Response estimates to efficiently capture dynamic response gain changes in multi-focus responses
By simplifying the multi-focus stimulation into symbolic stimulation groups and fitting the gain nucleus, the problem of difficulty in estimating the response of concurrent stimulation in the prior art is solved, and efficient evaluation and accuracy improvement of field of view functions are achieved.
Patent Information
- Application Number
- CN202380026852.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-03-11
- Filing Date
- 2023-03-10
- Publication Date
- 2024-11-08
AI Technical Summary
Existing multifocal stimulation methods are difficult to effectively estimate the response of concurrent stimulation to the field of view, especially when the stimulation switches from one eye to another, and fail to fully consider individual dynamic gain characteristics and nonlinear changes.
Simplify the multifocal stimulus fluctuations into symbolic stimulus groups, fit the gain nucleus of each group through the response measurement system, estimate the temporal impulse response of each multifocal stimulus area, and adopt a specific symbolic stimulus group to evaluate field of view function and reduce the gain nucleus. and improve response accuracy.
Improve the accuracy and efficiency of field function evaluation, enable more efficient capture of dynamic changes and individual-specific responses of the pupil system, and reduce the number of estimated coefficients and standard errors.
Smart Images

Figure CN118922124A_ABST
Abstract
Description
Technical Field
[0001] The present invention generally relates to methods and systems for concurrently assessing functional aspects of multiple portions of a person's or animal's visual field. More specifically, the present invention relates to determining how the history of stimuli presented within about 1 second affects responses to subsequent stimuli, most particularly when the balance of concurrently presented stimuli switches from one eye to the other. Background Art
[0002] Any discussion of the background throughout this specification should in no way be considered as an admission that the background is prior art or that the background is widely known or forms part of the common knowledge in the field.
[0003] Multifocal stimulation of the nervous system refers to presenting different sequences of stimuli to different parts of a sensory organ over time. For example, four buzzers can be attached to different parts of the skin of the forearm. Then, four different time sequences can determine when each buzzer stimulates the skin at its location. This is a multifocal presentation of concurrent tactical stimulation in the sensory field of the forearm. In the case of vision, the field of view of one or both eyes is divided into multiple regions, and then the visual properties of each region (such as color, contrast, brightness, or texture) are independently modulated over time according to a collection of time sequences, one time sequence for each region of one or more visual fields. The stimulation sequence is a record of the stimulation history of each stimulated region. In either example, some recording means is used to record the aggregated response of the nervous system to the multifocal stimulation.
[0004] A classic recording method is to record the electrical response of the nervous system to a stimulus. In the tactile or visual example just given, the electrical response of the brain can be recorded by placing electrodes on the scalp to capture the brain's response to the stimulus. These records contain a version of the sum of all responses to multiple concurrently presented stimuli. If only one area is stimulated at a time, then the brain activity will reflect the response of only one area, and the response can be estimated by averaging the onset of each stimulus in the sequence. When several stimuli are presented concurrently, the question arises of how to estimate the individual average responses to the stimuli delivered to each sensory area. This can be achieved when the time series driving the presentation of the stimuli at each area are statistically independent, that is, if they are substantially uncorrelated in time. Then, methods such as cross-correlation between the evoked response and the stimulus history, or some form of multiple regression between the evoked response and the stimulus history (i.e., the record of the stimulus sequence), can derive the average response of each stimulated area.
[0005] Patents related to multifocal methods to date have all dealt with specific classes of stimulus sequences, or the spatiotemporal coordination of patterns of sequences. Early examples included nearly orthogonal pseudorandom sequences designed to make response estimation by cross-correlation more efficient: EE Sutter, US 6,086,206. Those methods did not indicate that any particular temporal rate or density of stimulation was optimal. Later, the inventors of the disclosed inventions patented methods for generating stimuli with specific temporal or spatial densities, unexpectedly finding that the responses they produced were many times larger and more reliable: T Maddess and AC James, US 7,006,863 and US 8,583,223. Those patents indicate that the average gain (i.e., the average responsiveness of the system) is higher if the stimuli are kept separate in time or space. Thus, for example, US 7,006,863 focuses on the concept that "appropriate design of stimulus sequences may allow the nervous system... to produce larger or more reliable responses." US 8,583,223 focuses on the effect known as "lateral masking", whereby spatially adjacent stimuli can suppress each other's responses, and accordingly minimizes the deleterious effects of lateral masking, stating that "response size and reliability can be improved by ensuring that concurrently presented stimuli are spatially separated, and such stimuli are therefore referred to as being spatially sparse."
[0006] Subsequently, research shifted from evoked electrical responses to using pupil responses by recording time-varying changes in pupil diameter. Unlike evoked electrical responses, imaging and video monitoring of the pupils provides a non-contact form of multifocal assessment. Due to the connectivity of the neural system driving the pupils, each pupil reports a version of the sum of the activity of both retinas. Therefore, an estimate of the response recorded from a single pupil can yield an average response to stimulation of multiple stimulation areas presented concurrently to both eyes. Therefore, recording two pupils yields two sets of responses for each eye. The shift to pupil recordings yielded two pupil-specific patents.
[0007] The first of these patents relates to the so-called luminance balancing method: T Maddess and AC James; US 8,583,223. This involves keeping the average gain (i.e., responsiveness) on different test areas roughly the same. Average gain refers to the average response per unit stimulus intensity obtained on a collection of visual field areas that are stimulated concurrently during testing. The method is based on the following discovery: when the same intensity of stimulus is presented to each area, some highly responsive areas in the temporal visual field naturally respond more on average than other areas. These highly responsive areas set the average response gain of the pupil system to be lower, thereby reducing the relative responsiveness of other areas. By reducing the stimulus intensity delivered to the naturally highly responsive areas, the average gain during testing remains high, thereby increasing the response of the naturally less sensitive visual field areas. By inverting the static nonlinear stimulus response function, the stimulus intensity is balanced across the visual field. Therefore, US 8,583,223 states: "Reducing the luminance of the stimulus presented to the more responsive areas of the visual field reduces the contribution to the overall aggregated drive signal (of the pupil), thereby increasing the absolute response size of the generally less responsive areas.".
[0008] The second pupil specific patent concerns so called clustered salvos, T Maddess, CF Carle and AC James; US9848771. This concerns the finding that even for luminance balanced stimuli there is a further advantage to presenting stimuli in spatially adjacent clusters within a subset of regions of the visual field. In non-limiting demonstrations these are chosen to be clusters within the left vs. right half of the visual field or within the top vs. bottom half. This is quite different from the predictions of publication US 8,583,223. This approach keeps the number of stimuli presented at any one time step of the test relatively constant, thereby maintaining a reasonably balanced average gain. This does not occur in the previously patented stimulation methods where the average number of stimuli per time step varies more freely.
[0009] In the pupil system, the collection of inputs generated in the two retinas is passed to the left and right pretectal olivary nuclei (PONs), and then from each nucleus to the left and right Edinger-Westphal nuclei (EWNs), each of which innervates its corresponding left or right iris to produce a pupil light response. Each PON also obtains visual input from its left or right visual cortex. Each EWN in turn receives input from both PONs, which is how each pupil responds to the sum of the activity on the two retinas, as mentioned earlier. In a set of carefully designed experiments presented in US 9,848,771, it was concluded that "there is a gain control mechanism in the pupil path at or after the EWN" and that "this gain control system tends to attenuate the pupil response less when multiple visual stimuli are presented in a salvo of spatially adjacent stimulus clusters, compared to earlier methods such as temporally or spatially sparse stimuli." The patent further states that the proposed findings "suggest that EWN gain control involves a feedback mechanism that is too slow to suppress stimuli when they are delivered in salvos". Thus, US9,848,771 has two core ideas: the gain control mechanism that adjusts the average gain across the visual field occurs at the EWN level, when the combination of inputs to the two eyes and the visual cortex is complete, and the mechanism is too slow to be affected by sequences of parallel salvos, which in US9,848,771 occur at a rate of once every quarter of a second.
[0010] We found that interactions between specific consecutive volleys can have a significant effect on short-term gain, whereas previous publications (such as US 9,848,771) showed that the results of all volleys from across the bilateral retina are summed in the EWN and that this results modulates gain, leaving no room for a specific contribution from each eye.
[0011] All of the above-mentioned methods specify different types of multifocal stimulation sequences, or different spatiotemporal coordination of sequences, whereby each test subject receives the same, allegedly optimal stimulation. Similarly, the same response estimation method is used for each test subject and each patented stimulation sequence variant. In focusing on average gain and fixed response estimates, these methods do not indicate that there is any benefit in assuming that the dynamic gain characteristics of individuals will vary, or that there is any benefit in tailoring the response estimation method to the individual or pupil. In fact, those patents present evidence that no such dynamic effects occur, and as mentioned above, US 9,848,771 even states that the system for regulating responsiveness in the EWN is too slow to respond to random volleys of stimuli presented at 0.25 second intervals.
[0012] All response estimation methods employed so far in multifocal testing methods assume that, if nonlinearities exist in the system being examined, these are static. That is, those nonlinearities remain constant over the course of the test. As mentioned, the classical response estimation method involves the cross-correlation between the stimulus history (i.e., the set of test sequences presented for each area to be controlled) and the response of the neural system (the version containing the sum of all responses, the so-called aggregated response). The method was first proposed in 1965: in the article by YW Lee and M Schetzen entitled “Measurement of Weiner kernels of a non-linear system by cross-correlation”, published in
[0013] International Journal of Control, Vol. 2, pp. 237-254. The method is still in common use. The Weiner kernel mentioned in the title of the paper is also known as a linear and nonlinear weighting function. The nonlinear kernel captures the effects of additional nonlinear interactions between stimuli. The linear weighting function is also known as the temporal impulse response to repeated brief, impulsive stimuli. The paper describes methods for estimating the response in a system with dynamic split-gain control, where rapid changes in gain also depend on the stimulus history, and where those dynamics can vary from person to person and even from pupil to pupil. These nonlinearities are not captured using standard function extensions such as the Weiner kernel.
