There is an important aspect around trust and whether your partners, the surgeons, will come to trust the AI more than they trust you, or you more than the AI.
Interview with Professor Consolato Sergi,
Chief of Anatomical Pathology, Children’s Hospital of Eastern Ontario, University of Ottawa Professor of Pathology and Pediatrics, Emeritus Professor of the University of Alberta, Edmonton, Alberta, Canada
BIOSKETCH: Consolato M. Sergi is the Chief of the Division of Anatomic Pathology at the Children’s Hospital of Eastern Ontario, University of Ottawa and full Professor of Pathology and Pediatrics at the University of Alberta, Edmonton, Canada. Dr. Sergi was born in Rome (Italy), obtained his MD degree with honors at the University of Genoa, Italy, Human Pathology degree at the University of Heidelberg, Germany, the title of hon. Clinical Reader at the University of Bristol, UK, Ph.D./Habilitation at the University of Innsbruck, Austria, MSc Public Health in Austria, and FRCPC degree at the Royal College in Ottawa. Dr. Sergi’s specialty areas of interest are congenital heart disease and metabolic diseases, biliary diseases, carcinogenesis (bone/soft tissue/liver), and mitochondrial DNA-related cardiomyopathies.
Interview by Jonathon Tunstall – 26 Jul, 2021
Published – 09 Nov, 2021
JT – Professor Sergi, how did you first come acoss the field of digital pathology?
CS – I came across digital pathology with QA, (Quality Assurance) when I was reviewing different platforms for QA around 2003. These were different platforms for continuing medical education. I was looking at what was on the market to help pathologists stay on top of their medical education and it had to be a versatile platform. I was evaluating a few platforms from Australia, UK, Germany, US and also one from Canada. I found the Australian digital platform was probably one of the best and we published a paper on our study results. For me that was my first exposure to the technology and from that point onwards, I have seen more and more people become interested in digital pathology. Of course, there have also been many challenges along the way.
JT – So that was your primary initial use case, continuing education?
CS – It was actually a survey of the market, so we looked at different programmes and compared glass versus digital slides. We wanted to evaluate how the whole package could potentially contribute to the increase of medical knowledge of the doctor. At the same time, it also created a stimulus for a better understanding of how to deal with the storage of the slides. It was a very interesting time because obviously this was the beginning of digital pathology adoption and most of the people were a little hesitant about the new technology. That hesitancy still exists today amongst some of my colleagues. I still speak with several pathologists; some are more open minded than others and there are a few who are still not confident with the idea of digital pathology. There are obviously challenges and much of those lie around the modernization of the Lab and the LIS. I don’t think we will reach full digitization quickly. It is something that we need to implement gradually, but we are moving forward step by step and it is at least, a very interesting part of the lab modernization process.
JT – Were you scanning slides yourself back in 2003?
CS – At that time, no, we received the digital slides from a vendor and we evaluated the quality against the glass slides. When you do that and look at the digitization of slides, it is clear that there are some good vendors in the market. There are also challenges in making these comparisons because obviously you have to deal with digital slides in a different way than you would glass slides. That is not just about the difference between putting the glass slide on the microscope versus looking at it on the monitor. You also need to have the same systematic way of scanning the slide, first screening at low magnification and then at a higher magnification. This was an aspect that many people found challenging because when you look at a website for example, your eye will focus anywhere on the screen and then you tend to move your eye around looking at the elements that attract your attention. In microscopy, you need to have a systematic way to investigate the tissue. If you turn on your computer and open a file, you will see your digital image that has been saved and it is easy just to move around that image, rather than systematically screening the digital slide. Then you are going to waste a lot of time and actually, this is a problem, because it reinforces the view that the analysis is faster with a microscope than using a digital slide. So, we need a systematic way to investigate the slide. I think this is a key point and one I feel has not been adequately explained to pathologists and computer scientists. There are people on both sides of these discussions, but I have been in this field a few years now and I’ve heard a lot of this sort of feedback.
JT – I think there is also something to be studied in the way we use the zoom because experienced pathologists can glean a lot with a 20x objective by just scanning across the slide. With a digital system, it is much easier to move through the magnifications, but that is also a mew and very different way of working for an experienced individual.
