GAN-Based Novel Approach for Generating Synthetic Medical Tabular Data

Area of Science:

• Computational Biology and Bioinformatics
• Medical Informatics
• Machine Learning

Background:

• Growing demand for privacy-preserving synthetic medical data.
• Limitations in accessing real medical datasets for training generative models.
• Current methods' inability to generate tabular data solely from statistical inputs.

Purpose of the Study:

• To introduce a novel approach for generating tabular synthetic medical data from statistical information.
• To overcome the challenge of restricted access to real medical datasets.
• To enable synthetic data generation using only statistical properties.

Main Methods:

• Utilized a modified Generative Adversarial Network (GAN) architecture.
• Incorporated a custom loss function to improve generated data quality.
• Converted statistical data into tabular datasets.

Main Results:

• Achieved 'Good' similarity and 'Excellent' utility scores in evaluations.
• Demonstrated satisfactory performance for training machine-learning algorithms.
• Generated synthetic data shows promise for applications where real data is inaccessible.

Conclusions:

• The proposed method offers a viable solution for synthetic medical data generation from statistical inputs.
• This approach enhances medical data privacy and facilitates machine learning model training.
• The generated synthetic data, while not a complete replacement for real databases, is suitable for specific research and development needs.