An ocular surface monitoring method and system based on video analysis

Abstract

This invention proposes a video-based ocular surface monitoring method and system, belonging to the field of ocular surface monitoring technology. The method includes: acquiring ocular surface video sequences; using a geometrically constrained cascaded localization strategy to identify anatomical structures in each frame of the image, obtaining the location information of key anatomical structures, and calculating the palpebral fissure height; calculating the average tear meniscus height and the standard coefficient of variation of the tear meniscus height based on the palpebral margin position; calculating the corneal exposed area; detecting blinking events based on the temporal changes in palpebral fissure height from consecutive frames of the video sequence, and calculating the blinking frequency and the proportion of incomplete blinks; constructing a tear fluid dynamic balance model, substituting the average tear meniscus height, the standard coefficient of variation of the tear meniscus height, the corneal exposed area, the blinking frequency, and the proportion of incomplete blinks into the tear fluid dynamic balance model to calculate the tear fluid dynamic balance index, and performing a graded assessment of ocular surface health status based on this index. This invention enables a quantitative and comprehensive assessment of ocular surface health status.

Description

Technical Field

[0001] This invention relates to the field of ocular surface monitoring technology, and in particular to an ocular surface monitoring method and system based on video analysis. Background Technology

[0002] Ocular surface health monitoring plays a crucial role in the diagnosis of ophthalmic diseases such as dry eye and corneal diseases. Current ocular surface monitoring methods primarily rely on manual measurement of indicators such as tear meniscus height and palpebral fissure height, which suffers from problems such as strong subjectivity in measurement location selection, poor consistency of results among different doctors, and low efficiency in measuring multiple indicators separately. Furthermore, traditional comprehensive assessment methods often employ simple linear weighting, failing to fully reflect the dynamic balance between tear supply and demand on the ocular surface. Therefore, an automated, standardized monitoring method capable of comprehensively assessing the health status of the ocular surface is needed.

[0003] Chinese patent application CN115397306A discloses a method for measuring the tear trough. This method involves instilling fluorescein onto the surface of the patient's eye, capturing an image of the eye using blue light, identifying the tear trough region based on the color contrast between the fluorescent green area and the non-fluorescent blue area, and calculating the actual physical height of the tear trough using the outer diameter of the iris as a reference standard. This patent uses a regression multinomial to calculate the upper and lower contours of the tear trough, achieving automated measurement of tear trough height. However, this patent only measures the single indicator of tear trough height and cannot comprehensively assess the health status of the ocular surface. Summary of the Invention

[0004] In view of this, the present invention provides a video analysis-based ocular surface monitoring method and system, which uses a geometric constraint cascade positioning strategy to accurately identify key anatomical structures of the ocular surface, obtains spatial distribution information of tear meniscus height through adaptive multi-point measurement, and establishes a dynamic tear balance assessment model by combining parameters such as palpebral fissure height, blinking frequency, incomplete blinking, and corneal exposure area, thereby achieving a quantitative comprehensive assessment of the health status of the ocular surface.

[0005] The technical solution of this invention is implemented as follows: On one hand, the present invention provides a video analysis-based method for ocular surface monitoring, comprising: S1. Acquire video sequences of the ocular surface, preprocess consecutive frame images, and use a geometric constraint cascade localization strategy to identify anatomical structures in each frame image, sequentially locating the pupil, corneal boundary, and upper and lower eyelid margins to obtain the location information of key anatomical structures and calculate the palpebral fissure height. S2. Based on the eyelid margin position, an adaptive multi-point measurement strategy is used to obtain the spatial distribution information of the tear river height, and the average tear river height and the standard coefficient of variation of the tear river height are calculated. S3. Calculate the corneal exposure area based on the location information of key anatomical structures; S4. Based on continuous frame images of video sequences, blinking events are detected by temporal changes in palpebral fissure height, and blinking frequency and incomplete blinking ratio are calculated; S5. Construct a tear dynamic balance model, and substitute the average tear river height, standard coefficient of variation of tear river height, corneal exposed area, blink frequency, and proportion of incomplete blinking into the tear dynamic balance model to calculate the tear dynamic balance index, and classify and assess the ocular surface health status based on the index.

[0006] Preferably, in step S1, the geometric constraint cascade positioning strategy includes: The Hough circle detection algorithm is used to accurately locate the pupil and obtain the coordinates of the pupil center. and pupil radius The RANSAC algorithm was used to optimize and fit the detected circular edge points. Based on the proportional relationship that the corneal diameter is 3 to 5 times the pupil diameter, the search radius of the corneal boundary is set to be... Using the pupil center as the center, the radial gradient projection method is used to search for the corneal edge. The center of the corneal boundary is obtained by circle fitting of the search results from multiple directions. and radius .

[0007] Preferably, in step S1, the geometric constraint cascaded positioning strategy further includes using a unidirectional gradient algorithm for upper eyelid margin positioning and a bidirectional gradient convergence algorithm for lower eyelid margin positioning. For upper eyelid margin localization, the gradient function for downward scanning is defined as: ; Where t is the current vertical coordinate position when locating the upper eyelid margin. This represents the grayscale value at the vertical coordinate t on the scan line. For grayscale gradient, The Heaviside step function is used to filter negative gradients within the search range. Inside, among which , Upper eyelid margin position Corresponding to The maximum value point; For lower eyelid margin localization, the cumulative gradient function is defined by scanning downwards from the tarsal plate: ; Where s is the current vertical coordinate position when locating the lower eyelid margin. Let k be the starting point of the scan, and k be the vertical coordinate variable from the starting point to the current position. For vertical grayscale difference, This represents the grayscale value at the vertical coordinate k. For downward scanning, the distance weighting function is... The attenuation coefficient; Define the cumulative gradient function for scanning upwards from the tear river: ; in The endpoint of the scan. This is a distance-weighted function for upward scanning; Calculate the weighted fusion of cumulative gradients in two directions ,in and The weighting coefficient represents the position of the lower eyelid margin. Corresponding to The maximum value point.

