Tissue color images segmentation using artificial neural networks

Currently, most pathologists make their diagnosis of cancer based on a rough estimation of the density of the cell´s nuclei in the tissue sample, and also based on the morphological abnormality of the cancerous cells. The methods used to achieve their diagnosis are either too simple to diagnose a complicated tissue image or are depending on heavy human intervention and very time consuming. In order to assist pathologists to make a consistent, objective and fast diagnosis, we present in this paper a method of tissue color image segmentation as the main step of an entire system of cancer diagnosis. The segmentation approach is an unsupervised algorithm based on a modified Hopfield neural network (HNN). This algorithm is superior to HNN in the sense that it converges in a prespecified time to a nearby global minimum rather than an early local minimum. Two types of tissue (liver, lung) are presented, and three-color spaces (RGB, HLS and HSV) are used to investigate the efficiency of the algorithm in segmenting color images.