3D medical image analysis using knee MRI

This thesis works on 3D image analysis for effcient cartilage detection, Rician noise removal, visualization and shape analysis of the articular cartilage of the knee using MRI images. Due to the renewed focus on Osteoarthritis, emphasis is now laid on early detection of cartilage degeneration and the use of MRI imaging modality for imaging and visualization of the cartilage tissue. MRI offers superior contrast resolution which helps in visualization of soft tissues such as cartilage and can offer non-invasive diagnostic procedure for OA. Despite these improvements in imaging and quantitative analysis of the cartilage tissue, it is still faced with multiple challenges due to poor resolution, noise and imaging artifacts present in these images. This thesis addresses some of the key issues of noise and limited contrast in knee cartilage analysis for OA commonly observed when using MRI images. It specifcally addresses Rician noise observed in single coil magnitude MRI images which is an inherent noise and affects both image quality and contrast. This noise is both non-linear and signal dependent. This thesis investigates the presence of Rician noise on MRI signal and proposes new imaging techniques to effciently reduce the effect of noise while retaining crucial edge information in addition to improving the overall signal content which will enable in improved cartilage detection and quantifcation. This thesis also offers an improvement in cartilage detection technique despite its poor resolution. For this purpose it incorporates use of 3D un-decimated wavelets and their multiresolution capabilities to achieve better cartilage detection. A novel imaging technique is developed using wavelets multiresolution edge detection and optimization. Local and multiscale edge detection is attained with help of wavelet singularity and better time-scale localization. Role of Rician noise in presence of edge and its localization has also been addressed. Subsequently a 3D model of the cartilage tissue is reconstructed using the volume rendering function. This cartilage volume is also used for a quantitative assessment of the tissue for analysis of OA. In addition this thesis also considers geometric analysis of the cartilage shape as a smooth 2D Riemannian manifold. The variation in cartilage shape amongst individuals and during different stages of OA adds to the complexity of prediction of OA. Due to our assumption of cartilage as Riemannian manifold it can be used for shape analysis and feature extraction of the cartilage by making use of the intrinsic information of the manifold as a global feature. The Riemannian geometry enables us to compute the global nature of the cartilage shape and also its extrinsic variations in the form of Gauss and mean curvatures. As a final part of this study, we have included all the imaging algorithms to form the basis of a new software prototype for OA analyses. This software prototype incorporates wavelet multiresolution GUI for visualization of cartilage surface at higher wavelet resolutions and enables cartilage tissue analysis at these resolutions. The performance of the above imaging algorithms is assessed by their ability to improve the signal to noise ratio, computational effciency and improvement in cartilage analysis inconsideration with respect to other methods. The contributions of this thesis include effcient Rician noise removal and improvement in signal information and contrast ratio of the image, 3D cartilage detection using wavelets and multiresolution addressing the limited resolution information which primarily comprise chapters 2 & 3 of this thesis and have resulted in subsequent journal publications. In addition we address the role of cartilage curvature as Riemannian manifold to account for shape variations and curvature computation using the Riemannian manifold which may be implemented as a MRI imaging biomarker which comprises chapter 4. Finally we introduce a software prototype tool for OA analysis using MRI images in chapter 5 with final conclusions in chapter 6.