Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
An automatic seizure detection method from high-resolution intracranial-EEG (iEEG) signals is presented to minimize the computational complexity and realize real-time accurate seizure detection for biomedical implants. Complex signal processing on a large amount of iEEG signals captured via several electrodes is a crucial impediment in seizure detection when it comes to power consumption and real-time processing. Therefore, a subject-customized channel selection method correlated to a feature ranking unit is proposed to improve the computation efficiency and seizure detection accuracy by reducing the dimension of extracted features as well as the electrode channels. Nine popular time-domain features are extracted and ranked to constitute a customized feature subset. Subsequently, electrode channels are ranked with respect to the top four rank features obtained from the feature ranking unit. Then, the number of channels is optimized to reach the highest detection accuracy. The selected channels are compressed into a single channel to minimize the signal processing computation load. The suggested method is tested on seven patients with 37 seizure events from the SWEC-ETHZ dataset of the Bern University Hospital. The perfect sensitivity of 100%, the specificity of 92.98%, and the mean detection delay of 3.6 sec are achieved which outperform the state-of-the-art. In addition, the computation complexity is remarkably reduced which makes the presented method suitable for low-power real-time biomedical implants.
David Atienza Alonso, Amir Aminifar, Alireza Amirshahi, José Angel Miranda Calero, Jonathan Dan