A Blind Source Separation Algorithm for the Processing

and Classification of Electro-oculogram Data

Fernando Rojas

, Rodolfo V. García

, Olga Valenzuela

Luís Velázquez

and Belén San Román

Department of Computer Architecture and Technology, University of Granada, Spain

Network Department, University of Holguín, Cuba

Department of Applied Mathematics, University of Granada, Spain

Centre for the Research and Rehabilitation of Hereditary Ataxias, Holguín, Cuba

University of Granada, Spain

Abstract. Abnormalities in the oculomotor system are well known clinical

symptoms in patients of several neurodegenerative diseases, including

modifications in latency, peak velocity, and deviation in saccadic movements,

causing changes in the waveform of the patient response. The changes in the

morphology waveform suggest a higher degree of statistic independence in sick

patients when compared to healthy individuals regarding the patient response to

the visual saccadic stimulus modeled by means of digital generated saccade

waveforms. The electro-oculogram records of six patients diagnosed with ataxia

SCA2 (a neurodegenerative hereditary disease) and six healthy subjects used as

control were processed to extract saccades. We propose the application of a

blind source separation algorithm (or independent component analysis

algorithm) in order to find significant differences in the obtained estimations

between healthy and sick subjects. These results point out the validity of

independent component analysis based techniques as an adequate tool in order

to evaluate saccadic waveform changes in patients of ataxia SCA-2.

1 Introduction

The ocular movement records have been widely used in processing and classification

of biological signals and pathological conditions: clinical sleep scoring [10, 11],

cerebellar dysfunctions [12-14], diagnosis of the visual system [15, 16], amongst

others, also in human computer interface and visual guided devices [17-19]. The

Spino Cerebellar Ataxia type 2 (SCA-2) is an autosomal dominant cerebellar

hereditary ataxia with the highest prevalence in Cuba, reporting up to 43 cases per

100,000 inhabitants in the province of Holguin. In most families there is clinical and

neuropathological evidence of additional involvement of brainstem, basal ganglia,

spinal cord, and the peripheral nervous system [1]. This form of ataxia occurs

commonly in persons of Spanish ancestry in north-eastern Cuba, a figure much higher

than that found in western Cuba or in other parts of the world. The high prevalence is

Rojas F., García R., Valenzuela O., Velázquez L. and San Román B. (2009).

A Blind Source Separation Algorithm for the Processing and Classiﬁcation of Electro-oculogram Data.

In Proceedings of the 5th International Workshop on Artiﬁcial Neural Networks and Intelligent Information Processing, pages 50-56

DOI: 10.5220/0002263200500056

 SciTePress

probably the result of a founder effect, but might be due to an interaction between a

mutant gene and an unidentified environmental neurotoxin [2, 4].

Several studies have reported oculomotor abnormalities in SCA2 [1, 4-8].

Specifically, slowness of saccades has been suggested as a relatively characteristic

finding in SCA2[4, 8]. This fact determines significant differences in saccade

morphology between healthy individuals and patients with SCA-2, mainly for 60º of

stimulus amplitude. The electro-oculographical records are quite different in healthy

individuals and patients with a severe ataxia as it is shown in Figure 1 for a smooth

pursuit experiment.

Fig. 1. Electro-oculographic response to a smooth pursuit stimulus (top) obtained for a healthy

subject (center) and a patient of SCA-2 ataxia (bottom).

2 Using Blind Source Separation for Ataxia SCA2 Diagnosis

2.1 Hypothesis for the Proposed Method

Independent component analysis is aimed to find a linear transformation given by a

matrix W, so that the random variables y

, (i=1,…,n) of y=[y

,…,y

] are as

independent as possible in:

() ()yWxtt

⋅

(1)

This linear blind source separation approach is suitable for the signals obtained by

the EOG, as well as in other medical analysis such as electroencephalography (EEG),

electrocardiography (ECG), magneto-encephalography (MEG), and functional

magnetic resonance imaging (fMRI) [20-26].

