A method, system, device and readable storage medium for early warning of postoperative recurrence of strabismus

Abstract

The application provides a strabismus postoperative recurrence early warning method, system, device and readable storage medium, relates to the medical information technology field, and includes acquiring multi-source time sequence monitoring data, the multi-source time sequence monitoring data at least includes eye position sequence data and eye use behavior data; the difference of the eye position sequence data relative to postoperative baseline eye position data and eye position drift trend information are extracted; the recurrence risk score is calculated based on the eye position sequence data, the eye position drift trend information, and the corresponding warning level is determined according to the recurrence risk score; the collection cycle length of the next multi-source time sequence monitoring data is dynamically adjusted based on the warning level.The application realizes early discovery and accurate early warning of the hidden recurrence precursor by analyzing the gaze fluctuation characteristics of the eye position sequence data, the drift trend relative to the postoperative baseline and the eye use behavior through a multi-source time sequence monitoring index system for long-term home monitoring of postoperative strabismus patients.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and more specifically, to a method, system, device, and readable storage medium for early warning of recurrence after strabismus surgery. Background Technology

[0002] Strabismus surgery is not a radical cure, and a high recurrence rate is common in clinical practice, especially in children with intermittent exotropia, where the recurrence rate can reach 20% to 40% within 2 years after surgery. Timely identification and intervention of early signs of recurrence are key to avoiding a second surgery.

[0003] Current postoperative follow-up mainly relies on regular in-hospital check-ups, where doctors assess eye position and binocular vision using specialized equipment such as occlusion tests and synoptophores. This approach is highly dependent on patients actively coming to the hospital, making it impossible to achieve high-frequency, continuous monitoring. Furthermore, a single in-hospital measurement is insufficient to reflect the patient's true and dynamic eye position changes in daily life, easily missing early compensatory drift or intermittent strabismus.

[0004] There are some studies on remote eye position monitoring, but most of them only focus on the instantaneous deviation of eye position or single-dimensional drift, lacking a comprehensive analysis of the stability of eye position fluctuations, perceived eye position deviation, and multi-dimensional drift trends. The accuracy and lead time of early warning are still insufficient. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, device, and readable storage medium for early warning of recurrence after strabismus surgery, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: In a first aspect, this application provides a method for early warning of recurrence after strabismus surgery, comprising: acquiring multi-source temporal monitoring data of a strabismus patient after surgery, wherein the multi-source temporal monitoring data includes at least eye position sequence data and eye behavior data, wherein the eye position sequence data includes horizontal eye position sequence, vertical eye position sequence, and perceived eye position shift sequence collected during a first acquisition cycle when the strabismus patient is induced to perform a fixation task; extracting the difference between the eye position sequence data and the postoperative baseline eye position data, and the eye position drift trend information relative to the postoperative baseline eye position data within the acquisition cycle corresponding to the current fixation task, wherein the eye position drift trend information is obtained at least based on the temporal shift of one of the horizontal eye position sequence, vertical eye position sequence, and perceived eye position shift sequence relative to the baseline; calculating a recurrence risk score based on the eye position sequence data, the eye position drift trend information, and the eye behavior data, and determining a corresponding early warning level based on the recurrence risk score; dynamically adjusting the duration of the next acquisition cycle of the multi-source temporal monitoring data based on the early warning level, so as to switch the first acquisition cycle to a second acquisition cycle corresponding to the early warning level, wherein the acquisition cycles corresponding to different early warning levels are different.

[0006] Secondly, this application also provides a strabismus recurrence early warning system, comprising: The data acquisition module is used to acquire multi-source temporal monitoring data of patients after strabismus surgery. The multi-source temporal monitoring data includes at least eye position sequence data and eye behavior data. The eye position sequence data consists of horizontal eye position sequence, vertical eye position sequence, and perceived eye position deviation sequence collected during the induction of the strabismus patient to perform a fixation task within the first acquisition cycle. The feature extraction module is used to extract the difference between the eye position sequence data and the postoperative baseline eye position data, as well as the eye position drift trend information of the current task cycle relative to the postoperative baseline eye position data. The eye position drift trend information is obtained based at least on the temporal deviation of one of the horizontal eye position sequence, vertical eye position sequence, and perceived eye position deviation sequence relative to the baseline. The risk assessment module is used to calculate a recurrence risk score based on the eye position sequence data, the eye position drift trend information, and the eye behavior data, and determine the corresponding warning level based on the recurrence risk score. The cycle control module is used to dynamically adjust the duration of the next acquisition cycle of the multi-source temporal monitoring data based on the warning level, so as to switch the first acquisition cycle to the acquisition cycle corresponding to the warning level, wherein the acquisition cycle corresponding to different warning levels is different.

[0007] Thirdly, this application also provides a strabismus recurrence early warning device, comprising: Memory, used to store computer programs; A processor is used to implement the steps of the strabismus recurrence early warning method when executing the computer program.

[0008] Fourthly, this application also provides a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for early warning of recurrence after strabismus surgery.

[0009] The beneficial effects of this invention are as follows: This invention utilizes a multi-source time-series monitoring index system to conduct long-term home monitoring of strabismus patients after surgery. By analyzing the fixation fluctuation characteristics, drift trend relative to the postoperative baseline, and eye behavior of eye position sequence data, a multi-factor weighted fusion model is used to quantitatively assess the risk of recurrence. A four-level early warning system is established, and the duration of the next data collection cycle is dynamically adjusted according to the early warning level. This reduces disturbance to patients during the low-risk period and automatically intensifies monitoring during the high-risk evolution period, thereby achieving early detection and accurate early warning of hidden signs of recurrence.

[0010] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the strabismus recurrence early warning method described in this embodiment of the invention; Figure 2 This is a schematic diagram of the strabismus recurrence early warning system described in this embodiment of the invention; Figure 3 This is a schematic diagram of the strabismus recurrence early warning device described in an embodiment of the present invention.

[0013] The markings in the diagram are: 800, strabismus recurrence early warning device; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0015] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0016] Example 1: This embodiment provides a method for early warning of recurrence after strabismus surgery. This method can be guided by an interactive application running on a mobile terminal (such as a smartphone or tablet) that interacts with the strabismus patient after surgery, in the form of a gamified "gaze task." Some or all of the calculations can also be performed by a cloud server connected to the mobile terminal. For ease of description, this embodiment uses a mobile terminal as the execution subject, but the invention is not limited thereto.

[0017] To facilitate understanding, a general description of the data acquisition system upon which this invention relies will be provided first. This invention employs a hybrid acquisition mode that combines hierarchical, frequency-based, and long- and short-cycle approaches to construct a multi-source time-series monitoring database covering eye position, visual function, eye use behavior, and baseline reference dimensions.

[0018] First layer: Seamless continuous acquisition layer.

[0019] Eye-use behavior data is continuously processed in the background 24 / 7 by the mobile terminal's built-in sensors (ambient light sensor, distance sensor, screen usage status monitoring interface, etc.), requiring no manual intervention. The basic sampling granularity is at the second level, and short-term behavior segments are aggregated every 15 minutes. At 24:00 every calendar day, the full set of behavior statistics is automatically summarized, and the generated items include, but are not limited to: average daily near-field eye use duration, number of continuous eye fatigue segments (a single continuous eye use exceeding 40 minutes is counted as one segment), frequency of eye use in low light (ambient light intensity below 50 lux is considered a low-light environment), and nighttime eye use duration.

