IN GERMANY, AN EMERGING DIGITAL PATHOLOGY (DP) COMPANY has introduced a novel feature that could be a low-cost way for pathology groups to access artificial intelligence (AI) for digital image analysis. This would appeal to pathologists who may be reluctant to invest the substantial cost and time required to acquire and implement image scanners and digital pathology systems.
This novel approach uses the cloud to allow a pathologist anywhere in the world to upload an image for the DP company’s image analysis tools to analyze. Then, the system provides the pathologist with the results of the analysis.
Simple Access to AI Tools
This development is a strikingly simple way to access sophisticated pathology image analysis tools. Independent of how the German company, Mindpeak, charges for access to its digital-image analysis system, the referring pathologist needs only a digital image, a computer, and access to the internet to submit a digital pathology image for analysis.
Currently, Mindpeak focuses on the most common cancers, such as breast and lung cancer, and has a limited but growing number of digital image analysis algorithms that pathologists anywhere can access. Therefore, this DP company does not yet offer the full menu of tests that a busy pathology group would need every day. But the more important point is that a digital pathology company is willing to make its digital image analysis tools widely available to any interested pathologists.
Another insight that pathologists can take from this development is that when pathologists submit digital images for analysis online, such a system can help end the debate over whether it is better to use an open system or a closed system for such analysis. That choice is one pathology groups currently must make when preparing to purchase scanners and digital pathology systems.
At least one other company is using a system similar to the one Mindpeak uses to make pathology image analysis tools available via the cloud. That means, it could soon be possible for Mindpeak or other companies to offer best-of-class systems that provide digital image algorithms for each of the anatomic pathology subspecialties.
Mindpeak CEO and founder Felix Faber has said that offering AI-powered tools over the internet will allow anatomic pathologists to shorten the time required for diagnosis, while also improving the quality of care and diagnostic accuracy.
Deep Learning Tools
Founded in Hamburg in 2018, Mindpeak says it automates time-consuming visual tasks in clinical laboratories and pathology groups with state-of-the-art artificial intelligence (AI) and deep learning (DL) tools. Deep learning is defined as a subset of machine learning in which algorithms learn from analyzing large amounts of data.
Mindpeak’s strategy of allowing access to its AI and deep learning tools via the internet is a new development. Thus, it is still early for The Dark Report to identify and interview surgical pathologists who have established relationships with Mindpeak and have begun submitting images via the web for analysis.
Digital Workflow in Pathology
Pathologists can realize the full benefit of AI-utilization in a digital workflow where the AI algorithms are integrated into an image management system (IMS). An executive at Gestalt Diagnostics, a software company in Spokane, Wash., that has partnered with Mindpeak, is familiar with how Mindpeak has organized this service. Gestalt provides digital pathology solutions for pathologists.
In an interview, Gestalt’s COO Lisa-Jean Clifford clarified that the Mindpeak algorithm does not provide the diagnosis. “The algorithm provides the information that the pathologist then uses to make a diagnosis,” she said. “That’s called a pathologist-aided interpretation.
“One of the interesting features of the Mindpeak system is that anatomic pathologists or AP groups can have whole-slide images (WSI) or just a specific section from a whole-slide image analyzed when the AI algorithms are deeply integrated into IMS platforms, such as our PathFlow,” she added. “If the submission is just a section of a whole-slide image, that’s called a region of interest.
“When a pathologist submits a region of interest, the pathologist can use annotation tools to add a circle, a square, or other line drawing to a specific area,” Clifford noted. “Basically, the pathologist can then do a review of the AI-algorithms’ analysis and return the analysis quickly.”
In May, Mindpeak gained the European Union’s CE-IVD Mark for its BreastIHC AI software’s ability to detect and quantify breast-cancer cells for primary diagnosis. CE-IVD means the software has met the requirements of the European Parliament to sell its software to pathology practices as an in vitro diagnostic medical device.
