Cerebriu CEO Robert Lauritzen - Revolutionizing medical imaging with machine learning

Robert:
I started as an independent management consultant just to have some cash to start an endeavour, and then was looking for what to do, really. And I spent the first year just building sort of the basics of the business and started saving from what I would make there. And then I met with one of my co-founders of Cerebriu through a former colleague, who incidentally was his wife. And he is a professor at the University of Copenhagen and looking to somehow build a business based on machine learning and medical imaging to help clinical practice and radiology.

Intro:
My name is Philipp Stürmer, and in the first part of this episode of Deep Tech Stories, you will hear the story of how Robert Lauritzen and his team at Cerebriu are using AI to reach superhuman performance in medical imaging, transforming the way doctors see the human body. Have you ever lied inside an MRI machine, listened to loud buzzing and thumping sounds, wondering what was going on outside? It's a strange experience, lying there, helpless, relying on the machine to give doctors a glimpse inside the body. But despite all of the tech, in the end, the doctors still need to interpret that glimpse by themselves, often with the help of a radiologist, which takes time and resources. So it seems like it's the perfect task to automate and reach superhuman performance using machine learning, right?

That's exactly what Robert Lauritzen and his co-founders at Cerebriu thought as well. Robert himself has a long entrepreneurial history, which prepared him perfectly for revolutionizing medical imaging techniques, starting all the way back during high school.

Robert:
So I had my own company during high school doing, oh my god, that's a long time ago. I helped out with various computer-related problems or support, and also did work in the printing industry when that was a thing. Editing on computers, sort of visuals for commercials and stuff. I'm making it all compliant. And then it was ad hoc, technical work, whatever I could get my hands on. Then you picked it up during school somehow? Yeah, exactly. Yeah, just, yeah. I had something, created my own job. I mean, I was also employed in various fashions. I also was employed doing, in the day, Apple Macintosh installations for businesses, and stuff like that. It was the good old days of floppy disks, and floppy disk installs. So I'm that old. Yeah. So effectively, I've always been working with some sort of technology. But it started actually all the way back in high school.

Philipp:
As I said on your LinkedIn profile, it was for dental clinics.

Robert:
Yeah. But that was actually slightly after. So during my study at the university, I was also working in a sort of startup, a technology startup, where we made battery chargers, where I was developing software for battery charging, and controlling systems. And then I went independent, which transitioned into becoming a partner in another IT service company and software development company. Then I started working with one of my old buddies from school that was on the hardware side, electronics, to develop a solution for mobile dentistry.

Philipp:
For mobile dentistry?

Robert:
Mobile dentistry. So you could have a dental clinic in a backpack. Which is a combination of software, electronics, and sort of good old-fashioned hardware, because you need high-power drills to perform dental surgery, whatever and need to put it into a modular device, so you can have multiple, support multiple instruments. I mean, there are dozens of instruments in dentistry with different features, and different capabilities, and need to be able to control those. And those were the days before all the nice, flashy, smart devices and touchscreens, what have you. So it was building effectively on my side, I was heading R&D, which actually transitioned me out of my studies at the university and into just working with this. So it was so much fun. And developing sort of a medical device for the dental industry. And I was running, and I was sort of also heading the software development. So basically, back in the day, based on sort of the recent introduction of the Linux operating system, the Unix variant, built a sort of little micro-operating system for this device and developed all the controlling functionality to manage dental equipment.

Philipp:
How did you learn all that? 

Robert:
How did I learn all that?

Philipp:
Writing operating systems?

Robert:
Well, I guess that was learning by doing. I mean, obviously, a lot of self-study, I've always been very much digging into and learning new stuff. So my technical interest started in my very youth, I was a VIC-20 user. VIC-20 was before Commodore 64. So it was a 5kmemory unit and did sort of basic programming back in the day. Got into Amiga, got my first PC sort of around high school and stuff like that. So I've always been very interested in software and technology, and quite self-learned. I was also during university teaching software development, in some of the courses. The sort of real startup, which is my independence, was this dental equipment.

Philipp:
Yeah. But you did computer science during university? 

Robert:
Yes, computer science. That was just the real start of my entrepreneurial desire to create a difference and falling in love with the life science sort of industry, per se, because of the potential impact on healthcare. 

