Behavioral and Neuronal Correlates of Human Mood States - PROJECT SUMMARY Optimizing treatments in mental health requires an easy to obtain, continuous, and objective measure of internal mood. Unfortunately, current standard-of-care clinical scales are sparsely sampled, subject to recency bias, underutilized, and are not validated for acute mood monitoring. The recent shift to remote care also requires novel methods to measure internal mood. Recent advances in computer vision have allowed the accurate quantification of observable speech patterns and facial representations. The continuous and objective nature of these audio-facial behavioral outputs also enable the study of their neural correlates. Here, we hypothesize that video-derived audio-facial behaviors have discrete neural representations in the limbic network and can provide a critical set of reliable longitudinal estimates of mood at low cost across home and clinic settings. To test our hypothesis, we will enroll ten surgical epilepsy patients with comorbid major depression already undergoing surgical implantation for clinical reasons. We will obtain simultaneous video-derived audio-facial features, invasive brain recordings, and frequent assessments of mood. In Aim 1, we will combine continuous video recordings with frequent mood assessments to build a longitudinal mood prediction model using video- derived audio-facial features. In Aim 2, we will identify neural correlates of audio-facial dynamics using synchronized intracranial EEG under spontaneous and task settings. In Aim 3, we will use high frequency direct electrical stimulation to determine the causal influence of limbic activity on audio-facial features and internal mood. The result of this study is a mood-decoding model based on audio-facial behavioral features that are causally linked to limbic activity and mood. This model will allow for objective, passive, longitudinal, remotely-enabled, and low-resource measurements of internal mood. Multiple use cases exist to significantly advance psychiatric care, ranging from acute mood tracking to optimize rapidly-acting interventions such as psychedelics and neurostimulation, to longitudinal telehealth-enabled monitoring for suicide risk monitoring, treatment dose optimization, and relapse prediction. Across inpatient, outpatient, and at-home settings, this model can deliver important real-time mood and electrophysiological insights to both patients and providers. Relevance to NIMH RDOC Matrix: Negative Valence Systems: Construct: Loss; Circuit: vmPFC, Parietal cortex, default mode network; Physiology: LFP; Behavior: Rumination and Sadness.
