Diffusion and generative stochastic models applied to outstanding problems of modality synthesis, image enhancement and artifact reduction in magnetic resonance imaging

Background
The National Institutes of Health (NIH) intends to negotiate and award a contract without providing for full and open competition to the University of Maryland. The goal of this contract is to develop stochastic transport models, including diffusion models, generative networks, and optimal transport models, aimed at addressing outstanding problems in magnetic resonance imaging (MRI). This includes image synthesis, artifact correction, and resolution enhancements to improve the detection and characterization of normative central nervous system aging and neurodegenerative conditions associated with aging.

Work Details
The contractor will be responsible for the development of advanced stochastic transport models specifically tailored for MRI applications. This includes:
1. Designing diffusion models that enhance image quality;
2. Implementing generative networks for effective modality synthesis;
3. Creating optimal transport models aimed at artifact reduction in MRI images;
4. Conducting research focused on improving resolution for better detection of neurodegenerative conditions;
5. Collaborating with NIH researchers to ensure alignment with ongoing studies related to central nervous system aging.

Place of Performance
Bethesda MD 20892 UNITED STATES

Bidder Requirements
Bidders must demonstrate capability in developing stochastic transport models and have relevant experience in MRI technology. They should provide a technical proposal, a cost-price proposal, and evidence of their qualifications.

Incumbent Analysis (see Incumbents section for more detail)
The notice indicates a confirmed incumbent relationship with contract 75N95025P00404 awarded to the College Park University of Maryland (value $50,058; 09/29/25–07/28/26). The prior award focused on applying mathematical optimal transport theory to image enhancement and artifact reduction in magnetic resonance imaging, which aligns closely with the current effort’s diffusion and generative stochastic models for modality synthesis, image enhancement, and artifact reduction in MRI, suggesting the work may be a continuation or extension of the same modeling approach.