Digital Ready Slides

Thank you for the introduction. Hello and thank you to everybody who has joined today. I would like to introduce the concept of digital ready slides.

• What do we mean by that?
• What are the attributes associated with it?
• And also, to provide some practical guidelines and pragmatic tips on how best to create digital-ready slides.

As we look at digital-ready slides and creating digital-ready slides, one of the questions I often get asked is well, why now? And to do this, maybe we take a little bit of a step back and look at when artificial intelligence first became recognized. This was back in the 1950s when the term artificial intelligence was first coined by John McCarthy et al. and looking at a branch of machine learning where a computer could emulate what a human may do in the same situation. And then we jump forward to the last two decades and looking at the widespread adoption of digital pathology across several use cases, whether that's in research or clinical type applications.

I'm moving exclusively away from the microscope and starting to leverage digital pathology and on-screen review of slides. And it's kind of the convergence of these two that are leading us to digital-ready slides. But even then, the “why now?” questions still exist and I refer to this sometimes as the perfect storm. And we're now at a stage where the maturity of digital pathology technology is suitable for true high-throughput routine usage.

You also have acceptance of the usage of digital pathology by several of the governing bodies and regulatory bodies in different regions across the world. We've also seen massive developments in the infrastructure and IT capabilities that are available, both looking at disk speeds, the widespread availability of Internet connectivity and Wi-Fi. And, a huge body of evidence and proof of the value of the utilization of digital pathology.

And that puts us in a position where now we're starting to see as well the development of artificial intelligence-based tools for usage with digital pathology and the input to those to any development of an artificial intelligence based tool for pathology is going to be based on those digital pathology slides and that gives the whole requirement and the need for having digital-ready slides.

There are some key features that digital-ready slides should exhibit. You can see them here on screen. The first is that these will scan automatically and this is to minimize the user intervention. The idea here being that as you start using these routinely for high throughput, widespread scanning. You don't want to be burning 30 seconds or a minute on every slide because the hands-on time for that just becomes paralyzing.

The other piece is that you want to make sure you capture all tissue that's on the slide. This would include any small fragments of tissue if there's faintly stained tissue, particularly where you may have some IHC staining and there's faint expression in a fragment of tissue, you really want to make sure that that's still captured or low cellular density tissue as well, which can sometimes be problematic.

You want to make sure that all that tissue is captured on the slide, but also you want to make sure that it's done with the smallest possible scan area, and ideally in one focal plane. And really where this comes to is you could capture the whole slide region for every single slide, but you're going to capture a lot of white space at usually high-resolution scanning, and that is going to firstly take longer to scan and secondly, you're going to have the larger file size and then storage requirements that are unnecessary. These two pieces together where you want to capture all the tissue, but you want to do it in the smallest possible area. And if possible, in one focal plane because as you capture these digital pathology slides, they can be very large depending on the magnification that you're using.

Depending on the size of the piece of tissue that you're capturing, but these are in the range of anywhere between 250 megabytes to a gigabyte in size, give or take. Every time you capture a Z-stack, you're capturing another image, so you're multiplying, then the file size across the slide. Ideally for routine usage, if you can capture at a single focal plane, that's going to be advantageous both from a scan time and a file size perspective.

The final piece in the key features of digital-ready slides is to automate the data flow. And again, this is with a view to minimizing hands on time, not sitting there doing data entry, but also to avoid manual data entry and associated error risks.

As we look at bringing digital pathology into a laboratory, it's not as simple as putting a scanner in after the coverslipping stage and taking off from there. We need to look holistically at the complete workflow, that end-to-end workflow.

Within the lab and all the different steps in that tissue preparation stage, that will impact your final slide. How do we optimize each of these steps to create these digital ready slides? For this, I'm going to start with embedding which predominantly pertains to when you're embedding multiple tissues in the one block, such as needle core biopsies for instance.

The first piece is to cluster the samples together. And this follows some of the best practices that you would have typically anyway at embedding, but the idea being and you can see it clearly in the cluster samples in the one on the left samples are quite spread out. What this is going to do is from a scanning perspective, you're going to have a larger scan area there. That's going to take longer to scan and it will lead to a larger file size. This and then the version on the right shows you a few. Just put those a little closer together, still with sufficient space so that they don't overlap during sectioning and there’s a limited risk that that would happen.

