Intermittent external strabismus early screening device for ophthalmology department

Abstract

The invention relates to the technical field of strabismus detection, in particular to an ophthalmic intermittent external strabismus early screening device which comprises a gaze analysis module, a direction analysis module, a drift correction module, a screening calculation module and a result optimization module. According to the method, the eyeball displacement sequence in the watching process of the two eyes is collected, the direction distribution density of the displacement difference vector is analyzed, the tiny change of the eyeball movement is accurately captured, the eye position offset can be analyzed more finely, and the stability attenuation of the eyeball movement is revealed based on the direction set center coordinate and the density center transfer rate calculation and comparison. Abnormal signals are marked, the precision of early-stage abnormality discovery is improved, drift in an eyeball movement rule is corrected, a balance point is generated through probability correction, the situation of missed diagnosis is avoided, continuous deviation with poor convergence is analyzed, eye position changes are dynamically reconstructed, early-stage external strabismus is accurately judged, and technical means are combined, so that the precision of early-stage screening is greatly improved. And an accurate auxiliary judgment basis is provided for clinical screening.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of strabismus detection, in particular to an early screening device for ophthalmic intermittent exotropia. BACKGROUND

[0002] The technical field of strabismus detection includes detection and analysis methods for abnormal binocular vision function of the human eye, mainly studying the determination of eye position, movement state and coordination ability through optical imaging, gaze tracking and physiological signal monitoring. The core content includes obtaining eye position change data of the subject when gazing at different direction targets by using visual stimulation and imaging detection equipment, comparing and analyzing the offset and movement law of both eyes when gazing at the target, and judging whether there is strabismus and its type. It also relates to basic research directions such as image acquisition structure, pupil positioning mechanism, infrared irradiation detection method for clinical screening and auxiliary diagnosis, in order to realize accurate detection and early abnormal recognition of eye movement law.

[0003] Among them, the early screening device for ophthalmic intermittent exotropia refers to a detection device for detecting and judging the occurrence characteristics of exotropia of the subject in the natural gazing state. It mainly covers recording the eye movement trajectory when both eyes gaze at the target through the optical image acquisition unit, realizing the quantitative determination of eye position deviation through the principle of optical axis projection and reflection, and determining the time interval and frequency of exotropia occurrence combined with the real-time tracking of the reference target position. Through the structured detection means, the synchronous acquisition and comparison analysis of the eye under different gazing conditions are realized, and the eye position capture in different gazing directions is completed relying on adjustable light and imaging angle, so as to realize the screening of early characteristics of intermittent exotropia.

[0004] The main shortcomings of the prior art are reflected in the accuracy and dynamic change capture of eye movement law. In actual operation, the traditional technology obtains eye position data through optical imaging, gaze tracking and physiological signal monitoring, but it has certain limitations in capturing subtle changes in eye movement. Especially in the case of small eye position changes or low eye movement frequency, the existing technology may not be able to accurately capture these changes, thus missing the early warning signals of early strabismus. In addition, the existing technology mainly relies on rough comparison of eye position deviation and movement law in the analysis of eye movement stability, and lacks detailed analysis of direction concentration and density center shift. This leads to the inability to identify abnormal fluctuations in time when the eye movement is unstable, thus affecting the screening effect of early strabismus. Therefore, the existing technology has great shortcomings in the accuracy of judging small eye position changes and early strabismus, and may not be able to detect potential exotropia cases in time and effectively, thus affecting the effect of early intervention. SUMMARY

[0005] To solve the technical problems in the prior art, an early screening device for intermittent exotropia in ophthalmology is provided in embodiments of the present application. The technical solution is as follows: In one aspect, an early screening device for intermittent exotropia in ophthalmology is provided, which comprises: a gaze analysis module, which collects a binocular gaze displacement sequence of a subject during a gaze target point and divides sampling sections according to equal time, calculates a displacement difference vector group of adjacent sampling points in the plurality of sampling sections, and performs direction distribution density statistics to generate a direction concentration degree and deliver it to a direction analysis module; a direction analysis module, which extracts a direction set center coordinate of a continuous sampling section based on the direction concentration degree, calculates an angle change ratio between density centers of adjacent sampling sections, and analyzes a density center transfer rate, and if the density center transfer rate exceeds a stability threshold, marks a stability attenuation section and delivers it to a drift correction module; a drift correction module, which obtains a gaze center coordinate of an adjacent time window and a drift mean value of a previous stable section based on the stability attenuation section, and performs weighted fusion to calculate a drift balance point position, performs probability correction on the drift balance point position of a continuous time window, generates a drift balance point correction amount, and delivers it to a screening calculation module; a screening calculation module, which calculates a convergence difference based on the drift balance point correction amount and the density center transfer rate, and if the convergence difference continuously deviates from a convergence difference threshold in a plurality of time windows, performs nonlinear reconstruction on the section, and generates an early exotropia screening result; a result optimization module, which performs smoothing, re-weighting, and feature clustering based on the early exotropia screening result, the drift balance point correction amount, and the density center transfer rate, extracts an abnormal fluctuation section and performs correction, and generates an optimized early exotropia screening result.

[0006] As a further scheme of the present application, the direction concentration degree comprises a gaze displacement direction distribution density, a displacement difference vector group, and a direction offset amplitude, the density center transfer rate comprises a direction set center coordinate change rate, an adjacent section angle change ratio, and a stability threshold comparison result, the drift balance point correction amount comprises a drift mean value weighting, a drift probability correction parameter, and a drift balance point position deviation, the early exotropia screening result comprises a convergence difference distribution, a nonlinear reconstruction coefficient, and an abnormal section label, and the optimized early exotropia screening result comprises a smoothing drift balance point, a drift balance point correction amount, and an abnormal fluctuation correction.

[0007] As a further scheme of the present application, the gaze analysis module comprises: a gaze data submodule, which obtains a binocular gaze displacement sequence of a subject during a gaze target point, segments and samples continuous gaze positions according to equal time intervals, and performs time stamp synchronization and noise removal to generate a gaze displacement sequence set; A displacement difference calculation sub-module calculates, based on the gaze displacement sequence set, a difference value vector group in horizontal and vertical directions for displacement coordinates of adjacent sampling points in a plurality of sampling segments, analyzes a difference set of adjacent vectors and eliminates abnormal difference values, calculates an angle set of direction distribution, and generates a displacement difference vector group data set; A direction density sub-module performs frequency distribution statistics on a plurality of vector direction angles according to the displacement difference vector group data set, extracts a plurality of direction frequency calculation normalized density values, and iteratively calculates a concentration degree of all direction interval densities, and generates a direction concentration degree.

[0008] As a further scheme of the present application, the direction analysis module comprises: A density center extraction sub-module extracts a direction set center coordinate in all sampling segments based on the direction concentration degree, calculates a product of direction distribution frequency and a corresponding angle weight value, calculates a distribution gravity center coordinate according to a weight value sum, and generates a direction density center coordinate set. An angle change calculation sub-module calls the direction density center coordinate set, calculates an included angle difference value for density center coordinate points of adjacent sampling segments, calculates an angle change ratio according to a time interval between adjacent segments, and generates a density center angle change rate set. A stability determination sub-module compares a plurality of sampling segment angle change rate values with a set stability threshold according to the density center angle change rate set, identifies sampling segments that exceed the stability threshold, and generates a stability decay segment.

