A geometry-informed deep learning framework for ultra-sparse 3D tomographic image reconstruction.

Area of Science:

• Medical imaging
• Artificial intelligence
• Computational imaging

Background:

• Deep learning models in biomedical imaging offer potential but face generalizability challenges due to their data-driven nature.
• Ultra-sparse sampling in 3D tomographic imaging limits reconstruction quality and clinical applicability.

Purpose of the Study:

• To develop a novel deep learning framework that incorporates geometric information for enhanced 3D tomographic image reconstruction.
• To address the limitations of pure data-driven models in biomedical imaging by integrating prior knowledge.

Main Methods:

• Development of a geometry-informed deep learning framework.
• Introduction of a mechanism for integrating imaging system geometric priors.
• Application to 3D volumetric computed tomography (CT) with ultra-sparse sampling.

Main Results:

• Demonstrated that integrating geometric priors is essential for improving performance in ultra-sparse 3D CT reconstruction.
• Showcased the framework's ability to enhance image quality and model generalizability.
• Validated the effectiveness of the geometry-informed approach.

Conclusions:

• The geometry-informed deep learning framework significantly improves 3D tomographic image reconstruction from ultra-sparse data.
• Integrating known geometric priors is crucial for robust and generalizable deep learning in medical imaging.
• This approach offers promising advancements for clinical imaging and image-guided interventions.