RISING: A new framework for model-based few-view CT image reconstruction with deep learning.

Area of Science:

• Medical Imaging
• Computational Imaging
• Deep Learning in Medical Applications

Background:

• Medical image reconstruction from low-dose tomographic data is challenging.
• Deep learning offers superior results but can be unstable.
• Classical optimization methods have limitations in speed and accuracy.

Purpose of the Study:

• To propose RISING, a novel framework for sparse-view tomographic image reconstruction.
• To combine the strengths of model-based iterative algorithms and deep learning.
• To achieve accurate and fast image reconstruction from low-dose data.

Main Methods:

• RISING employs a two-step approach: an early-stopped Rapid Iterative Solver followed by an Iteration Network-based Gaining step.
• The first phase uses a regularized model-based algorithm for a few iterations.
• The second phase utilizes a convolutional neural network to complete the reconstruction.

Main Results:

• RISING demonstrates numerical and visual accuracy in reconstructed images.
• The framework is ground-truth free, combining data-driven flexibility with model-based sparsity constraints.
• Experiments on digital and real abdomen datasets confirm the method's effectiveness.

Conclusions:

• RISING achieves accurate medical image reconstruction from low-dose tomographic data.
• The framework offers computational speed and flexibility, making it suitable for clinical applications.
• RISING shows promise for real-time use in clinical imaging systems.