A spectrogram image based intelligent technique for automatic detection of autism spectrum disorder from EEG
Md Nurul Ahad Tawhid1, Siuly Siuly1, Hua Wang1
1Institute for Sustainable Industries & Liveable Cities, Victoria University, Melbourne, Victoria, Australia.
Plos One
|June 25, 2021
Summary
This study developed an efficient deep learning framework using electroencephalography (EEG) spectrograms for autism spectrum disorder (ASD) detection. The deep learning model achieved 99.15% accuracy, outperforming traditional machine learning methods for automated ASD diagnosis.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Artificial Intelligence
Background:
- Autism spectrum disorder (ASD) diagnosis relies on subjective clinical evaluation of EEG data.
- Current EEG analysis is time-consuming, costly, and prone to errors.
- There is a need for objective, automated methods for early ASD detection.
Purpose of the Study:
- To develop an automated diagnostic framework for ASD using EEG signals.
- To compare the efficacy of machine learning (ML) and deep learning (DL) models for ASD detection from EEG spectrograms.
- To identify potential EEG biomarkers for ASD.
Main Methods:
- EEG signals were pre-processed (re-referencing, filtering, normalization).
- Short-Time Fourier Transform converted EEG data into time-frequency spectrogram images.
- Spectrograms were analyzed using ML models with feature extraction and DL models (CNNs).
Main Results:
- The DL-based model achieved a high accuracy of 99.15% for ASD detection.
- The ML-based model achieved 95.25% accuracy.
- The DL approach demonstrated superior performance compared to ML and existing methods.
Conclusions:
- Deep learning models analyzing EEG spectrograms offer a highly accurate method for automated ASD diagnosis.
- This framework can aid in developing computer-aided diagnosis systems for ASD.
- The study highlights the potential of DL in discovering EEG biomarkers for ASD.
