Perimetry apparatus

Abstract

Described herein is an improved Perimetry test apparatus comprising software for predefining an optimised set of positions for perimetric test points, said software utilising a data map (Fmap) to determine said positions, the Fmap including features obtained from a digitised image of the retina to be perimetrically tested. The digitised image is an image of one or more of: retinal layers thickness; reflectivity or fluorescence of retinal structures; retinal vasculature; and functional maps derived as direct transformation of imaging data. The retina image is compared to a library map or reference, and the deviation between the retinal image and the library image is used to produce the Fmap. The data map includes values assigned to coordinates on the map, and optionally the software further includes smoothing the assigned values to provide a smoothed data map. The software further includes taking the second derivative of the values of the data map, or smoothed data map and selecting spaced test point positions based on locations where said second derivative has the greatest value.

Description

[0001] PERIMETRY APPARATUS

[0002] This invention relates to improvements in perimetry apparatus, in particular, the optimisation of test point locations used in automated perimetry tests, and software therefor.

[0003] Introduction

[0004] A perimetric eye examination, also known as a Visual Field (VF) test, is a diagnostic procedure that assesses the sensitivity to light at different point in a patient's field of vision. It is commonly used to detect and monitor conditions affecting the VF, such as glaucoma, optic nerve damage, and neurological and retinal disorders.

[0005] During a perimetric examination, the patient sits at a specialized device called a perimeter or VF analyser. The process typically involves the following steps:

[0006] Preparation: The patient is positioned, and their chin is placed on a chin rest. If necessary, one eye is covered with an eye shield, while the other eye is used for testing. The patient's corrective lenses, if applicable, are usually worn during the test.

[0007] Instruction: The patient receives instructions on the test procedure. They are informed to fixate their gaze on a central target throughout the examination while responding to visual stimuli presented at different locations within their VF.

[0008] Automated Perimetry: In this method, the patient looks straight ahead at the target, and a computer-controlled instrument presents stimuli at various points within their VF, known herein as test points. These stimuli are usually small dots or lights of varying intensity, and the patient responds to them by pressing a button or signalling when they perceive the stimuli. The instrument tracks the patient's responses and creates a map of their VF sensitivity. It should be noted that other methods can be used to detect a response, such as pupillary response or gaze movements.

[0009] Test Patterns: Different test point patterns may be used to assess various aspects of the VF, such as central, peripheral, or specific areas. Test points are typically organised in standardised, regular grids that cover a pre-defined area of the VF. Common test patterns include the Humphrey Field Analyzer (HFA) 10-2, 24-2 or 30-2, which assess central vision, and the 60-4 or 120-4, which examine the peripheral VF. Results and Analysis: Once the testing is complete, the data collected is analysed to generate a VF map. This map displays the patient's sensitivity to light stimuli across different areas of their VF. The results are often compared to age-matched norms or the patient's previous tests to detect any changes or abnormalities.

[0010] Perimetric eye examinations are valuable for diagnosing and monitoring various eye conditions. They provide essential information about a patient's VF, helping eye care professionals assess the extent of vision loss and tailor treatment plans accordingly. Regular perimetry tests may be recommended to track the progression of diseases or assess the effectiveness of interventions over time.

[0011] Standard Automated Perimetry (SAP) testing of the type described above is an effective tool. However, there are significant issues associated with such testing. Firstly, they can be timeconsuming and taxing for patients. Patient fatigue and inattention can therefore affect the accuracy of the results, leading to potential errors. Additionally, the test relies heavily on patient cooperation and subjective responses, which may introduce variability in the data. Finally, SAP may struggle to detect subtle VF defects, particularly in the early stages of certain eye conditions, which could potentially delay diagnosis and treatment. It is possible to test with higher resolution by reducing the spacing of test points within the grid, but this comes at the cost of increasing the number of test points and test time.

