Kidney stone disease is a common urological condition requiring timely detection to prevent complications. Non-contrast computed tomography (CT) is the gold standard for detecting renal calculi, but manual interpretation is time-consuming and subject to variability.
This study evaluates four deep learning object detection models-YOLOv8, YOLOv5, Faster R-CNN, and RetinaNet-for automated kidney stone detection in CT images. A dataset of 4,000 annotated CT slices from 170 patients was used. Performance was evaluated using mAP@0.5, precision, recall, false positive and false negative rates, and inference speed.
Faster R-CNN achieved the highest localization accuracy (mAP@0.5 = 0.93), while YOLOv8 demonstrated the best balance between accuracy () and computational efficiency, achieving real-time inference at 65 FPS.
The results highlight the trade-off between detection accuracy and processing speed across architectures. YOLOv8 provides an optimal balance for clinical implementation due to its strong performance and real-time capability.
Frontiers in medicine. 2026 Mar 18*** epublish ***
Yuguang Ye, Kavimbi Chipusu, Liuying He, Suo Shen, Jianlong Huang
School of Mathematics and Computer Science, Quanzhou Normal University, Quanzhou, China., Division of Biomedical Engineering, Department of Mechanical Engineering, University of Saskatchewan, Saskatoon, SK, Canada., Department of Urology, First People's Hospital of Fuyang, Hangzhou, China.