Computer-Aided Diagnosis (CAD) of Knee Osteoarthritis based on Magnetic Resonance Imaging for Quantitative Pathogenesis Analysis and Visualization

Abstract

Segmentation of articular cartilage (AC) from magnetic resonance images (MRI) plays a vital role in computer-aided diagnosis (CAD) of the presence and progression of knee osteoarthritis (OA). Segmentation allows a quantitative and qualitative analysis of the knee cartilage from the required morphological point of view and is an essential aspect of the clinical evaluation of knee osteoarthritis as it makes early diagnosis of pathological changes in the cartilage possible. This segmentation of the cartilage is often conducted manually or is done using various semi-automatic segmentation approaches that make the results of this segmentation not easily reproducible due to intra and inter observer variations. Given that the segmentation of AC is challenging due to its changing anatomical structure, semiautomatic segmentation methods have shown better technical advantages in terms of reliance on training data and user interactions. Conventional semi-automatic segmentation models require a large amount of user interaction; making them less ideal for real-time use in CAD systems. In recognition of this limitation, we propose a hybrid segmentation model combining random walkers with deep learning-based initialisation and level set based refinements of the segmented cartilage boundary. Finally, local cartilage degradation analysis will be performed to monitor the disease progression utilising fractal image features computed from the segmented cartilage and Ordered Value (OV) method computed using a thickness map coupled with AdaBoost classifier.