[0014] Subsequently, AC James, R Ruseckaite and T Maddess proposed a more flexible response estimation method based on multilinear regression in their 2005 paper entitled “Effect of temporal sparseness and dichoptic presentation on multifocalvisual evoked potentials”, published in Visual Neuroscience, Volume 22, Pages 45-54. Although those methods are more flexible, they are only suitable for systems with invariant Weiner nonlinearity. Unlike cross-correlation, regression response estimation allows features to be added to the response estimation model to account for various aspects of the recording or the response therein. For example, when recording evoked electrical responses, it is not uncommon for electrodes to pick up signals (buzz) from the AC mains supply. Thus, terms can be added to the regression model to partition the variance in the recording into independent components of physiological response and mains frequency, allowing neural responses to be estimated concurrently with the buzz from the mains. Biological responses often contain a certain degree of autocorrelation. Cross-correlation-based response estimation for multifocal stimulation cannot account for autocorrelation. As a result, the variance due to autocorrelation is misassigned in the estimate, appearing as noise. Concurrently estimating the response to multifocal stimulation and the autoregressive model as a form of interference is possible and may improve the estimate of the fit.
[0015] As demonstrated by James et al., 2005 (supra), another advantage of the regression framework is that the standard error (SE) of each estimated coefficient can be determined. Thus, if there are 88 stimulus sequences controlling the presentation of brief stimuli at 44 locations in the visual field of each eye, then the response estimate will extract the mean size of the response to the presentations at the 88 locations and 88 SEs, one for each response. In contrast, the classical cross-correlation method does not provide an estimate of the error. In principle, the regression framework can be extended to an iterative form of response estimation that can fit nonlinear coefficients.
[0016] Mathematically, estimating more coefficients from the same amount of data means that there are fewer data points in the pupil record for each estimated coefficient, so the standard error of each estimated coefficient will be larger, that is, the estimate will be worse. The process is like estimating the mean of a data set, that is, a single coefficient, and its standard error (SE). By definition, the SE is the standard deviation divided by the square root of the number of data points averaged. Therefore, the fewer points are averaged, the larger the SE. Therefore, as with any statistical estimation process, it is not desirable to try to estimate too much relative to the number of available data points in the pupil record. In principle, the multifocal test can be extended in order to collect more data, but this will make the test longer than tolerable or desired.
[0017] In a more common test, 44 regions of the visual field are examined for each eye, and both eyes are tested concurrently. The mean response amplitude for each region and the time to peak response to stimulus presentation for each region are estimated. This means that at least 2 coefficients or 176 coefficients are estimated for the 88 region / pupil recordings. If 2 more coefficients are added per region to account for changes in the systematic nonlinearity for that region, the total would be 2 x 176 = 352. Even this is possible and the standard error would grow by about the square root of 2, i.e., by more than 40%. Therefore, any strategy for estimating dynamic nonlinearity should have a strategy to minimize the number of additional coefficients to be estimated and still maintain the accuracy of the model by, for example, considering the variance (often referred to as model R). 2 or goodness-of-fit statistics) become the basis for better models to achieve enhanced performance.
[0018] As mentioned above, the regression framework can be extended to iterative forms of response estimation that can fit nonlinear coefficients. A series of such iterative response estimation methods have been established. The purpose of these methods is to determine whether the response estimates of pupil responses to multifocal stimulation can be improved by: 1) including the possibility of rapid fluctuations in responsiveness, i.e., dynamic gain; and 2) whether any such fluctuations can be attributed to recent changes in the stimulus history and estimating those changed coefficients in the form of so-called gain kernels; and 3) whether only a small subset of the possible numerous gain kernel coefficients can be efficiently estimated, thereby reducing the requirement for large amounts of data. Note that the normal time impulse response for each stimulus region will be estimated in the same process, so that the estimate of the gain kernel improves the estimate of the regional time impulse response. Another question is whether it is possible or desirable to estimate a separate gain kernel for each test subject or each pupil, or to fix the value of the gain kernel for each person that may depend on factors such as their gender and age.
[0019] Using those new methods, a series of findings that improve the accuracy of regional temporal impulse response estimates were obtained. These findings showed for the first time that changes in responsiveness associated with rapid stimulus histories should be considered for response estimation, and that these changes can be summarized in a very small set of coefficients as long as rapid changes associated with stimulus switching between eyes are included.
[0020] The overall goal of these methods is to efficiently assess the range and function of a person's or animal's field of view, in other words, to generate data about its field of view. An example animal may be a racehorse. Horses belong to animal husbandry, so a particular horse may have a field of view that is more or less suitable for being a successful racehorse. If the subject is a human, the field of view data may be used in combination with other data to determine whether the person is suitable for a given occupation, or to determine whether they should be allowed to drive a particular vehicle. Other data may be metabolic or physical data from a person. Field of view data may also be used to monitor a person's status. For example, a doctor may have diagnosed a person with diabetes using a blood sugar test. It may be useful for a doctor to monitor a patient's field of view data over time to understand any changes caused by diabetes or more common diseases in diabetes, such as simple cataracts. Finally, in combination with other data, a medical professional may use a combination of field of view data and other data to diagnose whether a person has a specific eye or brain disease, and the doctor or other medical professional believes that these data combined can be used as a basis for diagnosis. In a clinical setting, monitoring the stability of treatment results may be more common than assisting in diagnosis, because once a given person is diagnosed with a disease on a certain day, they may be monitored multiple times throughout their lifetime. Summary of the invention
[0021] Embodiments of the present invention simplify the rich set of multifocal stimulus fluctuations into a much smaller set of symbolic stimulus groups, and then the response measurement system fits a separate gain kernel for each of these groups (the gain kernel summarizes the dynamic changes in pupil response gain), while simultaneously estimating (i.e., measuring) the temporal impulse response for each multifocal stimulus region. That is, the term "gain kernel" should be considered as measuring the dynamic changes in pupil response gain. A set of symbolic stimulus groups with specific characteristics is shown to be more useful for evaluating the visual field. The gain kernel describes how the last few time steps of the stimulus history affect the responsiveness of the pupil system to the current stimulus selected from a specific symbolic stimulus group. In characterizing various aspects of the pupil neural system, the gain kernels associated with these stimulus groups are improvements to the response estimation (i.e., measurement) method of multifocal visual field testing using pupil records. To illustrate these steps, a non-limiting example of a set of clustered salvo stimuli is decomposed into symbolic stimulus groups, and by combining them into the response estimation (i.e., measurement), it is shown that the original clustered salvo method can be surprisingly improved by adding only a few additional coefficients. Stimulation methods other than the clustered salvo method can be selected. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Arrangements of methods, devices and systems are described below, by way of example only, with reference to the accompanying drawings, in which:
[0023] Figure 1The layout of a set of 44 multifocal stimuli is shown, and when the test subject looks at the center of the array, these stimuli will tile the central 60 degrees of the test subject's visual field. The outline enclosed by each diamond represents the outer boundary of a stimulus. When indicated by the control stimulus sequence, one or more of the stimuli take their active state within a certain period of time. In this specific non-limiting example, each active area is filled with bright yellow light for 33ms, and the background is much darker yellow. The 44 stimuli can be divided into three (A) or two (B) sets of non-overlapping rings. If the set of 44 stimuli is in active state at the same time, they will overlap, as shown in (C). The 3-ring or 2-ring set can be further divided into a series that either tiles the left half of the visual field or tiles the right half of the visual field, as shown in (D).
[0024] Figure 2 An apparatus is illustrated for: 1) presenting separate multifocal stimuli to two eyes separately; 2) obtaining video recordings of the two pupils under infrared illumination; and 3) a computerized system for controlling the display of the multifocal stimuli on two display systems, coordinating video capture of images of the pupils with the stimulation, and estimating (i.e., measuring) individual responses to the stimuli presented at each of a number of locations.
[0025] Figure 3 176 temporal impulse responses, also called linear weighting functions, are shown, which are the average responses of the pupil to the presentation of each of the 44 stimuli to each eye, as Figure 1 and Figure 2 In (A), there are pairs of impulse response waveforms that correspond to the locations of the stimuli that evoked them, following Figure 1 There are two waveforms for each stimulus position because, as indicated in the legend, one response comes from the left pupil and one response comes from the right pupil. This is because Figure 2 As shown in (A), both eyes are stimulated and both pupils are recorded concurrently. Each pupil receives information about the stimulus for each eye. As indicated by the inset axis in (A), each figure starts at stimulus presentation time 0 and shows the time evolution of the response over 1000ms. The vertical scale is 35μm and the downward deflection of the waveform indicates a reduced pupil diameter. Therefore, the average response to pseudo-random flash stimulation is a brief, smooth pupil contraction followed by a slow recovery to the baseline diameter. The amplitude and time to peak of each waveform can be estimated, as indicated in (B).
[0026] Figure 4 The diagram shows the use of Figure 2 The CYCLE 1 column shows that at time 0, a stimulus from the left visual field series (e.g., Figure 1 Six of the stimuli (stimulus elements) of D) are active and therefore brighten and are displayed to the left eye. The brightly presented stimuli are shown here in gray. In this specific example, all stimuli have a duration of 33 ms. The remaining six areas remain inactive and therefore only a darker background is shown at their locations, which are marked by a dotted border. At time 0.25 seconds, a different stimulus randomly selected from the right half field stimulus series is again displayed only to the left eye. Figure 2 The opaque wall material blocks the right eye from seeing any stimulus for the left eye. At time 0.5 seconds, the right eye receives its first stimulus. In this example (and other examples described herein), the time step is 0.25 seconds, but it will be understood that different time steps can be used. The order of the half-field series can present active stimulation (so-called salvo) and then repeat 8 steps per cycle. It will be understood that by selecting different stimulus series, there can be fewer or more time steps per cycle. The figure shows a possible set of stimuli presented on three of these cycles from row A to row X.
