Methods and systems for visual field test stimuli presentation and sensitivity map estimation

Abstract

A method for performing a visual field test that includes presenting, through a set of light emitters within a visual field test area, a first sequence of light stimuli; receiving, responsive to each light stimulus, a user response; and generating a sensitivity map of a visual field of a user as output of a machine learning (ML) model responsive to the user responses as input. The method includes determining that a subset of emitters are to present a second sequence of light stimuli based on the first predicted sensitivity map; presenting, through the subset of emitters, the second sequence of light stimuli; receiving, responsive to each light stimulus, the user response; and generating a second sensitivity map (e.g., by updating the first sensitivity map) as output of the ML model responsive to the user responses to the first and second sequence of light stimuli as input into the ML model.

Description

METHODS AND SYSTEMS FOR VISUAL FIELD TEST STIMULI PRESENTATION AND SENSITIVITY MAP ESTIMATIONFIELD

[0001] An embodiment of the disclosure relates to a method and system for optimizing a visual field test using a machine learning (ML) model to accurately and efficiently generate a sensitivity map of a user’s visual field. Other embodiments are also described.BACKGROUND

[0002] A person’s visual field refers to a visual area in which objects are visible at the same moment while the person’s gaze is fixed in a given direction. Generally, a human’s visual field for each eye may be approximately 60° inward from a vertical meridian of the eye, to 100° outward from the vertical meridian, and may be approximately 70° above a horizontal meridian of the eye and 80° below the horizontal meridian. Age or injury of a person’s eyes may lead to loss of peripheral vision, thereby reducing the field of view. In addition, diseases, such as glaucoma, may result in visual field loss, and in some cases loss of vision entirely.SUMMARY

[0003] A person’s visual field may be checked to determine an amount of vision loss and any acutely affected areas (e.g., based on age, disease, or injury), in order to provide a visual health care provider with information to provide a visual diagnosis and treatment. For example, a visual field test may be administered in which a health care provider sits directly in front of a subject, asking the subject to say when the provider’s hand comes into view while one eye is closed, and the provider moves a hand within a space in front of the subject. Such a manual test may only provide general boundaries of the subject’s visual field.

[0004] Another visual field test may include the presentation of light stimuli at specific locations and intensity (strength), and subsequent computation of a subject’s eye sensitivity. This type of test is traditionally done using handcrafted algorithms such as those in the Swedish Interactive Thresholding Algorithm (SITA) family or Zippy Estimation by Sequential Testing (ZEST). While these algorithms may produce reliable results, the rules used to design them are handcrafted and subject to several assumptions, and therefore may not be optimized between subjects. For instance, light stimuli may normally be presented in a staircase fashion at each possible location within a visual field test area (e.g., each location’s intensity dropping by 1 decibel (dB) each time light stimulus is provided at that location and the subject provides a response indicating that the light stimulus was seen by the subject), and following predefined rules, such as step size, number of crossing, etc. This formation leads to a trade-off between accuracy of a subject’s sensitivity and speed at which the test is administered, where a faster test can be achieved only by allowing a larger error in the estimation of the eye sensitivity. A longer examination, however, may be less desirable because it requires the subject’s attention for an extended period of time. While these traditional methods have handcrafted rules where testing at specific locations may be dependent on other locations predicted sensitivity these rules have been derived fromobservation of limited data and are therefore not optimized between different subjects. Therefore, there is a need to optimize light stimuli presentation and eye sensitivity map estimation during a visual field test.

[0005] The present disclosure provides a method and system for visual field test stimuli presentation during a visual field test and sensitivity map estimation using a machine learning (ML) model. In particular, a visual field test system may present a first sequence of light stimuli through a set of light emitters that are each located at different locations within a visual field test area, and responsive to each light stimulus, a user response may be received, where the user responses may be voluntary responses, such as a user input through an input device or involuntary responses, such as a user’s pupil constricting due to seeing a stimulus. As another example, a user response may be an absence of user input (absence of voluntary or involuntary responses) due to a particular stimulus. The system may generate a sensitivity map of a visual field of the user as output of a ML model responsive to the light stimuli sequence and user responses (stimulus / responses sequence) as input into the ML model. The system may also determine based on the sensitivity map generated from the given stimulus / response sequence a subset of light emitters to present a subsequent sequence of light stimuli. The system may also determine light intensity for each of the light stimuli based on a given stimulus / response sequence. The system may repeat these operations to further refine (update) the sensitivity map, whereby previous user responses to light stimuli may be used to update the sensitivity map. As a result, the ML model may provide the system to optimize intensity of and number of light stimuli presentation and computation of a resulting sensitivity map, without being subject to predefined handcrafted rules.

[0006] According to one aspect of the disclosure includes a method for performing a visual field test, the method including: presenting, through a set of light emitters that are each located at different locations within a visual field test area, a first sequence of light stimuli;receiving, responsive to each light stimulus of the first sequence of light stimuli, a user response; generating a first sensitivity map of a visual field of the user as output of a machine learning (ML) model responsive to the user responses to the first sequence of light stimuli as input into the ML model; determining that one or more light emitters of the set of light emitters are to present a second sequence of light stimuli based on the first sensitivity map; presenting, through the one or more light emitters, the second sequence of light stimuli; receiving, responsive to each light stimulus of the second sequence of light stimuli, the user response; and generating a second sensitivity map of the visual field of the user as output of the ML model responsive to the user responses to the first sequence of light stimuli and the second sequence of light stimuli as input into the ML model.

[0007] In one embodiment, each user response includes either a user input through an input device or an absence of the user input through the input device after a period of time from when a corresponding light stimulus is presented. In another embodiment, the second sequence of light stimuli includes fewer light stimuli than the first sequence of light stimuli. In another embodiment, the method further includes determining light intensities for the first sequence of light stimuli based on the locations of the set of light emitters and on an age of the user. In another embodiment, the light intensities for the first sequence of light stimuli are predefined. In another embodiment, the method further includes determining light intensities for the second sequence of light stimuli based on the sensitivity map, where at least some of the light intensities of the second sequence of light stimuli presented by the one or more light emitters are different than light intensities of the first sequence of light stimuli presented by the one or more light emitters.

[0008] In one embodiment, the first and second sensitivity maps include sensitivity parameters for the locations of the set of light emitters within the visual field test area, where the method further includes determining whether each sensitivity parameter of the firstsensitivity map for each location has an uncertainty value that is below an uncertainty threshold, where determining that the one or more light emitters are to present the second sequence of light stimuli includes determining that uncertainty values of sensitivity parameters for locations of light emitters other than the one or more light emitters of the set of light emitters are higher than the uncertainty threshold. In another embodiment, the first sensitivity map includes sensitivity parameters for the locations of the set of light emitters within the visual field test area, the method further includes determining light intensities for the second sequence of light stimuli based on sensitivity parameters for locations of the corresponding one or more light emitters. For example, the light intensities may be set to a strength indicated by the sensitivity map. In another embodiment, the method further includes, until uncertainty values of sensitivity parameters for all locations meet or exceed the uncertainty threshold, determining for any remaining locations with uncertainty values that are higher than the uncertainty threshold, that corresponding light emitters are to present a new sequence of light stimuli based on each sensitivity map previously generated during the performance of the visual field test, presenting, through the corresponding light emitters, the new sequence of light stimuli, receiving, responsive to each light stimulus of the new sequence of light stimuli, the user response, and generating a new sensitivity map based on at least some previous user responses. In another embodiment, the method further includes, responsive to sensitivity parameters for all of the locations having uncertainty values that are less than the uncertainty threshold, ceasing to present light stimuli through the light emitters; and providing a most recently generated sensitivity map. In another embodiment, the first and second light stimuli are presented while the user remains seated and engaged with a visual field test display.

[0009] According to another embodiment of the disclosure includes a visual field test system that has: a visual field test display with several of light emitters positioned at differentlocations within a visual field test area; at least one processor; and memory having instructions which when executed by the at least one processor causes the visual field test system to: sequentially emit light stimuli through a first set of light emitters of the several of light emitters in a first order, receive, responsive to each light stimulus from the first set of light emitters, a user response, determine that a second set of light emitters of the several of light emitters are to be used to emit light stimulus in a second order based on the user responses, where the second set of light emitters includes fewer light emitters than and at least one light emitter from the first set of light emitters, sequentially emit light stimuli through the second set of light emitters in the second order, receive, responsive to each light stimulus from the second set of light emitters, the user response, and provide a sensitivity map of a visual field of the user that is generated based on output of a ML model responsive to input based on the user responses to the emitted light stimuli.

