KCL enlists Nvidia to create AI platform for NHS hospitals

The platform uses deep learning algorithms run on supercomputers powered by Nvidia GPUs to discover new forms of medical treatment

health doc connect care telemedicine
Getty Images

King’s College London has teamed up with Nvidia to develop an AI platform that aims to facilitate more accurate and faster diagnosis by rapidly analysing NHS clinical data and medical imaging.

The research team believes that the system will develop new insights into the causes of diseases, treatments for patients and operational efficiencies, with the overall objective of providing better healthcare at a lower price.

The platform is the centrepiece of the new London Medical Imaging & AI Centre for Value-Based Healthcare at King's College London (KCL), where algorithms are being trained on an enormous trove of NHS medical images and patient pathway data to create new healthcare tools. The centre is focused on improving the experience for patients and their clinical outcomes across 12 pathways in oncology, cardiology and neurology.

"The idea is that all of these clinical pathways will utilise the platform that we're developing, to be able to make decisions that are relevant to the settings,” Dr Jorge Cardoso, CTO at London Medical Imaging & AI Centre for Value-Based Healthcare, KCL, tells Computerworld.

“And on top of that, we also have 10 startups that are working with us to prototype some of the tools we are developing, and transform them into algorithms and products that can be sold and have a wide impact beyond our own local Trust.”

Developing the platform

The platform will have to process a vast array of unstructured data, from imaging results to natural language text, and will therefore primarily use deep learning algorithms, which often require an enormous amount of computing power.

These needs led the research team to implement the Nvidia DGX-2 system, a two-petaflop GPU-powered supercomputer for AI research, which offers the memory and computing power necessary to rapidly and securely train the algorithms on large, 3D datasets.

Read next: What is NHSX? Inside the government's new healthtech unit

It will also use the Nvidia Clara AI toolkit, a selection of libraries for data, image processing, AI model processing and visualisation. Cardoso believes that the Clara framework will help the platform orchestrate algorithm training at scale and quickly make inferences. This combination of high throughput and fast predictions can be vital in critical hospital settings such as urgent stroke services, where life-changing decisions are sometimes needed in seconds.

National impact

The team of researchers from Nvidia and KCL is collaborating with clinicians from the London hospitals at Guy’s and St Thomas’, South London and Maudsley and King’s College Hospital, to ensure that their work is focused on medical needs.

Different models will be developed for each individual Trust to ensure that they fit their unique clinical attributes and patient demographics, but the data will be shared between a range of external organisations to build a larger, more encompassing overall model.

To ensure that the algorithms can be developed at multiple sites while protecting patient data, the system will use a federated learning approach, which allows it to both train models specific to each Trust while securely sharing the information between them. This allows data to be used from each individual hospital without it having to travel beyond the premises where it’s stored.

The solutions will be focused on local needs, but the aim is to extend the learnings across the country when the minimal viable platform is deployed next year. Cardoso believes that it could be used to conduct advanced research that healthtech unit NHSX could then translate into a national deployment strategy.

Read next: NHS promised £250m for AI: who gets the money and how will it be spent?

“A lot of the algorithms we are going to do will work locally and will work well in other hospitals,” he says. “NHSX could then go through the process of translating all of the work that we've been doing and other centres have been doing and translating that nationally to all of the other Trusts in the country.”

Copyright © 2019 IDG Communications, Inc.

Download: EMM vendor comparison chart 2019
  
Shop Tech Products at Amazon