Image registration of pulmonary SPECT and CT exams

Abstract

n the last years, medical imaging field has been had an enormous technological development. The application of 3D images extended its utility, becoming an essential tool to assist in the clinical diagnosis. There are several imaging modalities that allow to visualize different information about the body. In particular, with Computed Tomography (CT) anatomical information is obtained and with Single Photon Emission Computed Tomography (SPECT) the body metabolism is detected. By applying image processing methods it is possible to combine different visual information through multimodality registration and consequently improve the clinical diagnosis. In this thesis, a study of registration algorithms was done on 3D CT and SPECT images, in order to select the most appropriated transformation to identify bronchopulmonary segments with suspected pulmonary embolism. Rigid (rigid body, global rescale, specific rescale and affine) and non-rigid (diffeomorphic demons) transformations were applied. For each registration algorithm, processing time, image overlay and similarity measurements were analyzed. Affine transformation showed the best performance by allowing the combination of a good alignment and a low time to process. Therefore, it was considered the most suitable registration algorithm to identify suspected pathologic lung regions.