[0020] The second layer: the home-based proactive detection layer.

[0021] Eye position and visual function data are collected using a mobile interactive application. Under normal monitoring conditions, a first collection cycle is set, for example, two eye position tests and one simple visual function test per week. After an alert is triggered, the collection frequency will be automatically adjusted and encrypted.

[0022] The third layer: the home-based proactive and precise retesting layer.

[0023] Optional mobile phone optical peripherals (such as simple prisms or corneal reflection acquisition accessories) are used once a month under normal circumstances to obtain high-precision corneal reflection point position data and correct minor errors in mobile eye position detection.

[0024] Fourth layer: In-house professional review layer.

[0025] By connecting to the hospital's ophthalmology follow-up system, patients can undergo a professional follow-up examination every 3 months within the first year after surgery, every 6 months from 1 to 2 years after surgery, and annually after 2 years. The examination results are entered into the system and used as the gold standard data to cover and calibrate the simple home test data, while also updating the baseline reference values ​​for various postoperative conditions.

[0026] The following embodiments will describe in detail each step of the method of the present invention according to the data flow of the above-described acquisition system. See also Figure 1 The figure shows that the method includes steps S100, S200, S300 and S400.

[0027] Step S100: Acquire multi-source temporal monitoring data of patients after strabismus surgery. The multi-source temporal monitoring data includes at least eye position sequence data and eye behavior data. The eye position sequence data includes horizontal eye position sequence, vertical eye position sequence and perceptual eye position deviation sequence collected during the first acquisition cycle when patients after strabismus surgery are induced to perform fixation tasks. In this embodiment, the multi-source time-series monitoring data includes at least eye position sequence data and eye behavior data. Under initial routine monitoring conditions, the system is set with a first acquisition cycle as the basic acquisition cycle. For example, one month after surgery, active data acquisition tasks are triggered twice a week, with a corresponding acquisition cycle duration of 3.5 days. Of course, for those skilled in the art, this cycle can be flexibly set according to postoperative time and patient condition. For example, it can be set to twice a week during the core postoperative follow-up period (3 months to 2 years postoperatively). This embodiment does not impose specific limitations on this. All data are collected by the front-facing camera of the mobile terminal and interactive applications. In this embodiment, by combining multi-source data fusion, it surpasses single static eye position examination, capturing dynamic fluctuations in eye position, long-term drift trends, and their correlation with visual function and behavioral habits, laying the foundation for subsequent accurate early warning.

[0028] Specifically, the eye position sequence data mentioned in this embodiment can be acquired through the following methods: In one implementation, during the execution of a gaze task, the mobile terminal continuously acquires a sequence of facial images of the patient at a preset frame rate. The preset frame rate can be 25 frames / second, 30 frames / second, 60 frames / second, etc., and the acquisition duration can be 6 seconds, 8 seconds, 10 seconds, etc. This embodiment uses 30 frames / second for 8 seconds as an example, acquiring approximately 240 frames of facial images. The facial image sequence is acquired by the front-facing camera of the mobile terminal.

[0029] While capturing each frame of facial image, the mobile terminal simultaneously performs a validity check of the current environment. That is, it synchronously and in real-time captures the ambient light intensity and the distance between the face and the mobile terminal screen at the moment of capture for each frame of facial image. Specifically, it uses an ambient light sensor to detect the current ambient light intensity in real time, and simultaneously uses a front-facing depth sensor or estimates based on the facial image size ratio to detect the distance between the face and the screen. The detected ambient light intensity is compared with a preset illuminance range, and the distance between the face and the screen is compared with a preset distance range.

[0030] The preset illumination range can be set to 100 lux to 300 lux. This range is selected based on commonly used illumination standards in ophthalmic examinations, ensuring that corneal reflections are clearly visible without causing patients to squint due to excessive light, thus affecting eye position assessment. In other implementations, this range can be 50 lux to 500 lux to accommodate a wider range of home environments. The preset distance range can be set to 30 cm to 50 cm, which is the typical working distance for mobile phone facial recognition, ensuring that the front-facing camera clearly captures eye details.

[0031] If the ambient light intensity at the current acquisition time is within the preset illuminance range and the distance between the face and the screen is within the preset distance range, then the facial image frame acquired at that time is determined to be a valid sampling frame; otherwise, it is determined to be an invalid sampling frame.

[0032] If a preset number of invalid sampling frames appear consecutively during the data acquisition process (e.g., 10, 15, or 20 consecutive frames), indicating a persistently poor current acquisition environment or posture, the mobile terminal will suspend the current gaze task and, based on the type of anomaly, control the mobile terminal's interactive application to output targeted prompts for adjusting ambient lighting or facial observation posture and distance. This can be achieved through interface pop-ups, voice announcements, or other means. For example, if the ambient light is too low, the system might prompt, "The current environment is dark; please turn on the lights and re-detect"; if the distance is too close, the system might prompt, "Please move your phone away from your face by about an arm's length." Simultaneously, this gaze task will be marked as "aborted midway" and will not be counted as a valid acquisition for the day. In one optimized implementation, the system can automatically suggest a re-collection period and remind the patient via push notification.

[0033] The acquired and deemed valid facial image frames constitute a facial image sequence. Then, the locations of corneal reflective points and pupil centers are extracted frame by frame from this sequence. Specific extraction methods can employ image segmentation and feature detection algorithms from computer vision. For example, the eye region can be cropped into a Region of Interest (ROI), and then thresholding and Gaussian fitting can be used to locate the corneal reflective points (the brightest areas with the highest grayscale values). The pupil center can be located using Hough transform or elliptic fitting. Alternatively, a deep learning-based convolutional neural network model can be used to directly regress the pixel coordinates of the corneal reflective points and pupil center from the cropped eye region image.

[0034] Based on the extracted corneal reflective point position and pupil center position in each frame of the image, a two-dimensional relative offset vector (Δx, Δy) between the two is calculated for each corresponding frame, in pixels. Then, using a preset spatial calibration matrix, this pixel offset vector is converted into a prism power (Δ) metric commonly used in ophthalmology, yielding the horizontal and vertical eye position values ​​for each frame. The spatial calibration matrix can be determined through a standard fixation calibration procedure before shipment. This procedure requires the subject to fixate on multiple calibration points at known locations on the screen to establish a mapping relationship between pixel offset and prism power.

[0035] Through the above processing, in a single fixation task, by extracting from consecutive valid facial image frames, a set of horizontal eye position values ​​arranged in chronological order can be obtained from the horizontal eye position values ​​corresponding to each frame, forming a horizontal eye position sequence; at the same time, a set of vertical eye position values ​​arranged in chronological order can be obtained from the vertical eye position values ​​corresponding to each frame, forming a vertical eye position sequence.

