This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
In this paper, we propose a novel two-stage channel-aware Set Transformer Network to improve the seizure prediction performance while reducing the number of EEG channels to solve the size issue of wearable epilepsy seizure prediction devices. Experimental results using the CHB-MIT dataset show that before channel selection, the average sensitivity is 76.4% and the false positive rate (FPR) is 0.09/hour. After channel selection, the average sensitivity increases to 80.1% and the FPR slightly increases to 0.11/hour, even though the average number of channels decreases from 18 to 2.8. In addition, we confirm the necessity of more rigorous data segmentation through a seizure-unrelated data segmentation method.
Takeaways, Limitations
•
Takeaways:
◦
We present a novel network architecture that demonstrates improved seizure prediction performance despite a reduced number of EEG channels.
◦
Presenting the possibility of miniaturizing a wearable epilepsy seizure prediction device.
◦
Suggesting the importance of more rigorous data segmentation methods for seizure prediction.
•
Limitations:
◦
Experiments were conducted using only one CHB-MIT dataset, and further research on generalizability is needed.
◦
FPR slightly increased after channel selection, further research is needed to reduce FPR.
◦
Generalizability to patients with different types of epilepsy needs to be verified.
