Intracardiac thrombus formations despite continuous oral anticoagulation in atrial fibrillation patients undergoing catheter ablation procedures: pilot development of a machine learning prediction model

Figures

• 

  Figure 1. Flow chart demonstrating patient selection for analysis.

• 

  Figure 2. Clinical CHA2DS2-VASc scoring in the TF+ and TF− group; (A) Boxplot of the CHA2DS2-VASc in each group; (B) CHA2DS2-VASc score distribution in each group.

• 

  Figure 3. Receiver operating characteristic curve (A) and precision-recall-curve (B); XG-boost model (violet), random forest model (red), light GB model (blue), CHA2DS2-VASc score (brown).

• 

  Figure 4. P-wave morphology parameters for atrial cardiomyopathy in TF+ and TF− patients. Panels a-i display boxplots comparing quantitative P-wave parameters between patients with (TF+, dark blue) and without thrombus (TF−, light blue), Panel j shows the distribution of patients by number of abnormal P-wave parameters within each group. Panel k displays the proportions of patients in each group presenting with specific categorical p-wave morphology parameters. (Biphasic in V1, Biphasic P Wave Morphology in V1, Negative in V1, Negative P-Wave in V1).

• 

  Figure 5. Feature importance of the ML models; (A) Number of abnormal P Wave Parameters; (B) P Wave Duration in Lead II; (C) P Wave Amplitude in II; (D) P Wave Amplitude in V1; (E) P Wave Duration in V1; (F) P Wave Axis; (G) P Wave Dispersion; (H) PTFV1; (I) P Wave Area; (J) Number of Abnormal P Wave Parameters; (K) Proportions of patients presenting with categorial P Wave Morphology parameters.

• 

  Figure 6. SHAP-based feature importance of the 15 most relevant predictors in the three machine learning models. Panels (A), (C), and (E) show SHAP beeswarm plots for XGBoost, Random Forest, and LightGBM. Each dot represents an individual patient, with color indicating the relative feature value (purple = low, yellow = high). Panels (B), (D), and (F) display boxplots of the mean absolute SHAP values across cross-validation folds.