Deep Learning in Physiological Signal Data: A Survey.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Deep Learning (DL) shows great promise in 2D medical imaging.
• Its application to 1D physiological signals (e.g., ECG, EEG) is less explored.
• Physiological signals are crucial for diagnosing various medical conditions.

Purpose of the Study:

• To survey and analyze recent research (2018-2019) on DL for physiological signal analysis.
• To categorize and compare key DL parameters impacting performance in medical tasks.
• To provide a comprehensive understanding of DL's potential in this domain.

Main Methods:

• Systematic literature review of 147 papers published between January 2018 and October 2019.
• Categorization based on physiological signal data (modality, application) and DL concepts (model, architecture, datasets).
• Analysis of key DL parameters: input data type, task, model, training architecture, and dataset sources.

Main Results:

• Identified diverse DL approaches applied to electromyogram (EMG), electrocardiogram (ECG), electroencephalogram (EEG), and electrooculogram (EOG) signals.
• Key parameters influencing system performance were systematically reviewed and compared.
• Established a taxonomy for DL methods in physiological signal analysis.

Conclusions:

• Deep learning holds significant potential for advancing the analysis of physiological signals in medicine.
• Understanding and comparing key DL parameters is crucial for optimizing performance in medical applications.
• This review provides a foundational framework for future research in this interdisciplinary field.