Towards High-Fidelity Medical Visual Generation: Structured Generative Modeling for Restoration and Synthesis
Abstract
Generative models have shown strong potential for medical image restoration and synthesis, yet their clinical use remains challenging because visually plausible outputs may not preserve subtle anatomical structures. This thesis investigates structured generative modeling for high-fidelity medical vision, where explicit priors and conditions guide restoration and generation beyond direct pixel-level prediction. Five complementary frameworks are developed. LighTDiff introduces a lightweight T-shaped diffusion architecture for surgical endoscopic low-light enhancement, improving efficiency while preserving structural details. EDiTSR reformulates endoscopic super-resolution through sample-adaptive restoration operators and operator-space distillation, enabling structurally guided reconstruction without high-resolution inputs at inference. EndoIR addresses all-in-one endoscopic restoration through spatial--frequency prompting, task-adaptive embeddings, and noiseaware routing, allowing a single model to handle diverse and mixed degradations without explicit labels. SemanticIR integrates anatomical semantic priors and frequency-aware guidance into medical image restoration, improving semantic consistency and reducing structurally unsupported reconstruction. Finally, SurgSora extends structured modeling to controllable surgical video generation by incorporating object segmentation, depth cues, and trajectory-based motion conditions. Together, these studies form a progressive…
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.