Characterizing the Patterned Images That Precipitate Seizures and Optimizing Guidelines To Prevent Them

Abstract

The use of guidelines to prevent the broadcast of epileptogenic television program content has reduced the incidence of seizures in Britain and Japan. Epileptogenic content includes both flicker and patterns. The guidelines for flicker were developed on the basis of a model that related stimulus parameters to the proportion of patients affected. We here extend the model to pattern stimuli. A set of rules is advocated that keeps the level of risk to a consistent minimum and simplifies compliance. We propose that striped patterns that last > 0.5 s, occupy more than one fourth the area of the screen, and have bright stripes > 50 cd/m2 in luminance be restricted as regards the number of cycles admissible. The guidelines are estimated to protect at least two thirds of susceptible patients.

Full text

Wilkins
1
Characterising the patterned images that
precipitate seizures, and optimising guidelines to
prevent them
Arnold Wilkins, University of Essex, UK
John Emmett, Broadcast Projects Research, UK
Graham Harding, Aston University, UK
Keywords: photosensitive epilepsy, television, pattern-sensitivity, pattern-
sensitive epilepsy, guidelines, image safety
Running title: Guidelines for preventing seizures
Corresponding author:
Arnold Wilkins
Department of Psychology
University of Essex
Colchester CO4 3SQ UK
Tel +44 1223 812537
Fax +44 1206873590
Email arnold@essex.ac.uk

Wilkins
2
Abstract
The use of guidelines to prevent the broadcast of epileptogenic television
programme content has reduced the incidence of seizures in Britain and Japan.
Epileptogenic content includes both flicker and patterns. The guidelines for
flicker were developed on the basis of a model that related stimulus parameters
to the proportion of patients affected. We here extend the model to pattern
stimuli. A set of rules is advocated that keeps the level of risk to a consistent
minimum and simplifies compliance. We propose that striped patterns that last
longer than 0.5s, occupy more than one quarter the area of the screen and have
bright stripes more than 50cd.m-2 in luminance are restricted as regards the
number of cycles admissible. The guidelines are estimated to protect at least
two thirds of susceptible patients.
Introduction
In the general population about 1 in 6,000 is liable to photosensitive epilepsy
(PSE). Between 7 and 20 years of age the susceptibility to seizures from visual
stimulation is five times greater than in adulthood. Convulsions can be
triggered by a variety of visual stimuli, including flashing lights and steadily
illuminated patterns (1, 2).
In 1993, a broadcast advertisement (Golden Wonder Pot Noodles) precipitated
epileptic seizures in three viewers in the United Kingdom. On 17 December
1997, a broadcast children’s programme ‘Pokemon’ in Japan resulted in 685
admissions to hospital. On later investigation, 560 of these viewers were
shown to have suffered epileptic seizures, and 76 per cent of these patients had
no previous history of epilepsy (3). The use of alternating red and blue
backgrounds on successive frames was implicated in precipitating the seizures
(4).
The above incidents have led to national guidelines in the UK and Japan, now
extended internationally by the International Telecommunications Union
(ITU). The guidelines incorporate the following restrictions:-
1. Frequency. Flashes with frequency greater than 3Hz are prohibited.
2. Opposing changes in luminance. Flashes greater than or equal to
20cd. m- 2 are prohibited.
3. Area of flashes. Flashes greater in area than one quarter of the screen are
prohibited.
4. Colour. Flicker from saturated red light is prohibited.

