End-to-end Adaptive Dynamic Subsampling and Reconstruction for Cardiac MRI

TL;DR

This work introduces E2E-ADS-Recon, an end-to-end framework that jointly learns an Adaptive Dynamic Sampler (ADS), a Sensitivity Map Predictor (SMP), and dynamic MRI reconstruction to produce case-specific sampling patterns for accelerated cardiac MRI. It supports frame-specific or unified 1D/2D Cartesian subsampling and optimizes acquisition and reconstruction across varying acceleration factors $R$, using ACS-derived sensitivity maps and STE-based binarization for differentiable sampling. Across the CMRxRecon dataset, E2E-ADS-Recon yields superior reconstruction quality (SSIM, PSNR, NMSE), with the most pronounced gains at high accelerations and when frame-specific patterns are learned, demonstrating the value of leveraging temporal correlations in dynamic MRI. The framework is robust to reconstruction backbone choices (vSHARP or MEDL-Net) and emphasizes the practical significance of case-specific subsampling to improve image fidelity and patient comfort in dynamic MRI workflows, while noting limitations related to the need for fully-sampled training data and calibration data for SMP initialization.

Abstract

$\textbf{Background:}$ Accelerating dynamic MRI is vital for advancing clinical applications and improving patient comfort. Commonly, deep learning (DL) methods for accelerated dynamic MRI reconstruction typically rely on uniformly applying non-adaptive predetermined or random subsampling patterns across all temporal frames of the dynamic acquisition. This approach fails to exploit temporal correlations or optimize subsampling on a case-by-case basis. $\textbf{Purpose:}$ To develop an end-to-end approach for adaptive dynamic MRI subsampling and reconstruction, capable of generating customized sampling patterns maximizing at the same time reconstruction quality. $\textbf{Methods:}$ We introduce the End-to-end Adaptive Dynamic Sampling and Reconstruction (E2E-ADS-Recon) for MRI framework, which integrates an adaptive dynamic sampler (ADS) that adapts the acquisition trajectory to each case for a given acceleration factor with a state-of-the-art dynamic reconstruction network, vSHARP, for reconstructing the adaptively sampled data into a dynamic image. The ADS can produce either frame-specific patterns or unified patterns applied to all temporal frames. E2E-ADS-Recon is evaluated under both frame-specific and unified 1D or 2D sampling settings, using dynamic cine cardiac MRI data and compared with vSHARP models employing standard subsampling trajectories, as well as pipelines where ADS was replaced by parameterized samplers optimized for dataset-specific schemes. $\textbf{Results:}$ E2E-ADS-Recon exhibited superior reconstruction quality, especially at high accelerations, in terms of standard quantitative metrics (SSIM, pSNR, NMSE). $\textbf{Conclusion:}$ The proposed framework improves reconstruction quality, highlighting the importance of case-specific subsampling optimization in dynamic MRI applications.

Figures (26)

• Figure 1: Overview of the Adaptive Dynamic Sampler for frame-specific $k$-space sampling. Initial multi-coil dynamic $k$-space measurements $\tilde{\mathbf{y}}_{0}$ and sensitivity maps $\mathbf{S}$ are used to perform a SENSE reconstruction, which serves as the input to the encoder $\mathcal{E}_{\boldsymbol{\theta_1}}$ in the ADS module. The encoder output is flattened and passed to a multi-layer perceptron $\mathcal{M}_{\boldsymbol{\psi_1}}$, which generates probability vectors for each potential sampling location. These probabilities are softplused, rescaled to meet the acceleration factor $R$, and binarized using a straight-through estimator layer to produce the sampling pattern $\mathbf{U}_{\Lambda}$, ensuring AF($\Lambda$) = $R$. For brevity here we assume that $\tilde{\mathbf{y}}_{0}$ comprises only of ACS data and we also assume only a single ADS cascade ($N=1$).
• Figure 2: Overview of the E2E-ADS-Recon pipeline or frame-specific $k$-space sampling. Initial multi-coil dynamic $k$-space measurements $\tilde{\mathbf{y}}_0$ include ACS data $\tilde{\mathbf{y}}_{\text{ACS}}$, which are used by the SMP module to generate coil sensitivity maps $\mathbf{S}$. These sensitivity maps and the initial measurements are input to the ADS module, which generates adaptive sampling patterns $\mathbf{U}_\Lambda$ based on the desired acceleration factor $R$. These patterns guide subsampled $k$-space acquisitions during dynamic imaging. The subsampled data $\tilde{\mathbf{y}}$ are processed with the sensitivity maps $\mathbf{S}$ in the reconstruction module (e.g., vSHARP), yielding reconstructions $\hat{\mathbf{x}}$. The pipeline, including ADS, SMP and reconstruction model, is jointly optimized end-to-end to enhance imaging quality. For simplicity, the illustration assumes a single ADS cascade ($N=1$).
• Figure 3: SSIM ($\times 100$) metrics across all experimental settings and setups. Diamonds ($\Diamond$) on the box-plot median indicate the average best methods. A star ($\star$) on the upper whisker indicates non-significance in comparison to the average best method.
• Figure 4: Examples of 1D patterns across setups at $R=8$ for (a) frame-specific and (b) unified settings. Black: fixed/initial, red: learned pattern. Cyan boxes mark SSIM values.
• Figure 5: Example of reconstructions across setups for unified 1D sampling at $R=8$.