Automated segmentation and registration of the kidney in CT datasets

A framework for segmentation and registration of medical volumetric datasets is presented. Segmentation and registration are helpful for diagnosis and treatment planning because they allow determining change in datasets of a patient acquired at different times by a medical scanning device. Focus here is on segmentation and registration of lower torso Computerized Tomography (CT) datasets that contain the kidney, especially on kidney registration.The framework's two most significant parts are its segmentation and registration modules. The segmentation is done by a new technique that allows the kidney to be extracted from lower torso CT datasets. The technique combines active contour (snake)-, intensity-, and shape-based processing to extract the kidney. The registration is aided by two new kidney representation models. The models allow registration to be accelerated significantly. Two new registration techniques are also described. One registration technique extends the mutual information method to volumetric kidney datasets. The other registration technique creates a new Extended Gaussian Image (EGI)-based kidney representation model and performs registration on the EGI-based models. Both techniques follow a coarse-to-fine process.In addition, an extension of a kidney localization method is presented. This extension achieves a better performance than existing methods in localizing the kidney in rotated and truncated lower torso CT datasets.