Wilkins
3
Following the Pokemon incident and the introduction of guidelines, a decrease
in the number of patients with seizures from television has been noted in
Japan. There has been no corresponding change in the number of patients with
seizures from video games, suggesting that the broadcast guidelines have been
responsible for the decrease in seizures (5).
Over the three years that the revised ITC guidelines have been in place in
Britain only one seizure precipitated by broadcast programmes has been
reported. This related to flickering patterns. The ITC Guidance Note addresses
the risk from patterns as well as flicker and the following is an abstract of the
current guidance.
“5. A potentially harmful regular pattern contains clearly discernible stripes
when there are more than five light-dark pairs of stripes in any orientation.
The stripes may be parallel or radial, curved or straight, and may be formed
by rows of repetitive elements such as polka dots. If the stripes change
direction, oscillate, flash or reverse in contrast they are more likely to be
harmful than if they are stationary. If the patterns obviously flow smoothly
across, into, or out of the screen in one direction they are exempt from
restriction.
5.1. Potentially harmful patterns are not permitted when either of the
following two conditions apply:
i. the stripes are stationary and the pattern occupies more than 40% of the
displayed screen area; or
ii. the stripes change direction, oscillate, flash, or reverse in contrast and the
pattern occupies more than twenty five per cent of screen area; and in addition
to either of the above two conditions applying, when
iii. the screen luminance of the darker bars in the pattern is below 160 cd.m-2
and differs from the lighter bars by 20 cd.m-2 or more…
The above guidelines for patterns differ from those for flicker in that they
evolved on an ad hoc basis and were not developed from a model, as the
guidelines for flicker had been. The model has been described elsewhere (6).
The guidelines for patterns are difficult to understand and, it will be shown,
unnecessarily restrictive. The remainder of this paper is devoted to a revision
of the guidelines for pattern, a revision that extends to patterns the model
applied thus far only to flicker (6). We propose guidelines that are likely to be
less restrictive but (1) maintain or improve the theoretical level of risk (2)
simplify administration and (3) improve comprehensibility.
Starting assumptions
The probability of epileptic seizures can be estimated, without inducing
seizures, from the occurrence of certain epileptiform waves in the
electroencephalograph (EEG) recorded from scalp electrodes. The probability
of epileptiform activity has been studied as a function of a variety of stimulus

Wilkins
4
parameters under conditions in which one parameter at a time has been varied
while other parameters remain constant (2). The current ITC guidelines for
flicker and those of the ITU were based on a model that related the probability
of seizures to the obvious stimulus parameters (6). For the sake of simplicity it
was assumed that the effects of each parameter were independent. The
proportion of patients affected by flicker with particular parameter values was
estimated from the product of the proportions of patients affected when each
parameter took its particular value and the other parameters remained
unchanged. The assumption that the various parameters have effects that are
independent is almost certainly false. Nevertheless, when one parameter was
studied, the values of the other parameters were close to those values that are
maximally epileptogenic. The product of proportions is therefore likely to give
rise to an estimate of risk that is worst case.
We will now consider in turn each of the spatial parameters of patterns that
induce epileptiform EEG activity.
Effects of spatial frequency
We will begin by considering bright high-contrast patterns, i.e. those with
luminance and contrast sufficiently high to be a risk to all susceptible patients.
The proportion of patients affected by patterns of this kind is known to depend
upon the spatial frequency of the pattern as shown in Figure 1: Panel 11.
Spatial frequency refers to the number of cycles of the pattern (pairs of light-
dark stripes) subtending one degree at the eye. The function used to describe
the data is shown in the figure legend. Although continuous functions are used
for simplicity, here and elsewhere in this paper the predicted proportions are
limited to values between 0 and 1.
The shape of the function in Figure 1: Panel 1 has been shown to be similar for
stationary and vibrating patterns, and to be broadly independent of the
temporal frequency of vibration and its spatial extent (reference (2); p. 23,
Figure 2.15). Only the lower frequency limit will be considered. This is
because patterns with spatial frequency sufficiently high as to be above the
upper limit of sensitivity are likely to interfere with the screen raster,
generating low spatial frequency Moiré effects and flicker. In all the emerging
display methods, these secondary effects will be suppressed prior to
presentation.
Insert Figure 1 about here
1 Note that the function describes patterns that have a square-wave luminance profile (i.e. sharp edges).
Its shape is therefore unlike that of the contrast sensitivity function that describes the ability to detect
faint patterns (7).

