Combining EEG and MRI in a Multimodal Approach for Parkinson's Disease Detection

Currently, the diagnosis of Parkinson's disease (PD) relies essentially on the assessment of the symptoms by a clinician; it is therefore important to identify quantitative biomarkers to better characterize this disease. In this paper, we describe a way to combine data from MRI and EEG signals to help improve PD diagnosis. Features were extracted using texture analysis and diffusion parameters for the MRI data, and by calculating effective connectivity estimators on reconstructed sources for the EEG data. With linear discriminant analysis, for a binary classification between healthy subjects and Parkinsonian patients, we obtained an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{9 5. 5 \%}$</tex> with both modalities combined. Comparatively, we had <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{6 3. 6 \%}$</tex> for MRI alone and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{9 1 \%}$</tex> for EEG alone. These results highlight the relevance of a multimodal approach notably to improve the early diagnosis of PD.