Research map / Image generation / Fast diffusion sampling
CardioComposer: Leveraging Differentiable Geometry for Compositional Control of Anatomical Diffusion Models
Karim Kadry, Shoaib A. Goraya, Ajay Manicka, Abdalla Abdelwahed, Naravich Chutisilp, Farhad R. Nezami, Elazer R Edelman
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use inference-time geometric guidance on an unconditional model instead of a conditional model with dropout, with 0 constrained substructures (ablation) | CardioComposer · Compositional generation (0 constraints) | FD: 1622 -> 34.6 | Improved |
| use geometric guidance instead of implicit conditioning with dropout, with 1 constrained substructure (ablation) | CardioComposer · Compositional generation (1 constraint) | FD: 227 -> 38.5 | Improved |
| use geometric guidance instead of implicit conditioning with dropout, with 6 constrained substructures (ablation) | CardioComposer · Compositional generation (6 constraints) | FD: 31.1 -> 35.5 | Hurt |
| add guidance to the unconditional model with all geometric losses (ablation) | CardioComposer · Cardiac label maps (24 seeds) | Size fidelity (x1e5): 106.03 -> 9.34 | Improved |
| add RV geometric guidance (ablation) | CardioComposer · Cardiac label maps (24 seeds) | Aorta Betti error: 1.208 -> 1.458 | Hurt |
| increase RV target mass factor from 1.0 to 4.0 (ablation) | mass-only guidance · RV mass editing | FMD: 40.3 -> 1690.0 | Hurt |
| decrease RV target mass factor from 1.0 to 0.5 (ablation) | mass-only guidance · RV mass editing | Size error (x1e5): 12.60 -> 19.95 | Mixed |
| add shape loss guidance (ablation) | CardioComposer · Myocardium disentanglement | mass fidelity | Improved |
| increase training set from 20% to 40% and beyond (ablation) | latent diffusion prior · Cardiac label maps | conditional fidelity under guidance | No clear effect |
Closest papers by meaning
- A Diffusion Model for Simulation Ready Coronary Anatomy with Morpho-skeletal Control (ECCV 2024 · similarity 0.88)
- AnaDiffusion: Anatomically Compositional Latent Diffusion for Controllable 3D Brain MRI Generation (NeurIPS 2026 · similarity 0.84)
- VasTSD: Learning 3D Vascular Tree-state Space Diffusion Model for Angiography Synthesis (CVPR 2025 · similarity 0.83)
- EchoXFlow: A Beamspace Echocardiography Dataset for Cardiac Motion, Flow, and Function (NeurIPS 2026 · similarity 0.83)
- LiFT: Lifted Inter-slice Feature Trajectories for 3D Image Generation from 2D Generators (NeurIPS 2026 · similarity 0.82)
- Learning General-purpose Biomedical Volume Representations using Randomized Synthesis (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
