Applications of Eye Tracking in the Diagnosis of Early Stages of Autism

Spectrum Disorders

Giampiero Dalai, Sashidharan Komandur and Frode Strand Volden

Department of Design, Norwegian University of Science and Technology, Teknologiveien 22, 2815 Gjøvik, Norway

Keywords:

Eye Tracking, Eye Movements, Test Design, Autism Spectrum Disorders, Diagnosis, Children.

Abstract:

In this project we designed a computerized diagnostic test procedure to measure gaze parameters known

to be related to early symptoms of Autism Spectrum Disorders (ASD) (saccades and smooth pursuit eye

movements). An eye-tracker was used to gather gaze data. Custom visual stimuli and guidelines for collecting

eye tracking data from the subject group (children between 12-24 months of age) were developed. A ﬁrst

proof of principle study was performed on three children without suspected clinical diagnosis in the target age

range. The results were promising and the procedure seems to be applicable to small children. Further work

needs to be carried out in order to validate whether the procedure actually will be a good diagnostic support

tool in clinical settings.

1 INTRODUCTION

The aim of the research is to understand how eye

tracking technology can be used for the early de-

tection of Autism Spectrum Disorders (ASD) in child-

ren younger than the average age for diagnosis. ASD

is a complex neurodevelopmental disorder which af-

fects behavior, communication and social functioning

(APA, 2017). It can be diagnosed as early as 15 - 18

months of age. Even so, the average age of diagnosis

is about 4.5 years, and some subjects are not diagno-

sed until adulthood (APA, 2017).

As B

olte et al. (2016) describe, research into early

stages of ASD make use of various methods for ex-

amining the children development and responses to

interventions, right from the ﬁrst months and ye-

ars of life. The methods can be classiﬁed into two

groups: (1) informant and clinician-based behavio-

ral methods (e.g. questionnaires, observation scales,

interviews and developmental tests), which are more

based on observation, subjective and sometimes qua-

litative; (2) technology based and/or measurements of

basic cognitive or neurological processes and structu-

res (e.g. eye tracking, electroencephalography (EEG),

functional and structural magnetic resonance imaging

(MRI)), which are more direct, objective and mostly

quantitative. It is conceivable that eye tracking can be

used as an integrated part of screening and diagnostic

assessments in the future (Falck-Ytter et al., 2013).

Diagnosing ASD in the ﬁrst 24-30 months of life

of a child poses particular challenges to clinicians,

among which that there are no objective diagnostic bi-

omarkers for ASD (Samad et al., 2017). Issues in sen-

sorimotor control are involved in ASD (Johnson et al.,

2016), therefore tools for measuring these kind of de-

ﬁcits are promising. A particularly promising appli-

cation of eye trackers is the study on young children,

in order to capture early-emerging developmental me-

chanisms in this critical period of the development

and to illuminate the early course and characteristics

of ASD. Indeed, eye tracking has already been lar-

gely used in studies on people with ASD (for some re-

cent reviews, see B

olte et al., 2016; Falck-Ytter et al.,

2013; Frazier et al., 2017; Johnson et al., 2016; Papa-

giannopoulou et al., 2014).

Eye trackers can provide measurements on eye

movements which are impossible to assess with naked

eye (e.g. saccades, smooth pursuit). Remote eye trac-

kers (infrared / corneal reﬂection types) are unobtru-

sive and do not constrain movements, which makes

them ideal to use on small children for early diagno-

sis (B

olte et al., 2016; Falck-Ytter et al., 2013; Samad

et al., 2017).

The present on-going study does not aim to deve-

lop an eye tracking procedure with inherent diagnos-

tic value, but it aims to develop a framework (con-

sisting of a rationale of relevant eye parameters and

the methods to measure them) and a procedure which

proves to be applicable to the target children. The

study intends to apply the ﬁndings in the eye tracking

156

Dalai, G., Komandur, S. and Volden, F.

Applications of Eye Tracking in the Diagnosis of Early Stages of Autism Spectrum Disorders.

DOI: 10.5220/0006951201560162

In Proceedings of the 2nd International Conference on Computer-Human Interaction Research and Applications (CHIRA 2018), pages 156-162

ISBN: 978-989-758-328-5

scientiﬁc literature to clinical settings for ASD diag-

nosis. We started the research work as a Master The-

sis (Dalai, 2018) and we intend to reﬁne and improve

the tentative procedure outlined in this paper by doing

further research, until in the future we will be able to

assess its clinical diagnostic value.

