Marolize is in private practice with Dirk Booysen and Associates. She has applied to do her Masters on Eye Tracking next year. In her young career, she already has a list of impressive achievements such as:
Best Final Year Clinical Optometry Student – 2016: University of Johannesburg., Department of Optometry. Elected as best performing student in Clinic. Earned achievement as a final year student.
Best Clinical Optometry Student – 2015: University of Johannesburg, Department of Optometry.
Elected as best performing student in Clinic. Earned achievement as a third-year student.
Contributed to Medical Textbook – In Contact: Clinical Contact Lens Practice 2018
Wrote a chapter on contact lens optics and also created the digital illustrations for the book authored by Dr. Dirk Booysen
Eye tracking is not a new concept in vision science. It is believed that in the late 1800’s the French ophthalmologist Louis Émile Javal was the first to suggest that the eyes do not move smoothly across a page when reading, he hypothesised that more complex movements take place (Wade and Tatler, 2009).
Clear vision is limited to the Fovea. The peripheral retina detects objects of interest initiating the six extraocular muscles to rotate the eye and align the fovea (via saccades and motilities) with the object of interest. Even though our eyes are never stationary, the image remains clear without motion blur, due to perceptual mechanisms that supress the motion blur.
Several different methods have been invented to investigate eye movements. These methods include devices attached to the eyes, electro-oculographic analysis, limbal tracking and more recently, optical tracking which is a non-invasive method using the point of regard method (Poole and Ball, 2006). I use the Thomson Eye Tracker, which is an optical tracking method and is more affordable than other systems commercially available. The Thomson Eye Tracker can accurately pick up the position of the eyes within 1 degree of accuracy at a rate of 60 measurements per second (Thomson, 2017).
The software is loaded on a primary computer and requires a second monitor for stimuli to be presented on for the patient to view. The Tobii eye bar is a remote sensing system (Rados et al., 2012), consisting of three infrared lights inside a slim bar mounted on the bottom edge of the second computer monitor. The eye sensing bar locates the pupil centre and Purkinje images, dark and bright pupil margins and uses these as markers to establish the position of the eyes (Rados et al., 2012). The software measures a vector between the centre of the bright pupil and corneal reflections (first and second Purkinje images) and by making use of further algorithms, the point of regard is determined (Figure 1). This two point method helps to distinguish eye movements from head movements (Poole and Ball, 2006, Rados et al., 2012).
Before measurements can be made, a simple calibration should first be done. The patient is positioned in front of, about 50 cm from the screen and level with the centre of the screen. Rados et al., 2012 suggested the results are most accurate at a working distance between 30 -70 cm, and viewing the screen straight on, because the larger the viewing angle the poorer the accuracy (Rados et al., 2012).
The calibration screen shows a box containing two circles representing the eyes. The circles should appear bright, well defined, and close to the middle of the box. The software assists the user to orient the patient at the correct distance and position from the screen (Figure 2). Once the correct position is achieved, calibration begins. The patient is instructed to view the small pink target (Figure 3) at the centre of the screen until it disappears. By doing so the central point of regard for eye movements is established. Three other targets appear in the top left and right corners and the centre at the bottom of the screen. The patient should look at each of the targets separately in any order until each target disappears (Figure 3). The calibration is now complete and alterations in posture and position are allowed.
The position of the eyes can be monitored on the taskbar at the bottom of the primary screen at all times (Figure 4).
Several stimuli can be presented to evaluate eye movements. Basic preloaded reading text can be used or the user can upload reading passages according to language preference or level of difficulty. I have created and uploaded age appropriate reading material in both Afrikaans and English. The page width and height can be adjusted prior to starting the recording.
The next stimuli function is the rate of reading section. These are random words which are easy to read but do not make sense and have no structural sentence order. This gives the advantage that the reader is unable to “fill in the gaps” providing true and accurate eye movements. Adjustable variables in this section include row spacing, font size, font, number of rows, background colour and several text options including different languages.
