DATA ANALYSIS OF AGE-RELATED CHANGES IN VISUAL

MOTION PERCEPTION

Nadejda Bocheva, Olga Georgieva

Institute of Neurobiology, Bulgarian Academy of Sciences, Acad. G. Bonchev str. Bl. 23, Sofia, Bulgaria

Faculty of Mathematics and Informatics, Sofia University, “St. Kl. Ohridski”, Sofia, Bulgaria

Miroslava Stefanova

Institute of Neurobiology, Bulgarian Academy of Sciences, Acad. G. Bonchev str. Bl. 23, Sofia, Bulgaria

Keywords: Mixed ANOVA, Fuzzy clustering, Psychophysical experiments, Motion discrimination, Ageing, Visual

perception, Individual, Gender and group differences.

Abstract: Many cognitive abilities decline with age, but ageing is accompanied by great variability within older

population. The aim of the present study is to explore the possibility to differentiate the age-related and the

individual differences in visual information processing. Two different analytical methods – mixed ANOVA

and fuzzy clustering, were applied to the data of psychophysical experiments on motion direction

discrimination. The results suggest that the complementary analysis based on both methods offers new

opportunities to retrieve information from the psychophysical studies and to separate the differences due to

age and gender from the individual differences of the participants. The proposed data analytic approach

allows better understanding of the factors that caused variation in performance with age and can be used as

a diagnostic tool to distinguish pathological from normal ageing.

1 INTRODUCTION

Ageing depends on a multitude of factors -

biological, genetic, social, economic, etc. The most

wide-spread approach in studying cognitive ageing

is to compare the performance of the participants

from two age groups (e.g. Habak & Faubert, 2000;

Raz, 2004; Govenlock et al., 2009; Berard et al.,

2009; Pilz, Bennett & Sekuler, 2010; Allen et al.,

2010) or to find a correlation between different

behavioural measures and age (e.g. Billino et al.,

2008). While these analyses provide valuable

information about the factors that affect the age-

related changes in the cognitive processes, the

individual differences among the subjects are greatly

undermined or are examined mainly through a

correlation analysis (e.g. Rose et al., 2010; Busey et

al., 2010). However, the variability among the

participants might include different life experiences,

genetic influences, preferred strategies and

susceptibility to neuropathology (e.g. Hedden &

Gabrielli, 2004). For this reason it is important to

evaluate the individual differences in order to

determine behavioral norms and to distinguish

normal from pathological ageing. The

psychophysical studies have the potential not only to

describe human performance but also, when

combined with other data, to be used as an indicator

for degenerative processes.

The aim of the present study is twofold: First, to

suggest new data analytic strategies for individual

differences between the participants in

psychophysical experiments; second, to determine

the most appropriate conditions for evaluation of

these differences. In order to achieve this goal we

selected four experiments from our dataset of studies

on visual perception of dynamic information. The

experiments focus on the age-related changes in the

sensitivity to motion direction of dynamic noisy

stimuli. Here we will concentrate only on the

evaluation of the different analytical techniques and

experimental conditions that will provide better

segmentation of the age-related changes from the

individual characteristics of the participants.

556

Bocheva N., Georgieva O. and Stefanova M..

DATA ANALYSIS OF AGE-RELATED CHANGES IN VISUAL MOTION PERCEPTION.

DOI: 10.5220/0003147505560561

In Proceedings of the 3rd International Conference on Agents and Artiﬁcial Intelligence (ICAART-2011), pages 556-561

ISBN: 978-989-8425-40-9

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

2 METHODS

2.1 Subjects

Twelve younger subjects (mean age 19.5 yrs., range

16-24 yrs., six male) and 12 older subjects (mean

age 73.9 yrs., range 66-82 yrs, four male)

participated in the experiments. All of them have

normal or corrected to normal vision.

2.2 Stimuli

The stimuli consisted of 50 frame movie sequences

showing spatially band-pass elements. They moved

in circular aperture with radius of 7.0 deg,

positioned at the middle of the computer screen.

