All-in-one cardiac digital twin widget including personalized holographic 1.) healthy and 2.) abnormal heartbeat simulations and AI-ML driven risk prediction for long-term health outcomes to empower patient-provider interactions and provide additional considerations in holistic care. This streamlines the explanation of complex cardiological diagnoses such as atrial fibrillation. Our Bi-LSTM With Attention Mechanism together with MLP are constructed using genetic, environmental, lifestyle, and personalized health information, including data from patients' charts. Models provide predictive risk scores of future events or conditions occurring that existing tools cannot anticipate, such as iron-deficiency anemia in the case of atrial fibrillation.

All additional data files are available on Google Drive due to Github's storage limitations. See Installation & Deployment.

This includes input datasets, model training history, evaluation results, and final model output files:

• data/ECGModel: Contains raw ECG datasets for training and testing (mitbih_test.csv, mitbih_train.csv, ptbdb_abnormal.csv, ptbdb_normal.csv).
• data/heart: Heart dataset (heart.csv) for model training and evaluation.
• output/ecg:
  • confusion_matrix: Stores the confusion matrices generated for ECG-based model evaluations.
  • final_results: Stores the final evaluation results.
  • training_history.csv and training_history: Record the training progression of the ECG models.
• output/heart/models: Trained models are saved here, with the best model stored as best_model.pth.
• output/heart/plots: Contains various visualizations, such as:
  • confusion_matrix: Shows prediction accuracy.
  • correlation_matrix: Displays feature correlations within the heart dataset.
  • roc_curve: Illustrates the ROC curves for trained models.
  • training_history: Records the training progress.
• output/heart/results:
  • classification_report: Detailed classification metrics.
  • data_info: Dataset structure and statistics.
  • training_info: Overview of the training process.
• output/heart:
  • catboost_report: CatBoost-specific classification results.
  • data_info: Saved descriptions of the dataset used.
  • training_info: Contains details about training set configurations.

1. Clone the repository:

git clone https://github.com/HannahNJIT/HeartUp.git

2. Download the data/output files: Data/Output Google Drive

3. Try out the models with provided sample data:

python ecg-RNN.py
python heart.py