ORIGINAL RESEARCH article
Front. Physiol.
Sec. Computational Physiology and Medicine
Volume 16 - 2025 | doi: 10.3389/fphys.2025.1608441
ICLHD-Net: An Interpretable Contrastive Learning Network for Hemodialysis time series classification
Provisionally accepted
- 1School of Computer Engineering and Science, Shanghai University, Shanghai, China
- 2Department of Neurology, Shanghai Eighth People’s Hospital, Shanghai, China
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Hemodialysis is the most common treatment for uremic patients with end-stage renal disease (ESRD).However, due to the limitations of medical facilities, the adequacy of clinical hemodialysis cannot be met every time, which has resulted in a great impact on the prognosis and the patient's quality of life. Therefore, it is very valuable to adjust the quality of hemodialysis timely according to the indicators, giving priority to ensure that key indicators meet the standards. Through deep learning, the indicators importance can be obtained easily during the training process. Key hemodialysis indicators can be accurately found, and the patient's survival rate can be predicted based on existing indicators data.However, the number of samples in the collected dataset cannot be guaranteed, and for small training sets, the ranking of feature importance obtained from each training is often unstable. In this paper, we propose ICLHD-Net that fuses feature importance to find key indicators and predict patient survival rate. In particular, we combine two methods of calculating the feature importance ranking and improve the contrastive learning framework and loss function to ensure that the obtained feature importance ranking results are more accurate and stable, while improving the performance of classification. Our method is evaluated on a clinical environment dataset collected at Shanghai Municipal Eighth People's Hospital. Through five-fold cross-validation, this method has an accuracy of 92.8% for classifying the one-year survival rate of the patient, an AUC of 94.6%, a recall of 87.4% and a f1 score of 89.4%. Compared with the previous state-of-the-art model, the average accuracy of the data set increased by 1.3%.
Keywords: hemodialysis, feature importance ranking, Contrastive learning, Time series classification, Interpretable deep learning
Received: 09 Apr 2025; Accepted: 22 May 2025.
Copyright: © 2025 Song, Qi, Huang, Zhang and Liu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Anping Song, School of Computer Engineering and Science, Shanghai University, Shanghai, China
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.