The background colour function is very useful as it can be used to determine the effect of coloured filters on the reading metrics, which we often use on children with dyslexia and other learning related difficulties (Singleton and Trotter, 2005). Wilkins et al., 2016, suggested that an improvement of at least 15% in reading speed should be used to justify prescribing coloured filters (Wilkins et al., 2016).
The pure saccade test consists of a target jumping from one position to another initiating a saccadic eye movement. The amplitude determines how far each jump is from the next. The angle determines whether the jump is horizontal, vertical or at any other orientation along a 360° path. The timing function allows the examiner to alter the speed at which each jump takes place. Finally the background colour, stimulus colour and size of the target can be changed.
Similarly to the pure saccade test, there is a letter saccade test which shows letters or numbers in a vertical or horizontal line. The letters or numbers can either be in sequence or random order. Variables here include the distance between the rows, line spacing, font size and number of rows. By conducting this test the gross saccadic movements are evaluated in contrast to fine saccades measured during reading (Griffin and Borsting, 2010).
A pursuit test is also available consisting of an optokinetic nystagmus grating or a moving dot across the page. The speed, width and angle can be set as desired in addition to the colour of the gratings. When using the moving dot across the page, the speed, size angle and amplitude can be set as required as well as the colour of the dot and the background colour. This is a single dot moving across the screen for the viewer to follow.
The cover test function picks up eye movements through a transparent infrared cover which can be purchased additionally from the Thomson’s website.
After the stimulus is selected a recording can be made. The duration of the recording is determined by the examiner. I allow at least a minute of reading with the rate of reading test, and several lines when reading age appropriate material. Some readers improve as they continue reading, while others show deterioration due to fatigue.
The results are presented visually using several graphs, and the software makes use of special algorithms to provide statistics about the eye movements. The graphs provide details on the horizontal eye movements, vertical eye movements, convergence as well as the exact position of the eyes.
A good understanding of all the components of eye movements used while reading is essential before trying to interpret the results. These include saccades, fixations and regressions.
The first graph seen is known as the horizontal eye position graph. This represents how the eyes move from the left of the screen to the right while reading. On the horizontal axis the time lapse is plotted in seconds. The vertical axis represents the position of the eyes in either pixels or the angle in degrees. This scale will depend on your monitor where 0 (zero) represents the left hand side of the monitor and the maximum value represents the right hand side of the monitor (Figure 5). The position of the text determines where eye position starts and can be altered as required.
This specific graph can also be oriented with the time on the vertical axis and position on the horizontal axis. This is slightly easier to interpret as the examiner can easily visualize the movement pattern. This graph is still available on the Thomson software, it is the fourth graph labelled as “right and left eye horizontal”.
When viewing a normal graph you see can basic pattern being repeated several times (Figure 5). This pattern consists of a gradual climb from low pixels (left side of the screen) to higher pixels (right side of the screen) with a large vertical line returning back to low pixels. This was the first line which was read. The pattern repeats itself for each line as the recording commences. It’s easy to conclude that the patient used in the example (fig5) successfully read 5 lines.
Some literature suggests disregarding the first and last lines of the graph as it is often less effective and not typical of reading ability (Ciuffreda and Tannen, 1995).
The four basic components of typical eye movements, while reading, can be explained with the help of several coloured lines (figure 6). The horizontal lines representing fixations are red. Fixations are brief moments where no movements took place and lasts about 220-250 ms while reading silently and 275-325 ms when reading out loud. Fixations make up about 90% of eye movements. This means that while reading the eyes are mostly stationary (Ciuffreda and Tannen, 1995). Typically good readers have less fixations, however longer and more difficult words still require more fixations. Each fixation is connected to the next by a progressive saccade.