[0010] In one embodiment, each user response includes either a voluntary user response, an involuntary user response, or an absence of either type of response after a period of time from when a corresponding light stimulus is presented. In another embodiment, the memory has further instructions to determine light intensities for the light stimuli emitted through the first set of light emitters based on the first order and on an age of the user. In another embodiment, the sensitivity map includes sensitivity parameters, where the memory has further instructions to determine, for each sensitivity parameter, an uncertainty value based on the user responses to the emitted light stimuli, where the sensitivity map is provided responsive to the uncertainty values of the sensitivity parameters being less than an uncertainty threshold.

[0011] In one embodiment, the ML model is a first ML model, and the sensitivity map is a first sensitivity map, where determining the uncertainty value includes: generating the first sensitivity map as output of the first ML model responsive to the user responses asinput; generating a second sensitivity map as output of a second ML model responsive to the user responses as input; and determining uncertainty values for corresponding sensitivity parameters based on differences between the first sensitivity map and the second sensitivity map. In another embodiment, the sensitivity map is a first sensitivity map, where determining the uncertainty value includes: generating the first sensitivity map as output of the ML model responsive to the user responses as input; generating a second sensitivity map as output of the ML model responsive to user responses to the emitted light stimuli through the first set of light emitters in the first order; and determining uncertainty values for corresponding sensitivity parameters based on differences between the first and second sensitivity maps. In another embodiment, determining the uncertainty value includes generating several of sensitivity maps as output of the ML model responsive to the user responses as input, where each of the several of sensitivity maps is different; and determining uncertainty values for corresponding sensitivity parameters based on differences between the several of sensitivity maps.

[0012] In one embodiment, the sensitivity parameters are associated with the several of light emitters, where the instructions to determine that the second set of light emitters are to be used to emit light stimulus comprises instructions to determine light intensities for the second set of light emitters responsive to corresponding uncertainty values of corresponding sensitivity parameters being greater than the uncertainty threshold. In another embodiment, the light intensities for the second set of light emitters may be based on (or correspond to) sensitivity parameters of the (most recently) generated sensitivity map that correspond to the second set of light emitters. In another embodiment, the light stimuli is emitted while the user remains seated and engaged with the visual field test display. In another embodiment, providing the sensitivity map includes at least one of: displaying the sensitivity map of thevisual field of the user on a display; and transmitting the sensitivity map as an electronic message to an electronic device.

[0013] According to another embodiment of the disclosure, a method for performing a visual field test for a user, the method including: providing a first sequence of light stimuli at several of locations within a visual field test area, where at least two light stimuli have varying light intensity; receiving, responsive to each light stimulus of the first sequence of light stimuli, a user response; determining a subset of locations of the several of locations and light intensities for a second sequence of light stimuli at the subset of locations based on output from a ML model responsive to input based on the user responses, where two or more light stimuli of the second sequence of light stimuli at two or more locations of the subset of locations have different light intensities with respect to each other and with respect to light intensities of two or more light stimuli of the first sequence of light stimuli at the two or more locations; providing the second sequence of light stimuli having corresponding light intensities at the subset of locations; receiving, responsive to each light stimulus of the second sequence of light stimuli, the user response; and generating a sensitivity map of a visual field of the user as output of the ML model responsive to input based on the user responses to the first sequence of light stimuli and the second sequence of light stimuli.

[0014] In one embodiment, each user response comprises either a user input through an input device or an absence of the user input through the input device after a period of time from when a corresponding light stimulus is provided. In another embodiment, the method further includes determining light intensities for the first sequence of light stimuli based on the several of locations and on an age of the user. In another embodiment, the sensitivity map is a first sensitivity map, where determining the subset of locations and light intensities includes: generating a second sensitivity map of the visual field that as the output of the ML model responsive to input based on the user responses to the first sequence of light stimuli,where the second sensitivity map comprises a sensitivity parameter for each location of the several of locations; and determining an uncertainty value for each sensitivity parameter, where the subset of locations are determined based on their corresponding uncertainty values being greater than an uncertainty threshold. In one embodiment, the light intensities may be based on sensitivity parameters of the second sensitivity map for corresponding locations of the subset of locations. In another embodiment, the ML model is a first ML model, where determining the uncertainty value includes: generating a third sensitivity map as output of a second ML model responsive to input based on the user responses to the first sequence of light stimuli; and determining the uncertainty value based on differences between the first and second sensitivity maps.

[0015] The above summary does not include an exhaustive list of all embodiments of the disclosure. It is contemplated that the disclosure includes all systems and methods that can be practiced from all suitable combinations of the various embodiments summarized above, as well as those disclosed in the Detailed Description below and particularly pointed out in the claims. Such combinations may have particular advantages not specifically recited in the above summary.BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The embodiments are illustrated by way of example and not by way of limitation in the figures of the accompanying drawings in which like references indicate similar elements. It should be noted that references to “an” or “one” embodiment of this disclosure are not necessarily to the same embodiment, and they mean at least one. Also, in the interest of conciseness and reducing the total number of figures, a given figure may be used to illustrate the features of more than one embodiment, and not all elements in the figure may be required for a given embodiment.

[0017] Fig. 1 is a block diagram of a visual field test system that uses a machine learning (ML) model to present one or more sequences of light stimuli during a visual field test and to generate a sensitivity map.

[0018] Fig. 2 is a flowchart of one embodiment of a process to perform a visual field test using a ML model.

[0019] Fig. 3 is a flowchart of one embodiment of a process to stop a visual field test upon a determination that the sensitivity map is optimized.

[0020] Fig. 4 shows several stages in which a number of light stimuli presented through light emitters in a visual field test display is reduced during a visual field test based on output of the ML model.DETAILED DESCRIPTION

[0021] Several embodiments of the disclosure with reference to the appended drawings are now explained. Whenever the shapes, relative positions and other embodiments of the parts described in a given embodiment are not explicitly defined, the scope of the disclosure here is not limited only to the parts shown, which are meant merely for the purpose of illustration. Also, while numerous details are set forth, it is understood that some embodiments may be practiced without these details. In other instances, well-known circuits, structures, and techniques have not been shown in detail so as not to obscure the understanding of this description. Furthermore, unless the meaning is clearly to the contrary, all ranges set forth herein are deemed to be inclusive of each range’s endpoints.

[0022] Reference throughout this specification to “one embodiment” or “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, the appearances of the phrases “in one embodiment” or “in an embodiment” in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0023] Embodiments of the visual field test system disclosed herein optimizes the selection of stimulus strength and locations presented during a visual field test given subject responses (e.g., seen or unseen) to previously presented stimuli as well as for the accurate estimation of the subject’s eye sensitivity considering the collected subject responses to the stimuli. Unlike traditional methods that use handcrafted predefined rules, the disclosed testing system herein is capable of modifying and optimizing the presented light stimuli to better estimate (e.g., higher accuracy of) the subject’s sensitivity, while optimizing (e.g., minimizing) the required light stimuli. The system may use a machine learning (ML) modelthat is trained using a large dataset of visual field data and simulation of subject responses. In particular, the visual field testing system described herein estimates a subject’s sensitivity map from a set of light stimuli locations with a given light intensity (or strength) presented to the subject and collected responses from the subject (e.g., seen or unseen responses), optimizes the presentation of future stimulus locations and strength considering what has already been presented and collected responses, in order to guarantee a given accuracy in the sensitivity map estimation while minimizing the number of stimuli needed to be presented, and determines when the sensitivity estimation has achieved sufficient accuracy and the presentation of stimuli may be stopped, considering the test finished.

[0024] Fig. 1 is a block diagram of a visual field test system (hereafter may be referred to as “system”) 100 that uses a ML model to present light stimuli during a visual field test and to generate a sensitivity map. As described herein, the system may be used by a subject (user or patient) and / or a visual health care provider (or administrator) while administering a visual field test to measure (estimate) the patient’s visual field. The system 100 includes a visual field test analyzer 103, a display 104, an input device 101, and a visual field test display (or monitor) 102. In one embodiment, the system may include more or less elements, such as having multiple displays or by not having a display at all.

[0025] In one embodiment, each of the elements of the system 100 may be separate (electronic devices), as shown. In which case, each of the elements may be communicatively coupled to each other, such as via a wired or wireless (e.g., BLUETOOTH) connection. In another embodiment, at least some of the elements may be part of one device (e.g., housed within a shared housing). For instance, the display 104 and / or the visual field test display 102 may be a part of the visual field test analyzer 103, such that all three elements are housed (or a part of) one enclosure (or housing).

