Deep learning based approach for Behavior classification in diagnoses of Autism Spectrum Disorder using naturalistic...

Area of Science:

• Neuroscience
• Computer Science
• Developmental Psychology

Background:

• Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by social communication deficits and repetitive behaviors.
• Current ASD diagnosis relies on psychological screening and behavioral observation, particularly of repetitive actions like hand-flapping.
• Analyzing everyday activities in unstructured settings offers a novel approach to identifying potential ASD indicators.

Purpose of the Study:

• To investigate the potential of deep learning models for detecting Autism Spectrum Disorder (ASD) through video analysis of children's behaviors.
• To evaluate the performance of various deep learning architectures in classifying autistic behaviors from video data.
• To develop a robust and accurate system for early ASD screening and behavioral monitoring.

Main Methods:

• Utilized the publicly available Self-Stimulatory Behavior Dataset (SSBD) for classifying autistic behaviors.
• Pre-processed video data including region-of-interest detection and image cropping.
• Applied data augmentation techniques to enhance model generalization and training efficiency.
• Tested and compared deep learning models: CNN-GRU, 3D-CNN + LSTM, MobileNet, VGG16, and EfficientNet-B7 for spatiotemporal feature extraction.

Main Results:

• The proposed CNN-GRU model demonstrated superior performance compared to all other tested deep learning methods.
• Achieved a consistent accuracy of 0.9284 ± 0.0039 to 0.9294 ± 0.0038 with k-fold cross-validation, indicating robustness.
• The model effectively predicted behaviors in real-life, uncontrolled video settings.

Conclusions:

• The developed CNN-GRU model shows significant promise as an effective tool for the early screening and monitoring of Autism Spectrum Disorder (ASD).
• Action recognition systems based on this approach can assist clinicians in tracking behavioral trends for quicker, more accurate ASD assessment.
• The system's effectiveness in real-world scenarios highlights its potential for clinical implementation as a decision-support tool.