Multimodal Medical Image Registration Algorithms for the Kinematic Analysis of Knee Joints

Measuring the kinematics of joints and implants following orthopaedic surgery is important. While the Roentgen stereophotogrammetric analysis (RSA) method has high accuracy, it is an invasive method and not suitable for dynamic motion measurements. Alternatively video/optical tracking systems are non-invasive but suffer from low accuracy. 3D Computed Tomography (CT) to 2D single plane fluoroscopy image (3D-2D) registration methods are also non-invasive but have low accuracy in directions normal to the X-ray plane and require small initial displacements between the two images to be registered. It is these two limitations of 3D-2D registration algorithms that will be the focus of this thesis work. A new 3D-2D registration algorithm is presented for knee joints with large in-plane initial displacements. This approach uses a combination of the Log-Polar transform (LPT) and a gradient-descent (GD) based approach. A novel approach to pre-processing the images allows the algorithm to register the unsegmented fluoroscopy images. Experimental results show that the new algorithm is able to register images with much larger displacements than the GD based algorithm. An additional benefit of this approach is the reduced computational complexity. The previously proposed registration algorithm is then improved to handle large initial out-of-plane displacements. The improved registration framework utilizes a LPT mapping for out-of-plane translation, a conjugate directional search for out-of-plane rotations, linear pixel mapping between the multimodal images, and a fast digitally reconstructed radiograph generation method. A novel non-invasive, accurate kinematic analysis approach is also presented. The experimental results show that the proposed registration approach can increase the range of initial displacements while maintaining high accuracy in motion parameters. A limitation to accurate automatic tracking of knee motion by the proposed method is the presence of noise and blurring effects in the fluoroscopy frames. The previously proposed 3D-2D registration algorithm is improved using a dual-tree complex wavelet transform (DTCWT) based denoising approach. The experimental results show that the proposed approach can maintain high accuracy and repeatability. The kinematics analysis results for the algorithm without the de-noising show higher inter-patient variance than for the results provided by the algorithm using the proposed DTCWT based de-noising process