In Experiments 1 and 3 the moving elements

were 32, whereas in Experiments 2 and 4 they were

128. If any element left the aperture during its

motion, it re-appeared from the opposite side of the

aperture in order to keep the number constant. The

mean speed of motion in all experiments was 6.64

deg/s. In Experiments 3 and 4 the speed of the

individual elements deviated from the mean speed to

a by a random amount determined from a normal

distribution with spread of 1.32 deg/s. The motion

direction of each element was taken randomly from

a normal distribution of different spread. Six

different values of the spread were used: 2°, 5°, 10°,

15°, 25°, and 35°. They determined the level of

external directional noise.

2.3 Procedure

The task of the observers was to indicate whether the

mean direction of motion appeared to the left or to

the right of the vertical. In all experiments the

average direction of motion was downwards. After

each trial, an adaptive algorithm estimated the

angular deviation of the mean direction from the

vertical to be presented on the next trial.

Each subject participated in four experiments.

The experiments were divided in 3 blocks: a training

session of 60 trials and two experimental that

involved six separate adaptive QUEST (Watson &

Pelli, 1983) staircases of 40 trials for each noise

level. The subject sat at a distance of 114 cm from

the computer screen. The observation was

binocular.

2.4 Statistical Analyses

A non-parametric bootstrap procedure was applied

to the responses of each subject and experimental

condition to find a maximum-likelihood estimate of

the discrimination thresholds (tm) at which 82%

correct responses were obtained. A Weibull function

was used as a psychometric function model.

A between-within subjects ANOVA (mixed

ANOVA) with factors: age, gender and noise level

was applied to the log transformed thresholds

obtained in each experiment. A multivariate

approach (Rencher, 1995) was used to the within-

subject tests. A post-hoc Tukey HSD test was

applied to the results of ANOVA with factors: noise

and subjects to divide the participants in

homogeneous groups.

2.5 Clustering Algorithm

A fuzzy clustering technique that finds not well

separated data groups with vague and uncertain

boundaries was used. The clusters are described by

their centre. In the simplest case, this is a point in the

data space that is most representative for the cluster

in probabilistic sense. Every point of the data space

belongs to the distinct clusters with different degree

of membership – a value between 0 and 1, such that

if the data is close to the cluster centre, the

membership degree is closer to 1.

The widespread Fuzzy-C-Means (FCM)

algorithm (Bezdek, 1981) was applied. It is an

objective function-based algorithm with clustering

criteria defined as:







J ,

(1)

where N is the number of data points; c is the

number of clusters; u

and d

denote

correspondingly the membership degree and the

Euclidean distance of the data point x

, k=1,...,N, to

the i-th cluster centre, i=1,...,c. The coefficient



[1,) determines how much clusters may

overlap. Usually m=2 is taken.

As the number of data N=24 is relatively small,

meaningful partition could be expected for

clustering in no more than two or three clusters. The

appropriate c could be accessed through the cluster

validity measures that estimate the goodness of the

obtained partition (Babuska, 1998):

a) Average within-cluster distance (AWCD)





AWCD

(2)

monotonically decreases with the number of

clusters.

b) Fuzzy hypervolume (Vh)

DATA ANALYSIS OF AGE-RELATED CHANGES IN VISUAL MOTION PERCEPTION

557

eriment 3 Ex

eriment 4

Experiment 1 Experiment 2

2 5 10 15 25 35

tm [

]

2 5 10 15 25 35

Older Younger

2 5 10 15 25 35

tm [

]

2 5 10 15 25 35

Older Younger

2 5 10 15 25 35

tm [

]

2 5 10 15 25 35

tm [

]

2 5 10 15 25 35

males

females





21/

)](det[Vh ,

(3)

where F

is a fuzzy covariance matrix of i-th cluster.

Good partitions are indicated by small values of Vh.

3 RESULTS AND DISCUSSION

3.1 Age and Gender Effects

In all experiments the mixed ANOVA results show

significant effect of age (F(1,20)=8.67; 6.60; 7.01

and 18.59 for Experiments 1-4, p<0.05) due to the

higher discrimination thresholds and thus, to the

lower sensitivity to motion direction, of the older

subjects. The gender has insignificant effect at

p=0.05 (F(1,20)=0.17; 2.58; 1.56 and 0.72 for

Experiments 1-4). The interaction between age and

gender was significant only in Experiment 2

(F(1,20)=4.91; p<0.05) due to the differences

between the male subjects from the two age groups.