[0026] The input device 101 may be any device that includes one or more inputs that are designed to receive user input. For instance, the input device may include one or more buttons, which when pressed by a user may send a control signal to the visual field test analyzer 103 indicating that user input has been received. In another embodiment, the input device may include one or more sensors that may be arranged to receive user input (for example a gesture input with one or more fingers of a subject). For example, the input device may include one or more microphones that may be arranged to capture sound of an ambient environment. The system 100 may be configured to perform a speech recognition algorithm to detect speech contained within the captured sound, and may be configured to detect user input, such as a user spoken response, in the detected speech. As another example, the system 100 may include one or more cameras that may be arranged to capture images with which the system may detect user input (e.g., one or more gestures) by performing image recognition. In some embodiments, the system 100 may be configured to detect voluntary or involuntary user behaviors as user responses using one or more sensors, which may serve as user input, as described herein.

[0027] The visual field test display (or monitor) 102 may be a device with one or more light emitters 105 for use in testing a person’s visual field. The light emitters 105 may be located at different locations within a visual field test area on (or in) the test display 102. In one embodiment, the light emitters may be any device that may be capable of presenting light stimuli at various light intensities, such as a light emitting diode (LED). Fig. 4 illustrates an example of a visual field test area that includes light emitters within four distinct quadrants. The test display 102 may include bowl-shaped visual field test area, and include a structure (e.g., a chin rest) that enables a user to engage the user’s face with the structure to place the user’s face within the bowl-shaped area during a visual field test. In which case, the system 100 may present light stimuli through the light emitters while a user is (remains)seated and engaged with (e.g., having the user’s chin engaged with the chin rest of) the visual field test display.

[0028] The display 104 may be designed to present (or display) digital images or videos. The display may use liquid crystal display (LCD) technology, light emitting polymer display (LPD) technology, or LED technology, although other display technologies may be used in other embodiments. In another embodiment, the display may be a touch- sensitive display screen that may be configured to sense user input as touches or taps on the screen, and in response produce one or more control signals. In some embodiments, the display may use any touch sensing technologies, including but not limited to capacitive, resistive, infrared, and surface acoustic wave technologies.

[0029] The visual field test analyzer 103 may be any type of (computing) electronic device that may be configured to perform a visual field test, as described herein. For example, the analyzer 103 may be a desktop computer, a laptop computer, etc. In one embodiment, the analyzer may be a portable electronic device (e.g., being handheld operable), such as a tablet computer, a smart phone, etc. In another embodiment, the analyzer 103 may be a stand-alone electronic device that may be capable of providing a visual field test. In which case, the analyzer may include the test display 102 and display 104 built into the one device, having the input device 101 connected to the analyzer, thereby allowing a patient to conduct a visual field test and provide user input.

[0030] The visual field test analyzer 103 includes hardware (e.g., one or more processors, etc.) that allow the analyzer to perform a visual field test. For instance, the analyzer 103 may include a controller 106 and memory 107. The controller 106 may be a special-purpose processor such as an application-specific integrated circuit (ASIC), a general- purpose microprocessor, a field-programmable gate array (FPGA), a digital signal controller, or a set of hardware logic structures (e.g., filters, arithmetic logic units, and dedicated statemachines). The controller may be configured to perform visual field test operations. More about the operations that may be performed by the controller 106 are described herein. The memory 107 may be any type of non-transitory machine-readable storage medium that may include instructions which when executed by one or more processors (e.g., controller 106) may cause the (e.g., visual field test system 100 of the) visual field test analyzerl03 to perform a visual field test, as described herein. Examples may include read-only memory, random-access memory, CD-ROMS, DVDs, magnetic tape, optical data storage devices, flash memory devices, and phase change memory.

[0031] In one embodiment, at least some of the elements of the visual field test system 100 may be a part of a remote electronic device. For instance, at least a portion of the visual field test analyzer 103 may be one or more electronic devices (e.g., electronic servers) that are remote with respect to the input device, visual field test display 102, and / or the display 104. In which case, instructions stored within the memory 107 may be a part of and stored in a remote server, which may provide the instructions (e.g., to emit light stimuli) through the visual field test display 102 and receive responses through the input device 101 over a network (e.g., the Internet).

[0032] The memory 107 includes a visual field test software program 108, visual field test data 109, and one or more visual field test models 110. The visual field test software program 108 may be any type of software program, which when executed by the controller 106 may initiate and conduct a visual field test, as described herein. For example, the visual field test software program 108 may conduct a test, using the visual field test model 110, by causing one or more light emitters 105 to present (emit) light stimulus and may receive a user response to the light stimulus through the input device 101 (e.g., one or more sensors of the system 100), where a user response may include user input through the input device (e.g., a user selection of a button) or may be an absence of the user input through the input deviceafter a period of time from when a last (corresponding) light stimulus is presented. The visual field test software program 108 may use the visual field test model 110 to optimize the presentation of light stimuli and to generate an accurate sensitivity map of the user’s field of view. More about at least some of the operations performed by the software program are described herein.

[0033] The visual field test data 109 may include data for conducting a visual field test. The test data may include (e.g., predefined) light intensities for at least one sequence of light stimuli to be presented through the light emitters 105 in a particular order. For example, the test data may be based on one or more characteristics of a user to which the visual field test is to be performed. As an example, the test data 109 may indicate light intensities for particular locations within the visual field test area at which light stimuli is to be presented in a particular order (through one or more light emitters at that location) for a particular age of the user. In one embodiment, light (or luminous) intensity may be stored as a value (e.g., in dB). Such data may be presented initially to a user during a visual field test, user responses from which the visual field test software program may optimize the visual field test and estimate the user’s sensitivity map, as described herein. Test data 109 may be based on other characteristics, such as known visual maladies of the user (e.g., whether the user has glaucoma).

[0034] The visual field test data 109 may include one or more visual field tests conducted by one or more users and their corresponding user responses. For instance, the data 109 may include light intensities of a sequence of light stimuli (at their particular locations within the visual field test display 102) and corresponding user responses to the light stimuli for a given test. For example, the field test data may include a table (e.g., data structure) that associates light intensities of particular light emitters 105 with their corresponding user responses, indicating whether the user saw the light stimuli of the light emitters at thatparticular light intensity or not. In another embodiment, the visual field test data 109 may include sensitivity maps generated by the visual field test software program 108 based on user responses (and / or their corresponding sequence of light stimuli) in a given test. Sensitivity maps may include numerical values at locations within a plot that correspond to the locations of light emitters within the test display 102. In one embodiment, the numerical values may represent light intensity visible to the user based on user input (e.g., receiving user input in response to light stimuli for a given intensity). In which case, the plot may include numerical values ranging from 0 dB, meaning that the user was unable to see the brightest intensity of a light stimulus, to 50 dB, meaning that the user was able to see an intensity multiple times (e.g., 100,000 times) dimmer than a contrast between a background (e.g., of the visual field test display area) and a brightest intensity possible of the system. In another embodiment, the sensitivity map may be any type of plot (e.g., indicated by density of pixels) that may indicate visual acuity within the visual field test area. In some embodiments, at least some of the data may be simulated. As described herein, the visual field test software program 108 may use at least some of the visual field test data 109 to train and / or validate one or more visual field test models 110.

[0035] The visual field test model (which may be referred to as “model”) hereafter) 110 may be configured to generate an estimation of the user sensitivity as a sensitivity map based on one or more sequences of light stimuli presented during a visual field test and / or user responses to at least some of those stimuli. In one embodiment, the model may be a ML model that may be trained to generate sensitivity maps as output responsive to test data (e.g., light stimuli and / or user responses to the stimuli) as input. In particular, the model 110 may be trained using at least some of the visual field test data 109, including information about presented stimuli locations and corresponding intensities, user responses to those stimuli, and / or the resulting sensitivity maps (as ground truths). To train the model, the system 100may utilize supervised training techniques using labeled visual field training data. In which case, the model may be a neural network, such as a convolution neural network (CNN). In another embodiment, the model may be any type of machine learning model, such as a Random Forest, a Randomized Search, a recursive neural network, or a regression model. In another embodiment, other types of training approaches may be used, such as reinforcement learning. More about training the model is described herein.

