SynthEthics: Ensuring Digital Ethics and Performance with a Design Theory for Using Synthetic Image Data in Digital Health Deep Learning

This paper addresses the need for ethical and effective use of synthetic image data in digital health computer vision. It explores the design requirements and design principles for both responsible use of artificial intelligence in digital health and model robustness, focusing on privacy, ethical compliance, and domain adaptation. Using the design science research paradigm along with value-sensitive design and sociotechnical systems theory, this study presents a design theory that provides actionable guidance for the generation, selection, and integration of synthetic data in digital health. Through heuristic theorizing over two design cycles, the work provides a robust theory artifact and conceptual model to ensure ethical use and improve model performance in digital health through appropriate domain adaptation, generalization, and accuracy. In addition to contributing to theoretical knowledge, this research offers practical implications for health authorities to promote ethical standards and performance in synthetically trained AI applications.