Fat-based registration of breast dynamic contrast enhanced water images.

Purpose: In this study, a 3D fat-based deformable registration algorithm was developed for registering dynamic contrast-enhanced breast images.

Methods: The mutual information similarity measure with free-form deformation motion correction in rapidly enhancing lesions can introduce motion. However, in Dixon-based fat-water separated acquisitions, the nonenhancing fat signal can directly be used to estimate deformable motion, which can be later used to deform the water images. Qualitative comparison of the fat-based registration method to a water-based registration method, and to the unregistered images, was performed by two experienced readers. Quantitative analysis of the registration was evaluated by estimating the mean-squared signal difference on the fat images.

Results: Using a scale of 0 (no motion) to 2 ( > 4 voxels of motion), the average image quality score of the fat-based registered images was 0.5 ± 0.6, water-based registration was 0.8 ± 0.8, and the unregistered dataset was 1.6 ± 0.6. The mean-squared-signal-difference metric on the fat images was significantly lower for fat-based registered images compared with both water-based registered and unregistered images.

Conclusions: Fat-based registration of breast dynamic contrast-enhanced images is a promising technique for performing deformable motion correction of breast without introducing new motion. Magn Reson Med 79:2408-2414, 2018. © 2017 International Society for Magnetic Resonance in Medicine.