Trust but Verify: How AI Can Be Used To More Accurately Diagnose Pain

There’s an episode of Scrubs that I think of every time something hurts. In it, the main character JD says something about how pain is complicated for doctors because how people handle it varies greatly between individuals. Indeed, the understanding of the complexity and amorphous nature of pain is something that is shared by the entire human race, and yet we rely on an archaic pain chart with poorly drawn facial expressions and a few descriptors to accurately diagnose pain. Those descriptors clearly place certain adjectives, such as “gnawing” or “burning” at different levels, and yet, if we know that everyone feels and handles pain differently, how can we say for certain which is worse for an individual? We can’t, and that’s why a smart solution to diagnosing pain is so exciting.

Researchers at M.I.T published an article about their algorithm, which they named DeepFaceLIFT, about a year ago in the Journal of Machine Learning Research and in it, they detail how their two-stage hierarchical machine learning algorithm can actually give personalized predictions of pain because it is able to sort for the subjectivity, personalization, and interpretation of pain. As a kid, I suffered from really intense joint pain, and I always wished that the doctors prodding at me could just feel the sensation for themselves so that they could finally figure out what the problem was. While there is still no way to transfer one person’s feelings of pain to another one, this algorithm still proves much more useful than the most widely used measurement system today, the VAS, which is completely self-reported and thus subject to personal bias. With this algorithm, pain gets some objectivity.

Because pain is so personal and computer programs do have limitations in real-world applications, it is unlikely that the algorithm will ever completely replace the old system of self-reporting pain scores.   It has a really good chance of helping to sort out people in real pain from people who are faking it, however, which allows doctors to focus on the most critical patients first.

For anyone who has ever suffered from pain, the inability to accurately describe it is like an old friend. The issue crops up in pop culture and literature, describing a phenomenon we can all relate to. In fact, the experience is so universal that finding references to it is quite the opposite of the needle in the haystack; they’re everywhere. Thanks to machine learning, however, one day people will not need the words that so commonly escape us when trying to describe pain, because a doctor will be able to use this program and get an objective description, and that, is A-I-mazing.


Questions? Thoughts? Concerns? Let us know!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.