As it was shown in Section 1, in the analysis of EOG oriented to the detection of

SCA2 experts anticipate two possible behaviors of the individuals: sick and healthy

conduct. During an experiment over a healthy subject, the horizontal movement of the

eye is expected to follow the stimulus signal. Therefore, the horizontal eye movement

and the stimulus will hold a direct dependence between them, i.e. the signals are not

independent. In contrast, a sick individual may present a more chaotic response,

depending on the severity of the disease. Consequently, the subject response will not

depend in such a high degree on the stimulus signal, and the signals are independent

(or at least, “not so dependent”).

Therefore, the proposed methodology uses independent component analysis as a

classification algorithm criterion: if the independence measure (normally mutual

information) reveals independence between the individual response and the stimulus

signal, then it is rather possible that the individual presents some degree of ataxia or

related disease.

2.2 Description of the Blind Source Separation Algorithm

The proposed algorithm for ataxia SCA-2 diagnosis will go along the following steps:

1. Set both horizontal response and stimulus signal in the same phase, i.e. correct the

delay between the stimulus change and the saccade.

2. Normalize signals (x).

3. Apply ICA algorithm. Any well known ICA algorithm may be applied at this point

(FastICA [27], Jade [28], GaBSS [29-30], etc.).

4. Normalize estimations (y)

5. Calculate error measure between estimations (y) and mixtures (x) according to the

root mean square error expression:

[]

() ()

(,)

xt yt

RMSE

−

∑

(2)

6. Depending on the obtained error measure, a simple categorization algorithm (such

as C-means) may be applied in order to classify individuals. Otherwise, a human

expert may help in subject categorization based on the ICA results.

3 Results

The electro-oculogram recordings of six patients with severe ataxia and six healthy

subjects diagnosed and classified in the “Centre for the Research and Rehabilitation

of Hereditary Ataxias (CIRAH)” were used in order to perform the analysis of

repeated ocular saccadic movement tests for 10º, 20º, 30º and 60º divergence stimuli.

Fig. 2. Stimulus (1), response (2) and ICA components (3 and 4) obtained at 60º of stimulation

for patients (top) and control subjects (bottom).

All the records were carried out by the medical staff of CIRAH. Each individual

was placed in a chair, with a head fixation device to avoid head movements, the

variables were collected by a two channel electronystagmograph (Otoscreen, Jaeger-

Toennies). Recording conditions were set as follows: electrodes of silver chloride

placed in the external borders of right eye (active electrode) and left eye (reference

electrode), high pass filtering 0.002 Hz, low pass filtering 20 Hz, sensitivity 200

µV/division, and sampling frequency 200 Hz. For stimulus generation a black screen

CRT display showing a white circular target with an angular size of 0.7º was used.

The stimulus and patient response data are automatically stored in ASCII files by

Otoscreen electronystagmograph.

The patient response was filtered using a median filter, to obtain a clean waveform

of the patient response, afterwards it was phased with the stimulus. Finally FastICA

was applied to get the independent components (See Figure 2).

As Figure 3 depicts, results show that the error measure obtained for SCA-2

patients is clearly differentiable for the same measure obtained for control subjects.

That is due to the fact mentioned in the hypothesis (Section 2.1) that if the

independence measure reveals independence between the individual response and the

stimulus signal, then it is possible that the individual presents some degree of ataxia.

When the original signals (stimulus and response) were independent, the estimations

are close to those sources and, therefore, the RMS error decreases.

Sick 1 Sick 2 Sick 3 Sick 4 Sick 5 Sick 6 Control 1 Control 2 Control 3 Control 4 Control 5 Control 6

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

SCA-2 patients

Control

Experiments

RMSE (Error)

Fig. 3. Root mean squared error between the estimations and the sources after the application of

the algorithm to EOG data corresponding to SCA-2 patients (left) and control subjects (right).

4 Discussion

The results were obtained from six control subjects and six patients. Confirming our

hypothesis, starting from electro-oculography experiments, patients showed a

different behavior in terms of the visual response to a fixed stimulus (see Figure 2 and

Figure 3). Therefore, after applying our proposed approach to the raw EOG data,

classification and diagnosis can be made easily by simple human inspection of the

results. Nevertheless, further research in this line may help in the categorization of the

several stages of severity of SCA-2.

