Automated radiographic assessment of hands in rheumatoid arthritis

Rheumatoid arthritis (RA) is a chronic inflammatory disease leading to severe joint damage, particularly in the wrist, fingers and toes. To prevent irreversible joint damage, it is crucial that RA is treated in an early stage. The effect of treatment methods can differ per patient. Therefore it is important that disease activity is carefully monitored, such that a treatment can be altered if necessary. In order to visualize joint damage, radiographs of hands and feet are often used. In the past fifty years several methods have been developed to express visible joint damage in a quantifiable score. In general these methods are time-consuming and depend on subjective visual readings. With the introduction of the computer and the availability of digitized radiographs, researchers have been looking for new automated methods to measure the progression of joint damage. A measurable effect in early RA is joint space narrowing which is caused by degradation of the cartilage between the bones. A considerable part of this thesis deals with this subject. Firstly, different methods are evaluated on how to quantify the joint space width in an accurate and robust manner. Hereby methods developed by others have been evaluated. Subsequently, we present several image processing methods for analyzing radiographs. The first is a segmentation method based on a model of the hand skeleton. With this model it is possible to detect the locations of the bones and the joints in radiographs. The second part describes a method that accurately detects the bone margins within the joint. Finally, the joint space width is calculated using the detected joint margins. This method is evaluated by comparing measurement results with both manual measurements, and automated measurements developed by others. It is shown that automated measurements are more precise than manual measurements. It was observed from the image data, that a large variability of hand positioning is allowed during x-ray acquisition. This does not only complicate automated analysis, but can also have an effect on the projection angle within a specific joint. In successive radiographs, a change in the projection angle can negatively affect the precision of measurements. By using projection simulations it is demonstrated that this effect is relevant. Therefore, we recommend to standardize the protocol for radiographic acquisitions. A special positioning aid has been designed, which can be used to place a hand in a standard position. Another measurable effect of RA is the development of bone damage in the form of erosions. To be able to measure this effect, we investigated the possibility to make subtraction measurements between successive radiographs that have been taken with some time in between. It is demonstrated that different radiographs of the same bone can be compared using an image registration algorithm. By determining the difference between two images developing erosions can be revealed. However, with the current available radiographs it is not yet possible to quantify bone damage. We expect that in the future this will become feasible by applying the recommended positioning aid during x-ray acquisition.