)

=

exp
⁢

〈

-

[

log
⁡

(
f
)

-

log
⁡

(

f
0

)

]

2

2
×

log
⁡

(
b
)

2

〉

(
1
)

In equation (1), f₀may be the peak spatial frequency and b may control the bandwidth. Conversion from the Fourier domain shown in FIG. 8B to the spatial domain shown in FIG. 8A may give a log-Gabor dot element that can then be used to construct visual stimuli.

It should be noted from Equation (1) that the amplitude must be infinitesimally small when f=0. For practical purposes, the amplitude may be considered zero. This is important, as it means that for the log-Gabor dot, there should never be an overall change in screen luminance as a result of adding a dot to the display. The areas of increased luminance will be balanced out by areas of decreased luminance. This may have particular advantages for visual displays. It means that no matter the arrangement of spatially bandpass dots on the screen, the overall luminance will not be affected. For example, the shifting of the dot positions to create the disparity in the images may cause local increases or decreases in the density of the dots. If the dots changed the local luminance of the stimulus then these density changes would also cause there to be regions of the stimulus that were significantly brighter or darker than the rest. That may provide a non-stereo cue that would allow the patient to locate regions in the display where the disparity is being manipulated.

In addition to the log-Gabor dots discussed above, other spatially filtered dots may be used to construct visual stimuli. In fact, the log-Gabor dots may be only one of a vast number of options for spatially filtered dots. There may be particular interest in developing the visual stimuli used herein with control exerted over their amplitude spectrum. This may allow developers to design visual stimuli with desired characteristics. For example, high spatial frequency information, which may not be useful to patients with poor acuity, may be removed. It may be possible to develop such stimuli in the Fourier domain or in the spatial domain.

Another example of spatially filtered dots may be Difference of Gaussians (DoG) dots. FIGS. 9A-9C illustrate a DoG dot, in the spatial domain, the Fourier domain, and as an amplitude spectrum, taken as a radial slice from the center to the edge of FIG. 9B, respectively. As can be seen in FIG. 9B, the DoG element may include a center portion and a surround portion.

The DoG is a popular function which has a history of use for performing spatial filtering on images. As with the log-Gabors presented above, DoG elements may be used to directly replace dots in visual stimuli. Alternatively, stimuli can be generated with dots, which may approximate delta functions, and then those dots can be filtered with a DoG function.

A DoG element may be created by subtracting one Gaussian from another Gaussian. The two Gaussians may have the same mean, but a different standard deviation. FIG. 10 illustrates a Difference of Gaussian (solid line) created by subtracting a surround Gaussian (dashed line) from a center Gaussian (grey line).

The radial distance of each pixel from the center of the dot to be generated may be given by Equation (2) below. The center of the dot may be set at r_x,y=0.

r_x,y=√{square root over (x²−y²)} (2)

The center Gaussian may be defined by Equation (3) and the surround Gaussian may be defined by Equation (4), below.

$\begin{matrix} g_{centre} (r_{x, y}, σ_{centre}) = \exp (- \frac{r_{x, y}^{2}}{2 \times σ_{centre}^{2}}) & (3) \\ g_{surround} (r_{x, y}, σ_{surround}) = \exp (- \frac{r_{x, y}^{2}}{2 \times σ_{surround}^{2}}) & (4) \end{matrix}$

σ_surroundmay be greater than δ_centerand the two sigmas may control the shape of the DoG. The DoG may be created by taking the difference of the center Gaussian and the surround Gaussian using Equation (5).

d(x,y)=g_centre(r_x,y,α_centre)−a×g_surround(r_x,y,σ_surround). (5)

In Equation (5), a may control the relative contribution of the surround Gaussian. The value of a may be set to result in a DoG which has no global change in luminance by choosing a value which results in minimal amplitude at f=0. This is illustrated in FIGS. 9A-9C. Accordingly, it may be possible to achieve a similar amplitude spectrum (see FIG. 9C) with DoG dots, as with log-Gabor dots.

