Duke University researchers are using a grant from the North Carolina Biotechnology Center to “supercharge” vital medical information.
The Duke scientists say the instrumentation grant from NCBiotech enables them to use cutting-edge digital imaging and artificial intelligence (AI) technologies to advance cancer research, diagnostics and to support a vast new interdisciplinary data science initiative called AI Health.
NCBiotech’s $200,000 grant went to the Duke BioRepository & Precision Pathology Center (BRPC), Duke’s largest human tissue biobank. BRPC stores samples from consented patients for use in research.
Specifically, the March 2018 funding enabled BRPC to purchase of a Leica Aperio AT2 Digital Pathology Scanner. This state-of-the-art instrument performs high-throughput whole-slide imaging of both surgical pathology tissue slides and fluid sample preparations called cytopathology slides.
By digitizing these one-of-a-kind pathology slides, researchers are cataloging, analyzing and sharing the images with colleagues across North Carolina and beyond. Duke Cancer Institute provided an additional $146,000 to purchase a Visiopharm Image Analysis Software Algorithms platform with the “Oncotopix” software and the Pathcore Web, secure, web-based image management system to enable this process.
“There are specific areas in medicine where, I think, this is game-changing,” said Shannon McCall, M.D., director of BRPC and principal investigator for the grant.
The scanner has a 400-slide capacity (40 to 100 times more than previous scanners) and rapid scanning capability at the rate of 50 slides per hour, said McCall. This means technicians can load prepared slides into the Aperio and walk away to do other work for eight hours, or let it run overnight. Previously, they would have been tied to a scanner constantly setting up and reloading a few slides at a time by hand.
Not only does the Aperio improve speed and efficiency, it has significantly increased the amount of data for analysis available to researchers and clinicians. This opens new frontiers for discovery in cancer research and diagnostics. For example, it can scan and create separate digital images of a cell at 400X magnification in nine layers, or planes, explained McCall.
This is transformative because cytopathology slides, as aspirates of entire cells, can be different thicknesses, unlike traditional tissue slides which are prepared in a single 5-micron plane. A cross section of a human hair is 50 microns. Pathologists normally examine cytopathology slides by manually focusing up and down through their light microscopes. The Aperio’s ability to create Z-stack images allows cytopathologists to convert 100% of the slide’s information into digital output. This brings the digital pathology revolution to cytopathology as well as regular tissue slides.
A Boost to Duke’s AI Health Initiative
AI experts at Duke working alongside pathologists are also developing algorithms, or instructions for computers, to analyze the images and find new patterns leading to improved cancer screening and diagnostics. A recent experiment demonstrated that a proposed algorithm reading whole-slide images of thyroid tissue biopsies scanned by the Aperio achieved accuracy comparable to human pathologists reading the actual glass cytopathology slides using traditional light microscopes.
Based upon this and several other ongoing projects, Duke is now planning to buy another scanner to keep up with the demand for digitized slides and to accelerate the burgeoning university-wide initiative known as AI Health.
“We’re very excited to see one of our grant awards enable a ground-breaking initiative and new, interdisciplinary collaborations that will positively impact patient lives,” said Rob Lindberg, Ph.D., the NCBiotech vice president of science & technology development. “This is a great story that embodies the heart of our instrumentation grant program, which creates research capacity by providing cutting-edge, game-changing resources for which external funding is almost entirely unavailable through traditional grant mechanisms.”
AI Health is the umbrella term for an ambitious interdisciplinary initiative led by Duke’s Vice President of Research Larry Carin, Ph.D. The James L. Meriam Distinguished Professor of Electrical and Computer Engineering is an international expert in AI and leads the team conducting the thyroid tissue research. He’s also a member of the NCBiotech board of directors.
“It’s been said we are currently living in the fourth industrial revolution, which is being driven by AI,” said Carin. “Prior industrial revolutions were driven by things like electricity. So, the question one might ask is, what is the analogue of electricity in this industrial revolution? It is data, access to data at scale, computing power and advanced machine learning algorithms. All these things constitute the electricity of this industrial revolution. AI Health is Duke’s electric company for the fourth industrial revolution.”
He added that, just as our electric utility companies do not tell us exactly how to use electricity – we are free to decide whether to light our homes, run our refrigerators or power our computers – AI Health is not going to prescribe exactly how users across Duke use AI. It will facilitate use of AI.
Interestingly, in the same way NASA invented what it needed to achieve steps along the pathway to the moon, Duke is doing so in AI as well. McCall described how a computer scientist in her lab had to write a program to remove slide labels identifying patients from the images before they began sharing them. No information identifying patients leaves the confines of the computer firewall built around the pathology core, she said.
Also, the sheer size of the files of these data-rich images required building a robust computing pipeline for sharing images and develop other methods for secure transport of the data.
“A typical file size is one gigabyte,” said McCall. By comparison, that is more than 300 times the size of a single high-resolution cell phone photograph. “One of the costs associated with this work is storage and management of the data files.”
Nevertheless, Carin envisions AI Health to address these challenges and more as it facilitates use of AI in myriad ways by a range of departments – clinical as well as research – to “transform the health system.”
“We want to inject AI and data science into everything we do. The idea is that every clinical unit should be constantly learning about itself from data, constantly asking questions about how we can improve, constantly testing new ways of doing things,” Carin said.
Another element under the AI Health umbrella at Duke is The Crucible. This will be a place of innovation where technology meets AI Health to provide what is called data liquidity, or access to data at scale. To leverage all the fruits of AI, said Carin, one needs to have the capacity to get data out of the existing infrastructure and into a place where it can be analyzed.
Finally, the Duke Institute for Health Innovation is a place where pilot projects can test new prototypes for ideas at the intersection of AI and health, particularly with a focus on trying to transform the health system itself.
None of this has the intention to replace the important doctor-patient relationship, said Carin. Rather it is to “supercharge clinicians” by automating many processes that take time away from patients and to provide information that streamlines and improves diagnostics and care planning.