AI in digital pathology; increasing workflow efficiency and patient safety
Posted 11th February 2019 by Jane Williams
In the fifth of this six-part series, the experts considered the potential cross-border solutions and challenges for Digital Pathology, as well as the evidence for improved efficiency. In this final installment, they discuss the impact of AI on industry and medical practice.
If you weren’t able to make the panel discussion, you can watch the recording here.
Peter Hamilton Over the last year, we’ve seen a huge growth in the area of AI in pathology. In 2018, the UK government announced an initiative to actively fund, invest, and get behind pathology AI as an area for growth for the UK.
Neil, can you tell us about the Innovate UK programme and how that’s impacting UK industries and the partnerships that we’ve formed within Philips?
Neil Mesher The Innovate UK programme is a programme under the Industrial Strategy for Life Sciences. Pulling it together, we were looking for key areas and exemplars where we could work.
The team is looking for areas where we could demonstrate a benefit to the healthcare system, tackle some of the challenges around pathology and the healthcare system more generally, but also to create new business opportunities for UK PLC. There is a fantastic opportunity where you can improve things for the healthcare system, improve patient outcomes, create new business and job opportunities within the UK PLC, and attract new investments. That was what we were looking for within the industrial strategy programme, and digital pathology ticks all of those boxes.
There’s real commitment from the government in AI generally, but clearly in pathology and radiology. There are huge opportunities for us to create something which is world leading, and the centers that we’re building can be global exemplars. It’s that excitement in ticking off those multiple aims which makes this just a fantastic programme.
Peter Hamilton David, you are the leader of one of those programme centers? Tell us a little bit about PathLAKE and your expectations for it?
David Snead It’s an exciting time for us to be leading this big project into AI and digital pathology. PathLAKE has got three major objectives over the next three years. We want to develop exemplar projects and put them into practice; tools which really benefit pathologists. Automation may be some of the reporting that we do, and improved diagnostics in a number of cancer sites, particularly in prostate and breast cancer.
During the course of that, we aim to develop a data lake of high-quality whole slide image data, make that accessible both to research communities but also in SMEs within the UK. High-quality data, which entrepreneurs and scientists want to get to, providing new tools of benefit for both the economy and the health service.
The third part is to provide some know-how about making that happen. So we have an outward facing approach where we have expertise in pathology, digital pathology and computer analytics. Also on the commercial side, design and product planning; bringing products through from research lab to a research programme and then clinical trials programme, which makes them fit for use in clinical practice.
There are lots of important things to get right quite early on in the development pathway of these products, and we want to be able to offer that advice to those communities to help them get this right from the word go. We’ve been in conversation with regulators quite early on in the development of our PathLAKE plan, so they’re aware of what we want to do and have been able to offer advice and support on how it might go about it.
Neil Mesher Something else that has come out from this as well is the dialogue between academia, industry, the healthcare system, and the charities. It’s changed fundamentally on the back of this. We’ve all got a shared objective in terms of what we’re trying to achieve and a belief that we can’t do it on our own. But if we come together, the sum of our parts is far greater.
That excitement and that energy is partly because we’re working together in a way that I haven’t experienced previously. Even industrial partners working together is not necessarily something that naturally occurs.
Peter Hamilton Juan, you talk about AI being a key driver of digitization in pathology. How do you see AI being used routinely by pathologists in Europe?
Juan Retamero I don’t think we’re going to go out of work anytime soon! Something that’s frightened a lot of pathologists is comments like: “what’s the point of training new regulators when we’re going to be replaced?” Which is absolutely ridiculous! I mean you can buy a calculator for £5 and the calculator is going to get every calculation right 100 percent of the time, but despite that, accountants are still in full work. In other words, it’s a tool for them to deal with burdensome repetitive tasks, and the same is probably going to happen to professional pathology.
Peter Hamilton Jo, how are we preparing pathologists for this revolution that we all believe in?
