College of American Pathologists
Printable Version

  Digging their way in: digital pathology systems


CAP Today



February 2008
Feature Story

Dirk Soenksen, founder and CEO of Aperio Technologies, launched the digital pathology systems and services company out of his garage in 1999. Today, more than 350 of Aperio’s systems are installed in 25 countries. Soenksen has a bachelor’s degree in chemistry, a master’s in electrical engineering, and an MBA. He’s a former systems development engineer for General Electric and holds several patents for microscope-based diagnostics. He spoke at Aperio’s Pathology Visions Conference in San Diego Oct. 21. An edited version of his talk, “The Evolution of Digital Pathology,” follows.

How do we define digital pathology? Digital pathology is an environment for managing and interpreting pathology information. But this information is unique in the sense that it’s enabled by digitizing a glass slide. The ability to be independent of that glass slide is what enables digital pathology to have the value it does.

Three or four years ago, a digital pathology system comprised a slide scanner and viewing software. It’s a much broader picture now. There are a number of image acquisition modalities: a scanner, robotic microscope, camera on a microscope. That information is managed by an information management system that is central to today’s digital pathology system. In radiology, they call that a PACS, or Picture Archiving and Communication System. Here we simply call it an information management system. And it has to have certain attributes. You have to be able to manage the information securely, to view digital slides remotely, to manage workflow. And you want to be able to store your images. That could be a hospital PACS. And/or you want to be able to communicate with a laboratory information system or other type of lab equipment.

Digital pathology has two core capabilities. One is the ability to consolidate information in a way that you can’t by looking at a glass slide under a microscope. You can consolidate the digital slide, the gross image information from other modalities, in a cockpit—a pathologist cockpit that’s analogous to what your radiology colleagues are using. All the information is at your fingertips. At the same time, you now have new workflow tools—these are the second core capability—that allow you to view digital slides remotely and to conference, archive, and retrieve.

Digital pathology improves the quality and efficiency of pathology services, primarily by getting the right pathologist to look at the right slide. If you can get the right specialist to look at the right slide at the right time, that alleviates the need to ship slides, wait for glass slides, or transport pathologists.

Significant trends are creating pressures for change in pathology. The payers are one source of those pressures. You have cost pressures, you’re being asked to do more with less, and there’s an increased focus on lab expenses and quality. There’s also pay-for-performance. This is coming, and it’s something the insurance companies are striving for. There’s pressure from patients. They’re getting older, there are more tests, they’re becoming more knowledgeable about diagnostics, they have higher expectations of quality and services.

At the same time, pathologists’ customers —clinicians, oncologists, surgeons, and hospitals—are seeking differentiation in the quality and types of services they offer. These customers want to evaluate pathologists based on how well they perform. They need pathology coverage, they need to manage risk, and they largely have an interest in having an electronic health care record that combines the in vitro and in vivo components, radiology and pathology, imaging and microscopic information. And then there’s the lab market itself, where we see tremendous amounts of consolidation, new tests, the desire for a one-stop laboratory, and specialization of pathologists.

The pressures on pathology create needs and opportunity. Here’s how pathology as an industry can satisfy those needs and capitalize on those opportunities: You can say, “I’m going to tweak the current model and make minor adjustments,” or “I’m going to redesign the current model,” or “I’m going to be really innovative and create some new models.” If you want to tweak the current model, you could just work harder. You could spend more hours looking through your microscope. You could cut out waste. You could improve how you procure things and do billing and coding. Or you can redesign your model. You can apply Six Sigma processes. You can add subspecialty expertise to round out the two general pathologists who might be part of your system. You can implement bar codes. Or you could be innovative and say, “I’m going to adopt rapid tissue processing” or “I’m going to adopt a different workflow of slides for my lab. Instead of doing it based on a batch basis, I’m going to do it based on a single-slide basis.” Or you could be so innovative as to say, “I’m going to combine my radiology and my pathology departments,” which people have talked about but, to my knowledge, nobody’s done yet.

How can digital pathology help? Digital pathology can be used in any of those three modes of response. If you want to just tweak the model, you can use digital pathology to lower your courier and shipping costs and improve turnaround time. If you want to redesign the model, you could archive cases for easy recall, use image analysis to increase your objectivity, and reduce the time you spend on tumor boards. Or you could create new models, in which you might decide to generate virtual communities to discuss cases of interest. Or, you could enable content-based image retrieval, where you tell a system, “I have a cell here. I don’t know what it is. Find me other cases with cells that look like this one.” You could pre-scan slides to highlight areas of diagnostic significance, so you don’t have to spend your time searching for rare events or patterns of interest. Because the amount of information on a slide is so huge, there’s significant benefit. A computer can work tirelessly searching a slide at 40∞. As a pathologist, you don’t have the time to do that. Or you could enable the pathologist cockpit and integrate disparate sources of information into your cockpit and then sit there and read out cases. You could think about using digital image processing techniques to reveal details that aren’t available by looking at the glass microscope slide. The message here is that you can use digital pathology incrementally. It’s a question of where on the continuum of adoption you feel most comfortable.

