.png){kind=icon}
.png){kind=icon}
Return
Optimizing ST-Segment classification in ECG using multi-task learning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.bspc.2024.106591.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Objective: The accurate classification of ST-segments in electrocardiograms (ECGs) is crucial for evaluating and diagnosing myocardial ischemia. However, current algorithms for automatic ST-segment classification lack satisfactory accuracy. Methods: In this manuscript, we propose a deep learning-based ST-segment classification algorithm that incorporates multi-task learning to classify ST-segments as normal, elevated, or depressed. We utilize three datasets for model training and evaluation: the European ST-T Database (EDB), the China Physiological Signal Challenge 2018 (CPSC2018), and the STC Classification Database. To address the class imbalance in the data, we employ the Synthetic Minority Over-sampling Technique (SMOTE) and a novel waveform scaling augmentation (WSA) method. Pseudo-labels for each lead of the twelve-lead ECG signal are generated using a rule-based approach. Subsequently, we optimize the model and facilitate ST-segment classification by employing a customized loss function that combines signal reconstruction, classification, and lead localization losses. Results: Our proposed method outperforms other state-of-the-art approaches, as evidenced by the average F1 scores on the EDB (0.957), CPSC2018 database (0.807), and STC database (0.790). Conclusion: Multi-task learning outperforms single-task learning in effectively classifying ST-segments, offering a promising tool for automatically detecting ischemic cardiovascular disease.
Keywords:
ST-segment classification
Electrocardiogram
Deep learning
Multi-task learning
Class imbalance
Journal
IF:
4.9
Papers:
9.8K
Citations:
2.4W