Philipp:
Did you choose the dental in the beginning just because it came along? 

Robert:
Well, it was actually my friend. So he was working with this, which incidentally, the same sort of founder and inventor as also with the battery chargers. So I had in that early day, which was high school time, a connection with them. And we became sort of junior partners in this endeavour to create a solution based on several patents within sort of battery and charging, whatever modality and modularity of this kind of technology, which made it quite interesting to try to build a solution, a sort of a full-featured solution. And it was in the 90s, pre-dot-com. So there was a lot of frenzy about the opportunities of going viral on online sales and commercialization and stuff. But we did make a lot of, I think we built 500 working units with some of our partners, which were making the hardware and the electronics, et cetera. So then to some extent, there was some level of commercialization, but it never really took off. And I left before that. I wanted to move into, and then I started my career sort of in pharma, leaving that. And that was sort of another independent contractor job or a jump to get into pharma. Consulting pharma on developing solutions and that then moved into data
warehousing and stuff like that. Again, self-learned.

Philipp:
How did that work with pharma? Because as far as I know, they somewhat expect you to have a PhD.

Robert:
Yeah, obviously. So I was not doing, I was not making medicine type of thing. So it was within the IT systems or business applications, data warehouses, and stuff like that. So it's a lot about business processes and business process automation and support solutions and combining information across to get an overview of what's going on and stuff like that. So a data warehouse was about connecting sales and marketing with manufacturing and planning. So a lot of forecasting and planning solutions and stuff like that, which is about the backbone of any larger industry to make sure that you have an idea of how much you have to make at a given point in time. So that's not necessarily required to have a PhD, but a good mind is of course necessary. But that was sort of where I started my corporate life back in the day. And then I moved from Bering to the Nomen OrdiscusIT company, because I was on a joint project management training or sort of course, a couple of years where we did some extensive, I guess, this GMI certification, the certified project manager around the millennium, running quite substantial amounts of projects for years and could see that some of the projects there were quite exciting. And the opportunity of having, you know, many colleagues running larger projects and having sort of a lot of fragmented projects in a smaller company like Bering. I mean, it's not that it's a small company, but it's like five people for all the or eight people for all the projects. So it's, everyone is putting in a few hours a week on all the different projects. It's not very exciting. Not very efficient either, but that's, you know, you have a typical growth issue with trying to build a team that can manage all the different expectations and demands from the different business areas, which was also one of the reasons why I moved on because I thought, and I could see that the potential of working with a large corporation would be exciting and ran some substantial projects in that context, including sort of implementation of environmental monitoring systems across all manufacturing on a global scale, which of course is a huge endeavour and under the sort of most strict compliance
regulations from FDA, et cetera.

Philipp:
Did you notice any management changes in start-ups?

Robert:
Certainly, there were a lot of changes and I also developed, I mean, I had had also previous management responsibilities. I moved in specialized a little more in project management and then transitioned into what we call service management to really transform the company as part of the, the independence of daughter, a company or the subsidiary into a standalone sort of legal entity, which was the overall strategy. And part of that was actually building and transforming and developing the business into being a professional service delivery company. So, that was an opportunity that I got in that context, which of course was really exciting because it was, it was, it was adding onto the understanding of, of changing things from A to B, which is sort of what project management is about into, into transforming and operationalizing sort of all, all value adds and, and providing a transparent and well-functioning operational, excellent service organization and capability in a company, which was really, really exciting to build. And I helped build that as part of the journey before I moved into sort of senior management in the company with, you know, assuming the classic budget and people the responsibility of substantial parts of the teams.

Philipp:
And again, you picked all the necessary people and hard skills up on the way by learning and learning by doing. 

Robert:
Obviously, a lot of, a lot of development on the line and training and working with mentors and whatever to, make sure, to manage the development. But yes, it was not a school. It was a learning on the job. And I guess the last five years I had a few hundred employees globally under my responsibility and, and, and sort of refining and transforming also some of how we delivered across, you know, moving around from India to China to Eastern Europe, et cetera. But also came to a point some years back with a longing to get back to entrepreneurial work. I did start some entrepreneurial work in the company besides the transformations I've been through and driving the company and which was oriented more towards building solutions, building, you know, new ways of doing things that could be productized and sold on a different scale than sort of a fee for service approach. And ultimately decided, to leave, to start something else up together with someone else and then met through that.