Although the previous examples show only isolated spatially filtered dots in a grid arrangement, it should be understood that these dots can also be arranged into other forms. They may then be used to create spatially bandpass shapes as shown in FIGS. 13A-13B. The spatially bandpass shapes may be considered spatially filtered dot elements. It is mathematically equivalent to either i) arrange the spatially filtered dots to form a specific shape (FIG. 13B), or ii) begin with a sharp line image of the desired shape and then convolve that image with a single spatially filtered dot (FIG. 13A). A further method to generate spatially-bandpass shapes is to mathematically define shape outlines and then render those outlines such that their cross-section has a specific form. Two possible forms for those cross-sections would make use of the log-Gabor cross section shown in FIG. 8B or the DOG cross-section shown in FIG. 9B. Other functions described below may also be used.

Stimulus elements having spatially bandpass shapes in accordance with the present disclosure may created by adapting the fourth order derivative of the Gaussian function. The spatial frequency of such functions may be set by a, as given by Equation (6) below.

$\begin{matrix} σ = \frac{\sqrt{2}}{π \times f_{0}} & (6) \end{matrix}$

In Equation (6), f₀may control the peak spatial frequency of the amplitude spectrum. D_sq, the squared distance across the cross-section, relative to its center xo, for each position x may then be defined by Equation (7).

$\begin{matrix} D_{sq} (x, σ) = \frac{{(x - x_{0})}^{2}}{σ^{2}} & (7) \end{matrix}$

D_sq, the squared distance may be used to define the luminance at each position in the cross-section C. C may be defined by Equation (8).

$\begin{matrix} C (x, σ) = [4 \times {D_{sq} (x, σ)}^{2} - 12 \times D_{sq} (x, σ) + 3] \times \frac{\exp [- D_{sq} (x, σ)]}{3} & (8) \end{matrix}$

FIGS. 11A-11D illustrate an example of a spatially filtered dot element having a ring shape made using a cross-section defined by Equation 8. FIG. 11A shows the ring dot in the spatial domain. This is what would be visible in a stimulus made using such ring dots. The ring dot has the appearance of two narrow dark rings sandwiching a wider bright ring. This can also be seen in FIG. 11C, which gives a horizontal cross-section taken through the centre of FIG. 11A. The positive luminance deviation of the central bright ring is balanced out by the negative deviation of the adjacent dark rings. This results in there being no overall offset in luminance when this stimulus is added to the display. The transitions between the bright and dark regions appear smooth, as there are no higher spatial frequencies present to give the impression of sharp edges.

FIG. 11B shows the Fourier transform of FIG. 11A. Similar to the spatially filtered dots, the stimulus energy is located in a circular band around the origin. The amplitude spectrum is presented in FIG. 11D. It has a similar bandpass form to that shown previously for the log-Gabor and Difference of Gaussians stimuli. Accordingly, the ring dot may provide the same advantages as the other dots described above.

In some embodiments, the ring dot may be generated by taking the coordinate of each pixel relative to the center of the ring dot. In some embodiments, the coordinates of each pixel may be given in Cartesian coordinates, and the center of the ring dot may be (0,0). Using these coordinates, a radial distance from the center of the dot may be calculated for each pixel using equation (9) and the luminance at each radial position may be related to the luminance at the corresponding Cartesian position by equation (10).

r_x,y=√{square root over (x²+y²)}, (9)

R(x,y,r₀,σ)=C(r_x,y−r₀,σ) (10)

The method described above for creating ring dots may be extended to generate arbitrarily shaped dot elements. For example, FIGS. 12A-12C illustrate a ring dot, an annular square element, and an annular cross element. The outlines of the annular dots may or may not be defined mathematically, for example as a Fourier-like series of radial frequency pattern components.

The annular square shown in FIG. 12B may be created in the same way as the ring function outlined above using equations (9) and (10). The radius equation may be modified slightly from what is given in equation (9) and equation (11) below may be used instead.

$\begin{matrix} r_{x, y}^{'} = \sqrt[4]{x^{1} - y^{1}} . & (11) \end{matrix}$

The annular cross shown in FIG. 12C is an example of what can be achieved with radial frequency pattern components. Because its radius at a given pixel depends on that pixel's angular position around the center of the image, the angle must be computed from the Cartesian x and y coordinates as shown below in equation (12).

