--- title: 'GENRE-CMR: Robust Cardiac MRI Reconstruction' url: https://www.emergentmind.com/topics/genre-cmr type: topic --- # GENRE-CMR: Robust Cardiac MRI Reconstruction GENRE-CMR is a supervised, GAN-based framework for accelerated cardiac MRI reconstruction that is designed to remain effective across heterogeneous acquisition domains rather than only on the narrow distribution seen during training. Introduced as “Generalizable Deep Learning for Diverse Multi-Domain Cardiac MRI Reconstruction,” it addresses the standard accelerated MRI inverse problem under undersampled k-space while targeting the practical generalization gap created by variation across imaging centers, scanner vendors, protocols, image contrasts, trajectories, anatomical coverage, and patient populations [2508.20600]. In this formulation, “generalizable” denotes robustness to out-of-distribution CMR data not represented in the training distribution, and “multi-domain” denotes variation across imaging centers, scanner vendors, protocols, image contrasts, trajectories, anatomical coverage/views, and other acquisition conditions [2508.20600]. ## 1. Definition and clinical-technical setting GENRE-CMR operates on multi-coil undersampled k-space and reconstructs an image through a sequence of unrolled reconstruction stages [2508.20600]. The framework is positioned against reconstruction approaches that either perform well but are not robust to domain shift, rely on heavy diffusion models or external natural-image priors, or do not jointly combine unrolled optimization, adversarial learning, and explicit domain alignment [2508.20600]. Its stated objective is a single unified CMR reconstruction model that can handle diverse multi-domain data and maintain strong performance on unseen distributions [2508.20600]. The method is explicitly dual-domain. K-space is used for data fidelity, physical consistency, and the unrolled update; image space is used for learned refinement, SSIM-based reconstruction, and edge-aware supervision [2508.20600]. It is also explicitly adversarial: a generator performs reconstruction using a residual deep unrolled architecture, and a discriminator is trained adversarially to encourage more realistic reconstructions [2508.20600]. This positioning places GENRE-CMR within a broader trajectory of generative and structure-aware CMR methods. Earlier work on post-reconstruction artifact correction had already shown that adversarial learning and edge-focused supervision could sharpen CMR outputs in image space, although in a different problem setting centered on motion artifact correction rather than accelerated reconstruction [1902.11121]. A plausible implication is that GENRE-CMR extends that broader pattern from post hoc enhancement to physics-constrained undersampled reconstruction while shifting the emphasis from single-domain restoration to multi-domain robustness. ## 2. Residual deep unrolled reconstruction framework The central architectural novelty is a residual deep unrolled reconstruction framework [2508.20600]. GENRE-CMR builds on an earlier 2024 model and modifies it by adding residual connections between consecutive cascaded modules so that feature information from shallow stages is preserved and reused by later-stage reconstructors [2508.20600]. The authors argue that this improves convergence, hierarchical representation learning, and gradient flow [2508.20600]. The generator consists of 16 reconstructors, also described as cascades or unrolled steps [2508.20600]. At each step, the current k-space estimate is transformed to image space, coil sensitivity information is used for coil combination and processing, a convolutional reconstruction subnetwork predicts a feature or refinement term, that prediction is added residually to the propagated feature from the previous stage, the refined image representation is mapped back into multi-coil k-space, and a data-consistency-like update enforces agreement with acquired samples [2508.20600]. The algorithm is given as: \[ k^{(ACS)} \leftarrow \mathrm{Auto\_calibration\_signal}(k^{(0)}) \] \[ I^{(ACS)} \leftarrow \mathrm{FFT^{-1}}(k^{(ACS)}) \] \[ S_m \leftarrow \mathrm{APUnet}^{(SME)}(I^{(ACS)}), \qquad S_m' \leftarrow \mathrm{Conjugate\_Symmetry}(S_m) \] Residual features are initialized as \[ F^{(-1)} \leftarrow 0 \] and for each unrolled step \(t \in \{0,\dots,T-1\}\) with \(T=16\), \[ I_{MC} \leftarrow \mathrm{FFT^{-1}}(k^{(t)}), \qquad I_{SC} \leftarrow S_m' \cdot I_{MC} \] \[ \hat{F}^{(t)} \leftarrow \mathrm{APUnet}^{(t)}(I_{SC}, P^{(t)}) \] \[ F^{(t)} \leftarrow \hat{F}^{(t)} + F^{(t-1)} \] \[ I'_{RF} \leftarrow \mathrm{Repeat\_interleaved}(F^{(t)}), \qquad I_{SS} \leftarrow I'_{RF} \cdot S_m, \qquad G_k \leftarrow \mathrm{FFT}(I_{SS}) \] \[ k^{(t+1)} \leftarrow k^{(t)} - \eta^{(t)} \mathscr{M}(k^{(t)}-k^{(0)}) + G_k \] where \(\eta^{(t)}\) is a learnable step size, \(\mathscr{M}\) is the subsampling mask, and \(k^{(0)}\) is the acquired undersampled k-space [2508.20600]. This update reflects the unrolled optimization view: a learned reconstruction term \(G_k\) is combined with a measurement-consistency correction that pushes the estimate back toward the observed k-space data [2508.20600]. The paper states that the reconstructors are APUnet modules and that prompt learning, inspired by PromptMR, is incorporated to support adaptive reconstruction across different contrasts, trajectories, and anatomical views within one model [2508.20600]. However, the implementation specifics of the prompt design are not fully detailed [2508.20600]. Likewise, the paper does not provide exact numbers of convolution layers per APUnet, kernel sizes, feature channel counts, normalization layers, activation functions, or discriminator depth [2508.20600]. This suggests that the paper’s main architectural contribution lies more in the residual unrolled design and its integration with domain-generalization losses than in low-level module novelty. ## 3. Objective function and domain-generalization mechanisms GENRE-CMR defines a composite objective consisting of fidelity, Edge-Aware Region loss, and Statistical Distribution Alignment loss [2508.20600]: \[ \mathcal{L}_{\text{total}} = \lambda_1 \mathcal{L}_{\text{Fidelity}} + \lambda_2 \mathcal{L}_{\text{EAR}} + \lambda_3 \mathcal{L}_{\text{SDA}} \] In practice, adversarial loss is also used through GAN training, so the generator objective is described as \[ \mathcal{L}_{\text{Total-Gen}} = \mathcal{L}_{\text{Gen}} + \mathcal{L}_{\text{GAN-Gen}} \] [2508.20600]. The fidelity term combines a physical k-space consistency term and an image-domain SSIM loss [2508.20600]. For each stage \(t\), \[ \mathcal{L}^{(t)}_{\text{Phys}} \leftarrow \mathrm{MSE}\big(\mathrm{Mag}(k^{(t+1)}_{\text{central}}), \mathrm{Mag}(k^{(G)})\big) + \mathrm{MSE}\big(\Phi(k^{(t+1)}_{\text{central}}), \Phi(k^{(G)})\big) \] \[ \mathcal{L}^{(t)}_{\text{SSIM}} \leftarrow \mathrm{SSIM\_Loss}\big( \mathrm{CC}(\mathrm{FFT^{-1}}(k^{(t+1)}_{\text{central}})), \mathrm{CC}(\mathrm{FFT^{-1}}(k^{(G)})) \big) \] \[ \mathcal{L}^{(t)}_{\text{Fidelity}} \leftarrow \mathcal{L}^{(t)}_{\text{Phys}} + \mathcal{L}^{(t)}_{\text{SSIM}} \] where \(\mathrm{CC}\) denotes coil combination, \(\mathrm{Mag}\) and \(\Phi\) denote magnitude and phase, and \(k^{(G)}\) is ground-truth k-space [2508.20600]. The Edge-Aware Region loss is a central contribution. It is motivated by the tendency of deep reconstructions to appear smooth while blurring diagnostically important boundaries such as myocardial borders and small vessels [2508.20600]. Given ground-truth image \(I_{\text{gt}}\), Sobel gradients are computed: \[ G_x = S_x * I_{\text{gt}}, \qquad G_y = S_y * I_{\text{gt}} \] and the edge magnitude map is \[ M = \sqrt{G_x^2 + G_y^2} \] [2508.20600]. To enlarge the neighborhood around edges, the map is smoothed with a \(5\times5\) averaging kernel \(A\): \[ M_s = A * M \] A binary mask is then defined with threshold \(\tau=0\): \[ B_{i,j} = \begin{cases} 1 & \text{if } M_s(i,j) \ge \tau \\ 0 & \text{otherwise} \end{cases} \] The mask is applied to reconstructed and ground-truth images: \[ \tilde{I}_{\text{rec}} = B \odot I_{\text{rec}}, \qquad \tilde{I}_{\text{gt}} = B \odot I_{\text{gt}} \] and the loss is \[ \mathcal{L}_{\text{EAR}} = 1 - \mathrm{SSIM}(\tilde{I}_{\text{rec}}, \tilde{I}_{\text{gt}}) \] [2508.20600]. The intended effect is to penalize structural degradation near edges rather than treating all pixels equally. The Statistical Distribution Alignment loss is the most explicit generalization mechanism [2508.20600]. The training data are assumed to come from five domains, \[ \mathcal{D}_1, \mathcal{D}_2, \dots, \mathcal{D}_5 \] and groups of five consecutive samples are organized so that each group includes one sample from each distribution [2508.20600]. For each sample \(x_t^{(i)}\), feature vectors are extracted from each reconstructor \(l \in \{1,\dots,16\}\): \[ \mathbf{f}_t^{(i,l)} \in \mathbb{R}^d \] For each reconstructor \(l\), the domain-specific feature set is \[ \mathcal{F}_l = \{\mathbf{f}_t^{(1,l)}, \dots, \mathbf{f}_t^{(5,l)}\} \] Assuming each domain’s features at layer \(l\) follow a Gaussian \(\mathcal{N}(\mu_i^l,\Sigma_i^l)\), the SDA loss at layer \(l\) is \[ \mathcal{L}_{\text{SDA}}^{(l)} = \frac{1}{10} \sum_{i