Sir Michael Brady is Emeritus Professor of Oncological Imaging at the University of Oxford, having retired in 2010 as Professor of Information Engineering. He is co-Director of the Oxford Cancer Imaging Centre. He is distinguished for his work in artificial intelligence, and for his outstanding contributions to developing computer-based post-processing for a variety of medical images. He combines his work in oncology with a range of entrepreneurial activities. He was Deputy Chairman of Oxford Instruments, and also a founder of successful start-ups such as Guidance, Mirada Medical, Optellum, Perspectum Diagnostics, ScreenPoint Medical, and Volpara Solutions among others. Sir Michael was elected Fellow of the Royal Society, Fellow of the Royal Academy of Engineers, Fellow of the Institution of Engineering and Technology, Fellow of the Institute of Physics, Fellow of the Academy of Medical Sciences, and Fellow of the American Association of Artificial Intelligence, and also he is Membre Étranger de l’Académie des Sciences. In addition to this numerous academic fellowships and prizes he received a knighthood in 2004 for services to engineering.
Professor Sir Michael Brady was born in Prescot, Lancashire in April 1945. He grew up with his parents and two sisters in a close-knit family. Early Life
Having attended his local primary school, in 1956, Michael started at Prescot Grammar School, a boys-only school, where his favourite subjects were maths and physics. After taking A levels, he progressed to Manchester University to study maths. He says: “In those days there were relatively few kids who would go to university; the norm was that you went to work when you had finished school. I went to university and was the first person in my family ever to do that, and it changed me, completely.” In his first year at Manchester, Michael went to a series of lectures by an émigré Jewish mathematician by the name of Professor Bernhard Neumann; a Fellow of the Royal Society. He would go on to be the influence for Michael to study for his PhD in Australia, as Michael explains: “He absolutely captivated me, but then he (and his wife) moved to Australia, to take up chairs at the Australian National University. So when I was finishing my degree, and was scheduled to get a First Class Honours degree, I decided that I would like to do a PhD, and if I was going to do a PhD with anybody it was going to be with Bernhard. That’s why I went to Australia.” In 1967, as well moving to Australia, Michael also married Naomi Friedlander, a fellow student who was studying pharmacy. The couple have been married for 55 years. The pair moved to Australia, Michael had won a Commonwealth Scholarship to study mathematics for a PhD and Naomi worked as a pharmacist. Michael says of the time: “We travelled a lot throughout Australia. I played football for the Australian National University in a professional league. We played a lot of squash. We partied a hell of a lot. It was a world class research laboratory doing mathematics, and ideas were just flying around; they set a standard of doing mathematics that was just fantastic.” Education
In 1970, after completing his PhD, Michael and Naomi returned to the UK where Michael took a role in the computer department at the University of Essex, having decided that he didn’t want to work in pure maths. He explains: “I realised that what I really wanted to do was something that was applied. … I had the thought, I wonder what it would be like to work in computing, because computers were just becoming available.” Michael’s first experiences on an ICL machine were at Essex University. He adds: “I wrote simple ALGOL programs just to find out what it was like, but you weren’t allowed to go and see the computer, you just posted a card deck through the wall and you got back a sheet of paper the following day telling you that your program had made a bug etc.” Michael started his ten-year career at Essex working on Turing machines, recursive function theory, etc. He also wrote a book called The Theory of Computer Science. He also came across artificial intelligence (AI). He says: “I realised that what I had found myself doing in theoretical foundations of computing science weren’t that much different from pure mathematics and that really wasn’t what I was seeking. I was still looking around for an application and I came across AI, and in particular came across image analysis, and it was like a clap of thunder. I just thought, wow, this is what I’m going to do for the rest of my life.” In 1975 he applied to what was then the EPSRC for a grant to pay for him to travel to what Michael describes as “the twin Meccas of AI; MIT and Stanford”. He adds: “I asked for travel and accommodation for two months and to my astonishment they wrote back and said they would give me travel but they would like me to go for three months, and would I also go to Carnegie Mellon and various other places.” While he was at MIT, Michael struck up friendships with David Marr, Berthold Horn, and Patrick Winston, the director of the lab. He adds: “I just really loved being there. I got to meet my heroes, Marvin Minsky, Seymour Papert. Just to be at the AI Lab, surrounded by these various people, with these brilliant young students, like Gerry Sussman and Carl Hewitt, it was just like every day was Christmas. I just couldn’t get enough of it.” University of Essex
