Nonrigid registration of cardiac DSCT images by integrating intensity and point features. (January 2019)

• Record Type: Journal Article
• Authors: Xie, Qinlan; Chen, Zhao; Chen, Hong; Lu, Xuesong
• Is Part Of: Biomedical signal processing and control. Volume 47(2019)
• Page Start: 224
• Page End: 230
• Publication Date: 2019-01
• Subjects: Dual-source computed tomography -- Mutual information -- Marching cubes -- Adaptive stochastic gradient descent; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2018.08.039
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A novel registration method was proposed to obtain accurate anatomical information for the whole heart. The proposed method is used to improve the segmentation accuracy of each part of the heart due to the performance of nonrigid registration. Robust feature descriptors are used to generate the matched point pairs. The registration process is divided into coarse alignment and accurate registration. Two groups of experiments show that the proposed method achieves higher registration accuracy than traditional mutual information. Abstract: In order to obtain accurate anatomical information for the whole heart, we propose a nonrigid registration method combining mutual information with the marching cubes method. Certain points are regarded as prior knowledge of the shape landmarks of cardiac structures. The registration process for the heart image can be divided into two steps: coarse alignment and accurate registration. The coarse alignment uses affine transformations to localize and center the image of the heart, and the accurate registration uses a B-spline method to constrain the deformation field. Mutual information combined with feature point pairs are used as the similarity measure function. All 15-dimensional feature descriptors are used to identify matched point pairs between marching cubes points in atlas intensity images and other points in the neighborhood of target images needing segmentation. Adaptive stochastic gradient descent optimization is used to obtain optimal registration parameters. Two groups of experiments show that the proposed method achieves higher registration accuracy than traditional ones based only on mutual information. They indicate that accurate anatomical information of the whole cardiac structure can be obtained.