An Improved AlexNet Model and Cepstral Coefficient-Based Classification of Autism Using EEG.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Autism spectrum disorder (ASD) is a neurodevelopmental condition requiring early intervention for improved outcomes.
• Current ASD identification relies on subjective methods prone to observer variability.
• Machine learning and deep learning offer promising avenues for objective and early ASD detection.

Purpose of the Study:

• To evaluate deep learning networks (AlexNet, VGG16, ResNet50) for autism spectrum disorder (ASD) detection.
• To investigate the efficacy of cepstral coefficient features in ASD identification.
• To enhance ASD classification accuracy through architectural modifications of deep learning models.

Main Methods:

• Utilized cepstral coefficients (specifically Linear Frequency Cepstral Coefficients - LFCC) to generate spectrograms.
• Evaluated standard deep learning architectures: AlexNet, VGG16, and ResNet50.
• Developed a modified AlexNet architecture for improved ASD classification.

Main Results:

• Standard AlexNet with LFCC achieved 85.1% accuracy in ASD detection.
• A customized AlexNet architecture incorporating LFCC features reached 90% accuracy.
• Cepstral coefficients proved effective features for deep learning-based ASD identification.

Conclusions:

• Deep learning models, particularly customized AlexNet, demonstrate high potential for accurate and early ASD detection.
• The integration of cepstral coefficients enhances the performance of deep learning models for ASD identification.
• Objective, data-driven approaches like deep learning can overcome limitations of subjective ASD diagnostic methods.