[0036] In one embodiment, the system 100 may be configured to optimize the generation of a sensitivity map. For example, the model may be trained to generate one or more sequences of light stimuli at various light intensities, where user responses to those sequences may be used to generate a more optimized sensitivity map. More about sensitivity optimization is described herien.

[0037] In one embodiment, the memory 107 may include multiple models 110. For example, each model may be different (e.g., trained differently), where each model may be trained to generate sensitivity maps responsive to input based on user responses during a visual field exam. As described herein, multiple models may allow the system 100 to better determine the accuracy of data generated from user responses, thereby providing the system with a more optimized visualization of the user’s sensitivity strength.

[0038] Figs. 2 and 3 are flowcharts of processes 200 and 300, respectively, for performing a visual field test operations that may include presenting light stimuli and generating a sensitivity map using the visual field test model 110. For instance, at least some of these operations may be performed by the controller 106 of the (e.g., visual field test analyzer 103 of the) visual field test system 100. In particular, at least some of the operations may be performed by the visual field test software program 108 that may be executed by the controller 106.

[0039] In one embodiment, the processes 200 and / or 300 may be performed while a user is seated and engaged with the visual field test display 102, and in response to receiving instructions to conduct the test. In particular, instructions to conduct the test may be received from an administrator through one or more input devices. For example, the administrator may select one or more UI items displayed on the display 104 to start the exam. Once started, the analyzer 103 may perform at least some of the operations of the processes to generate a sensitivity map of the user who is taking the test. In another embodiment, the visual field test may start once an initial user input is received through the input device 101 (e.g., the user clicking a button for the first time).

[0040] Turning to Fig. 2, this figure shows a flowchart of one embodiment of the process 200 to perform an optimized visual field test using a visual field test model 110. The process 200 begins by the (e.g., visual field test software program 108 of the) controller 106 presenting, through a set of light emitters 105 that are each located at different locations within a visual field test area of the visual field test display 102, a sequence of light stimuli (at block 201). In particular, the controller 106 may cause one or more light emitters 105 to emit one or more light stimuli in a given order. For instance, the sequence of light stimuli may be systematically emitted through the set of one or more light emitters 105 over a period of time. In some embodiments, when two or more light emitters 105 emit light stimuli, each may emit their respective light stimulus over another period of time to allow a subject time to see the stimulus and provide a response. In one embodiment, at least some of the light emitters may emit light at varying light intensity. The controller 106 may be configured to determine light intensities and / or an order for this sequence of light stimuli based on various information. For instance, the controller may determine light intensities based on locations of the light emitters (e.g., based on the light emitters’ locations within the visual field test area). As another example, the controller may determine intensities based on characteristics of theuser, such as age, gender, health, etc., of the user who is conducting the exam. In one embodiment, the initial sequence of light stimuli (order and / or light intensity) presented by the system 100, which may be a sequence emitted immediately after a start of a visual field test, may be predefined (e.g., retrieved by the controller 106 from the visual field test data 109 stored in memory 107) and may be based on one or more of the characteristics of the user. In some embodiments, the first sequence of light stimuli may be presented sequentially and in a particular order through all of the light emitters 105.

[0041] The controller may receive, responsive to each light stimulus of the first sequence of light stimuli, a user response (at block 202). For instance, as each light emitter emits its corresponding light stimuli at a particular light intensity, the controller 106 may receive user input through the input device 101. In one embodiment, a user response may include a voluntary user response, such as a user input through the input device 101, or an involuntary user response that may include particular user behaviors or responses that the user may not be able to control voluntarily, such as a contraction of a user’s pupil in response to light stimuli. In one embodiment, an involuntary response may be detected by one or more sensors of the system 100, such as a camera of the input device, which may be arranged to capture one or more images of the user’s eyes in the case of detecting contraction of the user’s pupil. In another embodiment, a received user response may be a collection of one or more user responses to a given stimulus, such as the system 100 receiving input through the input device 101 and one or more involuntary user responses, such as eye-gaze, changes to pupil diameter over a given time, etc.

[0042] In another embodiment, a user response may be an absence of receiving a voluntary or involuntary response within a period of time from when the light stimulus occurs. For example, the absence may be of user input through the input device after a period of time from when a corresponding light stimulus is presented. In one embodiment, thecontroller 106 may keep track of user responses to light stimuli in a binary fashion with respect to their corresponding light stimuli. For instance, each time a light emitter emits light, the controller may wait a period of time to determine whether a user input is received indicating that the user has seen the stimulus. Once a user response is received, the controller may indicate (e.g., in a data structure) that the user has seen the light stimuli (e.g., indicated as a “1”) with respect to (and associated with) that light stimuli. The controller may then proceed to turn on a next light emitter and wait for a user response. If, however, no user response is received within a period of time, the controller may indicate that the user did not see the light stimuli (e.g., indicated as a “0” representing an unseen response). The controller 106 may sequentially go through the sequence of light stimuli, until all light stimuli in the sequence are presented and corresponding user responses are received.

[0043] The controller 106 generates a sensitivity map of a visual field of the user as output of the visual field test model 110 responsive to the user responses as input into the model (at block 203). In particular, the controller 106 may input the user responses into the model 110, which may output the sensitivity map. The input may also include the sequence of light stimuli, which may include the order of stimuli through the light emitters and / or the light intensity of the stimuli into the model. For example, the controller may include a table, in which each sequential row indicates the order and location within the visual field test area where light stimuli is presented, the light intensity of the light stimuli, and / or the user response (0 or 1). As a result, the controller 106 may compute an estimation of the user sensitivity using the trained visual field test model 110 and based on at least some stimuli presented and responses collected. In one embodiment, the controller may input all stimuli information and user responses to generate the sensitivity.

[0044] The controller 106 determines whether the sensitivity map is optimized (at decision block 204). In particular, the controller 106 may determine whether each or at leastsome sensitivity parameters of the sensitivity map have uncertainty that is less than an uncertainty threshold. The controller may be configured to determine an uncertainty value for each location within the visual field test area. The uncertainty value may represent an indication of certainty with the sensitivity prediction, where uncertainty may be a value between zero (e.g., certain) and one (e.g., uncertain). In another embodiment, the uncertainty value may be determined for each sensitivity parameter of the sensitivity map, which may be associated with a particular location (e.g., a particular light emitter) within the visual field test area. The controller may be configured to determine that the sensitivity map is optimized in response to all uncertainty values being less than (or equal to) the threshold. In another embodiment, optimization may be determined responsive to a percentage (e.g., 90%) of uncertainty values being less than the threshold. In some embodiments, the percentage of uncertainty values for determining optimization and / or the uncertainty threshold may be (e.g., user) configurable. For example, a higher uncertainty threshold may result in faster optimization of the sensitivity map, as opposed to a lower uncertainty threshold. More about uncertainty values is described in Fig. 3.

[0045] If uncertainty is low, and therefore the sensitivity map may be optimized, the process 200 may proceed to provide the sensitivity map (at block 209). For example, the system 100 may display the sensitivity map (generated at block 203) of the visual field of the user on the display 104. In which case, the system 100 may stop the visual field test. To stop the test, the system may cease presenting light stimuli through the light emitters 105 and / or may output a notification (e.g., an audible alert or a visual alert on the display 104) informing the user that the visual field test is complete. In another embodiment, the controller 106 may transmit the sensitivity map as an (or in an) electronic message to an electronic device (e.g., through electronic mail). In another embodiment, the controller 106 may store the results ofthe test (including the light stimuli and user responses) in the visual field test data 109, which may be used to further train the visual field test model 110, as described herein.

[0046] If, however, the sensitivity map is not optimized, the controller 106 may determine that a reduced set of light emitters (that may be from the previous set of light emitters that were used to present a (e.g., most) previous sequence of light stimuli) are to present an additional sequence of light stimuli based on the sensitivity map (at block 205). In particular, the controller 106 may determine that one or more light emitters of a set of light emitters, which may be emitters that presented the last or one or more previous sequences of light stimuli, to present another sequence of light stimuli. In another aspect, at least one of the light emitters that is to present another light stimulus may not have presented light stimulus in a previous (most recent) sequence of light stimuli, but may have presented light stimulus for another previously presented sequence.

