Abstract: Autistic Spectrum Disorder (ASD) is a neurodevelopmental disorder that leads to complexities in social interactions. Based on the statistics of the WHO, more patients diagnosed with ASD is steadily increasing. Most of the recent investigations focused on medical treatment and data collection, but they did not focus on ASD diagnosis related to Deep Learning (DL). To solve this issue, this work introduces an efficient model for ASD detection with multi-modal data utilizing Convolutional LeNet Forward Taylor network (CLeFTNet). Here, the CLeFTNet is developed by combining CNN and LeNet using the Taylor series. This work involves two kinds of inputs: First, an input brain image is pre-processed by an Adaptive Wiener filter (AWF) and Region of Interest (RoI) extraction. Then, the Box neighborhood search algorithm is employed for extracting the pivotal region regarding functional connectivity, and feature extraction is performed. Simultaneously, input autism data is normalized using Min–Max normalization, and then Matusita and Chord Distance is deployed for feature selection. Thereafter, the Synthetic Minority Oversampling Technique (SMOTE) is considered for data augmentation. The results from the above two processes are employed for detecting ASD by considering CLeFTNet. The analytic metric, namely accuracy, sensitivity and specificity attained 91.723%, 91.396% and 91.573%.

Key words: Autism spectrum disorder')">Autism spectrum disorder, Matusita distance, Chord distance, Convolutional neural network, LeNet