An Effective LRSF-DLNN-Based Autism Spectrum Disorder Prediction Using EEG and fMRI

In today's society, autism spectrum disorder (ASD) is a common neurological disorder that affects a person's behavior and communication. Hence, an early ASD prediction is essential for improving the lifecycle of ASD patients. Recently, many works have been deployed for diagnosing ASD using the electroencephalogram (EEG) and magnetic resonance imaging (MRI). But they are inefficient owing to the irrelevant noises and misclassification outcomes. Hence, a well-ordered logistic regression with scaled function-based deep learning neural network (LRSF-DLNN)-based ASD prediction using EEG and functional MRI (fMRI) is proposed. Initially, the inputs, like the EEG signal and fMRI image of the eye, are gathered and then subjected to the preprocessing phase. Here, for performing noise removal of EEG, the cosine-based Butterworth filter (CBF) approach is established. Similarly, the fMRI undergoes slice time correction and a GF filter-based smoothing process. Then, the preprocessed EEG is decomposed by utilizing weighted penalty factor-centric variational mode decomposition (WPFVMD), and the process of alpha and theta band estimation is performed on the basis of single-scale time dependent-based event-related spectral perturbation (SSTD-ERSP). Likewise, the features are extracted from both the decomposed EEG and smoothed fMRI. Next, the extracted features are given for feature selection (FS) utilizing distance functional green anaconda optimization (DFGAO). Lastly, the optimal features are given to the LRSF-DLNN classifier, which efficiently predicts ASD. Hence, the evaluation outcomes exposed that the proposed technique achieved better performance with a higher accuracy (98.8%) than other prevailing models.