Philipp:
What year was that? 

Robert:
This was back in, so I left in 16.

Philipp:
So you were in that company for nearly 13 years. All the way through the financial crisis.

Robert:
It was a long, it was a long journey. I got three kids and, and, you know, had a slipped disc and an operation and a lot of other stuff along the way. But, I, and I think originally I had not intended to stay so long, but I also had several different jobs in the company, right. And moved around and developed quite a lot. So I really had an exciting journey, but also could feel that if I was to go out and build something myself, I needed to start pulling the plug sort of in the early forties, I started having those thoughts and left there in 16. So I started and I really left without anything. I mean, I had no idea. I had no, I just needed more time to reflect. So I started as an independent management consultant just to, have some cash to start an endeavour and then was looking for what to do really. And I spent the first year just building sort of the basics of the business and started saving, from what I would make there. And then I met with one of my co-founders of Cerebriu through a former colleague, incidentally his wife. And he is a professor at the University of Copenhagen and looking to somehow build a business based on machine learning and medical imaging to help clinical practice and radiology, specifically. 

Philipp:
Why, why was he looking for that? 

Robert:
Well, he was looking for that because he had spent his entire life on building technology and designing and evolving science in a direction that was, that could provide utility and medical imaging, but had had an endeavour for many years within clinical research, but could see that there was a burning platform in clinical practice.

Philipp:
And it's reasonably rare that a technical professor wants to start a company.

Robert:
Yeah. Well, he had a company already, but that was sort of not really moving anywhere. And he wanted some help from someone with more sort of industrial expertise and experience and a broad set of skills, including also some commercial business management expertise that he, knew he could not fully himself, he and his partners, because two other new professors in computer science were part of the team. And then who is now our CTO, hisprotégé, PhD, and at a point in time postdoc. So a really strong technical team with decades of experience within machine learning, sort of on a global scale, leading science front end, front end level of science and the University of Copenhagen being sort of, I think, number four rated at that time within AI and medical imaging on a global scale. So quite a strong capability and all the expertise in the world within that, but with a need to balance that in a co-founding set-up with someone on the more business side of things. So at first I actually did think it was kind of interesting and it could be something I would pick up on, but he got back to me a few months later and then we got kind of excited and I met Akshay, my co-founder, which was great because I spent a day with him workshopping on some of the dilemmas and trying to figure out what we could do and that kind of capability and what the problem was. 

Philipp:
What were the dilemmas? 

Robert:
Well, I mean, so the overarching dilemma is the extremely growing demand within medical imaging, so radiology and diagnostic imaging, like MRI, CT, whatever, which is growing so fast that the volume of work is outgrowing capacity by nearly 10% per year and has been doing that for 20 years. So we all know what a work week looks like, just add three hours per week every year to your workload. The workload was driving burnout and error rates up, and it just seemed that there was a core problem that most likely could be helped and benefited by technology. And it did not seem, although there were quite several companies working in this space, it looked like in the landscape that the competitive landscape was focusing very much on, because of coming from research, focusing very much on biomarker quantification and stuff like that, because that's sort of the research dilemma being focused most on, I think, in science.

Philipp:
And what do you mean by biomarker quantification?

Robert:
So it's like a volume of a segment in the brain, a longitudinal assessment, which could be an indication of atrophy, stuff that is then useful to see whether or not a patient has a specific pathology or condition that can or cannot be treated. So the size of a tumour, the size of the count of something, something that you can somehow in imaging scientifically contest about, and there's a lot of focus on that, which is all about post-processing. And effectively, although we started in that space, really, we pivoted within a few weeks into what we're doing now, which is focusing intently on automating workflows.

Philipp:
So building a platform?

