RPI ID: 2023-022-301, 2023-022-401
Innovation Summary: This technology provides a full tomographic data estimation system that reconstructs complete tomographic datasets from sparse inputs. It includes preprocessing circuitry that divides sparse data into N subsets, parallel estimation circuitry that reconstructs each subset using a trained score model, and assembly circuitry that combines them into a full dataset. The estimation process may involve solving ordinary differential equations and conditioning on pseudo full data subsets derived from sparse inputs and reconstructed images. The system also includes training circuitry for unsupervised learning of the score model using artificial neural networks.
Challenges / Opportunities: Sparse tomographic imaging often results in incomplete or noisy data, limiting image quality and diagnostic utility. Traditional reconstruction methods like FBP or Fourier transform are insufficient for sparse datasets. This system addresses these limitations by leveraging patch-based parallel estimation and score-based modeling to reconstruct high-fidelity tomographic data. It opens opportunities for safer, faster imaging in CT and MRI by reducing radiation exposure and scan time.
Key Benefits / Advantages: ✔ Parallel processing of tomographic data subsets ✔ Patch-based reconstruction improves spatial accuracy ✔ Score model enables unsupervised learning from training datasets ✔ Reduces radiation exposure and scan duration ✔ Compatible with CT and MRI modalities ✔ Enables high-quality imaging from sparse datasets
Applications: • Medical imaging systems (CT, MRI) with reduced data acquisition requirements
Keywords: Sparse Tomographic Imaging, Patch-Based Reconstruction, Score Model, Parallel Estimation, CT, MRI
Intellectual Property: Patent Application PCT/US2023/080441 Published, US Application no. 19/131116 Pending