Pathology can be a tedious field that often requires sifting through a huge number of slides to identify the necessary information to score and grade a patient’s tumor sample. Much like imaging, this increasingly digital field is ripe for the application of artificial intelligence (AI) to reduce the burden on pathologists, provide confirmation of their work, marry up this analysis with other types of digital data such as genetic sequencing, and even to potentially uncover novel biomarkers.
To that end, AI-based, digital pathology startup Proscia Inc. has partnered with the University of California at San Francisco (UCSF) to advance the practice of pathology via AI. The pair will start with prostate cancer and then plan to move on to validate approaches in several additional pathology subspecialties.
The Philadelphia-based startup already has a skin cancer AI pathology product that’s known as DermAI; it recently published a study in the Feb. 21 issue of Scientific Reports that demonstrated that this approach is 98% accurate in classifying whole slide images of skin biopsies in real laboratory settings. This is “the first truly real-world-validated, deep learning system in pathology,” the researchers noted. DermAI is available for research purposes only thus far.
Pathologist’s practice
“We think the highest value applications today are going to be the ones that take the most highly repetitive mundane and time-consuming tasks that pathologists do today, and provide any level of automation, or consistency to that process,” Proscia chief product officer Nathan Buchbinder told BioWorld. “That leaves the pathologist behind the steering wheel when it comes to actually making the diagnosis and signing off on a case and making an impact on the patient’s life and their cancer journey. But it allows them to practice at the top of their license.
“They’re not spending the 10 to 15 minutes that it takes to review each and every one of these prostate cases to look at these slides – and then another 10 to 15 minutes to enter information into a report,” he continued. “They’re going to actually use their training, their knowledge and their medical expertise to interpret the results that have already been presented to them.”
Prostate cancer can be particularly burdensome for pathologists, since it requires a large number of slides per patient, complex reporting and a qualitative grading system that can lead to long turnaround times, the need for ancillary tests and lower confidence in treatment decisions.
A pathologist analysis of a prostate biopsy can be both time-consuming and involve tasks that can be conducted inconsistently. A prostate biopsy might include 12 core samples that would result in dozens of slides – with only one or two of those slides perhaps even containing tumor. Obtaining accurate and sufficient biopsy samples is a necessary step that itself isn’t always achieved.
Proscia envisions its system as potentially identifying the slides that are most likely to contain tumor, with the next potential steps of visually outlining the tumor for the pathologist and starting to make measurements of those regions. These could then offer a starting point for the pathologist’s review, or simply be used as a quality check alongside the pathologist’s own assessment to ensure consistency.
Digitization unbound
Beyond these basic tools, Proscia also hopes its AI-based analytics also have the potential to unveil new potential measurements. It could enable the routine quantification of what might previously been simply qualitative observations.
“We’re quite confident that there is a significant amount of information, of patterns in the morphology of this tissue, that correlates to clinical outcome or response to therapy or grading of the cancer. Pathologists have done a phenomenal job of translating into qualitative measurements and metrics that artificial intelligence and other computational applications can really drive towards a precision interpretation of who to treat and how to treat them and how well those treatments are going to work.”
In addition, these digital pathology analytics could ultimately be integrated and analyzed alongside all sorts of other digital information such as genetic or proteomic data on the tumor. That could unlock much greater diagnostic and treatment specificity since it would greatly aid in bridging the pernicious knowledge gap between phenotype and genotype, with pathology always having been confined to the former.
Most pathologists, like radiologists, are eager to have tools, even if they are based on the much-hyped inclusion of AI, that make their work simpler, faster and more accurate. But despite some advances at research hospitals, many community hospitals do not create digitized slides, and the process of scanning physical slides to make them digital can itself be time-consuming and expensive.
Proscia already offers the software platform Concentriq that’s designed for digital slide management. Now, the idea is to build AI tools on top of that to enable analytics. It launched in 2016 and is now widely in use by more than 3,000 pathologists and researchers. The company received a CE mark in November for its Concentriq Dx for use in primary diagnosis of cancer from digitized images of tissue biopsies.
The company has agreements with several academic research centers. In December, Proscia also partnered with another leading cancer research center, the Johns Hopkins School of Medicine, to develop AI pathology applications in multiple, undisclosed diseases. Thus far, it has amassed more than one million whole slide images from more than one dozen partners.
Proscia was founded in 2014 by three Johns Hopkins or University of Pittsburgh alums, including Buchbinder as well as CEO David West and CTO Coleman Stavish. Proscia raised an $8.3 million series A round in September 2018.
The startup expects that the UCSF relationship will enable it to amass significant technical adeptness, as well as clinical evidence that will enable it to develop a prostate cancer product, as well as additional pathology subspecialty products, that are then ready for testing in real-world laboratory settings.
“UCSF prides itself as being an institution in the intersection of research and clinical practice of medicine, continuously working to translate new findings into more effective prevention, diagnosis, and treatment,” said Zoltan Laszik, professor of pathology at UCSF. “Proscia’s focus on delivering practical AI solutions strongly aligns with our efforts, and we are pleased to work together to improve the routine pathology workflow.”
