Evaluation of the Maps

SOM can provide an efficient means of visualizing the relationships among the neurons and the important topological information can be obtained through an unsupervised learning process. A high dimensional EEG epoch is seen as a point which is projected onto the trained map. The locations of these projections differentiate among various EEG epochs, and indicate cognitive states during the allocation of brain resources. They are motivated by the face that representation of sensory information in the human brain has a geometrical order [55].

The unsupervised learning interactions allow clusters of neurons to win the competition, and then those neurons are adjusted to bring about a better response to the current input however the size of map is. Iterative application of this learning process, the specify clusters of this map that are topological close, being sensitive to clusters of inputs that are physical close in the input data. In other words, there is a correspondence between signal features and response locations on the map. We tried many different sizes (10*10, 15*15, 20*20, and 25*25) to get the suitable topology for the input data and the final maps are twenty five neurons square in this study. The different map size result that our SOM based exploratory data analysis using EEG suggests existence of distinct signatures among these five cases. The topographic structure of these unequal size maps is so consistent to represent the distraction levels during driving. The EEG epochs recorded from two single-tasks are projected onto the corner of each map, and the other EEG epochs with competition of brain resource are located in the middle of every map.

The results of 10*10 maps were showed in Fig. 5-3. We conclude that the structure of each 10*10 map is concerned with the distraction effect. The values of

quantify the clustering performance of the maps. The average accuracy of each case for each size map is shown in table-3 and Fig. 5-4. The correct accuracy of three dual conditions for each small map is just about 60% ant it is lower than the bigger ones. In the small size map, there is no any unlabeled neuron and we can assume the unlabeled neurons to be the boundary among the clusters. Another quality index for maps is the mixed neurons. There are EEG epochs of more than two conditions in these mixed neurons. The numbers of mixed neurons in different map size are shown in table-4.

The fewer neurons are mixed, the better results we get.

Fig. 5-3: The results of SOM with dimensions 10*10.

These maps are similar to the maps with bigger size shown in Fig. 4-6. The smaller maps make the classes (especially for the two single conditions) a little more compact.

(A) (B) two trained maps and (C) (D) are the labeling accuracy of these two maps.

Fig. 5-4: The relation between the map size and the accuracy of labeling.

This figure shows the accuracy increasing by the bigger map size. The number of map size means the length of one dimension. In our research, we use the square to represent the map. When the size is more than 24, the accuracy of all 5 cases is not significant changing. The map size is 25 in our research and the map with this size can provide the high resolution and be trained with less time.

The third index to evaluate the performance is the quantization error. At the end of the self-organization, each reference vector on the map resembles a cluster of similar input vectors which occurred during the self-organization [27]. The Euclidean distance between the reference vector of best-matching neuron and the input EEG epoch gives the quantization error. When samples differ substantially from those

Table-3: Total mixed neurons in each map The map size Training 1 Training 2

10*10 84/100 88/100

15*15 21/225 18/225

20*20 57/400 43/400

25*25 30/625 39/625

quantization errors in these maps from small size to bigger ones and the value of error is almost equal. The map with small size clustered together; on the other hand, the bigger maps could be more accurate in recognizing the difference feature among each case. Because the accuracy of each case and the number of un-mixed neurons are highest in the 25*25 map, we chose these square maps.

5.4 Distraction Level and Behavior Performance

When the subjects just controlled the car or answered the mathematic equations, the EEG signals were corresponding to one main cluster of each map (as showed in Fig. 4-1). And the other EEG epochs of dual tasks were mapped onto some subgroups to present the distraction levels with designed SOA conditions. When the subjects process two designed tasks, we can interpret brain activities by the trained maps.

In our results, we could clearly cluster the EEG epochs of different designed cases and the accuracy of each case is more than 90%. According the previous studies, the dual tasks involving driving and answering simple math questions in the stimulus onset asynchrony (SOA) are no significant difference in the behavior data among three dual-task conditions, such as response time and driving performance [17]. This study investigated how performance of two overlapping discrete tasks was organized and controlled that suggested that sequential performance of overlapping tasks was scheduled in advance and was regulated by initially allocating brain resources to one task and subsequently switching to the other task [56]. And the designed tasks were all visual-stimuli tasks. When the subjects responded these two tasks, the brain source in the frontal area and motor area would be compete. Therefore, these two visual-stimuli tasks interfered with each other and the interferences presented on brain dynamics. In this study, the EEG epochs of dual tasks were clustered into many

subgroups as showed in Fig.4-5, Fig. 4-6, and Fig. 4-7. We could find that the brain activities in three dual conditions were different.