[0047] In one embodiment, the controller may determine that a light emitter is to emit light stimulus based on whether the uncertainty value associated with the light emitter is higher than the uncertainty threshold. For example, upon determining that a location has an uncertainty value that is higher than the uncertainty threshold, the controller 106 may determine that the light emitter at that location is to emit light stimuli. In one embodiment, the uncertainty value may indicate a prediction of accuracy for a corresponding generated sensitivity parameter (or value) for that location by the visual field test model 110. If the uncertainty value is high, that sensitivity parameter may therefore be inaccurate. In one embodiment, the operations of block 204 and 205 may be performed at the same time. In particular, to determine whether the sensitivity map is optimized, the controller may determine whether sensitivity parameters of the sensitivity map have corresponding uncertainty values that are higher than the uncertainty threshold. Upon determining that some(or one or more) uncertainty values are higher than the threshold, the controller may determine that the corresponding locations need to be retested.

[0048] In one embodiment, the additional sequence of light stimuli may include fewer light stimuli than the previous sequence of light stimuli. In this way, each time another sequence of light stimuli needs to be presented in order to retest the user, less stimuli may be used. As a result, the test accuracy may focus on areas of the user’s visual field that may require more attention, while at the same time reducing the time necessary to perform the test since less light stimuli is used in every following iteration. In another embodiment, a subsequent sequence may test locations of the visual field test area that may not have been previously tested (by one or more previous sequences of light stimuli). In which case, the controller 106 may be configured to determine uncertainty of locations within the field regardless of whether that field has most recently been tested in order to determine whether recent testing may have increased uncertainty at other locations. If so, those locations may again be tested by the system 100.

[0049] In one embodiment, along with (or in lieu of) determining the additional sequence of light stimuli, the controller 106 may determine corresponding light intensity for the additional sequence of light stimuli. For example, the visual field test model 110 may be trained to output corresponding light intensity for the additional sequence of light stimuli responsive to input, such as user responses / light stimuli, as described herein. In one embodiment, the light intensity may be based on the sensitivity value predicted by the stimuli / response sequence collected up to that point. For example, the controller may adjust the light intensity for a given light emitter in the reduced set based on the currently predicted sensitivity map, offering an optimized guess of the stimulus intensity that will reduce the uncertainty of the collected location in the most significant manner upon collection of a response. In particular, a given light intensity for the additional sequence of light stimuli maybe set to (or based on) a corresponding sensitivity parameter of the most recently generated sensitivity map.

[0050] The controller 106 presents, through the reduced set of light emitters, the additional sequence of light stimuli (at block 206). In which case, the controller sequentially emits light stimuli through a set of light emitters in an order to present an additional sequence of light stimuli, where each of the light emitters emits light at a respective light intensity. The order in which the new light stimuli is presented may be different or the same as the previous order of light stimuli. In one embodiment, the presentation of the additional sequence of light stimuli may be performed seamlessly, such that the user taking the exam may not distinguish between the different sequences of light stimuli (e.g., when one sequence ends, and another begins). In another embodiment, the system 100 may notify the user that an additional round of testing is going to begin before the presentation of the additional light stimuli. The controller may receive, responsive to each light stimulus of the additional sequence of light stimuli, a set of user responses (at block 207).

[0051] The controller 106 generates a new sensitivity map of the visual field of the user as output of the ML model responsive to user responses (e.g., 1) the additional sequence of light stimuli and / or their respective user responses and / or 2) at least some light stimuli of past sequence(s) of light stimuli and their respective past user responses) as input into the ML model (at block 208). For instance, to generate a new sensitivity map (or update the previously generated map), the controller may include all user responses from all past and present sequences of light stimuli during this visual field test with this user. As a result, the ML model may be able to generate a more accurate sensitivity map due to the additional user responses.

[0052] The controller 106 returns to decision block 204 to determine whether the sensitivity map is optimized. Again, the controller may determine whether any sensitivityparameters of the recently generated sensitivity map have uncertainty values that are higher than the uncertainty threshold. Thus, until uncertainty values of sensitivity parameters for all locations meet or are below the uncertainty threshold, the controller 106 may iteratively repeat the operations of 204-208, continuing to reduce the set of light emitters that present light stimuli, until there is no uncertainty within the generated sensitivity map. In particular, the controller may determine for any remaining locations with uncertainty values that are higher than the uncertainty threshold, that corresponding light emitters are to present a new sequence of light stimuli based on previously generated sensitivity maps, present, through the corresponding light emitters, the new sequence of light stimuli, receiving user responses, and then generating a new sensitivity map as well as a new uncertainty map based on at least some (or all) previous user responses. In one embodiment, for each iteration, the controller 106 may determine varying light intensities. In some embodiments, at least one light emitter between iterations may have varying light intensities. Once there is no uncertainty (e.g., all locations have uncertainty values lower than the uncertainty threshold), the controller 106 may provide the sensitivity map, which may have been generated by the ML model 110 responsive to input based on user responses to at least some past light stimuli, as described herein. In particular, responsive to the sensitivity parameters for all of the locations having uncertainty values meeting or below the uncertainty threshold, the controller may cease to present light stimuli through the light emitters 105 and may provide a most recently generated sensitivity map.

[0053] Fig. 3 is a flowchart of one embodiment of a process 300 to stop a visual field test upon a determination that the most recently generated sensitivity map is optimized for the user conducting the test. In one embodiment, at least some of the operations of process 300 may be performed while a visual field test is being administered through the visual field test system 100. For instance, at least some of the operations in this process may be performedwith respect to operational blocks 204 and 205 of process 200 in Fig. 2 to determine whether a sensitivity map is optimized, and either determining a new sequence of light stimuli to be presented when not optimized or providing the sensitivity map when optimized. In which case, this process 300 may begin once an initial sensitivity map is generated (at block 203 in process 200). In another embodiment, at least some of these operations may be performed for one or more (or each) following generated sensitivity map. In some embodiments, at least some of the operations of process 300 may be performed by the model 110. For instance, the model 110 may be trained to generate a sensitivity map and uncertainty values, and based on those uncertainty values (with respect to an uncertainty threshold), either continue to test a subject (by producing additional sequences of light stimuli) or outputting the (most recent) sensitivity map.

[0054] The process 300 begins with the controller 106 determining for (e.g., a location associated with) a sensitivity parameter of a generated sensitivity map an uncertainty value (at block 301). In particular, sensitivity parameters may be associated with light emitters at different locations within the visual field test area. In one embodiment, the uncertainty value may be determined based on user responses to the emitted light stimuli. For example, the value may be determined based on deployment of several visual field test models through cross-validation or in a model ensemble manner. As described herein, multiple test models may be trained using different test data (or parameters) 109. These models may produce different results on the same set of presented stimuli and user responses to those stimuli. In which case, the controller 106 may be configured to generate a first sensitivity map as output of a first ML model from the models 110 responsive to at least some visual field test data 109 as input (e.g., user responses, the last series of light stimuli presented, etc.), and generate a second sensitivity map as output of a second ML model from the models 110 responses to at least some of the same visual field test data as input, whereboth ML models may be trained differently (e.g., using different training data). The controller 106 may be configured to determine an uncertainty value for a sensitivity parameter by comparing the sensitivity parameter of the first sensitivity map with a corresponding sensitivity parameter of the second sensitivity map, such as looking at the difference. Based on the difference, the controller may determine the uncertainty value. For instance, the uncertainty value may be a percentage difference between the two parameters. In another embodiment, the controller 106 may determine the uncertainty value based on a comparison between two or more corresponding sensitivity parameters that are generated using two or more different ML models. As a result, the variability of estimation produced by multiple ML models may be used to determine the accuracy and therefore uncertainty of a sensitivity parameter.

[0055] As another example, uncertainty may be determined based on variance with previous sensitivity map estimations. For example, the process 300 may be performed two or more times based on whether the controller determines that additional reduced sets of light emitters need to emit additional sequence of light stimuli. In which case, multiple sensitivity maps may have been generated based on past user references to past sequences of light stimuli. The results produced by the visual field test model 110 as more stimuli of different light intensities are shown at different locations tend to converge to a solution where the difference in the estimation after presentation of additional stimuli is minimal (e.g., below a threshold). In which case, the controller 106 may be configured to determine an uncertainty value for a sensitivity parameter by generating a sensitivity map as output of the ML model 110 response to user responses to a previous set of light stimuli as input, and generating a new sensitivity map as output of the ML model 110 responsive to user responses to the previous set of light stimuli and the most recent user responses to the most recent light stimuli. The controller may be configured to determine an uncertainty value for a sensitivityparameter based on differences between corresponding sensitivity parameters of both sensitivity maps. In one embodiment, the differences may be between a current sensitivity parameter and one or more previously generated corresponding sensitivity parameters (e.g., an average of previously generated parameters). The variance of estimates produced by the ML model before and after a new stimuli strength is presented can provide a value of the uncertainty of these estimations. In one embodiment, the uncertainty value may be based on the variance in a predicted sensitivity map that a new presentation in a given location may produce between collecting a possible positive response (e.g., seen) versus negative responses (e.g., unseen).

