Premium
TU‐CD‐BRA‐10: Hybridized Deformable Registration Framework for Contrast‐Enhanced Dedicated Breast CT
Author(s) -
Gazi P,
Boone J
Publication year - 2015
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4925607
Subject(s) - image registration , artificial intelligence , contrast (vision) , affine transformation , computer vision , segmentation , computer science , voxel , metric (unit) , image segmentation , cluster analysis , image processing , pattern recognition (psychology) , mathematics , image (mathematics) , operations management , pure mathematics , economics
Purpose: Utilization of contrast enhancement in dedicated breast CT (bCT) has been reported to be a reliable metric in determining the conspicuity of malignant breast lesions. In this study, we have designed and developed a framework of image segmentation and registration to align the structure of pre‐ and post‐contrast breast CT image. Methods: An iterative two‐means clustering method was used in image segmentation. The image segmentation algorithm results in segmenting the breast CT image to skin, adipose, fibroglandular and contrast‐enhanced lesions. These results are used in image registration method. A deformable image registration method code‐named Intensity Difference Adaptive Demons (IDAD) was developed based on the Demons image registration. Within the developed framework, the deformation field forces are calculated considering the contrast enhancement levels in the bCT image voxels. The performance of the developed framework was evaluated using mathematical simulations and patient breast CT images. Results: The proposed method outperformed conventional affine and other Demons variations for serial pre‐contrast and post‐contrast breast CT image alignment. In simulation studies, IDAD exhibited 1–11% improvement in Normalized Cross Correlation (NCC) compared to the conventional Demons approach with the improvement increasing with lesion size and contrast enhancement levels. Registration error measured by Target Registration Error (TRE) shows only submillimeter mismatches between the concordant anatomical target points in all patient studies. The implementation of the presented hybridized framework was implemented based on a parallel processing architecture, resulting in rapid execution time for the iterative segmentation and intensity‐adaptive registration techniques. Conclusion: Characterization of contrast‐enhanced lesions is improved using IDAD. Spatial subtraction of the aligned images yields useful diagnostic information with respect to lesion morphology.