Researchers at the University of Virginia have developed a novel generative artificial intelligence framework that significantly enhances the clarity of medical and coherent imaging systems, addressing longstanding diagnostic limitations caused by visual distortion and data scarcity. Spearheaded by Soumee Guha, a Ph.D. candidate in the Department of Electrical and Computer Engineering, the project successfully bridges computational algorithms with established physical imaging models to produce high-fidelity diagnostic visuals. Traditional ultrasound, endoscopic, radar, and laser imaging frequently suffer from speckling and multiplicative noise, which scatter acoustic or optical waves and obscure critical anatomical details. Standard deep learning solutions typically require massive, curated datasets to train, resources that are often unavailable or privacy-restricted in clinical environments. Guhas methodology circumvents this bottleneck by embedding mathematical physics directly into diffusion models. These models progressively strip visual noise from distorted input while preserving underlying structural integrity, generating enhanced images that reflect actual biological or physical conditions rather than synthetic artifacts. The approach delivers immediate clinical utility by producing parallel image sets that clinicians can compare, contrast, or merge with raw equipment output. This comparative analysis aids in accurate disease detection, tumor classification, and volumetric measurements. Furthermore, the high-quality synthetic images generated by the framework can serve as training data for subsequent diagnostic AI systems, creating a scalable pipeline for future medical technology development. Department Chair and research advisor Professor Scott Acton noted that Guhas work represents the first unified application of diffusion models specifically engineered for de-speckling and optical blur correction in coherent imaging modalities. Initiated in fall 2021 and defended in spring 2024, the research transforms abstract mathematical theory into practical healthcare tools. The interdisciplinary methodology addresses the recurring industry challenge where purely data-driven AI models fail to account for the physical constraints of imaging hardware, resulting in clinically unreliable outputs. By prioritizing physics-informed machine learning, the framework ensures that algorithmic enhancements remain consistent with real-world diagnostic requirements. Following her graduation, Guha will pursue postdoctoral research to expand the application of this imaging enhancement architecture across broader biomedical and remote sensing fields. The technology marks a significant step forward in making high-resolution diagnostic imaging more accessible, accurate, and adaptable, ultimately supporting earlier and more reliable disease intervention. As generative AI continues to evolve from content creation to scientific discovery, this framework establishes a new standard for integrating computational innovation with physical reality in medical diagnostics.