Abstract: Tanner-Whitehouse 3 (TW3) method assesses bone age by finding 13 Regions of Interest (ROIs) from left hand X-ray image and evaluating each region. TW3 method has complicated evaluation process and many assessment factors are subjective. Hence, an automated bone age assessment system that enables objective evaluation in short period of time is required. Automatic extraction of 13 ROIs is necessary for the implementation of the automated bone age assessment system. However, direct extraction based on deep-learning mechanism can produce erroneous extraction of ROI in wrong region because many regions in left hand bone have similar appearance. To prevent this problem, finding Big ROIs from the left hand X-ray image first and extracting the corresponding ROIs from each Big ROI can increase the probability of correctly finding ROIs. This study proposes a method of finding thumb Big ROI, middle finger Big ROI and little finger Big ROI from left hand X-ray image and evaluates the performance of the method.