Generative AI Models in Time-Varying Biomedical Data: Scoping Review

Area of Science:

• * Computational health sciences
• * Health informatics
• * Artificial intelligence in medicine

Background:

• * Traditional statistical and machine learning methods struggle with complex, multimodal health data and long-term dependencies.
• * Generative artificial intelligence (AI) excels at representing complex data distributions and patterns.
• * AI applications in finance and environmental sciences show promise for healthcare disease modeling.

Purpose of the Study:

• * Introduce basic concepts of generative AI for healthcare practitioners.
• * Discuss current generative AI algorithms and their applications in healthcare.
• * Address the complexity barrier limiting AI adoption in clinical practice.

Main Methods:

• * Systematic literature review of peer-reviewed papers on generative AI for time-series health data.
• * Included single- and multimodal generative AI models across structured/unstructured data, waveforms, imaging, and multi-omics.
• * Analyzed current methods, applications, limitations, and future directions.

Main Results:

• * Reviewed 155 articles on generative AI applications in time-series healthcare data, adhering to PRISMA-ScR guidelines.
• * Identified and categorized generative AI models across various healthcare data modalities.
• * Developed a systematic framework to aid clinicians in selecting appropriate AI methods.

Conclusions:

• * Critiqued existing generative AI applications for time-series health data to bridge the computational-clinical gap.
• * Highlighted shortcomings in current generative AI approaches for healthcare modeling.
• * Identified promising recent advances in generative AI for future healthcare applications.