CS – Yes exactly, and the correct systematic approach needs to be understood by the investigator, because you cannot just open the file, take a look around and expect the diagnosis to jump out at you. It’s not going to happen. You need to understand the correct investigative approach. I agree that you can grasp many interesting data and details from 20x, but all the magnification levels can give you some information and that is because all this data is processed simultaneously in your brain and assists in making a diagnosis. This is an important point and I sometimes make a comparison with vaccine hesitancy. Some people are unsure because there has been an issue with communication. If you communicate correctly with an audience or you take time to talk to the family of a patient for example, then the perception is going to be completely different than if you just use a flyer to communicate. Then people will be hesitant and that is the same whether we are discussing the vaccine or digital pathology. The key is communication, that is my personal perspective.
JT – If you started working in this field in 2003, you must have seen a lot of advances in image quality. Certainly, the resolution of scanning platforms has improved dramatically, and they have also got a lot faster. Would you say the image output from these instruments is now equivalent to the microscope? Is the microscope still an essential tool for the pathologist?
CS – The resolution is now very good and as you say, I have seen some impressive improvements in the quality of the machines. There are still some vendors who do not have acceptable image quality for the pathology lab and when you need a diagnosis of course you need the highest possible resolution. On the other hand, you can find pathology labs who have very poor-quality microscopes with lenses that are badly manufactured and the image quality from the microscope is therefore also poor. Ultimately it can boil down to how much money you want to spend on the modernization of the lab. That modernization however is not a simple or easy process, and it will still not be easy to implement a decade from now. The change to digital slides can be considered as only one part of the laboratory modernization. You cannot expect to have a lean laboratory just because you have digitized your system. However, getting back to your question, I am pretty confident that digital image quality will continue to improve in the future. You cannot expect to have electron microscopy resolution in a light microscopy system, but it will continue to improve.
JT – We are talking about the implementation of a digital ecosystem, aren’t we? You may have a slide scanner, but you need to be able to store your slides and you need some sort of digital workflow so that the pathologist can be presented with his or her cases for the day. You also need to be able to share those with colleagues in other hospitals for referrals and second opinion and all of this infrastructure is very costly.
CS – For the moment, storage is a big issue. We see that in the case of radiology with their smaller file sizes but higher volume, the PACS systems need 1,000 TB of storage or more. Storage prices however are going to be lower in the future and external drives will be available for each of the pathologists. My personal opinion, however, is that this is only one piece of the puzzle. I think the most important change will be cloud diagnosis and ultimately this will facilitate a clinician being able to use a tablet to make a diagnosis during a face-to-face meeting with a patient. We see this sort of thing in science fiction movies at the moment, but many of the pieces (the devices, cloud computing etc.) are already in place. I do believe that cloud digital microscopy is coming and that it will be hugely significant. Under this system, it is possible that the digital slides may not be held in the department but may belong to a cloud computing network. Then there is obviously a discussion about cyber security. There will be new challenges, but in digital pathology, many difficult questions have already been resolved. We will need a very sophisticated way to verify the lab workers, pathologists and clinicians and everyone involved in the process. Right now, we can sign out reports electronically and most of our systems are intranet based. However, in most of these systems there is no two-factor verification or facial recognition required to sign out a report. All of this stuff needs to be implemented and to date these topics have not been discussed too much or considered important. People focus only on the digitization itself, but if we want to make sure this technology is fit for the future, we need to build in very robust safety and security factors.
JT – Isn’t it a scary world if everyone’s medical information is going to a network of cloud servers?
CS – Yes, but that is how we deal with data now. Your data are harvested all the time. You are identified through Google or whatever software you use and even if you decide to use Unix or Linux, there are still opportunities to grab your ID and maybe even your physical address. I think cybersecurity will be key in digital pathology and this should be understood by the labs that want to adopt this technology. Obviously, the CEOs will be afraid of getting caught up in a ransom situation and there are several firewalls that you can put in place to make your systems safer. I think genetic data will also be available in the same way and we then need to find ways to protect all this data from attackers and from those who could profit from it, such as insurance companies. It is not fair to imply that someone will definitely develop a disease simply because they have a particular mutation. There are many challenges and ethical questions to resolve here, but I think if we can develop strong cyber security around our digital pathology platforms, we will create a system which works much more efficiently and which will be safer for the patient.
JT – Thinking about the size of digital whole slide images, wouldn’t we want to leave those in their original locations, and then find them through a web address with associated viewing permissions? We wouldn’t want to move the images around. Better to be able to access their location, share via a remote viewer and then apply analysis tools to those images from cloud-based resources.
