Core Concepts
AI model PTSD-MDNN offers objective and rapid PTSD diagnosis using video and audio inputs.
Abstract
The article introduces PTSD-MDNN, a deep neural network fusion model for detecting post-traumatic stress disorder (PTSD) by merging two unimodal convolutional neural networks. By utilizing only videos and audios as inputs, the model aims to provide a more objective and faster way of diagnosing PTSD. This approach could be beneficial in teleconsultation sessions, optimizing patient journeys, or human-robot interactions. Traditional diagnostic methods for PTSD involve health professionals using questionnaires, which may have limitations due to biases in self-reported data. The recent global traumatic event of the COVID-19 pandemic has highlighted the need for more objective diagnostic tools given the shift towards remote consultations. The study focuses on AI models for diagnosing PTSD, emphasizing the importance of structured data analysis and leveraging advancements in AI technology to improve diagnostic accuracy.
Stats
In France, PTSD affects 1-2% of the population.
The base "PTSD in-the-wild" contains 634 balanced videos for detecting PTSD.
The accuracy of the video classification is 0.89.
Fusion of video and audio modalities with L2 regularization achieved an accuracy of 0.92.
Quotes
"We propose PTSD-MDNN, a model that fuses audio and video modalities for detecting PTSD."
"Our model surpasses unimodal models by being non-invasive and processing sensitive patient information at low levels."