[0056] Another source of uncertainty may be determined by simulating multiple responses given to hypothetical additional presentation of stimuli within a range of values in consideration of strengths already presented. In particular, the controller 106 may be configured to simulate user responses based on a range of light intensities for a simulated sequence of light stimuli for a set of one or more light emitters. Based on the user responses (seen and unseen) that may be simulated, the controller may be configured to generate one or more simulated sensitivity maps. The controller may be configured to determine an uncertainty value based on a comparison between corresponding sensitivity parameters within the one or more simulated sensitivity maps. In one embodiment, the visual field test model 110 may perform the simulation, such that additional simulated responses may be passed to the same model to decide what it would predict next and their variance depending on simulated user responses. In which case, the visual field test model 110 may determine an uncertainty value based on a comparison between the simulated responses.

[0057] In one embodiment, previous user responses used for sensitivity map estimation may be altered to simulate the possibility of an incorrect response by the user. In which case, the controller 106 may be configured to generate multiple sensitivity maps usingthe ML model responsive to at least some altered user responses and / or at least some nonaltered user responses to presentation to past light stimuli. The controller may be configured to determine uncertainty based on differences between sensitivity parameters to altered user responses. In one embodiment, the degree or percentage of possible incorrect response used may be determined a priori or adapted to particular user behavior. As a result, the controller may be configured to monitor user response to light stimuli, and determine user behavior to particular light stimuli (e.g., location, light intensity, etc.) based on the monitored responses. From the monitored responses, the controller may be configured to alter user responses. The range of possible sensitivity estimations generated by this process may give a value of uncertainty.

[0058] In another embodiment, uncertainty may be derived from adjusting the visual field test model 110. For instance, the model may be a deep learning model in which dropout layers are employed to prevent overfitting to training data. In one embodiment, dropout layers may randomly nullify the contribution of some neurons in the network towards the next layer and leave unmodified other neurons. The randomness of this process may not guarantee that the same estimation is going to be produced by the same inputs, but it may provide a measurement of uncertainty in the model by running the estimation multiple times and considering the variability of the recorded results. Therefore, the controller may be configured to generate several of sensitivity maps as output of the visual field test model 110 responsive to the same user responses, where each of the sensitivity maps is different due to the dropout layers. The controller may determine an uncertainty value based on a comparison between corresponding sensitivity parameters of the several of sensitivity maps. For example, the uncertainty value may be a percentage of difference between a sensitivity parameter that uncertainty is being determined for and an average of the corresponding sensitivity parameters of the several sensitivity maps.

[0059] In one embodiment, uncertainty may be derived from one or more examples described herein. For example, the controller may determine multiple uncertainty values for a given sensitivity parameter, and determine uncertainty as an average of the value.

[0060] In another embodiment, uncertainty can be derived via other techniques. For example, an uncertainty value may be based on a consistency of user (voluntary and / or involuntary) responses. For example, the system 100 may generate an uncertainty value for a particular location over several (partial) iterations of process 300, where a voluntary response may always be received. In which case, when a user response to light stimulus is received for a particular location, the system 100 may generate a low uncertainty value (e.g., below a threshold). In another embodiment, an uncertainty value may be derived based on how close to a “decision threshold” a user response was made.

[0061] With an uncertainty value determined, the controller may determine whether another (or reduced) set of light emitters are to be used to emit light stimuli by determining light intensities for the reduced set responses to and based on corresponding uncertainty values of corresponding sensitivity parameters (e.g., being greater than an uncertainty threshold). As described herein, the controller may generate (or build) this reduced set one or more times in order to optimize the sensitivity map.

[0062] The controller determines whether the uncertainty value of the sensitivity parameter is below an uncertainty threshold (at decision block 302). In one embodiment, the uncertainty threshold may be a predefined threshold determined in a controlled environment (e.g., a laboratory). If not, meaning that uncertainty is high, the controller determines a light intensity for the light emitter at the location associated with the sensitivity parameter (at block 303). For example, the light intensity for the light emitter in a new sequence of light stimuli may be set (or based on) the corresponding sensitivity parameter (for the most recently generated sensitivity map). In one embodiment, the controller may determine lightintensity such that uncertainty at that location may be reduced. In other words, the intensity displayed at this location may be defined by the generated sensitivity map using the stimuli / user responses collected up to this point (e.g., based on previous strengths of that light emitter and / or other light emitters and / or user responses).

[0063] In another embodiment, the controller 106 may determine the intensity based on a light intensity model, which may output the intensity responsive to input, such as past intensities.

[0064] The controller 106 determines whether there are any additional sensitivity parameters to determine uncertainty for the sensitivity map (at decision block 304). Specifically, the controller determines whether there are other sensitivity parameters that need to be analyzed for uncertainty. If so, the controller returns to block 301 to determine an uncertainty value for the additional sensitivity parameters. As a result, the controller may analyze each sensitivity parameter for uncertainty, and if there is uncertainty, will determine a new light intensity for a light stimuli for an associated light emitter. In which case, a new light intensity may be based on previous light intensities and uncertainty values of sensitivity parameters associated with corresponding light emitters. This allows the controller to create a new sequence of light stimuli to be presented through light emitters associated with locations having sensitivity parameters with high uncertainty values. Thus, the controller determines locations and light intensities for another sequence of light stimuli at those locations based on output of the ML model, which may be responsive to input based on user responses to a previous sequence of light stimuli.

[0065] In one embodiment, the controller may analyze each sensitivity parameter in the generated sensitivity map. In another embodiment, the controller may only analyze sensitivity parameters associated with light emitters through which the previous most recent light stimuli have been presented. As a result, the controller may only analyze a subset ofsensitivity parameters, reducing computational operations. As a result, the system 100 may generate a sensitivity map of a visual field as output of the ML model responsive to input based on the user responses to one sequence of light stimuli, may determine uncertainty for the (or at least some of the) sensitivity parameters, and then may determine a subset of locations based on their corresponding uncertainty values being greater than an uncertainty threshold.

[0066] In some embodiments, the order through which the controller goes through sensitivity parameters may be based on past stimuli / user responses. As described thus far, the controller may go through all (or a portion of) sensitivity parameters to determine whether corresponding light emitters are to emit additional light stimuli. In another embodiment, a determination of uncertainty may define one or more other light emitters to emit stimuli. For example, upon determining that one sensitivity parameter associated with one light emitter has a high uncertainty, the controller may determine that one or more other light emitters are to be used for an additional sequence of light stimuli. Such a determination may be output of the model 110 responsive to input of light stimuli / user responses. In this case, the system 100 may define what location(s) may be used to emit light stimuli in order to provide a highest reduction of uncertainty in response to user responses to that stimuli.

[0067] If the controller has gone through the uncertainty values of (e.g., all) sensitivity parameters for the generated sensitivity map (or has gone through uncertainty values of a subset of sensitivity parameters), the controller determines whether there are any light stimuli to be presented (at decision block 305). In particular, the controller is determining whether all uncertainty values associated with the sensitivity map are less than the uncertainty threshold. If not, meaning that at least one light stimulus is to be presented, the controller 106 presents a light stimulus through a light emitter at its determined light intensity (at block 306). In particular, after going through each sensitivity parameter anddetermining a light intensity, the controller 106 may create a new sequence of light stimuli, which indicate locations of light emitters through which the stimuli is to be presented and at determined light intensities. Thus, at this point the controller may begin to present the new series of light stimuli. The controller receives a user response (at block 307). The controller 106 determines whether there are any additional light stimuli (e.g., in the newly created sequence of light stimuli) to be presented (at decision block 308). If so, the controller may present the next light stimuli in the sequence and await for a user response. Thus, the controller may run through the entire light stimuli sequence. Once all light stimuli have been presented, the controller 106 generates a new sensitivity map of the visual field of the user using at least some current and past user responses (at block 309). In one embodiment, the controller may input all past visual field test data (e.g., associated with this user’s visual field exam) into the visual field test model 110. For instance, the controller 106 may include a data structure of visual field test data including the series of light stimuli (e.g., locations of light emitters), associated light intensities, and / or corresponding user responses.

