WO2022018701A1

WO2022018701A1 - Method and device for the description of a field-of-view image

Info

Publication number: WO2022018701A1
Application number: PCT/IB2021/056684
Authority: WO
Inventors: Paola Andrea AMAR SEPULVEDA; Patricia AMARIS; Oliver ANTEQUERA; Jorge José MARTÍNEZ RAMÍREZ; Silvia Carolina MORENO TRILLOS; Carlos Andrés OCHOA PERTUZ; Juan Pablo Pestana Nobles; Andrés Felipe QUINTERO PARRA; José Luis RODRÍGUEZ LOCARNO; Reynaldo Villarreal González; Luis José Escaf Jaraba
Original assignee: Clínica Oftalmológica Del Caribe
Priority date: 2020-07-24
Filing date: 2021-07-23
Publication date: 2022-01-27
Also published as: CO2020009156A1

Abstract

The present invention relates to methods and a device for the description of a field-of-view image. Specifically, to methods for the description of the result of a field-of-view test provided by an automatic visual field analyser. One of the methods comprises the stages of obtaining a graphic probability map of the visual field image, defining levels of statistical significance, generating a first matrix in accordance with the levels of statistical significance and the probability map, generating a second matrix in accordance with a region growing method, and generating a list of classification results with a multi-class classification method. The device comprises a special-purpose computing unit and an image-acquisition unit connected to the special-purpose computing unit, said computing unit being configured to execute the methods disclosed.

Description

METHOD AND DEVICE FOR VISUAL FIELD IMAGE DESCRIPTION

Technical field of the invention

The present disclosure relates to methods for describing a visual field image. Particularly with methods for the description of the result of a visual field test delivered by visual campimetry equipment.

Description of the state of the art

The visual field is determined by a portion of reality that the eye captures each time it pans around. It is determined by imaginary lines that outline a square or a rectangle, where all the elements contained in each scene that is captured in the direct vision of reality are located. The width of the human visual field covers more or less 180 degrees, and only in a portion of those degrees are images clearly captured.

The visual field is measured by campimetry, which can be kinetic, with points of light moving inward until the observer can see them, or static, in which points of light are randomly lit on a white screen, so that the patient must press a button when visualizing the point. The most frequently used equipment is the Humphrey ^® automatic visual field analyzer.

The most widely used test patterns refer to 24 or 30 central degrees of vision, although most equipment allows the measurement of the entire visual field. To detect glaucoma and other eye diseases, a visual field test is applied, which is also called visual perimetry. This test corresponds to a medical examination used to evaluate changes in the visual field. Visual perimetry allows detecting defects in central and peripheral vision, it is used mainly to diagnose glaucoma, but it can also detect visual loss due to other causes, such as retinal diseases and optic nerve damage due to trauma, stroke or tumors. In practice, the interpretation of the visual field test is carried out by an ophthalmology specialist, this interpretation implies the investment of a certain time while the ophthalmologist analyzes the test and provides the patient with a diagnosis. Automated visual field analyzers, such as the Humphrey ^® Automated Visual Field Analyzer, have statistical databases of normal sensitivity values in different age ranges, which are used to compare with a patient's response. The result of the visual field test is a standardized document that comprises different graphs and indicators that show the patient's visual sensitivity thresholds in different areas of the visual field compared to normal values.

In the state of the art, publications related to the description of a visual field image are identified, such as, for example, the documents,

The document "A Probabilistic Multi-class Classifier for Structural Health" discloses a procedure for identifying glaucomatous visual field loss after applying a standard automated perimetry procedure (SAP). In said procedure In an example using a Humphrey ^® 24-2 testing algorithm, 54 individual points are tested, and the threshold value calculated for that point is compared to a database of people with similar vision and However, the methods used for the identification procedure are very complex and therefore may require a computer with great calculation power and take a long time to provide a result.

On the other hand, the document "Interpretation of Automated Perimetry for Glaucoma by Neural NetWork" discloses a method of interpreting automated perimetry for glaucoma through trained neural networks to classify patterns and be able to discriminate between a normal eye and eyes with glaucoma. Methods for classification include expert systems, multivariable statistical classifiers (eg, discriminant analysis), and neural networks. In particular, a neural network trained with the Backpropagation method is used to carry out a classification, finding a training of 1500 iterations as sufficient so that the percentage of concordance of the classification results between a human and a neural network of two layers is between 72% and 75%. However, the methods used for the classification procedure require a large number of iterations to reach an acceptable percentage of agreement, so it can take a long time to reach an optimal percentage, in addition, the methods are very complex and therefore can require a computer with great calculation power and take a long time to give a result.

