Idea
A CT reconstruction framework combining diffusion models and data consistency for improved sparse-view medical imaging quality.
Research Paper
Core Innovation
This paper presents DICE, a novel framework that integrates a two-agent consensus equilibrium into diffusion model sampling for sparse-view CT reconstruction. It uniquely balances a data-consistency agent with a diffusion model prior agent iteratively, improving image quality beyond traditional methods. This approach effectively handles undersampled data while capturing complex medical image structures.
Market Size (TAM)
$20–50B TAM for medical imaging software; $2–10B SAM from hospitals and imaging centers adopting advanced CT reconstruction. Driven by demand for lower radiation dose scans and improved diagnostic accuracy.
Potential Customers & Pain Points
- Hospitals needing faster high-quality CT scans with fewer views
- Medical imaging centers reducing radiation exposure
- CT device manufacturers seeking advanced reconstruction algorithms
Business Model
Licensing reconstruction software to medical device manufacturers and imaging centers; offering cloud-based reconstruction services; partnerships for embedded solutions in CT scanners.
Competitive Landscape
- GE Healthcare
- Siemens Healthineers
- Canon Medical Systems
Implementation Challenges
- Regulatory approval for clinical use
- Integration with existing CT hardware and workflows
- Computational resource requirements for diffusion models
Validation Strategy
- Conduct clinical trials comparing image quality and diagnostic accuracy
- Pilot deployments in partner hospitals for workflow integration
- Benchmark against existing reconstruction methods on diverse datasets
Research Paper Overview
DICE: Diffusion Consensus Equilibrium for Sparse-view CT Reconstruction
Summary
Sparse-view CT reconstruction is challenging due to undersampling causing ill-posed inverse problems. Traditional iterative methods use handcrafted or learned priors but struggle with complex medical image structures. Diffusion models (DMs) offer powerful generative priors for accurate image distribution modeling. This work introduces Diffusion Consensus Equilibrium (DICE), integrating a two-agent consensus equilibrium into DM sampling. DICE alternates between a data-consistency agent enforcing measurement consistency and a prior agent using a DM for clean image estimation. This iterative balance combines strong generative priors with measurement consistency. Experiments show DICE outperforms state-of-the-art baselines in reconstructing high-quality CT images under uniform and non-uniform sparse-view settings of 15, 30, and 60 views out of 180, demonstrating effectiveness and robustness.