Hybrid deep learning and feature selection approach for autism detection from rs-fMRI data.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychology

Background:

• Autism Spectrum Disorder (ASD) presents complex diagnostic challenges due to diverse manifestations and data limitations.
• Conventional machine learning methods struggle with high-dimensional neuroimaging data for ASD detection.
• Psychologists play a vital role in understanding ASD's cognitive, emotional, and behavioral aspects.

Purpose of the Study:

• To enhance Autism Spectrum Disorder (ASD) diagnosis by integrating deep learning (DL) for feature extraction.
• To improve diagnostic performance using a modified exponential-trigonometric optimization (ETO) algorithm for feature selection (FS).
• To evaluate a novel DL-based model incorporating Arithmetic Optimization Algorithm (AOA) and Guided Learning Strategy (GLS).

Main Methods:

• Utilized resting-state functional MRI (rs-fMRI) data from the Autism Brain Imaging Data Exchange (ABIDE I).
• Employed deep learning (DL) techniques for automated feature extraction from neuroimaging data.
• Implemented a modified exponential-trigonometric optimization (ETO) algorithm, integrating AOA and GLS, for feature selection (FS).

Main Results:

• The proposed DL model achieved superior performance compared to benchmark methods in diagnosing ASD.
• Demonstrated high accuracy (73%), sensitivity (78%), and Area Under the Curve (AUC) (79%) on average.
• The model effectively handled high-dimensional rs-fMRI data, overcoming limitations of conventional approaches.

Conclusions:

• The novel deep learning approach significantly improves the accuracy and efficiency of Autism Spectrum Disorder (ASD) diagnosis.
• The integration of DL with advanced optimization algorithms offers a promising direction for neurodevelopmental disorder detection.
• This method provides a robust tool for analyzing complex neuroimaging data in ASD research and clinical applications.