TransGAN-DX: A Hybrid Transformer-GAN Approach for Enhanced Cardiovascular Disease Diagnosis

Abstract

Cardiovascular disease remains one of the leading causes of mortality worldwide, highlighting the urgent need for more accurate and reliable diagnostic tools. TransGAN-DX is a novel deep learning framework integrating Generative Adversarial Networks (GANs) and Transformer-based architectures to enhance cardiovascular disease detection.

Link to the published article for this project: https://www.americanhhm.com/researchinsights/transgan-dx-a-hybrid-transformer-gan-approach-for-enhanced-cardiovascular-disease-diagnosis

The framework:

• Leverages GANs to generate synthetic data, improving training diversity and addressing data imbalance challenges.
• Utilizes a Transformer-based classifier trained on the enriched dataset to predict disease presence with high precision.
• Incorporates correlation and feature importance analyses for interpretability, aiding clinical decision-making.
• Addresses model overconfidence by calibrating predictions using calibration curves and expected calibration error (ECE).

Evaluated on the Heart Disease dataset, TransGAN-DX achieved:

• Accuracy: 89%
• F1-Score: 86%
• ROC-AUC Score: 88%

Features

• Data Preprocessing: Scales, normalizes, and handles data imbalance.
• GAN Module: Generates synthetic samples to augment the dataset.
• Transformer Classifier: Predicts disease presence with high precision.
• Evaluation Tools: Provides accuracy, F1-score, ROC curve, PR curve, and confusion matrix.
• Visualization: Includes correlation heatmaps, SHAP analysis, and loss plots.

License

This project is licensed under the MIT License. See the LICENSE file for details.

© ALI BAYANI 2025

Repository Structure

project/
│
├── data_preprocessing.py      # Data loading and preprocessing logic
├── gan_model.py               # Generator, Discriminator, and GAN training loop
├── transformer_classifier.py  # TransformerClassifier model and training
├── evaluation.py              # Accuracy, F1, ROC, PR curve, and confusion matrix
├── visualization.py           # Correlation heatmap, SHAP, and loss plots
├── main.py                    # Main script orchestrating the modules
└── requirements.txt           # Dependencies for the project