Accelerating Longitudinal MRI using Prior Informed Latent Diffusion

2024-06-29

arXiv.org

AI Summary

Long acquisition times for longitudinal MRI, difficulty modeling anatomical/pathological changes across sessions, and the scarcity of paired longitudinal k-space data hinder deep learning–based reconstruction. Method: We propose a self-supervised, prior-guided reconstruction framework that requires no paired data. It integrates a latent-space diffusion prior, a differentiable data consistency module, and an unsupervised prior fusion architecture to dynamically balance historical scan priors with current undersampled measurements while enforcing k-space data consistency—thereby avoiding overfitting. Contribution/Results: To our knowledge, this is the first unpaired, label-free, self-supervised training paradigm specifically designed for longitudinal MRI. Evaluated on public healthy datasets and multiple clinical longitudinal cases, our method significantly outperforms state-of-the-art approaches, enabling robust reconstruction at acceleration factors ≥8× while preserving anatomical fidelity and visibility of critical pathological features.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• User Modeling, Personalization and Recommendation: Studies of user behavior, including longitudinal effects of personalized systems

Abstract

MRI is a widely used ionization-free soft-tissue imaging modality, often employed repeatedly over a patient's lifetime. However, prolonged scanning durations, among other issues, can limit availability and accessibility. In this work, we aim to substantially reduce scan times by leveraging prior scans of the same patient. These prior scans typically contain considerable shared information with the current scan, thereby enabling higher acceleration rates when appropriately utilized. We propose a prior informed reconstruction method with a trained diffusion model in conjunction with data-consistency steps. Our method can be trained with unlabeled image data, eliminating the need for a dataset of either k-space measurements or paired longitudinal scans as is required of other learning-based methods. We demonstrate superiority of our method over previously suggested approaches in effectively utilizing prior information without over-biasing prior consistency, which we validate on both an open-source dataset of healthy patients as well as several longitudinal cases of clinical interest.

Problem

• Accelerating MRI acquisition using prior subject scans
• Overcoming anatomical changes between longitudinal scan sessions
• Eliminating need for paired training data in reconstruction

Innovation

• Uses diffusion model for MRI reconstruction
• Integrates prior DICOM scans during inference
• Eliminates need for paired training data