DIFFERENT APPROACHES TO SYNOVIAL-MEMBRANE VOLUME DETERMINATION BY MAGNETIC-RESONANCE-IMAGING - MANUAL VERSUS AUTOMATED SEGMENTATION

Automated fast (5-20 min) synovial membrane volume determination by MR I, based on pre-set post-gadolinium-DTPA enhancement thresholds, was e valuated as a substitute for a time-consuming (45-120 min), previously validated, manual segmentation method. Twenty-nine knees [rheumatoid arthritis (RAJ 13 osteoarthritis (OA) 16] and 17 RA wrists were examin ed. At enhancement thresholds between 30 and 60%, the automated volume s (Syn(x%)) were highly significantly correlated to manual volumes (Sy n(Man)) (knees: rho = 0.78-0.91, P < 10(-5) to < 10(-9) wrists: rho = 0.87-0.95, P < 10(-4) to 10(-6)). The absolute lute values of the auto mated estimates were extremely dependent on the threshold chosen. At t he optimal threshold of 45%, the median numerical difference from Syn( Man) was 7 ml(17%) in knees and 2 ml (25%) in wrists. At this threshol d, the difference was not related to diagnosis, clinical inflammation or synovial membrane volume, e.g., no systematic errors were round, Th e inter-MRI variation, evaluated in three knees and thr;tl wrists. was higher than by manual segmentation, particularly due to sensitivity t o malalignment artefacts. Examination of test objects proved tile high accuracy of the general methodology for volume determinations (maxima l error 6.3%). Preceded by the determination of reproducibility and th e optimal threshold at the available MR unit, automated 'threshold' se gmentation appears to be acceptable when changes rather than absolute values of synovial membrane volumes are must important e.g. in clinica l trials.