Fractional whale driving training-based optimization enabled transfer learning for detecting autism spectrum...

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Autism Spectrum Disorder (ASD) significantly impacts communication and social interaction.
• Early detection of ASD is crucial for managing developmental outcomes.
• Current diagnostic methods can be improved with advanced computational techniques.

Purpose of the Study:

• To introduce an automated method for identifying Autism Spectrum Disorder (ASD).
• To enhance the accuracy and efficiency of ASD detection using machine learning.
• To develop a novel hybrid optimization algorithm for training deep learning models.

Main Methods:

• Developed the Fractional Whale-driving Training-based Optimization (FWDTBO) algorithm by integrating Fractional Calculus (FC), Whale Optimization Algorithm (WOA), and Driving Training-based Optimization (DTBO).
• Employed Convolutional Neural Network-based Transfer Learning (CNN_TL) utilizing hyperparameters from pre-trained models like AlexNet and ShuffleNet.
• Implemented functional connectivity analysis on extracted nub regions, optimized by the Whale Driving Training Optimization (WDTBO) algorithm, with TL tuned by FWDTBO.

Main Results:

• The proposed FWDTBO-CNN_TL model achieved 90.7% accuracy in ASD detection.
• The method demonstrated high performance metrics: 95.4% sensitivity, 93.7% specificity, and 93% f-measure.
• The technique was validated using the ABIDE-II dataset, showing significant potential for clinical application.

Conclusions:

• The FWDTBO-CNN_TL approach offers a highly accurate and effective method for automated ASD identification.
• This research highlights the potential of hybrid optimization and transfer learning in neurological disorder diagnostics.
• Early and precise detection of ASD can be facilitated by advanced computational models like the one proposed.