Semi-automatic segmentation of knee osteoarthritic cartilage in magnetic resonance images

Knee osteoarthritis is one of the major socio-economic burdens of today. Magnetic resonance imaging facilitates analysis of disease progression by visualization of structural and biochemical changes in cartilage tissue. Segmentation of cartilages from magnetic resonance images is therefore important in clinical investigations. Today, segmentations are obtained using time-consuming manual or semi-automatic algorithms that are subject to some degree of inter- and intra-observer variabilities. Automated methods are rarely used in clinical practice but have obvious advantages over manual methods and the potential to improve clinical workflow. This paper presents an algorithm for segmentation of knee articular cartilage in magnetic resonance images. The method is semi-automatic and requires a minimal amount of manual intervention. The proposed method is tested on scans from 50 subjects with all degrees of knee osteoarthritis as defined by the Kellgren-Lawrence grading scale and achieves an average sensitivity, specificity and dice similarity coefficient of 0.853±0.093, 0.999±0.001, 0.800±0.106 and 0.831±0.095, 0.999±0.001, 0.777±0.054 on tibial and femoral cartilages respectively. The method allows for segmentation of pathological cartilage in clinical investigations.