“It’s a fact that the performance of algorithms is very closely correlated to the amount and breadth of data that we have,” Dr. Jorge Cardoso, Chief Technology Officer of the London Medical Imaging and AI Centre for Value-Based Healthcare (AI Centre) told the recent AIMed UK virtual Summit
But the sensitive nature and business value of such healthcare data also means its usage is highly regulated and not likely to be shared freely. Even if access were granted, it would still require considerable time, effort and expense to curate and maintain the kind of quality desired by fellow developers to train AI models. So to bypass the existing limitations, Dr. Cardoso and his team are now leveraging federated learning.
Federated learning is a machine learning technique where collaborating researchers share a partial-trained model, rather than the anonymized health data, as they develop the algorithm. It ensures nobody, including those who are involved, will see the data or a private version of it. In a paper recently published in Nature Digital Medicine, the AI Centre said it is now working with two industrial partners and 14 other institutions to promote the use of federated learning.
“It ensures data is safe and patient’s privacy is preserved, while providing unparalleled access to high-fidelity data,” said Dr. Cardoso. “The opportunities are immense. Federated learning is currently the best approach for scalable, safe, robust and fair AI in a healthcare setting”. The technique has already allowed clinicians to improve diagnostic tools for imaging analysis and enabled companies to accelerate drug discovery with decreased cost and time.
Facilitating healthcare data access is one of the many pathways that runs through the heart of the AI Centre. Its other projects include AI-powered 3D anatomical models that tackle the urgent backlog of cancer surgeries after the COVID-19 lockdown, helping to build the most powerful supercomputer for AI In healthcare and machine learning driven radiomic analysis for early detection of cardiac risk factors.
Established in February 2019 as part of the UK government’s Industrial Strategy Challenge Fund, the AI Centre’s mission is to utilize AI to speed up diagnosis and care for stroke, dementia, heart failure and cancer patients. As its name suggests, the AI Centre has a special focus on transformation and value-based healthcare. It envisioned a smooth pipeline to create, test and ultimately, deploy AI-driven solutions across the UK public health system which helps in the optimization of resources and improvement of clinical outcomes.
As such, the AI Centre brings together an ambitious consortium of three prominent higher institutions, the National Health Service (NHS), multinational industrial partners, 10 UK-based SMEs and the Health Innovation Network. The Centre is pulling expertise across the healthcare ecosystem and constructing appropriate infrastructure to support scaling of AI as it continues to harness some of the best research capabilities and clinical knowledge in the country.
Dr. Cardoso believes the drive for synergy comes out of necessity. The more the AI Centre wishes for a wide adoption of AI in healthcare, the more it needs to collaborate with different stakeholders. This is also the reason behind Dr. Cardoso’s keen interest in federated learning. He wants to make AI safe and it is definitely not a lone effort. Moreover, since technology tends to advance more rapidly than regulations, an institution like the AI Centre is set to play a pivotal role in staging the appropriate governance.
“There remains a gap between progress and regulation. Because some of the innovations we see right now, including federated learning, have never been fully tested or explored, so it’s very hard to determine what to do next and set the parameters. This is problematic and we need to work this out together”.