[0008] Preferably, step S1 further includes geometric rationality verification: Set geometric constraints: (1) Palpebral fissure height Is it within the range of 7 to 11 millimeters? (2) The vertical position of the pupil center in the palpebral fissure Does it meet the requirements? ; If any geometric constraint is not met, the geometric constraint cascade positioning strategy is re-executed. in, This refers to the lower eyelid margin. This refers to the position of the upper eyelid margin.

[0009] Preferably, in step S2, the adaptive multi-point measurement strategy includes: Extending to the left and right of the horizontal coordinate of the pupil center N measurement points are set within the range, and for each measurement point, the measurement is taken from the lower eyelid margin. Scan downwards and detect the point where the gray value jumps from a darker area to a brighter area as the lower boundary of the tear trough. The vertical distance from the lower eyelid margin to the lower boundary of the tear trough is the height of the tear trough. Calculate the average height of the Tears River and standard deviation Where N is the total number of measurement points, and j is the sequence number of the measurement point. Let J be the height of the tear river at the j-th measurement point; Eliminate satisfied Outliers.

[0010] Preferably, step S2 further includes calculating the standard coefficient of variation. Used to quantify the uniformity of the height distribution of the Tears River and to identify the highest point of the Tears River. and its location coordinates.

[0011] Preferably, step S4 includes: Extract the palpebral fissure height for each frame of a video sequence. Construct a temporal sequence of palpebral fissure height; Blinking events are identified by detecting local minima in the time series of palpebral fissure height. The criteria for determining a blinking event are as follows: ; in The baseline palpebral fissure height is the median of the time series. The threshold for determining blinking; Count the number of blinks within the duration of the video and calculate the blink frequency. ,in The number of blinks, The video duration is indicated by blink frequency, measured in blinks per minute. For each blink event, calculate the degree of eyelid closure. ,in The minimum palpebral fissure height during blinking, when This is considered incomplete blinking. Calculate the proportion of incomplete blinks ,in The number of incomplete blinks.

[0012] Preferably, the calculation of the corneal exposure area in step S3 includes: The exposed corneal area is defined as the intersection of the corneal boundary and the palpebral fissure, i.e., satisfying... and The set of pixels, where u and v are the horizontal and vertical coordinates of the pixels in the image, is used to obtain the corneal exposure area (CEA) by multiplying the number of pixels in the corneal exposure area by the actual area corresponding to each pixel.

[0013] Preferably, the formula for calculating the tear dynamic balance index in step S4 is: ; TDBI is the tear fluid dynamic balance index; This represents the average height of the River of Tears. BF is the blink frequency factor, where BF is the actual blink frequency. This is a reference value for normal blinking frequency. These are normalization parameters; The incomplete blinking factor (IBR) represents the proportion of incomplete blinks; CEA represents the corneal exposed area. This is a reference value for normal corneal exposure area; The standard coefficient of variation for the height of the Tears River; , , These are the model correction coefficients; Based on the TDBI value, the health status of the ocular surface is divided into four levels: If TDBI ≥ 0.8, it is considered normal; if 0.5 ≤ TDBI < 0.8, it is considered mildly abnormal; if 0.3 ≤ TDBI < 0.5, it is considered moderately abnormal; and if TDBI < 0.3, it is considered severely abnormal.

[0014] In addition, the present invention also provides an ocular surface monitoring system based on image analysis to implement the above-described method, the system comprising: The image acquisition module is used to acquire video sequences of the ocular surface and perform preprocessing. The anatomical structure recognition module uses a geometric constraint cascade localization strategy to identify the pupil, corneal boundary, and eyelid margin. The tear pancreas measurement module performs adaptive multi-point measurements based on the eyelid margin position to obtain spatial distribution information of tear pancreas height; The corneal exposure area calculation module calculates the corneal exposure area based on the corneal boundary and eyelid margin position. The blinking behavior analysis module detects blinking events based on the temporal changes in palpebral fissure height in continuous frame images, and calculates the blinking frequency and the proportion of incomplete blinks. The comprehensive assessment module integrates multiple indicators based on the tear dynamic balance model, calculates the tear dynamic balance index, and conducts a health classification assessment.

[0015] The present invention has the following advantages over the prior art: This invention employs a geometrically constrained cascaded localization strategy to sequentially and accurately identify key anatomical structures such as the pupil, corneal boundary, and eyelid margin. Combined with a bidirectional gradient convergence algorithm, it effectively suppresses eyelash occlusion interference, improving the accuracy and robustness of ocular surface structure localization. An adaptive multi-point measurement strategy is used to acquire spatial distribution information of the tear meniscus height, and the uniformity of tear distribution is quantified using the standard coefficient of variation, providing a more comprehensive reflection of tear reserve status compared to traditional single-point measurement methods. This invention achieves automatic monitoring of blinking behavior through video sequence analysis, accurately identifying blink frequency and the proportion of incomplete blinks. It incorporates dynamic blinking behavior into the ocular surface health assessment system, overcoming the limitations of traditional static image analysis in reflecting the dynamic updating ability of tears. A dynamic tear balance assessment model is constructed, comprehensively considering multi-dimensional parameters such as tear meniscus height, corneal exposure area, blink frequency, and the proportion of incomplete blinks, as well as their interrelationships. A nonlinear assessment formula is used to more accurately reflect the ocular surface tear supply and demand balance, providing objective and reliable comprehensive assessment indicators for ocular surface health in clinical practice. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of radial gradient scanning of the corneal boundary according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the principle of a bidirectional gradient convergence algorithm according to an embodiment of the present invention; Figure 4 This is a schematic diagram of multi-point measurement of the tear river according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the tear river labeling according to an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the calculation of corneal exposure area according to an embodiment of the present invention; Figure 7 This is an ocular surface image of a fully open eye state according to an embodiment of the present invention; Figure 8 This is an ocular surface image of a near-closed eye state according to an embodiment of the present invention; Figure 9 This is an ocular surface image of an incompletely open eye state according to an embodiment of the present invention; Figure 10 This is a schematic diagram of the system framework according to an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1As shown, this invention provides a video analysis-based ocular surface monitoring method, comprising: S1, acquiring an ocular surface video sequence, preprocessing consecutive frame images, using a geometric constraint cascade localization strategy to identify anatomical structures in each frame, sequentially locating the pupil, corneal boundary, and upper and lower eyelid margins, obtaining the location information of key anatomical structures, and calculating the palpebral fissure height; S2, based on the eyelid margin position, using an adaptive multi-point measurement strategy to obtain the spatial distribution information of the tear meniscus height, calculating the average tear meniscus height and the standard coefficient of variation of the tear meniscus height; S3, calculating the corneal exposed area based on the location information of key anatomical structures; S4, based on consecutive frame images of the video sequence, detecting blinking events through temporal changes in the palpebral fissure height, calculating the blinking frequency and the proportion of incomplete blinking; S5, constructing a tear fluid dynamic balance model, substituting the average tear meniscus height, the standard coefficient of variation of the tear meniscus height, the corneal exposed area, the blinking frequency, and the proportion of incomplete blinking into the tear fluid dynamic balance model, calculating the tear fluid dynamic balance index, and classifying and assessing the ocular surface health status based on the index.