Having developed a friendship with Patrick, David and Berthold, Michael was invited back to MIT annually until in 1978, Patrick invited Michael to be the lab’s Associate Director. Having obtained visas, the family move to the US in 1980. Michael says: “The thing I remember most is what a phenomenal place the AI Lab was but what struck me, and really informed the work we did when I came back to Oxford, was the sheer brilliance and power of the PhD students. I realised that the PhD students were really the engine of all that was innovative and great at MIT.” Together with a group of PhD students that included John Hollerbach, Tomás Lozano-Pérez, Matt Mason, and Eric Grimson, Michael decided he wanted to breathe new life into the robotics laboratory which had been started by Berthold Horn. The group put together a reader, which was Michael’s first book for MIT Press, started a journal and organised the first International Robotics Research Symposium to raise money. This resulted in a grant of around five million dollars from the System Development Foundation. Michael says: “So now we have a journal, a conference, a reader, and we started doing loads and loads of research. Tomás and Matt, John Hollerbach, Eric, and then other students like Demetri Terzopoulos, started writing superb theses, and then it really kicked off. … I basically had my own research group; I attracted a bunch of graduate students to come and work with me.” The group worked on a number of different projects including vision, image-guided control of robotics, tactile work, design of hands, design of touch sensing, path planning and collision avoidance. Michael adds: “Matt Mason was working on what became one of the most interesting ideas developed in the theory of mechanical assemblies. I also attracted John Canny, an Australian who did his MSc with me, and did the Canny edge detector. … Scientifically it was just a great time.” MIT
In 1983, during a visit to the MIT robobtics laboratory by the Department of Trade and Industry, Michael was asked what it would take to bring him home to the UK. Having laid out his terms, nine months later Michael was called by the University of Oxford asking him to visit for a chair; a position which he took up in 1985 despite never having applied for it. Michael outlined the three things he wanted in his new position; he says: “Firstly, I had been working in applications of computing, and I felt very comfortable with that, so I wanted to be a member of an engineering department rather than a computer science department. … I wanted my chair to be in engineering science. I wanted to declare that we were going to do practical stuff. I had this idea I wanted to do robotics, and in particular, I wanted to do mobile robotics, because we hadn’t had room at MIT, but it seemed to me that image-guided servoing would be fantastic. It would bring all these problems of robotics together. Secondly, I wanted to build a research lab, and surprise, surprise, after my experience at MIT, I wanted to build my lab around PhD students. The third thing I wanted to make sure that the faculty that we hired collaborated and so I insisted that in the letter of appointment to new faculty members, that the sentence be inserted at reappointment, which is the equivalent of tenure, credit will be given for having collaborated with colleagues. If you set the criteria for success, that determines the behaviour you get, so by saying I wanted and that we would value collaborations, all the young faculty that we hired all collaborated with each other.” Michael set out to hire the smartest young people he could find including Andrew Blake, Andrew Zisserman, Hugh Durrant-Whyte, Alison Noble, all of whom are now Fellows of the Royal Society and Lionel Tarassenko who is a Fellow of the Royal Academy of Engineering. The team set up the lab, seminars, reading groups and encouraged interdisciplinary action. In order for the lab to get EPSRC funding in robotics, they needed to have a company that would collaborate with them and so Michael began working with GEC Electrical Products. Michael goes on to explain some of his team’s developments: “The thing we did most of all in the Robotics Lab in those days was mobile robots. I was working together with GEC Electrical Products who had built electro-mechanical devices with a rotating infrared scanner that knew where it was using retro reflecting targets around the environment. It just didn’t know where anything else was and so we built ultrasound, infrared range sensors, vision sensors, and a kind of distributed computer architecture based around transputers, to do a whole series of hierarchical control from, making sense of the environment. In MIT, Rod Brooks was developing his subsumption architectures and our work was kind of a closely related alternative, with models of the environment, understanding various strategies for navigating around obstacles, either by re-planning a path or by slaloming around an object. The stuff that we did, both the device that was built, which was a transputer running software for doing signal processing, which we called a locally intelligent control agent (LICA), we licensed to what was going to be GEC Electrical Products, but effectively became Guidance (Michael’s first start-up).” After the success of the first project and