In this context, we see the eye tracking technolo-

gies as a supporting tool for ASD diagnosis, which are

designed to be used together with the already valida-

ted behavioral diagnostic praxis. Given the complex-

ity of ASD, it is fundamental to have a rich descrip-

tion of the children’s development condition, inclu-

ding quantitative and qualitative measurements. The

current screening and diagnostic tests (see Mag

an-

Maganto et al., 2017; Towle and Patrick, 2016; Char-

man and Gotham, 2013) already assess qualitatively

atypical gaze behaviors as a sign of ASD. Even so, the

clinicians need to assess many other complex behavi-

oral manifestations related to ASD in order to formu-

late a diagnosis. Eye tracking alone cannot provide a

deﬁnitive answer, but it can provide useful evidences

to support clinical decision making.

2 BACKGROUND

We reviewed literature with the aim of identifying

those eye-tracking metrics which seem to be suitable

to support early ASD diagnosis. The vast majority

of the collected literature shows results from experi-

ments done on adolescents and/or adults with ASD

(see for example the review by Johnson et al. (2016)),

while the number of studies involving small child-

ren in the early diagnosis age range is rather low (see

for example the review by Falck-Ytter et al. (2013)).

Even the few studies which have found differences in

speciﬁc eye parameters between typically developing

and ASD groups, have been carried out on subjects in

older age groups. This offers one of the motivations

to start a systematic experimentation with the target

children group.

So far, the eye parameters we have identiﬁed in

literature as potential useful early ASD markers are:

standard deviation of saccade gain (Johnson et al.,

2016), open-loop and closed-loop smooth pursuit gain

(Johnson et al., 2016; Takarae et al., 2004). Up to

this point, we have left aside the analysis on ﬁxati-

ons on Areas of Interest (AOI), due to the difﬁculty in

determining the ecological validity of the stimuli in

a controlled experimental setting. We left aside also

the analysis on pupil diameter, since this parameter is

sensitive to environmental lightning and it is difﬁcult

to assess consistently in a variety of different clinical

settings. The framework focuses on the assessment of

saccades and smooth pursuit eye movements, which

require high frequency and precision measurements

and also a more controlled experimental environment.

These two condition suit better eye tracking techno-

logy.

Discussing with a speech therapist expert in ASD

diagnosis (Minichiello, S., 2017, personal communi-

cation) it emerged that a possible validation process

for the usage of eye trackers in this clinical context

requires the following steps: (1) deﬁning the steps

of age for testing with the children (e.g. 18-24-36

months of age); (2) recording a series of eye tracking

measurements on a typically developing group, in or-

der to collect baseline data and to assess the reliability

of the measurements; (3) administering the same pro-

cedure on ASD children, in order to assess differences

between groups.

3 METHODS

Conducting eye tracking studies on small children po-

ses a series of challenges. Small children cannot be

instructed to behave in a certain way or to focus on a

stimulus for a certain period of time, due to their ob-

vious lack of linguistic competence for understanding

complex task instructions. Therefore, the visual sti-

muli need to be salient and interesting enough to be

followed with the gaze by the children, without furt-

her prompting by the researchers. Another issue is to

keep the children’s interest over the whole experiment

time. Short trials displaying simple eye-catching vi-

sual stimuli should be staggered with some audiovi-

sual contents as interstimulus materials, in order to

keep the children entertained and provide variety and

playfulness.

3.1 Apparatus

We carried out a ﬁrst iteration of testings, as a proof

of principle, by using a SMI



RED250mobile

eye

tracker. Its sample rate is up to 250 Hz, which is

sufﬁcient for tracking saccades, smooth pursuit mo-

vements and ﬁxations with great accuracy. It is moun-

ted on a laptop computer, underneath a 15.6 inch dis-

play monitor (refresh rate 60 FPS, covering around 39

degrees of visual angle horizontally and 22 degrees

vertically at 50 cm of distance, resolution 1920x1080

px). The setup is portable and can be carried around

and used in different settings.

Applications of Eye Tracking in the Diagnosis of Early Stages of Autism Spectrum Disorders

157

3.2 Setting

The research protocol follows the guidelines provided

by Sasson and Elison (2012) for eye tracking studies

on young children with ASD. The light in the room

should be kept a bit dim, in order to encourage the

child to focus on the display monitor on which the

visual stimuli are presented. The caregiver sits on

a chair (positioned in front of the screen at an ade-

quate distance) and the child sits on the caregiver’s

lap throughout all the experiment, in order to make

both the child and the parent feel more at ease. De-

pending on the location, the researchers can use an

external monitor connected to the eye tracker laptop

computer, and they operate the computer in a position

out of the child’s sight. If the use of an external moni-

tor is not practical or not feasible (as it was in the ﬁrst

proof of principle study), the researchers start the eye

tracking experimental routine on the laptop and let it

run until the end, placing themselves far away enough

to not disturb the experiment.

