Bone Scan Image Segmentation based on Active Shape Model for Cancer Metastasis Detection

Image segmentation is commonly used as a preliminary step before the recognition or detection system. In the segmentation process, firstly, we have to divide the image into several particular regions based on specific rules, while hoping the results highly represent the desired patterns. In the case of metastasis detection, bone scan image segmentation is a step that is necessary to be conducted before metastasis detection. The performance of these segmentation results will significantly influence metastasis detection results. Therefore, in this paper, we propose a bone scan image segmentation based on Active Shape Model. In the segmentation process, we divide the bone scan image into 4 (four) different regions, namely (1) head and spine region, (2) upper arms and collarbone region, (3) chest region, and (4) pelvis and upper thighs region. In the annotation step, we use 200 landmark points in total, with each region is represented by 50 landmark points. This annotation step is a necessary step of the model building based on Active Shape Model. We evaluate our proposed system by using 3-cross validation method and obtain a low cumulative error distribution, which is 0.004694.