[0020] In one embodiment of the present invention, step S1 includes: This invention acquires ocular surface video sequences lasting at least 30 seconds, with a frame rate set to 30 frames per second to ensure the capture of the complete blink cycle. During video acquisition, the distance between the capturing device and the ocular surface is kept stable to avoid scale errors caused by distance changes. First, each frame of the video sequence is preprocessed by converting the original image to grayscale. The grayscale conversion formula is as follows: ; in , , These are the pixel values ​​for the red, green, and blue channels, respectively. Then, Gaussian filtering is used for preprocessing to remove image noise. The Gaussian filter kernel size is [value missing]. Standard deviation ; A cascaded localization strategy, progressing from easy to difficult and from inside to outside, was adopted to sequentially identify key anatomical structures such as the pupil, cornea, and eyelid margin. In the precise pupil localization stage, due to the high contrast and near-circular characteristics of the pupil, the Hough circle detection algorithm is used to detect the pupil across the entire image. In practice, Canny edge detection is first applied to the preprocessed grayscale image to extract the edge image. The low threshold of the Canny algorithm is set to 50, and the high threshold is set to 150. Then, Hough circle transform is performed on the edge image, with the search radius set to 5% to 25% of the minimum side length of the image to accommodate changes in pupil size at different shooting distances. The accumulator threshold of the Hough circle transform is set to 30% of the number of edge points to ensure that the detected circle has enough edge support points. To improve the positioning accuracy, the RANSAC algorithm is used to optimize and fit the detected circular edge points. The number of RANSAC iterations is set to 1000, the in-point determination threshold is set to 2 pixels, and the center and radius of the circle are recalculated after removing outliers. Finally, the coordinates of the pupil center were obtained. and pupil radius , serving as the geometric reference point for locating all subsequent anatomical structures; like Figure 2 As shown, in the corneal boundary constraint search phase, based on prior anatomical knowledge, the horizontal diameter of the human cornea is approximately 11-12 mm, while the normal pupil diameter is approximately 2.5-4 mm. Based on the proportional relationship that the corneal diameter is 3 to 5 times the pupil diameter, the corneal boundary search radius is set to [range missing]. The radial gradient projection method is used to search the corneal edge with the center of the pupil as the center; In practice, starting from the center of the pupil, 36 radial directions are set at equal intervals from 0 degrees to 360 degrees, with each direction spaced 10 degrees apart. Scanning is performed outwards from the center of the pupil along each radial direction, with a scan step size of 1 pixel. The grayscale gradient is calculated in each direction. Where r is the radial distance, The grayscale value at a radial distance r; The corneal boundary corresponds to the peak position of the radial gradient, i.e., satisfies... And the location of the local maximum value, edge detection threshold Set to 1.5 times the mean of the global gradient of the image; perform circle fitting on each edge point detected in the radial direction, and use the least squares method to fit the center of the circle. and radius During the fitting process, outliers that deviate from the fitting circle by more than 3 pixels are removed. The final corneal boundary is obtained by fitting the circle from the multi-directional search results. like Figure 3 As shown, eyelid margin localization is a key challenge in the identification of ocular surface structures because the eyelid margin is relatively blurred and the lower eyelid margin area is often obscured by the root of the eyelashes. Traditional unidirectional edge detection methods are prone to false detection when dealing with eyelash obscuration. For upper eyelid margin localization, a scan line is established vertically upwards from the pupil center. A one-way gradient algorithm is used to search for gradient transition points from bright to dark within a reasonable range. The gradient function for downward scanning is defined as follows: ; Where t is the current vertical coordinate position when locating the upper eyelid margin. This represents the grayscale value at the vertical coordinate t on the scan line. The gray-level gradient is calculated using a first-order difference approximation. , The Heaviside step function is used to filter negative gradients and is defined as follows: when hour, when hour; Search scope Inside, among which , Upper eyelid margin position Corresponding to The maximum value is calculated by traversing all pixels within the search range. The maximum value is selected as the t-coordinate of the upper eyelid margin. For lower eyelid margin localization, a bidirectional gradient convergence algorithm is used, simultaneously projecting gradients from both the tarsal plate and the lacrimal meniscus. Utilizing the convergence characteristics of gradients in both directions corresponding to the actual eyelid margin position, eyelash interference is effectively suppressed. The cumulative gradient function for scanning downwards from the tarsal plate is defined as: ; Where s is the current vertical coordinate position when locating the lower eyelid margin. Let k be the starting point of the scan, and k be the vertical coordinate variable from the starting point to the current position. For vertical grayscale difference, This represents the grayscale value at the vertical coordinate k. For gradient magnitude, For downward scanning, the distance weighting function is... The attenuation coefficient is preferably set to 5 pixels; the calculation process of the cumulative gradient function is as follows: for each candidate lower eyelid margin position s, starting from the scanning start point... Start by calculating the gradient magnitude of each pixel k up to position s. Multiply by the corresponding distance weighting Then sum them up to obtain the cumulative gradient value at that position; Define the cumulative gradient function for scanning upwards from the tear river as: ; in The endpoint of the scan. The distance-weighted function for upward scanning is calculated as follows: for each candidate lower eyelid margin position s, scan downwards from that position to the scanning endpoint. Calculate the gradient magnitude of each pixel k, multiply it by the distance weighting, and then sum them. Calculate the weighted fusion of cumulative gradients in two directions ,in and The weighting coefficient represents the position of the lower eyelid margin. Corresponding to The maximum value point is found by traversing the search range. Calculate the value of each position within the entire range. The value is used to select the s-coordinate corresponding to the maximum value as the position of the lower eyelid margin; Weighting coefficient and The calculation range is adaptively adjusted based on the image quality of the area above the lower eyelid margin. Gradient strength standard deviation within Specifically, the calculation involves calculating the gradient magnitude of all pixels within that region. Find its standard deviation Where M is the total number of pixels in the region. This represents the average gradient magnitude. when Below the threshold When this occurs, it indicates that severe eyelash occlusion leads to weak gradient information, and the setting is... Increase the weight from the direction of the Tears River, when An image quality higher than the threshold indicates good image quality. Balanced weights, threshold Set to 50% of the overall image gradient standard deviation; The core principle of the bidirectional gradient convergence algorithm is that the real eyelid margin position is the only point that simultaneously satisfies the peak value of gradient accumulation in both the downward direction from the tarsal plate and the upward direction from the lacrimal lining. Eyelash occlusion only affects the downward scanning and does not affect the upward scanning, thus effectively eliminating eyelash interference. Subsequently, the geometric rationality of all positioning results was verified, and geometric constraints were set, including: (1) Palpebral fissure height Is it within the range of 7 to 11 millimeters? (2) The vertical position of the pupil center in the palpebral fissure Does it meet the requirements? ; If any geometric constraint is not met, the geometric constraint cascade positioning strategy is re-executed. Calibration is performed using a known anatomical structure with a corneal diameter of approximately 11.5 mm. The pixel-to-physical-size conversion factor is calculated as follows: mm / pixel, converting the pixel value of palpebral fissure height into the actual physical size.

