Growing Role for Digital Image Analysis in Pathology

Digital pathology systems used at Washington University for three years

CEO SUMMARY: For more than three years, pathologists at Washington University in St. Louis have worked with several different scanning products and digital pathology systems. Step-by-step, the Pathology Department has learned important lessons in how to capture digital images, archive them, then make them available on demand to the pathologists. WU currently captures images from about 25% of its surgical pathology cases and is making progress with the use of digital image assessment (DIA) to advance patient care.

DIGITAL PATHOLOGY IS EXPECTED TO BE both transformational and disruptive to the profession of anatomic pathology. For many surgical pathologists, this is not welcome news.

After all, such pathologists have decades of experience in the use of glass slides and microscopes. In this country, these are the tools that pathologists use to deliver a diagnostic service that is the envy of the world.

But, for many of the right reasons, digital pathology is poised to become the technology platform that allows pathologists to further advance the field of laboratory medicine. Use of digital whole slide images (WSIs) and sophisticated software algorithms will become essential complements to the skills and scientific savvy of surgical pathologists. Physicians and patients will benefit from the increased sensitivity and accuracy that results from use of these rapidly-evolving technologies.

To gain an early peek into this brave new world of digital pathology, THE DARK REPORT recently caught up with one of the nation’s most respected academic pathology departments. For more than three years, the Pathology Department at the Washington University School of Medicine (WUSM) in St. Louis, Missouri, has used the digital pathology systems and products from several different manufacturers. Few pathology labs in the world have the hands-on experience found at WUSM that comes from working with the digital pathology systems from different vendors.

At Washington University, pathologists are bullish on the future of digital pathology. “From this point forward, digital pathology will definitely play an expanding role in laboratory medicine,” stated John Pfeifer, M.D., Ph.D., who is a Professor of Pathology and Immunology at WUSM. He is also the department’s Vice Chairman for Clinical Affairs and a Professor of Obstetrics and Gynecology. “But digital pathology is a developing technology, which means it has limitations.”

In this exclusive interview with THE DARK REPORT, Pfeifer discussed at least four major insights that he and his colleagues have developed from their work with whole slide imaging and digital pathology systems.

First, the WUSM pathologists have come to better understand how the human mind works—given its evolutionary development— and how digital pathology can work in complement with the pathologist’s intellect and clinical experience.

Second, WUSM pathologists are learning how to use “digital image assessment” (DIA) to extract more information from the whole slide image. In turn, this provides enriched diagnostic information to the pathologist and opens the door to a more sophisticated analysis of the WSI.

Third, WUSM pathologists are expanding the capabilities that WSI and DIA can deliver in support of the pathologist. But they recognize that the current state of the technology will channel use of digital pathology toward specific functions.

The fourth insight relates to the performance of the current generation of digital pathology systems.

In discussing the first insight, Pfeifer noted that it is important to know how the human brain works with images. He offered a Darwinian metaphor about the human mind, including how human eyes “see” and how the human brain processes those images. This is a necessary starting point for the reader to understand Pfeifer’s predictions about how the pathology profession will take up and use whole slide images and digital pathology systems.

Human Visual Processing

“Bear with me for a moment,” asked Pfeifer. “Think about the human hunter, crouching in the jungle. He strains to distinguish his prey from the dense, leafy background of his surroundings. This is how human visual processing evolved over hundreds of thousands of years.

“At this time—when we were all hunters in the wild—our visual processing system did not evolve around the coat color of our prey in the wild,” he observed. “Human hunters had very few discussions about whether the color of the animal’s coat was sandy or brown—or whether the intensity of the hue was more red than yellow.

Seeing Patterns, Boundaries

“Instead, our visual system evolved around contrast boundaries and pattern recognition,” explained Pfeifer. “Humans are hard-wired to see the edge of a grass blade or the outline of a tree branch so that an animal crouching behind it can be seen.

“The point is that human brains process information based on the way our visual system evolved to help us recognize animals hiding in the grass and not by distinguishing between the color or the intensity of the grass itself.

