Deploying AI to Diagnose Cancer

When it comes to acute promyelocytic leukemia, an artificial intelligence diagnostic could be a quick and accessible lifesaver for millions

When Eugene Shenderov was five years old, he received a diagnosis that almost came too late. Born in Ukraine in 1983, Shenderov developed pediatric acute lymphoblastic leukemia in the wake of the 1986 nuclear explosion at Chernobyl.

“I was quite sick, pale, and bleeding,” Shenderov remembers. Because the region lacked adequate medical care at the time, Shenderov’s parents “knew we had to leave the country or basically I would die.”

The family eventually made it to the U.S., where Shenderov’s cancer was successfully treated. But the experience stayed with him, and today, as a doctor and researcher at Johns Hopkins, Shenderov is working to make sure others suffering from leukemia receive a diagnosis with time to act.

Specifically, Shenderov and his colleagues have developed an artificial intelligence diagnostic, called the “Leukemia Smart Physician Aid,” that has shown promising accuracy in diagnosing acute promyelocytic leukemia, or APL, a rare cancer that affects about 800 Americans a year.

Thanks to this rarity, though, funding has been tricky, says Mark Levis, MD, PhD, one of Shenderov’s collaborators, who provided support to hire a research analyst for the project through his patient gift fund. While the pharmaceutical industry generally supports lucrative endeavors like drug development, the government tends to fund research for common cancers. “The only way this project will get done,” Levis says, “is through patient donations.”

APL, which causes an accumulation of immature white blood cells in the bone marrow, “is considered one of the true oncologic emergencies,” Shenderov says. “It can kill you within hours.” The disease is resistant to chemotherapy and extremely aggressive, the oncologist explains, triggering sudden and deadly brain bleeds.

While APL is among the deadliest cancers, it’s also among the most treatable. If administered in time, an extremely high dose of all-trans retinoic acid, a form of Vitamin A typically used to treat acne, has been shown to cure the disease in more than 90 percent of patients.

“That’s very rare in cancer,” Shenderov says. “Diagnosis in this case almost equals cure.”

But diagnosis remains a problem. Currently, only specialized hospitals offer the genomic testing necessary to make a diagnosis, and typically takes days to return results, too long when time is of the essence.

Shenderov’s algorithm, on the other hand, can make a diagnosis in hours, meaning the patient can begin lifesaving treatment that day.

Best of all, the diagnostic is inexpensive and easily deployable, relying on a peripheral blood smear, a test that examines a patient’s blood under a microscope to identify abnormal cells.

“A peripheral smear costs dollars or cents depending on the region of the world; it’s one of the cheapest medical tests,” Shenderov says. “That’s why it’s also one of the most widely used medical tests in the world.”

This has big implications for rural and developing countries, where limited access to health care means APL is less likely to be diagnosed. Currently, there is “virtually no diagnosis on the continent of Africa,” Shenderov says. Often, patients in these regions are misdiagnosed with a related form of cancer called Acute Myeloid Leukemia, or AML.

“Currently, AML is a death sentence in most emerging economies of the world because … there are very few treatments available,” says Dan Milner, MD, an adjunct professor at Harvard’s T.H. Chan School of Public Health and the executive director of Access to Oncology Medicines, a global initiative working to bring cancer care to developing regions. “For APL there are excellent treatments available, but the diagnostics are a major barrier.”

Milner, who has also provided philanthropic funds to support its development, calls Shenderov’s AI diagnostic, “extremely promising.”

Shenderov and his research partners have been “bootstrapping” the project for nine years, he says, but the Leukemia Smart Physician’s Aid is finally ready for the trials necessary to ensure its efficacy — trials that require the kind of government funding for cancer research that has recently disappeared. Philanthropic sources, meanwhile, are flooded with requests for money.

“The pot has shrunk, and that’s made it harder to go for funding sources,” says Shenderov. “Granters now have to triage … and whole lines of research are under threat of being defunded.”

Because APL affects relatively few people, he knows his project can seem niche. But the diagnostic is “proof of concept,” he says, adding that the technology could eventually be used to identify a whole slate of blood diseases like malaria or other cancers.

As funding avenues narrow, Shenderov, who has spent almost the last decade on the algorithm, remains undaunted. “Good ideas in science take a very long time to operationalize,” he says.

But life-saving diagnoses shouldn’t.