Saccadic movements are represented by vertical lines and can be subdivided into forward saccades, backward saccades (or regressions) and return sweeps. Forward saccades are represented by the vertical green lines moving from the bottom to the top. Typically reading saccades last 10 to 30 milliseconds, no longer than 40 ms (Millodot, 1990) and range from 0.5 to 4 degrees in size, whereas gross saccadic movements are initiated by looking around a room (Griffin and Borsting, 2010) (Purves D, 2001). The length of a saccade varies from 1 to 18 characters with an average of 8 characters, this depends on the difficulty of the text and level of comprehension. Speed readers often have larger saccades but poor reading comprehension (Ciuffreda and Tannen, 1995).
Regressions are represented by the yellow lines and are similar in characteristics to forward saccades with the only difference being that they take place from the right to the left of the page. These movements are due to the eyes moving back to previous words and are often initiated by confusion or lack of comprehension. It is often noted that poor readers and children learning to read have a large amount of regressive movements. About 10-15% of saccadic movements are regressive in nature.
The purple lines represent return sweep saccades. These are large backward vertical lines with a slight oblique component. These movements mark the end of a line where a large return movements is made to direct the eyes to the beginning of the next line. The movement is initiated at about 6 characters from the end of one line to about 6 characters into the beginning of the next line. It is believed that the characters left out are interpreted by the visual system during the last and first fixations pause of each line, however it’s not uncommon to see small corrective movements (120-160 ms) either forward or backward know as an undershoot or overshoot respectively (Hofmeister et al., 1999). Return sweeps are larger movements of 12-20 degrees and last 40-54 milliseconds.
When viewing these graphs I prefer to toggle the “Model Data” button and the “Auto Scale” button. By using the Model Data button an orange line is activated over the blue lines of the eye movements which smooths out the movements into easily recognisable markers (Figure 7).
The convergence panel represents the visual posture of the two eyes while reading (figure 8). The near triad includes convergence, accommodation and pupil constriction (Emslie et al., 2007). It’s important to know the software takes into account the expected amount of convergence, which is used as a reference point at 0. Any deviation from 0 is seen as over- or under convergence. The convergence panel provides a new insight to binocular function regarding convergence. Traditionally convergence is measured where the patient views a static target, the playback function provides a real life situation to assess the effect reading has on convergence under different circumstances.
If the results are in the negative pixel scale an eso posture (convergence) is present and the patient is possibly over converging. Similarly, positive pixel scale refers to an exo posture (divergence), or under convergence. I find it useful to use the playback function with this graph. By doing this it’s easy to see how the convergence varies as the patient is reading, this can be done by viewing the figure at the right hand side of the convergence panel or viewing the prism dioptre value every millisecond of the playback.
The vertical eye position scale shows how the eyes gradually move from the top to the screen to the bottom while reading (figure 9). Large vertical jumps may indicate lines being skipped, this can easily be confirmed by making use of the playback function (figure 10).
All of the previously explained graphs can be viewed monocularly where a single line represents the movements of either the left and right eyes, or binocularly where two lines are shown in red and green representing the right and left eyes respectively (Figure 11).
Gaze metrics show the stimulus viewed together with a scan path and heat map which can be toggled as needed. The scan path shows a scribble like pattern across the words which are being read indicating where the eyes were looking at a specific time. The heat map function displays in colour the amount of time the eyes were fixating on a certain word. Cooler colours such as yellow, green and blue indicate short fixation periods typically indicating words which are easy to read or sight words. Warmer colours such as orange and red indicate longer periods of fixation typical for more difficult words. (Figure 12)
The final panel shown is the position panel. Two white circles can be seen on a black background. These two circles represent the two eyes. The position of the eyes are represented in coordinate format at the top of the panel. Horizontal pupillary distance (HPD) refers to the distance between the two pupils, Vertical pupillary distance (VPD) refers to the difference in height between the two pupils. The software automatically determines the head tilt in degrees by using the formula Head tilt = tan-1(VPD/HPD). The final numerical value on the panel labelled “Dist” indicates the working distance. I prefer the using the replay function, making it easier to get an idea of the average head posture while reading. This is useful to determine whether there is a significant head tilt present (figure 13).
Figure 14 is an example of the normal patterns expected for pure saccadic and pursuit eye movements.
