Predicting Delayed Phase Contrast-Enhanced MR Images from Early Phase Contrast-Enhanced MR Images Using Deep Learning-Based Iterative Network

Motivation: In breast MRI, early-phase contrast enhancement trends are key indicators for differentiating malignant from benign tissue. However, traditional imaging methods require extended time to capture these dynamic changes. Goal(s): This study proposed a deep-neural network to map time-dependent information in contrast-enhanced breast MRI, enabling the temporal prediction of enhancement and aiming for abbreviated scans. Approach: We designed an iterative network to sequentially generate post-contrast breast MR images at delayed time points, leveraging pre-contrast and early phase contrast-enhanced images. Results: Results from fibroglandular tissue and tumor regions demonstrated the potential of our method for predicting the delayed phase image from the early phase image. Impact: By enabling dynamic contrast prediction in breast MRI, our method aids in the characterization of enhancement patterns in breast tissue using only early phase post-contrast images. This approach potentially reduces scan times for dynamic contrast-enhanced MR applications.