[0021] like Figure 4 As shown, in one embodiment of the present invention, step S2 includes: Based on the lower eyelid margin position obtained in step S1, the lower eyelid lacrimal region is automatically measured, and an adaptive multi-point measurement strategy is used to obtain the spatial distribution information of the lacrimal height. Extending to the left and right of the horizontal coordinate of the pupil center N measurement points are set within the range. In specific implementation, N is preferably set to 5 measurement points, and the horizontal coordinates of the measurement points are as follows: , , , , It corresponds to five anatomical regions: inner canthus, inner middle, central, outer middle, and outer canthus, or seven measurement points are set according to the principle of equal spacing. For each measurement point, start from the lower eyelid margin. Scan downwards to identify the lower boundary of the tear trough. The tear trough region appears as a dark band below the lower eyelid margin in the image, with a distinct grayscale transition between its lower boundary and the cheek skin. The specific detection method is as follows: starting from the lower eyelid margin... Begin scanning downwards line by line, calculating the grayscale value of each line. When the grayscale value increment of three consecutive lines is greater than a threshold... At that time, it was determined to be the lower boundary of the Tears River, threshold. Set to 10 gray levels, detect the point where the gray value jumps from a darker area to a brighter area as the lower boundary of the tear trough, and the vertical distance from the lower eyelid margin to the lower boundary of the tear trough is the height of the tear trough; Noise interference is eliminated through morphological opening operations. The structuring element of the opening operation is: The rectangular kernel is then used to identify connected regions with an area greater than 20 pixels as effective tear river regions through connected component analysis. Small-area interference such as reflective points are filtered out to extract stable tear river boundaries. Statistical analysis was performed on the height values ​​of the Tears River at all measurement points to calculate the average height of the Tears River. and standard deviation Where N is the total number of measurement points, and j is the sequence number of the measurement point. For the height of the tear river at the j-th measurement point, eliminate those that satisfy the following conditions. Outliers are removed, and the mean and standard deviation are recalculated as the final result. Calculate the standard coefficient of variation The coefficient of variation is used to quantify the uniformity of tear river height distribution. It reflects the relative dispersion of tear river height. The larger the value, the more obvious the difference in tear river height at different locations, and the more uneven the tear distribution. Identify the highest point of the River of Tears The location coordinates of the highest point of the tear river are obtained by recording the horizontal coordinates of the corresponding measurement points. This information is clinically significant for understanding the asymmetry of tear distribution. Figure 5 An example ocular surface image with a central tear river annotation is shown. Through measurement and calculation, the height of the central tear river in this example image is 0.19 mm. To ensure the repeatability of measurements during subsequent follow-up examinations, the system automatically records the spatial coordinate offset of all measurement points relative to the pupil center. The offset is calculated as follows: During the patient's follow-up examination, the coordinates of the same measurement point are reconstructed based on the pupil position identified in the current image. The reconstructed coordinates are... ,in The pupil position in the new image. To record the coordinate offset, ensure that the measurements are compared at the same anatomical location.

