Electroencephalography-Based Machine Learning for Biomarker Detection in Dyslexia and Autism Spectrum Disorder: A...

Area of Science:

• Neuroscience
• Computational Biology
• Medical Informatics

Background:

• Autism Spectrum Disorder (ASD) and developmental dyslexia are complex neurodevelopmental conditions.
• Identifying reliable neurobiological indicators is crucial for early diagnosis and intervention.
• Electroencephalography (EEG) offers a non-invasive method for brain activity assessment.

Purpose of the Study:

• To review recent machine learning (ML) and deep learning (DL) advancements in analyzing EEG for ASD and dyslexia.
• To explore common methodological pipelines in EEG-based neurodevelopmental disorder research.
• To assess the diagnostic potential and limitations of ML/DL models in this domain.

Main Methods:

• Systematic review of 15 peer-reviewed studies (2013-2025) focusing on EEG analysis for ASD and dyslexia.
• Analysis of preprocessing techniques (e.g., MATLAB, MNE-Python) and feature extraction (e.g., spectral power, connectivity, entropy).
• Evaluation of ML algorithms, including Support Vector Machines, Random Forests, Deep Neural Networks, and Transformers.

Main Results:

• High classification accuracies (82%-99.2%) were reported for both dyslexic and ASD groups.
• Commonly used EEG systems (10-20) and preprocessing pipelines were identified.
• Advanced DL models demonstrated significant potential in distinguishing between neurodevelopmental conditions.

Conclusions:

• ML and DL applied to EEG show significant promise for identifying neurobiological markers of ASD and dyslexia.
• Current models offer potential for therapeutic applications but face challenges in interpretability and generalizability, particularly for the heterogeneous ASD population.
• Further research is needed to enhance model transparency and applicability across diverse clinical scenarios.