The four key capabilities of digital pathology are creating, viewing, analyzing, and managing digital slides. In 1997, the state of the art was partial-slide imaging. You could digitize only a portion of a slide. It wasn’t possible to digitize the entire slide because there was simply too much data. Whole-slide imaging didn’t become possible in a practical way until 2000. In 2002, soon after whole-slide imagers were developed, came virtual microscopy, which added the capability to efficiently view high-resolution—that is, large—digital slide images featuring image qualities acceptable to discriminating pathologists. Most important, it was possible to view digital slides remotely, including via the Internet. The proliferation of digital slide images created needs for analyzing and managing them—the two capabilities that transformed virtual microscopy technology into the information/data-management solution known today as digital pathology.

When it comes to image quality, where are we today? [See “Evolution of image quality,” page 70.] What you see on the left side—“low” image quality—is where we were five years ago, and on the right side—“high” image quality—is where we are now. Today, you can digitize a glass slide using a 100∞ oil immersion objective and get spectacular high-resolution digital slide images. When it comes to focus, many regions in early digital slides were out of focus. Today, it is typical for well-prepared slides to have excellent focus across the entire slide. In the past, it was also common to adjust the colors in the viewing software to compensate for the properties of different monitors. Today, color standardization software makes it possible to automatically adjust the color of the images displayed on the computer monitor to match what you would see under the microscope. In the early days, vendors had difficulty stitching together the image tiles or image stripes, or both, that compose digital slide images. You should now expect digital slide images to be virtually seamless. If you see seams today in a digital slide, it’s usually a sign that there is something wrong with the scanning instrument. Compression has also evolved. We all started with JPG compression and have evolved to JPEG2000, which is a preferred form of compression in the sense that you can get better image quality at the same compression factor. I think it’s fair to say that pathologists today acknowledge that the quality of digital slides is more than adequate for most needs.

There has been significant progress in scanning speed. This is a combination of how long it takes to scan the slide and how much time it takes to process the imagery data and generate an image that can be viewed via a network. In 2001, scanning speeds were over 20 minutes. Vendors were claiming scanning speeds of 10, 12 minutes, but it took another 10 minutes to process the image. Today, scanning speeds are down to less than two minutes. Scanning speed improvements have been enabled in large part by Moore’s law, which says that the processing power of CPUs will double every 12 to 18 months.

An interesting metric to understand is the scanning cost per slide, which is not necessarily proportional to scanning speed. For example, if your fast scanner breaks down every 20 minutes and you have to have a technician sitting by the system full-time to make sure it keeps running, that will add to the cost of scanning slides. If you have a system that requires a lot of re-scanning because the first scan is not very good, you’ll have to pay a technician to do quality inspections and re-scans of your digital slides. If you have a system with a small capacity and need someone to keep loading the autoloader, you’ll spend more money than an equivalent system with larger capacity. It’s important to factor maintenance costs and technician time into an overall scanning-cost-per-slide metric. Scanning cost has fallen dramatically from about $5 a slide with first-generation scanners, where most of the cost was technician cost, to about 50 cents a slide today, where the dominant cost is equipment cost. The 24/7 throughput of today’s scanners has also increased dramatically to about 200,000 slides a year with one scanning instrument.

Digital pathology is adopted in niches, and the number of applications that are now supported continues to increase. [See “Supported applications,” page 73.] There’s education, CME, and proficiency testing—applications for which you don’t need a particularly fast scanner and you don’t scan slides very frequently. Some applications such as frozen section telepathology require higher scanning speed, but the number of slides that must be digitized is typically low. The challenging end of the spectrum is the scenario where all digital slides are read on a computer monitor. We call this application the manual read of digital H&E slides, and it requires fast scanning of large numbers of slides. The applications of digital pathology cover the continuum of scan speed and frequency. Many applications don’t require the hypothetical ‘sub-30-second scanner that costs virtually nothing,’ which is one of the reasons you’re seeing rapid adoption of digital pathology.