[0022] like Figure 6 As shown, in one embodiment of the present invention, step S3 includes: The corneal exposed area is calculated based on the location information of key anatomical structures. The corneal exposed area refers to the visible corneal surface area within the palpebral fissure, reflecting the size of the ocular surface area actually affected by tear coverage and evaporation. The exposed corneal area is defined as the intersection of the corneal boundary and the palpebral fissure, i.e., satisfying... and The set of pixels, where u and v are the horizontal and vertical coordinates of the pixels in the image; The specific calculation method is as follows: traverse all pixels (u, v) in the image, and determine for each pixel whether it simultaneously satisfies two conditions, the first condition being... Ensure that the pixel is located within the circular area of ​​the cornea, the second condition. Ensure that each pixel is located within the palpebral fissure between the upper and lower eyelid margins, and count the number of pixels that simultaneously meet both conditions; denoted as . The corneal exposed area is obtained by multiplying the number of pixels in the corneal exposed area by the actual area corresponding to each pixel. ,in This is the pixel-to-millimeter conversion ratio. This is the conversion factor from pixel area to square millimeters; Considering that the eyelid margin curve may be asymmetrical and irregular in reality, the pixel-based statistical method is more accurate than the approximate calculation of the area of ​​a circular arc. The clinical significance of the corneal exposure area lies in the fact that the larger the exposure area, the larger the surface area that the tear film needs to cover, the greater the evaporation loss, and the higher the demand on tear film supply.

[0023] In one embodiment of the present invention, step S4 includes blink behavior analysis, which involves temporal analysis of palpebral fissure height based on consecutive frame images of a video sequence. The palpebral fissure height SLH(t) is extracted from each frame of the video sequence, where t represents the time index, to construct a temporal sequence of palpebral fissure height. This sequence records the dynamic changes in palpebral fissure opening and closing throughout the entire acquisition process. Figure 7The image shown is an ocular surface image with eyes fully open, at which point the palpebral fissure height reaches its maximum value. The palpebral fissure height in this state is used as a baseline reference value for blink event recognition. Blinking events are identified using a local minimum detection method, which determines blinking events by detecting patterns of significant decreases and recoveries in the temporal sequence of palpebral fissure height; the specific criteria are as follows: ,in The baseline palpebral fissure height is estimated using the median of the time series as a robust estimate, effectively avoiding the influence of extreme values. The preferred threshold for determining blinking is 0.3, which indicates that a blinking event is considered to have occurred when the eyelid closure reaches 70% or more. Figure 8 The image shown is of the ocular surface in a fully closed eye state. At this time, the upper and lower eyelid margins are in complete contact, and the palpebral fissure height is close to zero, which meets the criteria for determining a blinking event. Count the number of blinks detected throughout the entire video duration. Calculate blink frequency ,in The video duration is in seconds, and the blink rate is in times per minute. The normal blink rate for adults is about 15 to 20 times per minute, while the blink rate may be significantly reduced in people who use electronic devices for extended periods of time. For each detected blink event, calculate the degree of eyelid closure. ,in This is the minimum palpebral fissure height detected during blinking; a complete blink should close the palpebral fissure completely, i.e. Approaching 0 Approaching 1; when This is judged as incomplete blinking, that is, the palpebral fissure fails to close fully, such as... Figure 8 The image shown is an ocular surface image in the state of incomplete blinking. The upper and lower eyelid margins are not in complete contact, and the palpebral fissure height is significantly greater than in the state of complete eye closure. This incomplete blinking will lead to the inability of the tear film above the cornea to be effectively renewed, which is an important pathogenic factor of dry eye syndrome. Calculate the proportion of incomplete blinks ,in The number of incomplete blinks; this indicator reflects blink quality. Clinical studies have shown that the proportion of incomplete blinks among video terminal workers who stare at screens for long periods can reach over 50%, significantly higher than the 10% to 20% in the normal population. Blinking behavior analysis introduces an assessment dimension of the dynamic tear film renewal capacity, overcoming the limitation of traditional static image analysis in failing to reflect the tear film renewal frequency.

