<b>A Deep-Learning-Powered Pipeline for EEG-Based Dyslexia Detection Achieving ~94.66% Accuracy</b>

Dyslexia is a language-based learning disorder that significantly impacts reading, spelling, and writing abilities. Early identification of dyslexia is critical for targeted intervention and long-term academic success. Recent research has explored electroencephalography (EEG) signals to capture the neurophysiological correlates of dyslexia, often revealing distinctive patterns in brain oscillations and event-related potentials. In this study, we present a deep-learning-based approach for EEG-based dyslexia detection that achieves <b>94.66% classification accuracy</b> on a real-world dataset. Our methodology encompasses preprocessing, feature selection via mutual information, hyperparameter tuning using KerasTuner, and final network training with advanced regularization and callbacks. Notably, we benchmark our pipeline against classical dyslexia detection workflows, highlighting how our approach leverages modern techniques (feature normalization, batch normalization, dropout, automated hyperparameter optimization) to reach high accuracy while remaining computationally tractable. We also discuss potential translational applications in educational and clinical settings, challenges of EEG data variability, and the need for large-scale standardized EEG repositories for robust, replicable dyslexia detection research.

阅读障碍（Dyslexia）是一种基于语言的学习障碍，会对阅读、拼写及书写能力造成显著损害。早期识别阅读障碍对于开展针对性干预以及实现长期学业成功至关重要。近期已有研究借助脑电图（electroencephalography，EEG）信号捕捉阅读障碍的神经生理关联特征，这类研究通常能够揭示脑振荡与事件相关电位（event-related potentials）中的特异性模式。本研究提出了一种基于深度学习的EEG阅读障碍检测方法，在真实世界数据集上实现了**94.66%的分类准确率**。我们的方法涵盖了预处理、基于互信息的特征选择、借助KerasTuner实现的超参数调优，以及采用高级正则化与回调函数的最终网络训练流程。值得注意的是，我们将该流程与经典阅读障碍检测工作流进行了基准对比，阐明了本方法如何借助现代技术（特征归一化、批量归一化、丢弃法（Dropout）、自动化超参数优化）在保持计算可操作性的同时实现高精度。本研究还探讨了该方法在教育与临床场景中的潜在转化应用、EEG数据变异性带来的挑战，以及开展可靠可复现的阅读障碍检测研究所需的大规模标准化EEG数据集库建设需求。