AbstractsBiology & Animal Science

Visual scoring of magnetic resonance images for the early detection of rheumatoid arthritis is prone to human subjectivity and lacks sensitivity. In a bid to develop an objective and quantitative alternative using digital image processing, this thesis proposes automatic segmentation of the carpal bones, followed by the quantification of bone marrow edema, which is an important inflammatory imaging biomarker. Segmentation of the carpal bones is achieved using multi atlas-based segmentation. Compared to manual segmentations of the training data, an average Dice overlap of 0.85 was achieved. By examining contrast-enhanced MR images of the wrist, edematous bone is classified from normal bone marrow using knowledge based fuzzy clustering. Validation of the quantitative score against the existing RA MRI Scoring (RAMRIS) system showed a significant positive correlation. Segmentation error was seen to be a confounding factor, limiting the specificity of the BME measure. To increase agreement and maximise the available information, it is recommended that data from a complementary imaging plane is included.