KI-Expositionstherapeut*in für Angststörungen

*Wider research context.* Anxiety-related disorders - such as phobias, PTSD, social anxiety - are highly prevalent and pose a great burden on society. First-line treatment for such disorders is exposure therapy (ET), which entails safely exposing patients to the source of their anxiety. Although ET is often the best available choice, there is room for improvement: many patients do not respond to ET or experience residual symptoms and relapses. Scarcity of therapists - especially in low- and middle-income countries - also underscores the need for scalable, algorithmic versions of ET. This need has been made all the more urgent by the dramatic surge in anxiety disorders due to the COVID-19 pandemic. *Objectives.* Fundamentally, optimizing ET protocols requires solving two entwined problems: (i) estimating the patient’s anxiety level, and (ii) adaptively choosing experiences to which to expose the patient. This project addresses both of these problems, first separately, and then integrated into a novel bio-adaptive exposure therapy (B-ADEPT) paradigm. *Methods.* B-ADEPT paradigm will initially be developed using a lab analog of ET - extinction learning (EL) - whereby a mild aversive association is first experimentally induced in participants, and then extinguished. Finding optimized EL protocols is a high-dimensional search problem, with complexly interacting variables. To efficiently search this high-dimensional space, this project will use state-of-the-art artificial intelligence (AI) and biosignal analysis methods. Deep reinforcement learning AI algorithms will optimize the exposure procedure, by sequentially learning from each participant, and real-time psychophysiological models will improve the estimation of participants' anxiety levels. Finally, an outpatient pilot study will be conducted to test the translational potential of B-ADEPT. *Level of innovation.* The current research approach of optimizing ET by studying the effect of at most few variables using between-group trials, has produced mixed results, despite large efforts. The proposed BADEPT approach is profoundly different, and potentially vastly more efficient. Although aimed at anxiety disorders, B-ADEPT can potentially serve as a blueprint for optimizing behavioral therapies in general. *Primary researchers involved.* The PI, Dr. Melinscak, has extensive experience with applied statistical optimization (required for optimizing ET protocols) and model-based psychophysiology (required for optimizing the estimation of patient’s anxiety level). The main mentor, Prof. Scharnowski, is a leading expert on the real-time and closed-loop biosignal analysis. His expertise will facilitate the transfer of real-time techniques to model-based psychophysiology. Collaborators on the project - Prof. Tschiatschek, Mr. Lanzinger, and Prof. Dayan - will provide further expertise on applying technological means (AI, virtual reality, biofeedback) in the mental health domain.