Effect of Feature selection on the performance of Liver Tumor classification using Neural Networks

Neural network designs have pushed the boundaries of clinical picture analysis achieving remarkable performance in responsibilities together with tissue category and segmentation as well as prediction of numerous scientific consequences. There are different kinds of cancers happening in the liver. Various growths have diverse pictorial appearances and this changes after infusion of the differentiation medium. Hence, finding the location of liver malignancies is a difficult undertaking. In this paper, we are going to discuss liver tumor classification using neural networks. Here we have used a Pattern recognition network, with a training function as Scaled Conjugate Gradient, hidden neurons are 10, performance function as cross-entropy. Here we have collected 67 CT image which includes both Liver abscess and Hepatocellular carcinoma. 93 features were extracted from these images for training, validation, and test of the network. By including GLRLM features, the mean square error is reduced by 28.31%.