Guidelines for Ethical AI Medical Imaging Research

To regulate algorithm research, technology development, clinical validation and translation in AI medical imaging, mitigate ethical risks, and promote responsible innovation, the Ministry of Science and Technology released the Ethical Guidelines for AI Medical Imaging Research on August 27.

AI medical imaging shows strong potential in early diagnosis, workflow optimization and resource allocation. Wang Guoyu, a member of the Medical Ethics Subcommittee of the National Science and Technology Ethics Committee, which has developed the document, said medical imaging data is highly sensitive. Risks — including privacy breaches, algorithmic bias, technology misuse and unclear accountability — must be addressed throughout data collection, storage and sharing.

The guidelines set out six core principles for AI medical imaging research: advancing human welfare, promoting fairness and justice, respecting human subjectivity, protecting privacy and data security, ensuring safety and controllability, and enhancing transparency and trustworthiness.

"Patient interests must come first, with careful assessment of potential impacts on health and life safety," said Wang, adding that fairness requires preventing algorithmic discrimination and group bias, especially for vulnerable groups such as children, the elderly and patients with rare diseases.

For privacy and data security, the guidelines call for mechanisms for data authorization and withdrawal of consent to prevent unauthorized access and leaks.

Qu Jingjing, a young scientist at the Shanghai Artificial Intelligence Laboratory, said elevating safety and transparency in ethical principles means establishing full-lifecycle risk management, with preventive, monitoring and response mechanisms, and ensuring that technical logic and imaging interpretation decisions remain interpretable and verifiable.

The guidelines lay down tiered requirements — general, specific and ethics-review focused — to provide targeted and practical rules suited to the complexity and diversity of AI medical imaging applications.

For the research process, the document specifies rules on legal compliance, informed consent, accountability and conflict-of-interest management. It also clarifies when informed consent may be waived and calls for a collaborative system to clarify the roles of researchers, medical institutes, technology companies and regulators.

Data is both the foundation of AI medical imaging research and a major risk area. The guidelines set four data requirements: establishing a security system covering research, trials and translation to ensure traceability; using datasets with clear sources and scientific validity; enforcing data anonymization so that processed information cannot re-identify individuals or be restored; and exercising caution when using synthetic data to train AI models.

To tackle the "algorithm black box," core metrics such as sensitivity and specificity must meet appropriate standards. Algorithms must undergo repeatability testing and cross-device, cross-hospital and cross-population validation to prevent clinical bias. Bias must be identified and corrected at each stage — data collection, annotation, training and evaluation.

"The guidelines mark a shift from principle-based advocacy to detailed guidance," said Zhang Jun, director of the radiology department at the Huadong Hospital Affiliated to Fudan University.

Source: Science and Technology Daily