Non-Imaging Medical Data Synthesis for Trustworthy AI: A Comprehensive Survey

Synthetic data can help solve healthcare AI's data access problem: The paper identifies that trustworthy AI development in healthcare is limited by difficulty accessing large, high-quality datasets due to privacy regulations and technical challenges. Synthetic data—artificially generated data that mimics real clinical data distributions—offers a potential solution to this bottleneck without requiring access to actual patient information.

Non-imaging medical data has been overlooked in synthesis research: While deep learning algorithms have advanced significantly for generating synthetic imaging data (like X-rays or CT scans), the paper highlights a critical gap: synthetic data generation for non-imaging healthcare information such as electronic health records (EHRs), vital signs, and medical waveforms has received far less attention despite being equally important for AI development.

Data quality directly impacts AI trustworthiness in healthcare: The paper emphasizes that the quality of training data is fundamental to building trustworthy AI systems. Curated synthetic datasets with controlled variables can improve not only accuracy but also robustness and privacy protection of AI algorithms used in clinical settings.

Synthetic data enables privacy-preserving AI development: By generating artificial datasets that preserve statistical patterns from real clinical data without containing actual patient information, synthetic data approaches can help overcome strict ethical restrictions on healthcare data sharing, allowing researchers to develop and test AI systems responsibly.