Intra- and inter-field strength reproducibility of deep-learning based real-time cardiac MRI cine sequences with...

To assess intra- and inter-field strength reproducibility of volumetric parameters using deep-learning-based real-time cardiac cine MRI during breath-hold (BH) and free-breathing (FB). In this prospective single-center study, 56 healthy adults underwent cardiac MRI at 1.5 T. Of these, 33 had a follow-up scan after 2-7 weeks, and 23 received an additional same-day scan at 3 T with the same protocol. Real-time cine sequences (1RR), including short-axis and 2-, 3-, and 4-chamber views, were acquired in BH and FB. Left ventricular volumes were analyzed using automated segmentation. Intra-class correlation coefficients (ICC) and subjective image quality (sIQ) were used to assess reproducibility. At 1.5 T, BH sequences showed significant differences in stroke volume (SV) and ejection fraction (EF), while FB sequences revealed only minor, clinically irrelevant SV variation. End-diastolic volume (EDV) and left ventricular (LV) mass showed excellent reproducibility (ICC > 0.93); end-systolic volume (ESV) and SV had good reproducibility (ICC 0.79-0.88). Inter-field comparisons revealed significant differences for EDV (BH), and for SV and EF (FB), though most parameters remained consistent. EDV, ESV, and LV mass showed excellent reproducibility (ICC > 0.90), and SV showed good to excellent agreement. Deep-learning-based real-time cine MRI provides good to excellent reproducibility of cardiac volumetric parameters across field strengths and breathing conditions.