[0027] Figure 5 Shows an image Figure 3 Three possible allocations of one cycle of such clustered simultaneous multifocal stimulation to symbolic stimulation groups. Once assigned groups, they will be maintained in multiple cycles of the whole stimulation. Column 1 on the left shows all stimulation series as black, indicating that they belong to a symbol group (the ordinate label is A to H). Column 2 shows the assignment of series to 8 separate symbol groups (displayed as black, gray (from dark gray to light gray) and white, and the ordinate label is A to H). In column 3, according to the repeated pattern ABBA, the half-field series is assigned to two groups: black or light gray (and the ordinate label is A to H). The number of gain kernel coefficients required for the dynamic change of characterizing pupil responsiveness is reduced by the response estimation method described in this article using these types of group assignments. That is, each symbolic stimulation group only estimates (that is, measures) a gain kernel, and each gain kernel is for some time steps or lags in the stimulation history. By selecting a group that captures the important features of the pupil neural system, and estimating (that is, measuring) the gain kernels of the stimulation belonging to those groups, accurate multifocal responses can be efficiently estimated.
[0028] Figure 6Three alternative assignments of clustered stimuli to groups of symbolic stimuli are illustrated. Column 4 shows that four groups are assigned, whether they belong to a larger 3-ring, half-field series of 12 stimuli, or a smaller series of 10 stimuli arranged in 2 half-rings. Column 5 shows the assignment to two symbolic groups (shown by the gray and white coloring and the ordinate labels on A to H), where the groups are defined by the presented eye. Column 6 shows the assignment to 4 symbolic groups, which are defined by the larger and smaller half-field series and the stimulated eye, so that the group assignment switches eyes every other time step, so it is called the switched eye assignment method.
[0029] Figure 7 The diagram shows a cluster volley method using different types of stimulation trains that cycle in a cyclic manner, e.g. Figure 4 The series here are: Morph 1, upper vs. lower visual field; Morph 2, alternating quadrants; and Morph 3, alternating quadrants with sporadic extra stimuli (light grey) in the other quadrants.
[0030] Figure 8 The 88 control stimulus presentation sequences for each region r and 0.25 s time step n are shown. Transformed into a relative contrast parameter for each symbol group g and its summation version c n These are to estimate the pupil gain kernel coefficients for the signed stimulus group g at each time lag j In (A), the stimulus presentation sequence has been chunked into groups of eight symbolic stimuli (represented by eight gray shades and depicted by segments 8A to 8H), with Figure 5 The groups in column 2 of A correspond to the groups in column 2 of A. Only the first 50 seconds of the 240 second long sequence are shown. In each sequence in A, the upward tick indicates that the stimulus was presented in one of the 44 areas of each eye. Figures (B) and (C) show the steps in calculating Equation 1. D shows the sum of Equation 1 for each group: c n .
[0031] Fig. 9 The difference in bilinear basis function waveforms obtained when the gain kernel is included (black waveform, gain) or not (grey waveform, no gain) in the response estimation is shown. These are the median basis functions calculated after testing pupils, replicates, and 170 normal subjects using the P129 stimulus array.
[0032] Fig.10 An example of the estimated response to two orthogonal basis functions for each pupil, referred to here as basis functions 1 and 2, is shown. The waveform is an estimate of the left pupil response to stimulation of the left eye of a 43 year old male subject using the P129 stimulation array. Fig.10A shows at the top 44 estimated response waveforms, each with a different shade of grey, one for each of the 44 stimuli in the P129 stimulus array. Each waveform is a weighted sum of two basis function waveforms. Fig.10 B is the 44 weighted versions of the first basis function, which are the response components based on basis function 1. Fig.10 C is a weighted version of the second basis function, which is the response component based on basis function 2. Therefore, Fig.10 B and Fig.10 Summing the set of 44 pairs of response component waveforms in C, we get Fig.10 The estimated response in A.
[0033] Fig.11 The figure shows the median response waveforms calculated over 44 stimulus areas and the left and right eyes (LE and RE) of 170 subjects, as recorded by the left and right pupils. When the pupil reports the stimulus of its own eye, this is called a direct response. In contrast, when the pupil reports the stimulus of the other eye, this is called a congruent response. The legend indicates which gray level corresponds to the direct and congruent response of the two eyes (LE and RE). Each averaged waveform is obtained by Fig.10 and the two-component bilinear method of Eq. 10 created. Fig.11 A is the median response obtained with the estimated gain kernel, and Fig.11 B is the median response obtained when the gain kernel is not estimated. DETAILED DESCRIPTION
[0034] Figure 1 An array of 44 multifocal stimuli for presentation to the eyes of a person or animal is illustrated. Each stimulus (i.e., stimulus element) is diamond-shaped. Only the outer edges of the stimuli (i.e., stimulus elements) are shown here as black envelope lines. In this specific non-limiting demonstration, each part of the diamond-shaped stimulus area is bright yellow and is briefly presented on a constant darker background. During testing, the subjects look directly at the center of the array of stimuli, fixing their eyes on the small red cross located there. In the case of animals, they can be anesthetized. At the beginning of the test, there are no stimuli, only the background. When the test starts, the stimulus sequence determines when a given stimulus appears at its location. Each yellow stimulus is presented for 33ms each time the stimulus sequence requires it to be presented. During the test, the probability of any given area appearing in a pseudo-random manner is low. This randomness is required for the multifocal method to work because it makes the presentation of stimuli at one area statistically independent of the stimuli presented at other areas. That is, at any time step, each individual stimulus element has a probability of being presented of approximately 50%. The stimuli are arranged in 5 rings, which are Figure 1A, shown in Figure B. Although overlapping areas are never displayed together, if they are presented simultaneously, they will overlap in the way shown in Figure 1 C. Or Figure 1 The three rings of A or Figure 1 An example of a left and right half field set or series of two rings of B is shown in Figure 1 Shown in D. The array of 44 stimuli extended from the central fixation point to 30 degrees of the visual field as indicated by the axis labels.
[0035] In practice, two arrays of stimuli can be presented concurrently, with 44 stimuli per eye. Figure 2 An array is shown being presented to two eyes (200A, 200B) concurrently using a non-limiting stereoscopic arrangement. Figure 2 A computer system 201 with a computer 203 is shown. A light source 205 (e.g., such as an infrared light source) is controlled by the computer 203. The eyes of the subject are illuminated with infrared light from the light source 205 through a pair of cold mirrors (207A, 207B), which are transparent to infrared light. The cold mirrors are typically suitable for visible light, so using them and objective lenses (209A, 209B) allows each eye to see one of two display systems (211A, 211B), each display system presenting a multifocal array of stimuli to the pupil of each eye, such as Figure 1 As shown in . An opaque wall (213) isolates the left and right halves of the instrument so that the eyes cannot see the opposite side. The display system (211A, 211B) also includes a visual cue that causes the subject to automatically binocularly fuse the images displayed on the two display systems. Therefore, the subjects do not know that they are viewing two images, one image displayed on each of the two display systems, but they feel that they are viewing a single image. For example, two sensors (215A, 215B) in the form of infrared cameras capture moving pictures of the pupil, and the computer 203 of the computer system 201 continuously measures the pupil diameter to provide a time record of the pupil's response to multifocal stimulation. The computer 203 of the computer system 201 uses 88 stimulus histories (i.e., the display sequences used) and pupil responses to implement a response estimation (or calculation) method in order to estimate (or measure) the time impulse response to each stimulus sequence and the set of gain kernels.
[0036] Figure 3 A shows a collection of 176 estimated temporal impulse response waveforms, 44 for each eye and pupil. A downward deflection of the waveform indicates a decreasing pupil diameter. Figure 3B shows the peak amplitude of the contraction and the time to peak contraction, i.e., the time to reach the peak, which was estimated for each impulse response. The time to reach the peak provides a measure of the latency of the response. In conventional visual field testing, sensitivity is estimated by modulating the stimulus on a decibel scale; the μm contractions shown here are then converted to decibel amplitudes to derive sensitivity.
[0037] Figure 4 The diagram is generated Figure 3 In a specific non-limiting example of a response to a randomized cluster of stimuli, 88 stimulus sequences are combined to present the stimuli in a randomized clustered volley. The time (in seconds) is from the top to the bottom of each column in columns 1 to 3. There are pairs of panels on each row of a column. Each pair of panels shows the response to a randomized cluster of stimuli. Figure 2 The method shown in Figure 2 presents the stimulus to the left or right eye (column heading). At 0.25 second intervals, some of the possible stimuli within the half-field series are presented in bright yellow for 33 ms. The presented stimulus area is shown here as solid gray. This figure demonstrates that approximately half of the possible areas of the half-field series (such as Figure 1 D) is not shown, and the dotted border shows where those other stimuli of each half-field series will appear. Therefore, at the beginning of the test, that is, time 0 (A), six members of the series of left half-field stimuli may be activated for the left eye. At 0.25 seconds, about half of the possible areas in the right half-field series are active (3 in this case) and presented to the left eye. At 0.5 seconds, a new volley of left half-field stimuli is presented to the right eye. This situation lasts until 1.75 seconds, when 1 stimulation cycle is completed. Then, the sequence of the half-field series is repeated every 2 seconds in a cyclic manner, as shown in rows I to X of cycles 2 and 3, but at each time step, any member area of the series has a 50% chance of being active, that is, being presented. This randomized presentation of each area keeps the stimulus sequences statistically independent to allow responses to be extracted from pupil records using a multi-focal response estimation method. In this example, the cycle is repeated until each stimulus area is presented 96 times, taking a total of 6 minutes. Sometimes other sets of series are used (see below) and different numbers of repetitions are used per region. Different numbers or colors of stimuli may also be used.
[0038] However, continuing with the current example stimuli, it can be seen that sets of hemifield series can be assigned to different types of symbolic stimulus groups. The purpose of these groups is to further reduce the number of estimated gain kernels from the full set of possible gain kernels; there is one for each hemifield series. Each series is assigned to a specific group, and the groups have a common gain kernel. Therefore, the number of coefficients required to be estimated is smaller. However, there is a risk that too many or the wrong series will be assigned to a particular symbolic analysis group, resulting in a poor overall estimate. That is, an incorrect group assignment may fail to capture important features of the dynamics of the pupillary neural system. Efficient and advantageous symbolic groups were discovered experimentally.