Wilkins
5
Effects of area of retina stimulated.
The probability of epileptiform EEG activity in response to striped patterns of
different areas has been investigated in two studies summarised in Reference
(2): Figures 2.11 and 2.18, with a total of 19 patients. The data from these
studies have been combined in Figure 1: Panel 2. The proportion of patients at
risk from patterns increases linearly with the proportion of the visual cortex to
which the pattern projects.
Effects of pattern size
The size of a pattern was estimated from the spatial parameters of television
displays and the distance from which the displays are customarily viewed (8).
For example, a television with a 4/3 aspect ratio and 20-inch screen viewed
from a distance of 7 times screen height (2.1m) can be calculated to stimulate
an area of the visual field that projects to 25% of the visual cortex.
Assuming the relationship shown in Figure 1: Panel 2, it is possible to estimate
the proportion of patients affected by patterns that occupy different areas of
the screen, bearing in mind the simplifying assumptions outlined earlier.
Following previous work (6), we assume that 100% of patients is affected by
patterns that occupy all of the screen of a 20-inch television (diametric
measurement). The proportion affected by patterns that occupy one half, one
quarter, one eighth and one tenth of the screen area can be calculated on the
basis of Figure 1: Panel 2.
Insert Table 1 about here
Table 1 shows the proportion affected for stimuli (flashes or patterns) that are
eccentrically fixated as well as those that are centrally fixated. When the
stimulus occupies the upper or lower fields symmetrically (third column) the
proportion of patients affected is simply related to the area of the pattern.
When the stimulation is confined to one lateral visual field (fourth column)
only one cerebral hemisphere is stimulated, and it is necessary to allow for the
fact that the seizure thresholds in the two hemispheres can be quite different,
and that when one hemisphere is responsible for the seizures, only stimulation
in the contralateral visual hemifield is a risk. It was assumed that this was the
case in one third of patients (Reference (2); p.27). The data in Table 1
demonstrate that gaze position has little effect on the proportion of patients at
risk from a stimulus of a given size.
For the sake of simplicity, the effects of gaze position were ignored and central
fixation was assumed. The proportion of patients affected was calculated for
the case of a 20-inch screen with 4/3 aspect ratio and 24-inch screen with 16/9
aspect ratio, both of which have a screen height of 0.3m and a preferred
viewing distance of 7 times this height. It was assumed that 100% of patients
was affected by the full screen. Calculations were also undertaken for a 60-

Wilkins
6
inch screen with 4/3 aspect ratio, and a 73-inch screen with 16/9 aspect ratio,
at a (preferred) viewing distance of 5 times picture height. Figure 1: Panel 3
shows some of the results. The solid curve is for the 20inch television with a
4/3 aspect ratio at the preferred viewing distance: the broken curve is for a
screen with a 16/9 aspect ratio from the unusually close viewing distance of 4
times picture height (children often sit closer to the television than the
recommended “preferred” viewing distances). All the remaining functions lie
between these two curves. Note that the aspect ratio has little effect. The way
in which the proportion of patients affected decreases with the size of the
stimulus is similar for all screens and viewing distances.
Combining the effects of spatial frequency and pattern size
If we consider the number of pairs of light and dark stripes on the screen and
the percentage of the screen that the stripes occupy, we can calculate the
spatial frequency of the pattern and estimate the proportion of patients affected
from the product of the proportions shown in Figure 1: Panels 1 and 3. This
approach was used in the development of the guidelines for flicker, and, as has
been argued above, estimates the worst case risk, given that the stimulus
parameters have been investigated individually, maintaining other parameters
at or near their most epileptogenic values. The estimate obtained from the
product of the proportions is shown in Figure 1: Panel 4 for 4/3 screens with
20-inch diagonal measurement. Three important points emerge2.
First, the proportion of patients affected by any pattern that occupies less than
25% of the screen is small regardless of the number of lines it possesses, at
least for screens viewed from distances shown in Reference (8). If the screen is
viewed from closer, the spatial frequency is reduced but the area of the pattern
on the retina is increased. The two effects counteract one another.
Second, the proportion of patients affected by 5 stripe pairs is similar for
patterns that occupy the entire screen and those that occupy only a quarter of
the screen. The 5 stripe pairs filling the screen have a lower spatial frequency,
which compensates for the larger area of pattern. The same conclusions apply
when the functions shown in Figure 1: Panel 4 are calculated for 12-inch and
60-inch screens. The present limit of 5 stripe pairs would therefore seem to
provide adequate protection without specifying the pattern size, as is currently
the case. Note that 25% of the screen area corresponds to a solid angle of
approximately 0.006 steradians. This gives rise to the rule proposed at the end
of this paper.
2For those familiar with the area, there is also a fourth, abstruse consideration. In the range 1-5 stripe
pairs, an increase in the proportion affected is to be expected simply on the basis of the energy in the
pattern. Were this factor to be considered, it would increase the slope of all three functions in Figure 4
equivalently and leave unchanged the conclusions drawn from the figure.