The proposed method starts from the assumption that the response to a visual

stimulus is different in a healthy individual when compared to the response of an

individual afflicted by SCA-2. In the later situation, the response from the individual

is not dependent on the visual stimulus, so that the ICA algorithm estimations will be

similar to the obtained observations. This criterion has shown to be suitable in order

to distinguish between sick (patients) and healthy (control) individuals.

References

1. Bürk, K., Fetter, M., Abele, M., Laccone, F., Brice, A., Dichgans, J., Klockgether, T.:

Autosomal dominant cerebellar ataxia type I: oculomotor abnormalities in families with

SCA1, SCA2, and SCA3. Journal of Neurology 246 (1999) 789-797.

2. Orozco, G., Estrada, R., Perry, T.L., Araña, J., Fernandez, R., Gonzalez-Quevedo, A.,

Galarraga, J., Hansen, S.: Dominantly inherited olivopontocerebellar atrophy from eastern

Cuba: Clinical, neuropathological, and biochemical findings. Journal of the Neurological

Sciences 93 (1989) 37-50.

3. Velázquez Pérez, L., Sánchez Cruz, G., Canales Ochoa, N., Rodríguez Labrada, R.,

Rodríguez Díaz, J., Almaguer Mederos, L., Laffita Mesa, J.: Electrophysiological features

in patients and presymptomatic relatives with spinocerebellar ataxia type 2. Journal of the

Neurological Sciences 263 (2007) 158-164.

4. Velázquez, L.: Ataxia espino cerebelosa tipo 2. Principales aspectos neurofisiológicos en el

diagnóstico, pronóstico y evaluación de la enfermedad. Ediciones Holguín, Holguín (2006).

5. Rivaud-Pechoux, S., Dürr, A., Gaymard, B., Cancel, G., Ploner, C.J., Agid, Y., Brice, A.,

Pierrot-Deseilligny, C.: Eye movement abnormalities correlate with genotype in autosomal

dominant cerebellar ataxia type I. Annals of Neurology 43 (1998) 297-302.

6. Hutton, J.T., Albrecht, J.W., Kuskowski, M., Schut, L.J.: Abnormal ocular motor function

predicts clinical diagnosis of familial ataxia. Neurology 37 (1987) 698-701.

7. Yamamoto, H., Sobue, I.: Analysis of the abnormal ocular movements in spinocerebellar

degenerations and the influence of the thyrotropin releasing hormone. Clinical Neurology

20 (1980) 637-645.

8. Klostermann, W., Zühlke, C., Heide, W., Kömpf, D., Wessel, K.: Slow saccades and other

eye movement disorders in spinocerebellar atrophy type 1. Journal of Neurology 244

(1997) 105-111.

9. Velázquez-Pérez, L., Almaguer-Mederos, L., Santos-Falcón, N., Hechavarría-Pupo, R.,

Sánchez-Cruz, G., Paneque-Herrera, M.: Spinocerebellar ataxia type 2 in Cuba. A study of

the electrophysiological phenotype and its correlation with clinical and molecular variables.

Revista de Neurologia 33 (2001) 1129-1136.

10. Virkkala, J., Hasan, J., Värri, A., Himanen, S.-L., Müller, K.: Automatic sleep stage

classification using two-channel electro-oculography. Journal of Neuroscience Methods

166 (2007) 109-115.

11. Estrada, E., Nazeran, H., Barragan, J., Burk, J.R., Lucas, E.A., Behbehani, K.: EOG and

EMG: Two Important Switches in Automatic Sleep Stage Classification. Engineering in

Medicine and Biology Society, 2006. EMBS '06. 28th Annual International Conference of

the IEEE (2006) 2458-2461.

12. Spicker, S., Schulz, J.r.B., Petersen, D., Fetter, M., Klockgether, T., Dichgans, J.: Fixation

instability and oculomotor abnormalities in Friedreich's ataxia. Journal of Neurology 242

(1995) 517-521.

13. Yokota, T., Hayashi, H., Hirose, K., Tanabe, H.: Unusual blink reflex with four

components in a patient with periodic ataxia. Journal of Neurology 237 (1990) 313-315.

