Radiomic features of peri-left atrial epicardial adipose tissue and atrial fibrillation recurrence after ablation

Figures

• 

  Figure 1. Inclusion and exclusion criteria for the internal and external cohorts. AF, atrial fibrillation; PCTV, pulmonary CT venography.

• 

  Figure 2. Flow chart of key steps in radiomics. The left side illustrates the schematic segmentation of the entire EAT, whereas the right side shows the segmentation of the LA-EAT. Both follow the same feature extraction and dimensionality reduction processes. The SHAP beeswarm plot shows the top 15 important features based on their average impact on the model predictions. Each point represents an individual instance, and colours indicate feature values (red for high and blue for low). The horizontal spread of the points indicates how feature variations affect the output of the model. EAT, epicardial adipose tissue; LA, left atrial; SHAP, shapley additive explanations.

• 

  Figure 3. Clinical model nomogram for predicting recurrence within 1 year after RFA in patients with AF. The clinical nomogram presented here assigns a score to each selected predictor (including the LA-EAT volume) within the score range at the top of the plot. The total score is then matched with the scale at the bottom to estimate the probability of recurrence within 1 year after RFA in patients with AF. Age, LA-EAT volume and BNP/pro-BNP are the most significant factors identified through stepwise regression using the Akaike information criterion, whereas BMI, AF type and sex, recognised as influential factors, also predict AF recurrence. AF, atrial fibrillation; BMI, body mass index; BNP, B-type natriuretic peptide; EAT, epicardial adipose tissue; LA, left atrial; RFA, radiofrequency ablation.

• 

  Figure 4. ROC and calibration curves for models 1, 2 and 3. Model 1 (clinical factors and LA-EAT volume), model 2 (radiomics model), model 3 (model 1+model 2). ROC curves showing the predictive performance of each model in internal validation and external test cohorts. Calibration curves depicting the agreement between predicted probabilities and actual outcomes for each model. AUC, area under the curve; EAT, epicardial adipose tissue; LA, left atrial; ROC, receiver operating characteristic.

• 

  Figure 5. Decision curve analysis for models 1, 2 and 3. Model 1 (clinical factors and LA-EAT volume), model 2 (radiomics model), model 3 (model 1+model 2). Decision curves demonstrating the net benefit of each model for various threshold probabilities. EAT, epicardial adipose tissue; LA, left atrial.