3.3 Procedure

The experiment consist of four different phases:

• Introduction and setup phase: The caregiver and

the child take a seat, and they are administered the

informed consent form. In the meantime, a video

is shown on display monitor, in order to start to

capture the child’s attention;

• Calibration phase: When the caregiver feels that

he/she and his/her child are comfortable and re-

ady to start, a calibration routine is shown on the

display monitor;

• Visualization phase: A series of videos is shown

on the screen. Some of the them are stimuli mate-

rials, others are inter-stimulus materials with the

aim of keeping the child entertained, but which

are not suitable for eye tracking measurements;

• Conclusion phase: When the series of stimuli is

over, a message is displayed on the display moni-

tor and the test is ended.

The detailed experimental protocol has been sent for

approval to the Norwegian Regional Committees for

Medical and Health Research Ethics (REK), which

did not raise any issue concerning the procedure.

3.4 Stimuli

We identiﬁed appropriate experimental paradigms for

each eye parameter of interest, leading so far to the

development of four kinds of stimuli which we put

together in a single experimental procedure:

• Sinusoidal motion stimuli, assessing closed-loop

smooth pursuit gain (von Hofsten and Rosander,

1997);

• Triangular motion stimuli, assessing closed-loop

smooth pursuit gain (von Hofsten and Rosander,

1997);

• Step ramp task, assessing open-loop and closed-

loop smooth pursuit gain (Takarae et al., 2004);

• Step paradigm (Zalla et al., 2016) for visually gui-

ded saccade tasks, assessing the standard devia-

tion of saccade gain (Johnson et al., 2016).

We followed the recommendations from Smyrnis

(2008) in order to set up the parameters for the crea-

tion and the presentation of the stimuli, among which

sampling frequency (≥200 Hz), amplitude ranges, di-

rection of movement, number of cycles, etc.

In order to attract the children’s attention only on

the moving target in the stimuli, we designed it as the

only colored element in the scene, while the back-

ground was medium-gray.

We generated the visual stimuli programmatically

by developing scripts in the Processing 3 software,

which is based on the Java programming language.

The executive ﬁles are parametric and they allow the

researchers to manipulate the stimuli variables. The

software generates the visual stimuli automatically,

taking care for example of the conversions between

measurement units (pixels to degrees of visual angle),

the drawing of periodic waves starting from trigono-

metric parameters, the rendering of the target mo-

vement, etc. Due to the fact that the stimuli are pa-

rametric, they can be adapted to be used on different

devices and for different experimental designs. Figu-

res 1, 2, 3 and 4 illustrate schemes and visuals from

the experiment stimuli.

4 EXPECTED RESULTS

The experimentation has not fully started yet, but we

have started the recruitment process. We carried out

a ﬁrst small proof of principle study (Dalai, 2018, pp.

53–81) on three typically developing children (9, 15

and 24 months of age, all female) in order to assess

the applicability of the procedure to target subjects

and the effectiveness of the eye tracking technology

to collect the eye parameters of interest. We analy-

zed qualitatively the eye tracking records of the ex-

periment subjects and compared the records with the

ones of on an adult subject (male, 25 years old) with

no vision deﬁciencies an no clinical diagnosis. The

adult subject’s records acted as reference dataset. As

an example of the kind of data we collected, Figure 5

CHIRA 2018 - 2nd International Conference on Computer-Human Interaction Research and Applications

158

Figure 1: Scheme of the sinusoidal motion stimuli. Reworking from Dalai (2018, p. 59).

Figure 2: Scheme of the triangular motion stimuli. Reworking from Dalai (2018, p. 59).

Figure 3: Scheme of the Step-Ramp stimuli. Reworking from Dalai (2018, p. 60).

Applications of Eye Tracking in the Diagnosis of Early Stages of Autism Spectrum Disorders

159

Figure 4: Scheme of the visually guided saccades stimuli. Reworking from Dalai (2018, p. 61).

shows the eye tracking record of the adult subject, and

Figure 6 shows the record of the 24 months old sub-

ject. All the graphs have been outputted by the SMI

eye tracker software. In the graphs it is possible to

observe similar gaze patterns.

The framework and procedure seem to elicit the

expected gaze patterns in the children, while impro-

vements are needed in terms of timings, quantity of

recorded data, some stimuli parameters (e.g. target

velocity) and guidelines for the setting. Indeed, even

if the procedure elicits the right quality of gaze pat-

terns, the quantity of recorded data is still low overall.

