Neuroengineering

Epilepsy

Approximately 1% of the world's population (50 million people) suffers from epilepsy. Epilepsy is a neurological disorder where seizures arise in populations of neurons (brain cells) that become overly excited, through both electrical and chemical mechanisms. These seizures can have various behavioural manifestations, the most famous being the chaotic and uncontrollable movement of the body. Given that epileptic seizures can involve large portions of the brain, it is a global brain disorder, and can have a variety of affects on consciousness. While there are many drugs that can be used to prevent epileptic seizures, 25% of epileptics cannot be treated sufficiently by available therapies. Moreover, the exact cause of epileptic seizures in the brain is not well understood.

Our group is involved in several projects, aimed at understanding the underlying causes of epilepsy, and developing methods to help improve the quality of life for epileptics. This includes, trying to develop methods for the prevention or abortion of epileptic seizures. Such methods require the detection/prediction of epileptic seizures and the prevention of these seizures using such techniques as direct electrical stimulation of the epileptic brain region, or local drug delivery to the site of the seizure in the brain. These methods could be embodied in a device implanted in the brain. We are also trying to understand the underlying causes of epilepsy through both physiological experiments and neural modeling.

Research Projects

Epileptic Seizure Detection

If detection of the onset of seizures were possible, devices could be implanted within the brain in order to prevent a seizure from occurring by using electrical stimulation (drug delivery probably acts too slowly to abort a seizure). Current seizure detection algorithms analyse the electrical signals recorded from the brain. These algorithms draw on techniques from signal processing, non-linear dynamical systems theory, and machine learning. Moreover, they try to provide a seizure onset signal within milliseconds after the actual seizure onset.

Our group is developing seizure detection algorithms primarily based on signal processing and non-linear dynamical systems theory, since the methods involved are usually simpler and easier to implement in an implantable device. However, we are also interested in looking at machine learning based methods in order to determine the overall performance of all types of seizure detection algorithms. In the long term, our group aims to use seizure detection algorithms as part of an implanted device used to prevent seizures.

Epileptic Seizure Prediction

While seizure detection is necessary to activate an electrical stimulation device at the right time to abort a seizure, seizure prediction is also very important, and has a variety of benefits. First, it can be used to generate an electrical stimulus just before seizure onset in order to prevent a seizure from occurring. This is more useful than seizure detection, which relies on detecting the onset of a seizure, and thus there is a delay between seizure onset and abortion of a seizure through electrical stimulation. Second, anti-convulsant drugs can be slow acting (on the time scale of minutes), thus if a seizure prediction method could predict an event minutes in advance, it could activate a drug delivery system to diffuse a drug into the seizure region of the brain and prevent a seizure from occurring. Third, if a seizure can be predicted minutes in advance the patient could be warned and they can move themselves to safety. Fourth, patients often undergo brain scans to determine the focus of epileptic seizures for the purposes of surgical removal of the epileptic region, these scans require injection of a radioactively labeled marker used to indicate the focus. If a doctor could be warned about a seizure a few minutes in advance they could inject the marker at the right time to determine the seizure focus more accurately.

Current seizure prediction algorithms analyse the electrical signals recorded from the brain using methods similar to seizure detection. These algorithms draw on techniques from signal processing, non-linear dynamical systems theory, and machine learning. Instead of providing a seizure onset signal within milliseconds after the actual seizure onset, as is done by detection algorithms, prediction algorithms try to predict seizures on time scales of seconds, minutes, and even hours before seizures, depending on the algorithm.

Our group is developing seizure prediction algorithms. We have a primary interest in methods based on signal processing and non-linear dynamical systems theory, since they are usually simpler and easier to implement in an implantable device. However, we are also interested in looking at machine learning based methods in order to determine the performance of different seizure prediction algorithms. In the long term, our group aims to use seizure prediction algorithms as part of an implanted device used to prevent seizures.

Neural Modeling of Epilepsy

Despite years of research into epilepsy, little is actually known about the underlying causes of epilepsy. Largely, because it can occur in several parts of the brain, and involves large populations of neurons (brain cells) that interact with one another in a complex way. In an attempt to understand the underlying causes of epilepsy, the EEE Department, the Bionic Ear Institute , and the Howard Florey Institute are forming a collaboration to try to develop neural models of epilepsy at various scales of brain structure.

At a low level, neurons connect to other neurons via electrochemical connections called synapses. These synapses involve ion channels that are activated by neurotransmitters (i.e. a chemical messenger), and transfer electrical charge from one cell to another. Thus providing the basis of electrochemical signaling between cells. Certain types of ion channels are genetically related to epileptics. At a larger scale, networks of neurons that become overly excited are thought to be directly associated with epileptic seizures. Depending on the interests of the student we would like to begin developing neural models of epilepsy at different scales: (1) ion channels/synapses, (2) single neurons, or (3) networks of neurons. At each of these levels we are interested in modeling the changes in dynamics associated with changes in ion channel structure and properties. These changes in ion channels are related to genetic mutations that are associated with certain types of epilepsy.

Quantitative Analysis of Seizure Dynamics

Performing invasive experiments in humans with epilepsy presents us with various ethical dilemmas. Thus scientists have developed animal models of epilepsy as an alternative. In these models, the genetic, electrochemical, and/or structural make-up of the normal animal brain is altered such that 'epileptic-like' seizures can be generated in the brain. Thus animal models provide a means of trying to discover the underlying causes of epilepsy. The EEE Department, St. Vincent's Hospital of Melbourne and the Bionic Ear Institute have formed a collaboration to study animal models of epilepsy, as well as develop electrical stimulation or anti-convulsant drug delivery mechanisms that could be used to prevent seizures from occurring.

During physiological experiments in animal models, the responses of brain cells (neurons) will be recorded while seizures are generated in the brain. It is believed that seizures result from populations of neurons which synchronize their activities and become abnormally overactive.

Our group is analysing the responses of populations of neurons recorded during seizures in different parts of an animal brain. This will involve the application of statistical, signal processing, and non-linear dynamical systems theory methods to try and look for relationships, or interactions, between neural responses within specific regions of the brain. The main approach will be to look for synchrony between neural responses using such methods as cross-correlation, phase synchrony or non-linear interdependence. By performing a quantitative analysis of the neural responses we hope to understand the way in which neurons are recruited into a seizure, as well as the sequence of events involved in seizures.

This work is supported in part by the Australian Research Council

Publications on Epilepsy

Student projects in Epilepsy

A summary of this work can be found here:
Seizure Prediction

If you are interested in learning more about the epilepsy group, please contact:
Levin Kuhlmann
email: l.kuhlmann@ee.unimelb.edu.au
+613 8344 6689