Is it possible to identify anxiety, depression and/or alcoholism from EEG (resting EEG and event related potentials) and structural MRI data using machine learning algorithms?

Details

Supervisors

Macq, Benoît

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en mathématiques appliquées, à finalité spécialisée

Abstract

Identifying robust and non-invasive biomarkers of psychiatric diseases in the brain is still an open challenge. These unknowns prevent the effective treatment of patients suffering from alcoholism, depression and anxiety for example. In this work, we investigate whether it is possible using machine learning algorithms (SVM, random forest, ..) to predict from EEG and structural MRI data whether subjects are alcoholic, depressive or anxious. To do this, we used two datasets collected at St-Luc University Hospital. The first one includes 210 alcoholic, depressive, anxious and control subjects for whom structural MRI data have been collected. The second dataset includes 43 alcoholics with anxiety and depression comorbidities at the end of their withdrawal, for which EEG data (resting state and event related potentials) were collected. The results based on our two datasets showed that, (i) the structural MRI data allowed to differentiate alcoholics from controls, (ii) the most different variables between alcoholics and controls corresponded to the cortical thickness of the brain structures, and (iii) it was possible to identify slight differences between the ERP components of depressive and non depressive alcoholic patients. These results indicate that EEG and structural MRI data could be used to help diagnose psychiatric patients and that their potential should be exploited to devise new treatments and to study more complicated problems such as relapse in alcoholic patients.