Biomedical image segmentation for precision radiation oncology

The development of medical imaging technologies contributes to precision radiation oncology, cancer diagnosis, and treatment. With multimodality biomedical images generated during various treatment processes, computer-assisted diagnosis makes it possible for rich information from images to be analyzed and evaluated comprehensively and efficiently. The accurate region of interest or target object segmentation from medical images plays an indispensable role in computer-assisted diagnosis. In this chapter, we firstly introduce graph theories and graph models designed for image segmentation. Then we review the applications of graph theoretical models in target object segmentation from single-modality and multimodality biomedical images. Secondly, region-based neural networks in object detection and segmentation are reviewed, especially multiscale location-aware kernel representation. The applications of deep networks in medical image segmentation are also presented. We finally address the essential of complete tumor segmentation and quantitative computing of metabolic subvolumes in tailored dose painting for precision oncology and personalized treatment planning.