Physics-Guided Diffusion Priors for Multi-Slice Reconstruction in Scientific Imaging

2025-12-07

AI Summary

Medical and scientific imaging—particularly multi-slice reconstruction—faces ill-posedness, excessive GPU memory consumption, and limited accuracy due to data scarcity. To address these challenges, this work proposes a physics-constrained, patch-based diffusion prior framework. It tightly couples a differentiable physical forward model with a lightweight, partitioned diffusion prior, enabling patch-wise reconstruction that reduces GPU memory usage by over 60% compared to full-slice baselines. The framework supports joint optimization across modalities—including MRI and 4D-STEM. Quantitatively, it achieves superior in-distribution reconstruction fidelity over conventional physics-driven methods and end-to-end diffusion models, while demonstrating strong out-of-distribution generalization. Crucially, it is the first method to simultaneously ensure high-fidelity reconstructions and break the traditional memory–performance trade-off barrier.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Accurate multi-slice reconstruction from limited measurement data is crucial to speed up the acquisition process in medical and scientific imaging. However, it remains challenging due to the ill-posed nature of the problem and the high computational and memory demands. We propose a framework that addresses these challenges by integrating partitioned diffusion priors with physics-based constraints. By doing so, we substantially reduce memory usage per GPU while preserving high reconstruction quality, outperforming both physics-only and full multi-slice reconstruction baselines for different modalities, namely Magnetic Resonance Imaging (MRI) and four-dimensional Scanning Transmission Electron Microscopy (4D-STEM). Additionally, we show that the proposed method improves in-distribution accuracy as well as strong generalization to out-of-distribution datasets.

Problem

• Reduces memory usage per GPU while maintaining high reconstruction quality.
• Improves accuracy for both in-distribution and out-of-distribution datasets.
• Integrates partitioned diffusion priors with physics-based constraints for reconstruction.

Innovation

• Physics-guided diffusion priors for multi-slice reconstruction
• Partitioned diffusion priors reduce GPU memory usage
• Integrates physics constraints to enhance reconstruction quality