6. Conclusions

A number of measurement procedures for brain activities have been used to classify and predict cognitive states. In particular, Artificial Neural Networks have been widely employed to model cognitive states by performing EEG data classification. In this study, we proposed an unsupervised approach of visualizing and verifying the brain activities for distraction levels during driving. Our results show the topological relation of the input data. The SOM algorithm provides a new visualization method to analysis the EEG data. We extracted the EEG epochs from the continuous EEG data in Frontal and Motor components by the independent component analysis. The steps of feature processing which we proposed are effective to reduce the variation among the different subjects. The processed EEG epochs in the same case are clustered through the unsupervised learning processes. In the trained maps, the accuracy of each case is more than 90 %. The distributions of all recorded EEG epochs are consistent with the human brain activities. Our results demonstrated that five cases (three dual tasks and two single tasks) can be distinguished clearly by the SOM based method. Especially each single task is clustered in a distinct spatial area of the maps and the other dual tasks show several subgroups in the middle of the maps. We create two models to test the recorded EEG signal. The results of these two testing models show that the EEG epochs of single driving are clearly identified. For such dual tasks although there is no significant difference in the behavioral data, such as response time and driving performance, our SOM based exploratory data analysis using EEG suggests existence of distinct signatures among the five cases. The Frontal and Motor components which we extracted are the main activities area of responding multiple tasks at the same time during driving. Furthermore, the recognition of

distraction levels will help us to monitor the driver safety and warn them to pay more attention during driving to decrease or avoid the traffic accidents.

Reference

1. Young, K. and M. Regan, Driver distraction: A review of the literature.

Distracted driving. Sydney, NSW: Australasian College of Road Safety, 2007:

p. 379-405.

2. Dingus, T.A., et al., The 100-Car Naturalistic Driving Study, Phase II–Results of the 100-Car Field Experiment (Report No. DOT HS 810 593). Virginia Tech.

Transportation Institute-Sponsored by National Highway Traffic Safety Administration, 2006.

3. Horberry, T., et al., Driver distraction: The effects of concurrent in-vehicle tasks, road environment complexity and age on driving performance. Accident Analysis and Prevention, 2006. 38(1): p. 185-191.

4. Patten, C.J.D., et al., Using mobile telephones: Cognitive workload and attention resource allocation. Accident Analysis and Prevention, 2004. 36(3):

p. 341-350.

5. Dukic, T., L. Hanson, and T. Falkmer, Effect of drivers' age and push button locations on visual time off road, steering wheel deviation and safety perception. Ergonomics, 2006. 49(1): p. 78-92.

6. Lee, J.D., et al., Speech-based interaction with in-vehicle computers: The effect of speech-based e-mail on drivers' attention to the roadway. Human Factors, 2001. 43(4): p. 631.

7. Consiglio, W., et al., Effect of cellular telephone conversations and other potential interference on reaction time in a braking response. Accident Analysis & Prevention, 2003. 35(4): p. 495-500.

8. Horrey, W.J. and C.D. Wickens, Driving and side task performance: The effects of display clutter, separation, and modality. Human Factors, 2004.

46(4): p. 611.

9. Haigney, D.E., R.G. Taylor, and S.J. Westerman, Concurrent mobile (cellular) phone use and driving performance: task demand characteristics and compensatory processes. Transportation Research Part F: Psychology and Behaviour, 2000. 3(3): p. 113-121.

10. Engstrom, J., E. Johansson, and J. Ostlund, Effects of visual and cognitive load in real and simulated motorway driving. Transportation Research Part F:

Psychology and Behaviour, 2005. 8(2): p. 97-120.

11. Brookhuis, K., G. De Vries, and D. De Waard, The effects of mobile telephoning on driving performance. Accident Analysis & Prevention, 1991.

23(4): p. 309-316.

12. Pal, N.R., et al., EEG-based Subject-and Session-independent Drowsiness

Detection: An Unsupervised Approach.

13. Shen, K.Q., et al., A feature selection method for multilevel mental fatigue EEG classification. IEEE Transactions on Biomedical Engineering, 2007.

54(7): p. 1231-1237.

14. Vuckovic, A., et al., Automatic recognition of alertness and drowsiness from EEG by an artificial neural network. Medical Engineering and Physics, 2002.

24(5): p. 349-360.

15. Huang, R.S., et al. EEG pattern recognition- arousal states detection and classification. 1996.

16. Artaud, P., et al., An on-board system for detecting lapses of alertness in car driving. 1994.

17. Lin, C.T., et al. Distraction-related EEG dynamics in virtual reality driving simulation. 2008.

18. Deiber, M.P., et al., Distinction between perceptual and attentional processing in working memory tasks: A study of phase-locked and induced oscillatory brain dynamics. Journal of cognitive neuroscience, 2007. 19(1): p. 158-172.

19. Szameitat, A.J., et al., Task-order coordination in dual-task performance and the lateral prefrontal cortex: an event-related fMRI study. Psychological Research, 2006. 70(6): p. 541-552.

20. Stelzel, C., et al., The neural effect of stimulus-response modality compatibility on dual-task performance: an fMRI study. Psychological Research, 2006.

70(6): p. 514-525.

21. McFarland, D.J., et al., Bci meeting 2005-workshop on bci signal processing:

feature extraction and translation. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2006. 14(2): p. 135.

22. Lotte, F., et al., A review of classification algorithms for EEG-based Brain-Computer Interfaces. J Neural Eng, 2007. 40(4).