Table 1 shows some metrics of the proof of principle

study with the target group. The tracking ratio is a

quantitative metric outputted by the eye tracker soft-

ware, which describes how much eye movement data

was collected during a trial, and basically how efﬁ-

cient was the eye tracker in collecting data. The rese-

archers determined the percentage of correctly perfor-

med repetitions by analyzing visually the graphs out-

putted by the eye tracker software. A high percentage

of correct repetitions of the same gaze pattern streng-

thens the reliability of the measurements. It is more

of a qualitative metric: the lower it is, the more the

correct gaze patterns are scattered and not continu-

ous. A low percentage might be due to the children’s

lack of attention to the stimuli target, or to the children

moving too far away from the eye tracker or in other

directions. The last metric was determined qualitati-

vely by the researchers, stating if overall it was possi-

ble to detect in the graphs the expected gaze patterns

under investigation. The children’s records show the

expected gaze patterns for the custom visual stimuli

in 9 out of 10 total performed trials. Therefore, the

procedure shows potential to analyze speciﬁc types of

eye movements in the target children, if not younger

subjects.

We analyzed the eye tracker records with the aim

to highlight possible improvements for the procedure

and no judgement was done on the children’s perfor-

mance. More reﬁned and sophisticated algorithms for

the analysis of the data are needed in order to discern

clearly between the various types of eye movements

and analyze them statistically (see for example (Gior-

dano et al., 2017; Jansson and Medvedev, 2013; Lars-

son et al., 2015)). Nevertheless, we have already de-

veloped preliminary guidelines for the computation of

the parameters of interest from the eye tracking re-

cords and we have identiﬁed improvements for the

experimental procedure and stimuli. In particular, the

experimental design needs to balance better the dura-

tion of the presentation of the stimuli with the neces-

sary amount of stimuli repetitions needed for recor-

ding a reliable amount of data. The active collabora-

tion of the caregivers helps in redirecting the attention

of the children towards the display monitor.

The study will contribute to formulate a clearer

hypothesis for a future validation study of the pro-

cedure, which will aim to support clinical diagnos-

tic decisions. Systematic measurements conducted on

target children and in collaboration with clinical per-

sonnel will provide the necessary data about the eye

parameters of interest. The data will allow to assess

the validity of oculomotor performance as an early

ASD indicator. Collaborations with mathematicians

and software engineers will allow to complete the fra-

mework in its data analysis part. An on-going dialo-

gue with cognitive- and neuro-scientists will also al-

low to investigate further on which neural pathways

could be involved in the divergent development of the

oculomotor control in ASD.

CHIRA 2018 - 2nd International Conference on Computer-Human Interaction Research and Applications

160

Figure 5: Pilot test eye tracker record, 25 years old, male, no clinical diagnosis, sinusoidal motion target. Picture taken from

Dalai (2018, p. 66).

Figure 6: Participant’s eye tracker record, 24 months old, female, no clinical diagnosis, sinusoidal motion target. Picture taken

from Dalai (2018, p. 72).

Table 1: Summary of the experiment preliminary results. Reworking from Dalai (2018, p. 70).

Participant

(Age months) Test Trial n. Stimulus Tracking ratio (%) Repetitions (%) Correct pattern

P1 (15) 1 1 Sinusoidal 58,8% ≈50% (3/6) Yes

2 Triangular 71,6% ≈33.3% (2/6) Yes

3 Step-Ramp 79,3% ≈75% (6/8) Yes

4 Saccades 37,9% ≈20% (3/15) Yes

P2 (24) 2 1 Sinusoidal 78,7% ≈75% (4.5/6) Yes

2 Triangular 54,9% ≈33.3% (2/6) Yes

3 Step-Ramp 10,7% 0% (0/8) No

4 Saccades Not performed – –

P3 (9) 3 1 Sinusoidal 23,6% ≈16.6% (1/6) Yes

2 Triangular 52,9% ≈50% (3/6) Yes

3 Step-Ramp 29,1% ≈25% (2/8) Yes

4 Saccades Not performed – –

Applications of Eye Tracking in the Diagnosis of Early Stages of Autism Spectrum Disorders

161

REFERENCES

APA (2017). Diagnosing and managing autism spectrum

disorder (asd). Online. http://www.apa.org/

helpcenter/autism.aspx. Last accessed on June 13,

2018.

olte, S., Bartl-Pokorny, K. D., Jonsson, U., Berggren, S.,

Zhang, D., Kostrzewa, E., Falck-Ytter, T., Einspieler,

C., Pokorny, F. B., Jones, E. J. H., Roeyers, H., Char-

man, T., and Marschik, P. B. (2016). How can clini-

cians detect and treat autism early? methodological

trends of technology use in research. Acta Paedia-

trica, 105(2):137–144.