with a potential follow up grant, Michael approached GEC Electrical Products to see if they were interested in continuing the relationship. Unfortunately, the department had shut down and so he proposed to Malcolm Roberts, John Potter, and Russ Miles, with whom he worked at GEC, that they should start a new company to continue making robots. In 1991, Guidance was born. Michael explains: “Between Christmas and New Year we met up in my office in Oxford, and we planned what we would do to start this company; Guidance. To be honest, it was a bit of a lifestyle company during the 1990s, we made stuff, we sold the product that we did to FMC Corp in the United States, but the company didn’t really grow that much, there were only about ten, fifteen people by 2000.” After a scare that FMC Corp might not continue to buy at the same rate, the team realised that they needed to diversity and started a marine division for dealing with putting sensing technology on supply vessels going to offshore oil rigs and gas platforms. They were also approached by Securicor, (G4S), to develop tags for offenders. By 2010 the company had grown to 120 employees across its three divisions: automation, marine and monitoring. They sold the monitoring division to G4S, the automation division was sold in 2016 to Matthews, and the marine division was sold to Wartsila in 2017. Michael says: “It was my first start-up and, and it was a pretty interesting experience. It really got me in the mindset about what to do and what not to do by the time it came around to the time of doing my second one, which was in 1999.” Robotics Laboratory, University of Oxford
By 1994, Michael had been head of the department for 5 years and as well as trying to continue to be actively involved in research, he also found himself managing a department with a 1000 people, including 75 academic staff and sitting on a multitude of university committees. He says: “During that time I had time to reflect, and one of the things that struck me was that by 1990 I had been in Oxford for five years. We had built the mobile robots, done a lot of work in computer vision, I had done a lot of work funded by industry, by the Defence department, by research grants from Europe and EPSRC. I reflected on the fact that my wife’s mother had died of breast cancer, and her tumour was probably two centimetres across. The more I thought about it, the more I learnt about breast cancer, the more I was just appalled that a tumour had been able to that size without being detected.” It was at this time that the screening programme started up in the UK. He continues: “I was curious as to why they had missed this tumour so I went up to talk to the people in radiology up at the John Radcliffe Hospital, I went to Addenbrooke’s, I went to St George’s, and I went to the Institute of Cancer Research, just to try and find out what was known about early detection of cancer. I was appalled by the lack of decision support for radiologists, and so by 1991 I had determined that my future was going to be in breast cancer.” At the same time, Michael met a PhD student named Ralph Highnam, with whom he would go on to start Mirada Solutions and then Volpara Health Technologies. Michael says: “So Ralph walked into my office, and I said, ‘Listen, I’ve got this thing. I have a dream. My dream is that we could build a physics model of the formation of a mammogram, and we could use it to make a quantitative model of the breast. But,’ I said, ‘I don’t know whether we can do it. It’s pretty high risk.’ He just said, ‘It’s exactly the kind of thing I want.’ So, the two of us worked for three years and amazingly we came up with a model that from a mammogram gives a representation of the dense tissue in the breast, measured in centimetres. It was the first time anybody had ever made a quantitative model.” Unfortunately, Michael’s first application to EPSRC was rejected, though EPSRC senior staff encouraged a resubmission which they did and got the work funded second time around. The approach was probably too different, innovative than the norm at that time. So the pair focused on writing a book and then started Mirada Solutions in 2000. Michael explains: “Mirada Solutions had two pieces of technology, one was our mammography stuff, and the other was image registration.” However, the new technology which was in demand at the time was a hybrid imager called a PET/CT which needed software to make the fusion work. Reluctantly, they decided to put their mammography technology on the back burner and concentrate on supplying the needed software. Michael adds: “That eventually became Volpara but in the meantime, Mirada Solutions went with PET/CT image fusion and MR-CT, PET-MR, and so forth.” They sold Mirada in 2003 to CTI. In 2005, Siemens Molecular Imaging acquired the whole of CTI and they invested into the bits of Mirada’s portfolio that they were interested in but left the rest. At this point, Michael together with Hugh Bettesworth, Mark Evans, and Chris Behrenbruch, did a management buyout of the third-party business including the source code, the customer list and so forth, which they then named Mirada Medical. Today, Mirada Medical has 70 employees and it uses AI and, particularly machine learning, to do radiation therapy planning avoiding organs at risk. Mirada Solutions and Mirada Medical
