Features

Diagram by Florence Nightingale from Harriet Martineau, England and Her Soldiers, London, 1859.

Whatever Hollywood might say, AI is not all about killer robots in a far off future. It is more mundane, more everyday - and much more ubiquitous. And thinking it just belongs in a sci-fi film is dangerous, since this leads to a sense it is not relevant or that it is even unreal. Professor Ursula Martin bristles slightly at the very idea of robots, killer or otherwise, and points out that everyone’s lives are already affected by AI – and there are ‘bigger things at play’ than people realise.

The Maths professor and computer expert has put together a fascinating exhibition in the Bodleian  about the history of AI – and has raided Oxford’s collections  for treasures which showcase thinking about AI,  illustrate fundamental ideas, and  provoke debate.

AI is very useful. It’s all around us...It is changing our lives...Making...AI all about killer robots gets us off the hook of taking responsibility and distracts attention from the more everyday reality of AI

Professor Ursula Martin

Victor Frankenstein played God, she says, and Mary Shelley’s manuscript still has the power to shock. In the exhibition, it is open at the page where 'by the glimmer of the half-extinguished light' Frankenstein sees for the first time 'the dull yellow eye' of the creature he has created in his laboratory.  Another exhibit is Ada Lovelace, describing Charles Babbage’s calculating machines. She places them in a contemporary theological debate: did the the creator of the machines challenge God, or merely help us  to understand His works better?

Meanwhile, Ramon Lull’s colourful medieval diagrams present simple reasoning as an almost mystical process. But, for the 19th century economist Stanley Jevons, the goal was more mundane. His 'Logic Piano', a construction of ivory, wood, wire, showed, in principle, it was possible to mechanise human reason itself.

The logic pianoThe logic piano
Today’s AI can produce huge mathematical proofs built up by  similar chains of reasoning. But it cannot yet explain them, as a human mathematician could.

 Jonathan Swift, meanwhile, imagined a machine that could write a book on any topic ‘without the least assistance from genius or study’. And, in the early days of the computer, Christopher Strachey experimented with simple computer generated love-poems: the fore-runner of today’s AI software. It can mine millions of existing texts and 'learn' rules to create 'plausible' new texts about  any given topic.  Strachey’s programme used a very restricted vocabulary, which gave the poems an oddly prim and stilted tone.

Similarly, it is all too easy for modern AI to reflect biases in the texts it has learned from, propagating and amplifying those biases.

Professor Martin points out, to deliver  modern AI involves vast quantities of data, and fast computers that use clever algorithms, not just for calculation, but to reason and  find patterns too. Looking at early examples, such as Strachey’s poems, shows just how straightforward some of the underlying ideas are.  According to Wadham College-based Professor Martin, it is the scale of the data and the power of the computation that transforms these simple ideas into present-day AI. Understanding them in context can give us new ways to think about contemporary concerns as well.

Professor Martin and the teamProfessor Martin and the team
She says, take for example Florence Nightingale’s 'Rose diagram', what we would now call a 'data visualisation',  of  hospital  deaths during the 19th century British campaign in the Crimea. It demonstrates that most deaths were due to infection, and is based on data painstakingly collected and analysed by hand. Such techniques were, and are,  widespread in the British government and armed forces, and some of the underlying assumptions that shaped how the data was collected and used, have had long-lasting consequences.

‘There are so many stories you can tell from the Bodleian archives,’ Professor Martin says enthusiastically. ‘We’re trying to tell the history of AI here in 15 objects.'

So it is emphatically not all killer robots. Professor Martin adds, ‘AI is very useful. It’s all around us, for example your phone or your satnav or your bank are full of AI. It is changing our lives, and both as individuals and as society as a whole we need to think and act responsibly, just as we should with any other technology. Making conversations about AI all about killer robots gets us off the hook of taking responsibility and distracts attention from the more everyday reality of AI.’

How artificial intelligence is shaping medical imaging

Dr Qiang Zhang of the Radcliffe Department of Medicine explains how artificial intelligence is being used to help researchers and physicians interpret medical imaging.

Disruptive AI-based imaging technology might replace the injection of dye ‘contrast agents’ usually needed to show clear images of scar of the heart

Imagine you are a medical doctor, faced with a patient with suspected heart disease for symptoms such as chest pain, tightness, or shortness of breath. One way to find out what is happening, and help guide patient prognosis, is to do a cardiovascular MRI scan to look into any heart muscle abnormalities. The scan involves injecting a ‘contrast agent’ (a dye that will improve image contrast and show up scars on images) into a vein in the patient. Contrast-enhanced MRI has been the clinical standard to provide clear scar images, but it’s painful, and makes already expensive MRI scans even more so.

What’s more, this method is limited in patients with significant kidney failure – their kidneys have difficulty clearing the dye from their bodies, sometimes leading to irreversible complications. Some patients will be allergic to the contrast agent, and you might want to limit the use of injectable contrasts in some patients, such as pregnant women and children.

So how do you find out about what might be going on in your patient’s heart in that case, without injecting into them a contrast agent?

It turns out that injecting a contrast agent might not be the only way to get clear MR images to reveal scars in the heart muscle – in 2010, Professor Stefan Piechnik from the Radcliffe Department of Medicine at Oxford University came up with a method to study heart muscle properties, using a contrast-free MRI technique called T1-mapping. It produces an image of the heart with numerical values that change with different diseases.

Such contrast-free MRI contain a lot of information about heart tissue properties, some of which is subtle, or difficult to identify as a scar or other pathologies. As of now, researchers are still exploring the best ways to interpret and display the information from these contrast-free T1-maps, which is one of the reasons that they are not yet widely used by medical doctors.

This is why our cross-disciplinary team of AI scientists, magnetic resonance imaging specialists and cardiologists at the University of Oxford worked to find ways that artificial intelligence (AI) can enhance these contrast-free MRIs, to produce clear images of heart muscle scarring. AI effectively works like “virtual contrasts” to replace conventional intravenous contrasts.

We developed an AI-powered algorithm to combine multiple contrast-free MR images together with heart motion information, enhance the pathological signals in them, to reveal scars in a similar way to conventional contrast-enhanced MRI. This technology is called “virtual native enhancement”, or VNE, as it acts as an enhancer for the MR images, using only the ‘native’ (ie, non contrast agent enhanced) images produced by an MRI scanner.

In 2021, our team released the first proof of concept for this idea, by detecting scars in the heart muscle for patients with hypertrophic cardiomyopathy, a common genetic heart disease affecting 1 in 500 people, and the most common cause of cardiac death among young people.

Recently, we have found that VNE can also detect scars in patients who have had a heart attack. We compared contrast-free VNE with conventional contrast-enhanced MRI in these patients. We found that VNE highly agreed with the conventional MRI in detecting previous heart attack scars and their extent. Additionally, the VNE image quality was actually better, all without the patients needing to receive an injection.

Once completely validated, this new technology may slash the time that patients need to spend in an MRI scanner from the standard 30-45 minutes to within 15 minutes, saving more than half the scan cost, yet producing images that are clearer, more diagnostically useful, and easier to interpret.

AI imaging image1.jpg

Image: Development of VNE in detecting heart muscle scars for two different heart diseases. The right panels show our new contrast-free method, while the left panels show conventional contrast-enhanced which requires injecting contrast agents. Arrows point to the detected scars.

We think that these successive breakthroughs mark the beginning of a new era of diagnostic medical imaging, using AI instead of IV contrasts to reveal pathologies in the human body: we might finally be able to get rid of injections when it comes to heart MR imaging.

AI imaging image2.jpg

Image: Background of IV contrasts of MRI, and the emerging new era of AI “virtual contrasts”.

We are now working to further improve the capabilities of this technology, to detect more complex heart diseases and their underlying mechanisms, beyond the diagnostic power of current MRI. We plan to use these methods in large clinical studies as a diagnostic tool for novel investigations.

We think that this kind of Virtual Native Enhancement technology is an exciting and potentially game-changing advance for clinical MRIs in the future. Patients going in for a clinical MRI scan might not need an injection for most MRI scans, not just for the heart, but potentially for other organs as well. This would cut costs for healthcare providers, meaning that many more patients could access MRI scans; the risks of contrast-agent injections complications would disappear too. We hope adoption of this method could contribute to the digitalization of the NHS, something which is very much needed to address the backlog post COVID-19 pandemic.

With thanks to Professor Stefan Piechnik and Professor Vanessa Ferreira, joint senior authors of the study.

Further reading in Circulation: Zhang et al 2021, Zhang and Burrage et al 2022.

How decisions form in the brain: A physical basis for a cognitive process

Alberto Lazari of the Nuffield Department of Clinical Neurosciences explains the importance of insulation in our brains' wiring.

Our brains contain a striking amount of ‘brain wires’, which allow electrical signals to send important information from one corner of the brain to another. Although these brain wires are made up of biological material, they also bear surprising resemblances to the electrical wires you can see when you do a DIY job in your home. For instance, one key feature that allows the brain wires to work is that they are tightly insulated. A little bit like metal wires are coated with plastic, brain wires are also wrapped in an insulation material, called ‘myelin’. Myelin is essentially a fatty layer of insulation, wrapped around many of the wires in your brain.

Myelin is incredibly important. When this insulation layer breaks down, the brain struggles to transmit signals at its usual speed, which is what happens in conditions like multiple sclerosis. However, the insulation of brain wiring has often been overlooked by scientists. It is particularly difficult to measure non-invasively in a live human. On top of that, this insulation has long been considered a static part of the brain which is not particularly relevant to understanding the brains of healthy adults. While myelin is clearly important in multiple sclerosis, until recently very few scientists had studied myelin beyond the realm of disease.

However, recent studies have now called some assumptions about myelin into question. In particular, in the past decade many labs around the world, including here at Oxford, have shown that myelin is more complex and dynamic than previously thought. Ground-breaking new methods have also been developed to effectively measure fat-rich insulation through magnetic resonance imaging (MRI), allowing us to ask new questions about this ever-elusive insulation layer that envelops our brain wiring.

For example, we know that everyone has a different brain, and brain wiring is one way that our brains differ from each other. Do different people have different levels of wiring insulation? And do these differences between individuals influence how our brains work? As simple as these questions may sound, they had not been asked before - until now.

At the Wellcome Centre for Integrative Neuroimaging, we set off to find an answer, using new MRI techniques to study myelin. First, we scanned a large group of participants and captured detailed MRI brain scans which gave us information about myelin. We then tested the same participants with a type of non-invasive brain stimulation called Transcranial Magnetic Stimulation, or TMS. Using TMS, we can create fast electrical signals and track them across the brain on a millisecond scale. This technique allowed us to capture fast electrical communications between brain areas – even those on opposite sides of the brain. This was particularly useful, because this very rapid electrical communication along brain wires is exactly what we expect would be influenced by insulation, very much in the way that the insulation of metal wires in our homes changes their electrical conductance.

Our findings showed for the first time that variation in brain wiring insulation between people is associated with significant differences in how brain areas communicate. For example, participants with more myelin in a given “brain wire” connecting two brain regions also tend to have a stronger electrical connection between those two brain regions. This is important because it confirms the significance of myelin not just to disease, but also to the everyday functioning of the brain. It also demonstrates the utility of studying myelin to understand the fine details of how different regions of the human brain communicate with each other.

Finally, our results also carry important practical implications. If our individual brain wiring insulation is linked with how we respond to brain stimulation, could information about myelin be used in the future to study clinical responses to brain stimulation? For example, TMS is already being used as a promising therapy for major depression, but with huge variability in how people respond to this treatment. Could information about brain wiring insulation tell us more about why some people respond better than others to TMS? And could this eventually help us better tailor treatment? We still do not have answers to these questions. However, what is certain is that this fat-rich insulation of our brain wiring, once thought to be a totally uninteresting part of the brain, is likely to have some more exciting surprises in store for us.

The paper, 'A macroscopic link between interhemispheric tract myelination and cortico-cortical interactions during action reprogramming',  can be read in Nature Communications.

Finding out more about Parkinson’s by monitoring symptoms at home

Professor Chrystalina Antoniades of the Nuffield Department of Clinical Neurosciences explains how the COVID pandemic accelerated an innovation in one research project into Parkinson's Disease.

Parkinson’s is a progressive neurological condition, which affects around 145,000 people in the UK.

Symptoms start to appear when there isn’t enough of the chemical dopamine in the brain to control movement properly. People with Parkinson’s don’t have enough dopamine because some of the nerve cells that make it have died.

There are lots of symptoms, but the three main ones are tremor (shaking), slowness of movement, and rigidity (muscle stiffness).

Doctors typically diagnose and monitor the progression of Parkinson’s by assessing these symptoms using a ‘clinical rating scale’. This relies solely on the clinician’s own subjective impression of the person’s condition.

Since 2016, the NeuroMetrology Lab at the University of Oxford has been developing objective numerical measures to help doctors accurately diagnose disease and monitor the progression of Parkinson’s – which could lead to the provision of more targeted and timely treatment. Until recently this research team, based in the Nuffield Department of Clinical Neurosciences, has been carrying out their research via in-person clinics, attended by patients four times a year.

During the patient’s two-hour clinic visit, the researchers would measure subtle abnormalities in the speed and coordination of fast eye movements (known as saccades), hand movements, and gait. They would also assess cognitive performance using tasks on a tablet. Then they would try to work out whether these numerical measures could accurately and objectively quantify Parkinson’s, and track its progression over time.

The advent of home monitoring

One of the features of Parkinson’s symptoms is that they fluctuate both throughout the day and from day to day. So the research team always knew that they wanted to be able to monitor symptoms at home as well as in the clinic. They were aware that patients’ behaviour during short clinic visits every few months was probably not representative of the condition’s progression overall.

In 2020, the Covid pandemic put an immediate stop to research with human participants, making in-person clinics impossible. This apparent disaster in fact accelerated the researchers’ plans to roll out wearable technology and enable study participants to monitor their symptoms at home.

The biopharmaceutical company MSD is funding this new phase of the research project. The new grant has enabled the team, which I am leading, to work with MSD and the technology company Clinical Ink to capture data on participants’ symptoms at home. The wearable technology combines an Apple watch and phone to test a range of both motor and cognitive aspects of Parkinson’s.

I am delighted to be able to offer to our research patients the opportunity to be monitored so closely by such clever technology. My team has been working hard to make this a pleasant experience for all our patients and we are incredibly honoured to have such tremendous support from the Parkinson’s Disease community.

‘Digital health technologies offer tremendous opportunity to measure and objectively quantify the symptoms and progression of neurological disease,’ said Dr Marissa Dockendorf, Executive Director, Head of Global Digital Analytics and Technologies at MSD Research Laboratories. ‘MSD is excited to collaborate with the University of Oxford to further the development and characterisation of digital measures to support timely and reliable evaluation of potential new treatments for Parkinson's disease.’

How patients can monitor their symptoms at home

A member of the team sets everything up with participants remotely during a telemedicine appointment, explaining how to use the watch and the app on the phone. The participants receive instructions and the app gives step by step guidance on what to do. Participants are required to carry out testing at home once a month, performing tasks on the app such as reading, testing reaction times, and cognitive tasks.

David Williams, a participant in the study, said: ‘The wearable technology is very easy and comfortable to use. The instructions are very clear, the exercises are well explained and not at all difficult to accomplish. The staff are friendly, approachable people who always leave me with a sense of being valued as a contributor to what is obviously a very important research study. If you’re at all anxious about taking part, don’t be, just sign up. You won’t regret it!’

Kevin McFarthing, another participant, also stressed how easy it was to carry out home monitoring: ‘The OxQUIP team is very professional and thoroughly well organised, and are a pleasure to work with. They did a great job training me to use the remote devices’, he said.

The home monitoring does not replace clinic visits entirely; patients have a telemedicine appointment every four months, as well as the opportunity to come in to an in-person clinic if they wish.

Joan Severson, Chief Innovation Officer at Clinical Ink, said: ‘We are honored that our mobile and wearable technology plays an integral role in this study of Parkinson’s disease in Oxford. We are excited to collaborate with researchers who tirelessly work to increase objective numerical measures for diagnosing and monitoring disease progression.’

Looking to the future

The Covid pandemic was a dark period for many, and yet it accelerated this change in the way this research project is being carried out. The team is now able to gather richer, more nuanced and accurate data to feed into their analysis.

The outcomes of this project will improve the diagnosis, tracking and treatment of Parkinson’s. The insights gained about monitoring disease progression will make the assessment of clinical trials more efficient, leading to faster drug discovery not only for Parkinson’s, but potentially for a range of neurological conditions.

This work is part of the OxQUIP (Oxford Quantification in Parkinsonism) programme. If you’re interested in taking part in this study, please email Oxquip@ndcn.ox.ac.uk.

Professor Paul Riley at the IDRM Opening - July 2022

Professor Paul Riley, Director of the Institute of Developmental and Regenerative Medicine discusses how better-designed research buildings can help scientists break out of their silos.

The advances made in medical and biological sciences within our lifetime are staggering. It seems strange to think that the project to sequence the human genome took over a decade to complete back in 2001, yet similar sequencing technology is now portable and is used daily in the field.

Each advance in research has opened up new avenues of exploration and with each new stride whole new research disciplines have emerged. Sadly, in modern science it is impossible to be an expert in more than a few things, despite the fact that most scientists will have chosen their careers early on because of a fascination with all things to do with science, and a desire to keep learning and making new discoveries themselves.

Yet we are increasingly finding, in areas such as our response to pandemics or cancer treatments, it is vitally important that these related avenues of research should remain in contact with one another. The discovery of a new technique in one area could be just as useful to another, and research into complex medical issues is increasingly becoming multi-disciplinary.

Some years ago we began to plan a way of creating a new type of working environment for researchers, which could encourage scientists to mingle more with people from outside their own niche discipline. The logic is simple – scientists are passionate about their work and love talking about it. But the problem with many labs is that most researchers find themselves surrounded only by others from their own field and much of what is going on elsewhere is behind closed doors.

This was the idea behind the Institute of Developmental and Regenerative Medicine (IDRM) – to bring the related disciplines of cardiovascular science, immunology and neuroscience together under one roof, and to design it in a way that increases the chances of these groups mixing socially and professionally, promoting conversations to stimulate new ideas and collaborations between students, post-docs and PIs.

The newly-opened IMS-Tetsuya Nakamura Building is the home for the IDRM, and has been designed around shared common and break-out spaces which differ across each floor in flavour thus linking the laboratories and offices within each discipline vertically to promote mixing and collaboration from the ground-up.

My own research in cardiovascular science involves regular collaboration with colleagues in neuroscience and immunology, who I will now have on my doorstep, and who I can see coming and going each day. Georg Holländer’s group work on how immune cells learn ‘self’ from ‘non-self’ during development. Following a heart attack release of certain proteins from damaged heart muscle triggers a ‘non-self’ reaction, worsening the outcome and promoting heart failure. We can now work together to ask broad questions as to how cells identify ‘self’ and how can we intervene to ensure the body’s own immune response does not react badly during a heart attack. Each experiment and subsequent discovery can be discussed with the immunologists on the floor above in our breakout space.

The siting of the IDRM was also designed to put the researchers at the heart of the science and technology cluster in the Old Road Campus, and between the Headington hospitals where many researchers also work. For example, combining state-of-the-art imaging facilities across the road in the Kennedy Institute with newly purchased microscopes in the IDRM into a new Centre of Excellence will be a powerful tool for many researchers across both institutes and the wider campus.

But the drive to improve cross-disciplinary collaboration goes far beyond just the new building. We are working to improve our links to disciplines within maths to model disease, and the social sciences, such as law and ethics, which will have a major role in shaping and guiding research, as well as collaborations with industry on site, such as Novo Nordisk, and the BioEscalator facility to help turn new research into successful spinout companies that can develop real-world applications.

How effectively we manage to our cross-disciplinary collaborations will play a large part in the speed and efficiency of future research and development and the delivery our findings into clinical care but it will also make a career in science even more engaging.