Synthetic data generation methods in healthcare: A review on open-source tools and methods

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Data Science

Background:

• Data scarcity and privacy concerns hinder AI development in healthcare.
• Need for unbiased, large-scale datasets for robust AI model training.
• Synthetic data generation offers a potential solution to these challenges.

Purpose of the Study:

• To review the application and efficacy of synthetic data generation methods in healthcare.
• To analyze diverse medical data types including tabular, imaging, radiomics, time-series, and omics data.
• To identify trends in methods and programming languages used for synthetic data generation.

Main Methods:

• Systematic literature search of PubMed and Scopus databases.
• Categorization of synthetic data generation methods (statistical, probabilistic, machine learning, deep learning).
• Analysis of programming languages used for implementation, with emphasis on Python.

Main Results:

• Synthetic data generators are used to reduce clinical trial costs, enhance AI predictive power for personalized medicine, ensure fair treatment recommendations, and enable access to multimodal data without privacy risks.
• Deep learning methods dominate, utilized in 72.6% of studies.
• Python is the predominant programming language, used in 75.3% of implementations.

Conclusions:

• Synthetic data generation is a valuable tool in healthcare, particularly for rare diseases and personalized medicine.
• Deep learning-based approaches and Python implementation are current trends.
• Open-source repositories are documented to facilitate further research and accelerate adoption.