Optimized constraint satisfaction neural network for medical image segmentation

An image segmentation process can be favourably imaged onto a multi-particle system by a relaxation setup as a constraint satisfaction problem (CSP). One pixel can develop itself due to its own state and depending on the states of the adjacent pixels in direction of class intensity means. In this case, the relaxation process in which the segmentation decision about each image pixel is an iterative approach can be performed in parallel. The region-based image segmentation using a constraint satisfaction neural network (CSNN) is an innovative segmentation technique. Starting from this model, a technique for the segmentation of medical images using an optimized constraint satisfaction neural network (OCSNN) is presented, which boasts a significantly higher performance compared with the CSNN regarding the segmentation quality, time factor, and memory demand. We describe the structure of the neural network and the principle of the data compression, the initialization model adjusted by signal statistics, the weight function and temporary assessment of the neighbourhood influence and their influences on the segmentation process and the relaxation dynamics. The effect on the network efficiency is presented