Medical Image Registration Based on Moving Manifold Regularization
Wei Zhang, Huabing Zhou, Yicheng Yang, Changcai Yang, Zhenghong Yu · 2019
We propose a new registration framework, namely moving manifold regularization, for solving medical image registration problems. The proposed method first uses the Snake model to obtain the region of interest on the images, and then extracts the two landmark point sets from the segmented images. After that, a transformation model based on the two landmark point sets is used to register the medical images. This framework considers the error functional with moving manifold regularization as we formulate this transformation model as a vector-field interpolation problem. The regularized error function moves by pixels, and the manifold regularization is used as the prior of the transformation to capture the underlying geometry information of the input data. The transformation function for each pixel is specified in a special space namely Reproducing Kernel Hilbert Space. Extensive experiments on medical images showed that the proposed method, which overcomes the defects of manual selection error and improves the registration accuracy, is an effective and stable registration method.