[0039] Figure 5 Three possible methods of symbol group assignment are illustrated. In this figure, each column represents a different group assignment strategy that will be repeated in each cycle of the entire test sequence. In column 1 on the left (including rows A to H), all half-field series are painted black, indicating that they will be treated as a single group for response estimation. That is, a single gain kernel (i.e., a set of gain kernel coefficients) will be estimated for all of them, so this method is called shared gain. In the middle column 2, each half-field series is painted black, one of six gray levels from dark to light or white, indicating that each half-field series is treated as one of eight groups for response estimation. That is, a different gain kernel will be estimated for each of the eight symbol groups, so it is called the all half-field method. In column 3, for the duration of the stimulus, the half-field series are assigned to one of two groups, in the order ABBA, ABBA, and so on. In this figure and Figure 6 In , the ordinate labels of each row reflect the group assignments: Group 1, Group 2, etc. Those assignments will be maintained across all cycles of stimulation.
[0040] Figure 6 Three further alternative assignments of symbol groups are shown. In column 4 (A to H), there are four symbol stimulus groups, assigned 1 to 4 for the 3-ring stimuli in rows A to D, and this is then repeated for the similar 2-ring stimuli in rows E to H. Therefore, whether the half-field series has 2 or 3 rings, the assignment is called the half-field method. The middle column 5 shows the assignment of two groups corresponding to the stimulated eye (dark gray and white, and ordinate labels). Therefore, the assignment is called the intraocular method. In the right column 6, the half-field series is assigned to 2 groups (grey shading or white, and ordinate labels). The assignment always switches to a similar half-field of the other eye, so this assignment of symbol stimulus groups is called the switched eye method.
[0041] So far, all the examples have used series that were collections of stimuli from either the left or right visual field. Alternative types of series can also be used, Figure 7 Some examples are illustrated in . Figure 7Each column shows a different type of series that will be repeated in a cyclic manner, such as Figure 4 As shown in . Figure 4 As in the , different stimuli within each series are randomly active (grey areas). The leftmost column, variant 1 (rows A to H), uses series arranged by half-field, but with lower and upper halves instead of left and right halves. The middle column, variant 2 (rows I to P), illustrates series with alternating pairs of quadrants. So far, these series have included conventional subsets of the stimulus areas. Figure 7 The third column of , variant 3 (rows Q to X), is similar to the middle variant 2, but with occasional extra regions. To clearly indicate the extra regions, they are presented in a lighter grey. In practice, their brightness and colour can be the same or similar to the other stimuli. Variant 3 illustrates that irregular series are possible, where varying degrees of deviation from the regular pattern of the series can be tolerated. Clearly, Figure 7 The rows can be assigned to groups of symbolic stimuli, such as Figure 5 and Figure 6 in or other groups.
[0042] These examples demonstrate that the described method (and associated system) reduces the number of gain kernels required to be estimated while improving the accuracy of the impulse responses obtained from each stimulation region estimated along with the gain kernels.
[0043] Embodiments of the present invention are primarily developed for methods and systems for quantifying the visual field by estimating the response of the visual nervous system from a record of pupil size changes over time obtained in response to multifocal stimulation. Embodiments include methods and systems for quantifying stimulus-related dynamic changes in the pupil system (i.e., dynamic changes in pupil response gain). The dynamic changes in pupil response gain measured by a small set of coefficients are referred to herein as "gain kernels". This is different from only estimating linear and nonlinear weighting functions to characterize systems that contain nonlinearities. In principle, each stimulus sequence can have a gain kernel, but instead, they are estimated for smaller symbol groups of stimulus sequences. Therefore, if 88 stimulus sequences are divided into 8 groups, only 8 gain kernels will be estimated, not 88. If the stimulus is summarized by 2 symbol stimulus groups, only 2 gain kernels need to be estimated. Estimation of gain kernels for a limited number of symbol stimulus groups has never been considered or attempted before. In addition to estimating gain kernels, embodiments are also based on the discovery that specific symbol stimulus groups have surprising advantages.
[0044] Some features of specific non-limiting types of multifocal stimulation are described above. Figure 1 and Figure 2 As shown in , they displayed pseudo-randomly presented stimuli to 44 areas of the visual field of each eye. Figure 3 As shown in , the stimulation is delivered in a so-called cluster salvo. Under this scheme, one of a set of 8 series of stimulations distributed in the left or right half-field of the left or right eye is selected for possible presentation in each 0.25 second interval. The sequence of 8 half-field series is repeated in a cyclic manner. Figure 3 Three cycles of this repetitive process are shown. At any 0.25 second time step, each stimulus area in the half-field series has a 50% chance of being active, i.e., the stimulus area appears bright yellow within its diamond-shaped borders against a darker background. The total duration of the stimulus is 383 seconds, and during this time, each of the 88 stimulus areas is activated 96 times. When presented, each stimulus remains bright (active) for 1 / 30 second (33ms). The activation of each of the 88 areas is controlled by a sequence of 88 stimuli (also called stimulus trains or stimulus trains). These consist of 0s and 1s, such as Figure 8 As shown in A. The stimulus is activated when the sequence changes from 0 to 1, i.e., it is presented for 33 ms. The interval of the time steps of the stimulus sequence (i.e., their temporal resolution) is 1 / 30 second. The stimulus salvos occur on average every 0.25 seconds.
[0045] The pupil response is a continuous record of pupil diameter captured over time synchronously with stimulus presentation. The pupil diameter responds to the sum of all activity generated by the stimulus presented to both eyes. The pupil diameter contains this information because it is transmitted to the iris from the pretectal olivary nucleus (PON), which receives input from the extrastriate visual cortex and retina. The two PONs each provide information from both eyes to two Edinger-Westphal nuclei, each innervating one iris, so each iris responds to stimulation of both retinas. The stimulus sequence and pupil light response records are then submitted to a regression response estimation method or system to extract the response of the visual system to each of the 88 stimulus areas. There are two pupil records, one for each eye, so the response estimation process will be repeated twice to produce a total of 176 estimated responses. Figure 3 A shows a collection of 176 linear weighting functions (ie, average time impulse responses) which are the result of a standard regression response estimation procedure.
[0046] However, recent studies have provided evidence that the pupil system does not remain linear, and therefore a method to capture nonlinear dynamics in the form of a gain kernel is incorporated into the response estimation process. Figure 8 Steps in an illustrative process are shown. Figure 8 A shows the method for generating Figure 3 The responses were recorded for the first 50 seconds of 88 stimulus sequences. There was one sequence for each stimulus area and 44 sequences for each eye. Figure 1 and Figure 4 as shown in . Figure 8 A horizontal line is a sequence of 0s, with the occasional upward stick being a 1, indicating that a stimulus should be presented. In this non-limiting example, the stimulus sequence is presented in 8 different shades of gray. These represent Figure 5 The 8-symbol stimulus group of column 2. Strictly speaking, Figure 5 Groups 1 to 8 of column 2 are in Figure 8 Moving upward in the figure, the series of 12 regions per half field of view is at the bottom and the series of 10 regions per half field of view is at the top.
[0047] Signals representing the dynamic balance of stimulation within a group over time are developed to quantify any fluctuations in stimulation intensity within a group. These will be used in a regression framework to add estimates of the coefficients of the so-called gain kernel to the standard procedure for estimating the coefficients of the impulse response from each stimulus. This procedure is derived from Figure 8 A is marked as The stimulation sequence of region r starts with t = t, indicating the stimulation sequence of region r with time step n. That is, indicating that at t = t n The input of the stimulation activity of region r is expressed as n=1,…,n t , where n t is the number of time steps. If the time at which a salvo occurs is defined as Then when the flash occurs The value is 1 (n∈k0,k1,…,k p ), otherwise 0( Figure 8 A). A value of 0 indicates a darker background level. If is the median of the region selected from the symbolic stimulus group g, and G g is the set of regions in group g, then at time t n The relative contrast parameter can be defined as, for n∈k0,k1,…,k p :
[0048]
[0049] Otherwise it is 0. Note that G g yes Figure 5 and Figure 6 The gth symbol stimulus group in any group of stimuli. Figure 8 B shows the summation step: and Figure 8 C shows the output of equation 1 above, Note that, gives a measure of the relative instantaneous balance of stimulation relative to the median level defined by the stimulation frequency within the group. is positive when the number of stimuli in the group is greater than the average, and negative when the number of stimuli is less. For the example stimuli, only a random selection of stimuli from a single group is shown at any one time, so that a representation of each time t can be formed. n A single vector of stimulus contrasts: For n=1,…,n t ,like Figure 8 As shown in D. Once this framework is provided, a time-integrated contrast parameter is introduced for each group. This parameter depends on c n ( Figure 8 D) The current value and past value at this point, for n∈k0,k1,…,k p (Location where stimulation occurs):
[0050]
[0051] Otherwise it is 0 (i=0,1,…,p). Here is the pupil gain kernel of group g and M g is the number of steps after which the gain change becomes insignificant (i.e., when the dynamic change in responsivity becomes negligible). g is the number of time lags in the stimulation history of the group, i.e., the stimulation sequence of the group, which needs to be taken into account to capture the effect of dynamic gain changes, i.e., the number of lag coefficients per kernel. The time between two entries of is (k i-j -k i-j-1 ) / 30. In fact, The entries of are separated by 7 or 8 steps of 1 / 30 sec, depending on the group, so the average is 0.25 sec. Therefore, the pupil gain kernel is a is a sparse description of how the current (j=0) or previous (j>0) values of sum over time. Therefore, Equation 2, is based on the gain core Describes the temporal integration properties of the pupil as a measure of how recently a stimulus has been delivered.
