Abstract
OBJECTIVE: This study aims to develop a deep learning model for the early and accurate detection of hyper-acute large vessel occlusion (LVO) stroke using EEG data. METHODS: A pMCAO mouse model was used to simulate LVO stroke, with high-resolution EEG data collected during the hyper-acute phase. EEGNet, a specialized deep learning architecture, was employed to develop a model based on EEG signals for the detection of hyper-acute LVO strokes. Seven-fold cross-validation was conducted to evaluate the model's performance across multiple metrics, including accuracy, AUC, precision, recall, and F1 score. RESULTS: The model achieved an overall accuracy of 97.9% and an AUC of 0.977, demonstrating excellent diagnostic performance across the hyper-acute phase. Stroke detection was reliable within 1.5 h post-onset, with classification accuracies exceeding 95% in all five time intervals segmented by hour. t-SNE analysis confirmed effective feature extraction, and comparisons with sham-operated mice validated the model's specificity for stroke-related EEG changes. CONCLUSION: The EEG-based deep learning model showed robust performance in hyper-acute LVO stroke detection, achieving high accuracy and specificity. These results highlight its potential as a biomarker for early stroke diagnosis and as a foundation for real-time, non-invasive monitoring in clinical and prehospital settings.