This paper introduces the Simulated Interaction Task (SIT), a digital tool designed to automatically quantify social interaction deficits, focusing on Autism Spectrum Disorder (ASD). The SIT involves a standardized 7-minute simulated video dialogue, analyzing facial expressions, gaze behavior, and vocal characteristics. In a study of 37 adults with ASD and 43 healthy controls, machine learning achieved 73% accuracy in ASD detection using facial expressions and vocal features. Key indicators included reduced social smiling and mimicry, along with higher voice fundamental frequency and harmony-to-noise ratio. The SIT's automated analysis performed comparably to clinical expert ratings, offering a cost-effective and time-efficient alternative.

Hanna Drimalla, Tobias Scheffer, Niels Landweher, Irina Baskow, Stefan Roepke, Behnoush Behnia, Isabel Dziobek