Portal image feature extraction by hierarchical region processing technique

We describe in this paper a novel hierarchical region processing technique for the extraction of features from portal images. Feature extraction is an essential component in the computerized registration of portal images for radiation treatment verification and patient setup error minimization. Unlike the existing methods with global processing techniques, the proposed technique is able to extract features at different levels through forming a pyramid of regions-of-interest. Such pyramid is formed according to the inherent hierarchy of portal image features which are often evident only at corresponding levels. A sequence of operating steps, including histogram manipulation, Hough transform, optimal Canny edge detection, and pyramid feature merging, are employed to accomplish the processing at different levels of the regions-of-interest. With such processing hierarchy, we are able to overcome the difficulties encountered by the existing global processing-based techniques and produce accurate field edges and sufficient and meaningful anatomical landmarks for the subsequent feature matching and error correction procedures.