YOLO-Driven Kidney Stone Localization and Estimation in CT Images

The detection of kidney stones in medical imaging is crucial for timely diagnosis and treatment. This work presents a novel approach utilizing state-of-the-art deep learning techniques, like YOLO v8, You Only Look Once (YOLO) v10, and YOLO-NAS to optimize the reliability and speed of renal stone diagnosis from Computed Tomography (CT) scan. This work focuses on developing a renal stone identification system and counting the number of stones using YOLO-based object detection to assist medical professionals in early diagnosis. Three YOLO models are used YOLO v8, YOLO v10, and YOLO-NAS. Evaluation using a CT images dataset shows a mean Average Precision (mAP) of 90.2%, 89.04%, 88.13% for YOLO v10, YOLO v8 and YOLO NAS respectively. YOLO V10 achieves better performance, with a mAP 50 of 90.20%, mAP 50-95 of 94.20%, precision of 97.30%, and recall of 96.80%. These findings emphasize the model's ability to precisely detect and count the kidney stones, supporting precise diagnosis and treatment planning.