By Deborah Borfitz
April 18, 2019 | Pathologists around the world over—65,000 of them in 163 countries at last count—appear enamored of a free and easy-to-use digital pathology platform that allows slides to be aggregated irrespective of the scanner that created them. Equally appealing to many users of PathPresenter, “built by pathologists for pathologists,” is that they can seamlessly add both digital pathology and radiology images to their PowerPoint presentations.
So says Matthew Hanna, clinical instructor in breast pathology and informatics at Memorial Sloan Kettering Cancer Center (MSKCC), as well as one of the co-founders with Rajendra Singh, of the open-access platform. The first iteration of the platform, released in January 2017, was intended for peer-to-peer sharing of interesting cases and a source of images for rare diagnoses, he says. PathPresenter’s “public library” quickly grew to include 13,000 quality slides crowdsourced from users around the world.
PathPresenter is a “labor of love” and was initially considered more of an educational platform, an alternative to the otherwise cumbersome and time-consuming business of compiling content for conferences and tumor boards, he says. Traditionally, pathologists peered at glass slides one at a time under a microscope looking for interesting spots to photograph and import into their presentations. With PathPresenter, they could instead bookmark regions of interest on a stack of digital slides for immediate upload and review.
Increasingly, PathPresenter is seen also as a clinical tool for primary diagnoses and consultations. Among 124 MSKCC pathologists responding to a recent survey, 54% said they’d feel comfortable using digital pathology for primary diagnosis if they had the glass slides available upon request and less than one-quarter if they did not, says Hanna. “This is likely related to comfortability and familiarity and will assuredly increase in the future.”
Large academic centers such as the University of Pittsburgh, Ohio State, Cleveland Clinic, and Johns Hopkins already have a digital pathology consultation portal that uses whole slide imaging technology and accepts various forms of digital images and files, says Hanna, and MSKCC is soon to implement one. Pathologists at the referring institution can scan and digitally transmit their slides along with relevant patient information to subspecialty experts at the consulting institution in lieu of physically transporting glass slides that can break and fade over time. The turnaround time of digital consultations could also be much faster.
Pathologist-Friendly System
The biggest fault of most digital pathology platforms on the market is that they’re proprietary, standalone systems that are specific to a vendor’s slide-viewing software, which makes interoperability a key component of PathPresenter, says Hanna. With PathPresenter, pathologists can share and manage digital slides, radiology images, and almost all medical content from anywhere.
PathPresenter was also first to market with a feature set covering most everything a pathology department might need for educational use cases—the presentations and public slide library components as well as a repository for each pathologist’s slides, a folder for slides most useful for pathology training, and a quiz-and-scorekeeping module to track the academic progress of pathology trainees.
In a study soon to publish in the Archives of Pathology & Laboratory Medicine, Hanna shows a 93% decrease in glass slide requests at MSKCC once digital slides became available and a significant savings on operational costs, he says. Diagnosed glass slides traditionally go to an in-house slide archive and, after a certain number of years, are moved to an off-site storage facility, he explains. Accessing the slides requires writing a requisition and a day or two of waiting. With digital solutions, the digital slides are readily available in real time.
The biggest learning curve for those new to the system is in trying to organize the data, says Hanna. Pathologists working at large academic centers would also not have direct access to raw digital images for uploading into the PathPresenter system if they’re stored in a central repository or data center.
The platform continues to be developed, says Hanna, noting a “completely revamped version” is soon to publish that will include a website refresh and improved backend services. Rather than being a single, all-purpose platform, the next iteration of PathPresenter will have three arms—one for education that builds on the current capabilities, another for self-publication of digital pathology books or pathology-related digital content, and a third allowing the technology to be used for machine learning (ML) initiatives.
The "Race" for Answers
“Machine learning at Memorial Sloan Kettering is being used in a big way,” says Hanna, and it is hardly alone. “The race is on to see who will have the first or best medical-grade machine learning algorithm for pathology.” He foresees ML facilitating the creation of digital “prognosis markers” or providing efficient and objective quantification of biomarker assessment.
Pathologists have for years assessed stained tissue under a brightfield microscope to grade and stage tumors, says Hanna, but evidence suggests more is going on than meets the eye. For example, the elliptical fit or angle of continuous tumor regions would be undetectable—features that have been associated with breast cancer prognosis “with patients faring worse if the area or angle is smaller rather than large.”
Most ML algorithms today are looking to quantify features, such as mitosis, for purposes of grading tumors, says Hanna. Organizations routinely participate in grand challenges that “sort of gamify machine learning” by having contestants train their ML model on an open data set to identify something specific. In 2016, for instance, the Camelyon Challenge had teams use whole slide images to identify metastatic breast cancer in lymph nodes and the following year to quantify multiple cancerous lymph nodes and their size.
Remaining Challenges
Adoption of digital pathology in general has been slowed not only by interoperability issues but also by inefficient workflows, says Hanna. Pathologists “live in their laboratory information system,” yet it is often disconnected from other systems needed to view and analyze slides and make ML calculations. Only in recent years have digital pathology platforms even had an interface with various commercial laboratory information systems on the market, he adds.
Another barrier is that standard information systems in most hospitals—including those for transmitting patient metadata, and laboratory and imaging reports—is based on Health Level Seven International (HL7) standards and PathPresenter does not currently accommodate HL7 messaging communications, Hanna says. “That will likely change in future.”
The limitations of digital pathology in clinical settings is somewhat nebulous, Hanna says. Ohio State is the only academic center in the U.S. where a subset of pathologists are doing primary diagnostics without any glass slides using the first FDA-approved whole slide imaging device (Philips IntelliSite Pathology Solution) that scans at equivalent 40x magnification. With other scanners, true “magnification” (resolutions) can vary widely even among those with the same scan magnification objectives, he says. Attainable resolution depends on multiple hardware components, notably a unit’s image acquisition device.
Digital pathology is “somewhat lacking” in the subspecialties of cytology and hematopathology since they work with tissue at the cellular level and it is mostly in three dimensions, which requires scanning the glass slide in multiple planes, says Hanna. That means larger file sizes, longer scan times, and higher storage costs.
But something must give. The incidence of cancer is increasing at the same time the supply of pathologists is inexplicably dwindling, says Hanna. Among graduating medical students this year, the residency match in pathology was among the lowest across specialties.