Scope#30 | Enlitic

SAN FRANCISCO – Radiologists need to cut down on their screen time.

By James Simms

Dealing with an avalanche of medical images is hurting the ability of these specialists, who use X-rays and other imaging to help diagnose and treat illnesses, to spend more time talking with patients and other doctors, like neurologists and oncologists. Old-fashioned communication is key to getting the best patient outcomes. A hundred-fold rise in image volumes in recent decades, increasingly complex imaging and treatments and aging populations are adding to radiologist workloads – and healthcare costs. Exacerbating that, countries like Japan, Australia and the U.S. face a severe radiologist shortage too, as more of them retire.
For a non-physician, it’s like going through thousands of images on one’s smartphone and trying to accurately identify them all. If you label a crocodile an alligator – no big deal. But picture the workload is many times that, a long line of people waiting and those decisions fundamentally affecting a patient’s health.

U.S. medical software maker Enlitic wants to change all that through the use of artificial intelligence – a technology that has the potential to revolutionize medical image evaluation.

“The problem that we’re really trying to solve at Enlitic is to give radiologists back their time to do their job properly…to help sort through the data to find the needle in the haystack,” he says. “The more time we can spend talking to patients and referring doctors, the more context we are able to put on the images, and it gives a much more accurate result. If we have a[n AI] model that will find the abnormality first, even though it may or may not know what the problem is, it gives us the time to go and talk. That human part is hard to replace.”

AI Looks to Deliver Technology Benefits

Over the past two decades, digitalization has improved the management of images, including X-rays, which previously were processed like film and turned into physical slides for viewing. But that didn’t become a game changer for assessing them – nor did computer-assisted image reading.

AI coupled with massive patient data sets – over a billion anonymized images in Enlitic’s case –promises to help radiologists become more accurate, quicker and more efficient. Radiology studies include X-rays, ultrasounds, CAT/CT scans (computerized axial tomography is a series of X-rays taken in slices) and MRIs (magnetic resonance imaging is created using powerful magnets and radio waves). Doctors are more fascinated with CTs and MRIs than more mundane and difficult-to-read X-rays, which make up 45% of all radiology images, too.

And AI will help reduce radiology errors. In one out of four cases today, a diagnosis of no illness was made when there was one – a false negative; And in one out of five, an illness was diagnosed when there was none – a false positive, Enlitic says.

“There are so many ways that AI can positively impact many issues within the clinic. There are an estimated 450 million people each year that get some form of radiology examination,” says Enlitic CEO Kevin Lyman, who joined the company three years ago as a computer engineer. “It’s not about man versus machine but man plus machine and how together they can offer substantially better care.”

That potential is enormous.

UK-based research firm Signify expects a quadrupling of the global market to $2 billion in the four years through 2023. Lyman, however, says such estimates underestimate the potential. “Ultimately, there are dozens of types of studies in radiology, each of which represents a multi-billion dollar market,” he says.

In April, Japanese trading company Marubeni led a $15 million series-B round of funding for Enlitic and has been working with the firm since 2017 to develop Japan’s domestic market for its software.

X-rays to Feel First AI Impact

AI can do things that human beings alone can’t.

In the case of lung cancer, Lyman says the firm’s software has been able to detect cancer 24 months earlier than doctors because the model had past data on patients who eventually developed the disease.

AI can do that by using algorithms to determine the relationship between inputs and outputs. Machine learning, an AI subset, calculates what to multiply factors by to come up with a result. In a simplified credit score, for example, it would figure out what to multiply income, number of credit cards and net worth to produce a credit score, Lyman says. But experts often don’t know the factors to help diagnose a certain disease. In that case, deep leaning, a subset of machine learning, can determine what unknown factors or patterns contribute to a diagnosis, given a well-labeled data set, like the above lung cancer example, he adds.

