550 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 54, NO. 3, MARCH 2007 [5] D. D. Streeter, “Gross morphology and fiber geometry of the heart,” in Handbook of Physiology, R. M. Berne, Ed. Baltimore, MD: Williams & Williams, 1979, vol. 1, ch. 2: The Cardiovascular System, pp. 61–112. [6] P. A. Helm, H.-J. Tseng, L. Younes, E. R. McVeigh, and R. L. Winslow, “Ex vivo 3D diffusion tensor imaging and quantification of cardiac lam- inar structure,” Magn. Reson. Med., vol. 54, pp. 850–859, 2005. [7] R. Plonsey and R. C. Barr, “The four-electrode resistivity technique as applied to cardiac muscle,” IEEE Trans. Biomed. Eng., vol. BME-29, pp. 541–546, 1982. [8] A. van Oosterom, R. W. de Boer, and R. T. Van Dam, “Intramural resis- tivity of cardiac tissue,” Med. Biol. Eng. Comput., vol. 17, pp. 337–343, 1979. [9] P. Steendijk, G. Mur, E. T. van der Velde, and J. Baan, “The four-elec- trode resistivity technique in anisotropic media: Theoretical analysis and application on myocardial tissue in vivo,” IEEE Trans. Biomed. Eng., vol. 40, no. 11, pp. 1138–1147, Nov. 1993. [10] P. Le Guyader, P. Savard, and F. Trelles, “Measurement of myocardial conductivities with an eight-electrode technique in the frequency do- main,” in Proc. 17th IEEE-EMBS, 1995, pp. 71–72. [11] P. Le Guyader, F. Trelles, and P. Savard, “Extracellular measurement of anisotropic bidomain myocardial conductivities. I. Theoretical anal- ysis,” Ann. Biomed. Eng., vol. 29, pp. 862–877, 2001. [12] R. C. Barr and R. Plonsey, “Electrode systems for measuring cardiac impedances using optical transmembrane potential sensors and inter- stitial electrodes—Theoretical design,” IEEE Trans. Biomed. Eng., vol. 50, no. 8, pp. 925–934, Aug. 2003. [13] L. Tung, “A bi-domain model for describing ischaemic myocardial DC potentials,” Ph.D. dissertation, Massachusetts Inst. Technol., Cam- bridge, 1978. [14] B. M. Johnston, P. R. Johnston, and D. Kilpatrick, “A new approach to the determinination of cardiac potential distributions: Application to the analysis of electrode configurations,” Math. Biosci., vol. 202, no. 2, pp. 288–309, 2006. [15] R. M. Gulrajani, Bioelectricity and Biomagnetism. New York: Wiley, 1998. [16] H. P. William, P. F. Brian, A. T. Saul, and T. V. William, Numerical Recipes, The Art of Scientific Computing, 2nd ed. Cambridge, MA: Cambridge Univ. Press, 1992. [17] Y. Wang, P. H. Schimpf, D. R. Haynor, and Y. Kim, “Geometric effects on resistivity measurements with four- electrode probes in isotropic and anisotropic tissues,” IEEE Trans. Biomed. Eng., vol. 45, no. 7, pp. 8–19, Jul. 1998. [18] A. E. Pollard and R. C. Barr, “Cardiac micro-impedance measurement in two-dimensional models using multisite interstitial stimulation,” Am. J. Physiol.-Heart Circulatory Physiol., vol. 290, no. 5, pp. H1976–H1987, 2006. [19] F. X. Witkowski, K. M. Kavanagh, P. A. Penkoske, and R. Plonsey, “In vivo estimation of cardiac transmembane current,” Circ. Res., vol. 72, pp. 424–439, 1993. [20] C.-S. Kim, S. Ufer, C. M. Seagle, C. L. Engle, H. T. Nagle, T. A. Johnson, and W. E. Cascio, “Use of micromachined probes for the recording of cardiac electrograms in isolated heart tissues,” Biosensors Bioelectron., vol. 19, no. 9, pp. 1109–1116, Apr. 2004. [21] L. Clerc, “Directional differences of impulse spread in trabecular muscle from mammalian heart,” J. Physiol., vol. 255, pp. 335–346, 1976. [22] D. E. Roberts, L. T. Hersh, and A. M. Scher, “Influence of cardiac fiber orientation on wavefront voltage, conduction velocity and tissue resistivity in the dog,” Circ. Res., vol. 44, pp. 701–712, 1979. [23] D. E. Roberts and A. M. Scher, “Effects of tissue anisotropy on extra- cellular potential fields in canine myocardium in situ,” Circ. Res., vol. 50, pp. 342–351, 1982. Study of On-Line