However, the method disclosed in the document "Interpretation of Automated Perimetry for Glaucoma by Neural NetWork" does not disclose that multi-class growth region and decision tree methods are applied, although it does disclose that the methods for classification include expert systems, statistical classifiers multi-variable (such as discriminant analysis) and neural networks and that the visual fields are evaluated looking for the grouping of different amounts of adjacent points that exceed a certain sensitivity threshold to associate it with a particular condition.

Currently, it is common for several days to pass before the patient receives a description of the visual field test due to the lack of availability of medical specialists. The present invention makes it possible to generate a description of the visual field result at the time the graphic map of the visual field image is entered into the system, reducing the time between performing the test and timely treatment of the patient. In this way, the present invention allows medical care to be provided to a greater number of people.

As is evident, there is a need in the state of the art to have a visual examination description that is reliable and agile, in this way it is possible to provide timely treatment to a patient in addition to allowing medical care to a greater number of people. Brief description of the invention

The present disclosure relates to methods for describing a visual field image. Particularly with methods for the description of the result of a visual field test delivered by an automatic visual field analyzer, such as the Humphrey ^® perimeter or the Octupus ^® perimeter. One of the disclosed visual field image description methods comprises the following steps: a) obtain a graphic probability map of the visual field image, b) define levels of statistical significance, c) generate a first matrix assigning a value numerical value to each element of the matrix according to the levels of statistical significance and the graphical probability map of the visual field image, d) generate a second matrix with a set of numerical values adjacent to each other by applying a method of growth of region (130) with an initial value to the first array; and e) generating a list of classification results by applying a multiclass classification method to the second generated matrix, wherein the list of results describes the visual field image.

Another embodiment of the method comprises that in step c), conditions are checked iteratively based on the regions of a graphical probability map of the visual field image with pre-established significance levels to generate the first matrix.

Another embodiment of the method consists in that in step b), particular ranges are defined for the levels of significance.

Another particular embodiment of the method consists in that, in step d), the region growth method executes a series of steps according to a specific similarity criterion.

A specific embodiment corresponds to the fact that in step d2) of step d) the similarity criterion corresponds to a determined intensity level of a pixel.

Another embodiment of the method comprises a list of particular results that can be obtained in step e).

Another embodiment of the method corresponds to the fact that in step e) the applied multiclass classification method is based on the location of the values in the second matrix with respect to a blind spot region, a central fixation, a nasal region and a region paracentral of the graphical probability map of the visual field image.

A particular embodiment of the method consists in that in step e) specific conditions are followed based on particular statistical significance levels to generate a list with classification results. Step e) in a specific embodiment, precedes a step f) in which said list of classification results is exported to a peripheral, such as a printer, printing said result. You can also export the results list to a computer readable medium in a digital file that can be sent to a remote computer. The results list can also be exported to a display device like those mentioned above such as a screen or projector.

On the other hand, the device for visual field description comprises a specific purpose computing unit and an image acquisition unit connected to the computing unit, where the computing unit is configured to execute the steps of each one. of the methods for visual field description above.

Brief description of the figures

FIG. 1 shows a general flow chart of one of the methods disclosed herein.

FIG. 2 shows a specific flowchart of one of the methods disclosed herein.

FIG. 3 shows a pattern deviation map or pattern deviation probability plot of a visual field test for a left eye.

Detailed description of the invention

The present disclosure relates to methods for describing a visual field image. Particularly for the description of a visual field image of a campimetry report delivered by a visual campimetry equipment, visual perimetry or a visual field study. Said visual field image is used to assess alterations in the visual field of an individual.

For the understanding of this document, it will be understood that visual perimetry equipment also corresponds to a visual perimetry device, perimeter or automated perimeter. A result of a visual field test also corresponds to a visual field study or results report, which is a document delivered by a visual campimetry team and includes a graphic probability map of the visual field image, the latter is described later in this document. On the other hand, the visual campimetry equipment from which the result of a visual field test can come can be a Humphrey ^® campimeter or an Octopus ^® campimeter, among others, although it should be understood in this disclosure that it is not limited only to these devices, but any other that generates a visual field result.

For purposes of understanding this disclosure, a person of ordinary skill in the art would understand that the disclosed methods can take the form of being an entire process or method with a series of steps or being a device that executes the series of steps of said process. or method. In addition, the methods and systems may take the form of a computer program product on a computer-readable storage medium, having computer-readable program instructions, for example, a computer program that is incorporated or embedded in said storage medium. storage and subsequently executes each of the stages or steps of the disclosed method. Current methods can take the form of a computer program that runs on the web. Any suitable computer-readable storage media may be used including hard drives, CD-ROMs, optical storage devices or magnetic storage devices and combinations of these.

