The R-GAN model synthesized high-fidelity T2 parametric maps from OS-CMR images, achieving perfect correlation in edema detection with R² ≥0.9995.
Does a novel R-GAN model accurately synthesize T2 parametric maps from OS-CMR images for myocardial edema detection compared to original T2-weighted images?
A novel R-GAN model can synthesize high-fidelity T2 parametric maps from contrast-free OS-CMR images, potentially offering a time-efficient tool for myocardial edema detection.
Absolute Event Rate: 0% vs 0%
Abstract Background Myocardial edema is a hallmark of acute myocardial injury and is key to diagnosing conditions such as myocarditis and infarction. T2-weighted MRI, which detects edema via prolonged T2 decay times, is the clinical gold standard and often outperforms short tau inversion recovery (STIR). However, it is limited by long acquisition times and protocol variability. Oxygenation-Sensitive Cardiovascular Magnetic Resonance (OS-CMR), a contrast-free sequence, shows potential to reflect both T1 and T2 tissue characteristics. Given the promise of generative adversarial networks (GANs) in medical image synthesis and their limited prior use in T2 mapping we hypothesized that OS-CMR inherently encodes edema-relevant tissue information. Purpose We aimed to synthesize high-fidelity T2 parametric maps directly from native OS-CMR images for correct identification of myocardial edema without having to perform T2 mapping. Methods We developed a novel R-GAN using 2,189 paired OS-CMR and T2-weighted images from two cohorts, with preprocessing that included normalization, registration, and augmentation. Our R-GAN performance was compared to Pix2Pix using peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and Pearson correlation coefficient (PCC). Clinical validation included signal intensity analysis, Extended Phase Graph (EPG) simulations and quantitative T2-curve evaluations in edema and healthy myocardial regions. Results R-GAN achieved a PSNR of 21.56–24.95 dB, SSIM of 0.672–0.739, and PCC of 0.862–0.923 (Table 1). Signal intensities in representative cases closely aligned between original and synthetic maps (e.g., edema: 452 vs. 459; healthy: 283 vs. 269). EPG-based T2-curve simulations demonstrated perfect Pearson and Spearman correlations (1.0000), with MAE ranging from 0.0007 to 0.0185, MSE from 0.0000 to 0.0004, RMSE as low as 0.0008, R² up to 0.9997 and SAD ranged from 0.0029 to 0.0739. In edema ROIs, T2-curve correlation was ≥0.99, with MAE ranging from 0.0041 to 0.0286, MSE from 0.0000 to 0.0017, RMSE from 0.0041 to 0.0286, and R² ≥0.9995. In healthy ROIs, correlation was ≥0.98, MAE ranged from 0.0063 to 0.0595, MSE from 0.00001 to 0.0168, RMSE from 0.0063 to 0.1503, and R² ≥0.9671 and SAD with minimum values of 0.01 for edema and 0.02 for healthy tissue (Figure 1). Conclusion The proposed R-GAN model successfully synthesizes T2 parametric maps from OS-CMR images with high fidelity, enabling accurate detection and localization of myocardial edema, and demonstrating the potential to transform standard functional MRI into a time-efficient, non-invasive tool for structural tissue characterization in cardiac diagnostics.
Lotfikazemi et al. (Thu,) reported a other. The R-GAN model synthesized high-fidelity T2 parametric maps from OS-CMR images, achieving perfect correlation in edema detection with R² ≥0.9995.