Machine Learning Could Provide a Diagnosis After a Single Cough

If you see a doctor because you’re feeling under the weather, it’s not unusual for said health professional to ask you to cough to get a sense of what’s ailing you. What’s more unusual would be that same diagnosis happening without having to leave home at all — or even contact another human. According to Nature‘s Mariana Lenharo, a recent study showed that Google’s machine learning systems proved to be surprisingly effective in diagnosing illnesses though simply analyzing the sound of a cough.

As Lenharo explains, this Google system — Health Acoustic Representations (or HeAR) — used unlabeled audio data with a focus on the effects of COVID-19, tuberculosis and smoking on the human body. This differs from most methods of using machine learning to diagnose heath issues, which normally feature guidance from researchers as to what underlying conditions result in specific sounds.

The results of HeAR’s efforts sound promising, with the AI technology predicting both tuberculosis and COVID-19 more accurately than machine learning models using a more traditional method. It’s encouraging news — though it’s worth mentioning that, in 2020, a Google AI that showed promise for diagnosing eye diseases hit a few snags when deployed in the field.

How Do You Train Astronauts for a Medical Crisis in Orbit?

A competition in Canada is exploring this very issue

So, will you be able to get a diagnosis simply from coughing into your phone in the coming months? Not necessarily. As Lenharo writes, Google’s next stage will be sharing this technology with others in the medical field to see what they do with it. Or, as Google’s Sujay Kakarmath told Nature, “Our goal as part of Google Research is to spur innovation in this nascent field.”