θ_x,y=a tan 2(y,x) (12)

Thus, the shape is defined, similar to the ring in equation (10), by its radius. However, the radius is determined for each value of θ_x,yin the image. The annular cross is generated by equation (13) below.

R=1.4+0.6×sin(3×θ_x,y)+0.2×sin(9×θ_x,y)−0.12×sin(15×θ_x,y) (13)

One skilled in the art will recognize that it is possible to generate a variety of shapes from sinusoidal components following these examples.

Even for shapes that cannot be represented using a sum of annular components, as demonstrated in Equation (13), it is possible to produce a spatially filtered element. FIG. 13A illustrates how such a dot may be generated. The desired outline may be produced; in this example, the outline is a triangle. The outline may serve as a source image. The outline may then be filtered with a spatially-bandpass filter kernel. The log-Gabor or DoG functions described above may be used as the filter kernel. The convolution of the source image with the filter kernel may result in a spatially-bandpass annular shape having the given outline. One skilled in the art will recognize that a wide variety of dots may be produced in this manner. In some embodiments, dots may be designed for particular functions.

Any of the types of dot elements discussed above may be used to produce a visual stimulus 200 as shown, for example, in FIG. 2. Visual stimuli in accordance with the present disclosure will be discussed in more detail below.

In some embodiments, a visual stimulus 200 may be created by first rendering an image filled with placeholder dots. The placeholder dots may approximate delta functions. The image may then be convolved with one of the functions described above, for example with a log-Gabor function. The convolution may replace every placeholder dot with a spatially filtered dot element 202 at its location. This method may be particularly efficient. Alternatively, a plurality of spatially filtered dot elements 202 may first be rendered and may then be placed in random locations. These steps may produce a visual stimulus 200 with randomly placed spatially filtered dot elements 202.

It may then be necessary to create the impression of disparity between what is presented to the left eye and what is presented to the right eye when the visual stimulus 200 is viewed. The visual stimulus 200 may be modified using the following steps to produce such an impression.

To give the impression of disparity, the final x coordinates of the dots 202 in the grid 204 may be further adjusted according to the desired disparity (or depth) of the dots 202 relative to the surface of the screen on which the stimuli is presented, also referred to herein as the fixation plane. Shifting left and right eye dots inwards (i.e. moving the left eye dot to the right and the right eye dot to the left) gives the impression of crossed disparity (i.e. of a dot being closer than the fixation plane). Shifting the dots 202 outwards gives the impression of uncrossed disparity (i.e. of a dot being further away than the fixation plane). FIG. 5A illustrates the design of four potential target regions in stimulus and shows that the disparities of the dots 202 may be manipulated within the bounds of a wedge shape. This gives the impression of a floating wedge. A task assigned to a patient viewing the visual stimulus may comprise identifying where in the stimulus a wedge appears at a different depth than the rest of the stimulus. The task may be made easier by making the disparity larger and so the difference in depths greater, and harder by making the difference in depths smaller.

One skilled in the art will recognize that FIG. 5A merely illustrates an exemplary embodiment of a control pattern for dot disparity. The dots 202 in the grid 204 may be shifted to produce any shape known in the art, as discussed above with respect to the wedges 502 shown in FIG. 5A.

FIGS. 5B-5D illustrate other embodiments of target regions. FIG. 5B illustrates four symbols: a pentagon, a star, a triangle, and a square. In some embodiments a target region may be given the shape of one of these symbols or any other symbol known in the art. A patient may be asked to identify the symbol, provide a location of the symbol, or to determine whether the symbol appears to be floating in front of or behind the rest of the visual stimulus. FIG. 5C illustrates the letter E in four different orientations. In some embodiments the target region may be given the shape of a letter or other symbol whose orientation can be clearly recognized. A patient may be asked to identify a location and/or orientation of the symbol or to determine whether the symbol appears to be floating in front of or behind the rest of the visual stimulus. FIG. 5D illustrates four different letters. In some embodiments, the target region may be given the shape of a letter or number and a patient may be asked to identify and/or locate the letter or number or to determine whether it appears to be floating in front of or behind the rest of the visual stimulus.

