Background Current rheumatoid arthritis (RA) MRI synovitis assessment protocols are based on manual, semi-quantitative scoring using RAMRIS.
Objectives To investigate the feasibility of fully automatic synovitis quantification using a novel registration-based approach.
Methods T1-weighted coronal fat-saturated gradient-echo sequences of the wrist were acquired pre- and post-contrast (Gd-DTPA) on 1.5 Tesla scanners. The sequences were acquired at baseline (15 patients) and 18 months later (14 patients) from patients who had RA with active synovitis at baseline. One patient was excluded due to miscalibration between pre- and post-contrast sequences. Synovitis defined as enhancing synovial tissue was quantified for the distal radio-ulnar, radio-carpal, and inter-carpal/CMC joints for each patient at each time point using a novel, fully automatic technique. The synovitis volume was estimated in three steps: 1) the synovium was segmented using multi-atlas freeform registration from four binary atlases of wrist synovia (independent sequences), 2) positioning differences between pre- and post-contrast sequences were corrected using rigid registration, and 3) the synovitis volume was estimated by thresholding the difference image in the region of the segmented synovium. Synovitis was also scored by two radiologists using RAMRIS. As in , the technique is evaluated using the Spearman correlation coefficient (CC) to RAMRIS scores.
Results Correlation for static scoring is reported in Table 1. Correlation for longitudinal scoring, i.e., change in volume vs. change in RAMRIS score from baseline to follow-up, is reported in Table 2. We report the correlation to the RAMRIS scores of the individual readers and to the average scores of the two readers. Moreover, we report the CCs for summarized synovitis measurement, i.e., total RAMRIS scores vs. total automatic scores. The CCs between the radiologists were 0.70 and 0.88 for static and longitudinal summarized scoring.
Conclusions Our results show that fully automatic wrist synovitis measurement from MRI is feasible with good to excellent correlation to manual RAMRIS scores. For comparison, RAMRIS inter-reader correlation scores have been reported in the range 0.58-0.74 .
Chand, A.S., McHaffie, A., Clarke, A.W. et al: Quantifying synovitis in rheumatoid arthritis using computer-assisted manual segmentation with 3 tesla MRI scanning. Journal of Magnetic Resonance Imaging 33(5), 1106-1113 (2011).
Disclosure of Interest P. Mysling: None Declared, E. Dam Shareholder of: Biomediq, Employee of: Biomediq, S. Zaim Employee of: CCBR-SYNARC, H. Genant Shareholder of: CCBR-SYNARC, Employee of: CCBR-SYNARC, T. Fuerst Shareholder of: CCBR-SYNARC, Employee of: CCBR-SYNARC, M. Lillholm Shareholder of: Biomediq, Employee of: Biomediq