MELODY aims to develop and test a way for secure computer systems across the UK to work together without moving or sharing sensitive data, to train and improve Artificial Intelligence (AI) that can accurately detect skin cancer.
Skin cancer is very common and is increasingly putting pressure on NHS dermatology services. AI can be used to help doctors diagnose and treat patients, but a major challenge is ensuring that these AI systems are fair, inclusive and safe.
A new approach called ‘federated machine learning’ allows AI to learn from securely stored NHS data from multiple sites across the UK without moving or copying patients’ sensitive information. This not only helps protect patient privacy but also makes it easier for more NHS organisations to take part in research, allowing new AI technologies to be developed using more diverse data, leading to fairer and better AI that works well for people from all backgrounds.
MELODY (Federated Machine Learning for Dermatology) is a collaborative project, led by The Health Informatics Centre (HIC) at the University of Dundee. It involves a close partnership with NHS Tayside, Oxford University Hospitals NHS Foundation Trust and the Thames Valley and Surrey Secure Data Environment.
As part of the MELODY project, we will be adopting the use of various capabilities developed in the TREvolution programme, including:
- SACRO-Machine Learning (SACRO-ML) – a tool that checks whether a trained AI model inside a Trusted Research Environment (TRE) can be moved without risking patient privacy.
- Standardised Architecture for Trusted Research Environments (SATRE) – a blueprint that helps organisations build safe, consistent and trustworthy research environments.
- K8TRE – a new, flexible system for building secure research environments.
- RO-CRATE – a simple way to package research data, files and information together which makes research more accessible, allowing other researchers and computers to easily understand, reuse and reproduce research.
What MELODY aims to achieve
By the end of the project, the MELODY aims to develop a tool that can be used to train an AI model for the detection of skin cancer using data held securely within Dundee and Oxford research environments. To do this, their goals include:
- Building a federated AI infrastructure – Creating a secure system that allows AI to learn from data held securely in different NHS research environments.
- Establishing dermatology data resource – Setting up real-world collections of dermatology data at two NHS locations.
- Demonstrating federated ML – Training and testing an AI model using skin lesion images from both locations to show that this federated ML approach is secure and works in practice.
Progress so far
Within the first six months of delivery, the team has:
- Developed and set up a secure solution allowing for federated research between the Dundee and Oxford research environments
- Embedded three public members in all stages of the MELODY project
- Delivered online public involvement and engagement workshops to further inform the progress of the project
- Created a standardised dermatology dataset using skin lesion image data from NHS Tayside (with the development of a dataset from Oxford University Hospitals in progress)
- Produced outputs relating to standardising dermatology images for research, including a ‘how to’ protocol, a training session for researchers, and a practical tool to assess metadata availability, completeness and readiness.
- Begun training a machine learning model using data secured in the Dundee and Oxford TREs
- Adopted various tools created during the TREvolution programme, including ROCRATE and SATRE
In the next six months of the project, the MELODY team are excited to deliver:
- A trained and validated machine learning model for the detection of skin cancer
- A standardised skin lesion image dermatology dataset from Oxford University Hospitals NHS Foundation Trust
- A Collaboration Café to further inform the MELODY project
- Write a detailed report using feedback from the public to help guide policy around the use of AI in healthcare
- An evaluation of the suitability of the federated system
- Adoption of other TREvolution products, including K8TRE and SACRO-ML
Relevance to TREvolution
MELODY is helping to build the technology and evidence needed to make research across multiple TREs easier, more secure, and more representative of people across the UK. MELODY hopes to accelerate progress in skin research and build public trust in sensitive data research by demonstrating the safe, trustworthy and secure use of identifiable skin image data through federation.
About the DARE UK Research Exemplars
The DARE UK Research Exemplars are short, impactful research projects demonstrating how emerging Trusted Research Environment (TRE) capabilities can support real-world sensitive data research across the UK. Working across multiple TREs and research domains, these projects bring together researchers, TRE operators, public contributors and other partners to apply and evaluate innovations developed through TREvolution. Their role is to test new approaches in live research settings, generating practical evidence on what works well, what challenges remain, and how sensitive data research can be strengthened for the public good.