Wilkins
7
The third important point to emerge from Figure 1: Panel 4 is that the
proportion of patients affected approximately doubles as the number of stripe
pairs increases from 5 to 8, at least as regards patterns of reasonable size.
Again similar considerations broadly apply for smaller and larger screens. If it
were desired to relax the restrictions of the current guidelines, it is possible to
do so by specifying an equivalent risk for patterns that are static and those that
reverse in phase, vibrate or flicker. Patterns that change in this way are
approximately twice as likely to evoke seizures as stationary patterns. The rule
described in the guidelines proposed at the end of this paper therefore restricts
the number of light-dark pairs of stripes to 8 in the case of static pattern and 5
in the case of patterns that reverse in phase, vibrate or flicker. This proposal
limits the proportion of patients affected to about 0.5 for all pattern types, but
is generally less restrictive.
The above considerations apply to patterns that are of high contrast and
luminance, and the proportion at risk can be reduced to more acceptable levels
by specifying the contrast and luminance admissible, as will now be shown.
Effects of luminance and contrast
Space-averaged luminance. The data from Reference (9): Figure 2.12 have
been replotted in Figure 1: Panel 5 and show a linear increase in the number of
patients affected by a striped pattern with log luminance in the range 10-200
cd.m-2.
The proportion of patients at risk from a large bright pattern with the most
epileptogenic spatial frequency is shown as a function of the Michelson
contrast of that pattern in Figure 1: Panel 6. These data were obtained using a
stationary pattern (Reference (2) p. 20) and differ from those used previously
to estimate the effects of flicker (6).
At present, the guidelines limit the difference in luminance between the bright
and dark bars (stripes) to 20cd.m-2 when the luminance of the dark bar is less
than 160cd.m-2. This does not provide an even restriction on pattern contrast.
Patterns in which the luminance of the dark bar is 161cd.m-2 and the bright bar
200cd.m-2 provide a Michelson contrast of 0.108 and a space-averaged
luminance of 181 cd.m-2. The proportion of patients likely to be affected by a
pattern of this contrast and luminance has been estimated from the functions
shown in Figure 1: Panels 5 and 6 to be 0.66. If, however, the luminance of the
dark bar is less than 160cd.m-2, for example, 159cd.m-2, then the maximum
luminance of the bright bar is restricted by the guidelines to 179cd.m-2, the
contrast to 0.34, the space averaged luminance to 169cd.m-2 and the proportion
of patients at risk can be estimated to be 0.33. Given the present guidelines,
changing the luminance of the darker bar by a negligible amount (2cd.m-2) has
changed the proportion of patients at risk by a factor of two. This change is not
simply a reflection of the abrupt limit, but is due to an unfortunate choice of
threshold luminance for the darker bar.

