Collaborating with DeepMind and How to Get into MedTech

Really, my entry into the medtech world was through getting in contact with Mustafa — also a Mustafa — Mustafa Suleyman from DeepMind.

Maybe I'll just briefly tell you about my background, because it sets the scene for how I actually came to find myself in a room with him. From an early stage in my medical career I was interested in research, but also in new technologies, and for that reason ophthalmology seemed kind of cool. I was reading about bionic eyes, retinal prostheses, things like that, and I loved the idea of using new technology to restore sight — interfaces between electronics and humans, cyborgs, robots, all of this good stuff.

So one of the first things I did as a medical student, around 1999, was read about the bionic eye. I went to an early website for the retinal prosthesis project of a professor at Johns Hopkins in Baltimore — a guy called Mark Humayun.

On the website he mentioned he was open to medical students coming and spending time with him. So I sent him an email, and to my amazing excitement, within a couple of hours he replied: "Come on over, spend the summer with me." And so in 2000 I found myself spending the summer at Johns Hopkins and Wilmer, meeting all these people with this tremendous can-do attitude towards new technologies. One of the lessons I learned was that if you contact the right person, phrase things the right way, and the timing is good, you can open doors for yourself.

Then I went back to Ireland, graduated in medicine, and began my basic training in ophthalmology — three years there. I decided I wanted to continue as a clinical academic and go to the US. This was the old Irish training scheme, like the old UK one: basic training, then higher surgical training, and it was really hard to get into the higher surgical training, so people commonly took time out to do a PhD.

I had the idea that I wanted to look at retinal imaging — there was a new technology called OCT, optical coherence tomography. So I looked at this magazine website, US News and World Report, which every year ranks the top ten hospitals in the US by specialty. I found the top ten in ophthalmology, looked up the faculty on their websites, and found people in a few places who were leaders in OCT imaging, looked like nice people, and whose research seemed cool. I emailed six of them — Massachusetts Eye and Ear, Bascom Palmer, UCLA, various other places. And the nice thing was that three out of six invited me to come and work with them. So lo and behold, a couple of years later I found myself in Los Angeles at the Doheny Eye Institute for two years doing research. That's when I learned how to write papers.

I'm giving this story in a roundabout, extended way, but I think there may be something of interest here for people into new technologies and academia. What happened next was that I was thinking about staying in the States and applying for residency, but I had a chance meeting with an Irish doctor from Cork — a guy called Tom Flynn — at an international conference in Florida. He told me about this new thing in the UK called the NIHR, the National Institute for Health Research, and its new academic training pathways. There was something called an NIHR clinical lecturer, and Moorfields — the most famous eye hospital in the world — was going to be advertising one. It would let me do 50% clinic, 50% research and finish my training to consultant level. My life changed with that, because I thought: that is what I want. So through some swings and roundabouts, and a detour in Birmingham where I met Alastair Denniston, I ended up as an NIHR clinical lecturer at Moorfields in 2010.

I was privileged to finish that and then become an NIHR clinician scientist, where I actually had about 70% of my time protected for research — I was doing clinics one and a half days a week. That ran from 2015 until the end of February 2020. The pitch for it was that the research project would be about advanced retinal imaging, but more broadly it was: could I identify cool stuff in the tech world and apply it in healthcare, specifically ophthalmology? Having 70% of my time protected meant I could explore. I was hearing all about the advances in deep learning — the great AI awakening, as the New York Times called it, back in 2012, 2013 — and thinking, how do I do deep learning in ophthalmology? Who do I work with? I approached some people in university settings and, to be honest, was greeted with shoulder shrugs.

Then I thought, okay, who in industry? I knew this was a relatively limited pool of people with the expertise and the resources at that point. So I was looking for someone at a Google or a Microsoft or an Apple or an Amazon — someone not too junior, but not too senior. If you email Bill Gates, you're probably not going to get a reply.

What happened was that I was reading Wired magazine in June or July of 2015, and there was this brilliant profile of DeepMind. A lot of it focused on Demis Hassabis, but there was also a section with Mustafa Suleyman. He was the head of DeepMind Applied, and he was talking about his mother being a nurse on the NHS, his passion to use AI for good — in healthcare, climate change, energy, all this good stuff. That's when the light bulb went off: this is the person I need to contact.

