaths followed by cycles of 15 compressions (CC) and 2 ventilations (V). One of the participants performed CC via the two-thumb–encircling hands technique. At the same time, the other first responder performed rescue ventilation with the Ambu ® SPUR II Infant Resuscitator and Ambu ® Baby Face Mask number 0A (a round disposable silicon face mask for patient oxygenation and ventilation) (Ambu, Copenhagen, Denmark). Although a basic CPR infant manikin was used, a real situation was simulated as the manikin was being monitored by the LifePak ® 20E Defibrillator/Monitor (Physio-Control, Redmon, WA, USA), a capnography monitor of ventilation and the concentration of CO 2 exhaled air, and an electrocardiogram. The recorded heart rhythm was non-shockable.
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999996 |
PMC11276386_p13
|
PMC11276386
|
sec[1]/sec[1]/p[3]
|
2.2. Study Protocol
| 1.789063 |
biomedical
|
Study
|
[
0.9736328125,
0.01395416259765625,
0.0122833251953125
] |
[
0.84521484375,
0.1488037109375,
0.0022125244140625,
0.0036411285400390625
] |
The tests were carried out in a hospital in a randomized order and were made up of the following:
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999997 |
PMC11276386_p14
|
PMC11276386
|
sec[1]/sec[1]/sec[0]/p[0]
|
2.2.1. Control Test (CT)
| 2.498047 |
biomedical
|
Study
|
[
0.97021484375,
0.0243377685546875,
0.0053863525390625
] |
[
0.78076171875,
0.212890625,
0.001522064208984375,
0.004779815673828125
] |
This encompassed baseline CPR with a test in a static position using a manikin on a stretcher without a rigid board.
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
N/A
|
https://creativecommons.org/licenses/by/4.0/
|
en
| 0.999998 |
PMC11276386_p15
|
PMC11276386
|
sec[1]/sec[1]/sec[1]/p[0]
|
2.2.2. Stretcher Test (ST) ( Figure 2 )
| 3.322266 |
biomedical
|
Study
|
[
0.8994140625,
0.09619140625,
0.00424957275390625
] |
[
0.9052734375,
0.089599609375,
0.0006327629089355469,
0.004665374755859375
] |
This was comprised of a CPR test with an infant manikin carried on a stretcher in an intra-hospital transfer without a rigid board. The subject performing chest compressions stood on the victim’s right hand side and the subject responsible for ventilating was positioned to the left of the victim. Two other emergency staff directed the stretcher. The manikin was placed with its head in the direction of the way forward. The participants completed a predefined 2 min hospital route that consisted of the following:
|
[
"Myriam Santos-Folgar",
"Felipe Fernández-Méndez",
"Martín Otero-Agra",
"Roberto Barcala-Furelos",
"Antonio Rodríguez-Núñez"
] |
https://doi.org/10.3390/children11070865
|
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.