Affiliations:
1School of Information and Communications Engineering, Xi’an Jiaotong University, Xi’an, China
2Department of Computer Engineering, National University of Technology, Islamabad, Pakistan
Epileptic seizure detection from electroencephalogram (EEG) signals is important for early clinical intervention in patients with neurological disorders. This study presents a hardware–software co-design implementation of a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) model for automatic EEG classification on the Xilinx Zynq-7000 System-on-Chip (SoC) platform. In the proposed architecture, the CNN layers, which extract spatial features, are executed on the ARM Cortex-A9 processors, while the LSTM and fully connected layers are implemented in the FPGA fabric to enable real-time inference. The model was trained and tested on the Bonn University EEG dataset for three-class classification: normal, interictal, and seizure (ictal) states. The proposed system achieves a classification accuracy of 99.33% with an inference latency of 0.657 seconds per EEG segment. Hardware-oriented optimizations, including fixed-point quantization and efficient approximations of activation functions, reduce power consumption and increase processing speed. Compared with a full software implementation, the proposed co-design provides a 12.5× reduction in execution time. These results demonstrate that deep learning–based seizure detection can be effectively deployed on resource-limited embedded platforms for real-time medical applications.
EEG, FPGA, Epileptic seizure detection, CNN-LSTM, Zynq SoC
https://doi.org/10.21833/ijaas.2026.03.008
Akhtar, N., Buzdar, A. R., & Fan, J. (2026). CNN–LSTM-based EEG epileptic seizure detection: A hardware–software co-design approach. International Journal of Advanced and Applied Sciences, 13(3), 74–85. https://doi.org/10.21833/ijaas.2026.03.008