Interpolation and registration of medical image volumes from multiple modalities

In this thesis several methods of interpolating and registering anatomical MRI, CT, and PET image scans are reviewed. A modified linear interpolation proce-dure is developed to create three dimensional volumes from two dimensional image scan formatted according to the DICOM standard. This technique determines a weighted mean intensity for each unknown voxel using weights based on dis-tances from known values. The volumes are then processed using a principal axes registration method which translates the centroids of two volumes to the same position and rotates the volumes so that their principal axes are aligned. The principal axes needed to accomplish this task are determined by computing the eigenvectors of the volume’s covariance matrix. Merging the two procedures pro-vides an efficient method of determining a global rigid transform to register two sets of image scans. Both algorithms have linear computational complexity with respect to the number of voxels in a volume, but the speed of the interpolation method is highly dependent on the separation between image scans. Interpolation speed increases as the distance between scans decreases. This technique provides an excellent coarse registration of images which can be easily followed with more intensive, local methods.