[0052] The division normalization factor for the signed stimulus group g is defined as (for i∈k0,k1,…,k p , otherwise 0), using Taylor's theorem it can be approximated as p :
[0053]
[0054] If you define new inputs for each group g, For n=1,…,n t And r∈Gg , then we can remove the subscript and use vector notation where (⊙) is the element-wise product. n The pupil light response component due to region r in group g is modeled as:
[0055]
[0056] where M is the number of time steps after which the impulse response is not significant (about 60 time steps of 2 seconds or 1 / 30 of a second). Describe the time impulse response of the pupil in each region r, such as Figure 2 The basis functions in each region can be modeled by a lognormal function:
[0057]
[0058] Among them A r is the amplitude, is the time to reach the peak, and σ r is the time width parameter of the rth region. These are reasonable models for the linear weight functions, i.e., Figure 2 The time pulse response of . According to Equation 4, the time t n The total pupil light response at Set the time t n The pupil diameter y n The equation relating (μm) to the pupil light response is the discrete equation:
[0059] y n =f n +μ n +ε n (6)
[0060] where μ n is the interference signal and ε n is the error term. In reality, 382 seconds of stimulation is too long for individuals to tolerate. Therefore, the complete sequence Divided into 9 segments, each lasting just over 42 seconds. The 9 segment recordings are later reconnected into a full-length recording for analysis. This means that each of the 9 segments can have a different slope, constant drift. n Each of the nine stimulus segments is modeled as a separate constant drift (v1,…,v9) and the error term ε n is modeled as:
[0061]
[0062] where d0 = 1 by convention. Note that k here is the real-time step size of 1 / 30 second, and in this non-limiting example, 10 is considered. Therefore, ε n Is a white noise innovation n The result of the autoregressive process. The filter d = [d0, ..., d9] is called a pre-whitening filter. If the vector E = [ε1, ..., ε n ] and Z=[z1,…,z n ] is defined and D is the circulant matrix of d, then Equation 7 can be rewritten in matrix form as DE = Z. In other words, pre-whitening can be written as a matrix multiplication, using the notation below.
[0063] Fitting the data to the model—response estimation
[0064] An important assumption in the underlying fitting process is the innovation z n Does not contain autocorrelation. Consider one-step-ahead predictors
[0065]
[0066] in is at time t n The pupil diameter measured at n =f n +μ n If you define Then the column vector of stimulation at region r and the input matrix and a parameter vector θ containing all the parameters of the model:
[0067] , then F n is a function of X and θ: F n =F n (X,θ), and if the full model is used, then θ has n p =n r ×3+n k +n r ×M g +9 parameters.
[0068] This demonstrates the power of the regression framework, where θ contains the parameters of the impulse response basis function: A r ,t r ,σ r ; the coefficients d for the autoregressive filter, and the coefficients for the gain kernel for each group at each lag of n. All of these are therefore fitted simultaneously, thereby dividing the variance in the pupil recordings into its component sources. As a result, the SE in parameters such as the response amplitude for each stimulus region is reduced. By forcing the gain kernel to be a small set of sign groups for the chosen stimulus regions, The shared parameter values of , greatly simplify the model. However, it requires accurate selection of the groups to produce good estimates with fewer gain kernels and their coefficients.
[0069] The model fits the pupil data by minimizing the sum of squared errors using the Marquardt-Levenberg algorithm: This requires evaluation X(n t ×n r ) at which its Jacobian matrix J(X,θ)(n p ×n t ) and HessianH(X,θ)(n p ×n p ×n t ).
[0070] After developing this basis function for simultaneously estimating each stimulation region, and a gain kernel characterizing the dynamic changes in the system for the selected set of symbolic stimuli Next, we determine the parsimonious number and arrangement of groups of signed stimuli and the number of significant gain kernel lags for each group that yield improved response estimates, M g .
[0071] The regression framework can be summarized as follows. The standard method for measuring (or "estimating" in statistical terms) the impulse response (i.e., basis function) of each stimulated visual field region can be summarized as
[0072] Estimate of basis function = function (pupil response record, stimulus record) (a)
[0073] Equation a can be summarized as: The basis functions characterizing the average response to a stimulus and the estimate of their scale are functions of the pupil record and the stimulus record. Typically, a subject will have one or two time-varying pupil response records in a recording session, as well as Figure 3 Multiple stimulus recordings depicting stimulus onset and duration for each tested portion of the visual field of one or both eyes are shown in A. There may be more than one basis function per tested visual field region, and in that case the average response for each region will be a linear or nonlinear combination of the basis functions for each region. The measurement of the basis functions may be accomplished by regression methods or cross-correlation. In a regression framework, the basis function estimation process may be improved by including additional terms in the equation.
[0074] Estimate of basis function = function (pupil response record, stimulus record, other terms) (b)
[0075] θ in Equation 8 describes the basis function term consisting of the coefficients of the lognormal basis functions that characterize the fit: and other terms including the autoregressive model coefficients: Coefficients of the gain kernel The number of sign groups multiplied by the number of time lags per group; and the interference terms, v1,…,v9, for the nine segments. Of course, other terms can be added or different basis functions used, as will be described below. The key issue is to choose the minimum set of sign groups and lags, that is, the optimal gain kernel coefficients, which together with the basis functions minimize the sum of squared errors. To allow fitting the gain kernel, the stimulus history needs to be transformed into the relative contrast parameters of equation 1, and c n ,like Figure 8 as shown in .
[0076] Bilinear basis functions
[0077] As an alternative to the lognormal parameterized basis functions, the general time impulse response But this increases the number of parameters in the basis functions so much that the required test length will be too long. For example, 2 seconds would require 61 time points per region, increasing the number of parameters from 2 (amplitude and delay) × 88 = 166 to 61 (time steps) × 88 (regions) = 5368. To keep the test length reasonable, it can be assumed that the waveform of each region is a weighted sum of l common time components
[0078]
[0079] in is the weighting factor for the jth temporal component in region r. Equation 4 can be replaced by a bilinear form as suggested by Goh, XL in her 2008 PhD thesis, “Cortical generators of human multifocal visual evoked potentials and fields”, DOI: 10.25911 / 5d51548ee0131:
[0080]
[0081] If a single time component with 61 time steps is chosen, there are a total of 61 (time steps) + 88 (regions) = 149 parameters, which is less than the lognormal parameter case (177). For two components, there will be 298 basis function parameters. The Gauss-Newton method can then be used to estimate the bilinear, gain, interference, and autoregressive parameters concurrently, as suggested by Inverso, S. A., Goh, XL, Henriksson, L., Vanni, S., and James, A. C. in their 2016 publication "From evoked potentials to cortical currents: Resolving V1 and V2 components using retinotopy constrained source estimation without fMRI. Human brain mapping", Vol. 37, pp. 1696-1709.
[0082] Separate ON and OFF channels
[0083] Contrast sequence can be separated into ON and OFF components as defined in Equation 1, Where |·|' is half-wave rectification, that is, when x>0, |x|'=x, otherwise it is 0, that is, The ON and OFF kernels of group g are respectively p g and q g , then equation 2 becomes:
[0084]
[0085] Here M p and M q are the number of steps after which the gain in the ON and OFF channels becomes insignificant, respectively. Assuming that the ON and OFF sequences are independent, the kernels of the two channels can be estimated. For all j, In the case of , the linear model defined by Equation 2 is restored, where Therefore, Equation 11 defines a broader class of models than Equation 2.
[0086] Demo 1
[0087] To provide a dataset to demonstrate the procedures described herein, 94 normal control subjects (43 males) aged 49.6 ± 19.5 years (mean ± standard deviation) ranging from 18 to 91 years were tested, as well as 40 people with glaucoma in one or both eyes aged 67.2 ± 8.60 years ranging from 50 to 83 years. Glaucoma causes localized damage in the retina in the form of patches, which produce changes in sensitivity and latency, which can be characterized by making visual function maps such as those currently provided publicly.
[0088] Each person was tested twice approximately 2 weeks apart using 3 variations of clustered multifocal stimulation ( Figure 3 ). For simplicity, the three test variants are referred to as P129, P130, and P131. Figure 1 The P131 differs in that its stimulus array is isomorphically scaled down by a factor of 2. Thus, the P131 stimulus samples the visual field area 4 times smaller, but there are 44 scaled stimuli per eye. The P129 and P131 stimuli are presented in a dimmer 10 cdm -2 The background is bright yellow. P130 stimulus is 10cd / m 2 Bright green against a red background. Stimulus intensities vary to achieve luminance balance. P129 stimulus range is 73 to 150 cd / m 2 , P130 is 65 to 150cd / m 2 , while P131 is 134 to 288cd / m 2 The individual test areas were presented in a pseudorandom fashion such that each area presented a 33 ms stimulus at an average interval of 4 s. The test lasted just over 6.3 min, and each area was presented 96 times. In each test, the Figure 2 The device shown in tests both eyes concurrently.
[0089] During the test, pupil diameter was recorded every 1 / 60 s, synchronized with multifocal stimuli presented on two displays, one for each eye. Afterwards, the data were downsampled to a time step of 1 / 30 s. To allow the subjects to manage the test, each 6.3-min test was divided into 9 segments, each lasting approximately 42 s (Equation 6). The subjects were presented with segments separated by rest periods of 7 s or more, during which they could blink freely. Portions of the pupil recording in which blinks were present were removed. If more than 15% of a segment was lost due to blinking, the segment was repeated. This occurred in only about 1 in 200 segments.
[0090] The pupil response and stimulus history are then submitted to a variation of the response estimation method described above to estimate F n =F n(X,θ). This involves simultaneously estimating the mean pupil response for each region (e.g. Figure 3 ), that is, the basis function and gain core Variations of this method use different numbers and types of symbolic stimulus sets (e.g. Figure 5 and Figure 6 ), and different numbers of lag coefficients in the gain kernel. The goal is to find the equation that minimizes the total number of coefficients in the model (M in Equation 2). g ) also well characterize the specific group and number of hysteresis coefficients of the pupil response to 88-area multifocal stimulation.
