Borbath, Alice
[UCL]
Bol, David
[UCL]
Epilepsy is a neurological disease affecting more than 60 million people in the world. It is characterized by the sudden onset of seizures, which can be dangerous to the patient, and even cause death. Until now, seizures appear in an uncontrollable and unpredictable manner, rendering care and prevention difficult to implement. Recently, intensive research has been led in order to identify, based on different indicators, the onset of a seizure. One of the most reliable indicators seems to be electroencephalogram (EEG) signals measured directly inside the cranial cavity. This master thesis is focused on one specific algorithm under investigation, that allows interpretable detection of a seizure onset by comparing features generated by the patient’s EEG signals, with features generated by a neural mass model synthesizing EEG signals from three physiologically meaningful parameters. In this work, the move towards ultra-low-power implementation of this algorithm was started by transferring the code, previously implemented on Matlab, to C language, more adapted to a microcontroller such as ARM Cortex M4. Then, various ways to optimize memory and power usage of the algorithm were identified and investigated. It is shown that the resolution of the meaningful parameters fed to the neural mass model is of critical importance in these aspects. While the execution is still time-consuming, the algorithm has been brought a step closer to an implementation in a point-of-care device.
Bibliographic reference |
Borbath, Alice. Ultra-low-power implementation of a parameters identification algorithm for interpretable detection of epileptic seizures. Ecole polytechnique de Louvain, Université catholique de Louvain, 2020. Prom. : Bol, David. |
Permanent URL |
http://hdl.handle.net/2078.1/thesis:26660 |