Deep Learning-Based Feature Selection for Detection of Autism Spectrum Disorder

Ibrahim Nafisah¹Nermine Mahmoud²Ahmed A. Ewees³Mohamed G. Khattap⁴Abdelghani Dahou⁵Safar M. Alghamd⁶Ibrahim A. Fares^7*Mohammed Azmi Al-Betar⁸Mohamed Abd Elaziz^9*

• ¹Department of Statistics and Operations Research, College of Sciences, King Saud University, Riyadh, Kingdom of Saudi Arabia, Riyadh, Saudi Arabia
• ²Psychology and Mental health, Faculty of Human Science, Galala University, Suez, Egypt, Suez, Egypt
• ³Department of Computer, Damietta University, Damietta 34517, Egypt, Damietta, Egypt
• ⁴Technology of Radiology and Medical Imaging Program, Faculty of Applied Health Sciences Technology, Galala University, Suez 435611, Egypt, Suez, Egypt
• ⁵School of Computer Science and Technology, Zhejiang Normal University, Jinhua 321004, China, Jinhua, China
• ⁶Department of Mathematics and Statistics, College of Science, Taif University, P.O. Box 11099, Taif 21944, Saudi Arabia, Taif, Saudi Arabia
• ⁷Faculty of Science, Zagazig University, Zagazig, Egypt
• ⁸Artificial Intelligence Research Center (AIRC), College of Engineering and Information Technology, Ajman University, Ajman, UAE, Ajman, United Arab Emirates
• ⁹Faculty of Computer Science and Engineering, Galala University, Suez 435611, Egypt, Suez, Suez, Egypt

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder characterized by difficulties in communication, social interactions, and repetitive behaviors, with symptoms varying widely across individuals. Neuroimaging techniques, particularly resting-state functional MRI (rs-fMRI), have shown promise in identifying neural signatures of ASD. However, limitations such as high dimensionality, noise, and small sample sizes hinder their broader clinical application. Additionally, the interpretation of large-scale imaging data remains a significant challenge, often making it difficult to derive actionable insights from rs-fMRI data. This paper proposes a novel approach for ASD detection using deep learning and enhanced feature selection (FS) techniques. Specifically, a hybrid model combining a Stacked Sparse Denoising Autoencoder (SSDAE) and Multi-Layer Perceptron (MLP) is employed to extract relevant features from rs-fMRI data. Additionally, an enhanced 1 Nafisah et al. Running Title version of the Hiking Optimization Algorithm (HOA) is used for FS, incorporating Dynamic Opposites Learning (DOL) and Double Attractors to improve convergence towards the optimal subset of features. The proposed approach is evaluated using multiple ASD datasets, and its performance is compared with state-of-the-art techniques. The results highlight the superior performance of our model, achieving an average accuracy of 0.735, sensitivity of 0.765, and specificity of 0.752. These findings demonstrate the model's effectiveness in ASD detection.