This will give you about half the scan region and again that's going to make you twice as fast and half the file size. The other piece, when you're clustering your samples is to align them in the same direction and this also gives an advantage depending on the scanner that you're using or the type of focusing mechanism that it uses. If you can align your samples so that the focusing method or the scanning off your slides follows the direction of your tissue. You've got much the higher chance that all your tissue will be in focus. And that's because your scanner isn't going to jump from tissue, into white space, into tissue and kind of mapping that micro topography of the environment.

As we move on from embedding and into microtomy, there is a massive importance of best practice and some of these things aren't exclusive to digital pathology or digital-ready slides. However, they can be exaggerated when you move into a digital environment. And again, if you look at something like tears, they're annoying and bothersome when you're reviewing under a microscope.

When you start to utilize digital pathology, this can cause the focusing, depending on the type of scanner that you're using, to jump out of focus. Where in that white space it's trying to detect tissue and maybe it will detect a fragment that's at a different level in the micro topography and then cause some out of focus region. The very same is true for folds within tissue and even looking at the example that we have here, that tissue is going to be 2-3, maybe 4 times deeper in this folded region that you can see and again that's going to lead to out of focus regions or the necessity to then acquire in Z-stacks.

As we move across and look at contamination and this is an example of fungal contamination within the water bath and from not changing the water frequently enough. These little flecks of debris that appear on the slide can be recognized by your scanner as faintly stained tissue or potentially low-density tissue. That’s going to lead to a much greater region of interest being selected and scanned for the slide leading to those again slower scan times and larger file sizes. The same thing is true with chatter on your sides or in your tissue, very similar to tears or folds, which is all about that focusing mechanism and about being able to map that microtopography within your slides.

In addition to the standard best practice pieces at microtomy, some considerations where you may need to vary your practice when you start looking at introducing digital pathology is the section thickness. Often this can be a little bit thinner than the sections that are typically used, so it's typically somewhere in the region of three to four microns. This tends to be ideal for digital pathology, enabling you to scan in a single focal plane.

If we have a look at how this works and you can see an overview here of an objective lens. Very typical as to what you would see inside most scanners, and you'll have your layer of optimal focus, and then you'll have a depth of field. If you have thin sections and they will have that microtopography and variability along the surface of the tissue.

If all your tissue falls within that depth of field and we can map according to it, then there's a high chance that you will be able to capture all your tissue in focus in a single focal plane pass. However, if your sample is thicker then you may find there are areas where you cannot capture a single focal plane. That's going to push you to a situation where you're going to need to employ Z-stacking or have out of focus regions, neither of which are optimized when you're looking to do high-throughput routine scanning.

The other piece to take into consideration when you look at section thickness is the impact of thickness on staining. In the example here, I'm looking at IHC staining interpretation and it is documented in the literature and peer-reviewed publications on showing that section thickness affects our interpretation of IHC staining. Whether that is done manually or using automated analysis techniques, and I think this even becomes more important, as we start looking at either manual sharing of slides or manual review on screen with digital pathology. Where you could be under-calling or over-calling your IHC staining based on the section thickness. Or indeed, as we look at the further development of tools utilizing and leveraging AI and that standardization of your section thickness becomes even increasingly important.

Moving on, but still at the microtomy, is your tissue placement. When you're floating your sections, you've got nice ribbons. You're floating your sections and the placement on the slide. Best rule of thumb is to place centrally, but this is even more exaggerated when you look at digital pathology.

Because most scanners will have a no-scan zone towards the edges of the slide and this no-scan zone may even overlap with the edges of the cover slip. Again, this varies model-to-model and manufacturer-to-manufacturer. It's important to know about the scanner that you're looking to use within your laboratory.

What are the borders for the actual scanner? Login and make sure that all your tissue is within that scan region or that scannable region on the slide, and make sure that you don't have situations like they're shown on the left of the slide here. Where again, best practice, you don't have fragments of tissue outside of this coverslip area. All these things, if you have your tissue misplaced on the slide, can impact your tissue detection, your tissue focusing, and the ability to capture all that tissue in your digital slide.

Moving on then, away from microtomy to slide labeling. The first piece I'd like to touch on is in relation to automatic slide identification and I've said it previously, but the rule of thumb that I go by here is if you're looking to scan any more than 100 slides daily, you really will need to look at bar-coding your slides.

Digital pathology images are so large they do require specialized image management and viewing software. And typically, then you need to associate your metadata with those slides so you know where a slide is from, which case or animal or study is it related to. And that's when integrating with your laboratory data system, whether that's your LIS or your LIMS, becomes integral into having an end-to-end solution for your digital pathology. And that what that will allow you to do is based on your bar codes, automate the association of your slide data and your whole slide images and in aggregate those cases with those studies together.