Wilkins
8
The functions shown in Figure 1: Panels 5 and 6 have been used to estimate
the proportion of patients at risk from patterns composed of bright and dark
stripes of a wide range of possible luminances. The results are shown in Figure
1: Panel 7. In this figure the size of the points is proportional to the proportion
of patients affected. The small points adjacent to the diagonal represent
patterns associated with a risk that is similar and low. The points are those in
which the difference in luminance between the light and dark stripes is
constant. By limiting the maximum difference in luminance between bright
and dark stripes to 20cd.m-2 we obtain a risk that varies by a factor of less than
two over a wide range of pattern luminances, as shown when the data are
replotted as in Figure 1: Panel 8. Note that it is quite unnecessary to specify a
threshold luminance level, as at present.
Unfortunately, the limit of 20cd.m-2, stringent as it is, reduces the maximum
risk only modestly. The worst case risk can be estimated from Figure 1: Panels
8 as 56%. In other words 56% of patients sensitive to vibrating patterns are at
risk. Since a substantial proportion of photosensitive patients are sensitive to
vibrating patterns, this risk may be considered unacceptably high. A more
stringent limit is required, and it is difficult to provide this by limiting still
further the difference between the luminances of the bright and dark bars. The
variation in the settings of brightness and contrast of domestic receivers is
such that it is difficult to maintain a small luminance difference, even if the
difference is transmitted appropriately.
The various functions in Figure 1: Panel 8 give similar risk when the
luminance of the brighter stripe is 50cd.m-2, regardless of the luminance of the
darker stripe. This suggests that by restricting the luminance level of the
brighter stripe to 50cd.m-2 the level of risk is held similar for all luminance
differences, and is lower than that when a luminance difference of 20cd.m-2 is
specified, as is currently the case. With a limit of 50cd.m-2 a maximum of 45%
of patients is at risk.
Moving patterns
Patterns that drift continuously in one direction are generally less
epileptogenic than those that are stationary (2). For this reason, the current
guidelines do not restrict the use of patterns that “flow smoothly across, into,
or out of the screen in one direction”. The absence of restriction has permitted
two instances in which drifting patterns may have provoked seizures. In both
cases there is uncertainty as to whether the seizures arose from a brief freeze
frame at the end that may or may not have been broadcast, and in one instance
there is the possibility that the pattern moved and changed in size in such a
way as to provide flicker. Nevertheless in both cases the patterns had a spatial
frequency close to that at which seizures are most readily evoked, and the
patterns filled the screen. The similarity between the two examples would
seem to suggest that, despite the doubt concerning the exact cause of the

Wilkins
9
seizures, it would be sensible to prohibit large patterns with the worst spatial
frequencies, even when they drift. There is difficulty in separating continuous
drift from vibratory movement, particularly on refreshed screens, and one
option might be to consider moving patterns in the same way as those that are
stationary or that move in other ways. If this option is considered to be too
restrictive it is possible to maintain a similar level of risk for drifting patterns
by restricting those that have more than 12 pairs of stripes of whatever kind.
Twelve pairs of stripes rarely occur from natural or everyday scenes, with the
exception of the occasional views of railings and Venetian blinds.
Colour and colour contrast
Patterns that vary in luminance across the boundaries of the stripes can be
epileptogenic whatever the chromaticity, although for individual patients, there
may be certain chromaticities that are more epileptogenic than others (10).
Gratings with red/green stripes that do not differ in luminance are not
epileptogenic, at least for patients with normal colour vision (2). It is not
known whether isoluminant patterns of red/cyan stripes are epileptogenic,
although red/cyan isoluminant flicker is known to be so (4). Given the
available evidence, it will be assumed for simplicity that the epileptogenic
properties of a pattern depend on the luminance contrast across its stripes
without respect to chromaticity.
Pattern duration
The response to patterns is probabilistic, and so the longer a pattern is
presented, the greater the risk. There are no studies that have formally
investigated the effects of pattern duration, but in those studies that have been
undertaken by the authors it has been rare indeed for a paroxysmal response to
pattern to have occurred in less than 0.5s. The latency is typically in the order
of 1-2s. It is proposed that the guidelines apply to those patterns that last
longer than 0.5s. Patterns of shorter duration should not repeatedly be shown
because of flicker and the possibility of cumulative risk.
A suggested rewording of the Guidance Note as it concerns patterns
It is now possible to use the above data to propose a revision to the guidelines
for pattern.
A potentially harmful regular pattern contains clearly discernible stripes
when there are more than five light-dark pairs of stripes in any
orientation.
The stripes may be parallel or radial, curved or straight, and may be
formed by rows of repetitive elements such as polka dots. If the stripes
change direction, oscillate flash or reverse in contrast they are more
likely to be harmful than if they are stationary. If the patterns obviously

