Embedded AI for ECG Monitoring: Current Limitations and Emerging Perspectives

Embedded AI in ECG monitoring is a rapidly evolving field that is changing the landscape of wearable and implantable devices for monitoring cardiac health, although it is still limited by strong requirements for real-time analysis, power consumption, and memory storage. The current state of on-chip implementations for ECG analysis is still struggling with compute intensity, battery life, robustness against interference, and the need for secure on-chip analytical computing. Current directions in cutting-edge technologies aim to alleviate these challenges in ECG monitoring through the use of reduced neural networks, model approximation, and Tiny ML approaches specifically for microcontrollers and highly low power chips. Current directions in other areas for the advancement of ECG analysis in the real world encompass the use of hybrid processing, adaptive learning strategies, and the internet of things for intelligent processing on the edge, greatly improving real world robustness for clinical, ambulatory, and monitoring environments. Hardware-software co-design for the use of purposefully designed chips, field programmable chips, and neuromorphic chips is assisting in improving real-time analysis while reducing energy consumption in the field. In this paper, the current limitations in the development of artificial intelligence for electrocardiogram-based monitoring systems in practical settings are discussed, and some perspectives which might help in the collaborative development of real-time cardiac monitoring systems are presented.