Automatic Kidney Stone Detection using Low-cost CNN with Coronal CT Images

Abstract

A fast diagnosis of kidney stones is crucial to start the correct treatment, minimizing the risks of urinary complications. Machine learning approaches are valuable for an automatic diagnosis system for kidney stones from computer tomography (CT) exams. Recently, related studies achieved high accuracy in detecting kidney stones using deep-learning neural networks. However, their approaches were highly complex and time-consuming. This paper proposes a method for automatically detecting kidney stones on CT images using low-complexity deep-learning techniques. We compared three models based on Convolutional Neural Networks (CNN): 3-layered CNN (Conv3), 4-layered CNN (Conv4), and MobileNetV1. They were applied to kidney stone detection using 1799 CT images divided into 80% for training, 10% for validation, and 10% for testing. The proposed model Conv4 obtained the best performance, achieving a test accuracy of 97.2% and an F1-score of 97.6%, with a training time of 140 seconds.