Adaptive Discriminant Analysis for EEG-Based Brain Computer Interfaces C. Vidaurre*, A. Schlögl, R. Cabeza, R. Scherer, and G. Pfurtscheller Abstract—A study of different on-line adaptive classifiers, using various feature types is presented. Motor imagery brain computer interface (BCI) experiments were carried out with 18 naive able-bodied subjects. Experi- ments were done with three two-class, cue-based, electroencephalogram (EEG)-based systems. Two continuously adaptive classifiers were tested: adaptive quadratic and linear discriminant analysis. Three feature types were analyzed, adaptive autoregressive parameters, logarithmic band power estimates and the concatenation of both. Results show that all systems are stable and that the concatenation of features with continuously adaptive linear discriminant analysis classifier is the best choice of all. Also, a comparison of the latter with a discontinuously updated linear discriminant analysis, carried out in on-line experiments with six subjects, showed that on-line adaptation performed significantly better than a discontinuous update. Finally a static subject-specific baseline was also provided and used to compare performance measurements of both types of adaptation. Index Terms—AAR, automatic adaptive classification, band power esti- mates, BCI, Kalman filtering, LDA, on-line adaptation, QDA. I. INTRODUCTION An electroencephalogram (EEG)-based brain computer interface (BCI) is a system which enables people to control devices using signals recorded from the scalp. One of the goals of such a device is to assist patients who have highly compromised motor functions, such as completely paralyzed patients with e.g., amyotrophic lateral sclerosis [1]–[11]. A BCI is divided in different modules: preprocessing, feature extrac- tion, classification and feedback. Various signals are used in BCI sys- tems, but our experiences were based in EEG signals, which can vary in time. Therefore, adaptation of modules like feature extraction or/and classification is a very important issue in BCI research. Regarding the classification module, up to now classifiers were manual and discon- tinuously updated after a number of runs, and the moment when to perform this upgrade was decided by the researcher depending on sub- jective factors such as his/her experience, [4], [12], [13]. Because of the disadvantage of this manual adaptation, recently several BCI research groups have started to work on automatic adaptation of the classifiers [14]–[18]. Our group presents in this paper results from on-line experiments with two different automatic adaptive classifiers and three different fea- ture extraction techniques. The adaptation of the classifier is done in a Manuscript received November 17, 2005; revised August 11, 2006. This work was supported in part by the Spanish Ministry of Culture and Education under Grant AP-2000–4673, in part by the L. B. Gesellschaft, Austria, and in part by “Fonds zur Förderung der wissenschaftlichen Forschung”, Austria, under Project 16326-BO2. Asterisk indicates corresponding author. *C. Vidaurre is with the Department of Electrical and Electronic Engineering, Public University of Navarre, Campus Arrosadia s/n, 31006 Pamplona, Spain (e-mail: carmen.vidaurre@unavarra.es). A. Schlögl is with the Human-Computer Interface Department, Graz Univer- sity of Technology, 8010 Graz, Austria. R. Cabeza is with the Department of Electrical and Electronic Engineering, Public University of Navarre, 31006 Pamplona, Spain. R. Scherer and G. Pfurtscheller are with the BCI-Lab Computer Graphics Department, Graz University of Technology, 8010 Graz, Austria. Color versions of Figs. 4–6 are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBME.2006.888836 0018-9294/$25.00 © 2007 IEEE