[0091] Several methods have been used to assess the efficiency of different response estimation models. A simple measure is the model or R 2 The proportion of variance in pupil responses accounted for by the other measures examined response parameters such as peak amplitude A r and peak time ) reproducibility. Since there were control subjects with normal vision and patients with glaucoma, the diagnostic ability of pupil response parameters was also examined. Diagnostic ability was characterized as the percentage area under the curve (%AUC) of the receiver operating characteristic (ROC) plot of sensitivity to false positive rate. The obtained gain kernel coefficients were also fit using a linear mixed effects model. This allowed the determination of the mean gain kernel coefficient and its SE across subjects while controlling for multiple comparisons (multiple gain kernel coefficients within a subject). This was performed separately for the data of the control group and the glaucomatous subjects. The R in these models 2 Used as a measure of consistency in gain kernel coefficients across subject groups. By comparing these measures, the optimal number and type of symbol stimulus groups, as well as the number of gain kernel lag coefficients used to model pupil responses, were determined. Mean gain kernel lag coefficients were also examined to understand why some symbol groups outperformed others. Coefficients were also examined for variation across age and gender.
[0092] Table 1 shows the results of fitting an additive linear mixed effects model to the gain kernel lag coefficients obtained from the response estimates, where there are 4 signed stimulus groups, such as Figure 6 The gain kernels are defined in column 4 of , and each group has 3 lag coefficients, indicating the gain weight for each 0.25 second inter-stimulus interval. Therefore, the response estimation model contains 12 gain kernel lag coefficients. The stimulus was P129 and the subjects were 94 normal control subjects. Since there were 2 pupils and 2 repetitions, the data set contains 94×2×2=376 sets of 12 gain kernel lag coefficients. The linear mixed effects model contains a factor for each of the 12 kernel lag coefficients, as well as 12 factors for the lags for females. An age effect was also fitted, but was not significant, so it is not shown. Table 1 summarizes the results. The model F statistic is 179 and the adjusted R2 The value is 0.698.
[0093] The intercept, 0.350, is the average kernel lag coefficient for the 12 pupil records for males. Because this is an additive model, the next 11 rows: Gain Kernel Coefficient 2 to Gain Kernel Coefficient 12 indicate the differences from the intercept and the significance of those differences. Many of the p values are very small and are shown as 0.000. Among the p values marked as p=0.000, the least significant value is p<8×10 -10 For these rows, the relative significance of the differences is best understood by examining the t-statistic (t-Stat column). The effect for females is less significant but cannot be ruled out. The bottom line is that high model R 2 0.698 means that among the 376 data sets, the gain kernel coefficient is very consistent for this 4-group, 3-lag model gain kernel model.
[0094] Table 2 shows the R values of the other sign groups / lagged coefficient response estimation models calculated for 94 normal control subjects. 2 Values. Data for P129 to P131 are included here, so there are 376 sets of gain coefficients for each of P129 to P131. In fact, the data here are a little different from the data in Table 1. Here, the repeated data pairs for each pupil are submitted to a bootstrap sampling process with replacement. In this process, 64 synthetic data sets are created from each pair by randomly sampling and replacing 9 fragments of each data set. The responses of these 64 data sets are estimated. The idea of these bootstrap cross-validation methods in statistics is to obtain results that more closely reflect the population mean. Therefore, the responses of 376×64=24,064 data sets are estimated for each of P129 to P131. Subsequently, the median value is taken for the 64 values of each synthetic data set, and these more robust cross-validation data are submitted to the linear mixed effects model, as shown in Table 1.
[0095] Table 1
[0096]
[0097] The grayscale background of the cells in Table 2 helps to intuitively understand which response estimation method produces R 2 As mentioned, a larger value means that the linear model of the mean gain kernel coefficient (such as the linear model in Table 1) accounts for a large proportion of the variance, so when R 2 When , the coefficient varies little across the population. Group 1, 2, and 4 methods use Figure 5 The sign of stimulus group - column 1 (group 1, shared gain) and Figure 6The symbol stimulus groups - Column 4 (Group 4, by half field) and Column 6 (Group 2, switch eye). Surprisingly, the switch eye symbol stimulus group combined with the 3 lags, Group 2, Lag 3 model and its 6 gain kernel coefficients in Row 4 of Table 2, gave almost as good agreement (0.740 for P129) as the complex Group 4, Lag 4 model and its 16 gain kernel lag coefficients (0.757 for P129).
[0098] Table 2
[0099]
[0100] Table 3 compares the different group / lag response estimation methods in terms of diagnostic power and repeatability measures. The data were obtained from normal controls and glaucomatous subjects at P129. Because the glaucoma in the patients was generally mild and all 18 eyes of the patients would have been considered normal by any diagnostic criteria, the ROC values would not be expected to be very high. Therefore, the %ROC values for sensitivity and latency are intended to give a relative measure of diagnostic power. In turn, diagnostic power is a proxy for depicting true differences in visual fields, whether intended for diagnostic purposes or otherwise. The R 2 The value is the average variance accounted for by the response estimation model, and as with %ROC, larger numbers are better. In the grayscale coloring scheme, lighter backgrounds in cells perform better. Srep and Drep are measures of relative repeatability between replicate pairs, and for these, smaller numbers are better. Unless otherwise noted, the response estimation models have 3 lag coefficients per group in their gain kernel. The symbol stimulus group type is indicated by the Group column. These refer to Figure 5 and Figure 6 So COL2 means Figure 5 Column 2, COL5 indicates Figure 6 5 in , and so on. As shown in Table 2, the best performing model in terms of lowest complexity is the 2-group, switch-eye sign stimulus group assignment. It is much better than the 2-group, within-eye method with the same number of gain kernel coefficients.
[0101] Table 3
[0102]
[0103] The bottom two rows of Table 3 labeled Set Gain are another type of response estimation model. Here, cross-validated fits of two promising cases are taken from Table 2, the mixed effects model shown in Table 1 is fitted, and then the data for all subjects are forced to fit the average gain kernel of these two cases when performing response estimation. Performance is worse than allowing each pupil to find the model with the best fit. The Set Gain method demonstrates another non-restrictive variation on the basic demonstration. This shows that using a response estimation model tailored to the dynamic gain characteristics of each individual pupil outperforms the common practice of using standard function extensions such as the Weiner kernel.
[0104] Table 4
[0105]
[0106] The question arises, what makes the switching eye set efficient for both groups and the symbolic stimulus group? The values of the gain kernel lag coefficients for several conditions were investigated. Here, the model shown in Table 1 was fitted and the mean values of the gain kernel coefficients for the predicted 50-year-old population were examined. Table 4A is Figure 5 The 8 gain kernels assigned to the 8 groups in column 2 are fitted with two lags. Lag 0 is the instantaneous gain, and lag 1 is the effect of going back 0.25 seconds on the gain. There is a clear pattern at lag 0: groups 2 and 6 have the largest gains. Table 4 B and C are Figure 6 The group designations in column 4, where Table 4B is the data from females and Table 4C is the data from males. They are 3 lag fits. By examining columns 2 and 4, it is clear that they are similar, but in column 4, similar half-field series are equated by whether they are 3-ring series or 2-ring series. As might be expected, group 2 has the largest lag 0 gain, which corresponds to groups 2 and 6 in Table 4A. The gain for males is slightly less, but the pattern is similar for lag 0, and similar for lag 2.
[0107] Table 4D shows the best 2 groups: the switch eye symbol stimulus group assignment, i.e., Figure 6 Column 6, 3 lag coefficients are fitted. This has the interesting property that the half-field series for the same eye are summed together, meaning that each step for group 1 involves a jump to the other eye after a 0.25 second time step, and similarly for group 2, and for all cycles of the stimulus. This assignment is therefore called: switch-eye. Fitting the gain kernel for these sign groups retains the large lag 0 gain for group 2 seen in the more complex models previously. Table 4E shows the 2-lag version of the switch-eye group, with similar results for lag 0 as in Table 4D. Table 4F is a graph using Figure 6Column 5 shows an example of a symbol group assignment; a so-called intra-eye assignment. Here, group 1 is all the series shown to the left eye, and group 2 is all the series shown to the right eye. The pattern of high gain for group 2 lag 0 seen in Table 4 A to E seems to have disappeared. The results in Table 3 comparing the 3-lag version of switch eye to intra-eye show that, although switch eye has the same number of groups and gain kernel lag coefficients as intra-eye, it performs surprisingly better than intra-eye.
[0108] The fact that switching the eye works so well suggests that there must be an eye-dependent component to the gain. This was not expected because previous studies reported in US 9,848,771 showed that gain changes occur in the Edinger-Westphal nucleus, where inputs from both eyes are aggregated and summed, seemingly removing the possibility of eye-dependent gain changes. Switching the eye assignment groups together stimuli from the eye that has not seen a stimulus in the last 0.5 seconds, thereby capturing the higher dynamic gain state they produce. Taken together, this evidence suggests that there is an unexpectedly strong response modulation at the level of the pretectal-olivary nucleus or beneath the pathway from the retina to the pupil.
[0109] According to one example, where there can be more than one type of complementary set of half-fields or quadrants, for example, if the multifocal array consists of 5 stimulus rings centered on fixation, and half of the half-fields or quadrants are drawn from either rings 1, 3, 5 or from rings 2, 4, thereby forming two types of half-field or quadrant salvos, such that many cycles of the salvo will have 8 types of salvos, but again, only two symbol stimulus sets and two gain kernels can be measured to capture the switching eye patterns assigned by the symbol stimulus sets.
[0110] Overall, the objectives of the experiment have been achieved. The estimated gain kernel and the temporal impulse response of each region together improve the value of the estimated response. The specific parsimonious assignment of the symbolic stimulus group allows fitting response estimation models that are as efficient and consistent as estimation models with 2 to 4 times more gain kernel lag coefficients. These findings are completely unexpected and are not anticipated by any prior art. It is also advantageous to include the gain kernel in the response estimation model in order to capture the dynamic changes of gain.
[0111] Demo 2
[0112] In the first demonstration, the basis functions used were parametrically defined lognormal functions (Eq. 5). The paper mentions the possibility of using bilinear basis functions instead of parametric basis functions. It is thus shown that the advantages of estimating the gain kernel also appear when estimating alternative basis functions. For this demonstration, data from 170 normal control subjects were used. These individuals were tested with the previously described multifocal stimulation of P129 and P130 with 44 fields per eye, both eyes were tested concurrently and with the Figure 2 The device shown in Figure 2 records both pupils. Each subject was tested twice approximately 2 weeks apart. Figure 6 The symbol stimulus group in column 4 was assigned to half the visual field, and four sets of response estimates were obtained without and with the gain kernel. Therefore, the resulting fitted gain coefficients are similar to those in Table 4, B and C.
