Two-and-a-half order score-based model for solving 3D ill-posed inverse problems

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence in Medicine

Background:

• Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) are vital medical imaging modalities.
• Score-based models show promise for inverse problems like sparse-view CT and fast MRI reconstruction.
• Existing score-based models struggle with accurate 3D volumetric reconstruction, causing inter-slice inconsistencies.

Purpose of the Study:

• To introduce a novel Two-and-a-Half Order Score-based Model (TOSM) for improved 3D volumetric reconstruction in CT and MRI.
• To address the limitations of 2D-focused score-based models in generating consistent 3D medical images.
• To enhance the accuracy and reliability of solving 3D ill-posed inverse problems in medical imaging.

Main Methods:

• Developed TOSM, a novel score-based model that trains on 2D data distributions for efficiency.
• During reconstruction, TOSM leverages complementary scores from sagittal, coronal, and transaxial directions for precise 3D output.
• The model is grounded in robust theoretical principles for reliable performance.

Main Results:

• TOSM achieved state-of-the-art (SOTA) results on large-scale sparse-view CT and fast MRI datasets.
• Demonstrated an average improvement of 1.56 dB PSNR for sparse-view CT reconstruction across 29 views.
• Achieved an average improvement of 0.87 dB PSNR for MRI reconstruction with ×2 acceleration.

Conclusions:

• TOSM effectively resolves 3D ill-posed inverse problems by modeling 3D data distributions, unlike prior 2D approaches.
• The proposed method significantly reduces inconsistencies in reconstructed 3D volumetric images.
• TOSM offers a robust and effective solution for enhancing CT and MRI reconstruction quality.