AI-Driven CHD Detection Using an Ultra-Low Power Embedded System

Congenital heart disease (CHD) is a significant global health concern, particularly in disadvantaged communities that lack access to advanced diagnostic equipment and medical ability. Capitalizing on advances in digital stethoscope technology, this paper presents an innovative AI-driven CHD detection system implemented on an ultra-low-power embedded platform. Utilizing phonocardiogram data acquired using a digital stethoscope, our solution offers accurate CHD screening, while consuming less than 5mW, enabling 24-hour operation on a small 55mAh battery. This approach enhances portability, reduces pre-screening costs, and ensures the required accuracy for early diagnosis. The model used for inference implementation is based on an Extreme Gradient Boosting (XGBoost) algorithm. The system's design and optimization for low-power environments are discussed, along with its potential of integration with low power digital stethoscope platforms.