[0009] As a further scheme of the present application, the stability threshold is obtained by obtaining a direction concentration degree sequence of the measured individual in consecutive sampling segments, calculating an absolute amplitude of a difference value of the direction concentration degree between adjacent segments, and calculating an average fluctuation interval of the difference value set of all samples.

[0010] As a further scheme of the present application, the drift correction module comprises: A data extraction sub-module obtains adjacent time window gaze center coordinates and a previous stable segment drift mean value based on the stability decay segment, performs difference calculation in a spatial coordinate system, and generates a time window drift difference value set by statistically analyzing a drift change sequence in time windows and performing linear smoothing. A balance point position sub-module calls the time window drift difference value set, extracts a proportionality coefficient between a drift amplitude of adjacent time windows and a drift mean value of a previous stable segment, calculates a weighted drift correction of a plurality of time windows, and generates a drift balance point position sequence by performing sliding average analysis on a center trend of a smoothing interval. A probability correction sub-module extracts an occurrence frequency of point position distribution and a consistency of drift direction in consecutive time windows according to the drift balance point position sequence, calculates an occurrence probability of a drift direction difference in a plurality of time windows and analyzes a probability density, and generates a drift balance point correction amount by weighted fusion of the probability density and the drift consistency.

[0011] As a further scheme of the present application, the screening calculation module comprises: The convergence difference calculation submodule calculates the change gradient values of the drift equilibrium point correction amount and the density center transfer rate in the same time window respectively based on the drift equilibrium point correction amount and the density center transfer rate, analyzes the difference value sequence in multiple time windows, and performs weighted average on the difference value sequence to generate a convergence difference; The threshold deviation detection submodule calls the convergence difference to obtain the difference value change sequence in multiple continuous time windows, performs interval judgment on the difference values in multiple time windows according to a set convergence difference threshold, generates a continuous deviation judgment quantity by detecting the number and amplitude of time windows deviating from the convergence difference threshold; The section reconstruction submodule selects the drift equilibrium point correction amount and the density center transfer rate data of the corresponding section according to the continuous deviation judgment quantity, performs multidimensional interpolation and curvature adjustment calculation on the numerical sequence in the deviated section to generate an early exotropia screening result.

[0012] As a further scheme of the present application, the convergence difference threshold is set by calculating the mean and standard deviation of the difference sequence of the drift equilibrium point correction amount and the density center transfer rate of the measured individual.

[0013] As a further scheme of the present application, the result optimization module comprises: The smoothing and reweighting submodule performs smoothing on the drift equilibrium point correction amount based on the early exotropia screening result, the drift equilibrium point correction amount and the density center transfer rate, calculates a weighting coefficient according to the change amplitude of the density center transfer rate and performs point-by-point multiplication with the smoothing result to generate a stability weighting sequence; The feature clustering submodule calls the time distribution features of the drift equilibrium point correction amount based on the weighted difference values in multiple time windows in the stability weighting sequence, performs multidimensional aggregation on the weighted difference values, determines the similarity interval between the aggregated vectors according to the Euclidean distance, and obtains a feature clustering partition coefficient; The anomaly correction submodule calls the time variation trend of the density center transfer rate according to the feature clustering partition coefficient, performs deviation comparison on the weighted difference mean values in multiple clustering sections, determines the time period exceeding the interval threshold as an abnormal fluctuation period and performs weighted balance adjustment to generate an optimized exotropia screening result; The interval threshold is set by statistically setting the distribution law of the eye movement indicators of the measured user in multiple stages of measurement.

[0014] The technical scheme provided by the embodiment of the present application has at least the following beneficial effects: By collecting the eye displacement sequence during binocular fixation and analyzing the direction distribution density of displacement difference vector in multiple time periods, the mode of eye movement of the subject can be effectively extracted. Compared with the prior art, this method captures the small changes of eye movement more finely and can perform more accurate offset analysis in the early stage of eye displacement. Then, by calculating and comparing the transfer rate of the center coordinates of the direction set and the density center, the stability decay of eye movement can be revealed, and the possible abnormal signals can be marked. The improvement of accuracy not only helps to detect abnormalities in eye movement earlier, but also corrects small drifts in eye movement rules and generates a balance point through probability correction. This precise drift correction mechanism significantly improves the detection ability of eye movement rules and effectively avoids missed diagnosis that may occur in traditional methods when the eye position changes are small. Finally, by analyzing the sustained deviation of convergence difference, the eye position changes can be dynamically reconstructed nonlinearly, so that the occurrence of early exotropia can be more accurately judged. The combination of these technical means greatly improves the screening accuracy of early features of intermittent exotropia and provides more efficient and accurate auxiliary judgment basis for clinical screening. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0016] Figure 1 is a schematic diagram of the device of the present application; Figure 2 is a schematic diagram of the device framework of the present application; Figure 3 is a flowchart of the fixation analysis module in the present application; Figure 4 is a flowchart of the direction analysis module in the present application; Figure 5 is a flowchart of the drift correction module in the present application; Figure 6 is a flowchart of the screening calculation module in the present application; Figure 7 is a flowchart of the result optimization module in the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the present application will be described below with reference to the drawings.

[0018] In the embodiments of the present application, the words such as "exemplary", "for example", etc. are used to represent an example, illustration, or description. Any embodiment or design scheme described as "exemplary" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "exemplary" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be either one of the two.

[0019] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times, and it should be pointed out that their meanings are consistent when their distinction is not emphasized. "Of", "corresponding" and "corresponding" can be used interchangeably at times, and it should be pointed out that their meanings are consistent when their distinction is not emphasized.

[0020] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1, and their meanings are consistent when their distinction is not emphasized.

[0021] In order to make the technical problems, technical schemes and advantages to be solved by the present application more clear, the following will be described in detail with reference to the drawings and specific embodiments.

[0022] The embodiments of the present application provide an early screening device for intermittent exotropia in ophthalmology, such as Figures 1-2 An early screening device for intermittent exotropia in ophthalmology is shown in a schematic diagram, which comprises: A gaze analysis module acquires a binocular gaze displacement sequence of a subject in a gaze target point and divides sampling sections according to equal time, calculates a displacement difference vector group of adjacent sampling points in a plurality of sampling sections and performs direction distribution density statistics, generates a direction concentration degree and transmits it to a direction analysis module; A direction analysis module extracts a direction set center coordinate of a continuous sampling section based on the direction concentration degree, calculates an angle change ratio between density centers of adjacent sampling sections and analyzes a density center transfer rate, if the density center transfer rate exceeds a stability threshold, marks a stability attenuation section and transmits it to a drift correction module; A drift correction module obtains adjacent time window gaze center coordinates and a drift mean value of a previous stable section based on the stability attenuation section and weighted fusion calculates a drift equilibrium point position, performs probability correction on the drift equilibrium point position of a continuous time window, generates a drift equilibrium point correction amount and transmits it to a screening calculation module; A screening calculation module calculates a convergence difference based on the drift equilibrium point correction amount and the density center transfer rate, if the convergence difference continuously deviates from a convergence difference threshold in a plurality of time windows, a nonlinear reconstruction is performed on the section, and an early exotropia screening result is generated; The result optimization module performs smooth reweighting and feature clustering based on the early exotropia screening result, the drift balance point correction amount and the density center transfer rate, extracts an abnormal fluctuation section and performs correction, and generates an optimized early exotropia screening result.

