Three Image Experts which help Distinguish Lung Tumors from Non-Tumors

摘要

The design of three vision-expert programs is described. The vision experts are based on appearance models of objects in a typical chest radiograph which were derived using information from human experts. Each expert is based on a different technical concept. The rib expert uses relational constraints on parameter (Hough transform) space to determine if a rib has been detected; the vascularity expert uses a back-projected Hough transform; and the nodule expert uses features which were derived from studies of the nodule recognition process of radiologistics. The efficacy of the descriptive models and their implementations are evaluated. The process of development and implementation of image experts are discussed.