A novel approach to registration of RGB images of human hands

Spatial normalization permits pixel-level image comparison and analysis. It is used for the analysis of images of the same patient at different time points as well as different patients, imaged with the same or different imaging modalities. The method, mostly used on brain images, shows potential for its use on other body parts. Applying it to human hands could facilitate monitoring of a disease progression, such as rheumatoid arthritis, or comparison between patients.
We present semi-automatic registration method. It uses anatomical landmarks of a hand, and an extrapolation technique to wrap it to a template. The use of the algorithm is presented on 12 hands. Registration was performed by the use of landmarks selected visually on RGB image. Further, the registration was benchmarked against the ground truth registration, where the landmarks were determined by a rheumatologist using ultrasound. The average difference in pixel position between final images of both cases was 1.3 mm, with standard deviation of 1.4 mm.
We demonstrated that semi-automatic method for landmar selection successfully coregister images to pre-defined template. Besides neuroimaging, the method could make a major impact in the field of rheumatology.