Hand bone radiograph image segmentation with ROI merging

Bone segmentation in radiographic imaging is an intermediate level processing stage for an automated vision system for the skeletal assessment of children. It is one of the challenging problems in medical image analysis due to high noise levels and low contrast with non-uniform and complex intensity distribution of radiographic image. In this paper, we present a local merging algorithm for automatically segmenting bones from the hand radiograph. With an initial over-segmented image, in which the many primitive (homogeneous) regions are generated by watershed transform and image pre-processing, the hand bone X-ray image segmentation is performed by the local merging process on regions of interest (ROIs). Firstly, the hand is separated from the background to get hand boundary. In this phase, aiming the hand separation coincide with the region reduction, an merging algorithm based on the region adjacent graph (RAG) and nearest neighbour graph (NNG) is proposed. Next, the curvature information of the hand boundary is analyzed for determining the desired ROIs on the hand image. Finally, the sub-RAGs which are sub-graph of the RAG associated with the ROI are extracted, and the local merging process on each sub-RAG is individually executed. Experiments are carried out on 30 hand X-ray images of the young children where the carpal bones have distinct, non-overlapping boundaries. The experimental results show that with the proposed method, an accurate and robust segmentation can be achieved.