Digitizing Pathology

Turning glass slides into digital slides creates new analysis options.

The Department of Pathology at the University Medical Center (UMC) Utrecht processes some 300 to 500 histopathology slides per day, nearly 100,000 annually, each of which must be processed, scored, and—importantly—stored; if patients return to the hospital, their slides may need to be reexamined.

Oftentimes, though, that isn’t so simple. Glass slides can break or deteriorate over time. They can become lost; Utrecht’s archive holds some 7 million slides weighing 60,000 kg, and its 15 pathologists and 10 or so residents add another 600 kg annually. Perhaps most significantly, a slide can be under the lens of only one microscope at a time, creating logistical problems for consultations, conferences, and training sessions, and meaning, by extension, that if you mail off a slide for a second opinion, it’s gone until the colleague returns it.

To circumvent these issues, UMC Utrecht in 2007 began digitally archiving every slide it makes, says André Huisman, an information technology manager and researcher at the facility. To his knowledge, he says, “we are the only hospital in the Netherlands, and possibly in Europe or even the United States, that is digitizing slides on this scale.” That puts them firmly on the leading edge of a revolution that’s sweeping inexorably—albeit slowly—across the clinical landscape: Digital pathology.

What is digital pathology?
Fundamentally digital pathology is the digitization of microscope slides into image files called “digital slides”—the histologic equivalent of computerized x-rays. Yet according to Aperio CEO Dirk Soenksen, digitization is just the beginning. Digital pathology, Soenksen says, “is much more focused on the management of information than on the scanning technology for creating the digital slide.” Computer files can be analyzed, shared, archived, and compared in ways physical slides cannot.

At least 30 commercial systems exist, according to one 2006 study. Available from such firms as Aperio and BioImagene, MetaSystems, and Genetix, most are essentially automated brightfield microscopes in a box, though some, such as 3DHISTECH’s Pannoramic 250 and Genetix’s Ariol, handle fluorescent slides, too.

Whatever the platform, users invariably can view the resulting slides remotely (an application called “telepathology”), typically via a browser-based client often compared to Google Earth (that is, the software sends a low-magnification image that the user can quickly navigate and magnify). As a result, whether they work on a different floor in the same hospital or thousands of miles away, pathologists can make diagnoses and consultations as if in front of a microscope, all without endangering the glass slides themselves.

Pathologists at Toronto General Hospital (TGH), site of Toronto’s University Health Network (UHN) pathology department, have been using telepathology to interpret frozen neurosurgery sections at UHN’s Toronto Western Hospital (TWH) since 2004, says staff pathologist Andrew Evans. With TWH generating just two to four frozen section cases daily, it made no sense to station a pathologist there or have TGH pathologists drop everything to trek cross-town on demand, Evans says: “Telepathology allowed us to be in two places at one time.”

TGH also provides telepathology services for hospitals in Timmins and Sault Ste. Marie, some 400 miles away. In Timmins, Evans says, the need was particularly dire: The city has one pathologist servicing 11 hospitals, and some weeks, no on-site pathologist at all.

“Without telepathology, surgeons would have to schedule surgeries requiring frozen sections around the availability of the pathologist or rely on their own intra-operative judgment,” Evans says.

To date, UHN pathologists have remotely read some 1,600 frozen sections from 1,300 patients, Evans estimates. Initially, they used a custom robotic system based on a Leica TPS2 microscope to handle the workload. But in 2006, UHN switched to a more turnkey solution. Today, all the main UHN hospitals, plus those in Timmins and Sault Ste. Marie, use five-slide Aperio ScanScope CS scanners, says Evans, who coauthored a 2009 article detailing UHN’s experience. As a result, time from receipt of tissue from the surgical service to diagnosis has dropped from 20 minutes to 16 minutes, and slide interpretation time from 10 minutes to three minutes.

UMC Utrecht also uses Aperio hardware. The center’s three ScanScope XTs, each capable of digitizing 120 slides every six hours, run nearly 24/7, says Huisman. But Utrecht pathologists don’t yet make diagnoses from digital slides, he adds; “This would require scanning all slides before they leave the lab. For this the scanners are too slow.” Scanning occurs later, for archiving, education, and consultation.

Naturally, data can become overwhelming. At the Burnham Institute for Medical Research in La Jolla, Calif., one ScanScope XT, servicing over 150 users from as far away as China, burns through 1 TB per month, says Stan Krajewski, director of the Molecular Pathology Core. At Utrecht, files average 400 MB apiece, requiring about 200 GB daily (50 TB annually), Huisman says. The facility has some 130 TB of storage, but expansion is inevitable. “We assume storage costs will halve each three- or four-year replacement, so that we can double the storage for less,” Huisman says.