Wilkins
10
flow smoothly across into or out of the screen they are less likely to be
harmful than stationary patterns. The larger the patterned area, the
greater the risk.
When the light and dark stripes of any pattern collectively occupy a total
area greater than one quarter that of the screen
and
the luminance of the lightest stripe is greater than 50 cd.m-2
and
the patterns are presented repeatedly or for longer than 0.5s
then the following restrictions shall apply:
i. If any of the stripes change direction, oscillate, flash or reverse in
contrast, the screen should show no more than five light-dark pairs
of stripes;
ii. If all the stripes are stationary the screen should show no more
than eight light-dark pairs of stripes;
iii. If all the stripes obviously move smoothly across, into, or out of
the screen, the screen should show no more than 12 light-dark
pairs of stripes.
If the pattern is such as to produce a ‘flash’ by virtue of its movement or
an interaction with the screen refresh then the restrictions on flashes
shall apply.
Protection afforded by the guidelines
The guidelines for flicker are simple to administer and appear to be working
well. The ITC guidelines for pattern also appear to be effective, although they
are less easy to interpret, are more restrictive than they need to be, and provide
a variable level of risk. The new proposals provide a risk to patients that is
theoretically consistent over a wide range of pattern types, luminances and
contrasts. The estimated risk is lower than for the current guidelines, even
though the new proposals are likely to be less restrictive. Nevertheless, the
protection is poorer than it might otherwise be because of the threshold size
and luminance below which the guidelines are not held to apply. If a pattern
had maximally epileptogenic characteristics but an area slightly less than one
quarter that of the screen then about one third of patients would be at risk,
according to Figure 1: Panel 3. If a larger pattern passed the guidelines
because its bright bars had a luminance slightly less than the limit of 50cd.m-2,
a similar proportion of patients would be at risk, according to Figure 1: Panel

Wilkins
11
8. The protection afforded by the guidelines appears to be limited to about two
thirds of patients. It is quite possible to reduce this risk, but only by
introducing further restriction on patterns. When the stimuli are such that the
guidelines are applied, the protection is better: the risk is estimated to be
0.3*0.5=0.15; that is, 85% of patients who are sensitive to a pattern with the
most epileptogenic spatial properties filling an entire screen would be
protected.
The guidelines proposed above have been formulated for conventional video
screens viewed from conventional distances. If a more general formulation is
required, the specification of 25% of screen area could replaced by stipulating
a solid angle of 0.006 steradians at the minimum expected viewing distance.
Application of the above guidelines by automata
Devices have been built that automatically recognise video material likely to
cause seizures (5, 11, 12). The revised guidelines are insufficient for such
purposes as formulated above because they fail to define a stripe. If stripes are
defined in terms of the luminance difference across their boundaries, the
question arises as to the threshold luminance difference sufficient to
characterise a stripe. From the class of functions shown in Figure 1: Panel 8, it
can be shown that a luminance difference of less than 3cd.m-2 is likely to affect
fewer than 15% of patients. The small difference in luminance is attributable
to the low contrast at which patterns can provoke seizures: stripes that differ
by more than 3cd.m-2 are potentially a problem.
The automatic recognition of epileptogenic material is not a straightforward
matter and is beyond the scope of this paper. If Fourier analysis of the image is
used to recognise a striped pattern, it is worth bearing in mind that the addition
of energy in orthogonal orientations acts to decrease rather than increase the
epileptogenic potential of a pattern, (see ref (2), p. 14).
Correction of images that fail the above guidelines
The above guidelines are simple to understand. They also provide for a simple
remedy in the case of video material that fails the guidelines. It is necessary
only to reduce the luminance of patterns so that the luminance of the brightest
stripe is less than 50cd.m-2. Note that this is far simpler than reducing the
contrast of a pattern, as the present ITC guidelines might require. There are
various ways in which the luminance or its equivalence in voltage or pixel
values can be measured.
The steps necessary to evaluate material reduce to the following:-