[0036] Simultaneously, during the fixation task, the perceived eye position deviation is measured, that is, the sequence of perceived eye position deviations between the subjective fixation point clicked by the patient after strabismus surgery and the objectively measured actual fixation point is obtained. The specific implementation process is as follows: The interactive application control display interface of the mobile terminal continuously displays the fixation target at the center of the interface. At the same time, at multiple preset locations off the center, test light points flash briefly in a single location according to a preset time sequence. The multiple preset locations can be 4 (up, down, left, right), 8 (up, down, left, right and four diagonal quadrants), or 16. The duration of each light point flashing (i.e., the preset time sequence) is, for example, 200 milliseconds to 500 milliseconds, and the flashing interval is, for example, 1 second to 2 seconds, to ensure that the patient has sufficient time to respond.

[0037] During the above operation, the patient was instructed to immediately tap the perceived location of the light spot on the touchscreen using a stylus or finger upon seeing it. The mobile terminal received the screen coordinates of each tap as tap data.

[0038] That is, the mobile terminal receives the click data collected by the touch screen. The click data is the single subjective click position on the touch screen corresponding to each flash of the test light point by the strabismus postoperative patient. Each click position corresponds one-to-one with each flash of the test light point. The subjective click coordinates of each test point are compared with the preset actual coordinates of that test point on the display interface to calculate a two-dimensional offset vector for each test point. This two-dimensional offset vector contains direction and amplitude information. The direction indicates the offset of the patient's subjective perceived position relative to the actual position, and the amplitude, or modulus, reflects the degree of offset. The unit can be converted into visual angle (degrees) or prism diopters (Δ).

[0039] Multiple two-dimensional offset vectors corresponding to the orientations of all measured light points in this fixation task are collected into a set. All offset vectors are then sorted sequentially according to the flashing time sequence of each test light point. This ordered sequence of two-dimensional offset vectors constitutes the perceived eye position offset sequence. It should be noted that the "temporal" attribute of this sequence reflects the chronological order in which the light points flash; therefore, the recording of each offset vector also carries a corresponding temporal order.

[0040] Meanwhile, considering the abnormal scenarios of missed clicks and multiple clicks on the same light point, which can also lead to the problem of disordered offset sequence data, those skilled in the art can directly select the first valid click position as the subjective click position corresponding to the light point if multiple click operations are detected on the same test light point; if there are test light points with missed clicks, the sampling data corresponding to the light point is discarded.

[0041] The original horizontal eye position sequences, vertical eye position sequences, and perceived eye position offset sequences obtained above were cleaned by outlier removal and noise reduction preprocessing, respectively. Outlier removal aims to eliminate abnormal sampling points that significantly deviate from the main distribution due to factors such as blinking, head movement, and instantaneous changes in light. Noise reduction preprocessing aims to smooth random measurement noise, making the signal closer to the actual eye position changes. For outlier removal, various statistical methods can be used. One implementation uses the 3σ criterion (Laida criterion): the mean μ and standard deviation σ of each sequence are calculated, and data points deviating from the mean by more than ±3σ are marked as outliers and removed. Another implementation uses the median absolute deviation method: the median M of the sequence and the median MAD (absolute deviation of each data point from M) are calculated, and points deviating from the median by more than a preset multiple (such as 2.5 or 3 times MAD) are removed. In another implementation, the IQR (interquartile range) method can be used: data points smaller than the lower quartile value Q1 minus 1.5 times the IQR, or larger than the upper quartile value Q3 plus 1.5 times the IQR, are removed as outliers, where IQR = Q3 – Q1. For noise reduction preprocessing, various filtering methods can also be used. In one implementation, a moving average filter is used, replacing the current data point's value with the arithmetic mean of the current data point and its k nearest neighbors; the window width can be 3, 5, or 7. In another implementation, a median filter is used, replacing the current value with the median within the neighborhood window, which is particularly effective for removing impulse noise. Yet another implementation uses wavelet thresholding denoising, performing multi-scale wavelet decomposition on the signal, then applying soft or hard thresholding to shrink high-frequency noise components before reconstructing a smooth signal. In higher-order implementations, a Kalman filter can be used, performing optimal recursive estimation of the noisy observation sequence based on the established eye position state space model. After the above outlier removal and noise reduction preprocessing, we obtain the horizontal eye position sequence, vertical eye position sequence, and perceptual eye position offset sequence after data cleaning and updating. These sequences have a higher signal-to-noise ratio and are suitable for subsequent feature extraction and trend analysis.

[0042] Meanwhile, the aforementioned eye-use behavior data is collected automatically and continuously in the background by the mobile terminal without the patient's awareness, requiring no active operation. In one implementation, the mobile terminal obtains screen usage duration and usage time information through statistical interfaces (such as Android's UsageStatsManager or iOS's ScreenTime API) via the operating system's applications; obtains ambient brightness data through a built-in ambient light sensor; and estimates the eye-use distance through the distance estimation function of the front-facing camera.

[0043] The specific statistics collected in this embodiment are as follows: Average daily near-field screen time: The cumulative screen time when the daily viewing distance is less than 40 centimeters (based on the front-facing camera estimate or a preset conservative value), in minutes or hours.

[0044] Number of continuous eye fatigue segments: A segment of continuous screen use exceeding a preset fatigue threshold (e.g., 40 minutes) is counted as a "fatigue segment," and the number of fatigue segments is counted daily / weekly. The determination of continuous use duration can be configured with an interruption tolerance interval (e.g., allowing a short break of no more than 2 minutes).

[0045] Frequency of screen use in low light: The number of times the screen is used when the ambient light intensity is below the low light threshold (e.g., 50 lux) (each time the screen is unlocked and used for a certain period of time is counted as one time).

[0046] Nighttime screen time: The total screen time during the scheduled nighttime period (e.g., 11:00 PM to 6:00 AM the next day).

[0047] These behavioral statistics are aggregated and calculated on a daily basis. They are then uploaded to the cloud server via a secure encrypted channel during the off-peak hours of each day by the mobile terminal, or stored in a local encrypted sandbox for use in the risk score calculation in step S340.

[0048] Meanwhile, visual function attenuation data from multi-source time-series monitoring is used to assess the patient's binocular visual function status, particularly changes in fusion and stereopsis. To facilitate home monitoring, in a simplified implementation, the visual function attenuation data is obtained through manual input by the patient. Specifically, the mobile terminal interactive application can be set up with a "Visual Function Self-Assessment" interface, which integrates standardized and simplified visual function self-testing tools (e.g., a quick self-test program for stereo acuity based on random dot stereograms, or a quick self-test program for fusion range based on screen flicker). After completing the self-test, the application interface will directly display or guide the patient to input key indicator results, such as "stereo acuity value" (unit: arcseconds) and "fusion range" (unit: prism diopters Δ). Furthermore, after regular professional follow-up examinations at the hospital, patients can manually input data such as fusion range attenuation, stereo acuity attenuation, and monocular suppression duration obtained from the current examination in the "Data Synchronization" or "Manual Input" module of this application, based on the doctor's instructions or the examination report. This data will be linked to the time point of collection to form time-series data, which will be used for risk assessment in subsequent steps.

[0049] Step S200: Extract the difference between the eye position sequence data and the postoperative baseline eye position data, and extract the eye position drift trend information of the current acquisition period relative to the postoperative baseline eye position data. The eye position drift trend information is obtained based on the offset time sequence of at least one of the horizontal eye position sequence, vertical eye position sequence and perceived eye position offset sequence relative to the baseline.