However, the computer may include or be connected to a storage device, display device and/or a Human Interface Device (HID) such as a mouse, keyboard, it may also include peripherals such as an image acquisition device such as a scanner or a camera, may also be or include a specific or special purpose computing unit programmed to execute the method of this disclosure.

Thus, a display device of the invention includes, without limitation, any device that can be connected to a computing unit and display its output. Such a display device is selected from among others CRT (Cathode Ray Tube) monitor, flat screen, LCD liquid crystal display (Liquid Crystal Display), active matrix LCD, LCD Passive matrix LCD, LED screens, screen projectors, TV (4KTV, HDTV, Plasma TV, Smart TV), OLED screens (for Organic Light Emitting Diode), AMOLED screens (for Active Matrix Organic Light Emitting Diode), QD quantum dot displays, segment displays, among other devices capable of displaying data to a user, known to those skilled in the art, and combinations of these.

For its part, an HID device (for the acronym in English of Human Interface Device) includes, without limitation, keyboard, mouse, trackball, touchpad, pointing device, joystick, touch screen, among other devices capable of allowing a user to enter data in the computing unit of the device and combinations of these. For example, the HID device allows entering initial configuration values of the method such as significance levels, seeds to start the processing of a visual field image, etc.

On the other hand, the image acquisition device can include sensors sensitive to the visible spectrum or to other portions of the electromagnetic spectrum, used to capture an image that is in the visual field of the same. The image acquisition device is selected from the group consisting of compact cameras, APS (Advanced Photo System) cameras, SLR (Single Lens Reflex) cameras, digital cameras, TLR ( stands for Twin Lens Reflex), scanners and combinations of these.

Also, it should be understood in this document that when referring to a storage medium, it includes, but is not limited to, RAM memory (cache memory, SRAM, DRAM, DDR), ROM memory (Flash, Cache, hard drives, SSD, EPROM, EEPROM, removable ROM (eg SD (miniSD, microSD, etc), MMC (MultiMedia Card), Compact Flash, SMC (Smart Media Card), SDC (Secure Digital Card), MS (but not limited to Memory Stick), CD-ROM, Digital Versatile Disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, storage or any other medium that can be used to store information and that can be accessed by a computing unit or a processing unit. Instructions, data structures, computer program modules are generally embedded in memory registers. Some examples of data structure are: a text sheet or a spreadsheet or a database. For its part, the database can be selected from among others: hierarchical databases, network databases, transactional databases, relational databases, multidimensional databases, object-oriented databases, documentary data, deductive databases, and other databases known to a person of moderate skill in the art.

It is also to be understood herein that embodiments of the methods are described with reference to flow charts or method illustrations. It will be understood that each flowchart or method illustrations may be computer implemented as program instructions. These computer program instructions can be loaded into a specific or special purpose computer or special purpose computing unit or other data processing unit that is programmable, so that the instructions that are executed on the computer or other data programmable are a means of implementing functions specified in the disclosed flowcharts.

An example of a specific purpose device of the present invention is the one that has programmed the code or algorithm that describes each of the steps of the method sequentially.

Now, the disclosed method allows to describe, in natural language, the result of the visual field test in which the following types of visual field defects are described: paracentral scotoma, nasal step, double nasal step, Seidel scotoma, single arcuate scotoma , elevational scotoma, double arcuate scotoma, double scotomas, scotomas located in the upper quadrant of the visual field, scotomas located in the lower quadrant of the visual field, and combinations of these.

Referring to FIG. 1 shows the general flow diagram of an embodiment of the visual field image description method comprising the following steps: a) obtain a graphic probability map of the visual field image (110), b) define levels of statistical significance, c) generate a first matrix assigning a numerical value to each element of the matrix according to the levels of statistical significance and the graphic probability map of the visual field image (120), d) generate a second matrix with a set of numeric values adjacent to each other by applying a region growth method (130) with an initial value to the first matrix, and e) generating a list with classification results by applying a multiclass classification method (140) to the second generated matrix; where the list of results describes the visual field image.

Step e) of generating the list with the classification results, in a specific embodiment, precedes a step f) in which said list of classification results is exported to a peripheral, such as a printer, printing said result. You can also export the results list to a computer readable medium in a digital file that can be sent to a remote computer. The results list can also be exported to a display device like those mentioned above such as a screen or projector.

For the understanding of this document, the graphical probability map of the visual field image corresponds to a pattern deviation map or pattern deviation probability plot or PDPP (for the acronym in English Pattern Deviation Probability Plot). This PDPP is a graph provided by a perimeter and is also known as a visual field test result.

Said PDPP corresponds to an image that describes the amount of loss of visual sensitivity and measures how different this value is compared to the value of a normal person of the same age. Computerized perimeters commonly have databases with ranges of normal values for different ages that are used to generate the PDPP, particularly based on the result of a comparison of a measured visual sensitivity value and a reference value for an individual of the population. same age.

