Kolloquium: Gastvortrag am 22.06.22 mit Dr. Katharina Schultebraucks, Columbia University/ NYU Grossman School of Medicine

Im Rahmen unseres Kolloquiums laden wir Interessierte recht herzlich zu einem Vortrag von Dr. Katharina Schultebraucks zum Thema „Predicting Posttraumatic Stress and Resilience using Machine Learning, Computer Vision and Voice Analysis“ ein.

Der Vortrag findet am 22. Juni 2022 von 16.15 Uhr bis 17.45 Uhr  über Zoom statt: https://fau.zoom.us/j/8615734326


Approximately 40 million Emergency Department (ED) visits each year are due to exposure to potentially life-threatening events leading not just to physical morbidity but also to significant adverse mental health effects. Up to 30% of ED patients exposed to potential trauma report moderate-to-high symptom severity of post-traumatic stress disorder (PTSD) one year after discharge. The time immediately after trauma exposure offers a critical window for PTSD prevention, yet blanket provision of intervention is neither indicated nor cost-effective. Strong evidence shows that the individual response to trauma exposure is heterogeneous as are the underlying biological pathways.

I will present findings using routinely collectible data from Electronic Health Records (EHRs) to determine the risk of non-remitting PTSD in ED patients in the “golden hours” after trauma. Leveraging EHR data provides a feasible, minimally disruptive, and clinically important opportunity for prognosticating PTSD symptoms.

Besides using routinely available data in the acute care setting, I will present findings using a deep learning approach combining multiple polygenic risk scores for distinguishing resilience and depression after major life stressors.

Furthermore, I will present an application of deep learning to identify transdiagnostic markers of maladaptive stress responses by using digital phenotyping. This advanced machine learning approach used computer vision and voice analysis for extracting facial landmark features of emotions from video-recorded clinical interviews in combination with features of voice prosody, pupil dilatation, head movements, and speech content analyzed using Natural Language Processing. The extracted features were combined using deep learning to develop digital biomarkers as predictors of stress pathologies and neurocognitive performance after trauma exposure.

In the future, digital biomarkers could improve the scalability and sensitivity of clinical assessments using passive patient evaluations. In view of these recent advancements in prognosticating risk using computational methods, I will conclude by discussing the clinical implications of the future use of machine learning approaches for predicting and monitoring posttraumatic stress and resilience.