On the Automatic Detection of Comorbidity of Mental Disorders Using Audio-Visual Data

Abstract

It is reported that a patient with a mental disorder may develop other mental disorders over time. The condition of the simultaneous existence of multiple disorders is referred to as comorbidity. The treatment for a mental disorder is required to be modulated in the case of comorbidity. Thus, the detection of comorbidity is necessary. The comorbidity between major depressive disorder (MDD) and post-traumatic stress disorder (PTSD) is well investigated through several modalities, which are either intrusive or costly. Recently, the audio-video modality has been explored to detect several mental disorders for being non-intrusive and cost-effective. The comorbidity detection between MDD and PTSD on audio-video modality is yet to be reported. In this study, we present initial work on the detection of comorbidity between MDD and PTSD on a publicly available audio-video dataset. To detect MDD/PTSD comorbidity separately, we created two overlapping subsets of the dataset comprising participants diagnosed with MDD/PTSD. The overlapping part in the two subsets comprises data of participants diagnosed with both MDD and PTSD. The best performance for MDD comorbidity detection is found to be 0.789 in terms of macro-averaging F1score. In contrast, it is found to be 0.647 for PTSD comorbidity detection.