Hardware accelerated watershed based echocardiographic image segmentation

Abstract

The proposed poster presents a hardware accelerated left ventricle detection algorithm. Based on marker controlled watershed segmentation method, the algorithm core works sequentially: detection result from a given frame is taken as the segmentation marker to process the next frame. The complete algorithm is written in Matlab. It performs more then 20 elementary operations and needs tens of minutes to compute one movie. As this is a serious disadvantage in clinical application, hardware acceleration has been proposed. Authors built an experimental PC based system equipped with data translation frame-grabber and Alpha-Data ADM-XRC-II board with high capacity Virtex II field programmable gate array. Hardware implementation of particular algorithm stages shows significant speedup due to the possibility of computation concurrency in FPGA Virtex II device