Renal Lesion Classification in Kidney CT Images by Seven-Layer Convolution Neural Network

Abstract

AI techniques are pervading the medical field and facilitating the related educational applications, such as computer aided medical diagnosis, online surgery platforms and medical learning environments. Nowadays daily medical images and data come into being the big medical data need be processed as fast as possible. AI enables human to improve the accuracy and efficiency of diagnosis greatly relying on the techniques of radiological images analysis. In this paper one 7-layer deep Convolutional Neural Network (CNN) is designed to classify renal lesion in kidney Computed Tomography (CT) images. The CNN is trained on a middle-size dataset with 614 kidney CT images collected from real clinical data. Experiments show the mean and standard deviation of the overall accuracy of the binary classification reaches 90.36 ±1.02%. It has greatly better performance about 25% than the traditional Probability Neural Network (PNN) method with predefined features. The optimal structure of this CNN proves our method is rather promising to help doctors make medical diagnosis.