MInDI-3D: Iterative Deep Learning in 3D for Sparse-view Cone Beam Computed Tomography

2025-08-13

AI Summary

Sparse-view cone-beam CT (CBCT) reconstruction under low-dose conditions suffers from severe artifacts, limiting clinical utility in interventional imaging. Method: We propose the first end-to-end iterative reconstruction framework for real-world clinical CBCT, leveraging a 3D conditional diffusion model. Extending the 2D InDI paradigm to 3D medical image space, our method employs a differentiable inversion model that directly optimizes volumetric reconstructions. It is trained robustly on a large-scale synthetic CBCT dataset (16,182 cases) generated by CT-RATE. Contribution/Results: At only 50 projections, our method achieves a 12.96 dB PSNR gain and enables up to 8× radiation dose reduction. On 16 real cancer patient cases, it matches or exceeds 3D U-Net performance, with clinical evaluation confirming superior tumor boundary preservation. This work represents the first successful deployment of a 3D diffusion model for sparse-view CBCT reconstruction, establishing a new paradigm for low-dose interventional imaging.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

We present MInDI-3D (Medical Inversion by Direct Iteration in 3D), the first 3D conditional diffusion-based model for real-world sparse-view Cone Beam Computed Tomography (CBCT) artefact removal, aiming to reduce imaging radiation exposure. A key contribution is extending the "InDI" concept from 2D to a full 3D volumetric approach for medical images, implementing an iterative denoising process that refines the CBCT volume directly from sparse-view input. A further contribution is the generation of a large pseudo-CBCT dataset (16,182) from chest CT volumes of the CT-RATE public dataset to robustly train MInDI-3D. We performed a comprehensive evaluation, including quantitative metrics, scalability analysis, generalisation tests, and a clinical assessment by 11 clinicians. Our results show MInDI-3D's effectiveness, achieving a 12.96 (6.10) dB PSNR gain over uncorrected scans with only 50 projections on the CT-RATE pseudo-CBCT (independent real-world) test set and enabling an 8x reduction in imaging radiation exposure. We demonstrate its scalability by showing that performance improves with more training data. Importantly, MInDI-3D matches the performance of a 3D U-Net on real-world scans from 16 cancer patients across distortion and task-based metrics. It also generalises to new CBCT scanner geometries. Clinicians rated our model as sufficient for patient positioning across all anatomical sites and found it preserved lung tumour boundaries well.

Problem

• Reducing radiation exposure in sparse-view CBCT imaging
• Extending 2D InDI to 3D for medical artefact removal
• Generating large pseudo-CBCT dataset for robust training

Innovation

• 3D conditional diffusion model for CBCT
• Iterative denoising from sparse-view input
• Large pseudo-CBCT dataset for training