[0012] The retina is a complex and layered tissue at the back of the eye that contains several types of specialized cells. These cells work together to capture and process visual information before sending it to the brain, allowing us to perceive the world around us. Any dysfunction or damage to this tissue can result in vision problems or retinal diseases.

[0013] Another important tool for assessing the condition of the eye are images of the retina. Of particular use is an Optical Coherence Tomography (OCT) image of the retina which provides detailed visualization of the cell layers and structures within the retina.

[0014] OCT imaging utilizes light waves to create high-resolution, cross-sectional images of the retina. It provides indications of the structural integrity and health of the retina, aiding in the diagnosis and management of various eye conditions and diseases. During an OCT scan, the patient rests their chin on a support while looking at a target. The device captures a series of 2D innages using low-power, non-invasive light. These innages are then combined to generate a 3D representation or a series of cross-sectional images of the retina. The OCT image of the retina reveals the distinct layers within the retina. This detailed visualization helps ophthalmologists assess the thickness, structure, and any abnormalities within these layers. It is particularly useful in detecting and monitoring conditions like macular degeneration, diabetic retinopathy, glaucoma and other optic neuropathies.

[0015] Other images such as: OCT angiography, where a series of OCT images are captured to identify the presence of blood flow without the injection of any dye; Traditional angiography, where images are collected after administering the patient with blood dyes that fluoresce under the specific wavelength to obtain images highlighting blood vessels, abnormal vascular structures, haemorrhages and accumulation of fluid in the retina; Retinal fundus imaging, which employs a convention photographic technique to image the retina and can show features such as fat or lipid deposits, cell atrophy, and scar tissue features or reflectivity of the retinal surface; Autofluorescence imaging, where natural fluorescence of retinal pigment epithelial cells (RPE) is imaged and used to determine areas of altered signal, such as deposits of specific proteins or atrophic areas; Scanning laser ophthalmoscopy, where imaging is obtained by using a scanning laser to provide crisper images of the retina compared to traditional photographs.

[0016] All the imaging techniques above can provide a 2D or 3D representation of certain features of the retina, but do not provide the full extent of functional information provided by the VF test. Ultimately, only a perimetry test can provide a definitive measure of a patient's VF.

[0017] Summary of the invention

[0018] The inventor has realised that it is possible to reformat the known automated test point grid both spatially, and in number of points to increase both accuracy and speed of the perimetry test, by influencing the reformatting based on retinal feature images. The inventor was able to estimate the accuracy improvement using a map assumed to be the true target of the test. The map was reconstructed with a simulation using both the conventional grid and the reformatted configuration, and was compared to the original.

[0019] Accordingly, one aspect of the invention provides perimetry test apparatus comprising software for predefining an optimised set of positions for perimetric test points, said software utilising a retinal map obtained from a digitised image of the retina to be perimetrically tested, and utilising a data map (Fmap) to determine said positions, the data map (Fmap) including features obtained from a digitised image of the retina the features being deviations from a reference, or measurements taken from the digitised image.

[0020] Detailed description

[0021] Drawings

[0022] Examples of the invention are described in detail below with the aid of drawings, wherein

[0023] Figure 1 shows an example of a simulated modified image of features in a retina;

[0024] Figure 2 shows the image of Figure 1 further modified by software;

[0025] Figures 3a, b and c show the software-determined perimetric test point locations; and

[0026] Figure 4 shows a schematic of the software steps employed in an improved perimetric apparatus.

[0027] Figure 1 is an example of a map of a set "features" (Fmap) with an area denoted in degrees of inclination about an axis with its origin at the centre of vision. The Fmap is obtained firstly by making a scan of the retina (not shown) under investigation, in this example a digitised OCT image. That OCT image can be obtained by known techniques as described above.

