In 2020-21, there are few industries where that change is more visible – or more needed – than in the healthcare sector. We’ve spent the year helping healthcare organisations accelerate research using COVID-19 chest X-rays, allocate resources for the COVID-19 response, and define internal strategies for enabling the safe use of AI in practice.
It’s not just the NHS that can benefit from AI. From frontline patient care to lab research, our Faculty fellows helped to tackle issues across the healthcare system – and it’s clear that this is only part of a growing wave of AI experimentation in the sector.
Let’s take a closer look at the projects covered.
Birdie: Anticipating drug interactions and side effects with AI
If you spend time really getting to know the health and life sciences industry, you’ll have heard a story like Birdie’s several times. One organisation, working tirelessly to solve an issue that has plagued healthcare providers for years, but with one significant barrier standing in the way: the incredible complexity of patient medical data.
Birdie offers software that helps social care providers to support elderly patients living at home. This, of course, means working with patients that take a wide range of medications. For many elderly people, the list of medications grows longer every day. And every new medication brings an increased risk of side effects, falls, decreased mobility and adverse interactions with other medications.
It’s a significant issue: drug side effects and interactions lead to over 600,000 hospital admissions and 5000 deaths of elderly people each year. Tackling these issues costs the NHS £400 million every year.
Birdie had the data needed to anticipate and prevent these problems – staggering amounts of it, in fact. Over 7 million free text observations from carers visiting elderly patients in their homes, plus 5 million records of which medications patients were taking and when they were taking them.
Oscar, one of our Faculty fellows, joined a team working towards one goal: building a tool that could help Birdie’s customers identify the patients most at risk of adverse reactions.
Alongside the team at Birdie, with support from the team at Faculty, Oscar helped to build a tool that analysed anonymised versions of the patient data collected by carers. Using statistical inference, the tool analyses thousands of timelines for individual patients – recording which medications they took, when they took them, and when they experienced any adverse effects. The tool then highlights associations between medications that are linked to increased risk of adverse effects with instances of adverse effects recorded in the text logs.
The results could be life-changing – or even life-saving – for patients of carers using Birdie’s tool. With it, carers can get an even deeper insight into their client’s health, instantly understand medication effects and mitigate them where possible.
The Francis Crick Institute: using AI to identify cell types with a single dye
‘Solving Parkinson’s, one pixel at a time’ is a bold claim, but the Francis Crick team was well-placed to make it.
As Faculty fellow Mustafa described, many studies of Parkinson’s rely on studying two types of cells affected by the disease: neurons and astrocytes. By examining the neurons and astrocytes of Parkinson’s patients, researchers can uncover vital information about the effect of drugs or the progression of the disease.
But there’s a problem: it’s very difficult to tell the difference between neurons and astrocytes without dyeing the cells with their respective cell-type dyes, a process which kills them and makes them unsuitable for many forms of further study. To support the progression of several key research areas, the Francis Crick Institute needed a new way to distinguish between the cells.
Alongside the Francis Crick team and Faculty, Mustafa applied a novel approach to cell imaging, using one single dye for both neurons and astrocytes. The chosen dye is the same that is utilised for most studies of Parkinson’s disease and, most importantly, doesn’t kill cells in the process.
Mustafa then applied two machine learning techniques to the images of the cells in question to distinguish the two different cell types. The first method involved applying a convolutional neural network (CNN) directly to the images, delivering 96% accuracy. The second applied a regression model to biological features engineered from the images at the cell level, identifying the correct cell with 90% accuracy. Both approaches were capable of achieving cell-type identification, while using one single dye that is normally utilised in Parkinson’s disease and therefore does not preclude the utilisation of the cells for further studies.
This is groundbreaking work for the Crick Institute. These new approaches don’t just save time and unlock previously inaccessible insight; the technique can also be applied to a host of research areas concerned with neurones and astrocytes, not just Parkinson’s.
What’s next for healthcare and AI?
From fellowship projects like these to nationwide projects like the NHSX AI Lab, it’s clear that AI is playing an increasing role in the quality and efficacy of research and patient care.
You can find a sample of our other health and life sciences projects here, or get in touch to discuss your own AI needs with our health and life sciences team below.