Robust Heartbeat Detection From Multimodal Data via CNN-Based Generalizable Information Fusion.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Heartbeat detection is crucial for cardiac disease diagnosis and management, traditionally relying on electrocardiograms (ECG).
• Integrating additional physiological signals, such as arterial blood pressure (BP), can enhance detection robustness, particularly in critical care settings.
• Current multimodal approaches often indirectly fuse signal-specific estimates, limiting direct information integration.

Purpose of the Study:

• To develop a robust heartbeat detection method by directly fusing information from multiple physiological signals.
• To eliminate the need for manual feature engineering and ad hoc fusion strategies in multimodal signal analysis.
• To create a data-driven algorithm capable of learning optimal features from diverse signal inputs.

Main Methods:

• A convolutional neural network (CNN) was employed as a heartbeat detector, learning fused features directly from multiple physiological signals.
• The CNN-based information fusion (CIF) approach avoids intermediate signal-specific estimations and voting mechanisms.
• The algorithm is designed to be data-driven, enabling it to learn relevant features from any given set of signals.

Main Results:

• The proposed method achieved a 94% score using ECG and BP signals from the PhysioNet 2014 Challenge database.
• An accuracy of 99.92% was obtained using two ECG channels from the MIT-BIH arrhythmia database, outperforming previous results.
• The detector demonstrated high accuracy across various clinical conditions.

Conclusions:

• The CNN-based information fusion (CIF) algorithm offers a generalizable, robust, and efficient solution for heartbeat detection from multiple signals.
• This technique is valuable for medical signal monitoring systems, ensuring accurate heartbeat estimation even with partial channel reliability.
• The direct fusion approach enhances the reliability and accuracy of cardiac monitoring in diverse clinical scenarios.