[0028] Secondly, that retinal OCT image can then be mapped using software to provide a first data map (not illustrated), of discernible features of the image, in this example the data is the full retinal thickness. This is generally a known technique. Thirdly, again using software, the retinal thickness data map is compared to a similar second data map derived from the image of a normal, healthy retina obtained from a library of images, for example corresponding to the age and / or ethnicity of the patient. To aid the understanding of the invention the Fmap shown in Figure 1 represents the difference between the data in the first and second maps, to provide what is called herein the Fmap. To visualise the Fmap more easily, the greater that difference is, the darker is the grey used in the Fmap. So, in this example the core data needed is the deviation of the features obtained from a digitised image of the retina, when compared with the same features obtained from a digitised library image of an average retina with average vision, i.e., a heathy eye. For the purposes of illustrating the invention, the Fmap of Figure 1 has different greyscale intensities where the darker grey tones show the greatest deviation (indicative of an eye defect), and the brighter grey tones represent the least deviation, (indicative of healthier eye function).

[0029] The cell morphology of the retina (often referred to as "structure") obtained by the OCT scan image is a sensible choice for the Fmap because it has been found that topographic patterns of (adequately selected) structural features, like retinal thickness, correlate well with patterns of the visual function. With the image of Figure 1 as a starting point the methodology herein described optimises the location of perimetric test points locations based on, in this example, the Fmap of Figure 1.

[0030] Since, in this example, the Fmap of Figure 1 is a structural map from an eye with a defect (simulated in this example but similar to typical results), and is a comparison to a healthy eye as mentioned above, then the deviation between this map and that of an average healthy eye would provide a starting point to select a set of coordinates for test points rather than a predefined grid of test points, because visual function is likely to be affected in the vicinity of areas of greatest deviation. Additionally, locations where there is least deviation are also of interest, because those areas are likely to correspond to a healthy VF. The inventor has found that locations with higher change in spatial gradient in the Fmap provide more informative test location to approximate the Fmap with minimal error. Provided that the selected Fmap correlates well with a true VF sensitivity map (or an adequately scaled version of it), the spatial gradient in the Fmap can be used to identify the most informative locations for the perimetry test as well.The change in spatial gradient in the Fmap represents the level of "informativeness" that sampling the VF at a particular location would provide. In other words, the data shows how useful it would be to place a test point at that specific location. The data of Figure 1 is modified to provide VF test points based on the data in the Fmap. The modified data set is illustrated in Figure 2. That Figure 2 illustration, called information map (Imap), is obtained via software interrogation of the data of Figure 1. The resolution of the data for Figure 2 is chosen to substantially correspond with the resolution of the image resolution of the Fmap.

[0031] Software is optionally used firstly, to smooth the data in the Fmap so that the difference in values of local data is smoothed using a gaussian filter to make sure that the data is differentiable at every point of the considered area, as well as to make the software more robust to spurious readings.

[0032] Software is then used for the smoothed data set to find the absolute value of the planar gradient of the gradient (often referred to as the second derivative or the Hessian matrix) calculated from the smoothed Fmap. Using the second derivative means that the data set defines how the data's spatial gradient, or slope, is changing around the considered area. In simpler terms, values in the Imap of Figure 2 describe how much the gradient values change from one area to the next, and thereby define how much the retina features of interest (in this case cell depth compared to a normal eye) are changing across the retina. The higher the value in this Imap, the higher the chances that a test-point at that location can provide valuable information to assess boundaries of areas in the Fmap (and, consequently, in the VF for adequately selected features). Conversely, where the data's second derivative is zero, or a low value, then that is indicative of little or no rapid change in the spatial gradient of retina features and is of less interest.

[0033] From the Imap, informative test points are recursively placed on the test area derived from the most significant values of the Imap, i.e. where the second derivate values are highest, and thereby represent the most rapid change compared to the norm.

[0034] It should be noted that other maps directly derived from structural imaging can be used to build a Fmap, such as transformations of structural data into corresponding estimates of the VF (functional) map.