[0050] In this embodiment, postoperative baseline eye position data is considered the fundamental reference for assessing whether pathological regression of eye position has occurred. Therefore, in a preferred implementation, data measured by professional equipment in the hospital one month after surgery (i.e., the period of surgical wound healing and initial stabilization of the extraocular muscles) is selected as the fixed baseline. Specifically, it includes: Horizontal baseline eye position: the residual horizontal strabismus degree (unit: Δ) measured by the alternating prism occlusion method or synoptophore, under both corrective and uncorrected vision conditions. Ideally, it should be 0 Δ or close to orthogonal. Vertical baseline eye position: the residual vertical strabismus degree (unit: Δ) measured similarly. Perceptual baseline eye position: can be calibrated to zero deviation (i.e., consistency between subjective and objective perception), or determined in the hospital by perceptual eye position examination using a synoptophore. This baseline data remains unchanged after being entered into the system.

[0051] Meanwhile, to clarify the calculation method for eye position drift trend information, this embodiment uses a time series decomposition method to identify the implicit long-term pathological drift trend. In this embodiment, a target drift sequence (e.g., a horizontal drift sequence) is used as an example. The vertical drift sequence and the perceptual drift sequence can be processed in the same way. The target drift sequence is at least one of the horizontal drift sequence, the vertical drift sequence, and the perceptual drift sequence, specifically including: Step S210: Obtain the horizontal baseline eye position, vertical baseline eye position, and perceptual baseline eye position after strabismus surgery; Step S220: The mean horizontal eye position, mean vertical eye position, and mean perceptual eye position deviation corresponding to the current acquisition cycle and the historical acquisition cycle are respectively compared with the horizontal baseline eye position, vertical baseline eye position, and perceptual baseline eye position to calculate the difference, thereby obtaining the deviation value relative to the postoperative baseline in each acquisition cycle. All deviation values ​​are sorted according to the acquisition time to form a horizontal deviation sequence, a vertical deviation sequence, and a perceptual deviation sequence arranged in chronological order. In this embodiment, for each acquisition cycle, i.e. each successful fixation task, a representative index for that acquisition is calculated based on the preprocessed eye position sequence data: the horizontal eye position mean is obtained by taking the arithmetic mean of the preprocessed horizontal eye position sequence, the vertical eye position mean is obtained by taking the arithmetic mean of the preprocessed vertical eye position sequence, and the perceptual eye position offset mean is obtained by taking the arithmetic mean of the magnitudes of all offset vectors in the perceptual eye position offset sequence.

[0052] Then, the average values ​​obtained from the current acquisition cycle and multiple historical acquisition cycles, as well as all cycles prior to the current cycle, are compared with the corresponding fixed baseline values ​​to calculate the difference. For example: Horizontal offset (i) = Average horizontal eye position from the i-th acquisition – Horizontal baseline eye position; Vertical offset (i) = Average vertical eye position from the i-th acquisition – Vertical baseline eye position; Perceptual offset (i) = Average perceptual eye position offset from the i-th acquisition – Perceptual baseline eye position. Here, i represents the sequence number of the acquisition cycle. The difference sequence, arranged in chronological order of acquisition time, constitutes the horizontal offset sequence, vertical offset sequence, and perceptual offset sequence. Each offset represents the degree and direction of deviation of the patient's eye position from the ideal postoperative correction endpoint at the corresponding time point.

[0053] Step S230: Select the target offset sequence, perform time series decomposition, and extract the corresponding trend components. The target offset sequence is at least one of the horizontal offset sequence, vertical offset sequence, and perceptual offset sequence. The specific method for time series decomposition of the target offset sequence is as follows: The STL (Seasonal-Trend Decomposition using Loess) algorithm is employed. This algorithm iteratively applies the Loess smoother to decompose the original offset sequence into three components: a trend component (reflecting the long-term direction of change), a seasonal component (reflecting periodic patterns, such as day-night fluctuations and week-week differences), and a residual component (reflecting random fluctuations and noise). The decomposed trend component is extracted; this trend component is a smooth curve that removes short-term periodicity and random interference, focusing on reflecting the long-term evolution direction of eye position offset (gradual repositioning or gradual increase in drift). Next, a preset sliding time window is set on the trend component. This window captures a segment of the trend curve up to the most recent acquisition time (i.e., the current sliding time window ends at the last data acquisition time), for example, including the time span corresponding to the most recent 4 to 6 acquisition points. Within this window, the trend component data is fitted using linear least squares to obtain the slope of the fitted line. The slope is the eye position drift rate corresponding to the target offset sequence. Its absolute value reflects the speed of eye position drift, and the positive or negative sign indicates the drift direction.

[0054] It should be noted that those skilled in the art can also use the classical moving average decomposition method: apply a central moving average with a width of 3 or 5 sampling points to the target offset sequence (e.g., a horizontal offset sequence), and the resulting smoothed sequence is the trend component. In another implementation, the Hodrick-Prescott (HP) filtering method is used: by minimizing the objective function (sum of squared deviations between the actual sequence and the trend term + smoothing parameter λ × sum of squared differences of the trend term), the trend component sequence is directly separated. Alternatively, the Kalman filtering method can be used: model the eye position offset as a latent linear state (uniform or uniformly accelerated model) under noisy observation, and estimate the "true offset state" sequence, i.e., the trend component, through Kalman recursion. Regardless of the method used, after obtaining the trend component, the subsequent calculation of the eye position drift rate and the eye position drift acceleration factor remains consistent with the above, and will not be repeated in this embodiment.

[0055] Step S240: Based on the trend component, perform linear fitting within a preset sliding time window, and determine the eye position drift rate corresponding to the target offset sequence according to the slope of the fitted line; wherein, when there are multiple target offset sequences, multiple eye position drift rates are obtained respectively, and the multiple eye position drift rates are numerically merged to obtain a single integrated eye position drift rate. When the horizontal offset sequence, vertical offset sequence, and perceptual offset sequence are decomposed and slope extracted as described above to obtain multiple eye position drift rates (horizontal drift rate, vertical drift rate, and perceptual drift rate), these multiple eye position drift rates are numerically merged to obtain a final eye position drift rate from the eye position drift trend information. The numerical merging process includes methods such as calculating the mean (arithmetic mean or geometric mean) or taking the maximum / minimum value (taking the rate with the largest absolute value, i.e., the most significant deterioration, as the representative). For example, in one implementation, the final drift rate = (horizontal drift rate + vertical drift rate + perceptual drift rate) / 3. In another implementation, the final drift rate = max(|horizontal drift rate|, |vertical drift rate|, |perceptual drift rate|), taking the rate value with the fastest deterioration of deviation.

[0056] Further, to detect whether the drift trend itself is accelerating, the change in the drift rate is measured. See step S250 for details.

[0057] Step S250: Based on the rate of change of eye position drift rate corresponding to multiple consecutive preset sliding time windows, generate an eye position drift acceleration factor; when multiple eye position drift acceleration factors are generated, perform numerical merging processing on the multiple eye position drift acceleration factors to obtain a single integrated eye position drift acceleration factor; wherein, the numerical merging processing includes calculating the average or taking the maximum value; it should be noted that the preset sliding time window can be set to 1 month or 2 weeks.

