Deep Learning Framework for Gadolinium-free MR Imaging

Motivation: Potential health risks, higher costs, and longer acquisition times associated with gadolinium (Gd)-enhanced MR imaging underscore the need for alternatives. Goal(s): We leverage shared tissue-specific information from unenhanced multi-contrast MR images (T1w, T2w, and FLAIR) to synthesize high-quality T1-weighted contrast-enhanced images using deep learning. Approach: A model is pretrained with self-supervised contrastive learning to capture local tissue-specific MR contrast information from unenhanced scans. The representations from T1w, T2w, and FLAIR images then enable synthesis of gadolinium-free T1-weighted images. Results: Integrating tissue-specific MR contrast information in the synthesis of contrast-enhanced images improved image-quality evaluation metrics such as SSIM, PSNR, and LPIPS. Impact: Our approach leverages multi-contrast MR images to capture tissue-specific information through a self-supervised contrastive learning framework. By synthesizing gadolinium-free T1-weighted images from unenhanced scans, we aim to retain diagnostic quality while potentially reducing the related risks, costs, and acquisition times.