FIGS. 15A-15D illustrate an exemplary visual display in accordance with the present disclosure. The display shown in these figures may be similar to those shown in FIG. 2. Specifically, the visual display may comprise a plurality of dot elements, first arranged on a grid, then offset by a random amount, and finally having a subset offset so as to produce disparity between what is seen by each eye. FIG. 15C illustrates the visual display as it would be shown to a patient on a display device. A patient would view the visual display wearing glasses having a red lens over the left eye and a green or blue lens over the right eye. In some embodiments, different glasses, for example, glasses with differently colored lenses, may be used. FIGS. 15A and 15B illustrate, respectively, what the left and right eye would see individually. FIG. 15D illustrates what would be seen by a patient with stereoscopic vision when they view FIG. 15C with the glasses. Namely, the patient would see a wedge shaped portion of the visual display offset in front of the frame of the display device.

It should be noted from FIGS. 15A-15D that it would be difficult to cheat on such a test. The wedge shaped region which extends from the page when viewed by a patient with stereoscopic vision wearing colored glasses cannot be identified through any other means when viewing FIGS. 15A-15C. Rather, all regions of the visual stimulus appear to contain randomly placed dot elements and differences cannot readily be detected between the left eye stimulus shown in FIG. 15A and the right eye stimulus shown in FIG. 15B.

Some previous stereovision tests presented on digital displays have limited themselves to only presenting stimuli at disparities that can be achieved by shifting the stimulus features by an integer number of pixels (e.g. Hess, To, Zhou, Wang & Cooperstock, 2015. Stereo vision: the haves and have-nots. i-Perception 6, 1-5; Rodriguez-Vallejo, Llorens-Quintana, Montagud, Furlan & Monsoriu, 2016. Fast and reliable stereopsis measurement at multiple distances with iPad. arXiv 1609.06669). This severely limits the range and number of disparities that can be tested. It also means that the disparities that can be tested depend on the properties of the device being used, which is disadvantageous for creating a test that can be run on multiple devices. To produce more precise disparities, it may be desirable to shift the dots 202 by an amount that is not an integer number of pixels. For example, very small disparities may require the dots 202 to be shifted by less than one pixel. To achieve this, sub-pixel interpolation may be used. However, for the hard-edged stimuli used in previous tests this has the side-effect of blurring the dots that are presented in depth. This non-stereo cue may allow the patient to cheat. The design proposed herein avoids this problem in two ways. Firstly, the intrinsic “fuzzy” appearance of our spatially filtered dots 202 makes any slight blur introduced by the sub-pixel interpolation much harder to notice. Secondly, depth may be incorporated into the proposed stimuli by both moving the target wedge 502 forward (crossed disparity) and the background backward (uncrossed disparity) by the same amount. This means that the same degree of interpolation is applied to all dots 202 in the stimulus.

When the stimulus is presented to the subject (at step 104 of FIG. 1), the image that passes through the subject's eye is filtered by the optical transfer function of the subject's eye before the image reaches the subject's retina. This has the effect of blurring any input. The sub-pixel interpolation takes advantage of this, by presenting stimuli that appear (after the optical blur) to be placed between two pixels on the screen. Provided the optical blur is relatively broad compared to the pixel-to-pixel spacing on the display, this allows the presentation of disparities below the screen's resolution limit. It has been found that the interpolation provides a better approximation to an image presented in the desired disparity (that the sub-pixel interpolation is emulating) when applied to the proposed log-Gabor dots 202 than when applied to hard-edged pixel dots.

One skilled in the art will recognize that the types of spatially filtered dot elements described above which have blurred edges may be particularly advantageous for achieving sub-pixel interpolation. The sub-pixel interpolation may allow more minute shifts of the dots 202 to be made, than could be made using hard-edged visual stimuli. In particular, the centers of the spatially filtered dot elements may be able to be moved in sub-pixel increments. Such advantages may be achieved using any type of spatially filtered dot element. In contrast, hard-edged elements may only be movable in increments of pixels, and may thereby be limited by the resolution of the display on which they are shown.