In 2008/9, Ralph Highnam and Michael started Volpara Health Technologies having realised at a breast cancer conference in Tucson, Arizona, that while the world had not been ready for their invention in 2000, it was mostly because it was not truly understood. Michael says: “The more we started looking at breast cancer, the breast density, it became clear that the amount of dense tissue in a woman’s breasts four/five years after the menopause is probably the best single risk factor for whether that woman’s going to develop breast cancer, and we had the technology to quantify breast density. So, in Tucson we sat down and sketched out what we wanted to do to form a company.” Michael continues: “As well as focusing on breast density, the company has progressed to do a lot of AI for breast positioning, understanding image quality, automatic measuring of image quality, automatic measurements of how well positioned the breast is, whether or not you’re going to get the right information that you need for detecting both breast density and breast cancer. For figuring out, again using AI, what the performance of a technologist (a radiographer and radiologist in the UK), and whether we could do automatic assessment of the quality of a mammography machine and separate out these various things. That has become our enterprise product. We’re in 2,100 sites in the United States alone, and we’re in about 38 countries in software. So that’s been a pretty good story.” Volpara Health Technologies
In 2014, Michael helped Professor Nico Karssemejjer, from Radboud University in Nijmegen in the Netherlands to start ScreenPoint. Michael explains: “I met Nico at a conference in 1998 and we just got on like a house on fire. When we (Ralph Highnam and Michael) started Volpara in 2008/9, Nico was one of the co-founders.” Following a further meeting in 2014, Nico explained that he had been thinking about computer aided detection of breast cancer and had an idea of how to do it better: Nico invited Michael to help him start ScreenPoint, based in Nijmegen, which now has 32 employees and received major investment from Siemens around eighteen months ago. ScreenPoint
Having retired from Oxford, Michael was invited to be part of a new company called Perspectum Diagnostics in 2012. He says: “I figured I could take on one more company. I didn’t know which, but I was kind of just chewing on it when a Stefan Neubauer, who is Professor of Cardiology in Oxford, called me.” Stefan together with Professor Matt Robson introduced Michael to clinician Rajarshi Banerjee (Banjo), who explained about his PhD work applying quantitative MRI to the liver to assess the development of fatty liver disease, a consequence of obesity. Michael was impressed with the technology but advised them to licence the software, however, they wanted to form a company and convinced Michael to become a part of it. Michael says: “So we did it, Banjo is the CEO; I’ve been the Chairman; and Matt and Stefan, the co-founders, Matt’s now the CTO, Chief Technology Officer. We’ve now got this company that’s rooted in MRI physics, image analysis, software engineering, and what’s really nice is that we have 200 people working for the company. We’ve got branches in San Francisco, Dallas, Singapore, main office here in Oxford. Of those 200 people, 55 per cent are women, and nearly, 98 per cent of our employees are graduates. 30 per cent of our employees have got PhDs, of whom 60 per cent are women. We have people from 29 countries with 25 first languages: it’s a fantastic place to be. I really love working with and for Perspectum.” Perspectum Diagnostics
With Michael’s reputation as a start-up expert, he was invited to get involved with Optellum which he co-founded with one of his graduate students; Vaclav Potesil, from the Czech Republic. It focuses on the idea of applying AI to lung cancer, in particular for finding nodules which occur within the lungs. The company has around 20 employees. Optellum
In November 2018, Michael was asked by Ling Shao, a former PhD student, if he would sit on the advisory board of the Inception Institute of AI, in Abu Dhabi, where Ling was director of a research laboratory. Michael says: “I flew out to Abu Dhabi, and what I realised was that there were two groups. There was the Inception Institute of AI that was doing some really lovely stuff, mostly in speech recognition, natural language understanding, and image analysis, mostly based in machine learning, but not only, and there was a big data company called G42, that was run by Peng Xiao. There was a kind of gap between these two companies. There was no real collaboration.” Having given a talk about image analysis, medical analysis, both in the laboratory and its transfer to companies, Michael was then introduced to Dr Sultan, the Chief Executive of the Abu Dhabi National Oil Corporation, who explained the Mohamed bin Zayed University of AI and the national strategy for Abu Dhabi to transform their economy to reduce its reliance on oil. Michael adds: “They had identified renewable energy, financial services, healthcare, and material science and when they boil those down into, what are the enabling technology, they realised that one of the key ones was AI.” Having realised that they did not have the trained workforce that could fill the senior positions of these sectors that were based around AI, the Crown Prince invested his money into the Mohamed bin Zayed University of AI. Michael was asked to join the board of trustees and be interim President until they could recruit one. He adds: “I’ve been involved in recruiting some of the senior personnel, in putting together the processes for searching for students and so forth. Within the first week of the launch they had 1,900 applications for 50 studentships in the initial cohort.” The world’s first artificial intelligence university in Abu Dhabi