The noise level had significant effect on the

performance; the thresholds increased with the

increase in the noise levels (F(5,16) = 13.37; 11.47;

19.66 and 15.66 for Experiments 1-4; p<0.05). No

significant interaction is observed between the noise

and either age or gender. The triple interaction was

also insignificant at p=0.05. Figure 1 illustrates the

effect of the noise level on the mean thresholds in

each experiment.

An additional ANOVA was performed with age

and gender as between-group factors and dot

number, the speed variability (present or absent) and

the noise level as within-group factors. The results

show again a significant and independent of the

noise level effect of the age (F(1,20)=13.50;

p<0.05). The speed variability affects the sensitivity

to motion direction – when the speed of the

individual elements is different, the performance

improves (F(1,20)=4.55; p<0.05). This effect is

more profound for the denser displays as indicated

by the significant interaction between the dot density

and the speed variability (F(1,20)=6.62; p<0.05).

The interaction between the noise level and the

speed variability is also significant

(F(5,16)=4.99;p<0.05). This result might be an

indication for noise exclusion for a variable speed

(e.g. Lu & Dosher, 2009) due to a better match of

the spatial characteristics of the band-limited

elements to the optimal motion sensitive

mechanisms in the visual system.

To summarize, the results from the experiments

show that the sensitivity to differences in motion

direction declines with age. The lack of interaction

between the noise level and the age group indicates

that the age-related changes in performance are due

to an increase in the multiplicative noise in the

visual system i.e. the noise that depends on the

strength of the signal (e.g. Lu & Dosher, 2009). The

results suggest little effect of the gender on the

sensitivity to motion direction, implying that based

on their sensitivity to motion direction, the subjects

could be divided in two groups determined by their

age. The data do not allow distinguishing which

experimental conditions are best for evaluating the

changes in the performance due to either age or

gender.

Figure 1: Effect of the noise level (on the abscissa) on tm.

The error-bars represent one standard error.

In the mixed ANOVA each subject serves as

his/hers own control and the analysis permits

evaluating the contribution of the experimental

factors on the performance and the common trend of

the effects for the experimental groups. However, in

order to implement this analysis, we need to ensure

equal variance of the experimental groups and

therefore, independence of the variance on the size

of the noise level. In the cluster analysis such

restrictions are not obligatory and we seek for

groups based on the performance of the subjects

using the untransformed data (i.e. without the

logarithmic transformation of the thresholds).

First, the two-dimensional space determined by

the age and the individual discrimination threshold

tm for a particular noise level has been investigated.

The FCM clustering provides the trivial partition in

two clusters – the clusters of younger and of older

subjects. The cluster centres are not sensitive to the

noise level added to the stimuli and the clusters’

coordinates are at about 19.5 years and 73.5 years.

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

558

Experiment 1 Experiment 2

Experiment 3 Experiment 4

Figure 2: Clustering results for c=3. Red: cluster centre;

triangles – males; asterisks – females.

The diversity within the trivial groups could be

investigated by partition in three clusters. In this

case the clustering is not stable as it provides more

than one (local) minimum solution. Clustering, that

finds minima of the criteria (1), for different initial

partitioning was accepted. Additional assessment of

the clustering quality according to the performance

indexes (2) and (3) was done.

Except in Experiment 1 at noise level of 15º and

35º, the AWCD and Vh indexes decrease for c=3.

This means that the partitioning in three clusters is

more informative as the clusters are more compact

and well defined. The common tendency observed

is the splitting the older group (Figure 2). This

corresponds to the complexity of the ageing process

and it dependence on multitude of factors that leads

to larger individual differences at older age.

However, this fact is not equally expressed and

depends on the spatial and motion stimulus

parameters. Noise level of 15º for all data sets

provides unstable clustering for c=3 for different

initial partitioning.

How could we explain the differences between

these two analytical techniques? In the ANOVA we

have assumed that not only the age, but also the

gender of the subjects affects the performance and

we have compared the sensitivity to motion direction

of 4 groups defined by age and gender. In the

clustering, the grouping is determined by the age of

the subjects and their performance. The splitting of

the older group in two clusters does not necessarily

imply that this is due to the gender of the subjects.

Therefore, using two different methods of analysis

we have obtained complementary data about the

age-related effects in motion direction

discrimination.

