A deep learning algorithm-based visual strategy intervention study for children with autism spectrum disorders -...

Area of Science:

• Neurodevelopmental disorders
• Artificial Intelligence in Healthcare
• Behavioral Science

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition with unclear etiology and no targeted drug treatments for core symptoms.
• Effective interventions are crucial for improving social integration and quality of life for children with ASD.

Purpose of the Study:

• To develop and evaluate a novel visual strategy intervention for children with autism spectrum disorder.
• To enhance social integration through improved detection of abnormal behaviors.

Main Methods:

• A visual strategy intervention combining spatial-temporal feature fusion and abnormal behavior detection.
• Utilized MotionNet for behavioral feature extraction and an Optical Flow Feature (OFF) subnetwork for enhanced temporal feature analysis.
• Implemented a behavior detection method using an attention mechanism and clustering pool for dynamic behavior analysis in complex video data.

Main Results:

• The proposed model demonstrated superior performance in abnormal behavior detection compared to existing methods, achieving 92.57% accuracy with RGB frames.
• The method showed significant advantages in detecting abnormal behaviors in children, outperforming comparison descriptors by up to 3.64%.
• Experimental validation was conducted on SDUFall, Weizmann, and HMDB51 datasets.

Conclusions:

• The developed visual intervention method is effective in helping children with autism spectrum disorder overcome social barriers.
• This AI-driven approach offers a promising avenue for supporting social integration in children with ASD.