[0035] An example of software to modify data of the Fmap and arrive at the Imap is set out below in pseudo-code (where a double oblique indicates a text comment): / / smooth the Fmap map smooth_ Fmap = gaussianFilter(Fmap, filter_para meters)

[0036] / / Calculate the first gradient

[0037] [gx, gy] = gradient (smooth_ Fmap) absG = sqrt(gx.A2 + gy.A2 )

[0038] / / Calculate the second gradient

[0039] [ggx, ggy] = gradient absG )

[0040] Imap = sqrt( ggx.A2 + ggyA2 )

[0041] The resultant visualisation of the data from the Imap is shown in Figure 2. The informationmap calculated from the Fmap as the absolute value of the second derivative, where lighter greys represent more informative locations (more changeable data). This is the absolute value of the 2-dimensional second derivative.

[0042] Software is further employed to define the coordinates of test points in a recursive manner, starting from the location with the highest value on the information-map and progressing to locations with lower values. The placement of test points maintains a minimum distance between them and remains within the test area. This recursive process continues until either the entire desired test area is covered, or a maximum predetermined number of test-points is reached. The predetermined number of points can be predetermined to be equal or lower than the number of test points in the conventional grid used as a reference, thereby matching or increasing the test speed. By using this approach, the methodology aims to strategically distribute test points across the expected VF and its boundaries, based on their informative value, ensuring efficient and effective sampling for the reconstruction of the VF sensitivity map. Figure 3a illustrates the location of a first test point. Figure 3b illustrates where a second point would be placed, and Figure 3c illustrates the location of all the test points. An example of software for determining the location of the test points is:-

[0043] / / set initial parameters max_points = count(conventional_test_points) - x min_distance = r

[0044] / / initialise the list of test-points to empty so that the count is 0 optimised test points = [] n_points = count(optimised_test_points)

[0045] / / until points are less than predetermined number and there are still values > 0 to be

[0046] / / selected from the information map while (n_points < max_points) & any( information_map > 0)

[0047] {

[0048] / / find the max in the information map max_point = getCoordinates(max(information_map))

[0049] / / add the coordinates of the found point to the optimised list optimised_test_points.add(max_point)

[0050] / / update the count of the optimised list n_points = count(optimised_test_points)

[0051] / / set the neighbourhood of radius "min_distance" around the max point to 0 so that this / / area cannot be selected again in the following iterations information_map( neighbourhood) centre = max_point, radius = min_distance) )= 0

[0052] }

[0053] Figures 3a, b and c illustrate the placement of the test-points following the proposed methodology at different iteration steps and with the max number of test-points set to 68. Selected locations are represented with white dots. The Figures show the results of the iteration at step 1,2 and 58, respectively. It is possible to appreciate how the information map is updated at each iteration by setting a neighbourhood of the selected point to 0. The information maps shown in these figures are also 0 anywhere outside the test area, set to match the test area of a conventional grid. The white dots in Figure 3c show the optimised location for 68 test-points given the starting example Fmap of Figure 1.

[0054] Figure 4 is a simplified schematic illustration of the main software steps used to obtain the Imap of Figures 3a, b and c, in use with a perimetric apparatus 10. The perimetric apparatus 10 is schematically illustrated to show that the software described herein is operable within the apparatus via an internal computer or microcontroller 12 in this example. However, the software routine may be operable remotely, for example it may run on a server, or other computerised device based close to or remotely from the perimetric test apparatus, which is in continual or periodic data communication with the test apparatus 10. In that way existing perimetric testing apparatus that is unable to be adapted to the software routines described herein or where it is problematic to do that, can be adapted to use the software described herein.

[0055] Evaluation of the speed and accuracy of the reformatted test points according to the methodology described above, is possible based on the number of proposed locations compared to a conventional grid. For this example, the conventional grid "10-2" that consists of 68 test points covering an area of 10 degrees around the location of the central vision (fovea) is used as benchmark.

