Epilepsy affects about 50 million people worldwide and can seriously affect patients’ quality of life. Electroencephalography (EEG), which records the brain’s electrical activity in real time, plays an important role in epilepsy diagnosis and monitoring. However, EEG signals are nonstationary, noisy and highly variable between individuals, making manual interpretation time-consuming and vulnerable to subjective bias. Deep learning has become increasingly important for seizure prediction, but its performance often depends strongly on network architecture, while designing an effective architecture by hand can require substantial time and computing resources. Their framework, called CMS-NAS, combines convolutional neural networks, Mamba and spiking neural networks within a search space specifically designed for seizure prediction.
Rather than simply connecting the three types of networks, the researchers designed two hybrid building blocks, called ConvSpike and MambaSpike. The system first converts multichannel EEG signals into time-frequency representations using a short-time Fourier transform. It then generates a population of candidate network architectures. Each candidate is trained and evaluated, after which better-performing architectures are selected for crossover and mutation to produce new generations. The search simultaneously considers prediction performance and model size, allowing the algorithm to seek architectures that balance accuracy and efficiency. The resulting search space contains about 1.74 × 10¹⁷ possible architectures.
To evaluate seizure prediction, the researchers considered several measures. AUC reflects the model’s overall ability to distinguish different EEG states, with higher values generally indicating better discrimination. Sensitivity measures the proportion of actual seizures that are successfully predicted, while the study’s false-positive rate (FPR) measures the number of false alarm events per hour of interictal EEG. Thus, higher sensitivity and lower FPR are desirable. The researchers also calculated P values against a random prediction baseline, with P < 0.05 considered statistically significant. To reduce isolated false alarms, they used a k-of-n strategy, in which at least 20 of 30 consecutive predictions had to be positive to count as a positive event. The prediction horizon was set to 1 minute.
The researchers tested CMS-NAS on two public datasets, CHB-MIT and AES, and a private clinical dataset. They used both patient-specific and patient-independent evaluation. In the patient-specific setting, data from other seizure events of the same patient were used for training and validation, while one seizure event was held out for testing. In the patient-independent setting, all data from one patient were held out for testing, while the remaining patients were used for training, providing a more demanding test of generalization to previously unseen patients.
On CHB-MIT, CMS-NAS achieved an AUC of 0.930, sensitivity of 88.5% and FPR of 0.108 false alarms per hour in patient-specific testing. Seventeen of 18 subjects met the study’s statistical significance criterion. In patient-independent testing, the model achieved an AUC of 0.672 and sensitivity of 60.7%, remaining ahead of the compared methods on these two core measures.
On the AES dataset, CMS-NAS achieved an AUC of 0.878, sensitivity of 87.3% and FPR of 0.107 in patient-specific testing, while its patient-independent sensitivity reached 61.3%, the highest among the compared methods. On the private clinical dataset, the model achieved an AUC of 0.923, sensitivity of 84.4% and FPR of 0.138 in patient-specific testing. In patient-independent testing, it achieved an AUC of 0.662, sensitivity of 51.6% and FPR of 0.253.
The researchers also evaluated deployment efficiency. The architecture search took 11 hours, while the final model had only 0.034 million parameters, a model size of 0.131 MB, 0.293 million MACs, and an inference latency of 1.308 milliseconds.
Ablation experiments showed why the hybrid design works. CNN, Mamba and SNN were not simply stacked together but played complementary roles: CNN extracted local time-frequency features, Mamba modeled long-range temporal dependencies, and SNN used sparse spike-based processing to enhance transient abnormal signals. Removing any core component reduced overall prediction performance.
The researchers conclude that CMS-NAS can reduce some of the structural redundancy, search cost and computational burden associated with manually designed or conventional NAS-based seizure prediction models, while balancing prediction performance and lightweight deployment. However, the current framework is designed for fixed tasks and uses only single-modal EEG data. Future work will explore multimodal seizure prediction and architectures that can adapt to a wider range of tasks, further testing the framework’s generality and practical value.
Authors of the paper include Bo Yu, Chang Li, Rencheng Song, Xiang Liu, Ruobing Qian, Xun Chen.
This work is supported by the National Key Research and Development Program of China under Grant 2024YFE0201600; the Fundamental and Interdisciplinary Disciplines Breakthrough Plan of the Ministry of Education of China under Grant JYB2025XDXM109; the National Natural Science Foundation of China under Grants 32541010, 41901350, and 32271431; the Fundamental Research Funds for the Central Universities of China under Grant JZ2025HGTG0280; and the Anhui Provincial Natural Science Foundation under Grants 2408085MF156, 202523o09050017, and 2508085ZD005.
The paper, “A Hybrid Convolutional, Mamba and Spiking Neural Network Architecture Search for Seizure Prediction”, was published in the journal Cyborg and Bionic Systems on Sep 30, 2026, at https://doi.org/10.34133/cbsystems.0689.
Cyborg and Bionic Systems