Automatic Segmentation of Features in Medical CT Imagery

A method is described for fusing the results of different feature detection algorithms in order to segment regions in C T (computed tomography) images. While many feature detectors are available, each has its own characteristics strengths and weaknesses. In this work we aim to anticipate the complementary nature of such detectors, and so exploit strengths and tolerate weaknesses in each. To achieve this, the system is initially trained, so that the expected responses from each feature detector can be determined. A likelihood function then provides a combined measure of the similarity of each image pixel to the target region of interest and to the background region. This provides a means to segment the region of interest from the background using all information provided by the detectors. During segmentation, constraints are applied so that the shape of the extracted region closely resembles the anticipated appearance of the target region of interest.