A Neuromorphic Model With Delay-Based Reservoir for Continuous Ventricular Heartbeat Detection.

Area of Science:

• Neuromorphic Engineering
• Biomedical Signal Processing
• Computational Neuroscience

Background:

• Growing interest in neuromorphic hardware for bio-inspired algorithms.
• Need for hardware-based signal processing to avoid intensive digital operations.
• Conventional ECG classification methods are not ideal for analogue neuromorphic systems.

Purpose of the Study:

• To develop an intelligent, hardware-based neuromorphic model for continuous electrocardiogram (ECG) signal processing.
• To create an end-to-end dynamic system mimicking real-time signal flow in neuromorphic hardware.
• To avoid computationally intensive digital operations like signal segmentation and feature extraction.

Main Methods:

• Utilized delay-based reservoir computing as the core information processing unit.
• Implemented a novel training and labeling methodology.
• Employed the intrinsic memristive property of the reservoir for historical ECG information retention and high-dimensional mapping.

Main Results:

• Achieved 81% sensitivity and 98% accuracy in detecting ventricular ectopic heartbeat risk factors using the MIT-BIH database.
• Demonstrated a minimum memory requirement of 3.1 MB for inference due to analogue domain computation.
• The model processes raw ECG streams directly, outputting risk factor amplitude.

Conclusions:

• The proposed neuromorphic model offers an efficient, hardware-based solution for real-time ECG analysis.
• The system significantly boosts memory efficiency, making it suitable for future wearable devices.
• This approach bypasses traditional digital signal processing limitations in analogue neuromorphic systems.