When viewing the statistical analysis extensive information is given including viewing distance, number of lines read, words read per minute, fixations per minute, number of fixations, number of saccades and regressions naming only a few (figure 15).
Table 1 shows averages determined by Taylor, however the group size and ethnicity was not disclosed in the article (Taylor, 1958).
|Grade level||1st Grade||1st Grade||1st Grade||1st Grade||1st Grade||1st Grade||High School||College|
|Fixations per 100 words||240||200||170||136||118||105||83||75|
|Regressions per 100 words||55||45||37||30||26||23||15||11|
|Average span of recognition||0.42||0.50||0.59||0.73||0.75||0.95||1.21||1.33|
|Average duration of fixation (seconds)||0.33||0.30||0.26||0.24||0.24||0.24||1.23||1.23|
|Average reading rate (words per minute)||75||100||138||180||216||235||296||340|
Table 1: A summary of averages determined by Taylor.
Visual comfort is highly dependent on visual correction and binocular function. Previously, binocular function could only be evaluated by making use of stationary targets, the Eye Tracker evaluates binocular function continuously under different circumstances. Therefore the results give information on binocularity in active or dynamic situations, providing easy to understand and useful information for clinicians regardless of their level of experience.
Furthermore the results can be used to test or prove the effect of lenses or coloured filters on visual function. Currently eye tracking is the only method available to determine the effect of colour filters on reading objectively.
Finally eye tracking often sheds some light on cases where children suffer from learning difficulties where the visual system seems normal with small insignificant refractive errors, normal binocular function and fusional ability.
Current limitations include a lack of evidence relating to the repeatability of the results and a paucity of normative data relating to eye tracking, especially in South Africa. Therefore the statistics provided by the software supplier is limited in its usefulness and the practitioner must make conclusions based on the appearance of the graphs and personal experience. Population based research is needed to establish a normative data base for South Africans taking into account age, gender and ethnicity.
CIUFFREDA, K. J. & TANNEN, B. 1995. Eye Movement Basics for the Clinician, Mosby.
EMSLIE, R., CLAASSENS, A., SACHS, N. & WALTERS*, I. 2007. The near triad and associated visual problems. The South African Optometrist 66, 184-191.
GRIFFIN, J. R. & BORSTING, E. J. 2010. Binocular Anomalies: Treatments & Techniques, OEP Foundation.
HOFMEISTER, J., HELLER, D. & RADACH, R. 1999. The Return Sweep in Reading. In: BECKER, W., DEUBEL, H. & MERGNER, T. (eds.) Current Oculomotor Research: Physiological and Psychological Aspects. Boston, MA: Springer US.
MILLODOT, M. 1990. Dictionary of Optometry, Butterworths.
POOLE, A. & BALL, L. 2006. Eye tracking in human-computer interaction and usability research: Current status and future prospects.
PURVES D, A. G., FITZPATRICK D, ET AL. 2001. Types of Eye Movements and Their Functions. In: ASSOCIATES, S. M. S. (ed.) Neuroscience. 2nd edition.
RADOS, MANTIUK, A., MICHA, KOWALIK, NOWOSIELSKI, A. & BAZYLUK, B. 2012. Do-It-yourself eye tracker: low-cost pupil-based eye tracker for computer graphics applications. Proceedings of the 18th international conference on Advances in Multimedia Modeling. Klagenfurt, Austria: Springer-Verlag.
SINGLETON, C. & TROTTER, S. 2005. Visual stress in adults with and without dyslexia. Journal of Research in Reading, 28, 365-378.
TAYLOR, E. A. 1958. The Fundamental Reading Skill. Journal of Developmental Reading, 1, 21-30.
THOMSON, D. 2017. Eye tracking and its clinical application in optometry. Optician, June, 27-31.
WADE, N. J. & TATLER, B. W. 2009. Did Javal measure eye movements during reading? 2009, 2.
WILKINS, A., ALLEN, P., GILCHRIST, J. & MONGER, L. 2016. Visual stress and dyslexia for the practising optometrist.