Robert:
So building a platform. So effectively, well, I guess a platform sounds kind of corny because there are so many that are building a platform. I think our approach is really different to the extent that we move upstream. Instead of focusing on post-processing and providing these kinds of biomarkers or some sort of quantification, volumetrics, we move upstream to the earliest point where data is created or accessible, and to derive and pronounce whether or not a certain condition would warrant a different workflow than had been conditioned by symptoms or whatever the input was to radiology. And this is more or less the background that I have with service and operational excellence because effectively, hospitals are production. So they're looking at how can they be operationally excellent in what they do and as efficient as possible, which is inspired by processes from lean frameworks and others from manufacturing, which is about removing waste and improving quality at every single step in a given process. And there, it always makes time. Just like trying to do quality checks and testing, after you've gone through an entire R&D lifecycle is really expensive. It's always good to bring decisions and front-load questions where possible, but it requires information. So the concept and the idea of what we do is really when information is available, which means when the first images are reconstructed by the scanner, is there any new information that would warrant that we do something different than based on the basic symptoms?

Philipp:
So you did differently in the sense of a different diagnostic method?

Robert:
Exactly. A different diagnostic method is the imaging procedure. So you might get, you might go to your GP with a headache. It has been severe. It could be chronic or asymmetric, whatever, but nobody really knows. It could be a gazillion things. And it could also not be a pathology. I mean, at least not something we can detect and classify at this point. But then your GP would most likely refer you to the hospital if it found that of concern, or you could have a patient history with whatever, that migraine in your family or something else that would, that could be something dangerous, or it could be an age where it's dangerous or whatever. So eventually you might get up, and end up in an MRI scanner. But today that kind of sort of routine MRI would want to rule out quite a lot of stuff, which require very specific imaging procedures, which quite often are now stacked and include contrast and all sorts of stuff, which will render you sort of spending at least 30 minutes or even more typically in the MRI scanner and being pumped with chemicals. And the idea of what we're doing is effective because the minority of cases you actually have pathology. You don't have to have something, but if it's there, it could be quite severe. It could be critical. It could be even acute, like a stroke. So what we're doing is looking at if we have enough data within the first few minutes that is more specific than just a headache with whatever symptom, then let's look at that data then and be more specific and personalize the imaging output based on what we see early on and alert the healthcare practitioners if we see something that could be dangerous, that would require their more acute attention and prioritization of what's going on in the scanner.

Philipp:
So do you make sure that you're not missing anything?

Robert:
So we, like any other technology or even humans, can miss stuff. So we're not automating the diagnosis. What we're doing is looking at, and there's not enough information to make a diagnosis. That's why you actually add the additional images, the imaging procedure. But what we're looking at is how can we go from around brain protocol or something that sort of like scans everything into something where we can detect, see, is it a vascular condition? Could it be a stroke, a haemorrhage, a tumour, or some other significant lesion that would warrant a different imaging procedure? So if the procedures could change and provide better information to reduce recalls and get it right the first time, this is of value. And if we can simplify the operation of that, because today this is a manual thing that advanced MRI technologists that are operating the procedure, they're performing the image acquisition, that they sometimes detect something, sometimes don't. And if they do, they call the radiologist and they do that like several times an hour, maybe 10 times an hour, they call the radiologist. So there's a lot of interruptions, a lot of waste going on in the system. If we can provide something similar quite fast, early on, we could reduce the interruptions, and reduce the error rate of mis-imaging or over-imaging. That's the sort of intent. Ultimately, when doing that across all significant pathology, the high accuracy, this is the sort of long-term direction we're going at, then we can automate the workflow and become less and less dependent on the skill set of the operator and less and less dependent on having the right advanced sub-specialized neuroradiologists available for all scans, but spend more time on those patients and faster with those patients that have critical findings. So the idea here is to improve the quality of the examination procedure, be more efficient, and use specialized competencies. And this is what operational excellence is all about. But we're using technology and enhancing the scanners with this kind of approach. So putting it in a cartoon fashion, we're making 50 years of advanced MRI equipment from thumb scanners to smart scanners that can help the clinical process and improve the quality of medical procedures. 

Philipp:
And then you sat down with your co-founders, the two professors, and the back-then postdoc went through all of this.

Robert:
Yeah, that's what we spent the last four years on. So we created the company in 2018. I found someone who wanted to invest. I put in whatever I had scraped from being a management consultant for a year. And we started the company with a six-month runway.

Outro:
Thank you so much for listening to Deep Tech Stories. If you enjoyed this episode, don't forget to subscribe to Deep Tech Stories wherever you listen to a podcast or follow me on Twitter. You'll be hearing back from me in two weeks when we finally dive into the building process of the review and the data nightmare we'll experience with GDPR and patient data.