Image analysis is one of the drivers of adoption. Originally, image analysis was limited to particular regions of interest within a digital slide. You could look at a digital slide image, identify a small region, and say, I want to do a HER2 analysis here. Digital pathology image-analysis technology then evolved to support entire digital slide images, including support for image-analysis algorithms developed by third parties. Today, anybody can develop image-analysis algorithms that can be applied to an entire slide. We have also seen the advent of grid computing architectures wherein the processing requirements for digital pathology image analysis can, if desirable, be distributed among multiple PCs within an organization. Today, it is also possible to automate entire-slide image analysis by loading up racks of glass slides for overnight scanning and entire-slide image analysis. When you come to work in the morning, all of your glass slides will be digitized and the analysis results will be in a database. Or, if you want, you can select regions of interest ahead of time from volumes of digital slides and then initiate batch-mode analysis. These are examples of high-throughput image analysis—pharmaceutical companies that have adopted digital pathology are doing it.

Morphometric analysis is the simplest and most common type of image analysis today. It includes algorithms for protein expression—that is, ER/PR, HER2, and other immunohistochemical stains [see “Morphometric analysis,” page 74]. Many pathologists use it routinely to quantify what they see when they look through a microscope. Another type of image analysis—perhaps a bit more provocative—is something we call image quality, or IQ, analysis. One goal of IQ analysis might be to assess the quality of a digital slide. This would be useful as a basis for quantifying how well an autostainer is working. You could reject glass slides based on their image quality and say, ‘Do it again, because it doesn’t meet certain standards.’ IQ analysis also suggests that I can improve what can be seen through the microscope by doing some sort of image enhancement. Maybe I can improve the image quality of the digital slide I’m viewing in such a way that I can always see more on a computer monitor than through my microscope. Here’s an example of how to think about IQ enhancement. [See “IQ stain decomposition: colon cancer,” left.] The image in the upper left corner is a traditional H&E slide. Fundamental to the IQ enhancement technique is the concept of mathematically decomposing the traditional H&E image into an image containing only H and another image containing only E, shown in the lower left and right respectively. Based on the staining, you can then digitally enhance either image by boosting the amount of H relative to the amount of E or digitally increase or decrease nuclear detail. IQ enhancement has the effect of“re-staining” the entire digital slide image to provide an enhanced H&E image, shown in the upper right. There may even be benefit in examining the nuclear content of the hematoxylin-only image, without the clutter of the eosin staining. This technique is not limited to H&E images, but can be applied to IHC digital slides as well.

A more advanced form of image analysis is the world of pattern recognition, or CAD—computer-aided detection—where one goal might be to automatically highlight significant regions on a digital slide. Content-based image retrieval—useful for decision support—would fall into this category. When you think about it, CAD capabilities might seem a bit frightening. But CAD can be compelling if you think about the efficiency and quality needs we talked about earlier. For example, it’s always important to find tumor cell regions in an IHC specimen. [See “Pattern recognition/ CAD,” page 76.] In this example of a pattern-recognition tool that uses genetic algorithms, a paintbrush tool is used to mark the brown cells in green, and the background area is marked with a red color. After you just paint some cells, the segmentation and classification happens automatically and the tumor cells across the entire slide have been highlighted. That’s pretty powerful.

Let’s look now at the evolution of compliance and regulatory efforts. Our industry has seen tremendous progress in the emergence of standard file formats, interfaces, and color management. Progress has also been made in supporting organizations with Good Laboratory Practices and HIPAA compliance, and with obtaining FDA clearances for the use of digital pathology in selected applications such as HER2 image analysis.

Regarding the state of the digital pathology market, value propositions are now firmly established in several segments, particularly in education, biopharma, reference labs, and early-adopter hospitals. Early adopters are fueling demand for development. There is greater recognition that software is the primary source of value in digital pathology, enabled largely by the availability of high-quality scanners. There has been government-sponsored support in countries like Japan, Canada, and Spain, which have allocated significant sums to acquiring digital pathology solutions. Pathology leadership has recognized the importance of digital pathology. At the CAP’s FutureScape meeting, prominent pathologists talked about its importance. There is also an emerging awareness of its importance among LIS and PACS vendors.

On the supply side, companies, fueled by investors, are investing more time and effort in developing new and improved products. Vendors see the need to comply with standards and to obtain regulatory clearances. Companies are organizing for success, some by consolidating early players and others by business model innovation. These are all signs of a healthy industry. Adoption is well underway, with hundreds of systems being deployed in many countries. And there are different paths of adoption for different needs. You don’t have to take everything you’re doing today in analog—glass—and do it digitally tomorrow. You can start by adopting digital pathology for select niches and by deploying a system here or there. The future is exciting. Digital pathology is ready for prime time.