Utilizing deep learning models in an intelligent eye-tracking system for autism spectrum disorder diagnosis

Area of Science:

• Neuroscience
• Computer Science
• Medical Diagnostics

Background:

• Autism Spectrum Disorder (ASD) diagnosis relies on subjective methods, necessitating objective and efficient tools.
• Current ASD diagnostic procedures are often time-consuming, costly, and dependent on specialist expertise.
• Advancements in artificial intelligence (AI) and eye-tracking offer potential for automated ASD assessment.

Purpose of the Study:

• To develop and evaluate automated classifiers for early ASD diagnosis using eye-tracking data.
• To explore the efficacy of deep learning algorithms in improving the precision and efficiency of ASD screening.
• To provide a more objective and accessible method for identifying potential cases of ASD.

Main Methods:

• Utilized a dataset of eye-tracking pathways from 328 typically developing children and 219 children with autism.
• Employed image resampling techniques to mitigate overfitting and expand the training dataset.
• Developed automated classifiers using deep learning models: MobileNet, VGG19, DenseNet169, and a MobileNet-VGG19 hybrid.

Main Results:

• The MobileNet model achieved 100% accuracy, significantly outperforming existing systems.
• The VGG19 model demonstrated 92% accuracy in ASD classification.
• Deep learning approaches showed superior performance compared to traditional event detection algorithms.

Conclusions:

• Eye-tracking data, analyzed with deep learning, shows significant potential for efficient and accurate ASD screening.
• Automated classifiers can enhance diagnostic precision, addressing the need for prompt and reliable ASD assessment.
• These AI-driven tools can assist healthcare professionals in improving the accuracy and accessibility of autism diagnosis.