Hierarchical deep learning for autonomous multi-label arrhythmia detection and classification on real-world wearable...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Arrhythmia detection and classification are complicated by imbalanced data and complex arrhythmia combinations.
• Wearable electrocardiogram (ECG) monitoring presents unique challenges due to environmental noise, unlike controlled clinical settings.
• Accurate arrhythmia classification is crucial for timely diagnosis and intervention.

Purpose of the Study:

• To develop and evaluate a novel hierarchical model for robust arrhythmia detection and classification using wearable ECG data.
• To address the challenges of data imbalance and environmental noise in real-world arrhythmia monitoring.
• To improve the accuracy and reliability of automated arrhythmia diagnosis.

Main Methods:

• A hierarchical model combining Convolutional Neural Networks (CNN) with Bidirectional Long Short-Term Memory (BiLSTM) networks and an Attention mechanism was proposed.
• The model features a binary classification module for normal vs. arrhythmia beats and a multi-label module for complex arrhythmia combinations.
• Performance was evaluated on a proprietary dataset against several baseline models, including CNN+BiGRU with Attention, ConViT, EfficientNet, and ResNet.

Main Results:

• The proposed CNN+BiLSTM with Attention model demonstrated superior performance over existing baselines on the proprietary dataset.
• Achieved an average accuracy of 95%, F1-score of 0.838, and AUC of 0.906 for binary classification.
• Attained an average accuracy of 88%, F1-score of 0.736, and AUC of 0.875 for multi-label classification.

Conclusions:

• The developed model effectively detects and classifies real-world cardiac arrhythmias, even in noisy environments.
• This framework has the potential to revolutionize arrhythmia diagnosis, reduce cardiologist workload, and enable personalized emergency interventions.
• Real-time monitoring of arrhythmia occurrence via this technology can lead to faster patient care and improved outcomes.