MRecover: A Conditional Generative Model for Recovering Motion-Corrupted MR images Using AI Generated Contrast

2026-05-22 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition, Health & Medical Research · Depth: Expert, quick

Summary

MRecover, a conditional generative model, was developed to address motion artifacts in high-resolution T2w turbo spin echo (TSE) MRI, which often leads to significant data loss, particularly for hippocampal subfield segmentation. This model synthesizes routinely acquired T1w images to generate TSE images, employing autoregressive slice conditioning to ensure volumetric consistency. Trained on 577 subjects using 7T MRI data, MRecover demonstrated high in-domain fidelity with SSIM=0.84 and FSIM=0.94 on 148 subjects. It also generalized effectively to out-of-domain 3T data from 416 subjects, where subfield volumes from synthesized and acquired images showed strong correlation (r=0.87-0.97). Crucially, the model increased analyzable subjects in the motion-affected ADNI3 dataset by 31.8% (from 450 to 593) and improved effect sizes for diagnostic group differences in hippocampal subfield atrophy, with whole hippocampus ϵ^2 values ranging from 0.121-0.100 compared to 0.086-0.062 for left-right hemispheres.

Key takeaway

For research scientists analyzing neuroimaging data, particularly those dealing with motion-corrupted MRI scans, MRecover offers a robust solution to recover valuable data. You should consider integrating this conditional generative model to synthesize high-resolution T2w TSE images from routinely acquired T1w scans. This approach can significantly increase your analyzable subject count by 31.8% and enhance the statistical power for detecting diagnostic group differences in hippocampal subfield atrophy.

Key insights

MRecover uses a conditional generative model to reconstruct motion-corrupted MRI, significantly increasing analyzable data and diagnostic power.

Principles

• Generative models can synthesize high-resolution MRI from standard scans.
• Autoregressive conditioning ensures volumetric consistency in image synthesis.
• AI-generated contrast enhances diagnostic utility from corrupted data.

Method

MRecover is a conditional generative model that synthesizes T1w images into TSE images using autoregressive slice conditioning, trained on 7T MRI data to recover motion-corrupted MR images.

In practice

• Apply MRecover to salvage motion-affected T2w TSE MRI scans.
• Integrate T1w images as input for generating high-quality TSE images.
• Increase sample sizes for neuroimaging studies by recovering unusable data.

Topics

• Conditional Generative Models
• MRI Reconstruction
• Motion Artifact Correction
• Hippocampal Subfield Segmentation
• Neuroimaging
• Medical Image Synthesis

Best for: AI Scientist, Research Scientist

Related on AIssential

• Recover Semantics First, Generate Better: Improved Latent Modeling for 3D MRI Reconstruction and Cross-Contrast Synthesis — Prioritizing semantic recovery in 3D MRI latent space compression significantly improves reconstruction fidelity and cross-contrast synthesis.
• Impact of Synthetic Lesional MR Images in Automated Focal Cortical Dysplasia Detection in Low-Data Scenarios — Synthetic FCD-MRIs generated by conditional networks can augment real data, reducing labeled data needs for automated detection.
• Multi-Contrast MRI Motion Correction via Parameter-Informed Disentanglement and Adaptive Experts — A unified MRI motion correction framework disentangles contrast and adapts to severity for robust, generalizable artifact removal.
• Treatment-Conditioned Diffusion for Forecasting Neurodegenerative Disease Progression — A treatment-conditioned diffusion framework accurately forecasts neurodegenerative disease progression by integrating imaging and pharmacological data, preserving anatomical detail.
• From Baseline to Follow-Up: Counterfactual Spine DXA Image Synthesis in UK Biobank Using a Causal Hierarchical Variational Autoencoder — The CHVAE model synthesizes anatomically plausible, causally consistent spine DXA images by intervening on age, addressing anatomical variation challenges.
• BrainDINO: A Brain MRI Foundation Model for Generalizable Clinical Representation Learning — Large-scale slice-wise self-supervised learning yields a unified, transferable brain MRI representation for diverse neuroimaging tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.