Floppy Disks to Diagnostics

Many of you may remember a childhood spent with floppy disks, Atari joysticks, and computers whose operating systems were nothing more than green words glowing on a black screen. Graphical interfaces were still years away – and the idea that a computer might be able to assist with something as complex as medical care was still a twinkle in science fiction writers’ eyes. Yet in our lifetimes, we’ve gone from that fantasy to the point where we have to consider just how much, and in what ways, we can trust computers as diagnostic aids. It’s an exciting journey and one I’m privileged to be taking. But a journey that comes without obstacles is rare indeed!

Analog beginnings

My career as a pathology consultant began in 2008 when I completed my training. After three years working at the Royal Bournemouth Hospital, I relocated to Oxford, where I am now a specialist urological pathologist.

I initially joined the National Health Service (NHS) in Oxford as a consultant pathologist but, after a couple of years, I was invited to build a research group at the University of Oxford. As part of my work on prostate cancer with the Oxford Biomedical Research Centre, I was exposed to artificial intelligence (AI) and the building of algorithms. The process fascinated me – few people really know how much work goes into an algorithm to bring it up even to research grade, let alone make it viable for clinical practice. It’s through that experience that I recognized the need to build an infrastructure from the ground up on which to pin the algorithms my colleagues and I were developing.

Although I have no formal training in digital and computational pathology, I did most of my growing up in the 1980s, when computers were just starting to come into people’s homes. I used to sit with my friends and load up computer games, first on tape and then on disks. My family had one of the first disk computers to enter the market, so you can see I’ve always been tech-savvy! I’m amazed at how quickly we’ve gone from 5” floppy disks that would inevitably crash before they loaded to the kind of technology that can help us diagnose patients’ diseases and save their lives.

Early applications

I use digital pathology a lot in teaching – for instance, to set up online teaching portals where I can present a variety of interesting cases and include text sections that explain each case and provide answers. I recently taught the British Association of Urological Pathologists’ testis course that way; instead of lugging a bag of slides across London (and hoping that they all arrived at their destination intact), I uploaded my seminar to a secure, cloud-based site, and participants viewed all of the relevant and anonymized cases on that “virtual microscope.” I find it really exciting – and certainly much more efficient (and pleasant) than transporting physical slides.

My institution has now made significant moves into the digital space for primary clinical reporting. We have installed slide scanners and use the Philips IntelliSite Pathology Solution in our laboratory, as well as around the region, so that we can take digital referrals from other hospitals. At this point, we’ve finished retrospective validation in some subspecialties and are now ‘’live’’ for digital reporting for medical renal, hematopathology and testicular cases. The entire team at Oxford University Hospitals NHS Foundation Trust is very excited that we are changing the way that pathology is practiced in Oxford and the UK for the better. I’m optimistic that we will see benefits for patients in terms of quicker turnaround times and greater flexibility to work with colleagues across the region, providing expedited access to expert opinion. Benefits for pathologists include more flexibility to work offsite and access cases remotely and securely, for example, when urgent diagnosis is needed.

Urological pathologists, like many subspecialists, have seen a significant increase in workloads over recent months – not just numbers of cases, but also in terms of their complexity. There are only so many of us trained in this particular field so, at a certain point, there is no more “people power.” Many hospitals now find themselves in this situation, so we have to begin looking for ways to work smarter, rather than simply harder. Digital pathology can help us make more efficient use of the resources we have.

To validate case reporting at Oxford according to Royal College of Pathologists guidance, we’ve defined a number of small pilots to establish feasibility, determine the pipeline, and work out any issues we encounter along the way. Once we’ve done that, we’ll snowball our efforts into bigger specialties and begin using it more widely. I’ve taken the approach of not running before we can walk to make sure the processes are safe; hence the smaller pilots. We want to demonstrate to our pathologists and biomedical scientists first, that this can work, and second, how we might do it on a small scale so that we can scale it up rapidly, which we are planning to do over the next 18 months. We’d rather get the team on board by demonstrating feasibility than by dictating to them what we’re going to do, so I’m trying to gradually win their hearts and minds. Luckily, I have a supportive team who are all excited about the digital pathology revolution. We’ve implemented a steering group, we have an exciting research program with an engineering collaborator, Jens Rittscher, and we’re developing algorithms that might be of use in the near future. I’ve been fortunate to work with such keen early adopters. It really is a team effort.

Addressing uncertainties

I believe digital pathology is not only an exciting opportunity for the laboratory; I also think it has the potential to reinvigorate academic pathology. For me, the idea is intriguing because the efforts involved – building algorithms, setting up digital infrastructures – also highlight the value of the professionals in our field. For instance, it’s impossible to develop useful algorithms without involving pathologists, so we are given the opportunity to apply our expertise in a novel way. And it’s interesting to unpick the diagnostic process when I’m building algorithms, because a lot of what we do is pattern recognition. You may know a lymphocyte on sight, but the computer doesn’t, so you have to ask yourself, “What are the features? How can a machine ‘brain’ conclusively identify this object?”

On a practical level, I recommend that laboratories looking to enter the digital pathology sphere tap into the knowledge of those who have already done it. Such a transition involves a great deal of research, especially with respect to procurement. By talking with someone who has already undertaken a similar task, you can save a lot of time. I also recommend involving your IT department early – and you should keep an eye out for hidden costs that seem to come slightly further down the line as you work out the practicality of transitioning to digital. For us, in anticipation of high-volume routine work, a reliable high-volume scanner and a feature-rich viewer and case management workstation were clear choices.

