Fast 3D Model Reconstruction of the Knee Articular Cartilage and Automatic Detection of Chondral Damage/Osteoarthritic Craters using Image Processing Methodologies

Abstract

The motivation of the research is to develop a 3D visualization tool to assist a consultant radiologist in the diagnosis of Chondral injury/osteoarthritic (OA) craters that exist in Magnetic Resonance (MR) images of the femoral cartilage of the human knee. The thesis presented addresses this by making three scientific contributions. The first contribution is in the area of segmentation by presenting the development of a simple, robust and user-friendly semi-automatic segmentation method which we term the “Region-Based Segmentation and Bounding Box” (RBS&BB) method. A useful feature of this method is that it requires four mouse clicks to segment the articular cartilage from MR images. The second contribution is in the field of surface reconstruction by introducing a novel fusion of an adaptation of the contour method known as ‘Contour Interpolation (CI)’ with Radial Basis Functions (RBFs) which we describe as ‘Contour Interpolated/RBFs (CIRBF)’. In addition, we also present a spline boundary correction method which enhances the volume estimation of the method. The CI-RBF method significantly reduces data points required and significantly improved the computation speed over the comparable methods. The third contribution is in the field of hole-filling by the development of an image processing tool to rapidly and automatically detect 3D render and accurately quantify Chondral injury/OA craters that exist in MR images of the femoral cartilage. This was achieved by adapting the Moshtagh ellipsoid method and its novel combination with Harris’s corner detection method. The 3D rendering of the OA crater was achieved by our developed CI-RBF method. We validate the performance of our proposed method by performing three investigations. Firstly, we validate the method against synthetic craters, of known radii, generated in the femoral cartilage. Secondly, we demonstrate that the developed method can detect knees with pathology from a randomized sequence of knees with and without pathology). Finally, we show how the method can determine the volume of real OA craters and compare this to the manual delineation method. It is hoped that the techniques developed in this thesis will be of benefit to the future researchers in the field of segmentation, surface reconstruction and the detection of OA.