[0024] In one embodiment of the present invention, step S5 includes: A tear dynamic balance model was constructed, and multiple indicators measured in steps S2 to S4 were substituted into the model to calculate the tear dynamic balance index TDBI. The tear dynamic balance index of this invention is established based on the physiological principles of the ocular surface tear system, with blinking playing a crucial role in the dynamic balance of ocular surface tears. Normal blinking achieves tear redistribution and tear film renewal through the complete closure of the upper and lower eyelid margins, while the lacrimal glands are mechanically stimulated during blinking, resulting in reflexive tear secretion. Insufficient blinking frequency leads to prolonged tear evaporation time, increased corneal exposure time, and disruption of tear film stability. Incomplete blinking prevents sufficient tear renewal, resulting in uneven tear film thickness above the cornea and accelerating tear film breakup. Therefore, incorporating blinking frequency and the proportion of incomplete blinking into the evaluation model can more comprehensively reflect the dynamic balance of ocular surface tears. According to the tear mass and solute balance theory, the ocular surface tear system undergoes a dynamic balance process, which can be described by the basic mass balance equation: ; Where V is the tear volume. This represents the tear production rate. The evaporation loss rate, This refers to the nasolacrimal duct drainage rate. The rate of replenishment for the permeable water flow; Based on this theoretical framework and research on the relationship between tear reserve and ocular surface exposure area, the product of the tear meniscus height and the palpebral fissure height approximately represents the spatial scale of tear reserve in a physical sense, while the corneal exposure area represents the ocular surface area that needs to be covered and maintained by tears. Therefore, tear reserve coverage capacity can be expressed as... ; The process of constructing the calculation terms of the TDBI index formula of this invention is as follows: Establish a tear reserve capacity item, based on the fundamental role of tear reserve in protecting ocular surface health, and measure the average height of the tear river. As a core parameter, the tear meniscus height directly reflects the tear reserve at the lower eyelid margin and is the material basis for maintaining corneal moisture. This parameter is obtained through multi-point measurement and can objectively reflect the static tear reserve level. A blink behavior modification term was established, as blink behavior plays a key regulatory role in tear film dynamics; the blink frequency factor was defined as... Where BF is the actual blinking frequency, The normal blinking frequency is taken as 15 times / minute. As a normalization parameter, 30 blinks / minute is used; when the actual blinking frequency equals the normal value... Both excessively high and low blinking frequencies can lead to a decrease in this factor; a low blinking frequency prolongs tear film exposure time and accelerates tear evaporation; an excessively high blinking frequency may reflect ocular surface irritation or discomfort; the incomplete blinking factor is defined as... Where IBR is the proportion of incomplete blinking; when there is no incomplete blinking An increase in the proportion of incomplete blinking leads to a decrease in this factor, reflecting the negative impact of decreased blink quality on tear film renewal; incomplete blinking prevents complete contact between the upper and lower eyelid margins, resulting in ineffective tear film refresh in the area above the cornea; A corneal exposure area correction term is established, employing an exponential penalty function. Where CEA is the actual corneal exposed area. This is a reference value for normal corneal exposure area. This is the sensitivity coefficient; when the corneal exposure area is close to the normal value, this value is close to 1, and when the corneal exposure area deviates from the normal value, this value decreases rapidly; an excessively large corneal exposure area indicates an abnormally large eyelid opening and closing amplitude, an increased area that tears need to cover, and a reduced tear distribution per unit area; an excessively small corneal exposure area may indicate eyelid stenosis or abnormal eyelid position. Establish a distribution uniformity correction term, and define the penalty term for tear distribution non-uniformity as follows: ,when hour ,when When it increases The physiological basis for this is that even if the average tear river height is normal, but the distribution is extremely uneven, there may be insufficient tear fluid in some areas, which can still lead to ocular surface discomfort symptoms. Finally, the tear fluid dynamic balance index is established by combining the above three factors as follows: ; TDBI stands for Tear Dynamic Balance Index. The average height of the Tears River Blink frequency factor These two factors, representing incomplete blinking factors, incorporate the dynamic characteristics of blinking behavior into the assessment system, enabling the TDBI index to comprehensively reflect the static reserve capacity and dynamic renewal capacity of tear film. CEA represents the corneal exposed area. The reference value for normal corneal exposure area is set at 90 square millimeters. This value is based on statistical data of normal adults with a corneal diameter of 11.5 mm and an palpebral fissure height of 9 mm. The standard coefficient of variation for the height of the Tears River. , , Here are the model correction coefficients, where The dimensional normalization coefficients were used to make the TDBI value close to 1 under normal ocular surface conditions through regression analysis of clinical data. To ensure that the shape parameters of the exposure area deviation can reasonably reflect the degree of impact of the exposure area deviation on ocular surface health. The weighting parameter for uniformity of tear distribution can effectively distinguish different degrees of tear distribution abnormalities, and the preferred value range is [value range missing]. , , The specific value is determined through regression analysis of clinical data, and the recommended value is... , , ; This formula reflects the complex characteristics of the physiological system through a nonlinear function of multiplication and combination, forming a comprehensive quantitative assessment of the dynamic balance of tear film on the ocular surface. Compared with traditional single-index assessment, it has advantages such as multi-dimensional integration, nonlinear relationship processing, individualized assessment, and quantitative accuracy. The health status of the ocular surface is divided into four levels based on the TDBI value: If A normal ocular surface is defined as having a tear meniscus height greater than 0.25 mm and a uniform distribution, indicating a good balance between tear supply and demand. The diagnosis is mild abnormality, corresponding to mild tear reduction or mild uneven distribution, indicating minor problems with tear reserve or distribution. It is recommended to monitor the situation and have regular follow-up examinations. A moderate abnormality is diagnosed, corresponding to significant tear reserve insufficiency or distribution abnormalities, indicating a marked imbalance between tear supply and demand, and potential dry eye symptoms. Clinical intervention is recommended. The condition is classified as severe abnormal, corresponding to a severe dry eye condition, with severely insufficient or uneven tear production, requiring timely treatment. The system generates a visual evaluation report. The report generation process includes marking the detected pupil circular boundary (red circle), corneal boundary (green circle), upper and lower eyelid margin positions (blue line), and tear meniscus measurement point positions (yellow markers) on representative frame images, and displaying the values ​​of various measurement parameters, including... , , , The system measures SLH, CEA, BF, and IBR, and plots spatial distribution curves of tear meniscus height and temporal variation curves of palpebral fissure height; as well as the final TDBI comprehensive score and health level determination, to facilitate clinicians' rapid understanding of the patient's ocular surface condition.

[0025] This invention transforms the assessment process from static to dynamic by introducing video sequence analysis and blink behavior monitoring. Traditional methods measure static parameters such as tear height based on single-frame images, failing to reflect the dynamic updating process of tear fluid. This invention captures blink behavior characteristics through continuous frame analysis, using blink frequency and the proportion of incomplete blinks as important indicators of tear fluid dynamic balance. This allows for the identification of ocular surface problems caused by abnormal blinking, which are particularly common among video-based workers. This method provides a more accurate basis for the etiological diagnosis of dry eye syndrome and helps in developing targeted treatment plans.

