Radiomics-based atrial flow profile description from 4D Flow MRI in atrial fibrillation patients

Atrial Fibrillation (AF) is the most common sustained cardiac arrhythmia, associated with significant thromboembolic risk, driven by altered hemodynamics within the left atrium (LA). Characterizing these flow alterations in a quantitative and non-invasive manner remains an open challenge in clinical cardiology.This study represents the first application of a radiomics-based approach to hemodynamic parameter maps derived from 4D Flow MRI in the left atrium for the flow remodeling characterization in patients with atrial fibrillation. Radiomics features were extracted from five hemodynamic parameter maps computed across ten anatomical planes within the LA throughout the entire cardiac cycle. Given the limited sample size (n=23), a Leave-One-Out-Cross-Validation scheme with nested feature selection was adopted, ensuring that feature selection is performed exclusively on the training set at each fold. Within each fold, features were selected through a three-stage strategy: univariate statistical filtering, multiple comparison correction, and effect size filtering. The retained features were used to train a Support Vector Machine classifier. Stability and sensitivity analyses were subsequently performed to obtain a set of highly robust, reliable, and reproducible radiomics biomarkers. The top-ranked features, analyzed at atrial end-diastole and end-systole, were exclusively texture-based metrics derived from Flow Angle maps. Healthy controls exhibited spatially heterogeneous and complex flow organization consistent with coordinated atrial mechanical function, while AF patients showed simplified and more homogeneous patterns reflecting impaired contractile function and loss of physiological flow complexity.These results demonstrate that radiomics applied to 4D Flow MRI can quantitatively capture left atrial flow remodeling in AF, providing a foundation for future development of imaging biomarkers.