“Accept this premise about how the human eye ‘sees’ and how the human brain processes that image, and it becomes easier to understand why digital pathology systems can function in ways that expand the pathologist’s ability to tease out all types of clinically-useful information from the whole slide image,” he continued. “More specifically, this is where the role of digital image assessment (DIA) comes into play.

How Pathologists “See”

“Let’s take the example of quantitative or semi-quantitative analysis of immunohistochemical stains and how pathologists process this information,” Pfeifer noted. “I have a collection of images showing that—if you mask them in certain ways— you would call the color on that image a five on a scale of one to 10 because it is in the mid range.

“However, were you to mask that same color in a different way—by using a vertical mask or a horizontal mask—you might say it is only a two out of 10,” he commented. “It is the same intensity of color. Nothing else has changed except the mask and it is the mask which affects how our brains process the information.

“In general, we tend to think that pathologists are very good at identifying something that’s negative and when something is intense, pathologists call it intense,” he added. “But in the middle range, there is a lot of variability between pathologists.

“More specifically, it is becoming clear that there is significant variability between each pathologist day-to-day,” continued Pfeifer. “Lighting conditions and many other variables, including the quality of the imaging systems and microscopes, can contribute to this variability.

“The point is that pathologists are not very good in the middle areas because humans are not very good in the middle regions,” Pfeifer said. “That’s why digital image assessment has the potential to help pathologists in the middle regions.”

Hard-Wired to Be Correct

Pfeifer now discussed the second important insight gained from the use of digital pathology systems at Washington University. “DIA is an improvement because it is hard-wired to be correct,” he stated. “Compared to humans, DIA can be much better at distinguishing differences in the middle areas because DIA can be programmed both: 1) to recognize the prey crouching in the grass; and, 2) to precisely and objectively measure the color and intensity of the hue of the grass.

“DIA systems offer another benefit,” added Pfeifer. “They can be standardized in ways that other instruments, including microscopes, cannot.

“We now have reached the point in the conversation where pathologists become uncomfortable,” noted Pfeifer. “That is because we are talking about buying machines to do what pathologists do.

“While that idea may make pathologists uncomfortable, having machines do this work produces two distinct advantages,” he observed. “First, anyone looking at a substantial number of HER2/neu immunostains would welcome having a computer to help them do this work. After all, it can be repetitive and uninteresting.

“Second, pathologists using a digital pathology system will continue to issue a report based on their analysis of the specimen,” he affirmed. “They will use laboratory techniques to get the right answer and will continue to bill for the technical and professional components of this work. The only difference is that computer image analysis will help the pathologist make the analysis.

“Just as they do now, pathologists must still identify the area on the slide to be scanned and determine if specimen is positive for HER2/neu, for example,” he said. “The pathologist remains essential to the diagnosis. It’s just that computers will contribute more precision and accuracy to the pathologist’s findings.

“These are the reasons why whole slide imaging is a market niche poised to grow rapidly,” observed Pfeifer. “When DIA becomes mainstream, it will become possible to sell digital slide scanners to all of the 4,800 hospitals nationwide because they will need it to do standard pathology.

“There is a big difference between writing image analysis software that can quantitate a level of immunostaining and writing software that can recognize a pattern of cells and tell you whether it is cancer or not.”

“Currently, digital pathology is most commonly used in niche applications,” he noted. “WSI has a role in conferences, for telepathology, and for moving images across cities or continents. Here at Washington University, we continually learn more about how to use WSI and digital image analysis in ways that improve our analysis of specimens.”

Next comes the third insight about digital pathology. In their work with digital pathology systems, pathologists at Washington University believe that WSI and DIA will have a very defined role. Pfeifer explained that “there is a big difference between writing image analysis software that can quantitate a level of immunostaining and writing software that can recognize a pattern of cells and tell you whether it is cancer or not.