Image analysis
For cancer biologist William Gallagher of the School of Biomolecular and Biomedical Science at University College Dublin, digital slides help screen potential biomarkers. Gallagher’s process involves immunohistochemical staining of tissue microarrays containing hundreds of samples per slide—a high-throughput approach that saves time, tissue, and reagents. It also makes for tedious microscopy, as each core must be scored individually.

“Humans are very good at pattern recognition,” Gallagher says. “But computers are quite useful in quantitation.” Computers can easily distinguish hundreds of subtly different colors; humans cannot. As a result, pathologists manually score IHC assays by sorting into relatively crude, yet clinically significant, bins: 0, 1+, 2+, or 3+ staining, for instance. The trouble, says Gallagher, is, “Different pathologists may score the same sample differently, because it is a subjective decision, and what is 3+ for one may be 2+ for another.”

Software, though, can provide more nuanced, objective data. “A lot of manual pathology is focused on having to count a number of objects in a field of view as well as making calculations of how intensely stained particular objects are. Algorithms can do these tasks more consistently than a human can,” says Vikram Mohan, senior director of marketing at BioImagene.

BioImagene is one of several companies offering such algorithms, a handful of which (including assays from BioImagene, Aperio, and Abbott Laboratories) have been approved by the US Food and Drug Administration for use in the clinic. But there’s more to the problem than counting, says Elizabeth Chlipala, manager of Premier Laboratory, a Boulder, Colo.-based contract lab that offers digital pathology services. “When you scan an entire slide, the tissue doesn’t necessarily have only what you are interested in, it has more on it. You have to figure out a way to analyze only what you want.”

Tools like Definiens’ Tissue Studio and Aperio’s Genie software can identify regions of interest within a background of irrelevant material. Gallagher, who coauthored a 2008 review of automated image analysis systems in Expert Review of Molecular Diagnostics, prefers a custom solution.

“We have developed a self-learning approach that learns from the specimens what a tumor cell should look like, and can accurately quantify protein biomarker expression in different compartments within the cell,” says Gallagher. His team has validated IHC-Mark on “over 10 different cancer types,” he adds, including breast, colorectal, and lung.

HistoRx applies a different strategy. Its AQUAnalysis software relies on fluorescently labeled antibodies to distinguish tumor from normal cells, and cytoplasm from nucleus. It then quantifies, using a continuous variable called an AQUA score, the marker’s intensity within the tumor in each cellular compartment. In one proof-of-concept study, Yale University pathology professor (and HistoRx co-founder) David Rimm differentiated colon cancers based on their ratio of nuclear-to-membrane beta-catenin staining, an analysis that, the authors note, “is essentially impossible without continuous scoring.”

“Protein expression doesn’t follow a grid,” explains Jason Christiansen, HistoRx’s Senior Director of Operations, “you want to be able to measure on a continuous scale, like cholesterol or blood glucose.” And that, he adds, can pay dividends for personalized care via more precisely defined patient populations. “It’s not just positive-negative, there’s more nuance in there.”

In transition
According to Soenksen, pathology today is a field in transition. “I would say the vast majority of pathologists would say the field is going digital,” Soenksen says. “It’s really just a question of, when is my institution going to do that?”

Now, more and more companies are trying to help them make that decision. The newly formed Digital Pathology Association lists 17 corporate sponsors, including Leica, Olympus, Hamamatsu, and Definiens. Philips recently entered the field, as has Omnyx, a joint venture of GE Healthcare and the University of Pittsburgh Medical Center.

Omnyx CEO Gene Cartwright says, despite the competition, there’s room for innovation. Existing systems, Cartwright explains, are used, and really were designed, mostly for research and drug development. A clinical system must meet different standards of regulatory compliance, throughput, scalability, and IT integration, he says. “We are talking about a different sort of scale, a product designed from the ground up to be a complete hospital-wide digital clinical application.”

Regulatory approval, in particular, remains an issue. Though several IHC protocols have been FDA-approved, all are prognostic; no company has received clearance to automate the routine work of reading H&E-stained slides—a critical limitation given a worldwide shortage of trained pathologists. But that could be changing: FDA took up digital pathology in a October 2009 panel discussion, and will likely receive its first submissions soon; Ole Eichhorn, Aperio’s Chief Technology Officer, says Aperio plans to file this year. Should it or its competitors succeed, the entire industry will benefit, says Eichhorn. “It will put a certain imprimatur on this, that it really is safe and effective.”

This article was published in Bioscience Technology magazine: Vol. 34, No. 2, February, 2010, pp. 1, 8-12.