Abstract

We propose a method for segmentation of vascular structures and determination of blood flow velocity in coronary angiograms. The angiogram images of normal and abnormal-collateral patients are acquired at a rate of 15 frames/sec. In each frame, blood vessel is segmented from background using a backpropagation network. The input is given to two network topologies (121-17-2 and 4-3-2 layer configuration) and tested for their performance. The 4-3-2 configuration was able to classify blood vessel with less number of iterations comparatively and it can detect even small vessels with less computation time. The blood flow velocity in angiogram is determined in two methods. First method is by measuring the distance traversed by the contrast agent in each frame. The second method is based on determining the change in concentration of the contrast agent in two fixed region of interest. By first method, the flow velocity for normal and collateral angiograms are found to be 38 pixels/frame(p/f) and 15 p/f, respectively and by the second method, it is calculated as 45 and 28 p/f, respectively. The results show delayed arrival of contrast in abnormal collaterals than in normal images.