Wilkins
12
Look at the screen
Are there more than five stripes?
If so, do they last longer than 0.5s?
If so, does the brightness exceed the stated limit?
If so, categorise the motion of the pattern
Are the guidelines contravened?
If so, reduce brightness.
Emerging display technologies
In the home, flat-panel displays using either plasma or liquid crystal thin film
transistor (TFT) technologies are likely to become dominant over the
conventional cathode ray tube (CRT) displays within a decade, whilst in public
installations, the large light emitting diode (LED) matrix displays are gaining a
presence more rapidly still, both indoors and outdoors.
The growing “home cinema” market employing pre-recorded media
exemplified by the DVD family is likely to provide the first mass high
definition (HD) sources, although digital television transmission standards also
enable higher definition services to be offered to the public. All these services
will be viewed on displays that are intended to subtend roughly twice the
retinal area of standard definition displays. Although the plasma and TFT
displays are unlikely to exceed the CRT in terms of brightness, LED based
“digital billboards” normally have a surface luminance of 1200 cd.m-2, and
with diligent cooling they can reach 3000 cd.m-2 quite easily. These luminosity
levels are designed to enable daylight readability, although such levels are
often employed in dim indoor situations for added (advertising) impact.
Display resolution, is normally lower than that of broadcast television, often
by a factor of two, increasing the potential for harmful patterns.
None of the changes in display in technology reduce the need for broadcast
guidelines; rather the increases in screen subtense and brightness make
guidelines more necessary. If the current recommendations are to apply not
only to the current generation of display screens but also to future generations
it may be necessary to be more restrictive as screen size and brightness
increase. For now, the recommendations can be taken to apply to screens that
subtend about 10 degrees at the eye.
Acknowledgements.
The authors thank Paul Gardiner and the Independent Television Commission
for financial support and the American Epilepsy Association for prompting
this submission.

Wilkins
13
References
1. Harding GFA, Jeavons PM. Photosensitive epilepsy. London: Mac Keith
Press; 1994.
2. Wilkins AJ. Visual Stress. Oxford: Oxford University Press.; 1995.
3. Harding GFA. TV can be bad for your health. Nature Medicine
1998;4:265-7.
4. Shirakawa S, Funatsuka M, Osawa M, Fujita M, Oguni H. A study of the
effect of color photostimulation from a cathode-ray tube (CRT) display
on photosensitive patients: the effect of alternating red-cyan flicker
stimulation. Epilepsia 2001;42(7):922-9.
5. Takahashi Y, Fujiwara T. Effectiveness of broadcasting guidelines for
photosensitive seizures prevention. Neurology 2004;62(6):990-3.
6. Binnie CD, Emmett J, Gardiner P, Harding GFA, Harrison D, Wilkins
AJ. Characterising the flashing television images that precipitate seizures.
SMPTE Journal 2002;July/August:323-9.
7. Campbell FW, Robson JG. Application of Fourier analysis for the
visibility of gratings. Journal of Physiology (London) 1968;197:551-66.
8. Methodology for the subjective assessment of the quality of television
pictures. In: Recommendation ITU-R BT500-11; 2002.
9. Wilkins AJ, Binnie CD, Darby CE. Visually-induced seizures. Progress in
Neurobiology 1980;15:86-117.
10. Wilkins AJ, Baker A, Amin D, et al. Treatment of photosensitive epilepsy
using coloured filters. Seizure 1999;8:444-9.
11. Harding GFA, Takahashi T. Regulations: What next? Epilepsia
2004;45((Suppl 1)):46-8.
12. Nomura M. A comfortable brain interface to video displays. Neural Netw
1999;12:347-54.
13. Wilkins AJ, Darby CE, Binnie CD. Neurophysiological aspects of
pattern-sensitive epilepsy. Brain 1979;102:1-25.
14. Drasdo N. The neural representation of visual space. Nature (London)
1977;266:554-6.

