Confounders, diagnostic accuracy, and reproducibility in cardiovascular magnetic resonance-feature tracking-derived left atrial strain: A Berlin Research Network for cardiovascular magnetic resonance multi-software, multi-site comparison.

Abstract

Background: Cardiovascular magnetic resonance-feature tracking (CMR-FT) derived left atrial global longitudinal strain (LA-GLS) has prognostic relevance, even in the early stages of cardiovascular diseases. Identifying technical and subject-related confounders is essential for ensuring comparability across sites and for reliably distinguishing healthy from pathological conditions. This study aimed to evaluate the influence of post-processing software and subject-related factors on CMR-FT-derived LA-GLS, diagnostic accuracy, and to evaluate inter-site reproducibility.

Methods: This study included 149 healthy individuals and 40 patients with atrial fibrillation (AF; 19 persistent, 21 paroxysmal) from a single site. A subgroup of 18 traveling volunteers underwent CMR at four different sites. All participants underwent CMR in sinus rhythm. LA-GLS was assessed using three post-processing software packages (CVI42, TrufiStrain Research Prototype, Medis). Mixed models with repeated measures were applied to evaluate the effect of software, site, and subject-related factors on LA-GLS components. ROC curve analysis was used to assess diagnostic accuracy across software in distinguishing healthy controls from AF patients.

Results: All GLS components differed across post-processing software (p<.001). Reservoir and contractile GLS were lowest in CVI42 (23.9% ± 3.3%, 9.9% ± 2.2%), followed by TrufiStrain (27.4% ± 6.3%, 15.0% ± 4.8%) and Medis (45.4% ± 9.7%, 20.3% ± 5.7%). Conduit GLS was lowest in TrufiStrain (12.4% ± 4.8%), followed by CVI42 (16.3% ± 4.5%) and Medis (25.1% ± 8.2%). Among traveling volunteers, LA-GLS values were consistent across sites when the same software was used. Across all software, reservoir GLS negatively correlated with age. Diagnostic accuracy was comparable across software packages (AUC for reservoir strain: CVI: 0.81 [0.69-0.90], TrufiStrain 0.76 [0.64-0.88], Medis: 0.84 [0.72-0.94]).

Conclusion: Post-processing software is a significant confounder in CMR-FT-based LA-GLS analysis and age substantially influences LA-GLS. LA-GLS demonstrates excellent inter-site reproducibility when analyzed with the same software and offers comparable diagnostic accuracy across platforms.