Evaluation and Classification of Kidney Stone Detection Using Deep Learning Techniques

Abstract

Kidney stone detection is a crucial task in medical diagnostics where early identification can mitigate severe health complications. This research employs advanced deep-learning techniques to classify four types of renal ultrasound images: cyst, normal, stone, and tumor. Three pre-trained and customized neural network architectures-EANet, InceptionV3, and SqueezeNet-are utilized for this purpose. The methodology was rigorously evaluated on a testing dataset consisting of 3,734 renal ultrasound images. Results demonstrate an overall accuracy of 95.8% for EANet, 96.14% for InceptionV3, and 96.1% for SqueezeNet. Comprehensive comparative analysis employing metrics such as accuracy, precision, recall, F1-score, and ROC AUC score reveals that Inception V3marginally outperforms both EANet and SqueezeNet across multiple metrics. The research signifies a substantial advancement in the field of kidney stone detection and poses a promising direction for future clinical implementation.