Free-breathing single-beat exercise cardiovascular magnetic resonance with generative artificial intelligence for evaluation of volumetric and functional cardiac indices: A reproducibility study.

Abstract

Background: Exercise cardiovascular magnetic resonance (Ex-CMR) can reveal pathophysiologies not evident at rest by quantifying biventricular volume and function during or immediately after exercise. However, achieving reproducible Ex-CMR measurements is challenging due to limited spatial and temporal resolution. This study aimed to develop and evaluate a free-breathing, high-spatiotemporal-resolution single-beat Ex-CMR cine enhanced by generative artificial intelligence. We assessed image analysis reproducibility, scan-rescan reproducibility, and impact of the reader's experience on the analysis.

Methods: Imaging was performed on a 3T CMR system using a free-breathing, highly accelerated, multi-slice, single-beat cine sequence (in-plane spatiotemporal resolution of 1.9 × 1.9 mm² and 37 ms, respectively). High acceleration was achieved by combining compressed sensing reconstruction with a resolution-enhancement generative adversarial inline neural network. Ex-CMR was performed using a supine ergometer positioned immediately outside the magnet bore. Single-beat cine images were acquired at rest and immediately post-exercise. In a prospective study, the protocol was evaluated in 141 subjects. A structured image analysis workflow was implemented. Four expert readers, with or without prior training in single-beat Ex-CMR, independently rated all images for diagnostic and image quality. The subjective assessment used two 3-point Likert scales. Biventricular parameters were calculated. Inter- and intra-observer reproducibility were assessed. Fifteen healthy subjects were re-imaged 1 year later for scan-rescan reproducibility. Reproducibility was assessed using intraclass correlation coefficient (ICC), with agreement evaluated via Bland-Altman analysis, linear regression, and Pearson correlation.

Results: Free-breathing, single-beat Ex-CMR cine enabled imaging of the beating heart within 30 ± 6 s, with technically successful scans in 96% (136/141) of subjects. Post-exercise single-beat cine images were assessed as diagnostic in 98% (133/136), 96% (131/136), 82% (112/136), and 65% (89/136) of cases by four readers (ordered by descending years of Ex-CMR experience). Good image quality was reported in 74% (100/136) to 80% (109/136) of subjects. Biventricular parameters were successfully measured in all subjects, demonstrating good to excellent inter-observer reproducibility. Scan/rescan reproducibility over 1 year, assessed by two independent readers, showed excellent inter-visit ICCs (0.96-1.0) and strong correlations (R² ≥ 0.92, p < 0.001 for left ventricle; R² ≥ 0.95, p < 0.001 for right ventricle).

Conclusion: Single-beat Ex-CMR enabled evaluation of biventricular volumetric and functional indices with excellent reproducibility.