AI Can Assist Physicians in Unprecedented Ways

Artificial intelligence (AI) may still be an emerging technology, but that hasn’t stopped it from reshaping numerous aspects of our society already. To see this in action, look no further than its influence on consumer devices, internet services, and autonomous vehicles.

Deep learning, a subset of AI development, is breaking new ground in a field infamous for lagging behind innovation: healthcare. And it may be just what the doctor ordered.

A Panacea for Human Biases?

When it comes to diagnosis, physicians often try to take a systematic approach in identifying illnesses and ailments. But all too often, bias manages to creep in. Alternative possibilities are ignored or dismissed. And as a result, millions of Americans return from a trip to the doctor with a misdiagnosis.

Fortunately, researchers in China and the United States seem to have found a solution to this common problem: AI. In a paper recently published in Nature Medicine, the group of scientists reported their results with a new AI system they constructed.

After analyzing a variety of patient data (lab results, symptoms, medical history, etc.), this new AI system can automatically diagnose common childhood maladies like meningitis and influenza. It has shown itself to be extremely accurate. And the researchers think that it could one day help doctors diagnose complex conditions better.

The Insights Are All in the Data

The new AI system was trained on the data of approximately 600,000 Chinese patients who visited a pediatric hospital during an 18-month period. This shines a light on a clear-cut advantage China has in the race to become a global AI leader.

Not only does the country possess an incredibly large population, but it also places fewer restrictions on digital data sharing. In turn, it’s much easier for Chinese AI companies and researchers to access the data they need to train deep learning systems.

Recently, President Donal Trump signed an executive order intended to accelerate AI development in the United States. Coincidentally (or maybe not), a large part of this new American AI Initiative revolves around encouraging government agencies and academia to share data in the name of expediting progress.

Unfortunately for AI researchers, amassing large amounts of health care data is incredibly difficult in America. To put this in perspective, the researchers behind the new deep learning system we’ve discussed only had to visit one hospital in China to get all the data they needed. This data-gathering endeavor would never be so straightforward in American facilities.

Dr. George Shih, co-founder of MD.ai and associate professor of clinical radiology at Weill Cornell Medical Center, elaborates on the difficulties: “You have to go to multiple places. The equipment is never the same. You have to make sure the data is anonymized. Even if you get permission, it is a massive amount of work.”

Expanding AI’s Horizons to Healthcare

After spending the better part of this decade disrupting several industries, deep learning is now entering various areas of healthcare. Tech titans like Google and other famous San Francisco developers are doubling down on AI’s abilities to analyze and comprehend.

Some of the works-in-progress include systems that can examine electronic health records and flag medical conditions like diabetes, heart failure, and hypertension. Others analyze different forms of medical imagery such as X-rays, eye scans, and M.R.I.s to deduce diseases.

The new system that focuses on diagnosing childhood ailments was built by a team led by Dr. Kang Zhang, University of California San Diego’s chief of ophthalmic genetics. It relies on neural networks, algorithms composed of interconnected and conceptualized “artificial neurons”, to get the job done.

Prior to this, Zhang and his team built systems that could analyze eye scans for signs of diabetic blindness. This new system is much more versatile. Not only can it diagnose a wider range of conditions, but it can also recognize patterns in text instead of just medical images.

Zhang hopes this system can augment doctors’ abilities. “In some situations, physicians cannot consider all the possibilities,” he explains. “This system can spot-check and make sure the physician didn’t miss anything.”

So far, the results have been astounding, to put it mildly. When tested on unlabeled data, the system can perform just as well as experienced physicians. For example, in diagnosing asthma, a study found that physicians were 80 to 94% accurate. The system maintained an accuracy rate of over 90%.

A New Era of Deep Learning Diagnostics Is Coming

Neural networks have extraordinary pattern recognition capabilities that easily surpass any human’s. But why they make certain decisions and how they teach themselves continues to elude experts. Extensive clinical trials and testing would be needed to prove to doctors and patients that these systems are dependable.

With that being said, it could still be a number of years before we see deep learning systems enter clinics and hospitals for widespread use. It’s also much more likely that countries besides the United States will embrace this emerging technology first. For instance, doctors are limited in countries like India and China. Automated screening and diagnosing systems would be a huge help there.