Medical Image Registration using Geometric Hashing

a Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE. We present recent work at INRIA on registering medical images using Geometric Hashing. Here, registering means determining the rigid displacement that aligns two images to obtain the best overlap of anatomical structures represented inside the images. We analyze two methods that perform hashing on geometric invariants and use a voting procedure, inside a six dimensional accumulator implemented using a one dimensional hash table, to determine the best candidate rigid displacements. The rst method uses mostly unary geometric invariants, with an option to use binary invariants, while the second method uses binary invariants exclusively. The rst method uses characteristic ridge curves extracted on the surface of objects for registration purposes, while the second method uses characteristic extremal points. We present registration results using both methods , and report experimental registration accuracy data. We also analyze the computational complexity of the registration and compare with the complexity of alternate registration methods. Using data from a joint study with an industrial partner we show that Geometric Hashing can provide a practical, fast, and accurate method for registering CAT scan images of the same patient.