Real-Time Image Segmentation of Ultrasound Imaging to Assist Cardiac Disease Evaluation

Product introduction

Application scenario

Real-time ultrasound imaging processing

Specialized AI accelerators outperform CPUs /GPUs/NPUs in meeting medical imaging demands for real-time large-model inference with high performance/power efficiency.
ACCESS developed a portable ultrasound cardiac segmentation prototype using AC-Codesign-v1. This enables extraction of key cardiac parameters (size, volume, morphology) for functional assessment and anomaly detection. Integrated edge AI accelerators enhance diagnostic precision and operational efficiency in portable ultrasound systems.

In a standard AI-assisted cardiac ultrasound diagnostic setup, the system demonstrates a processing rate of 50 frames per second (FPS) with superior accuracy in image segmentation.
The mean dice similarity coefficients for segmentation of the left ventricle (LV), left atrium (LA), and myocardium (Myo) are 91.8%, 89.6%, and 84.3%, respectively.
The Center intends to collaborate with industry partners to further advance and commercialize this innovative technology.