A robust morphological algorithm for automatic radiation field extraction and correlation of portal images

Accurate extraction of the radiation field boundary from portal images is required in selective histogram equalization techniques for contrast enhancement of portal images, and is required in portal‐simulator image correlation. Local edge detectors are so noise sensitive that delineation of the field from anatomy edges and noise is very difficult. Optimization in edge localization and noise suppression can be achieved with optimal edge detectors, but the required calculation is very time consuming. A portal‐image segmentation algorithm based on morphological edge detection and morphological segmentation techniques is presented. Relying on two predefined criteria that are nonsensitive to variation of portal‐image types, the algorithm can automatically search the optimal threshold value, which is sensitive to the variation of the type and quality of portal images. With two stages of the searching procedure, the algorithm can accommodate a large variation of single and double exposure portal images obtained from different therapy machines. Results of our morphological edge detector are also compared to that of an optimal edge detector. Portal‐simulator image correlation using the inertia moments of the radiation field mask is investigated, and compared to correlation obtained using inertia moments of the radiation field contour. It is shown that the mask method is less sensitive to small variations and distortions in the field shape, resulting in a more accurate correlation. This would substantially simplify the task of treatment verifications.