A Trustworthy Health AI Development Framework with Example Code Pipelines.

Area of Science:

• Health Artificial Intelligence (AI)
• AI Ethics
• Software Engineering

Background:

• Health AI development currently lacks practical guidelines for ensuring trustworthiness.
• Existing general best practices are insufficient for the specific needs of health AI.

Purpose of the Study:

• To introduce the Trustworthy AI Development (TAIDEV) framework, a practical guideline for creating trustworthy health AI systems.
• To provide a structured approach for addressing ethical standards, robustness, and safety in health AI.

Main Methods:

• Developed the TAIDEV framework, featuring a matrix classifying technical methods across the AI lifecycle for EU Trustworthy AI requirements.
• Integrated customizable Python code pipelines and a validation checklist.
• Validated the framework using two open datasets (UCI Heart Disease, Diabetes 130-US Hospitals).

Main Results:

• The TAIDEV framework offers a comprehensive guideline for developing trustworthy health AI.
• Practical code examples and a validation checklist are provided to facilitate implementation.
• The framework's applicability was demonstrated on real-world health datasets.

Conclusions:

• The TAIDEV framework enables health AI developers to build ethical, robust, and safe AI systems.
• It serves as an extensible, open-source resource for advancing trustworthy AI in healthcare.
• The framework supports the development of reliable Clinical Decision Support Systems.