[0023] The direction concentration degree includes a gaze displacement direction distribution density, a displacement difference vector group and a direction offset amplitude, the density center transfer rate includes a direction set center coordinate change rate, an adjacent section angle change ratio and a stability threshold value comparison result, the drift balance point correction amount includes a drift mean weighted value, a drift probability correction parameter and a drift balance point position deviation, the early exotropia screening result includes a convergence difference distribution, a nonlinear reconstruction coefficient and an abnormal section label, and the optimized early exotropia screening result includes a smooth drift balance point, a drift balance point correction amount and an abnormal fluctuation correction.

[0024] Specifically, as shown in Figure 2 , 3 The gaze analysis module includes: A gaze data submodule acquires a binocular gaze displacement sequence of a subject during a gaze target point process, segments and samples continuous gaze positions according to equal-length time intervals, performs time stamp synchronization and noise elimination, and generates a gaze displacement sequence set; Firstly, the eye tracker with a sampling frequency of 1000 Hz is used to collect the raw gaze position data of the subject when gazing at a fixed target point with a diameter of 0.5 cm in the center of the screen for 10 seconds. The raw data collected is a series of data points containing time stamps (unit: milliseconds), left eye coordinates (pixels), and right eye coordinates (pixels). Subsequently, the continuous gaze position data is sampled in segments with a time interval of 100 milliseconds. In each 100-millisecond time period, the timestamps of the left and right eyes are synchronized. The specific actions are as follows: traverse all left eye data points in the time period, for each left eye data point, find the right eye data point with the closest timestamp in the same time period, if the absolute value of the difference between the two timestamps is less than 1 millisecond, then pair the two data points; if the difference is greater than or equal to 1 millisecond, discard the left eye data point. After successful pairing, take the average of the left eye coordinates and right eye coordinates in the pair of data points as the binocular fusion gaze coordinates at the synchronized time point. For example, in a time period, the left eye data point (time: 101.2 ms, X: 962, Y: 541) and the right eye data point (time: 101.5 ms, X: 960, Y: 543) are obtained, the difference between the two timestamps is 0.3 milliseconds, which is less than 1 millisecond, so a fusion point with coordinates ((962+960) / 2, (541+543) / 2), i.e. (961, 542) is generated. Next, noise removal is performed, where noise refers to the sudden jump in coordinates caused by blinking or loss of device signal. The displacement threshold for noise judgment is set to 50 pixels. This threshold is determined through a verification experiment involving 30 healthy subjects. In the experiment, each subject was asked to perform 5 blinking actions while recording eye movement data and manually marking the time window of the blinking. The maximum displacement between two consecutive sampling points (interval 1 millisecond) in the non-blinking state of all subjects was 48.7 pixels; while the minimum displacement during blinking was more than 60 pixels. Therefore, 50 pixels is selected as the threshold to distinguish normal eye movement from noise. The specific removal action is: calculate the Euclidean distance between each synchronized fusion gaze point and its previous fusion gaze point, if the distance is greater than 50 pixels, the current gaze point is judged as a noise point and removed. For example, the previous valid gaze point coordinates are (961, 542), the current point coordinates are (850, 430), the displacement distance is pixels, which is greater than 50 pixels, so the data point with coordinates (850, 430) is removed from the sequence. After processing all time periods, arrange all the remaining valid fusion gaze points in chronological order to finally obtain a gaze displacement sequence set.

[0025] Table 1 Cleaned gaze displacement sequence set fragment table

[0026] As shown in Table 1, which lists five consecutive data points in the generated fixation displacement sequence set after segmentation, synchronization and noise removal.

[0027] A displacement difference calculation sub-module, based on the fixation displacement sequence set, calculates a set of difference vectors in the horizontal and vertical directions for the displacement coordinates of adjacent sampling points within the multiple sampling segments, analyzes the difference set of adjacent vectors and removes abnormal difference values, calculates a set of angles of direction distribution, and generates a set of displacement difference vector data; The generated fixation displacement sequence set is received, as shown in Table 1. For two adjacent sampling points in the sequence set, the coordinate difference in the horizontal and vertical directions is calculated to form an initial difference vector. For example, using the first two data points in Table 1, the horizontal component of the first vector is 962-961=1 pixel, and the vertical component is 544-542=2 pixels, resulting in a difference vector (1, 2). All adjacent data point pairs are processed in turn to obtain a set of difference vectors, for example: the first group=(1, 2), the second group=(-2, -1), the third group=(3, 2), and the fourth group=(-1, 1). Next, the set of difference vectors is subjected to outlier removal. The criterion for judging abnormal difference values is the magnitude change of adjacent vectors. The specific implementation process is as follows: first, calculate the magnitude of each difference vector, i.e. its Euclidean length. Taking the above vector group as an example, the magnitudes are: the first group , the second group , the third group , and the fourth group . Then, a magnitude change threshold is set, which is set to 2.5 here. This threshold is obtained by analyzing the data of 50 subjects without known eye movement abnormalities in a stationary fixation task. The statistical results show that, under a sampling interval of 100 milliseconds, 99.8% of the adjacent microsaccade displacement vector magnitude changes have an absolute value less than 2.48. Therefore, 2.5 is used as a threshold with high discrimination. The removal rule is: calculate the absolute value of the difference between the magnitudes of two adjacent vectors, and if the value is greater than 2.5, the latter vector is considered abnormal. For example, calculate the magnitude difference between the second group and the third group: This value is less than 2.5, so the third group is retained. Calculate the magnitude difference between the third group and the fourth group: This value is less than 2.5, so the fourth group is retained. Calculate the magnitude of a fifth group vector, which is 6.20, then This value is greater than 2.5, so the fifth group is removed from the vector group. After removing all abnormal difference vectors, the direction angles of the remaining valid vectors are calculated. The direction angle is calculated using the arctangent function, which takes the vertical component and the horizontal component of the vector as input and outputs the angle with the horizontal positive direction, ranging from -180 degrees to 180 degrees. For example, for the vector first group=(1, 2), the direction angle is degrees. For the second set of vectors = (-2, -1), the direction angle is degrees. For the second set of vectors = (-2, -1), the direction angle is

[0028] The direction density sub-module, according to the displacement difference vector set data set, carries out frequency statistics on the direction angle of the multi-vector, extracts the multi-direction frequency calculation normalized density value, and iteratively calculates the concentration degree of all direction interval densities, to generate the direction concentration degree; The generated displacement difference vector set data is received, which contains a series of effective displacement difference vectors and their corresponding direction angles. First, the distribution frequency of the direction angles of all vectors is counted. In order to perform the counting, the 360-degree direction space is divided into 8 direction intervals of equal width, each of which is 45 degrees wide. The 8 intervals are (0, 45), (45, 90), (90, 135), (135, 180), (-180, -135), (-135, -90), (-90, -45), and (-45, 0). Assuming that after processing, a total of 200 effective displacement difference vectors are obtained. The number of vectors falling into each direction interval is counted by traversing the direction angles of the 200 vectors. For example, the counting results are as follows: there are 40 vectors in the (0, 45) interval, 80 in the (45, 90) interval, 20 in the (90, 135) interval, 10 in the (135, 180) interval, 10 in the (-180, -135) interval, 15 in the (-135, -90) interval, 15 in the (-90, -45) interval, and 10 in the (-45, 0) interval. Next, the normalized density values of each direction interval are calculated based on these frequencies. The normalized density value is calculated by dividing the vector frequency of the interval by the total number of vectors. For example, the normalized density of the (45, 90) interval is 80 / 200 = 0.4. The density values of the other intervals are respectively: (0, 45) is 0.2, (90, 135) is 0.1, (135, 180) is 0.05, (-180, -135) is 0.05, (-135, -90) is 0.075, (-90, -45) is 0.075, and (-45, 0) is 0.05. Subsequently, the concentration of all direction interval densities is iteratively calculated to generate the final direction concentration. The specific calculation process of the direction concentration is as follows: first, find the maximum value among all normalized density values, which in this example is 0.4. Then, compare this maximum density value with the expected density value under uniform distribution. In the case of 8 intervals, the expected density under uniform distribution is 1 / 8 = 0.125. The calculation formula of the direction concentration is the maximum density value divided by the expected density value. Therefore, the direction concentration in this embodiment is 0.4 / 0.125 = 3.2. This value indicates that in the most dominant direction interval, the occurrence density of eye movement displacement is 3.2 times that of the completely random (uniform) case. The direction concentration value is 3.2.