Michael talks about the things he’s learned through his career, saying: “One thing I learnt when I first started working in cancer is that the only thing that matters in research is the identification of a problem that you care about. It’s the problem that matters.” He goes on to say that departments in universities can get in the way. He explains: “Departments are a completely artificial administrative construct that can make your life difficult by erecting silos.” He says that the college system at Oxford helps to overcome this, however because it allows for cross-disciplinary work. He adds: “Every day, if you’re in a college, you end up talking to the fellow in history, or the fellow in English. You know what departments become irrelevant. People ask you what you are working on. I find that unbelievably liberating.” Michael goes onto say: “The most important thing in an academic’s life is that you create people. We don’t produce research, we produce people. That’s our main product. Even if we work with industry, we produce people. It doesn’t matter whether it’s undergraduates, masters or PhD students. Somebody asked me, what am I proud of in my academic career and the answer is: 115 PhDs. I’d like it to be 120 when I retire because I’ve got five at the moment. That’s what I care about and as I never tire of saying, just before his death, Sir Humphry Davy was asked, what was his greatest scientific discovery and he said, ‘Michael Faraday.’ It is so right. I could list you all my PhDs and give you a ten-minute talk on any single one of those theses. That’s who I care for.” Lessons learned
Michael is keen to differentiate between AI and machine learning. He says that AI “has become too much identified with machine learning”. He explains: “Of course, AI uses machine learning, but AI is not all machine learning. I think that some of the things that are going to really come to the fore are going to be causal reasoning. I see that happening more and more in medicine. If you want to understand the aetiology of diabetes. There are several different aetiologies that lead to diabetes and it depends on the interplay of the liver, the pancreas, the spleen, the kidneys and there are many ways in which they causally interact with each other. Frankly, correlations are ridiculous. As Judea Pearl says, the outbreaks of serious fires are highly correlated with the appearance of the fire service. So, presumably, if you want to reduce the risk of serious fires, you get rid of the fire service, because then you would break the correlation. Since the days of Ronald Fisher and right from the very foundations of statistics, it has embraced correlation, and correlation is not causation. Now, through Judea Pearl, and through AI, we have a representation of causal reasoning. I think that that’s going to have a resurgence.” He continues: “I think also, whilst machine learning will continue to be important, there will be a resurgence in planning and reasoning, reasoning in finance and medicine. I think Abu Dhabi have got it dead right. Finance, energy, energy deployment, energy budgets. I think, so much of what we will do will bring back explicit reasoning, will bring back image analysis, will bring back natural language into the core of AI, and we will understand how intelligence is situated in robots that interact and move in the real world. I think the whole idea of situated robotics is going to transform AI as well.” Challenges and opportunities for AI in the next ten years
Michael says: “I think the most important thing for anybody in any walk of life is that you only have one life and you should be fearless. You should figure out what it is you want to do in your life, and you should tackle a really hard problem that you care about. If you really want to have a life in science, and it’s a fantastic life, you should identify something that you care deeply about. … Just believe, chart your own life, chart your own course. That’s the most important thing. Be courageous. Control your own destiny.” Advice
Michael was knighted in 2004 for services to engineering. He is a Fellow of the Royal Society, the Royal Academy of Engineers, the Institution of Engineering and Technology, the Institute of Physics, the Academy of Medical Sciences, the American Association of Artificial Intelligence, and Membre Étranger de l’Académie des Sciences. In 2000, he was awarded the IET Faraday Medal and the IEEE Third Millennium Medal for the UK. In 2005, the Royal Institution awarded him the Henry Dale Prize for ‘outstanding work on a biological topic by means of an original multidisciplinary approach’, and in 2010 he was awarded the Whittle Medal of the Royal Academy of Engineering. Honours
Interview Data
Interviewed by: Elisabetta Mori on the 13th November 2019 at Perspectum Diagnosis, Oxford
Transcribed by: Susan Hutton
Abstracted by: Lynda Feeley