Referring to FIG. 3 shows an example of a visual field test pattern deviation map delivered by a perimeter. Said map comprises two visual test pattern deviation maps, a first map corresponding to a total visual field test pattern deviation map (1) and a second map corresponding to a total visual field test pattern deviation map. pattern visual field (2). The visual field test pattern deviation map further comprises significance levels.

In the example, both the total visual field examination pattern deviation map (1) and the visual field deviation map (2) comprise several regions of points that make up a pattern of significance and each pattern of significance is related to a level of significance represented in a white area in which they can be grouped up to a quantity of 6 by 7 black points or black pixels and a numerical value in percentages of 5 % and 2% and up to a quantity of 6 by 8 black points or black pixels and a numerical value in percentages of 1% and 0.5%.

In the particular example referred to in FIG. 3 the significance levels are related to said percentage value, which in turn is related to the significance pattern as indicated in the previous paragraph, as follows: a first significance pattern with 4 black dots in the white area when the significance is less than 5%, a second significance pattern with 16 black dots in the white area when the significance is less than 2%, a third significance pattern with 32 black dots when the significance level is less than 1%, and a fourth significance pattern with 48 black dots or an area of 6 by 8 black pixels when the significance level is less than 0.5%.

A person of moderate skill in the art would understand that the significance patterns are not limited by the number of points and that they generally represent a density of different points for the number of significance levels used. In other words, for the illustrated example, there are 4 levels of significance related to 4 patterns of significance that have different point or pixel densities between them.

In step b) of the method, in which the significance levels described in the previous paragraph are defined, the significance levels and their related significance patterns can be configured in the default perimeter, they can be acquired from a server with parameters defined configuration settings, or can be entered by a user. For example, a computer that executes the method can download the values of each of the significance levels from the network, which would allow these significance levels to be adjusted according to a specific geographical region where it is used to improve the results of the description. , given that said value of statistical significance may be associated with an ethnic or population factor. In another embodiment of step b) of the method, the levels of statistical significance can be defined by a group of experts. In step c) of the method of the present disclosure, a first matrix is used that corresponds to a data array in a computer memory register in which a reading of the model deviation map of the model is stored sequentially. examination of the visual field according to the patterns of significance, the matrix is useful for comparison operations to be performed on the values stored in it in the subsequent stages to obtain the regions of contiguous points to finally determine a particular value for a list of with classification results that are stored in another memory register. This will be explained in greater detail in each of the subsequent steps, and more specific modalities later.

In a particular example of the method, the reading of the model deviation map can be done by means of a peripheral connected to the computer, for example a scanner or a camera, also directly from a file stored or downloaded in a memory register of said computer. .

In a specific embodiment of the method, in step c), the first matrix is generated by comparing the regions of the deviation map of the visual field test model with the patterns of significance, checking the following conditions: el) if the region of the map probability graph of the visual field image to compare is equal to the significance level, then assign a numerical value in the first matrix according to the significance level and compare another region of the probability graph map of the visual field image, c2) if the region of the graphical probability map of the visual field image to be compared is different from the level of significance, then compare the region of the graphical probability map of the visual field image with another level of significance defined in step b) , c3) if the region of the graphical probability map of the visual field image to be compared is different from all the significance levels defined in step b) e Then assign a default numerical value in the first matrix, where the default numerical value corresponds to the fact that the significance level of the region could not be determined, and c4) the process of step c) is repeated until all the regions are compared. of the pattern deviation map, thus generating the first matrix. Alternatively, after stage c), a stage d) is executed in which the matrix of adjacent points is obtained, in which the region growth method is applied, which performs the following steps: di) select an element of the first matrix, d2) compare if the value of the selected element meets a similarity criterion with respect to a value of an element next to the selected element in the following way with the sub-steps: d21) if the value of the next element to the selected element meets with the similarity criterion and is not the last of the elements of the first array, then store in the second array the position of the adjacent element and return to step (di); d22) if the value of the element next to the selected element meets the similarity criterion and is the last of the elements of the first array, then store in the second array the position of the next element and finish; d23) if the value of the element next to the selected element does not meet the similarity criterion and is not the last element of the first array, then compare with another element next to the selected element and go back to step d2); and d24) if the value of the element next to the selected element does not meet the similarity criterion and is the last of the elements of the first array, then terminate.

In a step e) of the method, subsequent to step c), a multiclass classification is performed from the matrix of adjacent points obtained in step c). Possible outcomes include paracentral scotoma, nasal step, double nasal step, Seidel scotoma, single arcuate scotoma, altitudinal scotoma, double arcuate scotoma, double scotomas, upper quadrant visual field scotomas, lower quadrant visual field scotomas. visual field and combinations of these.

