BCI On Feature Extraction from Multi-Channel Brain Waves Used for Brain Computer Interface Hiroya SAITO Kenji NAKAYAMA Akihiro HIRANO Graduate School of Natural Science and Technology,Kanazawa Univ. E-mail: nakayama@t.kanazawa-u.ac.jp (MLNN) (BCI) (BCI) BCI (ICA) ( (Permutation ) ) MLNN [][2] BCI ICA Permutation MLNN AR [3][4] (SOFM) SOFM BCI []-[9] ABSTRACT BCI []-[2] FFT and Multilayer Neural Networks (MLNN) have been applied to Brain Computer Interface (BCI). In the BCI systems, how to extract individual features of mental tasks from multi-channel brain waves is an important problem. Independent Component Analysis (ICA) is one of feature extraction methods. However, there exits a Permutation Problem, that is, feature extracting order is easily changed data by data. Therefore, ICA cannot be effectively used for generating the MLNN input data. In this paper, we try to develop MLNN structures, which can reduce effects of the permutation problem. (ICA) [3] ICA (Permutation ) (SOFM) Furthermore, Self-Organizing Feature Map (SOFM) is applied to generating the MLNN input data. Classification performance can be improved by increasing dimension 2 Web of the SOFM output. []
Baseline Multiplication Letter-composing 3 Rotation Counting left C3 P3 O front back C4 P4 O2 right : : 3 :. C3 C4P3P4OO2EOG 7 EOG : MLNN 2Hz 2Hz sec=2, 7 3 (Independent Compornent Analysis:ICA) []-[2] (ICA) n s(t) = (s (t),..., s n (t)) T n 7 4 x(t) = (x (t),..., x n (t)) T A t 2 x(t) = As(t) (t =, 2, ) ICA s(t) A 7 7 s(t) A ICA 7 7 BP) 4. 4 [3] : : 2 : :. B M L R C [%] 9 9 6 8 84 : ICA FastICA [4]
6 MLNN 6. 7NN ICA Permutation NN 7 7 NN 2 4. 8= 4 :.4 2= N N N N N N N N 2 N N 2 N N 2 N N 7 N N 7 N N 7 2: 7NN 7 NN MLNN Permuation 3 NN NN 7 2 NN NN NN 7 NN 2 7NN (NN )7 (NN ) = 3 7 7 2 NN 7NN 7 NN NN Permutation 4 2 7NN 3 MT N N NN NN2 2 NN MT N N NN NN2 2 NN [%] B M L R C 6 7 7 4 8 6 7 9 9 4 7 2 7NN 6 6 8 8 66 2: ICA (7NN ) [%] B M L R C 4 2 4 2 4 3 3 32 2 7NN 3 4 4 34 3: ICA (7NN ) ICA 6% ICA 3% MLNN 6.2 NN NN7 NN7 3: 2 7NN
MT 2 7 MT2 2 7 MT 2 7 NN 2 7 4: NN MT2 2% 4. ICA :.4 : 6 [%] B M L R C 2 3 4 28 3 3 3 3 34 4: ICA (NN ) [%] B M L R C 3 2 3 3 24 3 6 7 44 : ICA (NN ) 7NN Permutation ICA SOFM ICA M (2 ) n 7, NN w j j =, 2,..., M SOFM 6.3 ICA 7 ICA (central) C3C4(parietal lobe) P 3P 4(occipital lobe) OO2 3 2 FastICA EOG Permutation 4. : 8 B M L R C [%] 4 7 6 8 8 66 6: ICA ICA BCI 7 7. (Self-Organizing Feature Map:SOFM) (SOFM) : SOFM SOFM Step w j () Step2
