NEW YORK (360Dx) – Deep Lens is developing an artificial intelligence platform that it anticipates will eventually be available to pathologists for primary disease diagnosis and could convert more of them to using digital pathology.
The startup said that it will use the $3.2 million seed financing that it announced in October to add more functionality and intelligence to its digital pathology system called Viper, or Virtual Imaging for Pathology Education and Research.
The platform is currently used by pathologists doing research, including those participating in clinical trials for drug development, and it could become available in the US and globally in about a year for use in the primary diagnosis of cancers in clinical settings, Dave Billiter, Deep Lens CEO and cofounder, said in an interview. The firm plans to obtain clearance from the US Food and Drug Administration after validating its technology for primary clinical diagnosis. It is also making its technology available for this purpose outside the US with appropriate regulatory clearances for each country, Billiter said.
The technology is available for free beta testing using up to 100 gigabytes of storage for whole-slide images as part of a lab-as-a-service offering. Its capabilities, convenience, and price point will help remove barriers for pathologists and pathology groups seeking to use digital pathology, Billiter said.
Historically, pathologists have painstakingly analyzed the content of glass slides through the lenses of a microscope, a process that involves tedious and repetitive tasks such as counting different cell types, denoting levels of stains, and annotating cells in different phases of a cell cycle — activities that are error prone, inconsistent, and time consuming.
Through its platform, Deep Lens is combining artificial intelligence and deep learning with computer vision across multiple cancer types to assist in visual analysis.
The Viper platform's capabilities include enabling prioritization of pathology cases, aiding peer-to-peer and pathologist-to-patient collaboration, and facilitating consultation on nuanced cancer diagnoses and case-specific details that can require many years of specialized medical training, Billiter said.
For future use, the firm is integrating artificial intelligence-based image detection, workflow support, telepathology, cloud storage, and built-in APIs for integration by hardware and software vendors and biopharma companies.
The current Viper single-user edition introduces customers to digital pathology by providing workflow and image management tools. A Viper enterprise edition is available for central pathology review and clinical trial enrollment among other functions.
Digital pathology but affordable
Columbus, Ohio-based Deep Lens is promoting use of high-quality imaging services at a low cost. Pathologists ship slides to the firm's accredited scanning labs where they are scanned and loaded to a cloud-based account. Through its website, the firm lists prices for slide imaging and scanning up to $199 for processing of light microscopy whole-slide images.
The system has been used for many years by pathologists doing research in large National Cancer Institute-sponsored clinical trials and has been refined based on input from hundreds of pathology users at more than 65 "major institutions" in eight countries, according to Deep Lens.
The firm has exclusively licensed the digital pathology platform and image analysis algorithms from Nationwide Children's Hospital in Columbus. Billiter and his colleagues began developing Viper more than 10 years ago when he worked as an informatics director at NCH, a pediatric hospital that receives more than 1.4 million patient visits annually.
NCH was an early adopter of digital pathology systems that enable histology slides to be robotically scanned to obtain high-resolution images of glass slides, Billiter said. Robotic scanning has worked well, but no systems have been available that can incorporate image scanning and accommodate the overall workflow of pathologists and the many different clinical use cases that they encounter, Billiter added.
The platform has been developed with input from expert pathology groups from around the world that focused on specific tumor types and subtypes, the firm said.
"Over time, we have worked on so many different projects and collected data about so many pathologists' reviews that we have been able to build a significant library of forms [associated with specific tumor types] that supports the pathology discipline," Billiter said.
Currently the platform's primary use is for pathology education and research, including supporting pharma clinical trials. Pathologists are a critical component of clinical trial design and evaluating whether patients are eligible for specific trials. "When you look at the number of precision treatments being developed in clinical trials, you can see that the role of the pathologist is increasing in importance, but at the same time we see decreasing numbers of pathologists," Billiter said. The Deep Lens system provides a solution by enabling pathologist to focus "on optimizing opportunities throughout the daily workflow," saving time and improving the quality of their diagnoses.
Logic embedded in the Viper system enables pathologists to prioritize their workloads based on the critical nature of a review, for example. The system also enables generalists to collaborate with experts on rare tumors that they may be seeing for the first time. After a pathologist scans an image, the firm's Deep Lens Assist feature enables them to match that image to others within a library of images.
"It's an assistance mechanism that really aids pathologists in certain cases, with certain tumor types, including in cases of rare disease," T.J. Bowen, cofounder and chief scientific officer of Deep Lens, said in an interview.
"We're seeing a lot of momentum in the growth of sequencing, genomics, and the processing of liquid biopsies, and we feel that digital pathology ties together all of these aspects of diagnosing a patient and getting them on the right therapy," Bowen said. The platform has been used for years by scientists in large cancer research initiatives looking to combine genomics, proteomics, and histology data to ensure that "the richest data set for patients so that they get the right therapies," he said.
Deep Lens is engaging with pharma companies that are using its technology for this purpose and is open to working with companies developing similar methods but for use cases not within its current scope and capabilities.
"We don't expect to be able to develop specific AI methods for all types of tumors," Billiter said. "However, we are developing the best AI-based methods for certain use cases and opening up our system to others."
The Deep Lens AI team is using next generation convolutional neural networks to add features to its existing workflow solution across dozens of cancer types and anticipates making cell counting, IHC quantification, mitotic index counting, and other critical tasks available to pathologists.
The firm anticipates that these advanced methods will be available in early 2019.
Adoption curve
Until the US regulatory approval in 2017 of the Philips IntelliSite Pathology Solution scanner, many small and mid-sized pathology labs were slow to embrace the use of digital pathology technology. Some pathologists believe that the FDA's approval of the system for sale in the US sent a clear signal that digital images could one day become the standard of care in anatomic and clinical pathology.
Seeing opportunity in digital pathology, companies and entities other than Philips and Deep Lens have been working to make digital pathology more broadly available.
San Jose, California-based Optrascan, for example, is selling a subscription service to pathologists who want access to digital pathology technology for research use.
Roche has launched a high-speed scanner for digital pathology that it said marked a step in the direction of more automated systems in a field with relatively slow adoption. The firm has initiated clinical trials in anticipation of getting clearance from the FDA that would permit marketing of the scanner for clinical use in the US.
Also, Cernostics is offering TissueCypher — a digital platform for multichannel fluorescence whole-slide imaging and includes image reading, image segmentation, biomarker expression, and tissue structure evaluation — to predict the risk of developing esophageal cancer in patients with Barrett's Esophagus. The test has been available since last year from the firm's CLIA-certified laboratory in Pittsburgh, Pennsylvania.
In a commentary published earlier this year in the journal JAMA Oncology, Balazs Acs and David Rimm, professors in the department of pathology at Yale University School of Medicine, noted that some deep-learning algorithms in a study achieved better diagnostic performance than a pathologists' panel.
Deep Lens anticipates that its system will be available for primary diagnostic use in clinical settings in about a year. "We have a roadmap for entering clinical applications and are conscious of regulatory requirements, [including the need to obtain FDA approval] for use of the technology for that purpose," Billiter said.