Sign up for our newsletter!    Become a member!
Home Events Officers Links Jobs Search Contact

Invited Talk

Signal Processing for Diagnosis and Treatment of Brain Disorders

Dr. Justin H. G. Dauwels
(homepage)

Stochastic Systems Group
Laboratory for Information and Decision Systems
Massachusetts Institute of Technology
Building 32D-566, 77 Massachusetts Avenue, Cambridge, MA 02139-4307, USA

Abstract

In the last three years, we have explored how signal processing may be used to improve the diagnosis and treatment of brain disorders. In this talk, I will report on progress made. I will focus on two projects: (1) early diagnosis of Alzheimer's disease (AD) from scalp EEG; (2) localization of epileptic brain tissue from interictal intracranial EEG. In the first project, we exploit loss in EEG synchrony as an indicator for early-stage AD. Early diagnosis of Alzheimer's disease helps ensure prescription of symptoms-delaying medications when they are most useful. Medical diagnosis of Alzheimer's disease is hard, and symptoms are often dismissed as normal consequences of aging. Diagnosis is usually performed through a combination of extensive testing and eliminations of other possible causes. Psychological tests, neurological examination, and increasingly, imaging techniques are used to help diagnose the disease. In particular, EEG is highly suitable in this context. An EEG recording system is a relatively simple and low-cost technology, at present available in most hospitals.

It has repeatedly been reported in the medical literature that the EEG signals of AD patients are less synchronous than in age-matched control patients. We investigate whether this phenomenon allows us to reliably predict AD at an early stage. To this end, we develop a novel family of measures to quantify the similarity of point processes, referred to as "stochastic event synchrony" (SES). SES tries to align events in the different point processes; the better the alignment, the more similar the point processes are considered to be. We apply SES together with classical synchrony measures to two different EEG data sets of early-AD patients and age-matched control subjects. We found that most measures indicate a loss in EEG synchrony. Both Granger causality and SES yield highly significant differences, consistently on both data sets. Combining both synchrony measures yields a strong indicator for early AD. In future work, we plan to verify those results on other data sets, and to combine EEG with other modalities.

In the second project, we exploit increased EEG synchrony as an indicator for epileptic brain tissue; in contrast to the first project, here we analyze intracranial EEG. Epilepsy is a devastating chronic disease affecting almost 3 million people in the US alone and 50 million people worldwide. For as many as 30% of patients seizures are poorly controlled with medications alone. For some of these patients, surgery may be the best option but success requires accurate delineation of the brain region responsible for seizure onset. Currently, the key to making this determination is the seizure itself, and the recordings must continue, usually for days, until enough seizures are obtained to determine the onset region. In some cases these recordings must be done using invasive electrodes, a procedure that includes substantial risk, discomfort and cost. 

Here we develop techniques that use periods of intracranial EEG that are between seizure to localize epileptogenic networks; we detect brain areas of hypersynchronous activity using signal processing methods. Our analysis of intracortical EEG of 11 epileptic patients shows that certain EEG channels and hence cortical regions are consistently more synchronous compared to others. Interestingly, hypersynchrony seems to strongly correlate with the seizure onset zone. By combining hypersynchrony with other electrophysiological signatures, we are able to reliably delineate the seizure onset area using interictal EEG. In the long term, this may enable shorter hospitalizations or even avoidance of semi-chronic implantations all-together.

This research is conducted in collaboration with various research partners, from MIT, MGH, RIKEN Brain Science Institute, Max Planck Institute Frankfurt, KAIST, and several other institutions.

Biography

Justin Dauwels received the diploma (M.Sc. degree) in engineering physics (magna cum laude) from the University of Gent, Belgium, in 2000, and the PhD. degree in electrical engineering from ETH Zurich, Switzerland, in 2005. He was a research scientist at the Amari Research Unit of the RIKEN Brain Science Institute, Japan, from Jan 2006 to December 2007. Since Jan 2008 he is a researcher at the Laboratory of Information and Decision Systems at M.I.T. He is also affiliated with the Neurology Department at the Massachusetts General Hospital. In the Fall of 2003 he was a visiting scientist at the M.I.T. Media lab, in  the Spring of 2004 he was an intern at Mitsubishi Electric Research Laboratories (MERL), Cambridge, MA.

His broader research interests include signal processing, machine learning, digital communications, applied information theory, and computational neuroscience. The aims of his research are to improve the diagnosis and treatment of neurological diseases such as Alzheimer's disease and epilepsy. To this end, he analyzes brain signals, using tools such as graphical models, combinatorial optimization and mathematical statistics. He recently also started to work on feedback control systems for brain stimulation (electrical and optogenetical), with the aim of preventing epileptic seizures.

He is a recipient of the ICASSP2006 Student Paper Award, a NIPS2007 and EMBC2008 travel award, a JSPS Fellowship in 2006, and a Fellowship of the BAEF in 2007. He is one of the two recipients of the 2007 Henri-Benedictus Fellowship of the King Baudouin Foundation (award ceremony honored by Princess Astrid, Princess of Belgium). He was technical committee member of a variety of conferences in the area of digital communications, machine learning, and computational neuroscience. He is associate editor of Computational Intelligence and Neuroscience, Hindawi Publishing Corporation.

Volker Koch, 09/2009