Detecting myocardial amyloidosis with native T1-weighted cardiac magnetic resonance imaging (MRI) remains challenging due to motion artifacts, long acquisition times, and variability across imaging protocols. Results presented at the European Society of Cardiology (ESC) 2025 show that an advanced residual generative adversarial network (R-GAN) can synthesize high-fidelity T1 parametric maps from original-signal cardiac magnetic resonance (OS-CMR).

The study analyzed 1,481 matched OS-CMR and T1-weighted images. R-GAN, enhanced with residual blocks and data augmentation, was compared to Pix2Pix using structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), and Pearson correlation coefficient (PCC). R-GAN with augmentation achieved SSIM 0.810, PSNR 16.1, and PCC 0.748, far exceeding Pix2Pix without augmentation (SSIM 0.158, PCC 0.03). Extended Phase Graph simulations confirmed near-perfect alignment between synthesized and ground-truth T1 curves (Pearson r = 0.99).

These findings indicate that AI-assisted T1 mapping using R-GAN offers a reliable, non-invasive method for myocardial amyloidosis detection. Future work will validate this approach in diverse patient populations and explore its integration into clinical workflows.