GraphSynth: A Graph-Diffusion Framework for Clinical Image and Video Scene Graph Synthesis
Abstract
High-fidelity surgical simulation is increasingly important for medical training, offering an ethical and scalable alternative to cadaver-based practice or training on real patients. However current Virtual Reality (VR) systems, remain expensive, hardware-bound, and often lack anatomical variability. These limitations have accelerated interest in data-driven generative models for surgical images and videos. While diffusion models achieve state-of-the-art visual quality, existing image- and video-generation approaches lack the structural precision needed to model fine-grained anatomy or instrument motion. Surgical scene graphs (SGs) have emerged as a promising structural representation, yet prior work has used them only as conditioning signals rather than generative targets. To address this gap, we introduce GraphSynth, the first diffusion-based framework for surgical SG-to-SG synthesis. Our approach extends two graph-diffusion paradigms: Discrete Denoising Diffusion for Graph Generation (DiGress) and Continuous Graph Denoising Diffusion Model (ConGress), to generate both image-level SGs and multi-frame video-graphs. Our models incorporate semantic node labels and, for the first time, continuous positional and temporal attributes, moving beyond prior diffusion work focused solely on categorical molecular or network graphs. Experiments reveal that discrete positional binning causes severe spatial collapse, making GraphSynth-Discrete unsuitable for modeling object positions. In…
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.