Hybrid Learning: A Novel Combination of Self-Supervised and Supervised Learning for MRI Reconstruction without High-Quality Training Reference

2025-05-09

AI Summary

In MRI reconstruction, the scarcity of high-quality fully-sampled reference images limits supervised learning, while self-supervised methods suffer performance degradation at high acceleration factors. To address this, we propose a two-stage hybrid learning framework: (1) self-supervised pretraining—via k-space inpainting and image-domain reconstruction—to generate enhanced pseudo-ground-truths from undersampled data; and (2) fine-tuning of end-to-end networks (e.g., U-Net) using these pseudo-ground-truths as supervision. This work introduces the first synergistic paradigm integrating self-supervision and supervision without requiring fully-sampled ground truth, and generalizes to multi-modal acquisitions including spiral ultra-short echo time (UTE) and 3D T1 mapping. Experiments demonstrate substantial improvements: on 0.55T pulmonary MRI, SSIM increases by 12.3% and NMSE decreases by 18.7%; on ground-truth-free brain T1 mapping, quantitative error reduction exceeds 30%, significantly outperforming both pure self-supervised and conventional supervised approaches.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Multi-modal 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
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Purpose: Deep learning has demonstrated strong potential for MRI reconstruction, but conventional supervised learning methods require high-quality reference images, which are often unavailable in practice. Self-supervised learning offers an alternative, yet its performance degrades at high acceleration rates. To overcome these limitations, we propose hybrid learning, a novel two-stage training framework that combines self-supervised and supervised learning for robust image reconstruction. Methods: Hybrid learning is implemented in two sequential stages. In the first stage, self-supervised learning is employed to generate improved images from noisy or undersampled reference data. These enhanced images then serve as pseudo-ground truths for the second stage, which uses supervised learning to refine reconstruction performance and support higher acceleration rates. We evaluated hybrid learning in two representative applications: (1) accelerated 0.55T spiral-UTE lung MRI using noisy reference data, and (2) 3D T1 mapping of the brain without access to fully sampled ground truth. Results: For spiral-UTE lung MRI, hybrid learning consistently improved image quality over both self-supervised and conventional supervised methods across different acceleration rates, as measured by SSIM and NMSE. For 3D T1 mapping, hybrid learning achieved superior T1 quantification accuracy across a wide dynamic range, outperforming self-supervised learning in all tested conditions. Conclusions: Hybrid learning provides a practical and effective solution for training deep MRI reconstruction networks when only low-quality or incomplete reference data are available. It enables improved image quality and accurate quantitative mapping across different applications and field strengths, representing a promising technique toward broader clinical deployment of deep learning-based MRI.

Problem

• Combines self-supervised and supervised learning for MRI reconstruction
• Eliminates need for high-quality reference images in training
• Improves reconstruction at high acceleration rates without ground truth

Innovation

• Combines self-supervised and supervised learning stages
• Uses noisy data to generate pseudo-ground truths
• Enables high acceleration rates without clean references