3.2 Characterization of the Individual

Differences

To evaluate the individual differences in task

performance ANOVA with factors: noise and

subject (regarded as a random factor), was applied

for each experiment. We wanted to evaluate how

stable is the performance in the different

experiments and whether data from one

experimental condition could predict the

performance if the conditions are changed. For this

purpose, a post-hoc Tukey HSD test was applied and

the subjects were divided in homogeneous groups

based on their sensitivity to motion direction. The

correlation coefficients between the ordering of the

subjects based on their performance in the four

experiments suggest that among the 6 comparisons

significant correlation is obtained only between the

orderings in Experiments 1 and 4 (correlation

coefficient =0.62; p<0.05). The homogenous groups

include subjects of different age and gender

The results of the analysis are presented in a

graphic form in Figure 3. Each column corresponds

to a different subject and each colour represents a

different homogeneous group. It is clear that no

distinct groups are formed between the different age

groups. The possibility to describe the individual

differences between the subjects by applying post-

hoc tests is limited due to the fact that in these

analyses we could not ensure equality of the

variance of the individual data by any

transformation. The measurements for each subject

are few, and therefore, the violation of the ANOVA

assumptions will have greater impact. Also, in the

test we have disregarded the potential interaction

between the subjects and the noise level. We have

no good measure to affiliate each subject to a group.

In this study only 24 subjects took part in the

experiments and the minimal number of

homogeneous groups that obtained by the post-hoc

Tukey HSD test is six. If more subjects participated,

DATA ANALYSIS OF AGE-RELATED CHANGES IN VISUAL MOTION PERCEPTION

559

the grouping based on their performance may

become more complicated due to the undetermined

separation between the groups and their large

number. For this reason, we have applied cluster

analysis to the data.

Figure 3: Representation of the homogeneous groups

defined by the post-hoc Tukey test. The first 12 columns

correspond to the older group, the second 12 – to the

younger one.

Information about the individual characteristics

of the subjects is searched by clustering in a higher

dimensional data space, defined by the noise level

values, independently of the subjects’ age and

gender. The division in three clusters is of a

particular interest as the participants are divided in a

group of high sensitivity (H) that corresponds to low

thresholds values, a group of medium (Md)

sensitivity and a low sensitivity group (L) that

corresponds to high thresholds values. The results

show that:

a) The cluster H of the low tm values contains

mainly younger people and a few older ones;

b) The medium cluster Md associates female and

male older subjects and few young females;

c) The cluster L is formed predominantly by the

older females.

The obtained fuzzy partition matrix assesses the

individual performance of a given subject to the

whole space partition. The level of association of a

person to every cluster is presented in Figure 4. It

could be seen that cluster H is more compact than

the others, which suggests small variation in the

sensitivity to motion direction of the members of

this cluster. As the others two groups predominantly

associate older people, this again confirms the

higher variability of the older group in comparison

with the younger one.

Figure 4: The distance of the individual discrimination

thresholds from the cluster centres. Each point represents

the data of a single subject. The order of the subjects in

each plot is the same. Triangles – male; asterisks – female;

blue – older participants; red – younger participants.

The general observation is that the clusters

overlap more for small noise levels as well as for

Experiments 1 and 2, whereas they are more

distinguishable for large levels of external noise.

When the external noise is low, the performance of

the subjects is limited mainly by the internal noise in

the visual system. At higher levels of external noise

the contribution of the internal noise decreases and

the performance is limited by the ability of the visual

system to “resist” to the external noise, for example,

by integrating the local motion information. The

clustering data suggest that the level of external

noise has little effect on the performance for the

participants in the high sensitivity group.

The cluster of the high thresholds depends

strongly on the external noise. In most experimental

conditions, except in Experiment 4, single subjects

belong to it, suggesting that they significantly

deviate in their performance from the rest of the

subjects. Experiment 4 allows the most distinct

separation of the subjects in different groups based

on their sensitivity to motion direction. It provides

the opportunity to classify reliably new subjects to

the existing groups and to treat the grouping as a

norm for characterizing people. The motion displays

in Experiment 4 contain a large number of moving

elements with varying speed where the different

realizations of the random samples might correspond

Medium sensitivity

Experiment 1 Experiment 2 Experiment 3 Experiment 4

Low sensitivity

High sensitivity

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

560

more closely to the characteristics of the

distributions associated with the noise level and the

speed of motion. In addition, the variable speed may

ensure optimal stimulation of the units in the brain,

sensitive to motion direction.

