An Energy Efficient ECG Ventricular Ectopic Beat Classifier Using Binarized CNN for Edge AI Devices.

Area of Science:

• Biomedical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Wearable Artificial Intelligence-of-Things (AIoT) demands resource and energy-efficient edge devices.
• Accurate and efficient classification of cardiac arrhythmias is crucial for wearable health monitoring.

Purpose of the Study:

• To design and implement an efficient binary convolutional neural network (bCNN) for classifying Ventricular and non-Ventricular Ectopic Beat images.
• To deploy the bCNN model onto a low-resource, low-power Field Programmable Gate Array (FPGA) for edge computing applications.

Main Methods:

• Developed an efficient bCNN algorithm incorporating function-merging and block-reuse techniques.
• Optimized the bCNN model for deployment on a low-power FPGA fabric.
• Evaluated the model's performance using key metrics including accuracy, sensitivity, specificity, precision, and F1-score.

Main Results:

• Achieved a classification accuracy of 97.3% for Ventricular and non-Ventricular Ectopic Beats.
• Demonstrated high sensitivity (91.3%), specificity (98.1%), precision (86.7%), and F1-score (88.9%).
• The deployed model exhibited extremely low dynamic power dissipation of only 10.5-μW.

Conclusions:

• The efficient bCNN model is suitable for resource-constrained wearable AIoT devices.
• The developed algorithm enables high-performance, low-power cardiac arrhythmia classification on edge devices.
• This work contributes to the advancement of efficient edge AI for wearable healthcare applications.