View master—an expert eyes digital
pathology’s future

September 2007
Feature Story

At the June Futurescape of Pathology CAP Foundation conference, Ronald S. Weinstein, MD, shared his thoughts on the barriers to adopting new technologies in pathology and pathology education. Dr. Weinstein is professor and former head of the Department of Pathology and director of the Arizona Telemedicine Program at the University of Arizona College of Medicine, Tucson. He is a co-inventor of an array microscope that decreases the processing times of virtual-slide processors and a co-founder and stockholder of DMetrix Inc. and UltraClinics Inc. His remarks follow.

Twenty years ago I was an invited speaker at a CAP Foundation meeting on new technologies. I used the term “telepathology” there for the first time at a professional conference. I described our then current work on the design of the first robotic telepathology system and presented our initial data on its diagnostic accuracy. I also introduced the concept of telepathology diagnostic networks. Speakers at that meeting were invited to make predictions. I proposed, among other predictions, that telepathology services might become widely available within a few decades. I also suggested that new pathology practice models could emerge that would incorporate telepathology. Virtual-slide telepathology was barely on the horizon back then; nevertheless, a number of those 1987 CAP Foundation meeting predictions have been validated. Once a so-called futurist makes good predictions, it’s tempting to take that person too seriously the next time around, so be forewarned.

The organizers of this CAP Futurescape meeting have invited me to make another set of 20-year predictions. I’ll take a shot at it in the form of two sets of predictions: one set with a near-term horizon (one to 10 years into the future) and a second set in the form of long-range predictions (11 to 20 years). My near-term predictions are on reasonably solid ground since the enabling technologies, and some of the practice models, are already being tested. The long-range predictions take us farther out on a limb and require leaps of faith.

For this presentation, I’ve been asked to focus my comments on reality checks on the status of digital imaging and virtual-slide telepathology implementations, specifically in education. I’ll consider several questions: Is the ground being laid for whole-slide digital slides (that is, virtual slides) to eventually replace glass slides in educational programs? Is virtual-slide telepathology diffusing from early adopter institutions at a significant rate? What is the current level of acceptance of digital pathology within academia? within broader constituencies? Is organized pathology embracing the technologies? Will this diffusion rate of the technology accelerate or stay constant?

For my long-range predictions, I will deal with a much more speculative area, namely the potential role of the neurosciences, including cognitive psychology, in defining pathology expertise in the future. Will we be using digital pathology imaging technologies, including telepathology virtual slides, to improve training strategies? Can advanced approaches, such as eye movement analysis and functional MRIs, be used to create profiles for certification of pathology subspecialty experts? Can we accelerate the transformation of novices into experts using neurosciences-based approaches to teaching and learning? Can we actually identify “biomarkers of expertise”? Can quantitative human performance data be used to help triage surgical pathology cases to the best pathologist available on diagnostic networks to sign out cases? At the University of Arizona, we are beginning to explore these questions, and our preliminary findings are promising and provocative.

For my first reality check, I’ll begin with an update on implementations of telepathology virtual slides, which James Bacus, PhD, pioneered in the early 1990s. Today, virtual slides are being used in educational programs at literally hundreds of institutions around the world. More than 250 virtual-slide scanners have been installed in the past five years, up from a handful, and the rate of installations is on the upswing. The range of applications of virtual slides is also expanding. Venture capitalists have spotted the virtual slide industry as an area of interest—generally a good sign. Digital imaging telepathology appears to have reached its tipping point for education and may become ubiquitous over the next decade. Virtual-slide telepathology is finally a growth industry.

Our experience with virtual slides at the University of Arizona provides us with a personal reality check. Our College of Medicine committed itself several years ago to renovating its outdated glass-slide-based histopathology laboratories. It invested an estimated $1.7 million in the project. Four 1970s student histopathology laboratories were gutted and retrofitted with interactive DMetrix virtual-slide training facilities. At Arizona, we have 116 medical students per class in Tucson. Two labs are used by the first-year class and the other two by the second-year class. The DMetrix virtual-slide laboratories became operational in August 2006, and our College of Medicine has used them successfully for an entire academic year.