CS – This is something which has not been discussed in detail, but I do think digital pathology and digital diagnosis will become more and more associated with cyber security. In the future you will need not only IT people but also cyber security people. We are currently seeing rapidly increasing salaries of IT staff and this reflects the importance of these skills.
JT – So in this future world, can a lab using traditional microscopy still exist?
CS – Yes, I think there will be a hybrid system for a long while and I don’t think the microscope is going to disappear any time soon. I’m not thinking just about in those countries which are poor in resources, but there will also be private practices where pathologists are accustomed to making diagnoses via a microscope. I think the microscope should always be available. We shouldn’t envisage a complete change to digital imaging, because freedom of choice is also an important aspect. We should not assume to be able to convince people to use a new technology just because we need to use it. This goes back to the hesitancy about digital pathology. If you communicate and educate people, there will be more and more people who feel able to make the change. Personally, I think we will see a slow and steady move over to digital, rather than a rapid change. There are only one or two labs right now which are fully digitized.
The other thing to consider is the storage and what to do with the glass slides once you have scanned them. There are two possibilities; you either keep the slide as a medical reference, or you recycle the glass. Also, glass slides fade eventually and then if you don’t have the block, you will just have a faded slide. You can re-stain perhaps but the digital image doesn’t fade and can potentially be preserved forever. That is a difficult issue, and I cannot give you an easy answer, but probably it the legal institutions or the lawyers will work out what to do with the glass slides.
JT – Where do you think image analysis and AI fit into all this? We have talked about digitization itself and that there are clear benefits for consultation, education and research, but image analysis is a second layer of technology. You can’t use image analysis and AI applications without first having digital slides. Some of the algorithms seem to be getting quite good now. Where do you see that going? Does AI change the role of the pathologists in the future?
CS – First of all, I think people are afraid of AI. Of course, in our case, it is a good thing. AI is an aid to the human pathologist, and we should not be afraid of these new tools. One of the topics I have been discussing with colleagues however is human bias, and there can also be bias in an AI system. We have protocols to avoid conflicts of interest and bias in humans, but we probably don’t have enough controls against bias in our current AI applications. There is also the risk to eventually see AI as the big brother and to not expect mistakes. We have to use AI in conjunction with QA. I don’t think that the pathologist is jeopardized by AI because the need of the pathologist will still be there. At this stage it is almost impossible to have a robot doing an autopsy, but maybe in the future, that will be possible. There is a lot of variability in the data you get during an autopsy and in this way, the pathologist will probably remain important.
In fact, I don’t consider AI to be the future in pathology, but to be the present. We have AI working now already helping us to do our jobs. This conversation is running with a webcam and you see me on your monitor. Of course, there is a lot of opportunity to do good things and bad things with AI, but I think most people should see it as a good thing. It’s like when we first got a calculator on a watch and nowadays you can have heartbeat, steps etc. There are always opportunities to improve, and I think pathology will be better for it. In the future we will certainly see more AI, probably an explosion of AI.
JT – So you seem to be saying that you see the pathologist remaining in control, with the AI acting as a digital assistant for screening, grading etc?
CS – There are fine nuances between some of the Gleason grades for prostate cancer in adult pathology and some biliary diseases of the infancy and childhood. Pattern recognition is going to get better, but it comes down to how you train and how you validate your AI. The other important aspect is around trust and whether your partners, the surgeons, will come to trust the AI more than they trust you, or you more than the AI. This is a tricky question. Do we trust the first values we see, or do we repeat two or three times to make sure? I think there will be different protocols to approach the digital slide from the AI perspective. Many of the concepts that we have in our mind now are changing because using pattern recognition systems we will capture parameters that we cannot capture with our eyes. These can then be used in Kaplan-Meier survival analysis and we will see how these parameters can be used practically. Using AI, we will certainly discover new parameters.
From this perspective, I am optimistic. I think there will be big improvements in the diagnosis of many cancers and in some of the current parameters that are too trivial and not sophisticated enough to assist in an indication of diagnosis / survival. In pharmacology and pharmacogenomics, the use of AI has already exploded. AI will also have a big role in quality assurance.
JT – And all these advancements should improve patient care.
CS – Yes, and we must continue to make sure that we have controls in the system. Those labs where you have a solo pathologist, that is also risky, even if they are the best. There are pros and cons to all the different ways of working. We need to use our QA systems in coordination with AI. That will improve healthcare processes and ultimately give a better, faster service to the patient.
JT – Professor Sergi, we’ll leave it there. Thank you for your time today.