In the coming months, Mindpeak plans to submit an application to the federal Food and Drug Administration for approval to market its BreastIHC AI software to pathologists in the United States, Faber said. BreastIHC AI can distinguish between tumorous and non-tumorous structures at the cellular level and is the first deep-learning software of its kind to get the CE-IVD mark, he noted. In the meantime, interested U.S. laboratories can follow the approach of introducing lab-developed tests, Faber noted.
The increased throughput that AI provides creates an opportunity for anatomic pathology groups that have invested heavily in digital scanners and imaging systems in recent years to reduce how much they spend on these systems, Faber said in an interview with The Dark Report.
Shift in Spending to AI
“Right now, pathology group budgets are usually allocated for scanners and image management systems,” he commented. “But in the next couple of years, they will shift that spending toward artificial intelligence because AI is the part of the process that—in the end—creates the value that pathologists seek from digital systems because it speeds up the process and it increases accuracy.”
During the same interview, Anil Berger, PhD, Mindpeak’s Vice President for Sales and Marketing, explained how Mindpeak’s product works. “The BreastIHC product includes three algorithms: one is for Ki-67, a protein used widely as an indication for proliferation for human tumor cells,” he said. “And the other is for the quantification of estrogen receptors (ER) and progesterone receptors (PR), and human epidermal growth factor receptor 2 (HER2, not yet CE-marked), which are markers for breast cancer.
“For BreastIHC, we have completed a study that’s not yet published about the variability of breast cancer diagnosis from different laboratories,” Berger added. “In that study, we showed that our product, BreastIHC, can work on different scanners and different stainers. In fact, we determined that BreastIHC works with six scanners and three stainers.
“What that means is that a wide range of pathology labs using different scanners and stainers could use the BreastIHC software almost right off the shelf,” he explained. “It’s a plug-and-play application that works without the need for any retraining and without the need to set up new parameters in the laboratory.”
Boosting Throughput with AI
During the interview with The Dark Report, Faber explained the role of artificial intelligence. “The AI software classifies cells in different ways, depending on whether they are tumorous or non-tumorous and whether they are positively stained or not,” he said. “On digital images, the software can identify where there might be a membrane staining, for example, and whether the membrane has been positively stained.
“At the same time, the AI software counts the number of cells that it identifies as tumorous and provides a clinical score. It then suggests that clinical score to the pathologist,” Faber added. “So, instead of having to classify and count, let’s say, 2,000 cells manually one by one, the AI software does the counting for the pathologist.”
By counting cells and providing a clinical score, such systems can save pathologists a significant amount of time each day. Therefore, AI software incorporated into digital imaging systems allows pathologists to review more cases per hour, thus improving throughput. Increased throughput means pathologists can provide faster results to referring physicians and better care to patients, Faber commented.
It was just last week when the federal Food and Drug Administration (FDA)issued its first-ever clearance allowing a pathology image analysis system to be used in clinical settings (See this article in this issue). Pathologists and pathology practice administrators can expect to see more AI-powered image analysis solutions submitted to the FDA for review.
Executive Explains AI Software Pricing
IN THE UNITED STATES, Mindpeak works with several partners in anatomic pathology, including Gestalt Diagnostics, PathPresenter and Augmentiqs, among others.
Gestalt COO and Chief Strategy Officer Lisa-Jean Clifford explained that Mindpeak has adopted flexible terms for pricing its products to anatomic pathology groups. “Your lab can pay a fee for a one-year term based on the number of pathologists who would use the AI tools, which amounts to a yearly subscription for unlimited use,” she noted, adding, “Or, your lab can choose an enterprise license. These are options that we resell for Mindpeak with our larger reference laboratory customers.
“An enterprise license works on a sliding scale as a per-month-per-use subscription fee, based on case volumes,” she commented. “In this pricing model, the number of whole-slide images doesn’t matter because it’s one set fee based on the number of cases.
“If a lab uses the Mindpeak software on, let’s say, zero to 600 cases, then the group would pay one set fee,” she noted. “But, if the pathology group used the software for 601 to 1,000 cases, then the set fee would be a higher per-month-per-use fee.”
Contact Anil Berger, PhD, at +49-40-35-67-67-97 or email@example.com; Lisa-Jean Clifford at 508-868-6827 or firstname.lastname@example.org.