In general, the mentioned multiclass classification method is based on the location of the values in the second matrix with respect to a blind spot region, a central fixation, a nasal region and a paracentral region of the deviation map of the exam model. of the visual field. Optionally, in a particular embodiment of the method, the multiclass classification method of d) generates a list with classification results according to the following conditions: di) if in the second matrix there are at least three contiguous elements with a level of statistical significance less than 5%, at least two contiguous elements with a statistical significance level of less than 1%, which are in a range between 5 and 21 degrees central fixation, and not contiguous to the blind spot region, then enter a paracentral scotoma result in the results list (116); d2) if in the second matrix there are at least three contiguous elements in the second matrix with a statistical significance level of less than 5%, at least two contiguous elements with a statistical significance level of less than 1%, and any of these elements is found in the periphery of the nasal region, then enter a nasal step result in the result list (105); d3) if in the second array there are more than five contiguous elements in the second array, then enter a large nasal step result in the result list; d4) if in the second matrix there are at least three contiguous elements with a statistical significance level of less than 5%, at least two contiguous elements with a statistical significance level of less than 1%, contiguous to the blind spot region, and none of the contiguous elements extending to the periphery of the nasal region, then enter a Seidel scotoma result in the result list (114); d5) if in the second matrix there are at least seven elements contiguous to the blind spot region, each with a statistical significance level of less than 1%, three contiguous elements that form a nasal step in the same visual hemifield, then enter a single arcuate scotoma result in the results list (109); and when all points in the same visual hemifield on the horizontal meridian have a statistical significance level of less than 1%, then enter an altitudinal scotoma result in the results list; where the visual hemifield corresponds to the upper or lower zone of the second matrix.

For the understanding of this document, it should be understood that when referring to the degrees of central fixation, it corresponds to a distance measured in degrees from the fixed point to which an individual looks during the course of a visual field test with a field meter.

The disclosed method, in some embodiments, combines a statistical significance level matrix and a graphical probability map of the visual field image and a region growth method with a second matrix. Due to the use of a region growth method, it makes the description of the field exams more agile. visual, it also allows the method to be run on computers that have reduced computing power.

The region growth method is used to obtain a visual field image segmentation of the result of a visual field examination. Said region growth method requires an initial seed point in the region where visual field image processing begins. Then from this seed point the neighboring pixels are calculated, and if they meet a similarity criterion then they are added to a region, using an array to store this value. Said similarity criterion corresponds to whether the pixel intensity is below or above a predetermined threshold corresponding to a particular value of statistical significance. For example, if the pixel intensity is greater than a statistical significance value of 0.5%. The process of calculating neighboring pixels is repeated for each pixel in the region, and the method continues until no more pixels are found that can be added to the region.

Optionally, in this method, the regions grow by aggregating pixels similar in value with respect to a probability P that is used to perform the segmentation, and alternatively the method allows a user to select a set of seed points in the image. These seed points will serve as starting points for the region growth process, whereby the final number of regions is equal to or less than the number of seeds sown by the user.

To carry out the pixel aggregation in the region growth process, it is done according to the similarity criterion explained above. However, other similarity criteria can be the difference between the value of the pixel to be added and the value of the seed, or the average value of the region already formed, being less than a predetermined threshold, analogously as explained above. .

Optionally, the described method for describing a visual field image can be implemented on a web platform. This allows a user to execute the method and obtain the description of a visual field result from anywhere in the world.

Said description of the visual field test report can be generated assisted by an operator, or without the intervention of an operator, that is, automatically. In an embodiment of the disclosed method, in step e) of generating a list with classification results, the following conditions are checked: el) if there are at least three contiguous elements in the second matrix with a statistical significance level of less than 5%, at least two contiguous elements with statistical significance level less than 2%, which are in a range between 5 and 21 degrees from the central fixation, and not contiguous to the blind spot region, then enter a result of paracentral scotoma in the results list, e2) if in the second matrix there are at least three contiguous elements in the second matrix with a statistical significance level of less than 5%, at least two contiguous elements with a statistical significance level of less than 1%, and any of these elements is located on the periphery of the nasal region, then enter a nasal step result in the result list, e4) if there are more than five contiguous elements in the second array s in the second matrix, then enter a large nasal step result in the result list, e5) if in the second matrix there are at least three contiguous elements with statistical significance level less than 5%, at least two contiguous elements with level of statistical significance less than 1%, contiguous to the blind spot region, and none of the contiguous elements extending to the periphery of the nasal region, then enter a Seidel scotoma result in the results list, e6) if in the second matrix there are at least seven elements contiguous to the blind spot region, each with a statistical significance level of less than 1%, three contiguous elements that form a nasal step in the same visual hemifield, then enter a result of arcuate scotoma unique in the results list; and e7) when all points in the same visual hemifield on the horizontal meridian have a statistical significance level of less than 1%, then enter an altitudinal scotoma result in the results list, where the visual hemifield corresponds to a zone of the second array.

