AI-Driven Generation of Post-Contrast T1 and ECV Maps from Native T1 Map in Cardiac MRI

Background/Objectives: This study aimed to develop an artificial intelligence-based method for generating virtual post-contrast T1 maps and extracellular volume (ECV) maps from native T1 maps and to evaluate its performance. Methods: The proposed method was based on a modified self-consistent recursive diffusion bridge framework to generate virtual post-contrast T1 maps from native T1 maps. Cardiac magnetic resonance (CMR) data were collected from consecutive patients with suspected myocardial disease. A total of 813 well-registered image slices were selected for model development and evaluation. On an unseen test set of native T1 maps, the trained model generated virtual post-contrast T1 maps, which were subsequently combined with the corresponding native T1 maps to compute ECV maps. Results: The myocardial T1 values derived from the reference and virtual post-contrast T1 maps revealed similar distributions, although a systematic offset between the distribution peaks was observed. Following ECV transformation, this offset was substantially reduced. In the held-out test cohort, the virtual myocardial ECV showed acceptable agreement with the reference ECV, achieving a mean root mean square error (RMSE) of 3.05%, despite noticeable slice-to-slice variability (R2 = 0.585; Bland–Altman 95% limits of agreement, −5.98% to +6.06%). Conclusions: The proposed method enabled the generation of post-contrast T1 and ECV maps directly from native T1 maps without the administration of gadolinium-based contrast agents during CMR. These findings suggest that the proposed approach represents a promising contrast-free, non-invasive alternative for myocardial tissue characterization, with the potential to reduce examination costs, eliminate contrast-agent-related risks, and improve patient safety.