Automatic Quantification of Destructive Changes Caused by Rheumatoid Arthritis

Rheumatoid arthritis is an incurable disease affecting predominantly peripheral joints of the appendicular skeleton. It can lead to severe disabling mutilations and even to the complete destruction of the joints. The accurate and reproducible quantification of the progression of the disease and of the destructive changes caused to the joints is a decisive factor during therapy and during clinical trials. The manual quantification methods are time consuming and lack accuracy as well as reproducibility. In this thesis an alternative approach to automatically quantify the destructive changes caused by rheumatoid arthritis is proposed. Based on a hand radiograph the positions of the bones and joints are determined by local linear mapping nets. They learn the visual appearance as well as the anatomical structure of the hand during a training phase. The ability to learn makes a straightforward transfer of the method to other anatomical structures possible. Based on the coarse position estimates of the bones, the contour is identified with active shape models and snakes. When these methods are combined in the ASM driven snakes algorithm, a control of a priori knowledge utilized during the search for the pathologically changed contour is possible. The resulting description and visual information of the bone contour and its surroundings are used for point wise classification of the contour with respect to the question whether or not individual points are affected by rheumatoid arthritis. The automatic determination of the extent of erosions in the joint region allows for an accurate and operator independent quantification of the disease progression. This research has been partly supported by Austrian Science Fund (FWF) under grant P14445-MAT and P14662-INF