A Deep Learning Framework for Early Autism Detection Using EEG Signals

Maha M. Hamzeh, Department of Computer Science, College of Science for Women, University of Babylon, Babylon, IraqFollow
Ali Y. Al-Sultan, Department of Computer Science, College of Science for Women, University of Babylon, Babylon, Iraq
Salah Al-Obaidi, Department of Computer Science, College of Science for Women, University of Babylon, Babylon, Iraq

ORCID

Maha M. Hamzeh: https://orcid.org/0009-0009-5113-5769

Article Type

Original Study

Abstract

Autism spectrum disorder (ASD) is a complicated neurodevelopmental illness, affecting social interaction, communication, and cognitive function. To lower healthcare costs and facilitate prompt intervention, early and accurate detection is crucial. However, behavioral assessments—which are inherently subjective and can lead to delayed diagnoses—are a significant component of traditional diagnostic procedures. This paper presents a convolutional neural network (CNN) and time-frequency analysis-based early ASD screening using EEG signals. EEG data undergoes a preprocessing step to remove noise and power interference. After that, the signals were divided into 5-, 10-, and 20-second time frames. Each EEG time frame segment is represented in the time-frequency domain using the Short-Time Fourier Transform (STFT). All the spectrograms of all EEG channels for a specific time interval were then averaged to provide a single time-frequency representative image… These spectrogram images were classified using a pre-trained Efficient Net. The dataset comprises 52 samples of 19-channel EEG recordings. The proposed method achieved a highest classification accuracy of 99.1% with 20-second length segments. The proposed approach outperformed existing methods, earning an 8% improvement, and demonstrated its potential for robust, non-invasive, and objective early ASD detection via EEG signal visualization and deep learning.

Keywords

Autism spectrum disorder (ASD), Electroencephalography (EEG), Deep learning (DL), Convolutional neural networks (CNN), Early detected autism, Short-time fourier transform (STFT)

How to Cite This Article

Hamzeh, Maha M.; Al-Sultan, Ali Y.; and Al-Obaidi, Salah (2025) "A Deep Learning Framework for Early Autism Detection Using EEG Signals," Journal of Intelligent Informatics, Networking, and Cybersecurity: Vol. 1: Iss. 1, Article 5.
Available at: https://jiinc.uobabylon.edu.iq/journal/vol1/iss1/5