Grading and quantification of hip osteoarthritis severity by analyzing the spectral energy distribution of radiographic hip joint space

An image analysis system is proposed for the assessment of hip osteoarthritis (OA) severity. Sixty four hips (18 normal, 46 osteoarthritic), corresponding to 32 patients of unilateral or bilateral hip OA were studied. Employing custom developed software, 64 Region Of Interest (ROI) images of Hip Joint Spaces (HJSs) were delineated on patients' digitized radiographs. The Fourier spectrum of each HJS-ROI was computed and expressed in polar coordinates. Spectral signatures, quantifying the radial and angular distribution of HJS spectral energy were formed. Signature descriptors were generated and utilized in the design of a two-level hierarchical decision tree, used for the grading of the severity of the disease. Accordingly, at Level 1, implemented by a multiple classifier system, the discrimination between normal and osteoarthritic hips was performed. At Level 2, the hips that had been successfully characterized as osteoarthritic at Level 1, were further characterized as of 'Mild / Moderate' or 'Severe' OA, by the Bayes classifier. A signature descriptors based regression model was designed, so as to quantify OA-severity. The system graded OA reliably, given that the accomplished classification accuracies for Level 1 and Level 2 were 98.4% and 100%, respectively. OA-severity values, expressed by HJS-narrowing, correlated highly (r = 0.9, p < 0.001) with values predicted by the model. The system may contribute to OA-patient management.