سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / اولین همایش بین المللی تحول خدمات سلامت: تلفیق مراقبت، درمان، بازتوانی و امیدبخشی

Automated Diagnosis of Autism Spectrum Disorder Using Electroencephalography Signals and Deep Learning Architectures: A Comprehensive Review

نویسندگان :

Furuzan Nourani¹ Neda Ghobadi Samian²

1- دانشگاه ازاد اسلامی واحد اردبیل 2- دانشگاه ازاد اسلامی واحد اردبیل

کلمات کلیدی :

Autism Spectrum Disorder (ASD)،Electroencephalography (EEG)،Deep Learning،Convolutional Neural Networks (CNN)،EfficientNet،Polysomnography (PSG)

چکیده :

Background: Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition traditionally diagnosed through subjective behavioral evaluations [3]. Electroencephalography (EEG) provides a non-invasive, objective modality with high temporal resolution to capture aberrant neural dynamics and functional connectivity [3,4]. Method: This review synthesizes advanced signal processing pipelines-including Down-sampling, Butterworth filtering, Short-Time Fourier Transform (STFT), and Continuous Wavelet Transform (CWT)-alongside modern artificial intelligence models (1D-CNNs, Transfer Learning via EfficientNet and SqueezeNet, Hybrid CNN-LSTM architectures, and classical classifiers) applied to pediatric EEG datasets [1,2,5]. Finding: Deep learning frameworks demonstrate robust performance, with multi-input CNNs and spectrogram-based EfficientNet architectures achieving diagnostic accuracies exceeding 98% to 99.18% [2, 5]. Classical classifiers (SVM, RF, GNB) operating on CNN feature representations achieve up to 100% accuracy, while lightweight SqueezeNet implementations achieve 85.5% to 87.8% [1, 5]. Conclusion: Combining time-frequency spectrogram representations with deep learning networks offers high diagnostic validity. Integrating wearable, low-power acquisition systems and Explainable AI (XAI) represents the primary pathway toward scalable, point-of-care clinical deployment [4, 5].