ROI Detection of Hand Bone Based on YOLO V3

China-05 method is the standard method for bone age assessment (BAA) in China. BAA is a typical target detection research project. In this study, a target detection algorithm was proposed combining image processing and deep learning based on the regions of interest (ROI) of 13 hand bones concerned by RUS-CHN method in China-05 method for bone age assessment. We sorted 13 actual ROIs blocks into 4 ROI collection regions (ROI-C), and detected ROI-C through image processing methods such as image equalization, binarization, rotation and positioning. After that, the YOLO V3 model is used to train the new data set composed of ROI-C to detect the actual ROI. This detection algorithm solves the problems of poor robustness, inaccurate detection area positioning, target detection error or detection duplication when we use traditional single image processing or deep learning algorithm. This method greatly improves the detection accuracy of ROI and has great significance for accurate assessment of bone age.