23. AlZoubi, O., I. Koprinska, and R.A. Calvo. Classification of Brain-Computer Interface Data. 2008.

24. Lee, F., et al. A comparative analysis of multi-class EEG classification for brain computer interface. 2005.

25. Qin, J., Y. Li, and A. Cichocki. ICA and committee machine-based algorithm for cursor control in a BCI system: Springer.

26. Jung, T.P., et al., Removing electroencephalographic artifacts by blind source separation. Psychophysiology, 2000. 37(02): p. 163-178.

27. Joutsiniemi, S.L., S. Kaski, and T.A. Larsen, Self-organizing map in recognition of topographic patterns of EEGspectra. IEEE Transactions on

28. Fukunaga, K., Introduction to statistical pattern recognition. 1990: Academic press.

29. Babiloni, F., et al., Recognition of imagined hand movements with low resolution surface Laplacian and linear classifiers. Medical Engineering and Physics, 2001. 23(5): p. 323-328.

30. Millan, J.R., et al., A local neural classifier for the recognition of EEG patterns associated to mental tasks. IEEE Transactions on Neural Networks, 2002. 13(3).

31. Kohonen, T., The self-organizing map. Neurocomputing, 1998. 21(1-3): p. 1-6.

32. Kohonen, T. and P. Somervuo, Self-organizing maps of symbol strings.

Neurocomputing, 1998. 21(1-3): p. 19-30.

33. Bezdek, J.C. and N.R. Pal, A note on self-organizing semantic maps. IEEE Transactions on Neural Networks, 1995. 6(5): p. 1029-1036.

34. Ritter, H., T. Martinetz, and K. Schulten, Neural Computation and Self-Organizing Maps; An Introduction. 1992: Addison-Wesley Longman Publishing Co., Inc. Boston, MA, USA.

35. Hewitson, B.C. and R.G. Crane, Self-organizing maps: applications to synoptic climatology. Climate Research, 2002. 22(1): p. 13-26.

36. Lin, C.T., et al. EEG Effects of Motion Sickness Induced in a Dynamic Virtual Reality Environment. 2007.

37. Stewart, D. A platform with six degrees of freedom(Platform with six degrees of freedom for flight simulation in pilot training). 1965.

38. Erickson, K.I., et al., Neural correlates of dual-task performance after minimizing task-preparation. NeuroImage, 2005. 28(4): p. 967-979.

39. Jasper, H.H., Report of the committee on methods of clinical examination in electroencephalography. Electroencephalogr Clin Neurophysiol, 1958. 10: p.

370-375.

40. Makeig, S., et al., Independent component analysis of electroencephalographic data. Advances in neural information processing systems, 1996: p. 145-151.

41. Comon, P., C. Jutten, and J. Herault, Blind separation of sources, part II:

Problems statement. Signal processing, 1991. 24(1): p. 11-20.

42. Cardoso, J.F. and A. Souloumiac, Blind beamforming for non Gaussian signals. IEE proceedings-f, 1993.

43. Comon, P., Independent component analysis, a new concept? Signal processing, 1994. 36(3): p. 287-314.

44. Bell, A.J. and T.J. Sejnowski, An information-maximization approach to blind separation and blind deconvolution. Neural computation, 1995. 7(6): p.

1129-1159.

45. Cardoso, J.F. and B.H. Laheld, Equivariant adaptive source separation. IEEE Transactions on Signal Processing, 1996. 44(12): p. 3017-3030.

46. Lee, T.W., M. Girolami, and T.J. Sejnowski, Independent component analysis using an extended infomax algorithm for mixed subgaussian and supergaussian sources. Neural computation, 1999. 11(2): p. 417-441.

47. Vesanto, J., et al., SOM toolbox for Matlab 5. 2000: Helsinki University of Technology.

48. Kohonen, T., Self-organization and associative memory. 1989.

49. Burgess, P.W., Strategy application disorder: the role of the frontal lobes in human multitasking. Psychological Research, 2000. 63(3): p. 279-288.

50. Sarnthein, J., et al., Synchronization between prefrontal and posterior association cortex during human working memory. 1998, National Acad Sciences. p. 7092-7096.

51. Fernandez, T., et al., EEG activation patterns during the performance of tasks involving different components of mental calculation. Electroencephalography and Clinical Neurophysiology, 1995. 94(3): p. 175-182.

52. Kuhlman, W.N., EEG feedback training: enhancement of somatosensory cortical activity. Electroencephalography and Clinical Neurophysiology, 1978.

45(2): p. 290.

53. Kuhlman, W.N., Functional topography of the human mu rhythm.

Electroencephalography and Clinical Neurophysiology, 1978. 44(1): p. 83.

54. Schoppenhorst, M., et al., The significance of coherence estimates in determining central alpha and mu activities. Electroencephalography and Clinical Neurophysiology, 1980. 48(1): p. 25.

55. Kumar, S., Neural Networks: a classroom approach. 2004: Tata McGraw-Hill.

56. De Jong, R., The role of preparation in overlapping-task performance. The Quarterly Journal of Experimental Psychology Section A, 1995. 48(1): p. 2-25.