[0056] Test speed is mostly dependant on the number of presentations for the same subject under the same conditions. Therefore, the number of test points presented in a perimetry test provides a sensible estimate of the test length.

[0057] For assessing perimetry test accuracy, it is possible to simulate the test. The simulation takes the true VF sensitivity (in our case, we take this as the Fmap) at each test-point. This is equivalent of assuming that the response from the patient at each location results in the measurement equivalent to the value indicated by Fmap. The same process is used for both the conventional grid and the optimised locations. These values are then used to reconstruct a dense VF map of the whole test area. The reconstruction is obtained by means of interpolation to "fill the gaps" between test-points. In this case, thin-plate spline interpolation was used. However, the choice of a different interpolation would not affect the comparison in relative terms. Finally, the two reconstructed VFs (from the conventional grid and the optimised locations) are subtracted from the known VF sensitivity (Fmap) to calculate the mean absolute difference (MAD). The lower the MAD, the closer the reconstruction to the known VF sensitivity.

[0058] The benchmark test grid (the "10-2") uses 68 test-points. The reconstructed VF based on this configuration results in a mean absolute difference with the known VF sensitivity of 1.44 (MAD). An optimised configuration obtained with the methodology proposed herein, setting the max number of test points to 68 and minimum distance equal to that of the 10-2, reduces the difference to a MAD of 1.09. This means that for a test of similar speed (same number of test points) the optimised configuration leads to higher accuracy than the standard grid under the same conditions. Reducing the number of points by 34% (down to 45 points) still shows a closer reconstruction of the VF benchmark (MAD = 1.39 < 1.44).

[0059] In conclusion, the proposed methodology offers benefits in terms of speed or accuracy, or both. These benefits could be pushed further by introducing dynamic distances between points. For example, distance could change based on the variance of the information map around the selected test point. The lowerthe variance around a point, the higherthe distance. One way to achieve this efficiently is to measure the variance of the neighbourhood around the selected point at each iteration, and to expand the neighbourhood until the measured variance exceeds a set threshold. Masking, reducing, or enhancing parts of the Fmap represents another way to control the speed and / or accuracy of the perimetric testing. Another strategy could be to use a lower number of test location than the conventional grid and use simulations to identify areas with the highest difference from the Fmap. Then, points could be added, recursively targeting these areas to reduce the estimated error.

[0060] The example above is just one way to put the invention into effect. Various refinements to processing the Fmap have been discussed above. The generation of the Fmap map data set can be carried out in various ways where retinal layer thickness obtained via OCT is just one feature set that can be mapped. Other known imaging techniques available in ophthalmology, for example as described above, can provide such measurement maps quickly and non- invasively, for example a measurement of the reflectivity of selected retinal structures (such as the nerve-fibres) can be used for the Fmap, with or without a comparison to the same image of a heathy eye or other reference, for example a historic reference. Angiograms (including OCT-Angiography) can capture the retinal vasculature can also be used to create Fmaps (again with or without comparisons). Retinal photography (based on visible light or specific light-wavelengths) can provide images (colour images or other images based on different wavelengths, such as fundus autofluorescence) that can highlight features for the Fmap. The Fmap can be based on direct measurements or on their deviation from library reference normative data or library historic images of the patient's retina. Various inference methods can also be used to estimate approximate functional maps from imaging data, which can serve as Fmaps for the method described above.

[0061] It is apparent from the paragraph immediately above that the Fmap starting point need not be a comparison with a healthy average eye, or previous image from the patient, as in Figure 1, but could be any image, or images, which has features that can be quantified perse as they appear in the image.

[0062] In a refinement, it is possible to use features of more than one type to influence the position / co-ordinates of the optimised test points.