In other embodiments of the disclosed method, the possible results include: paracentral scotoma, nasal step, double nasal step, Seidel scotoma, single arcuate scotoma, altitudinal scotoma, double arcuate scotoma, double scotomas, scotomas located in the upper quadrant of the visual field, scotomas located in the lower quadrant of the visual field and combinations of these.

Optionally, the disclosed method has a step subsequent to step e) in which a machine-readable document is generated showing the list of results from step e). On the other hand, the device for the description of a visual field image comprises: a specific purpose computing unit; and an image acquisition unit connected to the computing unit, where the computing unit is configured to: a) obtain a graphic probability map of the visual field image with an image acquisition unit connected to a computation and store it in a memory register, b) define levels of statistical significance in the computation unit, c) generate a first matrix assigning a numerical value to each element of the matrix according to the levels of statistical significance and the graphic map of probability and storing it in a memory register, d) generating a second matrix with a set of numerical values adjacent to each other by applying a region growth method (130) with an initial value to the first matrix; and e) generating a list of classification results by applying a multiclass classification method (140) to the second generated matrix, wherein the list of results describes the visual field image.

Referring to FIG. 2, a specific flow diagram of one of the methods disclosed herein is shown. The diagram process is carried out at least twice. In an example of execution of the method, in general, in a first exploration sequence, a scan can be made in the superior hemisphere of the visual field in search of visual field defects, and subsequently, an exploration in the inferior hemisphere of the visual field can be carried out. visual. A particular condition is that visual field defects are found in both hemispheres of the visual field, which makes it possible to distinguish double visual field defects, such as arcuate scotoma. In a first step (100) you read the PDPP exam result. Step (100) is equivalent to step a) described above. In a particular embodiment, reading the test result corresponds to acquiring it by means of an image acquisition device such as a scanner.

After step (100) a step (101) is executed in which a conversion of the test result into a numerical matrix is carried out, this corresponds to generating a first matrix assigning a numerical value to each element of the matrix according to the statistical significance levels and the graphical probability map of the visual field image, which is the same as step c) described above.

After step (101), a step (102) is continued in which a search is made for points in the nasal region that meet the condition of having a P<5%, that is, a level of statistical significance less than 5%, subsequently a step (103) of growing the nasal region is executed, which corresponds to a step equal to step d) described above. Afterwards, a step (104) is followed in which the condition of whether the nasal region has more than 3 points with P<5% or whether the nasal region has more than 2 points with P<1% is checked, if said condition is fulfilled, then a step (105) is passed in which a nasal step result is entered as in step d2) of step d).

Then in a step (106) the points in the region adjacent to the blind spot with P<5% are searched, that is, a level of statistical significance less than 5%, subsequently a step (107) is executed to grow the adjacent region , which corresponds to a stage equal to stage d) described above. Afterwards, a step (108) is followed in which the condition of whether the adjacent region has more than 7 points and the nasal region has more than 1 point is checked, if said condition is fulfilled then a step (109) is passed in in which an arcuate scotoma result is entered as in step d5) of stage d).

Otherwise, if the condition is not met, a step (110) is followed in which points are sought in the paracentral region with P<5%, that is, a level of statistical significance less than 5%, subsequently it is executed a step (111) of growing the paracentral region, which corresponds to a step equal to step d) described above. Afterwards, a step (112) is followed in which the condition of whether the adjacent region is equal to the paracentral region is checked, if said condition is fulfilled then a step (113) is passed in the that a Seidel scotoma result is entered as in step d4) of step d).

Otherwise, if the condition is not met, a step (115) is followed in which the condition of whether the paracentral region is greater than 3 points is checked, if said condition is met then a step (116) is passed. in which a paracentral scotoma result is entered as in step d4) of stage di).

The following is the particular pseudocode that does the region growth method procedure:

In a particular embodiment of the method for the description of a visual field image, a step is included in which a group of experts enter the configuration of the initial values of the method that correspond to criteria such as predetermined threshold, probability P, levels of statistical significance , number of pixels in an area that defines a pattern of statistical significance, size of the first matrix, size of the second matrix and seed value.

The configuration of the initial values of the method corresponds to a training step that can be executed iteratively before or during the execution of the steps of the disclosed method so that the method can give an interpretation of a visual field image accurately.