“For example, if you work at it, you can develop DIA specific to defining one disease, one kind of condition, and one type of specimen, such as with Pap smears,” he noted. “But that’s a long series of ifs: if you have one disease, if you have one type of stain, and if you have a standardized preparation. Were you to set up the DIA to handle all these boundaries, you can make it work.

Developing DIA Solutions

“The practical challenge for DIA pathology is the fact that thousands of biomarkers have been identified, many different stains are available, and the tumors have similar patterns but the background cells have a lot of variety,” observed Pfeifer. “It becomes a complicated process to develop a DIA application for each one of these unique combinations. For this reason, I don’t see DIA being used to support automated diagnosis by computer. That may be decades away.

“On the other hand, I do see DIA being used for quantitating immunostains,” he said. “This is what pathologists do with telepathology. A glass slide is scanned and the image is put on a computer screen. The pathologist then makes a diagnosis off that image on the com- puter screen. That’s different from taking a scanned image and having a computer make a diagnosis based on that image.

Building into the Future

“That is my take on the future of DIA,” said Pfeifer. “However, remember, the field of digital pathology is in its earliest stage in terms of the evolution of image analysis. At our lab, we have only three years of experience at scanning slides. Every day, we learn more about how these systems work and about some of the challenges they present to pathologists.”

Next, Pfeifer turned to practical lessons learned by the pathologists at Washington University. “There are certain realities with introducing a slide scanning system,” he said.

“The first lesson is that it requires a significant capital investment,” advised Pfeifer. “We have slide scanners from Aperio Technologies, from BioImagene, and from Trestle. Since we work with a number of different vendors, we are not a company store.

“We also work with multiple informatics and systems vendors,” he added. “As vendors, Cerner Corporation, which supplies our LIS, and Aperio Technologies, Inc., which supplies Spectrum hardware and software, have been particularly responsive in helping us adapt their products to our specific needs.

“These vendor collaborations played an important role in helping us learn how to get the most out of their software and instrument systems,” noted Pfeifer. “It is a reminder for all labs that one key to success is to have an effective, ongoing relationship with the vendors who supply information systems and analytical instruments.”

Expense of Digital Pathology

Digital pathology systems are expensive. “Regardless of vendor, these systems cost about $100,000 to $200,000 or more,” continued Pfeifer. “We estimate that the working life of these machines is about five years. Also, once installed, they are not free to operate.

“For a private pathology group, it will take as much as $250,000 to acquire the scanner and digital pathology system,” explained Pfeifer. “There is a need to hire at least one person to care for the machine. This covers the cost to acquire and to oper- ate the digital pathology system.

Cost to Store Digital Images

“It is important to also budget for the cost of storing the scanned images,” he added. “We calculate that, based on our analysis of storage costs, it costs about 25¢ cents to put each image on a server somewhere. Since we will scan between 20,000 and 25,000 images this year, we will spend about $6,000 to $8,000 to store our digital pathology images on a server rack.

“One reason storage is so expensive is that these are very large image files,” he said. “The size of the files depends on the region of the tissue being analyzed. The size of a file at 20x magnification could be 300 to 550 megabytes and at 40x, it could be 1 to 1.5 gigabytes in size. Those are huge image files.

“In addition to the capital costs of the machine, the personnel costs, and the direct cost of storing the images, a key component of the digital pathology system is the software that allows you to store the images, then retrieve them for viewing by the pathologists,” stated Pfeifer. “Because it is software, your pathologists will be required to learn how to use it.

Lessons Learned

“This brings up the question of using off-the-shelf software versus customizing software that is tailored to the specific needs of your practice environment,” he said. “Digital pathology vendors are willing to work with you in your practice environment, but there may be interfaces or some custom code that needs to be written and that will cost money. The benefits are significant because of improved workflow and the pathologist’s ability to electronically move information to colleagues or referring physicians.”

It is unusual for any pathology department or pathology group to buy and use multiple brands of scanners and digital pathology systems. Thus, the experience of Washington University pathologists over the past three years offers a number of unique insights and valuable lessons.