[0063] In the examples mentioned above, the objective is to obtain an optimised set of test points whose coordinates are based on data from the Fmap. So, even though the Fmap and the resultant Imap and test points have been illustrated in the Figures, it is not essential that they need to be illustrated and visually displayed in the process of obtaining the optimised test points. Rather the Fmap and Imap need only exist as data in the form of a value and a coordinate, and the test points need only be displayed once the optimised test is performed.

Claims

Claims1. Perimetry test apparatus comprising or using software for predefining an optimised set of positions for perimetric test points, said software utilising a retinal map obtained from a digitised image of the retina to be perimetrically tested, and utilising a data map (Fmap) to determine said positions, the data map (Fmap) including features obtained from the digitised image, the features being deviations from a reference, or measurements taken from the digitised image.

2. Apparatus as claimed in claim 1, wherein said digitised image is an image of one or more of: retinal layers thickness; reflectivity or fluorescence of retinal structures; retinal vasculature; and functional maps derived as direct transformation of imaging data.

3. Apparatus as claimed in claim 1 or 2, wherein said reference is a library map such as a historic retinal map of a retina to be tested or reference map, and the deviation between the retinal map and the library map is used to produce the data map (Fmap), or wherein said measurements are measurements of the reflectivity or fluorescence of the retina to be tested.

4. Apparatus as claimed in claim 1, 2 or 3, wherein said data map (Fmap) includes values assigned to coordinates on the map, and optionally the software further includes smoothing the assigned values to provide a smoothed data map.

5. Apparatus as claimed in claim 4, wherein the software further includes taking the second derivative of the change of adjacent data values of the data map, or smoothed data map.

6. Apparatus as claimed in claim 5, wherein said software further includes selecting spaced test point positions based on locations where said second derivative has the greatest values.

7. Apparatus as claimed in claim 6, wherein the minimum spacing of the test points is predefined.

8. Apparatus as claimed in claim 6 or 7, wherein the maximum number of the test points is predefined.

9. Apparatus as claimed in any one of the preceding claims, wherein said software further includes selecting areas of the data map (Fmap) which are excluded from the further software processing as defined in any one of claims 2 to 8, and / or selecting an area or areas of the data map (Fmap) which are intended for further software processing according to those claims.

10. Perimetry test software operable to predefine an optimised set of positions for perimetric test points, said software utilising a retinal map obtained from a digitised image of the retina to be perimetrically tested, and utilising a data map (Fmap) to determine said positions, the data map ( Fmap) including features obtained from a digitised image of the retina the features being deviations from a reference, or measurements taken from the digitised image.

11. Software as claimed in claim 10, wherein said digitised image is an image of one or more of: retinal layers thickness; reflectivity or fluorescence of retinal structures; retinal vasculature; and functional maps derived as direct transformation of imaging data.

12. Software as claimed in claim 10 or 11, wherein said reference is a library map, and the deviation between the retinal map and the library map is used to produce the data map (Fmap) or wherein said measurements are measurements of the reflectivity or fluorescence of the retina to be tested.

13. Software as claimed in claim 10, 11 or 12, wherein said data map includes values assigned to coordinates on the map, and optionally the software further includes smoothing the assigned values to provide a smoothed data map.

14. Software as claimed in claim 13, wherein the software further includes taking the second derivative of the change of adjacent data values of the data map, or smoothed data map.

15. Software as claimed in claim 14, wherein said software further includes selecting spaced test point positions based on locations where said second derivative has the greatest values.

15. Software as claimed in claim 15, wherein the minimum spacing of the test points is predefined.

17. Software as claimed in claim 15 or 16, wherein the maximum number of the test points is predefined.

18. Software as claimed in any one of the preceding claims 10 to 17, wherein said software further includes selecting areas of the data map (Fmap) which are excluded from the further software processing as defined in any one of claims 11 to 17, and / or selecting an area or areas of the data map (Fmap) which are intended for further software processing according to those claims.

19. Perimetry apparatus or software as claimed in any one of the preceding claims, wherein said software is run on a computer or the like operable remotely from the apparatus.

Images