The disclosed method can be programmed in the specific purpose computing unit using a programming language selected from Java, JavaScript, Perl, PHP and Python C, C++, #C, Python, SQL, Swift, Ruby, Delphi, Visual Basic, D, and other programming languages known to a person of moderate skill. For example, Python.

On the other hand, for the processing of visual field images, open source libraries such as opencv can be used without being limited to these libraries.

Likewise, the method can be executed from a Web Platform with the Django Framework.

The inventors tested the disclosed method for its accuracy, sensitivity, specificity, likelihood ratio, and ROC/AUC values.

The test included measurements in a sample of 212 visual field exams, said sample is composed of male and female individuals between 18 and 90 years of age. For the test, Humphrey ^® perimeters were used in a 24-2 SITA configuration, these in turn used machine learning with VFT algorithms (for the acronym in English of Value Function Transfer) each of the 212 visual field exams of the test are similar to those shown in FIG: 3. Each sample visual field test was processed by the disclosed method, as follows: First, a graphic probability map of the visual field image of the first visual field test was obtained, then a level of statistical significance was defined as follows: a first level of significance between 5% and 2%, a second level of significance between 2% and 1%, a third significance level between 1% and 0.5% and a significance level of less than 0.5%.

With this, a first matrix was generated assigning a numerical value to each element of the matrix according to the levels of statistical significance and the graphic probability map of the visual field image obtained.

Subsequently, a second matrix with a set of numerical values adjacent to each other was generated by applying a region growth method with an initial value to the first matrix; and finally, a list with classification results was generated by applying a multiclass classification method to the second generated matrix based on the location of the values in the second matrix with respect to a blind spot region, a central fixation, a nasal region and to a paracentral region of the graphical probability map of the visual field image.

At the end of the test, a list of results is obtained describing each of the visual field images of the sample with various classifications described, among which paracentral scotoma, nasal step, Seidel scotoma, single arcuate scotoma, altitude scotoma, arcuate scotoma were found. double and combinations of these.

To evaluate the disclosed method, the inventors made the following measurements according to the confusion matrix in Table 1 below.

Table 1. Confusion matrix, test results.

In the TP table it corresponds to True positive: the VFT (from the acronym in English for Visual Field Test) or visual field test has some defect and the prediction is positive.

In the FP table it corresponds to False positive: the VFT has no defect (Normal) but the prediction is positive. In the table TN corresponds to True negative: the VFT has no defects (Normal) and the prediction is negative.

In the table FN corresponds to False negative: the VFT has some defect but the prediction is negative. The precision of the method refers to the ability of the method to interpret a visual field test with some specific defect of those visual fields that do not have a normal visual field result. To calculate the precision of the method, the following mathematical equation was used:

Where Acc is the precision of the method. Substituting the values from Table 1 results in a precision of 97.64%.

The sensitivity of the method refers to the ability of the method to interpret a visual field test with some specific defect. To calculate the sensitivity of the method, the following mathematical equation was used:

where Sen is the sensitivity of the method. Replacing the values in Table 1 results in a sensitivity of 98.91%.

Method specificity refers to the ability of the test to correctly identify visual field tests without any defects. To calculate the specificity of the method, the following mathematical equation was used:

Where Esp is the specificity of the method. Replacing the values in Table 1 results in a specificity of 96.66%. The test likelihood ratio is defined as the ratio of expected test results in subjects with some (correctly predicted) defects on the visual field test to subjects with normal test.

The odds ratio for positive test results (LR+) indicates how much more likely it is that a positive test result will occur for subjects with a specific test defect (correctly predicted) compared to those who have been misidentified by the test. model. To calculate the probability index for positive results of the LR+ test of the method, the following mathematical equation was used:

Where LR+ is the probability index for positive results of the LR+ test. Substituting the values from Table 1 results in a likelihood ratio value for positive LR+ test results of 29.613.

The probability index for negative test results (LR-) represents the ratio between the probability that an incorrect interpretation of the visual field test will occur in subjects with some defect and the probability that the same result will occur in subjects without any defect. in the visual field test. To calculate the probability index for negative results of the LR-test of the method, the following mathematical equation was used:

Where LR- is the probability index for negative results of the LR- test. Substituting the values from Table 1 results in a value of 0.0112.

The present invention is not limited to the modalities described and illustrated, since as will be evident to a person versed in the art, there are possible variations and modifications that do not deviate from the spirit of the invention, which is only defined by the following claims.

Claims

1. A method of describing a visual field image, comprising the following steps: a) obtain a graphic probability map of the visual field image

(110); b) define levels of statistical significance; c) generating a first matrix assigning a numerical value to each element of the matrix according to the levels of statistical significance and the graphical probability map of the visual field image (120); d) generating a second matrix with a set of mutually adjacent numerical values by applying a region growth method (130) with an initial value to the first matrix; and e) generating a list with classification results by applying a multiclass classification method (140) to the second generated matrix; where the list of results describes the visual field image.