[0058] Based on the comprehensive eye position drift rate corresponding to multiple consecutive (e.g., the most recent 3 or 5) preset sliding time windows (each time window will generate a drift rate value), the changes of these consecutive rate values ​​are compared.

[0059] In one implementation, the change in rate between adjacent windows is calculated. , This represents the drift rate value corresponding to the first preset sliding time window. This represents the drift rate value corresponding to the second preset sliding time window. This represents the drift rate value corresponding to the third preset sliding time window, and so on. The arithmetic mean or maximum value of multiple changes is used as the eye position drift acceleration factor. If this factor is positive, it indicates that the drift rate is increasing, representing an accelerated deterioration of the eye position and a dangerous signal of relapse; if it is close to zero or negative, it indicates that the drift rate is stable or has slowed down. In another implementation, the comprehensive drift rate sequence of continuous windows can be linearly fitted, and the slope of the fitted line can be used as the acceleration factor. In yet another implementation, the sign and magnitude of the continuous rates can be statistically tested (such as the Mann-Kendall trend test), and the test statistic or its normalized value can be used as the acceleration factor.

[0060] When multiple acceleration factors are generated for multiple offset sequences, these multiple eye position drift acceleration factors are also numerically merged to obtain a comprehensive eye position drift acceleration factor. The numerical merging process includes averaging or taking the maximum / minimum value. Finally, the comprehensive eye position drift rate and the comprehensive eye position drift acceleration factor are combined as the output eye position drift trend information.

[0061] Step S260: Use the eye position drift rate and the eye position drift acceleration factor as the eye position drift trend information.

[0062] Step S300: Calculate the recurrence risk score based on the eye position sequence data, the eye position drift trend information, and the eye use behavior data, and determine the corresponding warning level based on the recurrence risk score; This step uses a multi-factor weighted fusion model as its core, quantifying risk signals from multiple dimensions into a recurrence risk score ranging from 0 to 100. The first step in this step is to calculate the risk factor scores for each dimension. Specifically, this includes: Step S310: Calculate the gaze stability fluctuation variance set based on the eye position sequence data; In this embodiment, the gaze stability fluctuation variance is calculated based on the preprocessed and updated horizontal eye position sequence, vertical eye position sequence, and perceived eye position offset sequence. The gaze stability fluctuation variance set characterizes the degree of stability fluctuation of the patient's eyes when fixing on the target within one acquisition cycle (single gaze task). The larger the variance, the more unstable the eyeball is, and the worse the eye muscle control may be.

[0063] For ease of understanding, this embodiment assumes a horizontal eye position sequence. ,in This represents the number of valid sampled frames. The first step is to calculate the mean of the horizontal eye position sequence. Then, the variance of horizontal fixation fluctuation corresponding to the horizontal eye position sequence is calculated. Similarly, those skilled in the art can also perform vertical gaze fluctuation variance analysis. The calculations will not be elaborated upon in this application.

[0064] Similarly, for a sequence of perceived eye position offsets, first calculate the magnitude of each offset vector to obtain a sequence of magnitudes, then calculate the variance of this sequence to obtain the variance of the perceptual eye position offset fluctuation. These three variances together constitute the gaze stability fluctuation variance set.

[0065] In other implementations, "variance" can be replaced or extended to other statistics that can represent the degree of dispersion, such as: standard deviation (the arithmetic square root of the variance, with the same dimensions as the original data), range (maximum value minus minimum value), mean absolute deviation, coefficient of variation, or interquartile range, collectively referred to as "variance". These alternative statistics also constitute equivalent content of the gaze stability variance set described in this invention.

[0066] The horizontal gaze fluctuation variance calculated above Vertical gaze fluctuation variance and variance of perceptual eye position deviation This constitutes the gaze stability fluctuation variance set, which serves as a three-dimensional feature vector. Output to subsequent steps.

[0067] Step S320: Input the eye position drift rate and the eye position drift acceleration factor into a preset first mapping function to obtain the eye position drift rate factor score; Among them, the first mapping function can be understood by those skilled in the art. Designed as a two-dimensional piecewise linear mapper or a two-dimensional lookup table. For example, based on drift rate. Divide the absolute value into segments: when At that time, the drift rate score is 20 points; when At that time, the drift rate score is 20 points; when At that time, the drift rate sub-score = 85 points; for the acceleration factor Segmentation: At that time, the acceleration factor score = 10 points; ,and At that time, the acceleration factor score = 50 points; At that time, the acceleration factor score was 80. Finally, the eye position drift rate factor score... = 0.7 × drift rate sub-score + 0.3 × acceleration sub-score. Those skilled in the art will understand that the specific thresholds, scores, and weights can be optimized and adjusted based on clinical samples.

[0068] Step S330: Input the mean of perceptual eye position shift and the variance of perceptual eye position shift fluctuation within the current sliding time window into the preset second mapping function to obtain the perceptual eye position shift factor score; Similarly, those skilled in the art can interpret the second mapping function. The design is a weighted summation function after min-max normalization. The mean and variance of perceived eye position shift within the current sliding time window are mapped to the interval [0,1], respectively. In the normalization, the maximum value of the mean perceived eye position shift can be preset to 15Δ, and the variance of the variance can be preset to 5Δ. 2 Then, based on this, the perceptual eye position deviation factor score is calculated. The perceptual eye position deviation factor score can be specifically set as follows: in, This represents the score of the perceived eye position deviation factor; In this embodiment, the adjustable weights are indicated. ; This represents a normalization function applicable to the mean of perceived eye position deviation; This represents the normalization function applicable to the variance of perceptual eye position deviation fluctuation; In this embodiment, the adjustable weights are indicated. ; This represents the variance of perceptual eye position deviation.

[0069] Step S340: Obtain visual function attenuation data within the same sliding time window. The visual function attenuation data includes at least the fusion range attenuation, stereoscopic acuity attenuation, and monocular suppression duration. Input the visual function attenuation data into a preset third mapping function to obtain the visual function attenuation factor score. In this embodiment, the visual function attenuation data includes at least the fusion range attenuation, stereoscopic acuity attenuation, and monocular suppression duration. Specifically, the fusion range attenuation is calculated as: baseline fusion range – current fusion range (unit: Δ); stereoscopic acuity attenuation is calculated as: current stereoscopic acuity value – baseline stereoscopic acuity value (unit: arcseconds); and monocular suppression duration attenuation is calculated as: current monocular suppression duration – baseline monocular suppression duration (unit: seconds / detection). These three attenuation values ​​are input into a third mapping function. Meanwhile, this embodiment provides an exemplary approach, which involves using the third mapping function. A three-dimensional segmented score lookup table is used, and its design is shown in Table 1:

[0070] That is, those skilled in the art can look up the corresponding interval score in a table based on the actual value of each attenuation, and then sum the three scores according to preset weights (such as fusion range weight 0.4, stereoscopic acuity weight 0.4, and monocular suppression duration weight 0.2) to obtain the visual function attenuation factor score. .

[0071] Step S350: Extract the average daily near-field eye use duration, number of consecutive eye fatigue segments, frequency of eye use in dim light, and nighttime eye use duration from the eye use behavior data within the same sliding time window. Input the above parameters into the preset fourth mapping function to obtain the score of poor eye use behavior factor. In this step, eye-use behavior data, including average daily near-vision time, is collected. Number of segments of continuous eye strain Frequency of eye use in dim light and nighttime screen time And input the fourth mapping function. For example, those skilled in the art can use the fourth mapping function A normalized weighted function based on clinical upper limit reference values ​​is used. Clinical safety upper limits are set for each parameter, for example: , , , Each value is divided by the upper limit and truncated to [0,1]. Then, it is multiplied by 100, and the final score for the poor eye use behavior factor is calculated as follows: in, This indicates the score of the poor eye use behavior factor; , , , The value represents the weight of eye use; in this embodiment, it is 0.25. However, those skilled in the art can set other values. This indicates the average daily duration of close-range visual use. This represents a normalization function applicable to the average daily duration of close-range eye use. This indicates the number of consecutive segments that cause eye strain. This represents a normalization function applicable to the number of consecutive segments of eye strain. Indicates the frequency of eye use in dim light; This represents a normalization function applicable to the frequency of eye use in low light conditions. Indicates the duration of nighttime screen time; This represents a normalization function applicable to nighttime eye use duration.

[0072] Step S360: Based on the horizontal and vertical offsets within the same sliding time window, calculate the total eye position offset in two dimensions, and input the preset fifth mapping function to obtain the surgical correction margin factor score. Specifically, total eye position deviation That is, take the root mean square value of the average offset within the window.

[0073] And the fifth mapping function For linear mapping functions: in, This represents the score of the surgical correction margin factor; k represents the proportionality coefficient, such that when When the score varies within the range of 0 to 20Δ, the score covers a range of 0 to 100 points. For example, if k = 5, then... hour, , hour, .

[0074] Step S370: Pre-configure corresponding weight coefficients for the eye position drift rate factor score, the perceived eye position deviation factor score, the visual function decline factor score, the poor eye use behavior factor score, and the surgical correction margin factor score, and generate a basic risk score by weighted summation; Specifically, the basic risk score is calculated as follows: in, This indicates the basic risk score; This represents the score of the eye position drift rate factor; This represents the score of the perceived eye position deviation factor; This represents the visual function attenuation factor score; This indicates the score of the poor eye use behavior factor; This represents the score of the surgical correction margin factor.

[0075] The above calculation formula reflects the most crucial decision-making role of eye position drift rate, followed by perceived eye position deviation, then visual function decline, with eye use behavior as a triggering factor and surgical correction margin as a boundary constraint. It should be noted that the above weights are preferred example values, and the weights can be fine-tuned for different types of strabismus or patients of different ages.

[0076] Meanwhile, in this embodiment, to improve the sensitivity of identifying the special high-risk pattern of multidimensional gaze instability combined deterioration, a correction factor is calculated based on the gaze stability fluctuation variance set, and the basic risk score is adjusted upward. See step S470 for details.

[0077] Step S380: Perform a fusion calculation on the variance values ​​of the gaze stability fluctuation variance set to obtain a risk correction factor, and superimpose a preset adjustment score or perform an upward weighting on the basic risk score to obtain the recurrence risk score.

[0078] Specifically, the horizontal gaze fluctuation variance calculated in step S200 within the same acquisition period Vertical gaze fluctuation variance and variance of perceptual eye position deviation Combined into a three-dimensional anomaly vector The corresponding vector magnitude is obtained by calculating the magnitude based on this three-dimensional anomaly vector.

[0079] The vector magnitude is calculated using the Euclidean norm: in, Indicates the magnitude of the vector; This represents the variance of horizontal gaze fluctuation; This represents the variance of vertical gaze fluctuation; This represents the variance of perceptual eye position deviation fluctuation. The modulus is calculated using the above formula. It comprehensively reflects the combined fluctuation amplitude of the three dimensions: horizontal, vertical, and perception.

[0080] At the same time, a preset joint anomaly threshold is set (e.g., 1.8△). 2 2.0△ 2 Or 2.5△ 2 This threshold can be determined through ROC curve analysis of clinical retrospective data (the optimal cut-off point can be determined). If the vector modulus corresponds to a preset number of consecutive acquisition cycles... When the number of consecutive preset number of data acquisitions (e.g., 2 or 3 consecutive acquisitions) exceeds the preset joint anomaly threshold, it is determined that a persistent, multi-dimensional gaze stability collapse has occurred, which is a signal indicating a general decline in eye muscle coordination control. At this point, the baseline risk score is... If the above-mentioned continuous abnormality determination conditions are not met, the basic risk score will be used as the recurrence risk score.

[0081] The correction method can involve adding a preset adjustment score to the basic risk score or performing a weighted upward adjustment. In one implementation, a preset adjustment score is added; for example, by giving... Add a preset fixed score (e.g., 8, 10, or 12 points), i.e., relapse risk score. In another implementation, an upward weighting method is used, for example, by adjusting the weighting... Multiply by an upward weighting factor (For example or (i.e., relapse risk score) The revised score will be used as the final recurrence risk score for determining the warning level. Of course, if the modulus is long If there are no consecutive exceedances, no correction will be triggered. .

[0082] Step S400: Dynamically adjust the collection cycle duration of the next multi-source time-series monitoring data based on the warning level, so as to switch the first collection cycle to the second collection cycle corresponding to the warning level, wherein the collection cycles corresponding to different warning levels are different.

[0083] This step is based on the aforementioned recurrence risk score. This is mapped to a pre-defined four-level early warning system to determine the corresponding early warning level: Level 1: Stable. The eye position was basically stable, with no persistent pathological drift; visual function was basically normal, and all indicators were within the physiological fluctuation range. Secondary observation: Slight or non-persistent eye misalignment, or mild abnormalities in a single indicator, but not yet showing a trend of continuous deterioration. Level 3 Warning: Persistent pathological eye displacement, coupled with progressive visual function decline or significant behavioral risk factors, shows a tendency for early, insidious recurrence. Level 4 High Risk: Significant and persistent regression of eye position and rapid decline in visual function strongly suggest that early overt recurrence is imminent or has already occurred.

[0084] Based on the current warning level determined by the above steps, the mobile terminal generates corresponding acquisition control commands, which affect the triggering frequency of the next and subsequent fixation tasks. The acquisition cycle duration refers to the time interval between two consecutive fixation tasks. The higher the warning level, the greater the risk of recurrence, requiring more intensive data sampling to finely track eye position dynamics; therefore, the acquisition cycle duration is set shorter.

[0085] The specific switching rules are implemented as follows: When the warning level is Level 1 (stable), the original first collection cycle is maintained. For example, the first collection cycle is 3.5 days (corresponding to approximately 2 collections per week), and this interval is maintained to trigger the next fixation task. When the warning level is Level 2 (observation), the collection cycle is shortened from the first collection cycle to the second cycle. For example, the second cycle is 2 days (corresponding to approximately 3.5 collections per week), increasing the monitoring density to closely observe the trend of minor abnormalities. When the warning level is Level 3 (alert), the collection cycle is further shortened to the third cycle. For example, the third cycle is 1 day (corresponding to 1 collection per day), using high-frequency sampling to precisely capture the evolution characteristics of hidden drift. When the warning level is Level 4 (high risk), the collection cycle is drastically shortened to the fourth cycle. For example, the fourth cycle is 0.5 days (corresponding to 1 collection each morning and evening, for a total of 2 collections), entering an emergency state of highest-density monitoring to keep abreast of the dynamic deterioration of eye position and buy time for emergency clinical intervention.

