At the Massachusetts Institute of Technology (MIT), researchers have developed a neural network model capable of analyzing text and audio data from various clinical interviews to find patterns that indicate depression.
According to a statement from MIT, these methods could be used to develop improved ways for doctors to diagnose depression and detect signs of depression in normal conversation.
In recent years, automatic learning has been a kind of artificial intelligence (AI) that has been very useful for this type of diagnosis, but beyond the recognition of words and intonations in a person’s language, this model usually tends to indicate that a person is depressed or not, depending on certain answers, which makes them dependent on the type of question asked.
In a paper presented at Interspeech 2018, MIT researchers describe a neural network model that can use raw audio and text data from interviews to detect speech patterns that indicate depression. In this sense, the neural network predicts whether a new subject is depressive without the need for additional questions or answers.
The researchers hope that this new method will enable them to develop tools to detect signs of depression in daily conversation. For example, this model could allow the creation of mobile applications that monitor the text and voice of users to search for depression patterns and send alerts, although this would, of course, be an invasion of privacy that would need to be assessed first.
This technology could be used, for example, to detect mental problems during occasional conversations in doctors’ practices, as each patient speaks and behaves differently and current methods are based on mechanical questions that are always asked in the same way.