Spine-Guided Kidney Localization in KUB Radiographs Using YOLO

The global prevalence of kidney stones among the adult population is estimated to be approximately 10–15%. Kidney–ureter–bladder (KUB) radiography is commonly employed for the detection of kidney stones and also possesses limited capability for identifying calcified lesions. Despite its advantages of being rapid and cost-effective, KUB radiography suffers from a relatively low sensitivity, ranging from 45% to 60%. Furthermore, image quality and diagnostic accuracy are often compromised by obstructions such as intestinal gas and rib overlap, which hinder precise localization of the renal region. These limitations, in turn, affect the performance of subsequent image segmentation and lesion detection tasks.To address these challenges, this study proposes a fully automated kidney localization approach based on the You Only Look Once (YOLO) deep learning model. The proposed method first utilizes the YOLO framework to accurately detect the lumbar spine and then infers the kidney location based on the anatomically stable spatial relationship between the kidneys and the lumbar spine. Experimental results demonstrate that the proposed method significantly enhances kidney localization accuracy and offers a reliable auxiliary tool for clinical analysis of KUB radiographs.