Referring back to FIG. 1, step 104 of presenting the visual stimulus to the subject will now be further described. The visual stimulus may be rendered to any suitable dichoptic viewing arrangement that allows for different images to be shown to the two eyes of the subject. In particular, the visual stimulus may be presented using a display system provided to the subject. For instance, the visual stimulus may be rendered on an electronic device (e.g., on the screen of a desktop computer, portable computer, laptop computer, tablet computer, on a head-mounted display, on a lenticular display, or the like). In some embodiments, the visual stimulus may be viewed using specialized glasses. The electronic device illustratively controls presentation of the stimuli, and more particularly controls the disparity presented to the subject using an adaptive procedure to locate the stereo threshold of the subject. In particular, the device presents the stimulus composed of spatially filtered dot elements in order to reduce the effect of acuity on the obtained stereo threshold. The device then analyses the data obtained from the subject after presentation of the stimulus (step 106 of FIG. 1), using a maximum-likelihood based fitting procedure that also obtains a measure of confidence in the thresholds (the error measurement) that are calculated from the subject's data.

In creating the impression of depth from stereoscopic disparity, different images are shown to the two eyes of the subject at step 104, using a suitable 3D display device and/or system. In one embodiment, this may be achieved by presenting the stimuli at step 104 on a screen associated with a computer, as discussed above. For example, a liquid crystal display (LCD) computer monitor with a 120 Hz refresh rate may be used. The screen may be synchronised through any suitable system, such as the NVIDIA 3D Vision® 2 system, to specialized glasses used by the subject to view the visual stimulus as presented. In one embodiment, wireless shutter glasses are used. It should however be understood that other suitable glasses (e.g., ferroelectric liquid crystal glasses) may apply and that these glasses can be connected to the screen wirelessly or not. By alternating between the left and right eye images on the screen in time with “opening” and “closing” the left and right eyes of the glasses, it is possible to stimulate the subject's eyes separately with stimuli shown on a same screen. Such a stereo vision test may be written using any suitable programming language (such as Matlab/Octave, Python or the like) and the presentation of stimuli on the screen may be controlled using any suitable toolbox (such as Psychtoolbox for Matlab/Octave) or by bespoke software.

In another embodiment, the stereo vision test may be presented on a handheld or portable computing device (e.g., a tablet computer) and the stimulus presentation achieved using red/green anaglyph. The stimuli may be generated in left and right eye versions by removing information from either the red or green colour channels. The two resulting images can then be overlaid using suitable computer software. Viewing with red/green filter 3D glasses worn by the subject then allows for the separate images to be viewed by the two eyes of the subject (with some crosstalk). It should be understood that other embodiments may apply. For example, the stereo vision test may be presented on a lenticular, polarised, or dichroic tablet or using a dichoric projection system.

In yet another embodiment, the stereo vision test may be presented on a head-mounted display. For example, the Oculus Rift virtual reality headset may be used to present the stimuli to the two eyes of the subject at step 104. In this embodiment, specialized glasses would not be needed to view the visual stimulus as presented.

It should be understood that the stereo vision test described herein may be developed using a static stimulus (as described above) or a dynamic stimulus. In the case of a dynamic stimulus, the dots (reference 202 in FIG. 2) would be presented (e.g., on the device's screen) for a random lifespan (e.g., between 100 and 300 milliseconds) before disappearing. Each time a dot 202 disappears, it would be replaced by a new dot 202 appearing at another location (with an appropriate disparity for its location in the stimulus). This dynamic stimulus may include more information than the static stimulus, and may accordingly help subjects perform a given stereo task. Additionally, the dynamic stimulus may reduce the ability of the subject to scrutinize the stimuli for any non-stereo cues to aid identification of the target location. This will further ensure that the measured thresholds only reflect the subject's sensitivity to stereoscopic disparity.

In one embodiment, the stereo task to be performed by the subject is introduced with an additional non-stereo cue to enable subjects with poor stereo to learn how the task works before the test begins. Additional trials with a non-stereo cue may be randomly introduced during testing, which, for instance, motivates children with poor stereo vision to continue attempting to perform the task. There may also be a built-in adjustment for subjects whose eyes are misaligned to bring the stimuli into registration.

