Exploring the challenges of representativeness in AI interpretation of medical imaging, a journal published in The Lancet this week introduces the ‘Medical AI Data for All (MAIDA)’ initiative, described as a “framework for global medical data sharing to address the shortage of public health data and enable rigorous evaluation of AI models across all populations”.
The framework’s creators state that MAIDA is designed to provide key ingredients for the thorough assessment of AI “through rich, diverse datasets”; it has been developed through collaboration with a range of global partners facilitating engagement, with 100 medical scans collected from a number of hospitals across the world in order to help assemble medical-imaging data at scale, across different populations and settings
The MAIDA workflow consists of four stages: project initiation, through which an initial meeting allows potential partners to review the workflow and confirm participation; ethics and data-sharing approval, with supply of relevant templates; data collection, including instructions for inclusion and exclusion criteria along with sampling methods and de-identification procedures; and finally data sharing, which sees an URL provided to partners so that they can directly upload their data.
Chest x-rays are a particular focus for the research team; the journal notes that although chest x-rays “are the most widely used radiological tests worldwide”, the generalisability of existing models “remains insufficiently evaluated as current public datasets are insufficient in size, diversity or scope”. As such, the MAIDA initiative seeks to improve quality and breadth of chest x-ray interpretation in three main areas: intensive care, neonatal intensive care, and the emergency department.
Through the MAIDA initiative, the researchers plan to progressively release these diverse global datasets, with the first release expected in 2024. They also aim to make the datasets publicly available, to support open research into the assessment and improvement of AI models for medical imaging.
Click here to access the journal in full.
July saw us cover the news that The Royal Marsden NHS Foundation Trust signed a five-year deal with health tech company Atos to modernise communications infrastructure and install hybrid, cloud-based tech, with senior vice president of Atos UK Samantha Jones highlighting that the project will “provide a communications ecosystem” to support staff and researchers “with access to fully integrated tools to communicate and collaborate.”
Earlier in the week we shared an exploration by Open Medical, with focus on the need to capture “rich, detailed data”.