[0026] like Figure 10As shown, the present invention also provides an ocular surface monitoring system based on video analysis for implementing the above method. This system includes the following modules: (a) Image acquisition module The image acquisition module is used to acquire video sequences of the patient's ocular surface. This module includes a video acquisition device and a preprocessing unit. The video acquisition device is preferably a digital ocular surface imaging device equipped with a front illumination system, capable of clearly capturing the pupil, cornea, eyelid margin, and lacrimal meniscus area of ​​the ocular surface, supporting continuous video acquisition at a rate of at least 30 frames per second to ensure the capture of the complete blink cycle. The preprocessing unit performs grayscale conversion and Gaussian filtering noise reduction on each frame of the video sequence, converting color images to grayscale and eliminating image noise, providing high-quality input data for subsequent anatomical structure recognition and blink behavior analysis. (ii) Anatomical structure recognition module The anatomical structure recognition module employs a geometric constraint cascaded localization strategy to sequentially identify key anatomical structures in the ocular surface image. This module includes a pupil localization unit, a corneal boundary localization unit, an eyelid margin localization unit, and a geometric verification unit. The pupil localization unit uses the Hough circle detection algorithm and the RANSAC optimized fitting algorithm to accurately locate the center coordinates and radius of the pupil, serving as the geometric reference point for subsequent localization. The corneal boundary localization unit, based on the anatomical proportions of the cornea and pupil, sets a constrained search range and uses the radial gradient projection method to search the corneal edge. Through multi-directional scanning and circle fitting, it obtains the center and radius of the corneal boundary. The eyelid margin localization unit uses a bidirectional gradient convergence algorithm. For the upper eyelid margin, it uses unidirectional gradient detection; for the lower eyelid margin, it simultaneously performs gradient scanning from both the tarsal plate and the lacrimal meniscus. By weighted fusion of the gradient information from both directions, it determines the eyelid margin position, effectively overcoming eyelash obstruction interference. The geometric verification unit verifies the anatomical rationality of the localization results, including verifying the palpebral fissure height range and pupil position proportion, ensuring the accuracy of the recognition results. This module processes each frame of the video sequence to obtain continuous temporal data on palpebral fissure height, providing basic data support for blink behavior analysis; (III) Tears River Measurement Module The tear trough measurement module performs adaptive multi-point measurements of the lower eyelid tear trough region based on eyelid margin localization results. This module includes a measurement point setting unit, a tear trough boundary recognition unit, and a statistical analysis unit. The measurement point setting unit sets multiple measurement points within a range extending horizontally to the left and right of the pupil center, preferably five to seven points to cover the inner canthus, central canthus, and outer canthus regions. The tear trough boundary recognition unit scans downwards from the lower eyelid margin at each measurement point, detects grayscale transition points between the tear trough region and the cheek skin, extracts the lower boundary of the tear trough, and calculates the vertical distance from the lower eyelid margin to the lower boundary of the tear trough as the tear trough height. The statistical analysis unit performs statistical analysis on the tear trough height of all measurement points, calculates the mean, standard deviation, and standard coefficient of variation, removes outliers, identifies the highest point of the tear trough and its location, and quantifies the spatial distribution characteristics of the tear trough height. (iv) Corneal Exposure Area Calculation Module The corneal exposure area calculation module calculates the visible corneal surface area within the palpebral fissure based on the corneal boundary and eyelid margin positions obtained by the anatomical structure recognition module. This module defines the corneal exposure area as the intersection of the circular corneal boundary and the palpebral fissure. By traversing image pixels, it counts the number of pixels that simultaneously meet the criteria of being located within the circular corneal region and between the upper and lower eyelid margins. Multiplying this number by the actual area corresponding to each pixel yields the physical size of the corneal exposure area. The corneal exposure area reflects the actual size of the ocular surface affected by tear film coverage and evaporation.

[0027] (v) Blinking Behavior Analysis Module The blinking behavior analysis module performs temporal analysis of palpebral fissure height based on continuous frame images of a video sequence. This module includes a temporal extraction unit, a blink detection unit, and a blink quality assessment unit. The temporal extraction unit extracts palpebral fissure height data from each frame of the video sequence, constructing a complete temporal sequence of palpebral fissure height that records the dynamic changes in palpebral fissure opening and closing throughout the acquisition process. The blink detection unit identifies blinking events using a local minimum detection algorithm. A blinking event is determined when the decrease in palpebral fissure height relative to the baseline value exceeds a set threshold. The median of the temporal sequence is used as a robust estimate for the baseline palpebral fissure height. The number of blinks throughout the entire video duration is counted, and the blink frequency is calculated. The blink quality assessment unit calculates the degree of palpebral fissure closure for each blink event. Incomplete blinking events are identified by comparing the minimum palpebral fissure height during the blinking process with the baseline palpebral fissure height. The proportion of incomplete blinks is calculated, quantifying blink quality. This module provides blinking behavior parameters for tear film dynamic balance assessment and is a key functional unit for achieving dynamic monitoring. (vi) Comprehensive Evaluation Module The comprehensive assessment module, based on a tear fluid dynamic balance model, provides a comprehensive quantitative assessment of ocular surface health. This module includes an indicator integration unit, a TDBI calculation unit, a health grading unit, and a report generation unit. The indicator integration unit receives multiple measurement parameters from the tear meniscus measurement module, corneal exposure area calculation module, and blink behavior analysis module, including average tear meniscus height, corneal exposure area, standard coefficient of variation of tear meniscus height, blink frequency, and incomplete blink rate. The TDBI calculation unit substitutes these parameters into the tear fluid dynamic balance index calculation formula, which comprehensively considers four aspects: tear fluid reserve capacity, blink behavior correction, exposure area deviation, and distribution uniformity, and calculates the TDBI comprehensive score through a nonlinear function. The health grading unit classifies ocular surface health into four levels—normal, mildly abnormal, moderately abnormal, and severely abnormal—based on the TDBI value, providing a quantitative basis for clinical diagnosis. The report generation unit generates a visual assessment report, annotating each measurement area and key structure on the original ocular surface image, displaying the values ​​of various measurement parameters, plotting the spatial distribution curve of the tear river height and the temporal change curve of the palpebral fissure height, and presenting the TDBI comprehensive score and health level determination results. In a graphic and textual format, it helps clinicians quickly understand the patient's ocular surface health status.

[0028] (vi) Data storage and management module The data storage and management module stores patients' ocular surface videos, measurement data, and assessment reports, establishing a patient profile database. This module records all measurement parameters, blink behavior parameters, TDBI scores, and health levels for each examination, supporting longitudinal monitoring and historical data comparison and analysis. Simultaneously, the module records the spatial coordinate offset of each measurement point relative to the pupil center, ensuring measurements are taken at the same anatomical location during follow-up examinations, improving the comparability and repeatability of follow-up results. The data storage and management module also supports data export functionality, facilitating clinical research and remote consultations.

[0029] The ocular surface monitoring system of this invention, through modular design, realizes a complete automated process from video acquisition, automatic identification of anatomical structures, accurate measurement of multiple indicators, dynamic analysis of blinking behavior to comprehensive health assessment, providing a standardized and intelligent technical solution for ocular surface monitoring, and improving examination efficiency and diagnostic accuracy.