The stereo vision test described herein may consist of a sequence of trials where the subject is presented with a stimulus containing a target at one of four locations. In each trial, the subject may be prompted to select which location contains the target, defined by the modulation of the disparity of the dots (reference 202 in FIG. 2). On trials where the subject is not sure of the location of the target, the subject may be instructed to guess. In the desktop computer version of the test (described herein above), the subject may respond by pressing a button on a keypad. For the tablet computer version of the test (described herein above), the subject may be able to tap on the screen at the target location. It should be understood that the test may also be extended to receiving verbal responses (e.g., “top”, “bottom”, “left”, and “right”) from the subject.

After responding, the subject is presented with the next stimulus. The disparity of the stimuli may be controlled by a pair of staircase routines that will respond to subject successes by making the stereo task more difficult (smaller disparity) and respond to failures by making the stereo task easier (larger disparities). After each trial, the disparity of the stimulus is recorded, along with whether the target location was successfully identified. This is later used to calculate the disparity at which the subject could reliably identify the target. An online rough estimate of the stereo threshold calculated from staircase reversals (disparities at which the subject transitioned from making incorrect to correct responses, or vice-versa) may also be obtained.

For the first few trials of each testing session, the stimuli may feature an additional cue to the location of the target. The contrast amplitude of the dots 202 in the non-target regions is reduced, so that even without any sensitivity to stereo the target is still apparent. This may be done to help train the subject on how to perform the task that will be required of them in the experiment. As discussed herein above, the trials may also be randomly introduced with additional cues during the testing to keep stereo-blind subjects motivated. The data from these trials may be discarded before any analysis of stereo function is performed.

Each staircase tracks the number of trials that the subject has responded to, as well as the number of reversals (where the task switched from getting easier to getting harder or vice versa). Upon reaching either the maximum number of trials or the maximum number of reversals, the staircase terminates. When both staircases terminate, the main part of the data collection is complete.

It may also be possible to perform a further test of the subject's ability to discriminate between crossed and uncrossed disparity at some factor above threshold (e.g. three times threshold). In this case, wedges in disparity are present at all four target locations (as shown in FIG. 5), with the background dots at a disparity of zero. Three of the wedges are in one direction of disparity (crossed or uncrossed) with a remaining odd-one-out wedge in the other direction. The subject then identifies the odd-one-out. This will test until the subject has reached either a certain number of responses (e.g., 12) or a lesser number of correct responses (e.g., 6). The subject's behaviour can then be converted to a measure of performance on this crossed/uncrossed disparity discrimination task.

The data collected during the experiment (at step 106 of FIG. 1) may then be presented in any suitable manner. FIGS. 6A-6C illustrate one method of presenting and analyzing the collected data. In one embodiment, the collected data forms a table with three columns: stimulus disparities (in arc seconds) in a first column, number of trials tested at each disparity in a second column, and number of correct responses at each disparity in a third column. The data may then be fit with a suitable psychometric function (such as a logistic function) 602, as shown in FIG. 6A, using any suitable fitting routine as discussed herein above. The psychometric function 602 fitting outputs an estimate of the subject's disparity threshold. Bootstrapping may also be used to calculate the standard error and confidence intervals of the threshold estimate.

Furthermore, that threshold and error may be used to assess when ordinary analysis has failed and a measure from a more robust secondary analysis should be used instead. This secondary analysis may perform a calculation on the percent-correct scores achieved by the patient at each disparity to give a measure of performance that forms a composite scale with the range of thresholds that can be obtained from psychometric function fitting. With improvement in performance, a patient can transition from that part of the scale to the part where their performance can be characterized by a psychometric function.

In other words, it should be understood that, for some subjects, it may not be possible to successfully constrain the fit of a psychometric function as in 602. In this case, the error associated with the estimate of threshold will be large. However, even in this case, it may be useful to be able to provide some measure of the subject's performance. As discussed above, a robust analysis may be required in such a case.

