Deep learning of MRI contrast enhancement for mapping cerebral blood volume from single-modal non-contrast scan with Mamba3D-CNN hybrid model

Motivation: MRI contrast agents, such as Gadolinium, are essential for high-resolution mapping of brain metabolism but pose health risks due to their invasive intravenous administration. Goal(s): To address these risks, we introduce a novel deep learning approach utilizing a 3D patch-based Mamba model to replace traditional contrast agents and enhance imaging quality. Approach: This study is the first to apply a 3D patch-based Mamba model specifically for this purpose. Results: Our model surpasses previous methods in estimating cerebral blood volume with sub-millimeter resolution, validated on MRI scans of aging and Alzheimer's patients, and is clinically applicable using a single T1-weighted pre-contrast scan. Impact: By accurately estimating cerebral blood volume, this approach eliminates risks associated with gadolinium administration, such as long-term tissue retention. This advancement enables functional imaging for researchers and clinicians, providing a safe and cost-effective alternative for studying and diagnosing neurodegenerative diseases.