Digital deployment naturally involves some capital outlay and developing a strong business case is quite complex – after all, the cost savings come from a variety of different places. Staff time is saved, for instance, but incrementally rather than as an easy to define single block. The argument I rely on most often? Quality! We want to decrease turnaround times, reduce the chances of error (because the barcoded workflow reduces the likelihood of mismatched cases), increase quality around network working and flexibility, and provide better access to second opinions.

Another argument that works quite well in a large academic center like Oxford is the development of clinical trial portfolios. Digital pathology platforms enable us to create a library of cases with easily accessible images; central review of patient slides or image analysis for new prognostic markers is much more efficient (with the appropriate consent, of course). You don’t have to scrabble around in the archives searching for old slides that might be missing or misfiled.

In short, it’s harder for decision makers to say “no” to a technology that offers speed, safety, and quality.

A digital future

For the generation of younger pathologists coming through the system, I think that digital pathology and image analysis will make the field more exciting. The next generation is much more likely to fully embrace the technology and use it to change pathology. I expect that we’ll start to use AI and image analysis tools to support our work. And though algorithms may replace some of our current tasks, I don’t feel threatened. I believe they’ll more likely take over the simplest duties, so that we can focus on the more complex tasks that rely on human instinct as well as intelligence. For instance, you might oversee a prognostic algorithm, but that doesn’t mean it will do the entire job for you; rather, it will give you a set of data points that you will need to verify you are happy with to incorporate into your report along with the information you generate yourself. Pathologists’ roles may shift to include more supervision of complex algorithms and AI, but it will still be humans who have the final word.

AI, and digital pathology in general, will do more than just remove tasks from our workflow; they’ll stratify our work into what needs immediate attention and what doesn’t. What requires extra testing? What needs a human eye to make a final diagnosis? I also think that pathology will increasingly feed into big data models; rather than having the pathology report as a standalone entity, it will be part of a pipeline and will be seen alongside the patient’s other results (from clinical examination, sequencing, and so on). Features extracted from the pathology images and report will be fed into that overall pipeline, which will lead to advanced predictive models. And that’s where I think pathology will fit into the big picture – as part of a multi-modal patient assessment.

The pathologist’s new clothes

Going forward, I expect that most – if not all – pathologists and biomedical scientists will need to have a working knowledge of IT. It’s likely that computational pathology skills will need to be embedded into the training for new pathologists – in much the same way that molecular pathology is now included. Why? Eventually, pathologists will be taking responsibility for the outputs of algorithms in their reports and, to do that, we need to have a basic understanding of how each algorithm generated that output – or at least understand the steps that have led to that result, so that we can make sure it’s valid. Otherwise, how can we take responsibility for it?

I think everybody, both trainees and existing consultants, will need to upskill. I also think there will be a place for a new subspecialty within pathology wherein people are trained to a higher level in these new technologies – “pathology informaticians” or something along those lines. Some institutions in the United States already have a designated specialist who takes responsibility for deploying digital platforms and setting up algorithms. I can see other countries taking a similar tack in future years.

Unfortunately, no amount of training will help if there simply aren’t enough laboratory medicine professionals to take on the amount of work we’ll be facing in the coming years – a workload that is steadily increasing as more and more pathologists approach retirement. Digital pathology will help us make better use of the workforce we have. That applies to aspects such as optimizing the diagnostic pathway, but also to tasks as simple as searching through stacks of glass slides to find the specific ones needed for a meeting or a case presentation.

Strengths… and weaknesses

The strengths of digital pathology are obvious. It’s exciting, it’s cutting-edge, and it will enable workforce flexibility to patient benefit. I think it will particularly help in highly specialized fields that need geographic cross-covering because of the small number of pathologists trained in those fields. These pathologists often lack a robust and well-designed archiving system – a key factor in the success of their daily work. My subspecialty, uropathology, is one that benefits greatly from digital and telepathology; others, like renal pathology, will probably see similar benefits. I also predict that the routine creation of library images will underpin personalized medicine, research, and clinical trials going forward. And, of course, the potential reductions in errors and turnaround time are great strengths.

At the moment, though, the volume of data these platforms generate is a challenge we must weigh against the requirement for pixel fidelity to be validated from the scanner to the eye of the pathologist. To me, at least, that challenge currently has no clear solution. If a lab becomes fully digital, it will churn out terabytes of data and it won’t be long before it runs out of storage space. Some laboratories are moving to off-site storage or low-cost cloud options, but it remains to be seen what’s best and most secure. I also see interoperability between platforms as a problem – although most companies are working to improve this. I think one of the main issues is that, when you’re trying to make a purchase in this area, it’s difficult to put all the information together. To succeed, you need experts across the table and design aimed at both the present and the future.

Finally, on a more general level, I think we’ll need to have extensive conversations about consent, patient and public involvement, and the building of algorithms. Just as digital pathology and AI are taking off, the General Data Protection Regulation (GDPR) has been implemented across Europe. We need to take a step back, look at what GDPR allows, and interpret that for digital pathology so that we can fully understand our responsibilities.

These weaknesses are not insurmountable, but they will require a strong focus and investment to overcome. Some of them must be addressed by industry and standardization – like the interoperability issue – whereas others, such as data storage, need to be fixed by institutions in collaboration with industry. I hope that centers that have already embraced digitization and found solutions to these problems will share what they’ve done so that others can follow in their footsteps. Other issues must be addressed on a national or even international level; for instance, those tackling consent, data protection, or public involvement. It’s true that there are many grey areas at the moment, but all great modernizations begin with unanswered questions. We’re getting there!