Epicardial Adipose Tissue Radiomic Features for Predicting Atrial Fibrillation Recurrence

A retrospective study of 302 atrial fibrillation patients who underwent pulmonary vein isolation, with data collected from 1,551 initially screened patients. The dataset was created by Guoxiang Ma and published on figshare in March 2026. It was used to develop a machine learning model integrating radiomic features from CT scans with clinical variables.

Use Cases

• Predicting atrial fibrillation recurrence after catheter ablation based on radiomic texture heterogeneity features.
• Training ensemble machine learning models for postoperative risk stratification using clinical variables like NT-proBNP.
• Validating segmentation models for epicardial adipose tissue on CT angiography images.
• Investigating the pathophysiological link between adipose tissue features and arrhythmia recurrence.

Strengths

• Dataset includes 302 patients with a balanced cohort of 151 with and 151 without recurrence after propensity score matching.
• Radiomic features were extracted using a SwinUNETR model achieving a dice similarity coefficient of 0.87 for segmentation.
• The resulting predictive model achieved an area under the curve (AUC) of 0.81 in internal validation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying feature data is unknown, which may limit suitability assessment.
• Data is from a single retrospective study, which may reflect inherent selection and temporal biases.

Provenance

Source

figshare, author Guoxiang Ma.

Collection Method

Retrospective study with data from patients who underwent pulmonary vein isolation; EAT segmented from CT images using a SwinUNETR model.

Freshness

Last updated 2026-03-18 07:42:50; freshness should be verified.