[0029] Specifically, as shown in Figure 2 , 4 The direction analysis module includes: The density center extraction submodule extracts the direction set center coordinates within all sampling segments based on the direction concentration, calculates the product of the direction distribution frequency and the corresponding angle weight, calculates the distribution gravity coordinates based on the weight sum, and generates the direction density center coordinate set. The direction concentration and corresponding direction distribution frequency data calculated for each continuous sampling segment in the previous stage are received. Here, a 6-second fixation task is taken as an example, which is divided into three continuous 2-second sampling segments (denoted as sampling segment 1, sampling segment 2, and sampling segment 3). For each sampling segment, its direction distribution frequency data is first extracted. The data is represented as the normalized density value of each of the 8 direction intervals (0, 45), (45, 90),..., (-45, 0). Then, an angle weight is set for each direction interval, which is the center angle value of the interval. Specifically, the weight of the (0, 45) interval is 22.5 degrees, the weight of the (45, 90) interval is 67.5 degrees, the weight of the (90, 135) interval is 112.5 degrees, the weight of the (135, 180) interval is 157.5 degrees, the weight of the (-180, -135) interval is -157.5 degrees, the weight of the (-135, -90) interval is -112.5 degrees, the weight of the (-90, -45) interval is -67.5 degrees, and the weight of the (-45, 0) interval is -22.5 degrees. Subsequently, the direction distribution barycenter coordinates of each sampling segment, i.e., the direction density center angle, are calculated. The calculation process is as follows: multiply the normalized density value of each direction interval by its corresponding angle weight, and then add all the 8 products. Since the sum of the normalized density values is 1, the sum is the weighted average of the barycenter angle. Taking the data of sampling segment 1 in the previous example as an example, the normalized density values of the 8 intervals are 0.2, 0.4, 0.1, 0.05, 0.05, 0.075, 0.075, and 0.05, respectively. The calculation of the direction density center angle is as follows: 22.5) + (0.4 67.5) + (0.1 112.5) + (0.05 157.5) + (0.05 (-157.5)) + (0.075 (-112.5)) + (0.075 (-67.5)) + (0.05 (-22.5)) = 28.125 degrees. The same calculation is performed for sampling segment 2 and sampling segment 3. The normalized density distribution of sampling segment 2 becomes {0.1, 0.2, 0.4, 0.1, 0.05, 0.05, 0.05, 0.05}, and the direction density center angle is: 22.5) + (0.2 67.5) + (0.4 112.5) + (0.1 157.5) + (0.05 (-157.5)) + (0.05 (-112.5)) + (0.05 (-22.5)) = 58.5 degrees. The normalized density distribution of sample segment 3 becomes {0.05, 0.1, 0.15, 0.2, 0.2, 0.15, 0.1, 0.05}, and the directional density center angle is: (0.05 (-22.5)) = 58.5 degrees. The normalized density distribution of sample segment 3 becomes {0.05, 0.1, 0.15, 0.2, 0.2, 0.15, 0.1, 0.05}, and the directional density center angle is: (0.05 22.5) + (0.1 67.5) + (0.15 112.5) + (0.2 157.5) + (0.2 (-157.5)) + (0.15 (-112.5)) + (0.1 (-67.5)) + (0.05 (-22.5)) = 0.0 degrees. Combining the directional density center angle calculated from the three sample segments generates the directional density center coordinate set.

[0030] Table 2. Directional Density Center Coordinate Set Example

[0031] As shown in Table 2, the table lists the directional density center angle calculation results of three consecutive sample segments. The advantage of this approach is that by converting the directional distribution into a single barycentric angle value, the overall trend of eye movement direction in each time period is quantified.

[0032] The angle change calculation submodule calls the directional density center coordinate set, calculates the included angle difference value for the density center coordinate points of adjacent sample segments, calculates the angle change ratio according to the time interval between adjacent segments, and generates the density center angle change rate set; The generated directional density center coordinate set, as shown in the data in Table 2, is called. The data set contains a series of sampling segments arranged in time sequence and their corresponding directional density center angles. For the density center angles of two adjacent sampling segments in time, the included angle difference between them is calculated, and the angle change rate is further calculated. First, the angle values of adjacent sampling segments are obtained, for example, the angle of sampling segment 1 is 28.125 degrees, and the angle of sampling segment 2 is 58.500 degrees. When calculating the included angle difference between the two, the periodicity of the angle needs to be handled. The specific action is: calculate the direct difference between the two angles, if the absolute value of the difference is greater than 180 degrees, subtract the absolute value from 360 degrees, and assign the opposite sign to the original difference. For sampling segment 1 and sampling segment 2, the direct difference is 58.500-28.125=30.375 degrees. The absolute value is less than 180 degrees, so the included angle difference is 30.375 degrees. For sampling segment 2 (58.500 degrees) and sampling segment 3 (0.000 degrees), the direct difference is 0.000-58.500=-58.500 degrees, and the absolute value is less than 180 degrees, so the included angle difference is -58.500 degrees. Next, the angle change rate is calculated according to the time interval between adjacent segments. In this embodiment, the duration of each sampling segment is 2 seconds, so the time interval between the center points of adjacent sampling segments is also 2 seconds. The calculation method of the angle change rate is to divide the included angle difference by the time interval. For the change from sampling segment 1 to sampling segment 2, the angle change rate is 30.375 degrees / 2 seconds=15.1875 degrees / second. For the change from sampling segment 2 to sampling segment 3, the angle change rate is -58.500 degrees / 2 seconds=-29.2500 degrees / second. Repeat this process for all adjacent sampling segments to combine all calculated angle change rate values in time sequence to generate a density center angle change rate set. In this example, the generated density center angle change rate set is {15.1875, -29.2500}. The result shows that the counterclockwise rotation rate of the direction center in the first time interval and the clockwise rotation rate of the direction center in the second time interval.