Wilkins
14
Tables and Figures
Table 1. The proportion of patients affected by stimuli (flashes or patterns) of
different size, assuming all are affected by stimuli that occupy the entire screen
of a 20-inch television. Data are shown for central and eccentric gaze.
Area of
screen
with
stimulus
Central
gaze Eccentric
Upper/
lower
margin
Eccentric
Lateral
margin
Weighted
average
Full 1.00 1.00
Half .63 .50 .80 .64
Quarter .32 .31 .50 .36
Eighth .09 .16 .25 .15
Tenth .03 .12 .19 .09

Wilkins
15
Figure legend
Figure 1: Panel 1. The proportion of patients with pattern-sensitive epilepsy at
risk from a large pattern of bright high-contrast stripes, shown as a function of the
spatial frequency of the pattern expressed in cycles per degree subtended at the
eye. The stripes had square-wave luminance profile. The data are from reference
(13) p. 4). The function has the following equation for 0<y<1, y=0.382(log(x))3-
2.020(log(x))2 + 1.285log(x) + 0.839. Panel 2. The proportion of a sample of 19
patients affected by a pattern of stripes, shown as a function of the area of visual
cortex to which the pattern projected, estimated from the formula proposed by
Drasdo (14). Data are from references (13) and (9), summarised in reference (2);
Figures 2.11 and 2.18). Note that no patient is affected when less than 8% of the
cortex is stimulated. The equation of the line for 0<y<1 is y=0.021x-0.184.
Panel 3. The proportion of patients with pattern-sensitive epilepsy at risk from a
pattern of bright high-contrast stripes, shown as a function of the overall size of
the pattern, expressed in terms of the proportion of the screen that it occupies.
Solid lines is for a 20-inch television with a 4/3 aspect ratio at the preferred
viewing distance of 7 times picture height. The dotted line is for a screen with an
aspect ratio of 16/9 viewed from 4 times picture height. Curves for other screens
at preferred viewing distance lie between the two black curves. The curves are
represented by the expression, y= -0.708x2 + 1.792x – 0.083, for 0<y<1, shown
by the white curve. Panel 4. The proportion of patients with pattern-sensitive
epilepsy at risk from a pattern of bright high-contrast stripes, shown as a function
of the number of bright-dark pairs of stripes constituting the pattern. The curves
show the effects of three different sizes of pattern expressed as a percentage of the
screen that the pattern occupies, see labels. The data are for a screen with 20-inch
diagonal measurement. Panel 5. The proportion of a sample of 9 patients affected
by a pattern of high-contrast stripes shown as a function of the luminance of the
pattern. The data are from reference (9): Figure 2.12. The proportion of patients at
risk from a large high-contrast pattern with the most epileptogenic spatial
frequency is shown as a function of the space-averaged luminance of the pattern.
The equation of the function for 0<y<1 is y=0.336ln(x) –0.745. Panel 6. The
proportion of patients with pattern-sensitive epilepsy at risk from a large pattern of
bright stripes, shown as a function of the Michelson contrast of the pattern. The
data are from reference (13); p.106, Figure 2.1. Panel 7. The proportion of
patients with pattern-sensitive epilepsy at risk from a large pattern of stripes with
luminance as shown on abscissa and ordinate. The size of the points is directly
proportional to the proportion of patients at risk, and the isolated white point in the
lower right hand side of the graph shows a proportion of 1.0 for comparison.
Panel 8. The proportion of patients with pattern-sensitive epilepsy at risk from a
large pattern of bright high-contrast stripes, shown as a function of the luminance
of the brighter bar. The curves show the effects of various differences in
luminance between the light and dark stripes (shown beside each curve and
expressed in cd.m-2).

Wilkins
16