To move from a situation where you are manually slide-sorting into the appropriate groups or cases to something that can then be truly automated. This is one of the areas where bringing digital pathology into a laboratory can add some extra steps to the lab. It can also help alleviate some of those traditional steps, such as case sorting, which can be done automatically. There are some workflow benefits that can be realized as well.

The other piece about using bar codes or slide labeling comes from minimizing your data error risk. Nobody wants errors within their databases. Looking firstly at the manual workflow and the slides here represent handwritten slides. And for those of you with a very keen eye, you may notice that the slide at the back says B46H5S0 and the one at the front is B46H5S0. Or is it a 0? Or is it a O? It's very difficult to tell and often impossible to distinguish with handwriting.

As we move along, this is clearly an area where you have risk. There's very much risk of handwritten errors but also interpreting that handwriting. Reading somebody else's handwriting can always be problematic. It's also very time-consuming compared to generating slide labels automatically.

It doesn't really impact too much at the scanning stage. But the next area then where it comes is once you have your slides scanned, how are you going to create a manual workflow where you have handwritten slides? You're going to manually have to create that association between your digital slide coming off your scanner and your database, whatever format that might be in. This requires significant data entry and with any data entry it's going to be time consuming. It’s another job to be done manually and it's also going to be inherently prone to keystroke errors. Ultimately, those two pieces together lead to potential errors within that final aggregated slide and data set that you have.

If we look at the bar-coded workflow, you can have human readable and bar-coded labels as the one label. You will eliminate that error that you may have with handwritten or transcription errors. Leveraging that integration pathway with the existing information management systems, you can be assured then that the right slide is being associated with the correct data within your information management system. And that aggregation is being done automatically.

By moving to this type of scenario, you are going to limit your risk and eliminate some of those manual workflow steps within the process that overall give you greater confidence and security in what you're viewing digitally on screen or indeed leveraging for AI or automated image analysis development.

When it comes to slide labelling, there are some best practices to consider with these things to do and things to avoid. If you have multiple overlapping labels, you can run into some difficulties or come up against some challenges. The first is that you may exceed the slide thickness tolerance. This is going to be very scanner dependent. Looking at how different scanners and different models move the slide from whatever kind of view the slide tray or slide carrier has and puts it in under that objective for high resolution scanning.

There are several different methodologies involved, whether it's a push pull mechanism on the slides, some scanners have little grabbers where they pick up the slides and move them. Or they put them into little clip-based slide trays for moving into the that kind of under the objective for the high-resolution scanning. All of them that will have a tolerance, no matter which mechanism is employed, each of them will have a thought tolerance to the slide size and the slight thickness, and if you exceed this, you run the risk where slides can get jammed, which is kind of the easiest one to resolve in a lot of cases, but still jammed slides mean it's unscheduled down time, you're going to have to remove them similarly to the way you'd remove a paper jam in a printer. And slides may not fit in the carrier. In the worst-case scenario, slides can get stuck. You may not be able to resolve it. You know on site you may need to call in an engineer or in worst case scenario slides could get broken meaning you're going to lose that sample, which is not a good place to be in if this is a rare sample where you can't go back and cut from another block or and potentially broken glass within your scanner mechanism, neither of which are positive things.

The same is true for sticky overhangs, so trying to ensure that when you place your slide label on your slide that it is fully within the boundaries of your glass slide because again, similarly sticky overhangs, they can stick the slides to the moving mechanisms to trays and can result in all those problems we just discussed with broken slides, potentially broken glass damage slides, slides jammed in the mechanisms. The best practice is where possible, single slide label within the boundaries of the glass slide.

The next step in the process that I'd like to speak to in relation to digital-ready slides is around the staining and whether this is your H&E staining, your standard routine, staining special stains, IHC or other staining modalities. Moving from reading slides on a microscope to reading on screen is a paradigm shift for the reviewer. Whether that's a pathologist or laboratory personnel or researchers.

By standardizing your staining, you're going to remove one of those obstacles or those pieces that need to be overcome as you're moving towards on-screen review. The other piece is that some of those added advantages of digital pathology, such as the multi-slide review options, you can see it here on the top right hand side where you can pull multiple slides onto screen at the same time potentially serial sections from the one piece of tissue or to compare the staining and the expression profile between different groups or different study groups and a research situation. That is going to emphasize any staining variation that you may have within your tissue preparation. You want to be able to have confidence to say that the staining that you're looking at is true staining and that, in the example that's shown here, that is a variation in staining and that the tissue on the left-hand side does have a higher expression profile.

