DocumentCode :
636855
Title :
Boosting specificity of MEG artifact removal by weighted support vector machine
Author :
Fang Duan ; Phothisonothai, Montri ; Kikuchi, Masashi ; Yoshimura, Yuki ; Minabe, Yoshio ; Watanabe, K. ; Aihara, Kazuyuki
Author_Institution :
Dept. of Electr. Eng. & Inf. Syst., Univ. of Tokyo, Tokyo, Japan
fYear :
2013
fDate :
3-7 July 2013
Firstpage :
6039
Lastpage :
6042
Abstract :
An automatic artifact removal method of magnetoencephalogram (MEG) was presented in this paper. The method proposed is based on independent components analysis (ICA) and support vector machine (SVM). However, different from the previous studies, in this paper we consider two factors which would influence the performance. First, the imbalance factor of independent components (ICs) of MEG is handled by weighted SVM. Second, instead of simply setting a fixed weight to each class, a re-weighting scheme is used for the preservation of useful MEG ICs. Experimental results on manually marked MEG dataset showed that the method proposed could correctly distinguish the artifacts from the MEG ICs. Meanwhile, 99.72%±0.67 of MEG ICs were preserved. The classification accuracy was 97.91%±1.39. In addition, it was found that this method was not sensitive to individual differences. The cross validation (leave-one-subject-out) results showed an averaged accuracy of 97.41%±2.14.
Keywords :
independent component analysis; magnetoencephalography; medical signal processing; signal classification; support vector machines; ICA; MEG artifact removal; MEG dataset; automatic artifact removal method; classification accuracy; independent component analysis; leave-one-subject-out cross validation; magnetoencephalogram; weighted SVM; weighted support vector machine; Accuracy; Electroencephalography; Feature extraction; Independent component analysis; Sensitivity; Support vector machines; Training;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Engineering in Medicine and Biology Society (EMBC), 2013 35th Annual International Conference of the IEEE
Conference_Location :
Osaka
ISSN :
1557-170X
Type :
conf
DOI :
10.1109/EMBC.2013.6610929
Filename :
6610929
Link To Document :
بازگشت