Three dimensional deformable image registration and registration verification
Metadata[+] Show full item record
Deformable image registration (DIR) is a fundamental problem in medical image processing. Due to the lack of ground truth, such problem known as registration verification without 'golden standard' standard is still an unsolved scientific question. We propose to apply well developed techniques in the computer vision field, including the feature/region/edge based detectors, descriptors and matching for deformable image registration and registration verification. Specifically, this thesis explores the feasibility of applying three dimensional scale invariant feature transform (SIFT), Harris-Laplace and maximally stable extremal regions (MSER), canny edge and iterative closest point (ICP) in the verification of registering magnetic resonance imaging (MRI) and computed tomography (CT) volume data. Our preliminary results show that the feature points/edges/regions provided by the aforementioned techniques are good candidates for the 'golden standard' key points. These features may be used as a powerful aid to human experts or even may be used alone in registration verification.