LONDON – Oxford University startup Ultromics Ltd. raised £10 million (US$13.4 million) in a series A round to fund the U.S. launch of its artificial intelligence (AI) image analysis system for diagnosing coronary artery disease.
The system, called Topological Analysis (TA) is designed to act as an aid to cardiologists in assessing echocardiograms of patients referred with symptoms including chest pain, shortness of breath and palpitations.
Ultromics cited data showing that currently, visual inspection of images results in the correct diagnosis in 80 percent of cases. That leaves one-fifth of patients to be referred for invasive confirmatory tests or treatment they do not need, or given an all clear and sent home when they do.
In a test of TA in 500 patients at six cardiology units in the U.K., diagnostic accuracy increased to 90 percent.
"The greater accuracy means you are reducing the number of patients sent home who shouldn't be, and reducing the number referred for surgery who don't need it," said Ross Upton CEO and co-founder of Ultromics. "Apart from the benefit for patients, there's a big cost saving for the National Health Service," he told BioWorld MedTech.
The algorithm that drives TA was trained on more than 5,000 echocardiograms held in what is claimed to be the largest fully consented commercial heart imaging database in the world, which was compiled by Paul Leeson, professor of cardiovascular medicine at Oxford University and a co-founder of Ultromics.
Upton wrote the algorithm as a doctoral student, going on to found Ultromics in May 2017.
The series A round was led by Oxford University's venture fund, Oxford Sciences Innovation, along with a collection of other venture capital funds and business angels.
Echocardiograms are the mostly widely used means of diagnosing coronary artery disease, with 40 million performed each year in the U.S. alone, but they contain a huge amount of information that currently is being ignored.
A cardiologist assessing an echocardiogram has to make a qualitative judgement based on a handful of visual elements that represent only a fraction of the data that is available on a scan.
TA meanwhile, is able to home in on 80,000 different data points, assembling these into a pattern that can be matched to confirmed diagnoses in the image database. It can pick up cues that cannot be discerned by the human eye.
Although TA "works in an instant," Ultromics is not promoting it for its greater speed or as a disruptive replacement for human experts, but rather has been careful to build it into existing care pathways.
"The software sits on the computer workstation on which the cardiologist types up reports," Upton said. "It fits neatly into clinical workflows." By the same token, no extra training is required to use TA.
Ultromics also has deftly steered around the concerns about commercial companies getting access to individual patient data. Echocardiograms are exported to the cloud for analysis, but they are anonymized. "No patient data has to leave the hospital," Upton noted.
While the algorithm underlying TA could, in principle, become more refined and accurate with greater use, it will be fixed in the expected launch mode advance of clinical trials. The six units testing TA in the U.K. are due to be scaled up to 20 hospitals this year, in preparation for formal launch in 2019.
Ultromics also is planning the initial U.S. launch for early next year, pending FDA clearance. The company has established a subsidiary in Seattle and is setting up trial sites.
Upton noted all the testing to date has been done in live clinics, meaning there already is a considerable volume of real world evidence demonstrating the value of TA.
Despite a proliferation of AI tools in cardiology (see BioWorld MedTech, June 12, 2018) as elsewhere in medicine, Upton said Ultromics' access to such a large imaging database means that as yet, there is no direct competitor product to TA.
Following on from coronary artery disease, Ultromics is planning to develop AI diagnostic tools for other heart diseases.