In each virtual-slide laboratory, students work in so-called pods. Ten pods are arranged around the periphery of each virtual-slide laboratory, each pod seating six students. The students have shared annotation input devices, tabletop space for their own wireless networked personal computers, and a high-quality, wall-mounted plasma screen. At the center of each laboratory is a laboratory instructor’s podium equipped with a large plasma screen on which the instructor can view a gallery of the video images displayed on the 10 student pod plasma screens. Each laboratory instructor, typically a staff pathologist, can import images from individual student pods and redistribute them to the other student pods. Students can also import examples from their own laptop computers onto their pod screen and then have their digital images redistributed to the entire class. Virtual-slide annotations, created by a student or lab instructor, can be shared with the entire lab. The flexibility of the system is remarkable and the system is user friendly.

Medical student performance, using the system, has been excellent. Student and faculty acceptance and interest have been high from the outset. We quantitated student acceptance of virtual slides and found that the students prefer to take their practical examinations using virtual slides in place of the traditional glass slides. Our pathology department staffing of laboratories is more efficient since our staff pathologists can interact with many students at a time, having shed the constraints placed on faculty members demonstrating histopathology features with the students’ conventional light microscopes in the past. It appears that subspecialty pathologist time in the student laboratories may be better used as well.

How long did it take the University of Arizona to convert from using histopathology glass slides to virtual slides in our medical student laboratories? Once construction of the laboratories was completed and the students’ glass slide study set had been converted into DMetrix virtual slides, implementation took a matter of weeks. Individual students imported the virtual slide files onto their personal computers in less than an hour. Use of the DMetrix virtual-slide viewer is very intuitive.

Today, the University of Arizona Department of Pathology uses virtual slides in our hospital laboratories for quality assurance programs, to support a same-day breast care service, and for providing second opinions. In the future, we plan to use DMetrix virtual slides for an international telepathology consultation service.

My near-term predictions regarding digital pathology and virtual slides in education are as follows:

Virtually all medical school courses will abandon traditional glass slides for teaching histopathology within five years. In addition, all medical students will have access to many other medical schools’ virtual-slide teaching sets over the Internet.
Resident training programs will incorporate virtual slides into their curricula at an accelerating rate. Board-certifying examinations and in-service proficiency testing will be largely in the virtual slide format. The American Board of Pathology has already incorporated virtual slides into its primary certifying examination, after carefully validating the suitability of virtual slides for “high risk” examinations. Pathology professional organizations have also incorporated virtual slides into their education programs.
Students, at all stages in the medical education process, will have relatively easy access to training from top subspecialists from other institutions over the Internet.
The need for institutions to have in-house pathology subspecialists could decrease, an unexpected change. Academic institutions could potentially save money in the future if they share the services of subspecialty pathology service providers using virtual-slide telepathology. This might be in the form of new pathology service consortia offering cost savings, access to a broader range of specialists, and improved cross-coverage.
There will be few, if any, technical barriers to implementing large virtual-slide diagnostic networks in the future. At the University of Arizona, we have operated for years a large statewide telemedicine network, offering 60 telemedicine services including telepathology. The Arizona Telemedicine Program has handled 500,000 telemedicine cases from dozens of rural sites over its internally operated telecommunications network. The program has been administratively housed in the Department of Pathology. Others can do it as well.
Reimbursement for educational services could erode in a competitive market. Subspecialty pathology faculty members may want to consider joining, and then work on expanding roles for themselves in, educational consortia.
Institutions with internationally recognized brand names—for example, Mayo Clinic, Harvard, Johns Hopkins—may choose to increase their use of their own “marquee pathologists” to market their institutions and their laboratory services.
Pathology education and training will become a global enterprise with the United States in a leadership role as an exporter nation for educational programs.

Now for my 11- to 20-year predictions regarding digital pathology and telepathology virtual slides.

The University of Arizona has a multicampus, interdisciplinary Laboratory of the Future research-and-development program. It encompasses teaching laboratories, hospital laboratories, and reference laboratories. Based on our experiences thus far, we make the following long-range prediction: Neurosciences will have an expanding role in the development of next-generation education and testing programs.

