Abstract
Conference Title: 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) Conference Start Date: 2018, Dec. 3 Conference End Date: 2018, Dec. 6 Conference Location: Madrid, Spain Knee osteoarthritis (OA) is the most common form of arthritis and the major cause of activity limitation and physical disability in older people. Quantitative measures of cartilage on MRI (Magnetic Resonance Imaging) represent potentially powerful surrogate endpoints in knee OA. However, manual segmentation and measurement of the knee cartilage are time-consuming tasks and are not sensitive to detect progression change. In this paper, we proposed a novel whole knee cartilage quantification method based on informative locations called Cartilage Damage Index (CDI). Instead of labeling the entire 3D MR sequence, we focused on the informative locations which are more likely characterized by cartilage loss. We conducted statistical studies for CDI, compared them with traditional manual segmentation, and found that CDI has high correlations with manual segmentation. CDI also shows other promising characteristics: good measurement reliability (ICC $0.90 \sim 0.98)$, a significantly shorter measurement time $(\sim 15$ mins VS 6 hours), and better sensitivity to detect the slow progression of cartilage loss (SRM: -0.65 VS -0.11). We compared the correlation of CDI and manual segmentation with other severity measurements (joint space width and knee alignment) of OA disease and the results demonstrated that CDI is comparable or better than manual segmentation as a novel biomarker to detect the progression of knee osteoarthritis.