[0068] The controller 106 returns to block 301 to repeat the process 300 until there are no longer any light stimuli to be presented. In particular, the controller may continue to repeat at least some of the operations of process 300 until all uncertainty values of the sensitivity map meet or exceed the uncertainty threshold. In response to all values being less than the threshold, the controller 106 may proceed from decision block 305 to provide the most recently generated sensitivity map (at block 310). For example, at this point the controller 106 may cease presenting light stimuli, alert the user that the test is complete, and / or provide the most recent map (e.g., transmitting the map electronically and / or displaying it on display 104.

[0069] In another embodiment, at least some of the operations described herein may be performed by the model 110. For instance, the model may be trained to determineuncertainty for at least some sensitivity parameters of a generated sensitivity map, and in response produce any additional sequences of light stimuli based on the determined uncertainty.

[0070] Fig. 4 shows several stages 401-403 in which the number of light stimuli presented through light emitters in a visual field test display is reduced during a visual field test based on output of the ML model. Each of the stages includes the visual field test display 102 with light emitters 105 arranged (e.g., in different locations) within a visual field test area 406 that includes four quadrants. As shown, the light emitters 105 are distributed evenly with respect to all of the quadrants. In another embodiment, the light emitters may be arranged differently in one or more quadrants. In one embodiment, the visual field test area may be bowl-shaped, where a patient may place the patient’s face inside the area. In one embodiment, each of the stages illustrates a different series of light stimuli being presented to a patient during a visual field examination.

[0071] The first stage 401 shows that each of the light emitters 105 is presenting light stimulus 404, illustrated as being white circles. In particular, this stage is showing a sequence of light stimuli being presented sequentially through all of the light emitters 105. For instance, each light emitter may emit light stimulus at a particular light intensity and in a particular order, which may be defined in the visual field test data 109. In one embodiment, this stage may represent the first sequence of light stimuli, where each light stimulus may have a predefined light intensity (e.g., due to an age of the patient), which may be the initial sequence of stimulus presented to the patient at the beginning of the examination. In one embodiment, only a portion of light emitters may emit light stimuli at this initial stage.

[0072] The second stage 402 shows that only a reduced set of light emitters 105 have presented light stimuli, while others have not illustrated light stimuli, which is illustrated as a black circle 405. In this case, the controller 106 may have performed at least some operationsof processes 200 and 300 to receive user responses in response to the light stimuli in the first stage 401, generated a sensitivity map using the visual field test model 110, and determined whether sensitivity parameters associated with the (locations of the) light emitters 105 have uncertainty values that are below the uncertainty threshold. In this case, the majority of the sensitivity parameters associated with light emitters in the top right quadrant (and some in the top left and bottom right quadrants) have uncertainty values that are higher than the threshold. As a result, the controller 106 has determined a second sequence of light stimuli and has presented it through this reduced set. In one embodiment, the controller 106 may determine light intensities for this second sequence of light stimuli based on (e.g., one or more of) the previously generated sensitivity map, where all light intensities of the second sequence may be different from intensities of the initial sequence presented by the light emitters 105. In particular, the controller may input the initial sequence and user responses into the visual field model 110, which may in response output the second sequence of light stimuli having light intensities based on the previously generated sensitivity map (e.g., based on uncertainty values of the map). In some embodiments, the light intensities of at least some of the light stimuli for this reduced set may be different than light intensities of light stimuli in the initial (or any precedent) sequence for corresponding light emitters. For example, a light stimulus emitted by a light emitter in the second stage 402 may have a strength greater or than a previous light intensity for the light stimulus that was emitted through the light emitter at the first stage 401 (during the initial sequence of light stimuli). In another embodiment, the light stimulus step-size may be greater or less than 1 dB, such as 0.5 dB.

[0073] The third stage 403 shows an even further reduced set of light emitters that have presented light stimuli that are only contained within the top right quadrant. Again, the controller 106 may have performed at least some of the operations described herein responsive to user responses to the light stimuli presented in the second stage 402, and havefurther reduced the light emitters used to emit light stimuli. In one embodiment, after this stage, all sensitivity parameters of a newly generated sensitivity map may have uncertainty values that are less than the uncertainty threshold, and therefore the visual field test may end.

[0074] As described herein, the visual field test model 110 may be trained to generate a sensitivity map based on visual field test data administered during a visual field test.Machine learning techniques provide the ability to learn relationships and correlations from a large dataset and optimize the approach of stimuli presentation and sensitivity map estimation beyond what has been used traditionally.

[0075] In one embodiment, training the visual field test model 110 may rely on a collection of a large set of sensitivity maps for the resulting model to understand the complex relationship among tested locations in a large population. Such a dataset may be derived in three subsets: training data, used for training the machine learning model, validation data used to compute accuracy and performance metrics during the training process in order for the model to select an optimal set of parameters and configuration that would make the training model perform accurately in a separate dataset, and test dataset that is not accessed during training or validation but instead used to assess the final performance of the complete trained model. In one embodiment, datasets may not be mixed together during training.

[0076] The visual field test model is to be trained to estimate a sensitivity map using a collection of presented stimuli at diverse locations and strengths and user responses to that stimuli. The data (e.g., visual field test data 109) available to train the model may include sensitivity maps, light stimuli used to generate the maps, and / or user responses to that light stimuli, which may be from a large dataset. In one embodiment, in order to construct the visual field test model 110, the system 100 may simulate user responses (seen / unseen) to collections of given stimuli of variable number, location, and strength. For example, during simulation stimuli may be presented at random strengths and / or locations. In oneembodiment, the strengths may be selected within a given range of strengths, such as 0 dB to 50 dB. In some embodiments, the first stimuli levels presented may be selected as the agecorrective normative values for sensitivity given a subject’s age. User responses to the presented stimulus intensities may be simulated by comparing the intensity value with the ground truth sensitivity map, deriving a binary response (seen or unseen) for each tested location and intensity value, depending whether the user sensitivity would detect the presented stimuli. As a result, the model may learn the best estimation of the final sensitivity map when presented with a set of variable number of stimuli presented at different locations at variable strengths and order of presentation and the associated simulated user responses.

[0077] In one embodiment, the system may validate the model with the purpose of also estimating the strength of the stimulus to be presented at each given location considering the strength and location of previous stimuli presented and collecting user responses to them. For example, during validation of the model 110 the system may simulate the first presentation of age-corrected normative values before any response is collected. The predicted sensitivity map from the model may be shown as the next stimuli strength and locations to display, considering the sequence of presentations and responses collected. New predictions may be updated and used as the next stimuli to be shown as more stimuli are presented and use responses are collected. This may provide the closest value where user responses may oscillate between seen and unseen, which may provide useful information for sensitivity map estimation.

[0078] As previously explained, an embodiment of the disclosure may be a non- transitory machine-readable medium (such as microelectronic memory) having stored thereon instructions, which program one or more data processing components (generically referred to here as a “processor”) to perform operations, such as visual field test operations performed by the visual field test software application using the test model 110 to generate sensitivitymaps, as described herein. In other embodiments, some of these operations might be performed by specific hardware components that contain hardwired logic. Those operations might alternatively be performed by any combination of programmed data processing components and fixed hardwired circuit components.

[0079] While certain embodiments have been described and shown in the accompanying drawings, it is to be understood that such embodiments are merely illustrative of and not restrictive on the broad disclosure, and that the disclosure is not limited to the specific constructions and arrangements shown and described, since various other modifications may occur to those of ordinary skill in the art. The description is thus to be regarded as illustrative instead of limiting.

[0080] To aid the Patent Office and any readers of any patent issued on this application in interpreting the claims appended hereto, applicants wish to note that they do not intend any of the appended claims or claim elements to invoke 35 U.S.C. 112(f) unless the words “means for” or “step for” are explicitly used in the particular claim.

[0081] In some embodiments, this disclosure may include the language, for example, “at least one of [element A] and [element B] ” This language may refer to one or more of the elements. For example, “at least one of A and B” may refer to “A,” “B,” or “A andB.” Specifically, “at least one of A and B” may refer to “at least one of A and at least one of B,” or “at least of either A or B.” In some embodiments, this disclosure may include the language, for example, “[element A], [element B], and / or [element C] ” This language may refer to either of the elements or any combination thereof. For instance, “A, B, and / or C” may refer to “A,” “B,” “C,” “A and B,” “A and C,” “B and C,” or “A, B, and C .”

