Training fairer medical AI needs diverse data, but hospitals are restricted in data they can share for privacy reasons. We built an easy-to-deploy platform for hospitals to take part in AI development without sharing data, and piloted it at 4 NHS Trusts https://t.co/VA41fXutdJ

What a better way to wrap up the #FlowerAiSummit2025 that an amazing panel of experts including @andrewsoltan @chongshenng , @IttaiDayan, Patrick Foley and Robert Norvill!

Join us in Oxford @StAnnesCollege on November 12th for a Workshop exploring Trustworthy AI & Healthcare. Exciting lineup of speakers from industry and academia. Registration is free but slots are limited! More details here -

Thrilled to see our Federated Learning paper highlighted in today's @FT, as part of a discussion around improving access to health data for AI. FL allows AI models to be trained and tested without ever moving patient data, and may help to reduce biases. https://t.co/cJSVFvJKXz

It's a delight to work with Flower. Thanks for having me at the 2024 AI summit!

Not every hospital collects and stores the same data in the same way; a big challenge in federated learning! Congrats Soheila Molaei & Anshul Thakur for leading this piece -accepted to AISTATS- using augmented graph attention networks for heterogenous FL! https://t.co/GgWUemXuZQ

Really pleased to speak at the Flower AI Summit 2024 next week, about our work on a rapidly-deployable federated learning platform for hospitals using the library. March 14 & 15 in London, UK. Full agenda here: https://t.co/EXuOhnqZ4T Free & virtual attendance available!

Flower did much of the heavy lifting in our work on full-stack federated learning for the NHS- a powerful and easily implementable library with excellent documentation! It'll be a pleasure to speak in support of Flower! https://t.co/QwDicUM62c

📢 New podcast out now! @andrewsoltan joins us to discuss the development, testing and deployment of a #FederatedLearning system across four UK hospital groups. @UniOfOxford @OUHospitals Listen here: https://t.co/mshLHGz5oD Read the paper here: https://t.co/D7bPQOtIWp

NEW Research: A scalable #FederatedLearning solution for secondary care using low-cost #microcomputing: privacy-preserving development and evaluation of a COVID-19 screening test in UK hospitals. @andrewsoltan @UniOfOxford @OUHospitals Read it here: https://t.co/D7bPQOtIWp

We also thank the participating hospitals/universities (@OUHospitals, @UniofOxford, @bedfordhospital, @PHU_NHS, @uhbtrust, @OxfordBRC) and the team, Anshul Thakur, @_yangjenny & Profs @drdavideyre, Tingting Zhu & David Clifton!

We particularly recognise @rhinohealth1 (Ittai Dayan et al), whose work at @nvidia was the first to implement FL in hospitals, @flwrlabs for their open-source framework, and @RaspberryPi_org for hardware! Grateful to @LancetDigitalH editors & reviewers https://t.co/SbYfgyihfl

Our results showed models trained with data from multiple hospitals, with federated learning, performed much better than models training just using data from one hospital. After training, the Pi's removable MicroSD storage was securely destroyed to protect confidentiality

Hospitals loaded anonymised patient data on to the Raspberry Pi's, and training took place without the data leaving their site. Models were aggregated using a server on @Azure. We trained & tested AI models to screen patients attending A&E for COVID-19.

We wanted to build a platform for any hospital to easily take part in developing AI models, without needing to share patient data. We used cheap microcomputers, called the Raspberry Pi 4B, and loaded them with all software & code needed for FL before sending out to hospitals.

Federated learning was developed by @Google researchers as a way to train AI models without moving data. But FL needs specialist expertise to set up at each hospital, and only a few researchers & companies have implemented it in healthcare. https://t.co/HvN8KoMo9N

Class imbalance -where one outcome is much more common than others- can be a major bias inherent in using routine health data. Here, @_yangjenny shows how an RL-based approach can train fairer multiclass classifiers, as a model-level approach to address bias within training data.

As the benefits of AI in healthcare continue to emerge, we investigate GPT-4's potential in radiology. Learn about research exploring GPT-4’s potential in assisting report structuring, classifying diseases, and generating comprehensive findings summaries:

@MSFTResearch Work kindly supported by @OUHospitals, @UniofOxford, Big Data Institute, @oxengsci & @OxfordCancer

Deciding how best to treat cancers needs expertise from multiple specialists & tests. I'm thrilled to share we've been awarded @MSFTResearch Accelerating Foundation Models grant, aiming to build models to improve communication & reduce delays in care. https://t.co/pV2eED57Gk