MELODY: Federated Machine Learning for Dermatology

Skin disease affects more than half of the UK population each year, placing growing pressure on dermatology services. Artificial intelligence has the potential to support diagnosis and care pathways, but many existing datasets lack diversity, risking biased or unrepresentative models.

MELODY will create dermatology image datasets within Trusted Research Environments in NHS Tayside and Oxford University Hospitals NHS Foundation Trust. Instead of centralising patient data, federated machine learning will allow models to be trained locally within each environment while sharing only model parameters.

The project will develop and test a skin lesion classification model using federated learning techniques, demonstrating how collaborative AI development can occur across TREs while protecting patient privacy.

Patient and public involvement will play an important role in the project. Engagement activities will involve patient representatives in discussions on dataset governance, transparency in AI evaluation, and the use of dermatology data in research.

By the end of the project, MELODY will:

• Implement a federated machine learning solution across two TREs
• Establish real-world dermatology image datasets in Scotland and England
• Develop and validate a federated skin lesion machine learning model
• Produce a technical and governance blueprint for federated AI development
• Create public-facing materials explaining federated AI development

Project information

Lead organisation: University of Dundee
Principal investigator: Richard Walls
Project duration: 12 months
Project partners: Health Informatics Centre (HIC), Thames Valley and Surrey Secure Data Environment (TVS SDE), NHS Tayside, Oxford University Hospitals NHS Foundation Trust
Funding provided: £486,340
Primary contact email: r.walls@dundee.ac.uk