[0033] The stability determination sub-module compares the multi-sampling segment angle change rate values with the set stability threshold according to the density center angle change rate set, identifies the sampling segments that exceed the stability threshold, and generates a stability decay segment. The set of density center angle change rates, in this example, is {15.1875, -29.2500} (unit: degree / s), each value in the set is compared with a preset stability threshold to identify the sampling section in which the gaze direction pattern changes significantly. The stability threshold is set as follows: first, 60 healthy adults without known visual or nervous system abnormalities are recruited to participate in a verification experiment. The experiment includes two tasks: task A requires the subject to continuously gaze at a single fixed point in the center of the screen for 5 minutes; task B requires the subject to covertly switch their attention between four invisible locations around the screen while gazing at the center point according to sound cues. The eye movement data of all subjects under the two tasks is collected, and the set of density center angle change rates for each person is calculated according to the previous steps. Statistical data shows that in task A (stable gaze), the absolute value of the angle change rate of 95% is less than 20.3 degrees / s. While in task B (attention switching), more than 85% of the absolute value of the angle change rate is higher than 25 degrees / s. Based on this, 20.5 degrees / s is selected as the stability threshold, which can effectively distinguish between stable gaze direction patterns and changing patterns. The specific implementation of the judgment is: traverse each value in the set of density center angle change rates, take its absolute value, and compare it with the stability threshold 20.5 degrees / s. If the absolute value of a change rate is greater than 20.5 degrees / s, the next sampling section corresponding to the change rate is identified as a "stability attenuation section". Taking the data in the previous example, the first angle change rate is 15.1875 degrees / s, its absolute value is 15.1875, which is less than the threshold 20.5, so sampling section 2 is not identified. The second angle change rate is -29.2500 degrees / s, its absolute value is 29.2500, which is greater than the threshold 20.5. Therefore, the next sampling section corresponding to this change rate, i.e. sampling section 3, is identified as a stability attenuation section.

[0034] Specifically, as shown in Figure 2 , 5 The drift correction module includes: The data extraction submodule obtains the adjacent time window gaze center coordinates and the drift mean value of the previous stable section based on the stability attenuation section, and performs difference calculation in the spatial coordinate system, counts the drift change sequence between time windows and performs linear smoothing, and generates a set of time window drift difference values. Based on the identified "sample segment 3" as a stability attenuation segment. First, this 2-second long stability attenuation segment is divided into 4 consecutive time windows of 500 milliseconds in length (denoted as time window 1 to time window 4). Then, all gaze point coordinates in these 4 time windows are extracted from the initially generated gaze displacement sequence set, and the average gaze center coordinates of each time window are calculated. Through calculation, the center coordinates of time window 1 are (965, 548), the center coordinates of time window 2 are (970, 552), the center coordinates of time window 3 are (968, 555), and the center coordinates of time window 4 are (972, 558). Next, the data of the stable segment immediately before the stability attenuation segment, i.e. "sample segment 2", is obtained, and the "pre-stability segment drift average" is calculated. The drift average here is defined as the difference vector of the gaze center coordinates of the first 100 millisecond time slice and the last 100 millisecond time slice in the stable segment. The starting gaze center of sample segment 2 is (961, 544), and the ending gaze center is (963, 546), so the pre-stability segment drift average vector is (963-961, 546-544), i.e. (2, 2). The next step is to calculate the difference in space coordinates to obtain the drift change sequence between time windows. The calculation is to obtain the difference vector between the gaze center coordinates of adjacent time windows. The specific calculation is as follows: the drift vector from time window 1 to time window 2 is (970-965, 552-548) = (5, 4); the drift vector from time window 2 to time window 3 is (968-970, 555-552) = (-2, 3); the drift vector from time window 3 to time window 4 is (972-968, 558-555) = (4, 3). Thus, the initial drift change sequence {(5, 4), (-2, 3), (4, 3)} is obtained. Finally, linear smoothing processing is performed on this sequence. The smoothing method uses a sliding average with a window size of 2 applied to the vectors in the sequence. The first vector (5, 4) remains unchanged as there is no previous sequence vector. The second smoothed vector is obtained by averaging the first and second original vectors: (5+(-2)) / 2, (4+3) / 2 = (1.5, 3.5). The third smoothed vector is obtained by averaging the second and third original vectors: (-2+4) / 2, (3+3) / 2 = (1, 3). Combining these smoothed vectors generates the time window drift difference set: (5, 4), (1.5, 3.5), (1, 3).

[0035] The equilibrium point sub-module calls the time window drift difference set, extracts the proportion coefficient between the drift amplitude of adjacent time windows and the pre-stability segment drift average, calculates the weighted drift correction of multiple time windows, and performs sliding average analysis to smooth the center trend of the interval, generating a drift equilibrium point sequence; The generated time window drift difference set, i.e. the vector sequence (5, 4), (1.5, 3.5), (1, 3). First, the amplitude of each time window drift vector is extracted, which is calculated as the Euclidean length of the vector. The amplitude of the first vector is ; the amplitude of the second vector is ; and the amplitude of the third vector is . At the same time, the "previous stable segment drift mean" vector (2, 2) calculated by the previous module is obtained, and the amplitude is . Next, the proportionality coefficient between the amplitude of each time window drift and the amplitude of the previous stable segment drift mean is calculated. The calculation process is to divide the drift amplitude of each time window by the drift amplitude of the previous stable segment. The first proportionality coefficient is ; the second is ; and the third is . Subsequently, the multi-time window weighted drift correction is performed, and the specific operation is to multiply each original vector in the time window drift difference set by its corresponding proportionality coefficient. The corrected vectors are: the first vector (5, 4) x 2.26 = (11.3, 9.04); the second vector (1.5, 3.5) x 1.35 = (2.025, 4.725); and the third vector (1, 3) x 1.12 = (1.12, 3.36). After that, a sliding average analysis with a window size of 2 is performed on the sequence of corrected vectors to smooth the data. The first corrected vector (11.3, 9.04) remains unchanged. The second smoothed corrected vector is ((11.3 + 2.025) / 2, (9.04 + 4.725) / 2) = (6.66, 6.88). The third smoothed corrected vector is ((2.025 + 1.12) / 2, (4.725 + 3.36) / 2) = (1.57, 4.04). Finally, based on these smoothed corrected drift vectors, the drift balance point position sequence is generated from the starting point of the stability decay segment, i.e. the center coordinates of time window 1 (965, 548), by accumulating the vectors.

[0036] Table 3 Drift balance point position sequence generation table

[0037] As shown in Table 3, by accumulating the corrected drift vectors, a series of coordinate points representing the core trend of gaze drift are obtained, which constitute the drift balance point position sequence.

[0038] The probability correction sub-module extracts the appearance frequency and direction consistency of point position distribution in consecutive time windows, calculates the appearance probability of the direction difference in multiple time windows, analyzes the probability density, and performs weighted fusion on the probability density and direction consistency to generate a drift balance point correction amount. According to the generated drift equilibrium point sequence, i.e., (965.00, 548.00), (976.30, 557.04), (982.96, 563.92), (984.53, 567.96). First, the offset vectors between the point positions in the continuous time window are extracted to analyze the distribution characteristics and the consistency of the direction. The offset vectors are calculated as follows: vector 1 (point position 0 to 1) is (11.30, 9.04); vector 2 (point position 1 to 2) is (6.66, 6.88); and vector 3 (point position 2 to 3) is (1.57, 4.04). All the vectors point to the first quadrant, showing consistency in the direction. To quantify this consistency, the angles of the vectors are calculated: the angle of vector 1 is degrees; the angle of vector 2 is degrees; and the angle of vector 3 is degrees. The standard deviation of these angles is degrees. The offset direction consistency score is defined as , i.e., . Next, the occurrence probability of the offset direction difference in multiple time windows is calculated. The offset direction difference is the angle difference between adjacent vectors: angle difference 1 (vector 2-vector 1) is degrees; and angle difference 2 (vector 3-vector 2) is degrees. The angle difference value space is divided into 4 intervals of 90 degrees, and both difference values fall into the [0, 90) interval. Therefore, the occurrence probability of this interval is 2 / 2=1.0, which is the probability density of this interval. Finally, the probability density and the offset consistency are weighted and fused to generate the drift equilibrium point correction. Here, the consistency weight is set to 0.7, and the probability density weight is set to 0.3. The setting of these weight values is based on a regression analysis of eye movement data of 40 subjects in a fatigue driving simulation. This weight combination has the smallest average angle error with the actual measured value when predicting the gaze point drift direction in the next 1 second. The direction of the fused correction is taken as the average angle of the offset vectors, i.e., 51.1 degrees. The amplitude is obtained by multiplying the average offset vector amplitude by the fusion weight factor. The average offset amplitude is . The fusion weight factor is (consistency score x consistency weight)+(probability density x probability density weight)=(0.928 x 0.7)+(1.0 x 0.3)=0.6496+0.3=0.9496. The correction amplitude is . The drift equilibrium point correction is a vector with an angle of 51.1 degrees and an amplitude of 8.27, and the coordinate components of the vector are .

