Correlation between myocardial radiomics features at the left atrium–left atrial appendage junction and stroke risk in atrial fibrillation patients: a retrospective case–control study

Objective This study aimed to systematically evaluate the microstructure changes of the myocardium at the left atrium–left atrial appendage (LA–LAA) junction in patients with atrial fibrillation (AF)-related cardioembolic stroke, and to explore its potential role as an imaging marker associated with stroke. Methods This retrospective study included four groups: AF-related cardioembolic stroke group ( n = 60), AF-non-stroke group ( n = 29), non-AF-stroke ( n = 53), and non-AF-non-stroke ( n = 53). All subjects underwent cardiac computed tomography examination. Myocardial thickness at the LA–LAA junction was measured, and radiomics features were extracted. Radiomics features included first-order statistical features, shape features, and texture features. Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to screen features, and a logistic regression model combining radiomics features and clinical parameters was constructed. Group comparisons were performed using t-tests or Mann–Whitney U tests. The predictive ability of the model was evaluated with the receiver operating characteristic (ROC) curve, and a nomogram was used for risk assessment. Results Compared with AF patients without stroke, patients with AF-related cardioembolic stroke significantly increased myocardial thickness at the LA–LAA junction [7.1 (6.43–8.33) mm vs. 6.58 ± 1.45 mm, p < 0.05]. In addition, multivariate logistic regression demonstrated that myocardial thickness was independently associated with AF-related cardioembolic stroke. Three radiomics features were selected by LASSO regression in AF-related cardioembolic stroke and non-AF-non-stroke patients, and a radiomics score (Rad-score) was generated. The Rad-score was statistically significant between the groups ( p < 0.001). The combined model incorporating Rad-score and clinical parameters achieved an area under curve (AUC) of 0.968 in the test cohort. Conclusion Both myocardial radiomics features and myocardial thickness at the LA–LAA junction may serve as potential imaging markers associated with AF-related cardioembolic stroke. Their incremental value for stroke risk stratification beyond established clinical risk factors requires further validation in larger prospective cohorts.