A deep learning model integrating structured data and clinical text for predicting atrial fibrillation recurrence

Figures

• 

  Fig. 1. Comparison of metrics among different models. A Radar chart comparing different LLM-based text extraction modules in the dual-branch deep learning network architecture on the training set. B Radar chart comparing different LLM-based text extraction modules in the dual-branch deep learning network architecture on the validation set. C Radar chart comparing different LLM-based text extraction modules in the dual-branch deep learning network architecture on the test set. D Receiver operating characteristic (ROC) curves of different prediction models on the training set. E Receiver operating characteristic (ROC) curves of different prediction models on the validation set. F Receiver operating characteristic (ROC) curves of different prediction models on the test set. G Decision curve analysis (DCA) of different prediction models on the training set. H Decision curve analysis (DCA) of different prediction models on the validation set. I Decision curve analysis (DCA) of different prediction models on the test set. J Bubble chart of DeLong’s test for different prediction models on the training set. K Bubble chart of DeLong’s test for different prediction models on the validation set. L…

• 

  Fig. 2. Kaplan–Meier survival curves for the optimal model. A Kaplan–Meier survival curves for the training set. B Kaplan–Meier survival curves for the validation set. C Kaplan–Meier survival curves for the test set.

• 

  Fig. 3. Model interpretability. A SHAP variable importance plot for the structured data channel. B SHAP token-level importance plot from the text channel. C Word cloud visualization of key features extracted from the text channel.

• 

  Fig. 4. Forest plot of sensitivity analysis. Forest plot displaying the results of sensitivity analysis across different subgroups and conditions.

• 

  Fig. 5. Flow chart of this study. A Study enrollment flowchart illustrating participant selection and exclusion criteria. B Schematic diagram of the dual-branch deep learning network architecture showing the integration of structured data and text processing channels.