Channel Stacking: A Rapid Classification Method for Parkinson's Disease Based on EEG Data.

This study introduces a technique known as "channel stacking," aimed at comprehensively representing electroencephalogram signals to identify Parkinson's disease accurately. This method combines information from multiple channels to prepare input signals for the model, enabling the deep learning architecture to capture information across all channels efficiently. For classification, a ResNet18 network is employed. Notably, the model achieves an accuracy of 96.43% with just 90 seconds of data per subject. This performance is validated through Leave-One-Subject-Out Cross-Validation, effectively simulating real-world scenarios. The experimental results highlight the robust detection performance of the proposed method on clinical datasets.