An Artificial Neural Networks Feature Extraction Approach to Predict Nephrolithiasis (Kidney Stones) Based on KUB Ultrasound Imaging

Abstract

Around 10% of the population will experience nephrolithiasis or renal calculi or kidney stones at their life time. Nephrolithiasis is the main cause for the haematuria, micturition and sever pain in the abdomen and during urination. More number of nephrologists are required to meet the rapid increase in the population. This paper deals with a supportive diagnostic system to the physician or nephrologist by applying neural network techniques. Generally, patients are supposed to undergo ultrasound scan test to identify the position, size and number of renal calculi in the renal system. Image processing techniques like preprocessing, segmentation and feature extraction were applied on the ultrasound scan images to extract the GLCM features. These extracted features were supplied as inputs to construct neural network and train the system by using feed-forward back-propagation algorithm. The proposed system is constructed with 22 input nodes, ten hidden nodes and one output node and thereby trained until it gets target or output. Then this trained neural network is allowed automatically to work on emerging new samples. Around 50 real-time images were collected from the patients, in which 60% of them were used for training and 40% for testing. The supportive software tool used for this proposed system is MATLAB 8.5.