DuoQ-EpiNet: a dual-track quantum–classical convolutional neural network for EEG-based epilepsy seizure detection

Epileptic seizure detection remains a critical challenge in clinical neurodiagnostics, particularly in low-data settings where EEG recordings are scarce. To address this, we propose DuoQ-EpiNet a dual-track hybrid framework that integrates quantum and classical deep learning models for robust seizure classification using the University of Bonn EEG dataset. In the first track, handcrafted statistical and spectral descriptors are extracted from the raw EEG signals and subsequently analyzed using a 1D Convolutional Neural Network (CNN) to learn discriminative temporal representations. In parallel, the second track with wavelet approach transforms the EEG signals into scalogram images, which are processed through a Hybrid Quanvolutional Classical Convolutional Neural Network (HQCNN) equipped with a Fixed Quantum Filter Circuit to generate expressive quantum feature maps followed by classical CNN. The latent representations obtained from both tracks are then fused and passed through fully connected layers to perform the final binary classification. Systematic comparison of the proposed DuoQ-EpiNet model by tweaking quantum hyperparameter based variants, state-of-the-art HQCNN architectures, as well as the best classical transfer learning models have demonstrated that the proposed model performs better than all evaluated variants. Among all evaluated configurations, the proposed DuoQ-EpiNet Binary Dual-Track (P-B-D) model achieved outstanding performance of 98.50% accuracy with its FQFC employed in Track 2 contrived with quantum hyperparameters of $n_{shots} = 1000$ and $n_{layers} = 1$ . Performance in data-scale studies ranging from 5% to 100% shows that DuoQ-EpiNet outperforms traditional baselines. Its generalization ability is confirmed by evaluation on the CHB-MIT scalp EEG dataset. The model maintains its stability at low noise densities with only slight performance deterioration, according to NISQ robustness study employing density matrix simulations with depolarizing, amplitude damping, phase damping, and readout noise.