[0086] Meanwhile, the retesting frequency of the third layer (external precision retesting layer) and the fourth layer (in-hospital professional retesting layer) mentioned above is also adjusted accordingly based on the warning level. For example, after a level three warning is triggered, the system will forcibly remind the patient to complete an outpatient retest within 15 days; after a level four high-risk warning is triggered, the patient will be reminded to have an emergency retest within 3 days.

[0087] Through the above methods, the entire system forms an intelligent closed loop of "periodic monitoring → multi-dimensional feature extraction → comprehensive recurrence risk scoring → early warning level determination → dynamic adjustment of collection cycle → next higher / lower frequency monitoring", which realizes adaptive matching of monitoring resources and recurrence risk. While reducing the follow-up burden of low-risk patients, it maximizes the early warning capability for high-risk patients.

[0088] Example 2: like Figure 2 As shown, this embodiment provides a strabismus surgery recurrence early warning system, the system comprising: The data acquisition module is used to acquire multi-source temporal monitoring data of patients after strabismus surgery. The multi-source temporal monitoring data includes at least eye position sequence data and eye behavior data. The eye position sequence data includes horizontal eye position sequence, vertical eye position sequence and perceptual eye position deviation sequence collected during the induction of strabismus patients to perform fixation tasks within the first acquisition cycle. The feature extraction module is used to extract the difference between the eye position sequence data and the postoperative baseline eye position data, as well as the eye position drift trend information relative to the postoperative baseline eye position data within the acquisition period corresponding to the current fixation task. The eye position drift trend information is obtained based at least on the offset time sequence of one of the horizontal eye position sequence, vertical eye position sequence and perceptual eye position offset sequence relative to the postoperative baseline eye position data. The risk assessment module is used to calculate a recurrence risk score based on the eye position sequence data, the eye position drift trend information, and eye use behavior data, and to determine the corresponding warning level based on the recurrence risk score. The period control module is used to dynamically adjust the duration of the next collection period of the multi-source time-series monitoring data based on the warning level, so as to switch the first collection period to the second collection period corresponding to the warning level, wherein the collection periods corresponding to different warning levels are different.

[0089] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0090] Example 3: Corresponding to the above method embodiments, this embodiment also provides a strabismus recurrence early warning device. The strabismus recurrence early warning device described below and the strabismus recurrence early warning method described above can be referred to in correspondence.

[0091] Figure 3 This is a block diagram illustrating a strabismus postoperative recurrence early warning device 800 according to an exemplary embodiment. Figure 3 As shown, the strabismus recurrence early warning device 800 may include: a processor 801 and a memory 802. The strabismus recurrence early warning device 800 may also include one or more of the following: a multimedia component 803, an I / O interface 804, and a communication component 805.

[0092] The processor 801 controls the overall operation of the strabismus recurrence warning device 800 to complete all or part of the steps in the aforementioned strabismus recurrence warning method. The memory 802 stores various types of data to support the operation of the strabismus recurrence warning device 800. This data may include, for example, instructions for any application or method operating on the strabismus recurrence warning device 800, and application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the strabismus recurrence early warning device 800 and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, and an NFC module.

[0093] In an exemplary embodiment, the strabismus recurrence warning device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the strabismus recurrence warning method described above.

[0094] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the strabismus recurrence warning method described above. For example, the computer-readable storage medium may be the memory 802 including the program instructions described above, which may be executed by the processor 801 of the strabismus recurrence warning device 800 to complete the strabismus recurrence warning method described above.

[0095] Example 4: Corresponding to the above method embodiments, this embodiment also provides a readable storage medium. The readable storage medium described below can be referred to in conjunction with the strabismus postoperative recurrence early warning method described above.

[0096] A readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the strabismus postoperative recurrence early warning method of the above-described method embodiments.

[0097] Specifically, the readable storage medium can be a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other readable storage medium capable of storing program code.

[0098] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0099] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for early warning of recurrence after strabismus surgery, characterized in that, include: Acquire multi-source temporal monitoring data of patients after strabismus surgery. The multi-source temporal monitoring data includes at least eye position sequence data and eye behavior data. The eye position sequence data includes horizontal eye position sequence, vertical eye position sequence and perceptual eye position deviation sequence collected during the first acquisition cycle when patients after strabismus surgery are induced to perform fixation tasks. Extract the difference between the eye position sequence data and the postoperative baseline eye position data, and the eye position drift trend information relative to the postoperative baseline eye position data within the acquisition period corresponding to the current fixation task. The eye position drift trend information is obtained based at least on the offset time sequence of one of the horizontal eye position sequence, vertical eye position sequence and perceptual eye position offset sequence relative to the postoperative baseline eye position data. A recurrence risk score is calculated based on the eye position sequence data, the eye position drift trend information, and the eye use behavior data, and the corresponding warning level is determined according to the recurrence risk score. The duration of the next collection cycle of the multi-source time-series monitoring data is dynamically adjusted based on the warning level, so as to switch the first collection cycle to the second collection cycle corresponding to the warning level, wherein the collection cycles corresponding to different warning levels are different.

2. The method for early warning of recurrence after strabismus surgery according to claim 1, characterized in that, The acquisition of multi-source temporal monitoring data of patients after strabismus surgery includes: In the same fixation task, a sequence of facial images of patients after strabismus surgery is continuously acquired at a preset frame rate. In the fixation task, the perceptual eye position offset sequence between the subjective fixation point clicked by the patient after strabismus surgery and the objectively measured actual fixation point is obtained. Based on the acquired facial image sequence, the corneal reflection point position and pupil center position are extracted from each frame of the facial image; Based on the relative offset between the corneal reflective point and the pupil center in each frame of the image, the horizontal eye position value and the vertical eye position value of the corresponding frame are calculated. The horizontal eye position sequence and the vertical eye position sequence are formed by the horizontal eye position value and the vertical eye position value of each frame respectively. The perceptual eye position offset sequence, horizontal eye position sequence and vertical eye position sequence are cleaned by performing outlier removal and noise reduction preprocessing respectively. After data cleaning and updating, the updated perceptual eye position offset sequence, horizontal eye position sequence and vertical eye position sequence are obtained.

3. The method for early warning of recurrence after strabismus surgery according to claim 2, characterized in that, Obtain the sequence of perceived eye position offsets between the subjective fixation point (based on subjective clicking) and the objectively measured actual fixation point of the strabismus patient after surgery, including: The display interface of the mobile terminal interactive application continuously displays the gaze target at the center of the interface, and at multiple preset locations off the center, the test light point flashes briefly in a single location according to a preset time sequence each time; it receives click data collected by the touch screen, the click data being the single subjective click position on the touch screen corresponding to each flashing test light point by the strabismus postoperative patient, and each click position corresponds one-to-one with each flashing test light point; The subjective click coordinates of each test light point are compared with the preset actual coordinates of the test light point in the display interface to calculate the two-dimensional offset vector corresponding to each test light point. All offset vectors are sorted sequentially according to the flashing time sequence of each test light point, and the perceptual eye position offset sequence is composed of multiple ordered two-dimensional offset vectors.

