Cognition and Modelling

The Cognition and modelling group focuses on the characterisation of cognitive decline in several neurological disorders with a specific emphasis on multiple sclerosis. The functioning of the degenerative brain is assessed by different neuroimaging modalities (EEG, MEG, fMRI) and these data are subsequently used as input features for artificial intelligence techniques (support vector machines, artificial neural networks) in order to develop biomarkers for neurological disorders.

This figure illustrates how graphs or networks are extracted from raw data. The first panel shows MEG data, but the process works just as well starting with EEG or fMRI data. We then use information from a brain atlas (C) to perform parcellation, dividing the brain into different regions of interest (ROI), the nodes. Another option to perform parcellation (not shown) is Independent Component Analysis (ICA). Data from the regions of interest is then summarised into a single time series per ROI (D). The data from the ROIs will then be used to perform edge detection (E), for example by calculating the correlation between ROI time series. Edges are collected in a correlation matrix (F) where the axes show all ROIs and the colour of the squares represents the correlation, or edge value, between two ROIs. Finally, the last panel shows the results of the analysis. The brain network can be shown in a topographic view (G), where the pink spheres represent the nodes and the blue cylinders the edges; or topological data can be shown (H), like the degree or clustering coefficient.

Cognitive Impairment in MS - Statistical and Neurophysiological aspects

PIs: Guy Nagels, Jacques De Keyser, Marie B D'hooghe

PhD student: Jeroen Van Schependom

Cognitive impairment is an important aspect of multiple sclerosis affecting about half of the MS population. In this project we have analysed clinical neuropsychological data collected during the period 2000-2014 at the National MS Center Melsbroek. By constructing survival curves for different cognitive tests, we were able to show the importance of information processing speed. Furthermore, we have demonstrated the value of the Symbol Digit Modalities Test in detecting general cognitive impairment (cross-validated sensitivity of 90 % at a specificity of about 60 %). Electrophysiological data allow us to assess the functioning of the brain at a timescale determined largely by the electronics used (typically well above 100 Hz). Clinical EEG data were analysed and substantial differences in network topology were found between a group of cognitively intact and cognitive impaired MS patients. In a follow-up study, we are collecting MEG data on a small group of MS patients.

A biomarker for cognition in multiple sclerosis, based on graph theoretical analysis of neurophysiological measurements

PIs: Guy Nagels, Jacques De Keyser, Marie B D'hooghe

PhD student: Jeroen Gielen

Reduced cognition is both hard to measure and hard to treat. We want to improve our ability to measure the impairment, in hopes of gaining a better understanding of what happens to the brain during the disease and being able to improve treatment. To do this we will gather brain scans from patients and look for regions that are more active during tasks concerning cognition, and how these regions interact with each other. This way we can visualise a brain network of interconnected regions. These networks will have certain properties, like the amount of connections the regions have or how many regions you would have to pass through to get from one region to another. We will summarise these properties in a biomarker. We are currently analysing brain scans of a group of students that performed two stimulus frequencies and modalities (auditory and visual) of Paced Serial Addition Testing (PSAT). Results show a lower effect of modality compared to stimulus frequency. This would imply that visual testing will be sufficient in cognitive testing over auditory testing, as it is also the preferred form of testing by patients.

Pattern classification techniques to improve the value of neurophysiological measurements for individual patients

PI: Guy Nagels

PhD student: Jorne Laton

Using electrophysiological measurements and machine learning we investigate classification and symptom estimation of individual patients suffering from neurological disorders. Classification algorithms are mainly suited for yes/no problems and are therefore useful for diagnosis and prognosis. Regression algorithms generate a continuous value, which is useful for estimating symptom severity, like fatigue and cognition. The data are EEG and MEG measurements, which can be analysed at sensor level, extracting features from the electrode signals or correlating these signals with each other. A 3D model can be generated from the 2D EEG/MEG data with source reconstruction, which allows to further analyse the data at source level. The focus of this project is mainly on schizophrenia and EEG, but a new study has been set up involving multiple sclerosis patients and MEG. In another study the effects of electroconvulsive therapy are to be estimated by comparing EEG measurements before and after therapy. In the first stage of the schizophrenia study, we have been able to show a reasonably high accuracy of 84% in separating patients from healthy controls. Next stages in this particular study are sensor space localisation of differences between the two groups and 3D source reconstruction to improve classification results.

Centre for Neurosciences • Vrije Universiteit Brussels • ©2017 • www.c4n.be • info@c4n.be
Faculty of Medicine & Pharmacie • Laarbeeklaan 103 • 1090 Brussel, BELGIUM • Tel: +32 (2) 477 64 10
Page generated in 0.0006 seconds.  •  Last update: 22 Oct 2015 13:41