engleski

A Method for Segmentation of CT Head Images

A novel method for automatic segmentation of computed tomography (CT) head images of patients having spontaneous intracerebral brain hemorrhage has been presented in this work. The images are segmented in five regions of interest: background, skull, brain, hemorrhage, and calcifications. The method consists of four major phases. The first phase performs a brightness normalization of a CT image by applying a K-means clustering algorithm to the pixel brightness values. A feature extraction based on a special receptive field is done in the second phase. The topology of the receptive field is inspired by the structure of the human retina. The third phase performs a pixel classification by means of a feed-forward error-back propagation neural network. In the last phase, a rule-based expert system is used to perform image labeling. The expert system knowledge is based on assumptions on region sizes and region neighborhood relations. The proposed method has been applied to real patient CT images and has shown encouraging results.

nije evidentirano

nije evidentirano