Multiresolution non-rigid image registration method and its application to removal of normal anatomic structures in chest radiographs

摘要

We have developed a novel method for multiresolution image registration based on the wavelet transform, which allows flexible non-rigid (i.e., non-linear) registration of two similar images. In this method, a non-linear mapping that deforms one image to the other is obtained by minimization of a functional representing the squared difference between the two images. A combined approach of a steepest descent algorithm and a multiresolution (i.e., coarse-to-fine) approach is used for the minimization based on the wavelet-domain representation of the non-linear mapping. Our new method was applied to the registration of the left and right lung regions in chest radiographs for removal of the normal anatomic structures such as ribs while abnormal structures such as lung nodules are preserved. We employed the result to reduce the number of false detections obtained with our computer-aided diagnosis scheme for detection of lung nodules in chest radiographs.