Nowadays, doctors are told to regularly watch out and screen for depression, a mental health disorder which affects 300 million people worldwide.
However, depression is a difficult disorder to diagnose. For example, when a medical professional is attempting to diagnose a patient, they rely on methods such as blood tests or other lab tests. But, these methods have proven to be rather unsuccessful.
Instead, talking with a patient is likely to result in a better understanding of their mental health. However, making sure that the right questions are asked and the answers are well understood is a big challenge as they differ for every patient.
In order to combat this problem, a Massachusetts Institute of Technology (MIT) research team is using artificial intelligence (AI) to detect depression and take a little of the strain off doctors.
The researchers’ work is a neural-network AI model pilot which can predict depression from natural conversations. This AI method works by looking for patterns in a patient’s speech from text transcriptions and audio. The method doesn’t take all of the work from a doctor but can analyze the results of a conversation.
To develop the AI model, the MIT researchers took 142 recorded patient interviews and aimed the neural-network at detecting depression.
They then tested this model. One of the first tests was to check how accurate a prediction could be for text and audio features. Next, they worked on understanding predictive performance “when conditioning on the type of question asked, and independent of the time it was asked during the interview session.” Finally, the team tested “modeling temporal changes of the interview”.
Although there is still some work to do on this model before it can accurately diagnose depression, it’s certainly an interesting use of the technology and could one day be used to save many lives.
Source: Psychology Today
License: The text of "MIT Artificial Intelligence Predicts Depression from Patient’s Speech" by All3DP is licensed under a Creative Commons Attribution 4.0 International License.