Continuous electrocardiogram (ECG) monitoring is essential for the early detection of cardiac disorders, yet most state-of-the-art algorithms rely on computationally intensive machine learning models that are difficult to deploy on low-power wearable devices. Recently, neuromorphic computing has emerged as an energy-efficient alternative by exploiting sparse, event-driven computation. While existing training-free spiking neural networks can successfully classify normal rhythms and simple ECG abnormalities, clinically relevant arrhythmias such as atrial fibrillation, atrial flutter, and premature contractions require the analysis of more complex temporal relationships between successive heartbeats.

This project aims to extend a spiking neural network for real-time ECG interpretation by developing neuromorphic computational primitives for cardiac arrhythmia detection. Instead of relying on supervised learning, the network will exploit electrophysiological knowledge through biologically inspired mechanisms including delay lines, coincidence detection, winner-take-all competition, and inhibitory interactions. The resulting architecture will detect temporal irregularity, premature atrial and ventricular contractions, and repetitive ectopic patterns while remaining suitable for deployment on neuromorphic hardware for wearable cardiac monitoring.

 

Methodology

1. Analyze ECG recordings from publicly available datasets and characterize the temporal and morphological signatures of clinically relevant arrhythmias.

2. Develop spiking neural network modules for beat-to-beat regularity estimation, premature beat detection, and temporal pattern recognition using biologically inspired computational primitives.

3. Integrate the proposed modules into an existing neuromorphic ECG classifier and evaluate their ability to detect complex arrhythmias in real time.

4. Compare the proposed architecture with conventional machine learning approaches in terms of classification performance, interpretability, computational complexity, and energy efficiency.

 

Available Material

• Public ECG datasets (e.g., PTB-XL).

• Existing training-free spiking neural network framework for ECG classification.

• Access to neuromorphic hardware for deployment and validation.

 

Requirements

• Python programming.

• Interest in computational neuroscience, biomedical signal processing, and neuromorphic computing.

 

Contacts

Elisa Donati (elisa@ini.uzh.ch)

 

References

1. Chen, J., Tian, F., Yang, J., & Sawan, M. (2022, May). An event-driven compressive neuromorphic system for cardiac arrhythmia detection. In 2022 IEEE International Symposium on Circuits and Systems (ISCAS) (pp. 2690-2694). IEEE.

2. Amirshahi, A., & Hashemi, M. (2019). Ultra low-power and real-time ecg classification based on stdp and r-stdp neural networks for wearable devices. arXiv preprint arXiv:1905.02954.