EEG-based Epileptic Prediction via a Two-stage Channel-aware Set Transformer Network

2025-07-21

AI Summary

To address the practical limitations of wearable epilepsy prediction devices—namely, excessive EEG channel count and large physical footprint—this paper proposes a two-stage channel-aware Set Transformer network, integrated with seizure-agnostic data partitioning and a channel importance evaluation mechanism, enabling highly reliable prediction with minimal channels. Methodologically, the approach employs a two-stage attention mechanism to model dynamic inter-channel dependencies while decoupling temporal correlations between training and test data to enhance subject-level generalization. Experiments on the CHB-MIT dataset comprising 22 patients achieve an average sensitivity of 80.1%, a low false alarm rate of 0.11 per hour, and a dramatic reduction in average channel usage from 18 to 2.8—substantially improving model lightweightness and clinical deployability. The core contribution lies in the first synergistic integration of channel-aware structural learning and seizure-independent segmentation for low-channel epilepsy prediction, effectively balancing predictive performance, robustness, and real-world applicability.

Technology Category

• Machine Learning: Hardware-aware ML
• Computer Vision: Segmentation
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Security and Privacy: Large-scale security measurements
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Epilepsy is a chronic, noncommunicable brain disorder, and sudden seizure onsets can significantly impact patients' quality of life and health. However, wearable seizure-predicting devices are still limited, partly due to the bulky size of EEG-collecting devices. To relieve the problem, we proposed a novel two-stage channel-aware Set Transformer Network that could perform seizure prediction with fewer EEG channel sensors. We also tested a seizure-independent division method which could prevent the adjacency of training and test data. Experiments were performed on the CHB-MIT dataset which includes 22 patients with 88 merged seizures. The mean sensitivity before channel selection was 76.4% with a false predicting rate (FPR) of 0.09/hour. After channel selection, dominant channels emerged in 20 out of 22 patients; the average number of channels was reduced to 2.8 from 18; and the mean sensitivity rose to 80.1% with an FPR of 0.11/hour. Furthermore, experimental results on the seizure-independent division supported our assertion that a more rigorous seizure-independent division should be used for patients with abundant EEG recordings.

Problem

• Predict epileptic seizures using fewer EEG channels
• Improve seizure prediction accuracy with channel selection
• Ensure rigorous seizure-independent data division for reliability

Innovation

• Two-stage channel-aware Set Transformer Network
• Seizure-independent division method
• Reduced EEG channels to 2.8 average