[0039] Specifically, as shown in Figure 2 , 6 , the screening calculation module includes: The convergence difference calculation sub-module calculates the change gradient values of the drift balance point correction amount and the density center transfer rate in the same time window based on the drift balance point correction amount and the density center transfer rate, analyzes the difference value sequence in multiple time windows, and performs weighted averaging on the difference value sequence to generate the convergence difference; The drift balance point correction amount calculated for each time window in the stability decay segment and the corresponding density center transfer rate are received. In this embodiment, the stability decay segment (sample segment 3) is divided into four 500 ms time windows, so there are three time window transitions. Through calculation, the drift balance point correction amount vector sequence of the three transitions is {(5.22, 6.43), (5.50, 6.80), (5.30, 6.50)}. At the same time, the density center transfer rate during the same time window transition needs to be obtained. The transfer rate is defined as the absolute value of the density center angle change rate, and the data is obtained by calling the result of step 5 and recalculating according to the 500 ms time window, and the sequence is {18.2, 25.5, 30.1} (unit: degree / s). First, the change gradient values of the two groups of data in the same time window are calculated. For the drift balance point correction amount, the change gradient is obtained by calculating the difference between the magnitudes of adjacent vectors. The magnitude of each vector is: ; ; . The change gradient sequence is ( ), ( ) = {0.46, -0.35}. For the density center transfer rate, the change gradient is the difference between adjacent values: { ), ( )} = {7.3, 4.6}. Next, the difference value sequence in multiple time windows is analyzed. The sequence is obtained by subtracting the two gradient values during the same transition. The first difference value is . The second difference value is . The difference value sequence is {-6.84, -4.95}. Finally, the difference value sequence is weighted averaged to generate a single convergence difference. The setting of the weight coefficient is based on an eye tracking experiment involving 80 subjects, and the results show that when analyzing the development trend of unstable fixation, the data of the later time window has a greater impact on the prediction result. Based on this, increasing weights are set for consecutive difference values, and in this embodiment, the two difference values are set weights of 0.4 and 0.6, respectively. The calculation of the convergence difference is: )+( )= . The advantage of this method is that by combining the dynamic change gradients of the macro position change and the micro direction distribution change of the drift in two dimensions, the synchronization of the two types of eye movement characteristics in the time evolution is quantified. The generated convergence difference is -5.706.

[0040] The threshold deviation detection submodule calls the convergence difference, obtains the difference value sequence in the continuous multiple time windows, and performs interval judgment on the difference values in the multiple time windows according to the set convergence difference threshold value, and generates a sustained deviation judgment quantity by detecting the number of time windows continuously deviating from the convergence difference threshold value and the deviation amplitude; The calculated convergence difference is called. In order to perform time series analysis, the current obtained is the difference value sequence in the continuous multiple time windows without weighted average, that is, {-6.84, -4.95}. In order to show the detection of continuous deviation, the sequence is expanded to {-6.84, -5.50, -7.10} here, representing the convergence difference in the transition period of the continuous three time windows. Next, each difference value in the sequence is compared with the set convergence difference threshold value. The convergence difference threshold value is set to 5.0. The determination of this threshold value is derived from a comparative study including 100 subjects (50 patients clinically diagnosed with early stage intermittent exotropia and 50 healthy controls with normal visual function). The study collected eye movement data of all subjects when performing long time fixation tasks, and calculated the convergence difference sequence. The data shows that 98% of the absolute values of the convergence difference of the healthy control group are less than 4.8, and 95% of the subjects in the early exotropia patient group have continuous multiple absolute values of the convergence difference exceeding 5.5. Therefore, 5.0 is selected as the threshold value for distinguishing between health and potential abnormalities. The specific interval judgment action is: calculate the absolute value of each element in the sequence, and compare it with the threshold value 5.0. The first value: , which is greater than 5.0, is judged as deviation. The second value: , which is greater than 5.0, is judged as deviation. The third value: , which is greater than 5.0, is judged as deviation. Subsequently, a sustained deviation judgment quantity is generated by detecting the number of time windows continuously deviating from the convergence difference threshold value and the deviation amplitude. In this example, the number of time windows continuously deviating is 3. The deviation amplitude is defined as the average of the absolute values of all deviation values and the part exceeding the threshold value. The exceeding parts are , , . The average exceeding amplitude is . The calculation method of the sustained deviation judgment quantity is to multiply the number of time windows continuously deviating by the average exceeding amplitude. Therefore, the sustained deviation judgment quantity is . The result shows that the eye movement control of the subject exhibits sustained and significant macro and micro movement feature dissociation within 1.5 seconds (transition of 3 500 ms time windows).

[0041] The segment reconstruction submodule selects the drift balance point correction quantity and density center transfer rate data of the corresponding segment according to the sustained deviation judgment quantity, performs multidimensional interpolation and curvature adjustment calculation on the numerical sequence in the deviated segment, and generates an early exotropia screening result; According to the generated sustained deviation determination quantity (whose value is 4.44, a number greater than 0), subsequent operations are performed. First, the section corresponding to the determination quantity is selected, i.e., all original data in the stability attenuation section (sample section 3), mainly including the drift balance point correction quantity sequence (5.22, 6.43), (5.50, 6.80), (5.30, 6.50) and the density center shift rate sequence {18.2, 25.5, 30.1} in the section. Next, multi-dimensional interpolation is performed on the value sequence in the deviation section. The purpose of interpolation here is to increase the density of data points to construct a smoother motion trajectory. The specific operation is: between every two consecutive vectors of the drift balance point correction quantity sequence, a new vector is inserted based on the time center point. For example, between the vectors (5.22, 6.43) and (5.50, 6.80), a new vector is inserted, whose coordinates are the average of the corresponding components of the two vectors, i.e., ( ) / 2, ( ) / 2)=(5.36, 6.62). This operation is performed on the entire sequence to double the number of data points. Subsequently, curvature adjustment calculation is performed. The purpose of curvature adjustment is to modify the shape of the reconstructed motion trajectory according to the severity of the deviation. Here, a curvature adjustment factor is defined, whose value is positively correlated with the sustained deviation determination quantity. The calculation formula is , and the numerical value is . The adjustment action is: decompose each vector in the interpolated drift balance point correction quantity sequence into its radial component and tangential component, and then multiply only the amplitude of its tangential component by the curvature adjustment factor 1.444. After completing the curvature adjustment, all adjusted vectors are sequentially accumulated to obtain a reconstructed drift displacement vector. After the above interpolation and adjustment calculation, the sum of all vectors is (20.5, 25.8). Finally, the early exotropia screening result is generated based on the reconstructed drift displacement vector. The setting of the judgment benchmark comes from a retrospective data analysis of 300 ophthalmic cases (including 150 cases of diagnosed early exotropia patients), and the results show that when the amplitude of the reconstructed drift displacement vector exceeds 28 pixels, its screening sensitivity for early exotropia is 93% and the specificity is 91%. In this embodiment, the amplitude of the reconstructed drift displacement vector is pixels. The value 32.95 is greater than the set threshold of 28 pixels. Therefore, the generated early exotropia screening result is: positive.