[0113] Fig. 9 The results of averaging the estimated bilinear basis functions for pupils, subjects, and repetitions are shown, demonstrating the effect of including a gain kernel in the response estimation for the P129 stimulation protocol. Therefore, the waveform is the average of 170 subjects × 2 pupils × 2 repetitions = 680 estimated bilinear basis functions. The legend indicates that the black curve (gain in the legend) is the average basis function under the condition of estimating the gain kernel as described above. The gray curve (no gain in the legend) is the result when the gain kernel is not fitted. The basis function waveforms are significantly different. The basis function obtained with the estimated gain kernel reaches a peak later and recovers to the baseline pupil size (0 on the vertical scale) more slowly. Therefore, fitting the gain kernel changes the timing of the estimated response.
[0114] Along with estimating the basis functions for each pupil, the response estimation process of Equations 9 and 10 also estimates 44 coefficients (i.e., weights) to be applied to each basis function to create an estimated response waveform for each region. For a given pupil, there is a 61×1 vector bf containing the basis function waveforms. There is also a 44×1 vector cf containing the values (i.e., weights) of the coefficients of each basis function to be applied to each of the 44 P129 test regions for each eye. Then the matrix multiplication: wf=bf×cf T A 61×44 matrix wf of the estimated response waveforms of the individual basis functions is created, where T is the transpose. Given this outer product of vectors, the process can be called bilinear. In fact, bf and cf are estimated simultaneously for the 88 regions covered by each pupil recording.
[0115] Table 5 summarizes the differences in the results for the two types of response estimation methods: estimation with (Gain) and without a gain kernel (NoGain). All table entries are medians of the 680 estimates. Given the large numbers, any differences in the means are highly significant, even allowing for multiple comparisons. The amplitude of peak pupil constriction is much larger for the gain kernel case (second row, labeled Gain) than for the case without fitting a gain kernel (first row, labeled NoGain). This is true for both P129 and 130. The t-statistic (i.e., the amplitude of constriction for each region divided by its SE for each region) provides a measure of the signal-to-noise ratio for each region and is much larger for the gain kernel version. Similarly, the gain kernel contains the mean R for the model 2 The value is larger. Fig. 9As mentioned, the time to reach the peak is longer. Overall, adding the gain kernel during the response estimation process improves the goodness of fit and the signal-to-noise ratio when using a single bilinear basis function per pupil.
[0116] Table 5
[0117]
[0118] Demo 3
[0119] As indicated in Equation 10 and the associated text, there may also be two or more (l) bilinear basis functions per pupil, where during iterative fitting, singular value decomposition is used to estimate for j=1,...,l For the current demonstration with 2 orthogonal basis functions per pupil, use Figure 6 The Hemifields sign stimulus group assignments in column 4 estimate gain kernels for the 4 groups, as in the previous demonstration. The resulting fitted gain coefficients are therefore similar in form to those in Table 4, B, C, and like those, this preserves the large lag 0 gain of group 2 seen for the switch eye assignments to the sign group.
[0120] Fig.10 The figure uses two orthogonal basis functions to estimate the response of a single pupil. In this example, the pupil is the left pupil and Fig.10 The responses shown in A are those produced by stimulating the left eye of a 43-year-old male with the 44 area P129 stimulus described above. The 44 estimated responses are Fig.10 44 waveforms with different gray levels are shown in A. It is noteworthy that the 44 estimated waveforms per region have different peak contraction amplitudes and times to peak. These are achieved by summing different mixtures of the two basis functions found by the iterative fitting process, as shown in the text associated with Equation 10 above.
[0121] For a single bilinear basis function, matrix multiplication is used to construct from two basis functions Fig.10 44 responses of A. Consider the 61×2 matrix bf, where the values of the columns describe the l=2 basis function waveforms. Next consider the 44×2 matrix cf containing the values d as the coefficients (i.e., weights) of each basis function to be applied to each of the 44 P129 test regions for each eye. Then the matrix multiplication wf=bf×cf T Create a 61×44 matrix wf of the estimated response waveform, where T is the transpose. Alternatively, each basis function of each column bf1 and cf1 can be calculated in two steps: and Therefore wf = wf1 + wf2. In this case, wf1 and wf2 can be considered as two sets of waveforms of each region response component based on its respective basis function. Fig.10 This process is illustrated for one pupil of one eye of a subject, where Fig.10 The waveform of A is Fig.10 The sum of the waveforms in B and 10C.
[0122] Table 6
[0123]
[0124] Table 6 shows the results of including and not including the gain kernel in the estimation of the bilinear basis function method. The top row, labeled No Gain on the left, is the data obtained when the gain kernel was not included in the estimation, and the second row labeled Gain is when the gain kernel was included in the response estimation. The bottom row is the ratio of the gain / no gain cases. As in Demonstration 2, the data are summarized for both eyes and pupils of 170 normal subjects who took two P129 tests. The first column is the median peak to peak amplitude of the response contraction in each region, and the second column is the t-statistic for those amplitudes. Next is the R 2 Goodness of fit statistics ± their standard deviation (SD). T statistics for each regional weight are also given (cf above), and response components 1 and 2 refer to the response components generated by basis functions 1 and 2. In all cases, including a gain kernel will improve all aspects of the fit value. The two basis function fits provide flexibility in terms of time to peak and waveform shape. If the stimulus has a long non-pulse time course, the flexibility of the waveform shape can be useful.
[0125] The data for component 2 in Table 6 do indicate a possible disadvantage of the bilinear basis function method: the t-statistic for component 2 is small. In fact, the absolute values of the t-statistics are taken before averaging, because they can be of any sign (e.g., Fig.10 C). However, they are less than 2, indicating that their average differences from 0 are not significantly significant. This may indicate that flexibility is gained by using a parametric basis function such as the lognormal function of Equation 5, but also including its time derivative in the response estimation process, thereby adding an additional parameter per region.
[0126] Demo 4
[0127] So far, it has been assumed that when estimating the gain kernel, the contrast sequence of Equation 1 and c nPositive and negative values of should be treated equally. As shown in Equation 11 and the associated text, the stimulus data can be split into positive (ON) and negative (OFF) contrasts, and separate gain kernels can be estimated for each contrast, allowing a wider class of response models to be estimated. In particular, the model can cover the possibility that the pupil system has different gains for positive and negative contrast sequences. Here a single lognormal parametric basis function is used for each test region, and each basis function has its own peak value estimated for each region, i.e., amplitude and time to peak.
[0128] Table 7 shows Figure 6 Column 4 shows the fitted gain kernel coefficients for the ON and OFF contrasts for the four groups assigned by Hemifields for the sign stimulus group. As before, three lags are included. Therefore, Table 7 follows the format of Tables 4B, C, but with separate entries for the ON and OFF gain coefficients estimated simultaneously. As with Tables 4B, C, the largest gain is for Group 2 at lag 0. Upon investigation, the OFF gain is actually greater than the ON, especially for Group 2 at lag 0, which is an unexpected and favorable result. The t-statistics for the coefficients labeled Groups 1 to 4 in the left column are given in the four rightmost columns labeled similarly. The lag 1 t-statistic for the OFF gain coefficient is surprisingly significant compared to the ON gain. This is also unexpected and reveals more information about the pupil system.
[0129] Table 7
[0130]
Claims
1. A method for evaluating a subject's visual nervous system, the method comprising the following steps: presenting a multifocal set of stimuli within portions of one or more visual fields of the subject, the multifocal set of stimuli being suitable for eliciting a pupillary response from at least one pupil of the subject, the multifocal set of stimuli comprising a plurality of separate stimulus elements presented to different components of the one or more visual fields, the separate stimulus elements being presented concurrently in a statistically independent sequence, wherein said multifocal set of stimuli is presented based on a set of symbolic stimuli formed by subsets of said statistically independent sequences, using sensors to detect responses to a multifocal set of presented stimuli, Analyze the detected responses, measuring dynamic changes in pupil response gain based on the detected response, and A temporal impulse response to each of the plurality of individual stimulus elements is measured based on the measured dynamic changes in pupil response gain.
2. A method as claimed in claim 1, wherein at any time step, the multifocal sets of stimuli are presented in series of the total number of the individual stimulus elements, and those series are in turn assigned to the same or less number of symbolic stimulus groups, and the dynamic changes in pupil response gain are measured only for the individual stimulus elements defined by the same or less number of symbolic stimulus groups.
3. The method of claim 2, wherein: At a particular time step, a multifocal set of stimuli is presented to the fields of view of the two pupils in a cyclic salvo, wherein the salvo of the multifocal set of stimuli includes individual stimulus elements randomly selected from stimulus elements of complementary left or right halves of the fields of view of the two pupils, and at successive time steps, the salvo of the multifocal set of stimuli is presented to the left half of the field of view of the left pupil, then to the right half of the field of view of the left pupil, then to the left half of the field of view of the right pupil, and then to the right half of the field of view of the right pupil, and wherein the probability of individual stimulus elements from either the left and right halves of the fields of view of the left and right pupils being presented is approximately 50%, and the sequence of salvos is cyclically repeated multiple times, and the stimulus elements for the left and right halves of the fields of view of the left and right pupils are assigned to symbolic stimulus groups.
4. A method as claimed in claim 3, wherein the stimulus elements presented to the left half of the visual field of the left pupil and the right pupil are assigned to a first symbol stimulus group, and the stimulus elements presented to the right half of the visual field of the left pupil and the right pupil are assigned to a second symbol stimulus group, and the dynamic change of the pupil response gain is measured only for the first symbol stimulus group and the second symbol stimulus group, and the impulse response is measured for each stimulus element based on the measured dynamic change of the pupil response gain.
