An IoT-Cloud-Based Detection Approach of Generalized Seizure across Ages Using Harmonic-Guided Neural Networks with Edge Optimization

Automated seizure detection faces critical challenges in generalizing across age-specific EEG patterns, particularly for neonates and the elderly, where seizures are frequently missed. This study proposes a novel harmonic-guided neural framework optimized for IoT-cloud environments, enabling accurate and low-latency seizure monitoring for all age groups. The proposed architecture combines three key innovations: First, wavelet-based preprocessing harmonizes EEG signals across age groups, accounting for developmental variations like neonatal delta brushes and elderly focal slowing. Second, adaptive neural networks (CNNs + Transformers), trained on standardized time-frequency representations, detect seizure patterns with high accuracy. Finally, a distributed edge-cloud system ensures efficient processing—lightweight wavelet analysis runs locally on Raspberry Pi devices, while complex model inferences are handled remotely in the cloud. This study validates the TUH EEG Corpus (covering neonates to the elderly) and NICU datasets, comparing them against traditional SVM and raw-EEG CNN baselines. The system achieves: 93.2% sensitivity and 91.7% specificity across ages (vs. 70–85% for non-harmonic methods in neonates/elderly), <150ms latency on edge devices (60% faster than cloud-only processing), 40% lower energy use via harmonic-guided feature pruning. Our work bridges the age-generalization gap in seizure detection by unifying harmonic signal processing with edge-cloud optimized neural networks. The framework’s low-cost deployment potential (∼$50 edge hardware) makes it viable for NICUs, aged-care facilities, and resource-limited settings. This study introduces the first harmonic-guided neural framework for cross-age seizure detection with IoT-ready scalability.