What makes psychotherapy efficient? Searching for the psychotherapy X factor

Ben-David Sela Tal# and Zilcha-Mano Sigal - Department of Psychology
Hagit Hel-Or - Department of Computer Science
Ilan Shimshoni - Department of Information Systems

 Machine Learning
Deep learning
Neural Networks
Computer vision
Image processing


PhD Grant 2020

sad balloon with copy space blue monday


Psychotherapy is one of the primary tools used to treat a variety of mental illnesses, including depression. We tend to believe that psychotherapy is an efficient tool, but actually, psychotherapy is effective for no more than 50% of the individuals seeking treatment (Cuijpers et al., 2014). Thousands of studies have attempted to improve treatment efficacy, but they have failed to produce meaningful and consistent findings (Zilcha-Mano, 2020). It has been argued that the methods employed in psychotherapy research today have reached the limit of their ability to teach us about the factors associated with therapy session efficacy (Lorenzo-Luaces & DeRubeis, 2018). Therefore, it has been suggested that new methods such as an exploratory data-driven approach based on Machine Learning (ML) algorithms have the potential to enhance our understanding of what predicts a treatment session’s efficacy. Thus, our study aims to reveal the components of therapy sessions associated with effective reduction of depressive symptoms by analyzing in-session patients’ emotional response from their facial expressions and eye contact. 

For this purpose, we formed an interdisciplinary team, involving the labs of Prof. Shimshoni from the Information Systems dept and Prof. Hel-Or from the Computer Science dept, and Prof. Zilcha-Mano from the Psychotherapy lab in the dept of psychology, all part of the University of Haifa’s Data Science Research Center.  

We have already processed almost 100 therapy sessions with the OpenFace software, which assists in extracting the movement of each facial muscle. Our challenges today focus on examining patients’ eye contact with the therapist. We are trying to examine this while using only the patient’s video. The facial motion and the gaze information will then be used to train a ML algorithm to detect points of emotional interest in the psychotherapy session.

This research can help us learn what makes a psychotherapy session efficient, by examining in-session interventions and their immediate influence on patient symptoms. We believe that this kind of innovative knowledge may have a meaningful impact on psychotherapy research and practice.