First, for a digital pathology system to be useful, the pathology laboratory needs to accomplish two things with equal proficiency. One is to establish a productive workflow to acquire the pathology image, archive it, and make it available on demand to the pathologist.

The other is the importance of an integrated suite of pathology informatics capabilities that support both the operational workflow in the lab and the pathologists’ diagnostic practice patterns in a seamless manner.

Working with Digital Images

It is noteworthy that WUSM pathologists spent the resources necessary to allow pathologists to access and view digital pathology images in ways that were consistent with their individual practice preferences. Once this capability was in place, 100% of the department’s pathologists began to regularly work with digital pathology images.

Another noteworthy insight is how digital pathology is now an essential component of the WUSM pathology department’s consultation service. For subspecialist pathologists, this development points to one valuable way that use of digital images and digital pathology systems can expand their network of referring clinicians—and the revenue that comes with these additional consultations.

Because They Tend to Be ‘Lone Wolves’, Pathologists May Need Custom Software to Work with Digital Images

INSTALLING A DIGITAL PATHOLOGY imaging system in a large academic medical center may require custom software to ease the transition to the new system for all pathologists on staff,” stated John Pfeifer, M.D., Ph.D., Professor, Pathology and Immunology, at Washington University School of Medicine in St. Louis, Missouri. When the Pathology Department at the School of Medicine wanted to get the most from its digital imaging systems, it needed custom software.

“We have about 25 to 35 people who sign out pathology cases,” he explained. “And academic pathologists are lone wolves. Each one has an individual style. Some are early adopters and some are never adopters. This makes it very difficult to tell these people in academic medicine that they need to do something one way if it’s difficult to work that way. For our practice, it meant we had to develop various interfaces to make the digital pathology system work for each of our pathologists.

“Clinical labs have a laboratory information system (LIS) to open up reports, sign out reports, and issue reports,” continued Pfeifer. “But for digital imaging, another software program may be needed with separate passwords and log-ons for viewing scanned images.

“It’s likely that the second program will run in the background— but it will require a second monitor on the pathologist’s desk,” he noted. “This change in work practice might mean some pathologists will not use the ability to view images digitally.

“To solve this problem, we worked with our digital imaging and LIS vendors to write separate interfaces for our pathologists,” explained Pfeifer. “Now we have a system that everyone can use. So if the pathologist is using Cerner CoPath to write a report, he or she will see a tab in the background.

“Clicking the tab instantly opens an interface to the digital imaging software. A second mouse click brings up the specific pathology image,” stated Pfeifer. “To write this interface cost about $50,000 but it was worth the expense because it established seamless work flow on the desktop for our pathologists.

“Here’s how it works,” he explained. “When a pathologist needs to sign out, for example, a case of an excision, the system will tell him or her if there is a previous case for this patient. By clicking a tab, the pathologist sees—in half a second—that four slides from that previous case are available to view digitally. By clicking on a slide, the system opens the digital image to allow the pathologist to navigate to it.

“This means that pathologists use just one monitor and log into only one system each morning,” Pfeifer said. “They don’t have to remember different user names and passwords for the different systems. It is seamless. Before we got this interface, there were about three of us who were willing to work without it. But now that we have this interface, everyone uses it, and we all love it.”

 

Digital Pathology Systems Help Boost Academic Center Role for Patient Referrals

ONE USE OF DIGITAL PATHOLOGY SYSTEMS at the Washington University School of Medicine in St. Louis is to support the academic center’s role in patient referrals. John Pfeifer, M.D., Ph.D., Professor, Pathology and Immunology at the Washington University School of Medicine, in St. Louis, explained how the university uses whole slide imaging (WSI).

“Before acquiring image scanners and digital pathology systems three years ago, we rigorously evaluated how we would use digital pathology and where it would add value,” he recalled. “We discovered it could help us in several places, such as consults where institutions want their pathology slides returned after the consultation.

“We need a permanent record whenever a case comes here,” explained Pfeifer. “We review the case, issue a report, and maybe two months later, the patient ends up coming here for definitive therapy.

