Related Experiment Videos
Seamless integration for enhanced seizure prediction using HybridConvMobileNet on Typhoon HIL
Prabhat Kumar Upadhyay1, Priyaranjan Kumar1, Manoj Kumar Panda2
1Department of Electrical and Electronics Engineering, Birla Institute of Technology, Ranchi, Jharkhand, 835215, India.
Scientific Reports
|November 27, 2025
Summary
This study introduces HybridConvMobileNet, a novel deep learning model for efficient and accurate real-time seizure prediction using electroencephalogram (EEG) data. The model achieves high performance on benchmark datasets, demonstrating its clinical potential for seizure monitoring.
Area of Science:
- Medical Technology
- Artificial Intelligence
- Signal Processing
Background:
- Real-time seizure prediction is crucial for timely medical interventions and improved patient outcomes.
- Existing methods for seizure prediction face challenges in efficiency and accuracy for real-time applications.
Purpose of the Study:
- To introduce HybridConvMobileNet, a novel hybrid deep learning model for efficient and accurate real-time seizure prediction.
- To leverage 1D convolutional neural networks (CNN) and MobileNet for enhanced feature extraction and computational speed.
Main Methods:
- The HybridConvMobileNet model integrates 1D CNN for spatial feature extraction from frequency-domain electroencephalogram (EEG) data.
- Input features are derived from 1D Short-Time Fourier Transform (STFT) coefficients of pre-processed EEG signals.
- The MobileNet component enhances computational efficiency and speed for real-time processing.
Main Results:
- The model achieved high performance on the CHB-MIT dataset (99.70% accuracy, 99.31% sensitivity, 99.43% F1-score) and Siena dataset (99.67% accuracy, 99.08% sensitivity, 99.57% F1-score).
- HybridConvMobileNet outperformed eight existing seizure prediction methods across both benchmark datasets.
- Real-time implementation showed a low mean detection latency of 0.1 to 1 second.
Conclusions:
- HybridConvMobileNet demonstrates significant potential for effective and efficient real-time seizure prediction.
- The model's performance and low latency underscore its suitability for clinical seizure monitoring and control applications.
