Search

Obsessive compulsive disorders (OCD) are marked by distress, negative emotions, mental processes and behaviors that are reflected in physiological signals such as heart rate, electrodermal activity, and temperature. Continuous monitoring of physiological signals associated with OCD symptoms may make measures of OCD more objective and facilitate close monitoring of prodromal symptoms, treatment progress and risk of relapse. Thus, we explored the feasibility of capturing OCD events in the real world using an unobtrusive wrist worn biosensor and machine learning models. Nine children and adolescents (mean age = $12.3$, SD = $2.6$) with mild to severe OCD were recruited from child and adolescent mental health services. Participants were asked to wear the biosensor in the lab during conditions of rest and exposure to OCD symptom-triggering stimuli and for up to eight weeks in their everyday lives and register OCD events, i.e., every time their OCD symptoms caused distress. We explored the relationships among the amount of physiological data, registered OCD events, age, OCD symptom severity and symptom types. In the machine learning models, we considered detection of OCD events as a binary classification problem. A two-layer cross validation strategy with either random 10-folds, leave-one-person out, or leave-week(s)-out in both layers was used. We compared the performance of four models: logistic regression, random forest (RF), feedforward neural networks, and mixed-effect random forest (MERF). To explore the ability of the models to detect OCD events in new patients, we assessed the performance of participant-based generalized models. To explore the ability of models to detect OCD events in future, unseen data from the same patients, we compared the performance of temporal generalized models trained on multiple patients with personalized models trained on single patients. Eight of the nine participants collected biosensor signals totaling $2,045$ hours and registered $1,639$ OCD events. RF and MERF models outperformed the other models in terms of accuracy in all cross-validation strategies, reaching 70\% accuracy in random and temporal cross validation. Better performance was obtained when generalizing across time compared to across patients. Generalized temporal models trained on multiple patients were found to perform better than personalized models trained on single patients. Our pilot results suggest that it is possible to detect OCD episodes in the everyday lives of youths using physiological signals captured with a wearable biosensor. Large scale studies are needed to train and test models capable of detecting and predicting episodes.

Behavioral observation is an important tool for research and practice in clinical, developmental, and social psychology. Traditional,expert-based methods for behavioral observation from videos are limited by the amount of time and resources it requires. Artificialintelligence tools are available that can automatically and reliably code specific emotions and behaviors. However, the generatedoutput is not specifically aligned to applications in psychology. We present methods for automating behavioral coding using opensource software and provide an overview of available sources and instructions on how they can be applied to generate codes fromwidely used behavioral coding manuals. Using two examples, we describe methods for applying open source computer visionand hearing tools to code behavior in two settings: a parent-child interaction and a psychotherapeutic exposure session. We use amodular approach, defining specific behaviors with features extracted with computer vision and hearing tools that can be combinedinto higher order constructs. Modules can be combined in different ways to allow a flexible application to a variety of codingmanuals or observational tasks. In addition, we propose ways of testing the validity of the automated codes. Some modalities arebetter suited than others for specific tasks. For example, the performance of computer vision tools is excellent for detecting emotion,whereas speech signal processing is less sensitive to detecting specific emotions. However, speech is better suited for detectingvocalization than computer vision. Issues of privacy and bias are discussed as well as methods for mitigating these risks. Weidentified behaviors that can be computationally coded. Our proposed automated methods will enable researchers to code more oftheir video and audio data than is possible with human coding for meaningful behavioral information and investigate mechanisms atplay in social interactions. Some of these methods can be directly applied by psychologists whereas others necessitate collaborationwith computer scientists.