In some embodiments, a suitably robust measure may be derived by calculating the area between the subject's percent-correct data (with an additional rightmost point extending the polygon horizontally to some arbitrarily high x-value) and the line that represents chance performance (also referred to herein as the 25% “guessing” line). This area can be calculated using trapezoidal integration. FIG. 6B illustrates the robust area-based measurement method. In FIG. 6B, the x-axis may be log 2 transformed before calculating the areas. To prevent outlier points having an undue influence on the area calculation, nonparametric bootstrapping can be performed from the raw trial-by-trial data and the median area calculated from those resampled bootstrap data sets may be used. The calculated area was found to be highly correlated with the fitted psychometric function threshold. The threshold and error from the psychometric function fit may be used to assess when ordinary analysis (such as the logistic function fit) has failed and a measure from a more robust analysis should be used instead. One skilled in the art will recognize that the trapezoidal integration method illustrated by FIG. 6B is merely an exemplary extrapolation method. Other methods may be used to analyze the data to which a function cannot be fit without departing from the scope of the present disclosure.

Furthermore, this secondary measure may be utilised to create a composite performance scale (or composite performance axis). That scale may have a first part corresponding to the thresholds estimated from the psychometric function as in 602 (up to a given maximum) and a second part where the scale extends to include values from a more robust analysis. FIG. 6C illustrates this composite scale, in which a first portion gives the levels of performance that can be measured through fitting a psychometric function. The second portion of the scale extrapolates the threshold values, with the location of the subject on that part of the scale determined using the bounded area method. With improvement in performance, a patient can transition from that part of the scale to the part where their performance can be characterized by a psychometric function.

In one embodiment, the overall outcome of the analysis of the subject data (as performed at step 106 of FIG. 1 and as discussed herein above) may therefore consist of the following: (1) a graph of the psychometric function as in 602, allowing the person administering the test to get a view of the subject's performance; (2) an estimate of the threshold for detecting stereoscopic disparity, with the associated standard error and confidence intervals; (3) a measure derived from the area between the subject's data points and the 25% guessing line, used as a backup indicator of stereo performance in subjects whose data could not be fitted; and (4) a score on the direction-of-depth discrimination task, the score being between 6/6 (i.e. a perfect score) and 0/12 (i.e. a score below chance, which would be 3/12).

One skilled in the art will recognize that the method illustrated in FIGS. 6A-6C and described above may be used with any type of test for stereo vision. Namely, data may be collected at different disparity levels, indicating the performance of the patient at each level. The data may then be fit with a logistic psychometric function 602, as shown in FIG. 6, using any suitable fitting routine as discussed herein above. The collected data may come from a stereo vision test discussed herein or from any other type of stereo vision test known in the art. Such analysis may allow more accurate data to be collected from patients using new and traditional testing methods.

The systems and methods described herein may be further extended to make measurements of stereo ability at different locations in the visual field. This could serve as a diagnostic tool for diseases which cause a local loss in sensitivity to disparity. For example, conditions affecting the retina of one eye would prevent a subject from being able to use disparity information at the affected location. This may allow deformations of the retina to be identified at an earlier stage than would be possible otherwise.

As discussed above, the present disclosure measures the ability of a patient to detect depth from disparity in images. The perception of depth from disparity requires compatible inputs from corresponding regions of the retina in the two eyes. Where a disease of the retina deteriorates the quality of the output of one eye (or otherwise causes it to no longer correspond to the output of the other eye), this will impair stereoacuity at the location of the deterioration. Examples of diseases that could degrade the retina in this way include macular degeneration, glaucoma and diabetic retinopathy. Proper treatment of these diseases is helped both by early detection and by the ability to monitor disease progression. In current clinical practise, the health of different regions of the visual field can be assessed in a variety of ways. These include the Amsler grid, where a square grid is presented to the patient and they are asked to report any abnormalities in its appearance. Preferential Hyperacuity Perimetry works on a similar principle; the patient is presented with a row of dots and asked to indicate any deviations from a straight line. Both of these tests rely on the patient's subjective report of their visual perception. There are also visual field tests in which a patient is presented with targets (flashing lights or other luminance-contrast stimuli) at different locations in their visual field and scored on their ability to detect targets at each location. These tests will not identify cases where the percept is visible, but distorted or where the later brain areas “perceptually fills-in” the distorted or unresponsive region. Furthermore, each eye must be tested separately.