Use of neural networks for feature based recognition of liver region on CT images

Medical diagnostic support systems are gaining popularity due to easy access of computers in this field and the increasing workload of radiologists. The automatic processing of X-ray images, segregation of different regions, and detection of certain features are a few of the objectives of this task. Neural networks have been successfully applied to various pattern recognition problems. High-resolution images, such as X-ray computed tomography (CT) images, where real time processing is desirable, present a challenge to image processing. Neural networks, due to their parallel processing nature, present an attractive prospect to the solution of such images. Since these medical images require a priori knowledge for their analysis, knowledge is stored in the network through training. A neural network has been trained to learn the texture of the liver region that may be used in 3D rendering and automatic segmentation of normal and abnormal liver regions. The network is based on a backpropagation neural network (BPNN) that is trained on a set of features calculated in a window of fixed pixel size. The system learns the texture of the liver region through supervised training and gives acceptable results for an independent set of images not used during training.