Automated Detection of Kidney Stone Using Deep Learning Models

Abstract

Kidney stone is the most common disease nowa-days. Proper diagnosis of the disease is required to cure and lead a healthy lifestyle. Various methods are posed for kidney stone detection using different imaging techniques. This paper proposes an automated system for detecting kidney stones using deep learning models. The experiments are performed using an open-source Computed Tomography (CT) image dataset. These datasets are made to perform on deep learning models. On analyzing the efficiency of different deep learning models, it is found that VGG series performs the best. The accuracy obtained for kidney stone detection using VGG16 architecture is 99%. The performance of the model has also been evaluated by using a method called stratified K-fold cross validation. Further, the area of kidney stone is detected using Gradient-weighted Class Activation Mapping, which is referred to as Grad-CAM. In short, here we are giving the CT image to VGG model server for classifying and then using Grad-Cam to find the area of interest and finally, the expert checks the output to verify the result.