“At that time, we can compare our archived digital image of what was on the definitive excision from what was there initially,” commented Pfeifer. “This is especially important in those cases where there is a difference. Use of digital images enhanced our clinical practice by allowing us to have a permanent record of these slides. This is a big deal, because there are many consults and everyone wants the slides returned to them.

“The other need we have is that patients are often presented at conferences even before they come to our academic center for definitive care,” he said. “We need digital images to project at those conferences.

“One area that is growing in volume and which makes digital imaging incredibly helpful is ancillary testing modalities that result in the destruction of the slide,” Pfeifer stated. “Many molecular testing paradigms result in destruction of the sample.

“When this happens, the question comes up about whether there was a specimen mix, or a contaminant, or perhaps it was very small tumor that was present only on the biopsy,” explained Pfeifer. “When you collect tissue off that slide, it is obviously not going to exist in your files anymore. So we get a whole slide image of the slide and archive that image in our database.

“Then we can destroy the slide as part of testing because we have an electronic image of the slide,” he added. “That is incredibly helpful because we used to fret about the destruction of these samples.”

 

University Hospital’s IT Director Outlines the Nuts and Bolts of Digital Pathology

IT WAS 2008 WHEN THE FIRST DIGITAL PATHOLOGY SYSTEM was put into use by the Pathology Department at the Washington University School of Medicine in St. Louis, Missouri. “It took our information systems department about nine months to build the capability and two months to implement the informatics infrastructure to support digital pathology,” recalled Mike Isaacs, the Director of Information Systems, in the Department of Pathology & Immunology.

“In the first year, while everyone was learning how to work with digital pathology images, about 4,000 cases were scanned,” he said. “Last year, that number increased to 9,500 cases. For 2010, we are on pace to scan about 12,000 images, which is about 35 cases scanned per working day this year.

“In the beginning, only different research projects and study sets were scanned,” Isaacs noted. “Next, we began scanning slides to use in conferences.

Interface to Patient EMR

“One factor that limited the utility of digital pathology images was that our pathology informatics platform was not yet linked to the patients’ electronic medical record (EMR),” he explained. “The digital pathology images resided in a Spectrum database but were initially not accessible from the Cerner CoPath laboratory information system (LIS). In 2009, we established an electronic link to the patients’ records, went live, and did about 9,500 scanned images.

“It is important to know that, in 2010, the 12,000 scans represent just a small percentage of the total number of pathology cases, as well as about one quarter of the 42,000 surgical cases handled by the Pathology Department in 2009,” noted Isaacs. “However, it is important to note that, in that 12,000 number is about 95% of all of our inside-outside consult cases.

“Remember that many of these cases will generate about 10 slides each,” continued Isaacs. “Pathologists typically select two or three of the diagnostic slides for the digital image archive. Because the images are so big, before scanning slides, it is important that there be real benefit for later diagnostic purposes.

No Workflow Bottlenecks

“As mentioned, these are large images,” observed Isaacs. “We are set up to scan about 1 gigabit per second from the scanner to the server. That is very fast—meaning there are no workflow bottlenecks as the pathologist waits for the system to pro- duce the needed images.

“Another lesson is the need for one full-time person to manage the digital images,” he stated. “In our arrangement, two people work part time and each does different aspects of the workflow. One gets the cases when they are scanned, reads the bar code to identify the patient and the test (such as the H&E slide), and then sends it to the Cerner CoPath LIS, where the system matches the bar code to the patient.

“Our second part-time staff member loads the scanner itself and chooses the correct area of the slide to scan,” Isaacs said. “The goal is to avoid scanning the entire slide, where appropriate. If the region of clinical interest is smaller, the digital file of the scan is smaller.

“Once a digital image is captured by the scanner, this staff member checks the focus for the image,” concluded Isaacs. “All these steps show that some staff time is required to manage the capture, storage, and distribution of digital pathology images.”

Tags: , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

Enter Your Login Credentials
This setting should only be used on your home or work computer.

×

Send this to a friend