And that this isn't to do with manual variation in staining that I didn't leave it to bind if I needed to. And really, it's that ability of digital pathology to highlight and emphasize some of these items. And the other piece, you'll see it then down on the H&E slides towards the bottom of the screen. These are serial sections of the same tissue cut to the same thickness and sent to different labs asking them to just do their normal H&E staining on it.

What you'll see here is a massive variation into the color and the intensity of the staining of their H&E. As you start looking at leveraging digital pathology either for sharing across multiple sites where you have your tissue coming from different labs and either into a centralized or distributed review that can highlight some of these differences and really bring them to the fore. It also potentially has that impact as you start looking at AI development. What's the variability in your input to those training sets and what impact does it have then on the size of the training set that you need and the performance of your AI tools?

The next piece at the staining stage is to make sure that your slides are free of any staining debris or artifacts, and the easiest and most simple way to do this is to ensure you have the appropriate post-stain washes and what they will do is obviously remove any residual stain.

If you've got any streaks of stain across your slide, and sometimes you'll get those DAB clumps, and if you're doing any kind of manual IHC work or just general debris. And again, these pieces of debris can be problematic on several different levels.

When you start looking at digital pathology and scanning those slides. And the first is that and in the situation of the slide on the top here where this piece of debris towards the left hand side could be recognized as tissue. That's going to increase then the scanner area which is going to give you a bigger file slide size and slower scanning. That's going to give you slower throughput overall. If this is a recurring issue, it will really start to impact on that throughput of scanning within your laboratory.

It can also lead to out-of-focus regions depending on the scanning mechanism of the instrument that you're utilizing. It can cause it to recognize the debris as the tissue and set the kind of growth focusing like your course focus on a microscope that it would be set there and then your actual tissue of interest may be out of focus and it can also lead to missed tissue. Again, if you had a situation where you've got a big clump of dark debris and then you've got some faintly stained tissue that may recognize the debris as the tissue and scanned that instead of the sample of interest. And all of this can lead to numerous problems, as we've discussed.

The last step then, as we move on from staining is coverslipping. This is the last process before either going into your microscopy-based review or before scanning. I would say if you're really looking to employ digital pathology and create digital ready slides to give very strong consideration to automated coverslipping, if it's not already employed within your laboratory. And this is for several reasons.

The first is the elimination of bubbles. Bubbles can be problematic when focusing when they get recognized as tissue. If they are collocated with the tissue, like reviewing under a microscope, you may not be able to pull that tissue into sharp focus because of the bubble. If it is located away from the tissue, again, this can be identified as tissue and then captured within the scan region.

The other piece is no excess mounting media, and this is quite similar in a lot of ways to the same situation that we had with the sticky labels on the slide. Excess mounting media has two phases when before it's fully cured and hard, when it's in that sticky and tacky state. It then runs the risk of sticking to your moving mechanisms within the scanner, whether that's a slide tray or moving on to your scanning stage for that high resolution scanning. Sticky slides are more likely to get stuck in your instrument and lead to several problems and jammed slides, broken slides glass in your scanner. The other pieces then, if that excess mounting media has hardened again, you may have lumps of this hardened mounting media at the edges of your slide that you're going to mean that your side exceeds or could exceed the tolerance of your scanner, again leading to the range of problems.

The other piece is to have clean dry slides. Not all scanners keep the slides in a completely horizontal position. Some of them will tip them either at an angle or into a vertical position in some situations. If your slides aren't fully dry, and if your cover slips aren't fully dry, there is a risk that they will start to slide off and pull your tissue with them. The other piece is if you've been handling the slides between cover slipping and scanning there's a risk of putting debris or fingerprints or glove dust on these slides. Finally, you need to make sure that your cover slip is on straight and again this is due to the tolerance that the different scanners will have in relation to movement of the slide or the slide trays within the scanner.

To summarize, as we look at digital ready slides or digital pathology, the adoption of digital pathology is such a multifaceted project and affects multiple stakeholders within any pathology department. As I mentioned at the top of the presentation, it's not simply just about adding a scanner after cover slipping and adding it to your workflow. It really affects the whole laboratory workflow and requires careful consideration on taking that holistic view of the slide creation to really generate those optimized digital ready slides. So, with that in mind, that brings me to the end of my presentation today and I would be very happy to take any questions that you may have. Thank you very much.