Related Experiment Video
Updated: Nov 2, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.6K
Automated ASD detection using hybrid deep lightweight features extracted from EEG signals
Mehmet Baygin1, Sengul Dogan2, Turker Tuncer2
1Department of Computer Engineering, College of Engineering, Ardahan University, Ardahan, Turkey.
Computers in Biology and Medicine
|June 13, 2021
Summary
This study developed an automated autism detection model using electroencephalogram (EEG) signals. The hybrid deep learning approach achieved 96.44% accuracy, offering a valuable tool for early autism diagnosis.
Area of Science:
- Neuroscience
- Medical Technology
- Artificial Intelligence
Background:
- Autism spectrum disorder affects approximately 1 in 54 children.
- Electroencephalogram (EEG) signals exhibit distinct patterns in children with autism.
- Automated detection models can aid in identifying autism spectrum disorder.
Purpose of the Study:
- To design and implement an automated autism detection model using EEG signals.
- To develop a hybrid lightweight deep feature extractor for high classification performance.
- To create an adjunct tool for neurologists in autism diagnosis.
Main Methods:
- A novel signal-to-image conversion using 1D_LBP and STFT to generate spectrograms.
- Hybrid deep feature extraction combining MobileNetV2, ShuffleNet, and SqueezeNet.
- Feature selection using a two-layered ReliefF algorithm and shallow classifiers with 10-fold cross-validation.
Main Results:
- A Support Vector Machine (SVM) classifier achieved 96.44% accuracy.
- The proposed model demonstrated high classification performance on a large EEG dataset.
- Features extracted by the hybrid deep lightweight feature extractor were highly discriminative.
Conclusions:
- The hybrid deep lightweight feature extractor is effective for autism detection via EEG signals.
- The developed model shows potential as an adjunct diagnostic tool in clinical settings.
- This approach facilitates earlier and more accurate autism diagnosis.