x = (x, x 2,..., x n ) 7.2 SOFM MLNN SOFM Step3 () : : 87 x w c = min x w j j =, 2,..., M () c, Step4 4 4 3 3 2 2 w j (t) = { wj (t) + η(t)[x(t) w j (t)] j Λ c w j Otherwise (2) η(t) Λ c c η() =. η(t) = η()( ( t T )2 ) T t T = 6 SOFM (c+d, c+d), (c d, c+d), (c+ d, c d), (c d, c d) Λ c d d() = round(.3 (M row )(M cul )) d(t) = round(d()( ( d() )( t d() T )2 )) round() () T = d(t) 7. 2 2 3 3 4 4 8: SOFM Baseline Multiplication Letter-composing Rotation Counting 7: SOFM MLNN 9 4 4 3 3 2 2.9.8.7.6 4. 4 3. 2 2 3 3 4 4..4.3.2. 3 2. 2. 9: SOFM MLNN 2 3 4 6 7 8 9 6: η(t) Step Step2 2 3 4 6 7 8 9 7: d(t) 7.3 SOFM Step2Step MLNN
() : : 4. 8 [%] B M L R C 9 8 7 9 9 84 9 7 9 9 88 9 7 9 9 8: SOFM [3] C.Anderson and Z.Sijercic, Classification of EEG Signals from Four Subjects During Five Mental Tasks, In Solving Engineering Problems with Neural Networks: Proceedings of the Conference on Engineering Applications in Neural Networks (EANN 96), ed. by Bulsari, A.B., Kallio, S., and Tsaptsinos, D., Systems Engineering Association, PL 34, FIN-2 Turku, Finland, pp.47 44, 996. [4] G.Pfurtscheller and C.Neuper,Motor imagery and direct brain-computer communication, Proc. IEEE, vol. 89, no. 7, pp.23-34, July 2. [] J.R.Millan, J.Mourino, F.Babiloni, F.Cincotti, M.Varsta, and J.Heikkonen, Local neural classifier for EEG-based recognition of metal tasks, IEEE- INNS-ENNS Int. Joint Conf. Neural Networks, July 2. 7 =2 SOFM 36% SOFM 8 BCI (ICA) [2] K.Nakayama and K.Inagaki, A brain computer interface based on neural network with efficient pre- (SOFM) processing, Proc.IEEE, ISPACS26, Yonago, Japan, pp.673-676, Dec.26. SOFM [3] REBOLLEDO MENDEZ Jovan David Brain Computer Interface Based on ICA and Neural Network BCI 27 [4] 2 [] G.Pfurtscheller, C.Neuper, C.Guger, W.Harkam, H.Ramoser, A.Schlögl, B.Obermaier,and M.Pregenzer, Current trends in Graz braincomputer interface(bci) research, IEEE Trans. Rehab.Eng., vol.8, pp.26-29, 2. [2] B.Obermaier, G.R.Muller, and G.Pfurtscheller, Virtual keyboard controlled by spontaneous EEG activity, IEEE Trans. Neural Sys. Rehab. Eng., vol., no. 4, pp.422-426, Dec. 23. [6] K.R.Muller, C.W.Anderson, and G.E.Birch, Linear and non-linear methods for brain-computer interfaces IEEE Trans. Neural Sys. Rehab. Eng., vol., no. 2, pp.6-69, 23. [7] J.R.Millan, On the need for on-line learning in brain-computer interfaces, Proc. IJCNN, pp.2877-2882, 24. [8] G.E.Fabiani, D.J.McFarland, J.R.Wolpaw, and G.Pfurtscheller, Conversion of EEG activity into cursor movement by a brain-computer interface (BCI), IEEE Trans. Neural Sys. Rehab. Eng., vol. 2, no. 3, pp.33-338, Sept. 24. [9] C.W.Anderson, S.V.Devulapalli, and E.A.Stolz, Determining mental state from EEG signals using neural networs, Scientific Programming, Special Issue on Applications Analysis, vol.4, no.3, pp.7-83, Fall, 99. ICA [] http://www.cs.colostate.edu/eeg/ [],,,,, Vol. No.74 pp.2-3, SIP2-4, 2.7.