Opportunities and challenges of deep learning methods for electrocardiogram data: A systematic review

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Electrocardiogram (ECG) is a vital diagnostic tool in medicine.
• Deep learning (DL) methods show great promise for predictive healthcare tasks using ECG signals.

Purpose of the Study:

• To systematically review DL methods applied to ECG data.
• To analyze DL models, tasks, and data from both modeling and application viewpoints.
• To identify current challenges and future research directions in DL for ECG analysis.

Main Methods:

• Conducted a systematic literature review of DL models applied to ECG data.
• Extracted 191 relevant papers published between January 2010 and February 2020.
• Analyzed papers based on tasks, DL models, and datasets used.

Main Results:

• The number of publications on DL for ECG data has grown exponentially.
• DL methods achieve accuracy comparable to traditional approaches, with ensembles yielding superior results.
• Hybrid architectures, particularly CNN-RNN ensembles with expert features, demonstrate the best performance.

Conclusions:

• DL is rapidly advancing ECG analytics, with hybrid models showing exceptional results.
• Key challenges include improving interpretability, scalability, and efficiency of DL models.
• Future research should explore novel datasets and methods for new ECG applications.