Contouring blood pool myocardial gated SPECT images with a neural network leader segmentation and a decision-based fuzzy logic

The authors have developed an algorithm to extract ventricular contours in gated single photon emission computed tomography (G-SPECT) images of the blood pool. In this kind of images, the authors have to deal mainly with 3 problems. First of all, there is a superposition of nuclear emission sources within the epicardium making difficult the separation of the 2 myocardial ventricles. Secondly, due to the great variety of ventricular forms, the process is hard to automate and, thus, human participation is required although it is time consuming. Third, the authors have to deal with noise and diffusion resulting from the nature of the technique itself. Their algorithm employs wavelets filters. To overcome the first and third difficulties. A neural network leader type algorithm, combined with a decision-based fuzzy logic system, enables the authors to segment the G-SPECT images and to recognize the left ventricle. This way, the endocardial contour is extracted and the second problem can be solved. Experimental results and a comparison with other methods, among them the one used in the authors' laboratory in clinical routine, show that the performance of the authors' algorithm is very high.