An evaluation of image registration methods for chest radiographs

Image registration is commonly used in medical applications for revealing changes in different series of images. In this study, the performances of different registration scenarios based on different feature extraction and matching methods were assessed in the context of chest radiographic images. For this purpose, combination of three well known key point descriptors (SIFT, SURF and ORB) were used as feature detectors. For feature matching, SIFT and SURF methods were also employed individually. The tests were conducted on chest X-ray images of real patient data taken at different times. The accuracies of the registered images were assessed by two different validation algorithms. The experiments revealed that the highest registration accuracy is achieved when SIFT and SURF descriptors are used together for key point extraction, and SIFT algorithm is used for feature matching.