Charman, T. and Gotham, K. (2013). Measurement is-

sues: Screening and diagnostic instruments for au-

tism spectrum disorders - lessons from research and

practise. Child Adolesc Ment Health, 18(1):52–63.

Dalai, G. (2018). Eye tracking as a supporting diag-

nostic tool for Autism Spectrum Disorders. NTNU,

Gjøvik. http://hdl.handle.net/11250/2502572. Last

accessed on July 30, 2018.

Falck-Ytter, T., B

olte, S., and Gredeb

ack, G. (2013). Eye

tracking in early autism research. Journal of Neuro-

developmental Disorders, 5(1):28.

Frazier, T. W., Strauss, M., Klingemier, E. W., Zetzer, E. E.,

Hardan, A. Y., Eng, C., and Youngstrom, E. A. (2017).

A meta-analysis of gaze differences to social and non-

social information between individuals with and wit-

hout autism. Journal of the American Academy of

Child & Adolescent Psychiatry, 56(7):546–555.

Giordano, D., Pino, C., Kavasidis, I., Spampinato, C., Pie-

tro, M. D., Rizzo, R., Scuderi, A., and Barone, R.

(2017). An eye tracker based computer system to

support oculomotor and attention deﬁcit investigati-

ons. In 2017 IEEE 30th International Symposium

on Computer-Based Medical Systems (CBMS), pages

538–543.

Jansson, D. and Medvedev, A. (2013). Visual stimulus

design in parameter estimation of the human smooth

pursuit system from eye-tracking data. In 2013 Ame-

rican Control Conference, pages 887–892.

Johnson, B. P., Lum, J. A. G., Rinehart, N. J., and Fiel-

ding, J. (2016). Ocular motor disturbances in autism

spectrum disorders: Systematic review and compre-

hensive meta-analysis. Neuroscience & Biobehavioral

Reviews, 69(Supplement C):260–279.

Larsson, L., Nystr

om, M., Andersson, R., and Stridh, M.

(2015). Detection of ﬁxations and smooth pursuit mo-

vements in high-speed eye-tracking data. Biomedical

Signal Processing and Control, 18:145 – 152.

Mag

an-Maganto, M., J

onsd

ottir, S. L., S

anchez-Garc

ıa,

A. B., Garc

ıa-Primo, P., Hellendoorn, A., Charman,

T., Roeyers, H., Dereu, M., Moilanen, I., Muratori, F.,

Paz, M. P. d. l., Rog

e, B., Oosterling, I. J., Yliherva,

A., and Canal-Bedia, R. (2017). Measurement issues:

Building a theoretical framework for autism spectrum

disorders screening instruments in europe. Child and

Adolescent Mental Health.

Papagiannopoulou, E. A., Chitty, K. M., Hermens, D. F.,

Hickie, I. B., and Lagopoulos, J. (2014). A systema-

tic review and meta-analysis of eye-tracking studies in

children with autism spectrum disorders. Social Neu-

roscience, 9(6):610–632.

Samad, M. D., Diawara, N., Bobzien, J. L., Harrington,

J. W., Witherow, M. A., and Iftekharuddin, K. M.

(2017). A feasibility study of autism behavioral mar-

kers in spontaneous facial, visual, and hand movement

response data. IEEE Transactions on Neural Systems

and Rehabilitation Engineering, PP(99):1–1.

Sasson, N. J. and Elison, J. T. (2012). Eye tracking young

children with autism. Journal of Visualized Experi-

ments, (61):3675.

Smyrnis, N. (2008). Metric issues in the study of eye

movements in psychiatry. Brain and Cognition,

68(3):341 – 358.

Takarae, Y., Minshew, N. J., Luna, B., Krisky, C. M., and

Sweeney, J. A. (2004). Pursuit eye movement deﬁcits

in autism. Brain, 127(12):2584–2594.

Towle, P. O. and Patrick, P. A. (2016). Autism spectrum dis-

order screening instruments for very young children:

A systematic review. Autism Research and Treatment,

2016:29.

von Hofsten, C. and Rosander, K. (1997). Development

of smooth pursuit tracking in young infants. Vision

Research, 37(13):1799 – 1810.

Zalla, T., Seassau, M., Cazalis, F., Gras, D., and Leboyer,

M. (2016). Saccadic eye movements in adults with

high-functioning autism spectrum disorder. Autism,

0(0):1–10.

CHIRA 2018 - 2nd International Conference on Computer-Human Interaction Research and Applications

162