4 CONCLUSIONS

The common analytical approaches used to portray

the age-related changes in the different cognitive

processes provide a good description of the trends

associated with ageing and the experimental

conditions when differences occur. This makes them

a useful tool to characterize the process of ageing

and to seek association between the behavioural data

and the physiological changes in the brain. However,

these methods are less efficient when the individual

differences in the ageing process need to be

evaluated.

Our study is a first attempt to apply clustering

algorithms to differentiate the effects of age, gender

and individual differences on performance of a

behavioural task. The interpretation of the clustering

results allows detecting the deviation level of a

subject from the respective age group. Whether this

is related to degenerative processes or not, could not

be determined only by the results of the present

study; it requires tracking the changes in the

cognitive abilities of the participants in longitudinal

studies. However, our data analytic technique

provides opportunities to use psychophysical

methods for early diagnostics of the deterioration in

the cognitive abilities of the individual with age. The

results of the cluster analysis suggest also, that it is

easier to detect the individual deviations in

demanding tasks and difficult experimental

conditions.

ACKNOWLEDGEMENTS

This work was supported by grant TK01-200 of the

National Science Fund, Bulgaria.

REFERENCES

Habak, C., & Faubert, J., 2000. Larger effect of aging on

the perception of higher-order stimuli. Vision

Research, 40: 943–950.

Raz N., 2004. The aging brain observed in vivo:

differential changes and their modifiers. In: Cabeza R.,

Nyberg L. & Park D.C. eds. Cognitive neuroscience of

aging: linking cognitive and cerebral aging. New

York: Oxford University Press, 17-55.

Govenlock, S., Taylor, C. P., Sekuler, A. B. & Bennett , P.

J., 2009. The effect of aging on the orientation

selectivity of the human visual system. Vision

Research, 49: 2273-2284

Berard, J. R., Fung, J., McFadyen, B. J. & Lamontagne,

A., 2009. Aging affects the ability to use optic flow in

the control of heading during locomotion.

Experimental Brain Research, 194(2): 183-190.

Pilz, K. S., Bennett. P. J. & Sekuler, A. B., 2010. Effects

of aging on biological motion discrimination. Vision

Research, 50(2): 211-219

Allen, H. A., Hutchinson, C. V., Ledgeway, T., & Gayle,

P., 2010. The role of contrast sensitivity in global

motion processing deficits in the elderly. Journal of

Vision, 10: 15, 1–10,

Billino, J., Bremmer, F., & Gegenfurtner, K. R., 2008.

Differential aging of motion processing mechanisms:

Evidence against general perceptual decline. Vision

Research, 48: 1254–1261.

Rose, N. S., Rendell, P. G., McDaniel. M. A., Aberle, I., &

Kliegel, M., 2010. Age and individual differences in

prospective memory during a "Virtual Week": the

roles of working memory, vigilance, task regularity,

and cue focality. Psychology and Aging, 25: 595-605

Busey T., Craig J., Clark C. & Humes L., 2010. Age-

related changes in visual temporal order judgment

performance: Relation to sensory and cognitive

capacities. Vision Research, 50(17):1628-1640.

Hedden T. & Gabrieli, J. D., 2004. Insights into the ageing

mind: a view from cognitive neuroscience. Nature

Revues Neuroscience, 5: 87-96.

Watson, A. B. & Pelli, D. G., 1983. QUEST: a Bayesian

adaptive psychometric method. Perception &

Psychophysics, 33(2): 113-20.

Rencher, A. C., (1995). Methods of multivariate analysis.

John Wiley & Sons, Inc., New York.

Bezdek, J. C., 1981. Pattern Recognition with Fuzzy

Objective Function Algorithms. Plenum Press. New

York.

Babuska, R., 1998. Fuzzy Modeling for Control. Kluwer

Academic Publishers. Boston.

Lu, Z.-L., Dosher, B. A. 2009. Mechanisms of Perceptual

Learning. Learning and Perception, 1: 19–36.

DATA ANALYSIS OF AGE-RELATED CHANGES IN VISUAL MOTION PERCEPTION

561