TU‐CD‐BRA‐07: A Comparison of Demons Image Registration Algorithms to Monitor Longitudinal Changes in Knee Cartilage: Data From the OsteoArthritis Initiative (OAI)

• Published in: Medical physics (Lancaster) Vol. 42; no. 6Part32; p. 3607
• Main Authors: Hoang, U, Sinha, U
• Format: Journal Article
• Language: English
• Published: American Association of Physicists in Medicine 01.06.2015
• Subjects: Image registration; Jacobians; Magnetic resonance; Medical imaging
• ISSN: 0094-2405; 2473-4209
• DOI: 10.1118/1.4925604

Purpose: To develop accurate tools to quantify small and localized changes in cartilage morphology to facilitate comparisons between patient cohorts with varying degrees of osteoarthritis (OA) as well as to track longitudinal changes(normal progression and response to treatment). Methods: We implemented and compared the registration accuracy of four variants of the demons algorithm on magnetic resonance cartilage image volumes. The algorithms are four variants of the Demon's algorithm: symmetric simple demons (SMSD), symmetric evolved demons (SMED), inverse‐consistent simple demons (ICSD), and inverse‐consistent evolved demons (ICED). The registration algorithms were also evaluated for the accuracy of the average Jacobians. The average Jacobian of the cartilage was compared to the ratio of volume change for validation. Evaluation was performed on 36 subjects using the baseline and later time point images acquired after 12 months. Results: The symmetric evolved demons algorithm provided the best in registration accuracy evaluated using quantitative metrics of mean squared error and voxel overlap. The symmetric simple demons and symmetric evolved demons performed equally well in terms of the Jacobians. Conclusion: Approximately 27 millions adults age 25 and older diagnosed with OA; however, cartilage loss with disease progression is small and localized to sub‐regions of the cartilage. It is desirable to develop accurate methods that can automatically extract measures of cartilage volume and thickness (global and local). The techniques developed here will be used, in future studies, to explore differences in cohorts segregated by disease severity and correlation of local changes to clinical variables.