View in EDS

DSVTN-ASD: Detection of Stereotypical Behaviors in Individuals with Autism Spectrum Disorder using a Dual Self-Supervised Video Transformer Network.

Saved in:
Bibliographic Details
Title: DSVTN-ASD: Detection of Stereotypical Behaviors in Individuals with Autism Spectrum Disorder using a Dual Self-Supervised Video Transformer Network.
Authors: R., Asmetha Jeyarani1 (AUTHOR) asmetha.r@gmail.com, Senthilkumar, Radha1 (AUTHOR) radhasenthil@annauniv.edu
Source: Neurocomputing. Apr2025, Vol. 624, pN.PAG-N.PAG. 1p.
Subjects: Receiver operating characteristic curves, Autism spectrum disorders, Child behavior, Anomaly detection (Computer security), Dual diagnosis
Abstract: Early detection of Autism Spectrum Disorder (ASD) is essential for improving the standard of life for those with autism. Traditional diagnostic approaches in hospitals are challenging and expensive, due to time constraints, difficulties in contextual understanding, the intermittent nature of behaviors, and clinicians' subjective observations. Recognizing the limitations of traditional methods, there is a rising interest in using technology, such as video analysis, to determine stereotypical behaviors associated with autism. Existing activity recognition techniques using supervised learning are based on the datasets with labels. However, this is practically impossible due to the difficulty of precisely labeling the recurring behaviors and the size of the dataset. To alleviate this, a self-supervised model called the Dual Self-Supervised Video Transformer Network (DSVTN-ASD) is proposed with three proxy tasks to enhance its learning capabilities. This model is trained using a few samples of labeled autism videos, and unlabeled video samples of normal and ASD behavior. During inference, pseudo-labels are generated for the unlabeled samples that help in anomaly detection. The pose key point extraction phase and repeated behavior detection are incorporated to predict each pose's key points and identify abnormal behavior. The experimentation is validated on the extended Self-Stimulatory Behavior Dataset (SSBD) and achieves Micro and Macro Area Under the Receiver Operating Characteristic Curve (AUROC) of 95.01 % and 93.13 %, respectively. This AUROC value is better than the other unsupervised strategies in the literature, and thus it can serve as a benchmark for carrying out work based on self-supervised learning. • This work presents a Dual Self-Supervised Video Transformer Network to detect stereotypical behaviors in ASD children. • Self-supervised learning was utilized for finding the anomalies in the video. • Proxy tasks are employed as the pretext tasks for anomaly detection. • Model performance was evaluated by comparing the different number of frames. • Model achieves good results on the Enhanced Self-Stimulatory Behavior Dataset (ESSBD). [ABSTRACT FROM AUTHOR]
Copyright of Neurocomputing is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Early detection of Autism Spectrum Disorder (ASD) is essential for improving the standard of life for those with autism. Traditional diagnostic approaches in hospitals are challenging and expensive, due to time constraints, difficulties in contextual understanding, the intermittent nature of behaviors, and clinicians' subjective observations. Recognizing the limitations of traditional methods, there is a rising interest in using technology, such as video analysis, to determine stereotypical behaviors associated with autism. Existing activity recognition techniques using supervised learning are based on the datasets with labels. However, this is practically impossible due to the difficulty of precisely labeling the recurring behaviors and the size of the dataset. To alleviate this, a self-supervised model called the Dual Self-Supervised Video Transformer Network (DSVTN-ASD) is proposed with three proxy tasks to enhance its learning capabilities. This model is trained using a few samples of labeled autism videos, and unlabeled video samples of normal and ASD behavior. During inference, pseudo-labels are generated for the unlabeled samples that help in anomaly detection. The pose key point extraction phase and repeated behavior detection are incorporated to predict each pose's key points and identify abnormal behavior. The experimentation is validated on the extended Self-Stimulatory Behavior Dataset (SSBD) and achieves Micro and Macro Area Under the Receiver Operating Characteristic Curve (AUROC) of 95.01 % and 93.13 %, respectively. This AUROC value is better than the other unsupervised strategies in the literature, and thus it can serve as a benchmark for carrying out work based on self-supervised learning. • This work presents a Dual Self-Supervised Video Transformer Network to detect stereotypical behaviors in ASD children. • Self-supervised learning was utilized for finding the anomalies in the video. • Proxy tasks are employed as the pretext tasks for anomaly detection. • Model performance was evaluated by comparing the different number of frames. • Model achieves good results on the Enhanced Self-Stimulatory Behavior Dataset (ESSBD). [ABSTRACT FROM AUTHOR]
ISSN:09252312
DOI:10.1016/j.neucom.2025.129397