Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
A Practical Application of Neuromorphic Computing: Detecting Human Epileptic Seizures From Short-Term Intracranial Electroencephalograms (iEEG)
This project aims to build brain-inspired computing hardware to implement always-on analysis of iEEG recordings to optimize diagnostics and therapies for epilepsy
patients, by developing Spiking Neural Network models and validating them using neuromorphic prototype chips
Epilepsy affects ~1% of the general population in western countries and is thus one of the most prevalent chronic neurologic disorders. The hallmark of epilepsy are seizures, caused by transiently dys-coordinated neuro-glial activity, which gives rise to typically faster and larger-amplitude electric brain signals. These signals are best recorded by intracranial electrodes (iEEG). Recent studies have impressively demonstrated that ultra-long-term iEEG recordings may help to optimize diagnostics and therapies of epilepsy patients (Baud et al., 2018) and have kindled strong interest in developing novel and
implantable devices for monitoring human electric brain activity. To be implantable these devices have to be small and low-power, which mandates extremely energy-efficient methods for signal processing.
Epilepsy affects ~1% of the general population in western countries and is thus one of the most prevalent chronic neurologic disorders. The hallmark of epilepsy are seizures, caused by transiently dys-coordinated neuro-glial activity, which gives rise to typically faster and larger-amplitude electric brain signals. These signals are best recorded by intracranial electrodes (iEEG). Recent studies have impressively demonstrated that ultra-long-term iEEG recordings may help to optimize diagnostics and therapies of epilepsy patients (Baud et al., 2018) and have kindled strong interest in developing novel and implantable devices for monitoring human electric brain activity. To be implantable these devices have to be small and low-power, which mandates extremely energy-efficient methods for signal processing.
One of the main characteristics of neuromorphic processors systems is their very low
power consumption. Thus the goal of this project is to test, whether Spiking Neural Networks
implemented on neuromorphic processing systems may be used to reliably detect epileptic
seizures from intracranial short-term iEEG recordings as provided by one of the supervisors
in an open-access anonymized data set (http://ieeg-swez.ethz.ch/).
The student will investigate the use of recurrent Spiking Neural Networks (rSNNs) to analyze
the iEEG data-set and train them to detect and possibly predict the onset of an epileptic
seizure. The challenge will be in finding network configurations that are robust to variability in
both their internal parameters and the input signals, so that their hardware implementation on
mixed-signal analog/digital neuromorphic circuits will work robustly.
We will use both software simulations to investigate the rSNNs configurations and hardware neuromorphic processors currently available at INI to validate them.
The project fits within the research carried out at INI to use neuromorphic chips to process
bio-signals. The student will, therefore, benefit from the infrastructure already available at INI,
and from the interactions with the INI members working on related projects.
One of the main characteristics of neuromorphic processors systems is their very low power consumption. Thus the goal of this project is to test, whether Spiking Neural Networks implemented on neuromorphic processing systems may be used to reliably detect epileptic seizures from intracranial short-term iEEG recordings as provided by one of the supervisors in an open-access anonymized data set (http://ieeg-swez.ethz.ch/). The student will investigate the use of recurrent Spiking Neural Networks (rSNNs) to analyze the iEEG data-set and train them to detect and possibly predict the onset of an epileptic seizure. The challenge will be in finding network configurations that are robust to variability in both their internal parameters and the input signals, so that their hardware implementation on mixed-signal analog/digital neuromorphic circuits will work robustly. We will use both software simulations to investigate the rSNNs configurations and hardware neuromorphic processors currently available at INI to validate them. The project fits within the research carried out at INI to use neuromorphic chips to process bio-signals. The student will, therefore, benefit from the infrastructure already available at INI, and from the interactions with the INI members working on related projects.
• Prof. Giacomo Indiveri, Institute of Neuroinformatics, University of Zurich and ETH
Zurich, Winterthurerstrasse 190, 8057 Zurich, Switzerland; giacomo@ini.uzh.ch
• Prof. Kaspar Schindler, University Department of Neurology, Inselspital, 3010 Bern;
kaspar.schindler@insel.ch
• Prof. Giacomo Indiveri, Institute of Neuroinformatics, University of Zurich and ETH Zurich, Winterthurerstrasse 190, 8057 Zurich, Switzerland; giacomo@ini.uzh.ch • Prof. Kaspar Schindler, University Department of Neurology, Inselspital, 3010 Bern; kaspar.schindler@insel.ch