Hand Bone X-rays Segmentation and Congregation for Age Assessment using Deep Learning

Bone Age Assessment (BAA) is one of the greatest clinical practices to diagnose growth potential and genetic disorders of a child. Bone X-ray shows the human inside body using ionizing radiation, helping doctors in diagnosing diseases or providing treatment. In this work, we utilize the scans of hand bone X-ray, which is a type of medical images in pediatric hand radiographs for BAA. We propose a 3-step scheme for this task, including segmentation, regression and congregation. Considering 3 main components of one hand including phalanges, metacarpals and carpals as Regions Of Interest (ROI), we segment the images and create 6 new datasets. Thanks to the precisely segmentation of ROIs, our datasets facilitate better BAA scheme by ignoring unnecessary parts, e.g., background of the images. Our proposed regression method with congregation achieves the best mean absolute error of 5.69. Furthermore, extensive evaluations are performed for different ROIs, demonstrating the impact of each part of hand bone images in BAA.