Lyman believes that the high quality of Enlitic’s data and the comprehensiveness of its AI radiology software separates it from rivals, both large and small. That comes from training the specialists labeling the data, which trains the algorithm, to ensure accuracy and consistency, and literally working side-by-side with radiologists in software development. Dr. Masahiro Jinzaki, a professor of diagnostic radiology at Keio University’s medical school, says Enlitic’s software “has shown to have a high diagnostic accuracy and so appears to be effective.”

By the end of 2019, the company expects to cover 95% of the global volume of X-ray types, such as chest and head, and 95% of CTs and MRIs in the same way by the end of 2021. The ease of employing Enlitic’s software in the radiology management software of different manufacturers globally is another distinction, he adds. The most promising use for AI in radiology, Dr. Jinzaki says, is in X-rays because it would likely increase accuracy and reduce time spent by doctors on an unwanted task.

Hurdles to Introducing AI and Enlitic’s Outlook

While there is great promise for AI in radiology, there are hurdles to overcome before patient usage and deeper development of the technology. These include securing regulatory approval, constructing a payment model for the service, developing strict universal rules for patient data usage and overcoming any remaining resistance from radiologists concerned that AI might replace them.

Eventually, Enlitic’s CEO says the best way to show the company’s impact would be a counter on its website tallying misses prevented, needless procedures avoided and doctors’ time saved.

“It’s really the ability to affect clinical outcomes in a way that no individual doctor could,” Lyman says. “And in painting a picture of how we would grow over the next ten years, that would always be the key performance indicator that would drive us the most.”

INTERVIEW: Dr. Masahiro Jinzaki, Professor and Chairman / Dept. of Diagnostic Radiology, Keio University School of Medicine, Vice Director / Keio University Hospital

Dr. Jinzaki, a recognized expert in diagnostic radiology in Japan talked about Enlitic’s artificial intelligence software used to help read medical imaging, namely X-rays, and the overall use of AI in medicine. He is also involved in the “AI Hospital Project” sponsored by the Cabinet Office on how medical institutions can introduce AI to improve their efficiency. (The interview has been edited for clarity.)


Question: How do you evaluate Enlitic’s AI software?


Dr. Jinzaki: Right now, the only way to evaluate it is to look at the academic articles that evaluate diagnostic accuracy – as there aren’t that many approved AI medical software programs. Among those, Enlitic has shown to have a high diagnostic accuracy and so appears to be effective.

Question: What areas do you think that Enlitic’s AI software could be used?


Dr. Jinzaki: The diagnostic accuracy for X-rays read by people is already pretty low, so any increase in accuracy would be very welcome. On the other hand, since human beings have high accuracy for chest CTs (computerized axial tomography), rather than CTs, chest X-rays look to benefit the most from AI.

In the case of chest X-rays, nobody really wants to read them because they are pretty hard to decipher. Because things can be hidden behind the shadows of organs, there is a high risk of overlooking medical problems. So, it would be really significant to raise the accuracy of chest X-rays with AI; This would be an example of where AI is more accurate than human beings.

Enlitic is working on AI for chest X-rays, and I think that could be introduced into diagnostics.

Question: What role do you think AI could play in medicine overall?


Dr. Jinzaki: The expectations for AI are a bit excessive. AI is strong in dealing with the past. The software is developed with vast troves of imaging already done to help flag medical problems. It would be great to relegate the past to AI and to enable human beings to create forward-looking technology, such as developing new diagnostics methods.

For Japan, using AI to ensure that expert knowledge is transmitted from the current generation of doctors to the next should really be effective, as the number of experienced doctors is decreasing because of retirements.

At Keio, we also are working on an “AI Hospital.” We’re looking into areas where we can introduce AI in the medical field. We are now introducing AI to reduce the amount of labor required in medicine, such as by using robots. In the future, we plan to expand into the area of expert judgement, such as in diagnosing illnesses.

All information contained in this article is based on interviews conducted in July 2019.

* This software is not approved by the FDA (U.S. Food and Drug Administration) for the diagnosis, cure, mitigation, treatment, or prevention of any disease.
* This software has not taken medical device approval, and is not currently manufactured or sold in Japan. The above information is based on the development and preparation stage for obtaining approval in Japan.