[0042] Specifically, as shown in Figure 2 , 7 , the result optimization module includes: The smoothing and reweighting submodule, based on the early exotropia screening results, the drift balance point correction amount and the density center transfer rate, smooths the drift balance point correction amount, calculates the weighting coefficient according to the change range of the density center transfer rate and multiplies it point by point with the smoothing result to generate a stability weighted sequence. The system receives the output of a "positive" early exotropia screening result and retrieves the associated drift equilibrium point correction sequence (5.22, 6.43), (5.50, 6.80), (5.30, 6.50) and density center shift rate sequence {18.2, 25.5, 30.1} (unit: degrees / second). First, the drift equilibrium point correction sequence is smoothed using a moving average with a window size of 2. The first vector (5.22, 6.43) remains unchanged due to the absence of a preceding vector. The second smoothed vector is obtained by averaging the corresponding components of the first and second original vectors, calculated as (…). ) / 2, ( ) / 2, the result is (5.36, 6.615). The third smoothed vector is obtained by averaging the second and third original vectors, calculated as ( ) / 2, ( The result is (5.40, 6.65). The smoothed sequence is (5.22, 6.43), (5.36, 6.615), (5.40, 6.65). Then, weighting coefficients are calculated based on the variation amplitude of the density center transfer rate sequence. The variation amplitude is the absolute value of the difference between adjacent transfer rate values. The first variation amplitude is... The second change range is The formula for calculating the weighting coefficients is as follows: The baseline rate of change was set based on an eye-tracking experiment involving 50 healthy subjects. The experimental data showed that during a stable fixation task, the average change in the density center shift rate over a continuous 500-millisecond time window was approximately 5.0 degrees / second; therefore, the baseline rate of change was set to 5.0. The first weighting coefficient was applied to the second point of the smoothed sequence and calculated as follows: The second weighting factor is applied to the third point, and the calculation is as follows: The first point of the smoothed sequence has no preceding changes and its coefficient is 1.0. The resulting weighted coefficient sequence is {1.0, 2.46, 1.92}. Finally, the smoothed vector sequence is multiplied point-by-point with the weighted coefficient sequence. The first vector is... The second vector is The third vector is These calculation results are combined to generate a stability-weighted sequence: (5.22, 6.43), (13.19, 16.27), (10.37, 12.77).

[0043] the feature clustering sub-module, based on the weighted difference values in multiple time windows in the stability weighted sequence, calls the time distribution feature of the drift balance point correction amount, multi-dimensionally aggregates the weighted difference values, judges the similarity interval according to the Euclidean distance between the aggregated vectors, and obtains the feature clustering partition coefficient; The generated stability weighted sequence (5.22, 6.43), (13.19, 16.27), (10.37, 12.77) is received. First, the weighted difference values generated by multiple time windows in the sequence are calculated, specifically the difference values between consecutive vectors. The first difference vector is The second difference vector is The weighted difference value sequence (7.97, 9.84), (-2.82, -3.50) is obtained. Next, the time distribution feature of the drift balance point correction amount is called to multi-dimensionally aggregate the weighted difference values. The "time distribution feature" here refers to the time point at which each difference value occurs. Assuming that the time points corresponding to the difference values are 1.0 seconds and 1.5 seconds respectively, a three-dimensional aggregated vector containing the time dimension is constructed. The first aggregated vector is (7.97, 9.84, 1.0); the second aggregated vector is (-2.82, -3.50, 1.5). Then, the similarity interval is judged according to the Euclidean distance between the aggregated vectors. The Euclidean distance between the two aggregated vectors is calculated as The judgment reference value of the similarity interval is set to 12.0. The setting of this reference value is derived from an analysis of eye movement data of 100 subjects, which shows that in consecutive single-type eye movement events, the distance between aggregated vectors is less than 12.0 in 95% of cases, while the distance increases significantly when the eye movement mode switches. In this example, the calculated distance 17.16 is greater than the reference value 12.0, so it is judged that the two aggregated vectors do not belong to the same similarity interval, i.e. they do not belong to the same cluster. Based on this judgment, the feature clustering partition coefficient is obtained, which is an array used to identify the cluster to which each weighted difference value belongs. Since the two difference values are judged to belong to different clusters, the first difference value is assigned a cluster label 1 and the second difference value is assigned a cluster label 2, and the generated feature clustering partition coefficient is [1, 2].

[0044] The anomaly correction sub-module, according to the feature clustering partition coefficient, calls the time variation trend of the density center transfer rate, offsets and compares the weighted difference mean values in the multi-cluster section, judges the time period that exceeds the interval threshold as an abnormal fluctuation period and performs weighted balance adjustment, and generates an optimized exotropia screening result; The interval threshold is set by statistically setting the distribution of the eye movement indicators of the tested user in multiple stages of measurement; According to the generated feature clustering partition coefficient, an operation is performed. The coefficient indicates that there are two different clustering sections. At the same time, a time-varying trend of the density center shift rate, that is, the sequence {18.2, 25.5, 30.1}, is called. First, a shift comparison is performed on the weighted difference mean in the multi-clustering section. Cluster 1 only contains one difference vector (7.97, 9.84), and the mean thereof is itself, and the amplitude thereof is . Cluster 2 only contains one difference vector (-2.82, -3.50), and the mean thereof is also itself, and the amplitude thereof is . Next, a time period exceeding an interval threshold is determined. The interval threshold is set according to a baseline study on a visual healthy population, which shows that 99% of the weighted difference mean amplitudes are less than 8.0 under stable fixation conditions. Therefore, the interval threshold is set to 8.0. The mean amplitude of each clustering section is compared with the threshold. The mean amplitude 12.66 of cluster 1 is greater than 8.0, and therefore the time period corresponding thereto (that is, the transition from the first 500 ms time window to the second) is determined to be an abnormal fluctuation period. The mean amplitude 4.49 of cluster 2 is less than 8.0, and is not determined to be abnormal. Subsequently, a weighted balance adjustment is performed on the abnormal fluctuation period. The adjustment aims to correct the initial screening result and reduce the overreaction caused by a temporary abnormal fluctuation. The adjustment amount is calculated by multiplying the reconstructed drift displacement vector amplitude (32.95 pixels) of the initial screening by an adjustment factor. The adjustment factor is calculated as . The adjusted displacement vector amplitude is pixels. Finally, the adjusted amplitude is compared with the judgment reference value 28 pixels of the early exotropia screening. Since 20.82 pixels is less than 28 pixels, the initial “positive” conclusion is corrected. The advantage of this manner is that the inconsistent fluctuation in the data stream is identified and quantified through clustering analysis, and the initial judgment is internally verified and corrected. Finally, an optimized exotropia screening result: negative, is generated.

