Convolutional Neural Networks for Segmented Liver Classification

Abstract

The classification of liver disease is of paramount significance for an early diagnosis of patients. In this paper, suggesting a way for classifying the liver in two categories: normal and abnormal based on CT scans is the target. For this experiment, a special earlier focus for getting the best rate by using the Convolutional Neural Networks (CNN) is made. This process has been done by using many different layers to increase the accuracy and reduce the error probabilities by invoking training, validation, and test database, each of these contains a set of images under testing. The process followed through extracting the features and the characteristics found in the segmented liver led up to the classification of testing group into normal and abnormal categories. Initially, and in order to get the best results, the extraction of the liver as a mono-element in the classification there were a need to use Rayleigh, GMM, THRESHOLDING, and finally GVF. These latest results are used as CNN inputs. Experimental results show that CNN features have achieved a rating performance of up to 99.84 %.