Autonomous cars are controversial, but proponents point out that they are far safer than traditional cars because the artificial intelligence systems controlling them are not vulnerable to human error. The computer, they say, won’t fall asleep at the wheel, get distracted checking its text messages, or drive while intoxicated. It’s better at math than humans, accurately calculating when to slow down or change lanes safely. Now this same technology could bring new levels of accuracy to the medical field and even save lives.

Deep learning computers developed by Case Western Reserve University have been handily outperforming medical doctors in detecting cancer and diagnosing heart failure. With $9.5 million in funding from the National Cancer Institute, they’ve developed tools that can analyze digital pathology images of cancers of the lung, head, neck and breast to accurately identify those patients who could be spared undergoing risky chemotherapy.

For example, the researchers have found that their system is better than human experts in distinguishing malignant and benign lung nodules using CAT scans. This is a very valuable advancement when you consider the fact that around 98 percent of the nodules flagged by human radiologists as being suspicious turn out to be benign. The system could save people a lot of worry, not to mention unnecessary additional testing.

The artificial intelligence system is so advanced that it predicted which people out of a group of 105 patients showed evidence of a pending heart failure with a 97 percent accuracy rate; the two pathologists who assessed the same group were only correct 73 and 74 percent of the time.

In another study that looked at prostate cancer scans in three countries, the computation imaging algorithms caught clinically significant prostate cancer in MRIs in more than 70 percent of the cases where radiologists failed to detect it. Similarly, in half of the cases where radiologists had mistakenly identified someone as having clinically significant prostate cancer on MRIs, the machines correctly determined that the disease was not present.

Deep learning AI could prove to be a lifesaver

All of this is extremely promising, but the researchers would like to see further validation in bigger studies. Nevertheless, there’s a huge amount of potential here to spare people unnecessary treatment and better identify those who need interventions.

The machines are excellent at comparing and contrasting hundreds of tissue samples in the same amount of time that a pathologist needs to pore over a single slide, and they can compile data and make predictions with ease.

That doesn’t mean that doctors should start worrying about losing their jobs to machines any time soon, however. While the computers excel at doing things quickly and in high volumes with a great degree of accuracy, there will always be the need for doctors to manage people’s health. What it can do, however, is make their work easier and let them know which patients they need to focus on and which cases aren’t as concerning as they might appear.

This technology, therefore, isn’t going to replace doctors; it will just help them do their work better than before. Case Western Professor of Biomedical Engineering Anant Madabhushi said that in a country like Botswana, where there is only one pathologist for 2 million people, it could allow many more individuals to get much-needed help.

Of course, like any technology, there are ways it could go wrong, but the high accuracy rates that have been demonstrated so far show that this technology could prove to be a lifesaver.

Sources for this article include:

TheDaily.Case.edu

OnlineLibrary.Wiley.com