[0045] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An ophthalmic intermittent exotropia early screening device, characterized in that, The device comprises: a gaze analysis module, which collects a binocular gaze displacement sequence of a subject during a gaze target point process, divides sampling segments according to equal time lengths, calculates a displacement difference vector group of adjacent sampling points in the multiple sampling segments, and performs direction distribution density statistics to generate a direction concentration degree and deliver it to a direction analysis module; the direction analysis module, based on the direction concentration degree, extracts a direction set center coordinate of a continuous sampling segment, calculates an angle change ratio between density center points of adjacent sampling segments, and analyzes a density center transfer rate, if the density center transfer rate exceeds a stability threshold, marks a stability attenuation segment and delivers it to a drift correction module; the drift correction module, based on the stability attenuation segment, obtains adjacent time window gaze center coordinates and a drift mean value of a previous stable segment, and weightedly fuses to calculate a drift balance point position, performs probability correction on the drift balance point position of a continuous time window, generates a drift balance point correction amount and delivers it to a screening calculation module; the screening calculation module, based on the drift balance point correction amount and the density center transfer rate, calculates a convergence difference, if the convergence difference continuously deviates from a convergence difference threshold in multiple time windows, the segment is nonlinearly reconstructed, and an early exotropia screening result is generated.

2. An ophthalmic intermittent exotropia early screening device according to claim 1, characterized in that: The direction concentration degree includes gaze displacement direction distribution density, displacement difference vector group and direction offset amplitude, the density center transfer rate includes direction set center coordinate change rate, adjacent segment angle change ratio and stability threshold comparison result, the drift balance point correction amount includes drift mean value weighting, drift probability correction parameter and drift balance point position deviation, and the early exotropia screening result includes convergence difference distribution, nonlinear reconstruction coefficient and abnormal segment label.

3. An ophthalmic intermittent exotropia early screening device according to claim 1, wherein, The gaze analysis module comprises: a gaze data submodule, which obtains a binocular gaze displacement sequence of a subject during a gaze target point process, segments and samples continuous gaze positions according to equal time intervals, and performs time stamp synchronization and noise removal to generate a gaze displacement sequence set; a displacement difference calculation submodule, based on the gaze displacement sequence set, for the displacement coordinates of adjacent sampling points in multiple sampling segments, calculates a difference vector group in the horizontal and vertical directions, analyzes the difference set of adjacent vectors and removes abnormal difference values, calculates the angle set of direction distribution, and generates a displacement difference vector group dataset; a direction density submodule, according to the displacement difference vector group dataset, performs distribution frequency statistics on multi-vector direction angles, extracts multi-direction frequency calculation normalized density values, and iteratively calculates the concentration degree of all direction interval densities to generate a direction concentration degree.

4. The device as claimed in claim 1, wherein, The direction analysis module comprises: a density center extraction submodule, based on the direction concentration degree, extracts the direction set center coordinates in all sampling segments, calculates the product of the direction distribution frequency and the corresponding angle weight value, calculates the distribution barycenter coordinates according to the weight sum, and generates a direction density center coordinate set; an angle change calculation submodule, which calls the direction density center coordinate set, calculates the included angle difference for the density center coordinate points of adjacent sampling segments, calculates the angle change ratio according to the time interval between adjacent segments, and generates a density center angle change rate set; The stability determining sub-module compares the multi-sampling segment angle change rate values with the set stability threshold according to the density center angle change rate set, identifies the sampling segments exceeding the stability threshold, and generates stability decay segments.

5. An ophthalmic intermittent exotropia early screening device according to claim 4, wherein, The stability threshold is set by obtaining the direction concentration sequence of the tested individual in the consecutive sampling segments, calculating the absolute amplitude of the difference value of the direction concentration between adjacent segments, and calculating the average fluctuation interval of the difference value set of all samples.

6. The device for early screening of intermittent exotropia in ophthalmology as claimed in claim-1 wherein, The drift correction module includes: The data extraction sub-module obtains the adjacent time window fixation center coordinates and the drift mean value of the previous stable segment, and calculates the difference in the spatial coordinate system based on the stability decay segments, counts the drift change sequence between time windows, and performs linear smoothing to generate a time window drift difference set; The balance point sub-module extracts the proportion coefficient between the drift amplitude of adjacent time windows and the drift mean value of the previous stable segment, calculates the weighted drift correction of multiple time windows, and performs sliding average analysis on the center trend of the smoothing interval to generate a drift balance point sequence; The probability correction sub-module extracts the appearance frequency and offset direction consistency of the point position distribution in the consecutive time windows according to the drift balance point sequence, calculates the appearance probability of the offset direction difference in the multiple time windows, analyzes the probability density, and performs weighted fusion on the probability density and the offset consistency to generate a drift balance point correction amount.

7. The device as claimed in claim 1, wherein, The screening calculation module includes: The convergence difference calculation sub-module calculates the change gradient values of the drift balance point correction amount and the density center shift rate in the same time window, analyzes the difference value sequence in the multiple time windows, and performs weighted average on the difference value sequence to generate a convergence difference. The threshold deviation detection sub-module obtains the difference value change sequence in the consecutive multiple time windows by calling the convergence difference, judges the interval of the difference value in the multiple time windows according to the set convergence difference threshold, detects the number and amplitude of the time windows deviating from the convergence difference threshold, and generates a continuous deviation determination amount. The segment reconstruction sub-module selects the drift balance point correction amount and the density center shift rate data of the corresponding segment according to the continuous deviation determination amount, performs multi-dimensional interpolation and curvature adjustment calculation on the numerical sequence in the deviated segment to generate an early exotropia screening result.

8. An ophthalmic intermittent exotropia early screening device according to claim 7, wherein, The convergence difference threshold is set by calculating the mean value and standard deviation of the difference sequence of the drift balance point correction amount and the density center shift rate of the tested individual.

9. The device as claimed in claim 1, wherein, The device further includes: The result optimization module performs smoothing, re-weighting, and feature clustering based on the early exotropia screening result, the drift balance point correction amount, and the density center shift rate, extracts abnormal fluctuation segments, and performs correction to generate an optimized early exotropia screening result. The optimized early exotropia screening result includes a smoothed drift balance point, a drift balance point correction amount, and an abnormal fluctuation correction.

10. The device for early screening of intermittent exotropia in ophthalmology according to claim 9, characterized in that, The result optimization module includes: The smooth reweighting submodule is configured to smooth the drift balance point correction amount based on the early exotropia screening result, the drift balance point correction amount, and the density center transfer rate, calculate a weighting coefficient according to a change range of the density center transfer rate, and multiply the weighting coefficient with the smooth result point by point to generate a stability weighting sequence; The feature clustering submodule is configured to call a time distribution feature of the drift balance point correction amount, aggregate the weighted difference values in the stability weighting sequence in multiple time windows, determine a similarity interval according to an Euclidean distance between aggregated vectors, and obtain a feature clustering partition coefficient. The anomaly correction submodule is configured to call a time variation trend of the density center transfer rate according to the feature clustering partition coefficient, offset and compare weighted difference mean values in multiple clustering sections, determine a time period that exceeds an interval threshold as an abnormal fluctuation period and perform weighted balance adjustment, and generate an optimized exotropia screening result. The interval threshold is set by statistically setting a distribution rule of an eye movement index of a user under test in multiple stages of measurement.

Images