<![CDATA[Kidney stone diagnosis requires fast and reliable evaluation, yet ultrasound interpretation still largely depends on clinical expertise. This study evaluates four Convolutional Neural Network (CNN) architectures, VGG16, ResNet50, MobileNetV2, and EfficientNetB0 for classifying kidney ultrasound images into Normal and Stone categories. Using a public dataset of 9,416 images, the models were assessed in terms of predictive performance and computational efficiency. MobileNetV2 achieved perfect classification performance, recording 100% accuracy, precision, recall, and F1-score, while maintaining the smallest parameter size (≈3.6M) and fastest training time (~44 s/epoch). VGG16 and ResNet50 also delivered near perfect accuracy (99.79% and 99.89%) with full recall for Stone cases. In contrast, EfficientNetB0 failed to generalize, yielding only 51.62% accuracy due to severe misclassification of Normal images. These results demonstrate that MobileNetV2 provides the most reliable and efficient solution for ultrasound based kidney stone classification, highlighting its strong potential for practical clinical deployment.]]>
Copyrights © 2026