4. The method for early warning of recurrence after strabismus surgery according to claim 2, characterized in that, The process of acquiring the facial image sequence of the strabismus patient after surgery also includes the following steps: Simultaneously and in real-time, the ambient light intensity and the distance between the face and the mobile terminal screen are collected at the moment of each frame of facial image acquisition. If the ambient light intensity at the current acquisition time is within the preset illuminance range and the distance between the face and the screen is within the preset distance range, then the facial image frame acquired at that time is determined to be a valid sampling frame; otherwise, it is determined to be an invalid sampling frame. All invalid sampled frames are removed, and only the valid sampled frames are combined according to the acquisition time sequence to obtain the facial image sequence; If a preset number of invalid sampling frames appear consecutively in the time sequence, the current gaze task is terminated, and the mobile terminal interactive application is controlled to output targeted prompts for adjusting ambient lighting or adjusting face observation posture and observation distance according to the type of anomaly.

5. The method for early warning of recurrence after strabismus surgery according to claim 1, characterized in that, Based on the changes in the aforementioned eye position sequence data and postoperative baseline eye position data, information on eye position drift trend is obtained, including: Obtain the horizontal baseline eye position, vertical baseline eye position, and perceptual baseline eye position after strabismus surgery; The mean horizontal eye position, mean vertical eye position, and mean perceptual eye position offset corresponding to the current acquisition cycle and the historical acquisition cycle are respectively compared with the horizontal baseline eye position, vertical baseline eye position, and perceptual baseline eye position to calculate the difference, thereby obtaining the deviation amount relative to the postoperative baseline in each acquisition cycle. All deviation amounts are sorted according to the acquisition time to form a horizontal offset sequence, a vertical offset sequence, and a perceptual offset sequence, respectively. The task cycle is the data acquisition cycle in which a fixation task is performed. Select a target offset sequence for time series decomposition and extract the corresponding trend components. The target offset sequence is at least one of the horizontal offset sequence, vertical offset sequence and perceptual offset sequence. Based on the trend component, linear fitting is performed within a preset sliding time window, and the eye position drift rate corresponding to the target offset sequence is determined according to the slope of the fitted line; wherein, when there are multiple target offset sequences, multiple eye position drift rates are obtained respectively, and the multiple eye position drift rates are numerically merged to obtain a single integrated eye position drift rate. Based on the rate of change of eye position drift rate corresponding to multiple consecutive preset sliding time windows, an eye position drift acceleration factor is generated; when multiple eye position drift acceleration factors are generated, the multiple eye position drift acceleration factors are numerically merged to obtain a single integrated eye position drift acceleration factor; wherein, the numerical merging process includes calculating the mean or taking the maximum value. The integrated eye position drift rate and the eye position drift acceleration factor are used together as the eye position drift trend information.

6. The method for early warning of recurrence after strabismus surgery according to claim 5, characterized in that, Claim 5 calculates a recurrence risk score based on the eye position sequence data, the eye position drift trend information, and eye use behavior data, specifically including the following steps; All data used in each step are taken from the same preset sliding time window, and the end time of this sliding time window is the last time all monitoring data is collected. A set of variances for gaze stability fluctuations is calculated based on the eye position sequence data, wherein the set of variances for gaze stability fluctuations includes horizontal gaze fluctuation variance, vertical gaze fluctuation variance, and perceptual eye position deviation fluctuation variance. The eye position drift rate and the eye position drift acceleration factor are input into a preset first mapping function to obtain the eye position drift rate factor score; Input the mean of perceptual eye position shift and the variance of perceptual eye position shift fluctuation within the current sliding time window into the preset second mapping function to obtain the perceptual eye position shift factor score; Obtain visual function attenuation data within the same sliding time window. The visual function attenuation data includes at least the fusion range attenuation, stereoscopic acuity attenuation, and monocular suppression duration. Input the visual function attenuation data into a preset third mapping function to obtain the visual function attenuation factor score. Extract the average daily near-field eye use duration, number of consecutive eye fatigue segments, frequency of eye use in dim light, and nighttime eye use duration from the eye use behavior data within the same sliding time window. Input the above parameters into the preset fourth mapping function to obtain the score of poor eye use behavior factor. Based on the horizontal and vertical offsets within the same sliding time window, the total eye position offset is calculated in two dimensions. The total eye position offset is then input into a preset fifth mapping function to obtain the surgical correction margin factor score. The corresponding weighting coefficients are pre-configured for the eye position drift rate factor score, the perceived eye position deviation factor score, the visual function decline factor score, the poor eye use behavior factor score, and the surgical correction margin factor score, and the weighted sum is used to generate a basic risk score. The risk correction factor is obtained by fusion calculation of the variance values ​​of the gaze stability fluctuation variance set; the recurrence risk score is obtained by superimposing a preset adjustment score or by weighted adjustment on the basic risk score.

7. The method for early warning of recurrence after strabismus surgery according to claim 6, characterized in that, The relapse risk score is obtained by correcting the basic risk score based on the gaze stability fluctuation variance set, specifically including the following steps: The horizontal gaze fluctuation variance, vertical gaze fluctuation variance, and perceptual eye position deviation fluctuation variance within the same acquisition period are normalized respectively, and the three sets of normalized variance values ​​are combined to form a three-dimensional anomaly vector. The corresponding vector magnitude is obtained by calculating the magnitude of the three-dimensional anomaly vector based on the Euclidean norm. If the vector magnitude corresponding to a consecutive preset number of adjacent acquisition cycles is greater than the preset joint anomaly threshold, the basic risk score is superimposed with a preset adjustment score or weighted upward to obtain the recurrence risk score; if the consecutive anomaly determination condition is not met, the basic risk score is used as the recurrence risk score.

8. A strabismus surgery recurrence early warning system, characterized in that, include: The data acquisition module is used to acquire multi-source temporal monitoring data of patients after strabismus surgery. The multi-source temporal monitoring data includes at least eye position sequence data and eye behavior data. The eye position sequence data includes horizontal eye position sequence, vertical eye position sequence and perceptual eye position deviation sequence collected during the induction of strabismus patients to perform fixation tasks within the first acquisition cycle. The feature extraction module is used to extract the difference between the eye position sequence data and the postoperative baseline eye position data, as well as the eye position drift trend information relative to the postoperative baseline eye position data within the acquisition period corresponding to the current fixation task. The eye position drift trend information is obtained based at least on the offset time sequence of one of the horizontal eye position sequence, vertical eye position sequence and perceptual eye position offset sequence relative to the postoperative baseline eye position data. The risk assessment module is used to calculate a recurrence risk score based on the eye position sequence data, the eye position drift trend information, and eye use behavior data, and to determine the corresponding warning level based on the recurrence risk score. The period control module is used to dynamically adjust the duration of the next collection period of the multi-source time-series monitoring data based on the warning level, so as to switch the first collection period to a second collection period corresponding to the warning level, wherein the collection periods corresponding to different warning levels are different.

9. A strabismus surgery recurrence early warning device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the strabismus recurrence early warning method as described in any one of claims 1 to 7 when executing the computer program.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the steps of the strabismus postoperative recurrence early warning method as described in any one of claims 1 to 7.

Images