So I found his profile on LinkedIn and took out a one-month premium subscription, just so I could send him a message without being connected to him. This is all true, by the way — I have the messages, I have the screenshots recorded for posterity. I sent him a message saying: I'm a consultant at Moorfields.

We're doing a thousand OCT scans a day. There are people losing sight and going blind — not just at Moorfields, not just in the UK, but around the world — because they can't be seen and treated quickly enough. We should apply deep learning to these OCT scans so we can identify the people with sight-threatening disease and get them in front of someone like me, a retina specialist, as quickly as possible. And by the way, we're two stops away from you on the tube.

To my excitement — third time lucky in my career — it was, "Okay, come and have a coffee." A few days later I found myself at the Google headquarters in Pancras Square in London, talking to Mustafa Suleyman. And five years later, I've been working with DeepMind and then Google Health ever since.

Ultimately, if you want impact at scale, you do need to collaborate with industry. There are pros and cons, of course. But my dream is to have something that begins with an idea — by me or by someone who works with me — that we make into an algorithm and ideally publish in a top-level journal. And then not end there: okay, we've published a Nature paper, but how do we turn this into something that can be used by a million people?

It turns out a lot of clinicians and academics are very naive about how hard that latter part is. Some people have this misconception that the hard part is publishing the paper. Well, let me tell you: the hard part is implementation. The hard part is execution and making things happen. So working with industry is a chance for me to, hopefully, have patient benefit at scale.

I'd add something to that. There's a book I just reread called The Innovators by Walter Isaacson — he's the guy who wrote the Steve Jobs biography. The Innovators is essentially a history of all the people who drove the digital revolution from the 1940s to the present: the first personal computer, the microchip, the microprocessor, the web, the internet, all of it. What struck me was that a lot of the digital revolution of the last fifty years was driven by foundations laid during World War Two, when the first computers were being created — and that involved very close collaboration between government, industry and academia. A kind of virtuous triangle where they worked together closely, and that made amazing things happen.

As I was rereading it on lockdown in my home in North London, I was thinking: wow, what are the parallels with the pandemic? With the current restrictions, are there opportunities to finally pull the trigger on transforming healthcare with new technologies, driven by the pandemic?

That's a great question. I often think about it, because I sometimes have imposter syndrome — that I'm the person in the room who knows least about what I'm talking about. And I sometimes feel negative about the long medical apprenticeship. I finished medical school in 2002 but only became a consultant in 2015. Granted, that was an academic pathway, so it was a bit longer. But I do think: imagine if I could have taken one or two of those years and done a master's in data science, or something actually useful, instead of being in the middle of nowhere learning some operation I was never going to do again.

On the more positive side, though — I like to think it's a good thing to have one major area of expertise, one minor area, and a working knowledge of multiple others. My major area is ophthalmology, particularly retinal disease. Put me in a room to talk about macular degeneration and I'd feel confident I know as much or more than most people. That's my foundation. Related to that, I'm a clinical academic, so I know how to publish papers and get grants — and those are transferable skills, whether to the tech world or to setting up your own company. My particular research interest has been OCT and retinal imaging, so I know a decent amount there. But along the way I've had to pick up a working knowledge of statistics — even if you work with a statistician, you need to do a lot of it yourself; you can't be completely ignorant of it as a clinical academic. Then maybe there's a health-economic component to a grant, and I have to get a working knowledge of that too.

And I need to know a bit about machine learning — but I know I'm never, ever going to compete with a Jeffrey De Fauw or an Olaf Ronneberger from DeepMind, people who've dedicated their whole lives to it. So the question is: what's the amount of knowledge I need to have meaningful conversations with people like that, and a conceptual framework I can apply?

Then it turns out there are so many other things. Project management. Leadership. Managing finances. Not to mention the entrepreneurial side, which is largely something I'm oblivious to. So that's a bit of an extended monologue, but the answer is: lots of different areas.

Yes, I totally agree — and now I hope there aren't any machine learning people listening throwing things at their MP3 players. I'm hugely biased, but I just think medics are the best. There's a tendency, at least for me, to sometimes feel out of my depth talking to an experienced computer scientist and think, wow, these guys are on another level.