5. The method of claim 3, wherein the stimulus elements presented to the left half of the visual field of the left pupil are assigned to a first symbol stimulus group, the stimulus elements presented to the right half of the visual field of the left pupil are assigned to a second symbol stimulus group, the stimulus elements presented to the left half of the visual field of the right pupil are assigned to a third symbol stimulus group, and the stimulus elements presented to the right half of the visual field of the right pupil are assigned to a fourth symbol stimulus group, and the dynamic changes of the pupil response gain are measured only for the first symbol stimulus group, the second symbol stimulus group, the third symbol stimulus group, and the fourth symbol stimulus group, and the impulse response is measured for each stimulus element based on the measured dynamic changes of the pupil response gain.
6. A method as claimed in claim 3, wherein the stimulus elements presented to the upper half of the visual field of the left pupil are assigned to a first symbol stimulus group, the stimulus elements presented to the lower half of the visual field of the left pupil are assigned to a second symbol stimulus group, the stimulus elements presented to the upper half of the visual field of the right pupil are assigned to a third symbol stimulus group, and the stimulus elements presented to the lower half of the visual field of the right pupil are assigned to a fourth symbol stimulus group, and the dynamic changes of the pupil response gain are measured only for the first symbol stimulus group, the second symbol stimulus group, the third symbol stimulus group and the fourth symbol stimulus group, and the impulse response is measured for each stimulus element based on the measured dynamic changes of the pupil response gain.
7. A method as claimed in claim 3, wherein the symbol stimulus group is arranged based on a first quadrant of stimulus elements for the left pupil and a second quadrant of stimulus elements for the right pupil, wherein the first quadrant and the second quadrant are diagonally opposite to each other, and the dynamic change of the pupil response gain is measured only for the symbol stimulus group, and the impulse response is measured for each stimulus element based on the measured dynamic change of the pupil response gain.
8. The method of claim 7, wherein one or more of the symbolic stimulus groups further comprises one or more additional stimulus elements outside of the first quadrant and the second quadrant.
9. The method of claim 3, wherein two symbol stimulus sets and two gain kernels are measured to capture the switching eye patterns assigned by the symbol stimulus sets.
10. The method of claim 3, wherein if there are N types of half-field or quadrant salvos, then N / 2 symbol stimulus groups are assigned to capture the switching eye patterns assigned by the symbol stimulus groups.
11. A method as claimed in claim 2, wherein the field of view of two pupils is stimulated by presenting a multifocal set of the stimulus, the multifocal set of the stimulus comprising four subsets of stimulus elements, the four subsets of stimulus elements stimulating a first complementary part and a second complementary part of the field of view of the two pupils, wherein the assignment of the two symbol stimulus groups follows the following order: the first complementary part presented to the first pupil is assigned to the first group of symbol stimulus groups, the second complementary part presented to the first pupil is assigned to the second group of symbol stimulus groups, the first complementary part presented to the second pupil is assigned to the first group, the second complementary part presented to the second pupil is assigned to the first group, and the impulse response is measured for each stimulus element based on the dynamic change of the pupil response gain measured only for the first group and the second group.
12. The method of claim 11, wherein the first complementary portion and the second complementary portion of the multifocal set of stimuli include sporadic deviations from a set pattern of the first complementary portion and the second complementary portion.
13. A method as claimed in any one of claims 1 to 12, wherein the temporal impulse response is measured using a log-normal distribution as a parametric function of a finite number of parameters for each portion of the field of view.
14. The method of claim 13, wherein the time derivative of the parameter function is used in the response measurement.
15. The method of any one of claims 1 to 12, wherein the temporal impulse response is measured as a set of one or more bilinear orthogonal basis functions.
16. A computer system for evaluating a subject's visual nervous system, the computer system comprising: computer, a computer-controlled light source arranged to present a multifocal set of stimuli within portions of one or more visual fields of a subject, the multifocal set of stimuli being suitable for eliciting a pupillary response from at least one pupil of the subject, the multifocal set of stimuli comprising a plurality of individual stimulus elements presented to different components of the one or more visual fields, the individual stimulus elements being presented concurrently in a statistically independent sequence, wherein the light source is further arranged to present, under control of the computer, a multifocal set of said stimuli based on a set of symbolic stimuli formed by a subset of said statistically independent sequences, at least two sensors arranged to detect responses to a multi-focal set of presented stimuli, The computer is arranged as: Analyze the detected responses, measuring dynamic changes in pupil response gain based on the detected response, and A temporal impulse response to each of the plurality of individual stimulus elements is measured based on the measured dynamic changes in pupil response gain.
17. A computer system as claimed in claim 16, wherein the light source is controlled by the computer to present the multifocal set of stimuli in series of the total number of said individual stimulus elements at any time step, and those series are in turn assigned to an equal or lesser number of symbolic stimulus groups, and the computer is arranged to measure dynamic changes in pupil response gain only for the individual stimulus elements defined by the equal or lesser number of symbolic stimulus groups.
18. A computer system as described in claim 17, wherein the light source is controlled by the computer to present a multifocal set of stimuli to the fields of view of the two pupils in a cyclic salvo at a specific time step, wherein the salvo of the multifocal set of stimuli includes randomly selecting individual stimulus elements by the computer from the stimulus elements presented to the complementary left or right halves of the fields of view of the two pupils, and at consecutive time steps, the light source is controlled by the computer to present the salvo of the multifocal set of stimuli to the left half of the field of view of the left pupil, then to the right half of the field of view of the left pupil, then to the left half of the field of view of the right pupil, and then to the right half of the field of view of the right pupil, and wherein the probability of individual stimulus elements from any one of the left and right halves of the field of view of the left and right pupils being presented is approximately 50%, and the sequence of salvos is repeated multiple times in a cycle, and the stimulus elements for the left and right halves of the field of view of the left and right pupils are assigned to the symbolic stimulus group.
19. A computer system as claimed in claim 18, wherein the stimulus elements presented to the left half of the visual field of the left pupil and the right pupil are assigned by the computer to a first symbol stimulus group, and the stimulus elements presented to the right half of the visual field of the left pupil and the right pupil are assigned by the computer to a second symbol stimulus group, and the computer is arranged to measure the dynamic change of the pupil response gain only for the first symbol stimulus group and the second symbol stimulus group, and is also arranged to measure the impulse response for each stimulus element based on the measured dynamic change of the pupil response gain.
20. A computer system as claimed in claim 18, wherein the stimulus elements presented to the left half of the visual field of the left pupil are assigned by the computer to a first symbol stimulus group, the stimulus elements presented to the right half of the visual field of the left pupil are assigned by the computer to a second symbol stimulus group, the stimulus elements presented to the left half of the visual field of the right pupil are assigned by the computer to a third symbol stimulus group, and the stimulus elements presented to the right half of the visual field of the right pupil are assigned by the computer to a fourth symbol stimulus group, and the computer is arranged to measure the dynamic changes of the pupil response gain only for the first symbol stimulus group, the second symbol stimulus group, the third symbol stimulus group and the fourth symbol stimulus group, and is also arranged to measure the impulse response for each stimulus element based on the measured dynamic changes of the pupil response gain.
21. A computer system as described in claim 18, wherein the stimulus elements presented to the upper half of the visual field of the left pupil are assigned by the computer to the first symbol stimulus group, the stimulus elements presented to the lower half of the visual field of the left pupil are assigned by the computer to the second symbol stimulus group, the stimulus elements presented to the upper half of the visual field of the right pupil are assigned by the computer to the third symbol stimulus group, and the stimulus elements presented to the lower half of the visual field of the right pupil are assigned by the computer to the fourth symbol stimulus group, and the computer is arranged to measure the dynamic changes of the pupil response gain only for the first symbol stimulus group, the second symbol stimulus group, the third symbol stimulus group and the fourth symbol stimulus group, and is also arranged to measure the impulse response for each stimulus element based on the measured dynamic changes of the pupil response gain.
22. A computer system as described in claim 18, wherein the symbolic stimulus group is arranged by the computer based on a first quadrant of stimulus elements for the left pupil and a second quadrant of stimulus elements for the right pupil, wherein the first quadrant and the second quadrant are diagonally opposite, and the computer is arranged to measure the dynamic change of the pupil response gain only for the symbolic stimulus group, and is also arranged to measure the impulse response for each stimulus element based on the measured dynamic change of the pupil response gain.
23. The computer system of claim 22, wherein one or more of the symbolic stimulus groups further comprises one or more additional stimulus elements outside of the first quadrant and the second quadrant.
24. The computer system of claim 18, wherein the computer is arranged to measure two symbol stimulus sets and two gain kernels to capture the switching eye patterns assigned by the symbol stimulus sets.
25. A computer system as described in claim 18, wherein if there are N types of half-field or quadrant salvos, then N / 2 symbol stimulus groups are assigned by the computer to capture the switching eye patterns assigned by the symbol stimulus groups.
26. The computer system of claim 17, wherein: Under the control of a computer, a light source is arranged to stimulate the field of view of two pupils by presenting a multifocal set of said stimuli, said multifocal set of stimuli comprising four subsets of stimulus elements stimulating first and second complementary parts of the field of view of the two pupils, wherein the assignment of the two symbolic stimulus groups follows the following order: the first complementary part presented to the first pupil is assigned by the computer to the first group of symbolic stimulus groups, the second complementary part presented to the first pupil is assigned by the computer to the second group of symbolic stimulus groups, the first complementary part presented to the second pupil is assigned by the computer to the first group, and the second complementary part presented to the second pupil is assigned by the computer to the first group, wherein the computer is arranged to measure an impulse response for each stimulus element based on a dynamic change of pupil response gain measured only for the first group and the second group.
27. The computer system of claim 26, wherein the first complementary portion and the second complementary portion of the multifocal set of stimuli include sporadic deviations from a set pattern of the first complementary portion and the second complementary portion.
28. A computer system as claimed in any one of claims 16 to 27, wherein the computer is arranged to measure the temporal impulse response as a parametric function of a finite number of parameters for each part of the field of view using a log-normal distribution.
29. A computer system as claimed in claim 28, wherein the computer is arranged to use the time derivative of the parameter function in the response measurement.
30. A computer system as claimed in any one of claims 16 to 27, wherein the computer is arranged to measure the time impulse response as one or more bilinear orthogonal basis functions.