Our work on the Laboratory of the Future program is in three main areas at this time: 1) human performance assessments including a search for physiologic measurements of what we like to refer to as biomarkers of expertise; 2) facility design, construction, and testing for what we envision to be the Laboratory of the Future, a decentralized enterprise based on a multilaboratory consortium; and 3) the creation of advanced workflow management software and systems for use in a next-generation laboratory command-and-control center. We are modeling such a center now in Phoenix at the new branch campus of the university’s College of Medicine. Some of the design features of our Laboratory of the Future are based on discoveries made recently in our human performance labs.

Our 11- to 20-year predictions involve the neurosciences and cognitive psychology. Our hypothesis is that the scientific methods of the neurosciences will be applied to the analysis of pathologists’ mental processes, including learning skills. Our focus on the neurosciences and cognitive psychology stems in part from two of our current interests at the university: scientific differentiation of novice versus expert performance and the use of command-and-control systems for laboratory workflow management. With respect to expert performance in pathology, managing the use of expert pathologist second opinions in decentralized virtual group practices will have its challenges. The following will require much effort and thought: developing an infrastructure for profiling experts’ competencies, recruiting and selecting experts for specific case assignments, tracking their availability and performance preferably in real time, assessing experts’ performance on an ongoing basis, and setting a value for their services, factoring in case complexity and degrees of difficulty, as a basis for compensation. In Arizona, we have been working our way through these complicated issues for several years. Progress has been made, especially in facility design and testing and in workflow management software development.

With regard to the novice versus expert topic, expertise per se in subspecialty surgical pathology is often ill-defined and can be highly subjective. It is well known in academic circles that well-recognized experts not infrequently disagree on difficult diagnoses. Although cell and genetic markers can resolve some of the diagnostic issues for certain types of surgical pathology cases, we have a long way to go in resolving significant interobserver variability issues that seem to hamper our field.

What would be the best criteria for certifying the designation “expert”? Should the designation “expert” be based on numbers of cases of a certain type of pathology handled per year, training pedigree, the prestige of the expert pathologist’s institution, scholarly publications, or invitations to speak at state and national pathology organizations? Generally, we do know whom we cite as authorities in the field and would recommend for case referrals. Pathologists usually know whom to send difficult cases to for second opinions. Lawyers know which experts might add additional authority to their arguments. But a precise definition of the surgical pathology expert can be elusive.

Traditionally, some of the training of pathology fellows by experts has been done in an apprentice mode, using a multiheaded light microscope to simultaneously view histopathology slides. This doesn’t guarantee that both participants are examining the identical features in a histopathology section at the same time. Cognitive psychologists, using eye position tracking devices, can show that the mentor and trainee viewing activities are not identical.

At Arizona, we have a program using cognitive psychology research methods to study surgical pathologists’ performance. One goal is to find out how to accelerate the progression of a junior faculty member—a novice—into a bona fide expert pathologist. We are also exploring the possibility of using scientifically validated physiologic parameters, such as eye movements, to stratify levels of surgical pathology expertise.

Elizabeth A. Krupinski, PhD, a distinguished cognitive psychologist in the University of Arizona Department of Radiology, started our program in 2005. Dr. Krupinski has had a career-long interest in eye movement studies. She is well known for her work documenting the search strategies of radiologists viewing digital mammograms. Her research involves the use of accurate and sensitive eye-tracking equipment.

The availability of virtual slides has made it possible to apply the eye-tracking technology used in radiology clinical research to pathology, since pathologists viewing virtual slides no longer have their eyes blocked by a microscope’s eyepiece. For the first time, a pathologist’s eye pupils’ X- and Y-coordinates could be instantly recorded during actual slide viewing. This represents a breakthrough for investigators with an interest in analyzing the physical profiles of novices and surgical pathology experts.

