Real Time GPU-Based Segmentation and Tracking of the Left Ventricle on 2D Echocardiography

摘要

Left ventricle segmentation and tracking in ultrasound images present necessary tasks for cardiac diagnostic. These tasks are difficult due to the inherent problems of ultrasound images (i.e. low contrast, speckle noise, signal dropout, presence of shadows, etc.). In this paper, we propose an accurate and automatic method for left ventricle segmentation and tracking. The method is based on optical flow estimation for detecting the left ventricle center. Then, the contour is defined and tracked using convex hull and spline interpolation algorithms. In order to provide a real time processing of videos, we propose also an effective and adapted exploitation of new parallel and heterogeneous architectures, that consist of both central (CPU) and graphic (GPU) processing units. The latter can exploit both NVIDIA and ATI graphic cards since we propose CUDA and OpenCL implementations. This allowed to improve the performance of our method thanks to the parallel exploitation of the high number of computing units within GPU. Our experiments are conducted using a set of 11 normal and 17 disease hearts ultrasound video sequences. The related results achieved automatic and real-time left ventricle detection and tracking with a rate of 92 % of success.