Jo Martin It’s not just pathologists – at the college we’re going to have to change everything we do. We’ve started to digitize our exams and we will go fully digital at some stage. We’re looking at digital training. We’re looking at ways in which we present training material to people. We’ll have a digital learning center for excellence, which will allow us to train and validate.
We’ve already issued guidance. We’ve got strategies around digital, so a lot of work for us as well as our trainees. The trainees are increasingly learning digitally, not necessarily on the whole slide imaging, although there’s a lot of whole slide imaging out. We’ve issued our exam cases of whole slide imaging, putting them on the Leeds website and social media. There’s lots of learning materials out there where we’re offering a range of digital materials already.
Peter Hamilton David, what areas do you think are going to benefit the most from AI?
David Snead Personally, I think the one area which is going to benefit the greatest is the screening of tissue with no areas of interest in it. If you look at what pathologists do, a large part of their time is looking for things which aren’t there. It’s obviously important to establish what’s not there, but the pathologist doesn’t need to do that. The pathologist needs to recognize what is there and make sense of it. If we can shift the burden of attention for the pathologists to that area and leave the computers in the first area, that will address a lot of the concerns over capacity.
The CAMELYON Challenge is a fantastic example of just how good AI is at finding metastasis in lymph node: as good basically as a pathologist really. It will do it all day and all night. It’s not going to get tired. It’s not going to get interrupted and drop a slide and forget about it. It will look at everything.
Juan Retamero At the end of the day, this is going to translate into two things: more efficiency and greater patient safety.
Jo Martin At the congress, I saw a lovely piece of AI looking at H. pylori from a group. Looking at H. pylori is painful and wearing. That sort of tool could be very beneficial. Perhaps, we won’t have to do the additional CF feeds. We can save the lab time and money doing that. It’s not just cancer; it’s all the other stuff we do too.
[Audience question] If in the future – let’s say 20-50 years – when AI is powerful enough to do work that pathologists can do, do you think the standard to become a consultant in pathology will be much lower?
Jo Martin I think then as now, we’ll be left with all the difficult stuff. It may be the other way around because we have to recognize when the AI is actually not picking stuff up, and then you’ll be left with that for which there is no algorithm. For example, I’ve got three cases never described before. It’s not in the books and I don’t have an algorithm associated with it. It’s down to me. There will be routine stuff that will help, but if anything, we’ll become more skilled.
Neil Mesher There’s other data sets to bring into the equation as well, which will also potentially complicate things, so you need that human interaction. Where are we with AI in radiology? Where does genetics come from? Where does family history come into it? There’s a whole bunch of different factors you can start to bring in once you’ve got that data set that you can do something with and far more prescriptive, and the promise of precision medicine starts to come in as well. I think that the role changes, it doesn’t mean that it’s gone.
Peter Hamilton If you’re talking to a new pathologist coming through, a trainee, or resident, or even a medical student that’s considering pathology as a career, how would you describe your hope for pathology 10 years from now?
Jo Martin My hope for pathology is that the already interesting job that we do will be slightly more fun with less rummaging around in the file trace: more fun, high tech, faster, and more flexible.
David Snead My hope is that pathologists will remain the nerve center of the hospital providing the key bits of information that are needed to treat the patient, make the diagnosis, and provide the best treatment for the patient. I think that will inevitably involve the use of AI interpreting digital signatures on areas of slides.
Neil Mesher There’s a huge opportunity for the pathology function to actually make an even bigger contribution. The efficiency gains, the effectiveness gains will come. Technology will develop, and it will accelerate in terms of what it can do. It’s a tremendously exciting time for the profession.
Juan Retamero We have to realize that it’s the beginning of something that is going to change the profession forever. It’s actually a profession that hasn’t really changed that much for 150 years, and this is really exciting, so welcome onboard and enjoy the ride!
Take a look at the agenda for the upcoming Digital Pathology & AI Congress: USA. Alongside expert keynote speakers, the conference will comprise of a vibrant exhibition room full of technology providers showcasing their technologies and other solutions.
Leave a Reply