But actually, getting into medicine and progressing through it is bloody hard work. You have to be very smart, but also extremely diligent. To survive as a doctor you need problem-solving skills that aren't really taught — it's like being able to survive in a war zone. So if you can do that, and write papers at the same time, and apply for grants at the same time, you're a pretty useful person to have around. Even if you don't know something, you can probably learn enough about it to say or do interesting things pretty quickly. If I were a CEO of a company, medics would be a top-priority hire for me.

Yes — a doctor with some knowledge of coding is probably more useful than a world-leading doctor with zero coding plus a world-leading computer scientist with zero medicine, in terms of translation, ideas and innovation. On that note, after my NIHR clinician scientist award finished, I recently got funding from UK Research and Innovation as a Future Leaders Fellow. It's an amazing scheme because it's specifically about research and innovation at the boundary between disciplines, and between industry and academia — because that's where all the cool stuff happens.

At the highest level: just look for cool stuff, stuff that interests you. If you read about something exciting, it's probably being done by someone not that different from you, and in many cases those people are approachable. My own experience, even as a medical student in Dublin back in the 90s, was that you can find opportunities to meet cool people and be involved with cool projects.

Related to that — a bit of advice I always give — is that life is too short to work with assholes, basically, if I can say that. You may find yourself contemplating working with somebody who has a reputation for Nature papers or Lancet papers, but also a reputation for not being very nice, or being a bully. I'd say just don't do it. Work with people who are cool; you'll enjoy yourself, you'll be more productive, and that's the way to pursue things.

At a less philosophical level: pursuing research is a very good approach, because it keeps all your options open. Getting involved in research projects during your summer electives, trying to publish some papers and case studies — first of all, they tick a lot of boxes on your CV, so even if you decide research isn't for you, they help you get the job you want after medical school, or the consultant post you want. But beyond that, they give you skills either way. If you want an academic pathway — which is what I'm on, a dream job for me — you can go down it. And if you want to move into the tech world, you've got some data science and statistics, you know how to write papers, how to be organised — so you can step into that world more easily.

When I started working with DeepMind and DeepMind Health, there were these amazing young doctors, younger than me — people like Joe Ledsam and Cían Hughes — who just took the step into that world. For the most part their backgrounds were as clinical academics, and they fit very nicely. And if you don't want to work with a big tech company and want to go down the startup road instead, these are the skill sets that'll be really useful.

Let me think, because I've got all these books in my office that I've underlined and read through a few times. One I reread recently is Originals by Adam Grant, about how nonconformists change the world. I've got it right beside me here, and I've underlined almost every page. One thing that struck me is that in the first chapter it says there are two routes to success: conformity and originality. It talks about how, if you look at child prodigies, they don't necessarily prove to be great originals — because to become a chess player at a very high level, or a violinist, you have to be so focused, and it's not necessarily focused on being original.

There's a risk in medicine that it's so hard to get into, you have to work so hard, that you can't go too off-piste on the pathway. You get into medical school, you just do your exams, you finish — and then, oh my God, a load more exams. It's super competitive to get the specialty and the hospital you want, and by the time you become a consultant you're just tired and want to do private practice and live a comfortable life, because you're 40 and can't take it anymore. So to be original you have to be a little bit different from that. But that doesn't mean doing crazy, risky stuff. There's a misconception that big tech innovators go all in on huge bets and take enormous risks — but Larry Page and Sergey Brin didn't give up their PhDs at Stanford until two years after they formed Google.

Innovation often comes not from 21-year-olds but from people who are older, and they manage their risk portfolios — they have backup plans. That really influenced me.

The other book I'd highlight is The Patient Will See You Now by Eric Topol. Eric Topol is my hero. I'm fortunate to have got to meet him, interact with him, get advice from him — to me it's like being a musician and getting to talk to John Lennon or Paul McCartney, or jam with the Beatles or the Rolling Stones. The Patient Will See You Now is about the digitisation and democratisation of healthcare. It's full of cool stuff, but also really grounded in patient benefit — you can tell he's a working doctor who would do anything for his patients. That would probably be my desert-island book, if I were only allowed to choose one.