| Zugriffsnummer | 25980 |
| Dokumenttyp | Konferenzartikel in Zeitschrift |
| Peer Review | unbekannt |
| Sprache | Englisch |
| Titel | Classifying ICA components of evoked MEG data |
| Autor(in); Institution |
Ghaemi, Dorsa; TU, Berlin, GERMANY
Kohl, Florian; 8.4, Mathematische Modellierung und Datenanalyse, PTB-Berlin
Orglmeister, Reinhold; TU, Berlin, GERMANY
|
| Quelle/Jahr | Proceedings BMT 2010, 44. DGBMT Jahrestagung, 3-Länder-Tagung D-A-CH, Rostock. Biomedizinische Technik: 55 (2010), Suppl. 1, 4 S. |
| Availability | [CD-ROM] ; file name: 1569315608.pdf |
| ISSN | 0939-4990 |
| Verlag | Berlin: de Gruyter |
| Konferenzangaben | BMT 2010, 44. DGBMT Jahrestagung, 3-Länder-Tagung D-A-CH, Rostock, Rostock, 05-08, October, 2010, Germany |
| Zusammenfassung | Independent component analysis (ICA) has become a popular tool to decompose stimulus evoked magnetoencephalographic (MEG) data. In principle, independent interfering signals are separated from the evoked neuronal signals of interest by ICA. However, the user is often left to choose the ICA component of interest out of many recovered components manually. Only recently, work has been done to automate this process and to provide the user with richer information than the bare ICA results. In this work we propose an automated selecting scheme using support vector machines (SVM) based classification of event-related signals and commonly encountered interfering signals. In contrast to previous work, we allow hybrid components to be classified by considering the classes: ‘event-related’, ‘cardiac’, ‘alpha’ as well as any mixture of these and ‘unknown’. It follows that the user obtains information about the ICA extracted components such as ‘this component has event-related characteristics’ or ‘this component has event-related and alpha wave characteristics’. We show the usefulness of the procedure for auditory-, visual- and auditory/visual-evoked MEG data. |
