Deep Radon Prior: A fully unsupervised framework for sparse-view CT reconstruction.

Area of Science:

• Medical Imaging
• Deep Learning
• Computed Tomography

Background:

• Sparse-view CT reduces radiation but causes artifacts.
• Existing deep learning methods require large labeled datasets and lack interpretability.

Purpose of the Study:

• Introduce a fully unsupervised deep learning framework for sparse-view CT reconstruction.
• Mitigate dependency on labeled data and improve interpretability.

Main Methods:

• Propose the Deep Radon Prior (DRP) framework, integrating a neural network as an implicit prior.
• Utilize image and Radon domain gradient feedback for progressive neural network optimization.
• Narrow the solution space in the Radon domain for under-constrained protocols.

Main Results:

• DRP demonstrates convergence properties and produces high-fidelity images.
• Significantly reduces artifacts compared to existing methods.
• Achieves performance comparable or superior to supervised methods.

Conclusions:

• DRP advances sparse-view CT by leveraging Radon domain self-correlation with neural networks.
• Offers a scalable, interpretable, and unsupervised approach to CT reconstruction.
• Represents a paradigm shift with potential to redefine medical imaging.