The initial eye movement and human performance study of Dr. Krupinski’s group (using DMetrix telepathology virtual slides) was published in Human Pathology last year (Krupinski EA, et al. Hum Pathol. 2006;12: 1543–1556). For Dr. Krupinski’s virtual slide experiment, she displayed, individually, a set of 20 virtual slides of breast core biopsies. The observers (three medical students, three residents, and three pathology faculty members) viewed them at very low magnification on a high-resolution IBM video monitor screen. The observers were tasked with identifying, by pointing at the video screen, three regions of interest where they would want to zoom to higher magnification. While carrying out this search task, an eye-tracking apparatus, mounted on their foreheads, documented the X- and Y-coordinates of one of the observer’s pupils. Sampling was at a rate of 60 coordinate pairs per second.

Slide C illustrates the scan-path of a pathologist. This is superimposed over the virtual slide. The pathologist selected the three regions of interest in under five seconds. Upon re-review, regions of interest did contain diagnostic information. The straight lines in the scan-paths trace so-called saccadic eye movements. The circles show locations where the observer stopped and dwelled on the image. Foveal (central) vision can be mapped and is indicated by the dashed circle (slide D). The circles labeled one, two, and three indicate the positions of the selected regions of interest. For each observer, it was determined if decisions to zoom to higher magnification were made while viewing that location with central or peripheral vision. Slide A is a scan-path of a medical student in our post-sophomore fellowship program.

Dr. Krupinski and her collaborators reported that the virtual slide search strategies were different for each group of observers. Typically, medical student searches took about 10 times as long as pathologist searches for identifying three regions of interest for zooming. Also interesting to find was that the trainees—that is, the medical students and residents—made almost all of their zoom decisions based on imaging with their foveal vision. In contrast, pathologists often based zooming decisions on viewing areas of the virtual slides with their peripheral vision. Despite these differences in search strategies of the novices and the fully trained pathologists, they all selected comparable regions of interest for zooming to higher magnification.

The lessons learned from our work in cognitive psychology aid our current work on developing facilities. We recently completed construction of a first-of-a-kind multifunctional conference amphitheater, located on the new branch campus of the College of Medicine in Phoenix. The T-Health Command and Control Amphitheater will be used as, among other things, a cognitive psychology lab in which we will explore uses of various areas of visual fields (that is, central versus peripheral vision) for medical decisionmaking, and as a test bed for developing pathology command-and-control laboratory software systems to manage the workflow in telepathology virtual group practices.

At the front of the amphitheater is a large video screen that consists of a 2 x 6 video cube array measuring 5 feet by 20 feet (Weinstein RS, et al. J Interprofessional Care. 2007; 21 (S1): 1–13). For this to function as a triage center for telepathology cases, we would plan on creating a distributed group practice with pre-credentialed pathologists on call to handle specialty surgical pathology cases. Cases could be prioritized for distribution to panels of telepathologists by case managers who may be in the amphitheater or at satellite call centers. Surgical pathology cases would be prioritized and assigned to pathologists according to pre-set criteria (patent pending, University of Arizona and UltraClinics—another university spinoff company).

It remains to be seen if our research on vision proves to be relevant to the Laboratory of the Future work effort, but we are optimistic. For example, our work on eye movements and visual perception could be helpful in designing presentation layers for the laboratory command-and-control system software we are envisioning and designing. The UltraClinics company is working on next-generation pathology command-and-control workflow software tailored to the specifications of a decentralized laboratory group practice (Weinstein RS, et al. IBM Systems Journal. 2007; 46: 69–84). The amphitheater will eventually be equipped to allow us to test laboratory workflow software according to various virtual group practice scenarios.

Some of what I’ve said may sound like science fiction, but that’s the very nature of true innovation. Author Arthur C. Clarke points out that true innovations often seem like magic. As a point in fact, there are literally hundreds of papers that have already been published on topics related to what I’m describing. And some of that work provides the foundation for our own Laboratory of the Future program. Klaus Kayser, MD, PhD, of Germany, Bela Molnar, MD, PhD, of Hungary, and I summarized some of this literature in a book, Virtual Microscopy: Fundamentals, Applications, Perspectives of Electronic Tissue-based Diagnosis (VSV Interdisciplinary Medical Publishing, Berlin, 2006). The book contains 534 references. We’ll see how much of what I have described today actually happens in the decades ahead, but the book might be a reasonable starting point for your personal reality check.