NEW YORK (360Dx) – DeepMed IO, a Manchester, UK-based company specializing in digital pathology, has received £1 million ($1.3 million) in government funding to support the development and roll out of its platform.
Relying on artificial intelligence, the company has crafted a software tool to assist pathologists in diagnosing cancers and delivering clinical reports. The company plans to introduce its first application for detecting breast cancer metastases on stained lymph-node sections by 2020.
SBRI Healthcare, a competition launched by NHS England to support small businesses developing innovative products and services, granted DeepMed IO the funding last month as part of a second phase of support. SBRI granted DeepMed IO £100,000 in seed funding last year, and the government funding is the only financial backing the company has received to date.
According to Co-founder Konstantinos Vougas, the support from SBRI should enable the year-old firm to bring to market its first module for detecting metastatic regions in breast cancer, which relies on images of whole sides stained with hematoxylin and eosin to make a diagnosis.
"We are using the SBRI phase II funding to complete the first module and to get it CE marked," Vougas said. Such a tool would be classified as a class A medical device under the new In Vitro Diagnostic Regulation in Europe, requiring certification by a notified body prior to receiving clearance for clinical use, Vougas noted. Once that is achieved, the company will seek to address the UK market first, followed by other markets in Europe, while reaching the US is a more distant goal.
"We have made contacts and have been networking with the heads of procurement in several hospitals in the Greater Manchester area," Vougas said. "We are looking at expanding to large markets in Europe, such as Germany and France, where the digital pathology concept has been adopted," he said. "After 2021, we would consider US adoption. That's the overall roadmap."
Vougas co-founded DeepMed IO with Vassilis Gorgoulis, a molecular pathologist at the University of Athens, last year to deliver new tools supported by deep-learning methods for addressing what Vougas called the "global pathology workforce crisis," where pathologists, particularly in the UK, are unable to keep up with a rising demand for diagnoses.
More specifically, given the aging populations in developed countries, an increasing rate of cancer, and the heavy reliance on a limited pool of pathologists, technologies that can improve upon the efficiency and accuracy of diagnosis are in demand and are being funded by government agencies such as SBRI, Vougas said.
In June, for instance, the UK announced the availability of £50 million to create a network of centers to advance digital pathology and medical imaging. In July, Philips announced an unrelated deal with Oxford University Hospitals NHS Foundation Trust to create a digital pathology network across a number of UK hospitals.
One reason the UK is so keen to see uptake of new technologies in the digital pathology area is that its own internal guidelines are creating a bottleneck in terms of delivering diagnoses and treatments to patients. The NHS mandates that 80 percent of patients referred to a specialist from a general practitioner be see within two weeks. It also mandates that at least 80 percent of cases referred to specialists reach a treatment decision within 60 days after first being seen.
While the NHS has been able to meet its guidelines on the first point, it has consistently been unable to meet the second. "This is a point where NHS has failed since 2013 and it's getting worse and worse," said Vougas. "The SBRI call was for technologies to address this diagnostic bottleneck," he said. "That's why they funded digital pathology."
Another issue is a dearth of pathologists. While the UK has experienced a 4.5 percent increase in oncology-related pathology requests year on year, the workforce to handle them only increased by about 1 percent a year. "Being a pathologist is a difficult specialty," noted Vougas. "It's a lab practice, meaning you have to work over a microscope all day," he said. "Compared to other specialties it is not so well paid. That constitutes another bottleneck with increasing pressure."
Vougas portrayed DeepMed IO's system as an offering that "costs less, solves the problem, and increases the quality of diagnosis." The company has been working with partners at Christie and Manchester Foundation Trust Hospitals as well as the Manchester Cancer Research Center Biobank, the University of Manchester, and the histopathology department of Athens Medical School in Greece to obtain the samples to train its deep-learning methods to make diagnoses.
Vougas noted that DeepMed IO, which currently employs around 10 people, maintains an office in Manchester, as well as a Greek office in Athens. Both are involved in product development.
"Through the biobank we are collecting patient-consented samples for training the module we are developing," Vougas said. "We are digitizing these samples and are annotating the digitized microscope slides with the help of expert pathologists," he said.
Ultimately, the company hopes to shave time off the evaluation process by delivering an AI-powered system for the automated identification of metastatic regions in lymph nodes on stained telescopic slides. Resulting information will then be presented to the pathologist to be included as part of a final evaluation.
"We draw pathologists' attention to these loci, so that [they] can identify whether there is a tumor or not, therefore increasing the diagnostic efficiency," said Vougas. "On average, histopathologists have two or three minutes to scan a lymph node section for metastasis," he added. "So, it supports pathologists' decisions and helps them create reports quickly."
DeepMed IO aims to offer its module via a pay-as-you-go portal. Users would obtain the software for free, but pay a rate for every slide analyzed by the system. Vougas suggested it might cost around £2 per slide analyzed, but declined to provide more specific pricing terms. Given the volume of evaluations done, that would not only generate income for the firm, but reduce the cost of outsourcing cases to external pathologists, he claimed.
The system will also operate via a server-cloud architecture, with all data held locally, he added.
Though DeepMed IO and its collaborators have not yet described the new approach in a publication — a paper has been submitted for review — the company does claim that its alpha prototype achieved a 53 percent decrease in diagnostic time and a 4.6 percent increase in diagnostic accuracy in a pilot clinical evaluation undertaken during the first phase of its SBRI project.
While the initial application of DeepMed IO's product will be for breast cancer, the company aims to create modules for melanoma, lung, colon, and prostate cancer, Vougas said. It will continue to work with its partners to develop these modules. The company will also seek additional investments in the next nine months to support product development, Vougas said.
While some international players, such as Philips, Roche, and Leica, have already established themselves in the UK digital pathology tools market and are investing heavily in the technology, Vougas said it was possible that DeepMed IO could partner with these firms rather than compete against them.
"It's one of our strategies to build a very efficient and worthwhile product and then approach the big players with a strategic alliance concept, where these tools could be integrated with their digital pathology platforms," Vougas noted. "They are jumping on the AI boat, the same as us. We would like to build bridges between players and, in a good scenario, bundle our software with their digital pathology platforms."