2. The method of Claim 1, wherein in step c) the generation of the first matrix is performed by comparing regions of the graphic probability map of the visual field image that correspond to a group of points that are equivalent to a level of statistical significance with each of the statistical significance levels defined in step b) according to the following conditions: el) if the region of the graphic probability map of the visual field image to be compared is equal to the level of significance then assign a numerical value in the first matrix according to the level of significance and compare another region of the graphical probability map of the visual field image; c2) if the region of the graphical probability map of the visual field image to be compared is different from the level of significance, then compare the region of the graphical probability map of the visual field image with another level of significance defined in step b); c3) if the region of the graphical probability map of the visual field image to be compared is different from all the significance levels defined in step b) then assign a predetermined numerical value in the first matrix, where the value default numeric corresponds to that the significance level of the region could not be determined; and c4) repeating the region comparison until all regions of the graphical probability map of the visual field image are compared.

3. The method of Claim 2, wherein in step b) the levels of significance are: a first level of significance between 5% and 2%, a second level of significance between 2% and 1%, a third level of significance between 1% and 0.5% and a significance level of less than 0.5%.

4. The method of Claim 1, wherein in step d) the region growth method (130) performs the following steps: di) selecting an element from the first array; d2) compare whether the value of the selected element meets a similarity criterion with respect to a value of an element contiguous to the selected element; d21) if the value of the element next to the selected element meets the similarity criterion and is not the last of the elements of the first array, then store in the second array the position of the next element and return to step (di); d22) if the value of the element next to the selected element meets the similarity criterion and is the last of the elements of the first array, then store in the second array the position of the next element and finish; d23) if the value of the element next to the selected element does not meet the similarity criterion and is not the last element of the first array, then compare with another element next to the selected element and go back to step d2); and d24) if the value of the element next to the selected element does not meet the similarity criterion and is the last of the elements of the first array, then terminate. 5. The method of Claim 4, wherein in step d2) the similarity criterion corresponds to an intensity level of one pixel.

6. The method of Claim 1, wherein in step e) the list has results that are selected from the group consisting of paracentral scotoma, nasal step, seidel scotoma, single arcuate scotoma, altitudinal scotoma, double arcuate scotoma and combinations of these.

7. The method of Claim 1, wherein in step e) the multiclass classification method (140) applied is based on the location of the values in the second matrix with respect to a blind spot region, to a central fixation, to a nasal region and to a paracentral region of the graphical probability map of the visual field image.

8. The method of Claim 1, wherein step e) of generating a list with classification results follows the following conditions: el) if in the second matrix there are at least three contiguous elements with a statistical significance level of less than 5 %, at least two contiguous elements with statistical significance level less than 2%, that are in a range between 5 and 21 degrees from the central fixation, and not contiguous to the blind spot region, then enter a result of paracentral scotoma in the results list; e2) if in the second matrix there are at least three contiguous elements in the second matrix with a statistical significance level of less than 5%, at least two contiguous elements with a statistical significance level of less than 1%, and any of these elements is found in the periphery of the nasal region, then enter a nasal step result in the results list; e4) if in the second array there are more than five contiguous elements in the second array, then enter a large nasal step result in the result list; e5) if in the second matrix there are at least three contiguous elements with a statistical significance level of less than 5%, at least two contiguous elements with a statistical significance level of less than 1%, contiguous to the blind spot region, and none of contiguous elements that extend to the periphery of the nasal region, then enter a result of seidel scotoma in the results list; e6) if in the second matrix there are at least seven elements contiguous to the blind spot region, each with a statistical significance level of less than 1%, three contiguous elements that form a nasal step in the same visual hemifield, then enter a single arcuate scotoma result in results list; and e7) when all points in the same visual hemifield on the horizontal meridian have a statistical significance level of less than 1%, then enter an altitudinal scotoma result in the results list; where the visual hemifield corresponds to an area of the second matrix.

9. A visual field image description device, comprising: a specific purpose computing unit; and an image acquisition unit connected to the computing unit; wherein the computing unit is configured to: a) obtain a graphic probability map of the visual field image with an image acquisition unit connected to a computing unit and store it in a memory register; b) define levels of statistical significance in the computation unit; c) generating a first matrix assigning a numerical value to each element of the matrix according to the levels of statistical significance and the graphical probability map and storing it in a memory register; d) generating a second matrix with a set of mutually adjacent numerical values by applying a region growth method (130) with an initial value to the first matrix; and e) generating a list with classification results by applying a multiclass classification method (140) to the second generated matrix; where the list of results describes the visual field image.