Self-Supervised Weighted Image Guided Quantitative MRI Super-Resolution

TL;DR

The paper tackles the bottleneck of time-intensive high-resolution qMRI by introducing SWIG qMRI SR, a physics-informed, self-supervised framework that uses routinely acquired HR weighted MRI as guidance to upscale LR qMRI without needing HR qMRI ground truth. It formulates SR as MAP with two complementary loss terms, enabling accurate quantitative maps (PD, T1, T2) while preserving measured LR data and leveraging guide information from wMRI. Through synthetic data and in-vivo MuPa-ZTE validation, the approach demonstrates cross-sequence generalizability and improved synthesis of unseen contrasts like T2-FLAIR, suggesting a practical route to integrate fast qMRI into clinical workflows. Ablation studies confirm contrast-specific guidance and the necessity of LR qMRI input, highlighting the framework’s potential to enhance diagnostic quantification without extra acquisition burden.

Abstract

High-resolution (HR) quantitative MRI (qMRI) relaxometry provides objective tissue characterization but remains clinically underutilized due to lengthy acquisition times. We propose a physics-informed, self-supervised framework for qMRI super-resolution that uses routinely acquired HR weighted MRI (wMRI) scans as guidance, thus, removing the necessity for HR qMRI ground truth during training. We formulate super-resolution as Bayesian maximum a posteriori inference, minimizing two discrepancies: (1) between HR images synthesized from super-resolved qMRI maps and acquired wMRI guides via forward signal models, and (2) between acquired LR qMRI and downsampled predictions. This physics-informed objective allows the models to learn from clinical wMRI without HR qMRI supervision. To validate the concept, we generate training data by synthesizing wMRI guides from HR qMRI using signal equations, then degrading qMRI resolution via k-space truncation. A deep neural network learns the super-resolution mapping. Ablation experiments demonstrate that T1-weighted images primarily enhance T1 maps, T2-weighted images improve T2 maps, and combined guidance optimally enhances all parameters simultaneously. Validation on independently acquired in-vivo data from a different qMRI sequence confirms cross-qMRI sequence generalizability. Models trained on synthetic data can produce super-resolved maps from a 1-minute acquisition with quality comparable to a 5-minute reference scan, leveraging the scanner-independent nature of relaxometry parameters. By decoupling training from HR qMRI requirement, our framework enables fast qMRI acquisitions enhanced via routine clinical images, offering a practical pathway for integrating quantitative relaxometry into clinical workflows with acceptable additional scan time.

Figures (4)

• Figure 1: (a) Synthetic data generation for training the framework. (b) Training CNN and inference scheme for SWIG qMRI SR. (c) The ResNet-like CNN.
• Figure 2: Super-resolved parametric maps from ablation study experiments on a representative slice. Rows: PD (top), T1 (middle), T2 (bottom). Columns: ground-truth $Q_h$ (left), model predictions $M_{1}^{++}$ through $M_{12}^{--}$ (center), LR input $Q_l$ (right). Slice-specific SSIM and HFEN relative to $Q_h$ shown below each map.
• Figure 3: Quantitative SR evaluation across 17 test subjects. SSIM (left) and HFEN (right) for PD, T1, and T2 maps. Each point on y-axis represents per-subject metrics averaged over brain-masked axial slices. Red bar: baseline $Q_l$; blue bars: SWIG qMRI SR models.
• Figure 4: (a) Model $M_{12}^{++}$ applied to MuPa-ZTE data. Rows: PD, T1, T2 maps. Columns per protocol (Acq1, Acq5, Acq5$^{\text{R}}$): input $Q_l$, SR output $\hat{Q}_h$. (b) Acquired guides (left) and synthesized images from $Q_l$ and $\hat{Q}_h$ using forward models with acquisition-matched parameters of acquired guides. (c) T2-FLAIR synthesized from $Q_l$ and $\hat{Q}_h$ (never used as guide) versus acquired reference. Metrics in B and C computed for shown slice.