14. Hubner, J., Sprenger, A., Klein, C., Hagenah, J., Rambold, H., Zuhlke, C., Kompf, D.,

Rolfs, A., Kimmig, H., Helmchen, C.: Eye movement abnormalities in spinocerebellar

ataxia type 17 (SCA17). Neurology 69 (2007) 1160-1168.

15. Güven, A.l., Kara, S.: Classification of electro-oculogram signals using artificial neural

network. Expert Systems with Applications 31 (2006) 199-205.

16. Irving, E.L., Steinbach, M.J., Lillakas, L., Babu, R.J., Hutchings, N.: Horizontal Saccade

Dynamics across the Human Life Span. Invest. Ophthalmol. Vis. Sci. 47 (2006) 2478-2484.

17. Kumar, D., Poole, E.: Classification of EOG for human computer interface. Annual

International Conference of the IEEE Engineering in Medicine and Biology - Proceedings,

Vol. 1 (2002) 64-67.

18. Brunner, S., Hanke, S., Wassertheuer, S., Hochgatterer, A.: EOG Pattern Recognition Trial

for a Human Computer Interface. Universal Access in Human-Computer Interaction.

Ambient Interaction (2007) 769-776.

19. Komogortsev, O.V., Khan, J.I.: Eye movement prediction by Kalman filter with integrated

linear horizontal oculomotor plant mechanical model. Proceedings of the 2008 symposium

on Eye tracking research & applications. ACM (2008) 229-236.

20. Vaya, C., Rieta, J.J., Sanchez, C., Moratal, D.: Convolutive Blind Source Separation

Algorithms Applied to the Electrocardiogram of Atrial Fibrillation: Study of Performance.

Biomedical Engineering, IEEE Transactions on 54 (2007) 1530-1533.

21. Anemüller, J.r., Sejnowski, T.J., Makeig, S.: Complex independent component analysis of

frequency-domain electroencephalographic data. Neural Networks 16 (2003) 1311-1323.

22. Funase, A., Yagi, T., Barros, A.K., Cichocki, A., Takumi, I.: Single trial analysis on

saccade-related EEG signal. Neural Engineering, 2007. CNE '07. 3rd International

IEEE/EMBS Conference on (2007) 371-374.

23. Hironaga, N., Haruhana, K., Liu, L.C., Fenwick, P.B.C., Ioannides, A.A.: Monitoring of

eye movement and its use for artifact elimination. International Congress Series 1270

(2004) 134-137.

24. Milanesi, M., Martini, N., Vanello, N., Positano, V., Santarelli, M., Landini, L.:

Independent component analysis applied to the removal of motion artifacts from

electrocardiographic signals. Medical and Biological Engineering and Computing 46

(2008) 251-261.

25. Guilhon, D., Medeiros, E., Barros, A.K.: ECG Data Compression by Independent

Component Analysis. Machine Learning for Signal Processing, 2005 IEEE Workshop on

(2005) 189-193.

26. Kocyigit, Y., Alkan, A., Erol, H.: Classification of EEG Recordings by Using Fast

Independent Component Analysis and Artificial Neural Network. Journal of Medical

Systems 32 (2008) 17-20.

27. Hyvarinen, A., Oja, E.: A Fast Fixed-Point Algorithm for Independent Component

Analysis. Neural Computation 9 (1997) 1483-1492.

28. Cardoso, J.-F.: Source separation using higher order moments. ICASSP, IEEE International

Conference on Acoustics, Speech and Signal Processing - Proceedings, Vol. 4 (1989) 2109-

2112.

29. Rojas, F., Alvarez, M.R., Salmerón, M., Puntonet, C.G., Martin-Clemente, R.: Adaptive

and Heuristic Approaches for Nonlinear Source Separation. Proceedings of the

International Joint Conference on Neural Networks, Vol. 1 (2003) 720-725.

30. Martin-Clemente, R., Puntonet, C.G., Rojas, F.: Post-nonlinear blind source separation

using methaheuristics. Electronics Letters, Vol. 39, Issue: 24. (2003) 1765- 1766.