Claims

CLAIMSWhat is claimed is:

1. A method for performing a visual field test, the method comprising: presenting, through a set of light emitters that are each located at different locations within a visual field test area, a first sequence of light stimuli; receiving, responsive to each light stimulus of the first sequence of light stimuli, a user response; generating a first sensitivity map of a visual field of the user as output of a machine learning (ML) model responsive to the user responses to the first sequence of light stimuli as input into the ML model; determining that one or more light emitters of the set of light emitters are to present a second sequence of light stimuli based on the first sensitivity map; presenting, through the one or more light emitters, the second sequence of light stimuli; receiving, responsive to each light stimulus of the second sequence of light stimuli, the user response; and generating a second sensitivity map of the visual field of the user as output of the ML model responsive to the user responses to the first sequence of light stimuli and the second sequence of light stimuli as input into the ML model.

2. The method of claim 1, wherein each user response comprises either a user input through an input device or an absence of the user input through the input device after a period of time from when a corresponding light stimulus is presented.

3. The method of claim 1, wherein the second sequence of light stimuli includes fewer light stimuli than the first sequence of light stimuli.

4. The method of claim 1 further comprising determining light intensities for the first sequence of light stimuli based on the locations of the set of light emitters and on an age of the user.

5. The method of claim 4, wherein the light intensities for the first sequence of light stimuli are predefined.

6. The method of claim 1 further comprising determining light intensities for the second sequence of light stimuli based on the sensitivity map, wherein at least some of the light intensities of the second sequence of light stimuli presented by the one or more light emitters are different than light intensities of the first sequence of light stimuli presented by the one or more light emitters.

7. The method of claim 1, wherein the first and second sensitivity maps include sensitivity parameters for the locations of the set of light emitters within the visual field test area, wherein the method further comprises determining whether each sensitivity parameter of the first sensitivity map for each location has an uncertainty value that is below an uncertainty threshold, wherein determining that the one or more light emitters are to present the second sequence of light stimuli comprises determining that uncertainty values of sensitivity parameters for locations of light emitters other than the one or more light emitters of the set of light emitters are higher than the uncertainty threshold.

8. The method of claim 7 further comprising, until uncertainty values of sensitivity parameters for all locations meet or exceed the uncertainty threshold, determining for any remaining locations with uncertainty values that are higher than the uncertainty threshold, that corresponding light emitters are to present a new sequence of light stimuli based on each sensitivity map previously generated during the performance of the visual field test, presenting, through the corresponding light emitters, the new sequence of light stimuli, receiving, responsive to each light stimulus of the new sequence of light stimuli, the user response, and generating a new sensitivity map based on at least some previous user responses.

9. The method of claim 8 further comprising, responsive to sensitivity parameters for all of the locations having uncertainty values that are less than the uncertainty threshold, ceasing to present light stimuli through the light emitters; and providing a most recently generated sensitivity map.

10. The method of claim 1, wherein the first sensitivity map includes sensitivity parameters for the locations of the set of light emitters within the visual field test area, wherein the method further comprises determining light intensities for the second sequence of light stimuli based on sensitivity parameters for locations of the corresponding one or more light emitters.

11. The method of claim 1, wherein the first and second light stimuli are presented while the user remains seated and engaged with a visual field test display.

12. A visual field test system, comprising: a visual field test display with a plurality of light emitters positioned at different locations within a visual field test area; at least one processor; and memory having instructions which when executed by the at least one processor causes the visual field test system to: sequentially emit light stimuli through a first set of light emitters of the plurality of light emitters in a first order, receive, responsive to each light stimulus from the first set of light emitters, a user response, determine that a second set of light emitters of the plurality of light emitters are to be used to emit light stimulus in a second order based on the user responses, wherein the second set of light emitters includes fewer light emitters than and at least one light emitter from the first set of light emitters, sequentially emit light stimuli through the second set of light emitters in the second order, receive, responsive to each light stimulus from the second set of light emitters, the user response, and provide a sensitivity map of a visual field of the user that is generated based on output of a machine learning (ML) model responsive to input based on the user responses to the emitted light stimuli.

13. The visual field test system of claim 12, wherein each user response comprises either a voluntary user response, an involuntary user response, or an absence of either type of response after a period of time from when a corresponding light stimulus is presented.

14. The visual field test system of claim 12, wherein the memory has further instructions to determine light intensities for the light stimuli emitted through the first set of light emitters based on the first order and on an age of the user.

15. The visual field test system of claim 12, wherein the sensitivity map includes sensitivity parameters, wherein the memory has further instructions to determine, for each sensitivity parameter, an uncertainty value based on the user responses to the emitted light stimuli, wherein the sensitivity map is provided responsive to the uncertainty values of the sensitivity parameters being less than an uncertainty threshold.

16. The visual field test system of claim 15, wherein the ML model is a first ML model, and the sensitivity map is a first sensitivity map, wherein determining the uncertainty value comprises: generating the first sensitivity map as output of the first ML model responsive to the user responses as input; generating a second sensitivity map as output of a second ML model responsive to the user responses as input; and determining uncertainty values for corresponding sensitivity parameters based on differences between the first sensitivity map and the second sensitivity map.

17. The visual field test system of claim 15, wherein the sensitivity map is a first sensitivity map, wherein determining the uncertainty value comprises: generating the first sensitivity map as output of the ML model responsive to the user responses as input; generating a second sensitivity map as output of the ML model responsive to user responses to the emitted light stimuli through the first set of light emitters in the first order; and determining uncertainty values for corresponding sensitivity parameters based on differences between the first and second sensitivity maps.

18. The visual field test system of claim 15, wherein determining the uncertainty value comprises: generating a plurality of sensitivity maps as output of the ML model responsive to the user responses as input, wherein each of the plurality of sensitivity maps is different; and determining uncertainty values for corresponding sensitivity parameters based on differences between the plurality of sensitivity maps.

19. The visual field test system of claim 15, wherein the sensitivity parameters are associated with the plurality of light emitters, wherein the instructions to determine that the second set of light emitters are to be used to emit light stimulus comprises instructions to determine light intensities for the second set of light emitters responsive to corresponding uncertainty values of corresponding sensitivity parameters being greater than the uncertainty threshold.

20. The visual field test system of claim 12, wherein the light stimuli is emitted while the user remains seated and engaged with the visual field test display.

21. The visual field test system of claim 12, wherein providing the sensitivity map comprises at least one of displaying the sensitivity map of the visual field of the user on a display; and transmitting the sensitivity map as an electronic message to an electronic device.

22. A method for performing a visual field test for a user, the method comprising: providing a first sequence of light stimuli at a plurality of locations within a visual field test area, wherein at least two light stimuli have varying light intensity; receiving, responsive to each light stimulus of the first sequence of light stimuli, a user response; determining a subset of locations of the plurality of locations and light intensities for a second sequence of light stimuli at the subset of locations based on output from a machine learning (ML) model responsive to input based on the user responses, wherein two or more light stimuli of the second sequence of light stimuli at two or more locations of the subset of locations have different light intensities with respect to each other and with respect to light intensities of two or more light stimuli of the first sequence of light stimuli at the two or more locations; providing the second sequence of light stimuli having corresponding light intensities at the subset of locations; receiving, responsive to each light stimulus of the second sequence of light stimuli, the user response; andgenerating a sensitivity map of a visual field of the user as output of the ML model responsive to input based on the user responses to the first sequence of light stimuli and the second sequence of light stimuli.

23. The method of claim 22, wherein each user response comprises either a user input through an input device or an absence of the user input through the input device after a period of time from when a corresponding light stimulus is provided.

24. The method of claim 22 further comprising determining light intensities for the first sequence of light stimuli based on the plurality of locations and on an age of the user.

25. The method of claim 22, wherein the sensitivity map is a first sensitivity map, wherein determining the subset of locations and light intensities comprises: generating a second sensitivity map of the visual field that as the output of the ML model responsive to input based on the user responses to the first sequence of light stimuli, wherein the second sensitivity map comprises a sensitivity parameter for each location of the plurality of locations; and determining an uncertainty value for each sensitivity parameter, wherein the subset of locations are determined based on their corresponding uncertainty values being greater than an uncertainty threshold.

26. The method of claim 25, wherein the light intensities are based on sensitivity parameters of the second sensitivity map for corresponding locations of the subset of locations.

27. The method of claim 25, wherein the ML model is a first ML model, wherein determining the uncertainty value comprises: generating a third sensitivity map as output of a second ML model responsive to input based on the user responses to the first sequence of light stimuli; and determining the uncertainty value based on differences between the first and second sensitivity maps.

Images