A Hybrid Model for Bipolar Disorder Classification from Visual Information

Bipolar Disorder (BD) is one of the most prevalent mental illnesses in the world. It has a negative impact on people's social and personal functions. The principal indicator of BD is the extreme swing in the mood ranging from manic to depressive states. This paper addresses the challenge of detecting the BD states by monitoring affective information extracted from video recordings of structured interviews. Our goal is to classify the condition of patients suffering from BD into the clinically significant states of remission, hypo-mania, and mania. To this end, we apply a Convolutional Neural Network (CNN) model to extract facial features from video signals. We supply the features' sequence to a Long-Short-Term Memory (LSTM) model to resolve the BD state. Our framework achieved promising results on the development set of the Turkish Audio-Visual Bipolar Disorder Corpus with the Unweighted Average Recall (UAR) of 60.67%.