The random forest-based fusion model combining EAT radiomics and clinical variables predicted atrial fibrillation recurrence after pulmonary vein isolation with an AUC of 0.81 (95% CI: 0.59–0.87).
Observational (n=302)
No
Does a machine learning fusion model integrating epicardial adipose tissue radiomic features from pre-procedural CT and clinical variables predict atrial fibrillation recurrence after pulmonary vein isolation?
A machine learning fusion model incorporating epicardial adipose tissue radiomics from pre-procedural CT and clinical variables can effectively predict atrial fibrillation recurrence after pulmonary vein isolation.
Effect estimate: AUC 0.81 (95% CI: 0.59–0.87) for random forest fusion model (95% CI 0.59–0.87)
Purpose This study aimed to develop and validate a machine learning model that integrates radiomic features of epicardial adipose tissue (EAT) from pre-procedural CT angiography with clinical variables to predict atrial fibrillation (AF) recurrence after pulmonary vein isolation (PVI). Materials and methods This retrospective study initially included 1,551 AF patients who underwent PVI. After data integrity screening and 1:1 propensity score matching (PSM) to balance confounding factors, the final analysis cohort consisted of 302 patients (151 with recurrence and 151 without recurrence). EAT was segmented from preoperative CT angiography images using a SwinUNETR model, which was pre-trained via transfer learning on manually annotated images. Following segmentation, radiomic features were extracted. Subsequently, six machine learning models were developed and evaluated. Results The SwinUNETR segmentation model achieved a dice similarity coefficient of 0.87. For AF recurrence prediction, the fusion model demonstrated superior and robust performance in internal validation. The random forest-based fusion model achieved the highest area under the curve (AUC) of 0.81 (95% CI: 0.59–0.87). Key predictive features included NT-proBNP and texture heterogeneity features from EAT, which align with known pathophysiological mechanisms involving systemic inflammation, metabolic dysregulation, and local atrial adipose tissue remodeling. Conclusion A fusion model incorporating EAT radiomics and clinical variables effectively predicts AF recurrence after PVI, with ensemble methods showing optimal performance. This study provides a multiscale, interpretable computational tool for individualized postoperative risk stratification, highlighting the complementary role of EAT imaging biomarkers to systemic clinical factors.
Ma et al. (Thu,) conducted a observational in Adults with atrial fibrillation undergoing catheter ablation by pulmonary vein isolation with pre-procedural CT imaging (n=302). Machine learning fusion model integrating epicardial adipose tissue radiomic features from pre-procedural CT angiography and clinical variables vs. Either clinical model alone or radiomics model alone was evaluated on Atrial fibrillation recurrence defined as any episode of AF, atrial flutter, or atrial tachycardia lasting more than 30 seconds recorded after a 3-month blanking period post-PVI (AUC 0.81 (95% CI: 0.59–0.87) for random forest fusion model, 95% CI 0.59–0.87). The random forest-based fusion model combining EAT radiomics and clinical variables predicted atrial fibrillation recurrence after pulmonary vein isolation with an AUC of 0.81 (95% CI: 0.59–0.87).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: