Interview with Sir Michael Brady Part 2

Born just outside Liverpool, Michael was the first in his family to go to university where he fell in love with mathematics. He says: “I was really enraptured by a German mathematician, Bernard Neumann, and when I finished my degree I wanted to do a PhD in Bernard’s Australian lab.” Having spent two years in Australia, he  returned to the UK to do “something more practical than pure mathematics” and got involved in the developing field of computing science.  He was awarded a university lectureship in Essex, one of the biggest computing science departments in the UK at the time. After a few years, he realised that computing was a “pretty dull” except for AI and image analysis which he found “fascinating and intellectually stimulating” and allowed him to combine computing and mathematics.

In 1985, after six years as Associate Director of the AI Laboratory at MIT working on image analysis, robotics and AI, Michael returned to the UK and took up a role at the University of Oxford in information engineering. He says: “It seemed that they didn’t know what the hell that meant, but, that was OK, because I didn’t know what it meant either.”

Michael launched a Robotics Lab at Oxford. He says: “In the late Eighties, I thought it should be possible to build free-ranging mobile robots. The Research Councils in the UK thought this was completely nuts, they thought this would never happen within the next 50 years, but they were keen to support my research and we ended up building mobile robots in collaboration with GEC in Rugby.”

Despite the project’s success, GEC decided to close down its robotics support and Michael, as now Head of Engineering Science, decided to form his first spin-out company called Guidance in association with the three colleagues from GEC who had worked on the EPSRC project. Guidance eventually grew to three divisions in factory automation, marine, and monitoring for smart tagging systems. Each division was sold separately during the period 2010-2018.

Meanwhile, Michael’s mother-in-law died of breast cancer. He says: “Given that we were being supported so much by industry and the military for doing image analysis, I found it hard that anybody could miss a tumour that was that big. I was outraged by that. I went to a whole series of hospitals to try and find out why that was the case. I realised that doctors just did not have the kind of information and imaging technology that people in industry and the military took for granted. I thought, this is just nonsense.”

In 1994, when his tour of duty as head of department was finishing, Michael left robotics to work in breast cancer.  He says: “I dug in to working in breast cancer and that got me into medical imaging in general, with so-called deformable image registration because it’s very rarely the case that one kind of image, MRI or CT, will give you all the information you need. You quite often have to combine, to stratify. For example, if somebody has a mammogram and then they go for an MRI, you want to transfer the information that you got from the mammogram to the MRI.”

Michael, together with Alison Noble built a medical imaging lab in the Department of Engineering Science. He also wrote a book with Ralph Highnam. He says: “We tried to get our technology used by industry. Large industry wasn’t interested, they thought it was too academic. So, I figured we would do it ourselves, and so we formed two companies and then merged them to become Mirada Solutions.”

In 2003 Michael sold Mirada Solutions to CTI Molecular Imaging in the States, and became the CTI Mirada division.  Two years later, Siemens bought the whole of CTI Molecular Imaging to form Siemens Molecular Imaging.

Michael says: “It then became clear that they weren’t going to use the mammography work, which was my original driver. So, in 2008 I met up again with Ralph Highnam and we started another company which is now called Volpara Health Technologies.”

The company grew steadily through the next 8 years and in 2016 was floated on the Australian Stock Exchange.  Volpara Health Technologies currently has 250 employees and is based in New Zealand, Australia, and the United States.

Michael says: “By this stage I had got a bit of a name for starting companies based on technology, and so when I retired in 2010, I figured I could take on one more company: well I’ve ended up taking a lot more than one more company.”

Michael was introduced to Rajarshi Banerjee by Professors Stefan Neubauer and Matt Robson from the Oxford Centre for MRI, and together they formed a company called Perspectum which has already grown to 250 people with offices in Oxford, Dallas, San Francisco, Singapore, and Lisbon.

Michael says: “We started off by working on quantitative imaging of the liver. Our initial focus was fatty liver disease (steatosis) and non-alcoholic steatohepatitis (NASH), which is essentially the serious downstream consequence of steatosis. We have subsequently progressed towards liver cancer, towards cirrhosis, and then to multi-organ conditions such as diabetes. The most recent multi-organ condition is COVID, which is a viral infection that attacks in an inflammatory way a whole series of one’s organs, like the liver, the pancreas and so forth.”

In 2014, a long time friend and colleague, Nico Karssemeijer who worked in breast cancer in the Netherlands called Michael with thoughts about starting up a company to develop second generation computer-aided detection of breast cancer. Michael and Nico formed a company called ScreenPoint Medical which now has FDA clearance for both 2D and 3D mammography and which is growing rapidly based on investments from Siemens Healthineers and from New York-based Insight Partners.  Michael is also involved with a company called Optellum, started by two of his graduate students, Timor Kadir, Václav Pot?šil who invited him to join them doing assessment of lung nodules to either malignant or benign.

Alongside all of this, Michael and colleagues did a management buyout of Miranda to form Miranda Medical that specialises in radiation therapy planning.  He is also involved with Naitive Technologies.

More recently in 2018, after advising an ex-graduate student who had assumed leadership of a new AI lab in Abu Dhabi, Michael has been appointed to the board of trustees of the new graduate college called the Mohamed bin Zayed University of AI, MVCUAI, in Abu Dhabi.

Michael says his ambition has always been to “take the fruits of the science and technology that we build in the university out and to help people.  If I’m working in medical technology, it’s because I want to help people. Just writing papers was never enough for me. I found that big companies were very conservative, and so, if you really wanted to innovate, and you wanted to get innovative solutions out into helping people, then you have to form new companies.  It’s the start-ups that are generating most of the innovation.”

As a result, Michael finds himself nearing seventy seven and a serial entrepreneur but with continuing links to university research. He says: “So, one way and another, I find myself at my ripe old age being involved with, Perspectum, ScreenPoint, Optellum, Naitive, MBZUAI, a little bit with Oxford University still, and with Mirada and with Volpara.

“I have become absolutely convinced that there is a fundamental and deep relationship between working on hard problems in science and making things work in practice with clinicians, usefully, reliably, 99.99 per cent of the time, 24/7. I’ve really believed that there is this close coupling between the hard problems that clinicians face and the kind of science that we want to do in universities.  They are not separate, they are joined together.”

Michael Brady, believes three things have fed off of each other over the last forty years to have an impact on his work and on the diagnostics of breast cancer. He explains: “The first obviously is the power of computing.  A mobile phone now has roughly 1,000 times the computing capacity of the computers that were on the Apollo landing craft, and that’s just a mobile phone. In my backpack when I cycle to and from home, I carry something like ten times the amount of compute power than was in my entire laboratory of 200 people at MIT 40 years ago.  It’s a staggering amount of compute power and it’s getting more and more and more all the time.

“As part of that increase in compute power, has been an increase in graphics.  If you look at the quality of images that we see on computers now, compared with what they were even ten or twenty years ago, it’s extraordinary, so much so that the images that clinicians can see on computer screens now are as good as they ever used to see on film thirty years ago.  So, computing and imaging has reached the point where we can do things that were considered inconceivable thirty years ago, and can display the results in a way that engages with clinicians to inform their judgement to give decision support.  And, that power has been harnessed by equally massive developments in software engineering, and, for me, the apotheosis is AI.  It always has been and always will be.

“The second is that the emergence of the Internet and then the cloud. We now have a fantastic global infrastructure of information, with not only distributed storage, but distributed compute power. That’s what the cloud gives us.  So we can now, not only take one single computer which has got more, more powerful, but we can engage 100 computers in a completely geographically distributed sense. For example, my company here in Perspectum, if we want to train technology to learn new a method for segmentation, we will quite often train it on 50,000 cases, but we won’t do it on one computer.  We’ll do it through the cloud on 64 to 128 computers that might be in four different continents. It’s cheap, it’s effortless. Almost infinite storage, almost infinite compute power, and globally distributed information.

“The third thing is that the world has been transformed by molecular biology.  It is not just looking at strings of genes, and it’s not just looking at wonderful pieces of technology such as CRISPR-Cas9; it’s also the wider genome, understanding amino acids and proteins, protein structure, it’s understanding the role, for example, of Messenger RNA, which we have just seen so spectacularly with the development of vaccines against COVID. When you take those three technologies:  you end up with a totally transformative way to think about medical problems.”

As an example, Michael points to the first digital mammography systems which were developed in the year 2000. Until 2000, mammography images were done via film which needed to be digitised and run offline. All of which took considerable time. Michael says: “With the advent of digital mammography images were captured digitally, this meant that we could process them digitally, which gave rise to computer-aided detection, automatic detection of tumours and microcalcifications.

“However, the bad news was that the algorithms weren’t very good; they created huge numbers of false positives, and as a result there was over-diagnostics.  There were far too many false recalls with that.  All of that was just 20 years ago. However, ScreenPoint Medical, has developed decision support technology, not only for 2D but also for 3D mammography, which now outperforms all but the best breast radiologists, at a time when the number breast-specific radiologists in Europe is reducing.

“So although it’s mandated that all mammograms are read by two radiologists, in practice even twenty years ago that became one radiologist and one technologist or radiographer. Quite often now it’s two radiographers.  However, what I think is vastly more important is that we have shown that if you take one reader, plus our Transpara technology, the combination will outperform both the radiologists by themselves, or the technology by itself.    Every single radiologist, no matter what their skill level, has their ability improved.  That’s game-changing.  It also tells me that we’re beginning now to be able to take 2D, 3D mammography, and link it to MRI, to look at things like, for example, breast hormonal composition, breast density, and to look at following the progression of breast cancer over years.

“So, computers, plus the Internet, cloud, plus biology, have completely revolutionised how we detect, how we treat, how we choose treatments, how we monitor the treatment of breast cancer.”

Through the companies that Michael and his colleagues have started, they have got installations in around twelve and a half thousand hospitals around the world.

Breast cancer is just one example of a condition which has been and is being transformed by technological advance, but Michael says: “We’ve barely started yet, there’s a vast amount more that will be done, can be done. But the same is true for liver disease and heart disease.  We are beginning to move now from treating people who are sick to beginning to anticipate disease, to treat people when we find, perhaps even incidentally or accidentally, the earliest stages of disease and intervene when the prognosis will be vastly better.

Michael says that the longer he has worked in the field the more he cares about taking the science into the real clinical work, the real world of drug development, and in what people call impact. He adds: “You would never have impact with any technology unless it works essentially all the time, and to make it work all the time, when you’re dealing with the diversity of humans, with the diversity of pathology, you can only achieve that kind of result if the technology is based in really deep science. There is this combination between deep science and impact.”

In looking at how AI can help in the current situation, Michael is keen to differentiate between machine learning and AI. He says: “Machine learning is only one tiny piece of AI. The two terms, AI and machine learning, are used as though they’re the same thing and that’s just nonsense, they’re not.”

He believes that although AI is having a very considerable impact, it’s not only machine learning that is going to have an impact as we go forward.

Michael is currently involved in a project called COVERSCAN to help pull together information needed to help manage COVID care.

He explains: “COVID is a very complex viral pathology which impacts upon multiple organs.  The major impact of COVID that we have seen so far is in inflammation of the heart, myocarditis, but it’s also been an inflammation of the pancreas, and of the liver. It is in fact a classic instance of a multi-organ condition.  COVID, just like type 2 diabetes, is a paradigm of a multi-organ condition.  In fact you quite often get the same thing with metastatic cancer as well.  So more and more we are beginning to see chronic multi-organ conditions.

“That’s a challenge for the medical profession, because medicine has become more and more and more specialised. You don’t get general internists; you’ll get liver specialists, pancreas specialists, kidney specialists, heart specialists, or lung specialists. COVID impacts all of them.

“Therefore, we need methods that can provide, in the first instance, information about all of these various organs to a team of clinicians who will work collaboratively. If you have COVID, and long COVID particularly, and your version of long COVID happens to be mostly impacted on your pancreas, it’s not going to be helped that much if you see a cardiologist who knows nothing about the liver and nothing about the pancreas.

“AI and image analysis, first of all can play a role in pulling together the information from these different organs. That, in essence, is what we’ve been doing with the COVERSCAN, a project now, a product, which is cleared by the UK authorities, and is now being cleared by the FDA in the United States.  We expect to roll out COVERSCAN as a product in Europe and the USA over the next twelve months.

“AI is not just about machine learning.  Machine learning can find a whole series of relationships.  It can detect things, but it currently cannot find relationships.  Very much like in statistics we find correlations, but correlations are weak.  What we really need to do is to understand the causal structure that underlie many of these complex multi-organ conditions.  Each of the chains of causality that we see is what is known as an aetiology, a disease course. For example, in the case of diabetes, there are something like five, six major different aetiologies.

“There are now methods for building formal methods for causality. Combining them with image analysis and machine learning will be the next wave of AI in medicine. I have two colleagues here at Perspectum who are working on applying these causal methods, one to diabetes, and one to primary liver cancer; hepatocellular carcinoma.  And I think that, causal reasoning, plus machine learning, plus signal and image processing, brought together, will be the next great wave of AI in medicine.”

Michael says: “The technology of AI, of image analysis, of biology, compute power, and the Internet/cloud, have the potential to take all of these technologies and distribute them through the world to level up the awful discrepancies there are between, for example, Sub-Saharan Africa, and Europe and the United States.  Will they do that?  That’s not a question for technology; that’s a question for politics and social, and the international political community.

“All the evidence over the past thirty years has been that despite the emergence of these technologies, the gap between the wealthiest in our society and the poorest in our society has gotten bigger.  It has not shrunk.  In every single place, the gap has gotten bigger.

“We have seen that most recently with lockdowns following COVID, where governments simply assumed that everybody in society will have the Internet, and multiple computers in their homes, and have been surprised to find out that the poorer parts of society don’t have Internet, don’t have computers, don’t have lots of iPads.

“It shows just how totally out of touch most politicians are of the reality of life amongst the poor people where I was born into. The technology is completely neutral about levelling up.  It provides the potential for it.  Politics, I’m afraid, will drive it in exactly the opposite direction.”

Interviewed by Elisabetta Mori

Transcribed by Susan Hutton

Abstracted by Lynda Feeley