The Working Group on Recommended Practices for Generalisability of Artificial Intelligence for Medical Imaging (IEEE 3350 Working Group) was formally established in September 2022 under the chairmanship of Associate Professor Tantao of the Faculty of Applied Sciences of the Macao Polytechnic University (MPU), with significant technical support from standards experts of the Alibaba Group, and with the participation of PhD student Jiang Mingfu and MSc student Ong Wengsheng of the Faculty of Applied Sciences of the MPU in the relevant work of the Task Force. This standard (practice) is the first time that a university in the Macao SAR is leading an IEEE level international standard/practice in the field of medical artificial intelligence. The working group is committed to provide a framework and recommendations to improve the generality of AI models in medical imaging. The main topics include: overview of the generality of AI models in medical imaging, processes and specifications for data processing (i.e., data collection, data annotation, data coordination, and data expansion), methods for composite factor removal, specifications for AI modelling, methods for continuous learning and federated learning, and metrics for assessing scalability perspectives.

After more than half a year of hard work and unremitting efforts by members of the IEEE 3350 Working Group, the Recommended Practice for Enhancing the Generalisability of Artificial Intelligence in Medical Imaging was formally approved as a project on 30 March 2023 in the IEEE Standards Organisation. This standard of practice proposes a framework to provide recommendations for enhancing the generalisability of Artificial Intelligence (AI) models in medical imaging. It includes the following components: a review of AI solutions in medical imaging from a scalability perspective; processes and specifications for data handling - including data collection to ensure diversity and data annotation to achieve rigour; methods for removing compounding factors to mitigate bias between various datasets; AI modelling specifications, medical imaging small training data, unimodal/multimodal data in medical imaging, enabling flexible deployment in different usage scenarios; introduction of continuous learning and joint learning approaches to adapt AI models to local populations; and evaluation metrics from a scalability perspective.

In the future, the working group will unite domestic and foreign industry partners to carry out the research and development, testing and validation of a series of standards and practices for the generalisation of medical imaging artificial intelligence, and actively construct a standard ecosystem for the generalisation of medical imaging artificial intelligence, so as to turn the standard for the generalisation of medical imaging artificial intelligence into a cutting-edge technological standard with international influence.