[0030] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A video analysis-based method for ocular surface monitoring, characterized in that, include: S1. Acquire video sequences of the ocular surface, preprocess consecutive frame images, and use a geometric constraint cascade localization strategy to identify anatomical structures in each frame image, sequentially locating the pupil, corneal boundary, and upper and lower eyelid margins to obtain the location information of key anatomical structures and calculate the palpebral fissure height. S2. Based on the eyelid margin position, an adaptive multi-point measurement strategy is used to obtain the spatial distribution information of the tear river height, and the average tear river height and the standard coefficient of variation of the tear river height are calculated. S3. Calculate the corneal exposure area based on the location information of key anatomical structures; S4. Based on continuous frame images of video sequences, blinking events are detected by temporal changes in palpebral fissure height, and blinking frequency and incomplete blinking ratio are calculated; S5. Construct a tear dynamic balance model, and substitute the average tear river height, standard coefficient of variation of tear river height, corneal exposed area, blink frequency, and proportion of incomplete blinking into the tear dynamic balance model to calculate the tear dynamic balance index, and classify and assess the ocular surface health status based on the index.

2. The ocular surface monitoring method based on video analysis according to claim 1, characterized in that, In step S1, the geometric constraint cascaded positioning strategy includes: The Hough circle detection algorithm is used to accurately locate the pupil and obtain the coordinates of the pupil center. and pupil radius The RANSAC algorithm was used to optimize and fit the detected circular edge points. Based on the proportional relationship that the corneal diameter is 3 to 5 times the pupil diameter, the search radius of the corneal boundary is set to be... Using the pupil center as the center, the radial gradient projection method is used to search for the corneal edge. The center of the corneal boundary is obtained by circle fitting of the search results from multiple directions. and radius .

3. The ocular surface monitoring method based on video analysis according to claim 2, characterized in that, In step S1, the geometric constraint cascaded localization strategy also includes using a unidirectional gradient algorithm for upper eyelid margin localization and a bidirectional gradient convergence algorithm for lower eyelid margin localization. For upper eyelid margin localization, the gradient function for downward scanning is defined as: ； Where t is the current vertical coordinate position when locating the upper eyelid margin. This represents the grayscale value at the vertical coordinate t on the scan line. For grayscale gradient, The Heaviside step function is used to filter negative gradients within the search range. Inside, among which , Upper eyelid margin position Corresponding to The maximum value point; For lower eyelid margin localization, the cumulative gradient function is defined by scanning downwards from the tarsal plate: ； Where s is the current vertical coordinate position when locating the lower eyelid margin. Let k be the starting point of the scan, and k be the vertical coordinate variable from the starting point to the current position. For vertical grayscale difference, This represents the grayscale value at the vertical coordinate k. For downward scanning, the distance weighting function is... The attenuation coefficient; Define the cumulative gradient function for scanning upwards from the tear river: ； in The endpoint of the scan. This is a distance-weighted function for upward scanning; Calculate the weighted fusion of cumulative gradients in two directions ,in and The weighting coefficient represents the position of the lower eyelid margin. Corresponding to The maximum value point.

4. The ocular surface monitoring method based on video analysis according to claim 2, characterized in that, Step S1 also includes geometric rationality verification: Set geometric constraints: (1) Palpebral fissure height Is it within the range of 7 to 11 millimeters? (2) The vertical position of the pupil center in the palpebral fissure Does it meet the requirements? ; If any geometric constraint is not met, the geometric constraint cascade positioning strategy is re-executed. in, This refers to the lower eyelid margin. This refers to the position of the upper eyelid margin.

5. The ocular surface monitoring method based on video analysis according to claim 2, characterized in that, In step S2, the adaptive multi-point measurement strategy includes: Extending to the left and right of the horizontal coordinate of the pupil center N measurement points are set within the range, and for each measurement point, the measurement is taken from the lower eyelid margin. Scan downwards and detect the point where the gray value jumps from a darker area to a brighter area as the lower boundary of the tear trough. The vertical distance from the lower eyelid margin to the lower boundary of the tear trough is the height of the tear trough. Calculate the average height of the Tears River and standard deviation Where N is the total number of measurement points, and j is the sequence number of the measurement point. Let the height of the tear river be the height at the j-th measurement point; Eliminate satisfied Outliers.

6. The ocular surface monitoring method based on video analysis according to claim 5, characterized in that, Step S2 also includes calculating the standard coefficient of variation. Used to quantify the uniformity of the height distribution of the Tears River and to identify the highest point of the Tears River. and its location coordinates.

7. The ocular surface monitoring method based on video analysis according to claim 1, characterized in that, Step S4 includes: Extract the palpebral fissure height for each frame of a video sequence. Construct a temporal sequence of palpebral fissure height; Blinking events are identified by detecting local minima in the time series of palpebral fissure height. The criteria for determining a blinking event are as follows: ； in The baseline palpebral fissure height is the median of the time series. The threshold for determining blinking; Count the number of blinks within the duration of the video and calculate the blink frequency. ,in The number of blinks, The video duration is indicated by blink frequency, measured in blinks per minute. For each blink event, calculate the degree of eyelid closure. ,in The minimum palpebral fissure height during blinking, when This is considered incomplete blinking. Calculate the proportion of incomplete blinks ,in The number of incomplete blinks.

8. The ocular surface monitoring method based on video analysis according to claim 3, characterized in that, The calculation of the corneal exposure area in step S3 includes: The exposed corneal area is defined as the intersection of the corneal boundary and the palpebral fissure, i.e., satisfying... and The set of pixels, where u and v are the horizontal and vertical coordinates of the pixels in the image, is used to obtain the corneal exposure area (CEA) by multiplying the number of pixels in the corneal exposure area by the actual area corresponding to each pixel.

9. The ocular surface monitoring method based on video analysis according to claim 1, characterized in that, The formula for calculating the tear fluid dynamic balance index in step S4 is as follows: ； TDBI is the tear fluid dynamic balance index; This represents the average height of the River of Tears. BF is the blink frequency factor, where BF is the actual blink frequency. This is a reference value for normal blinking frequency. These are normalization parameters; The incomplete blinking factor (IBR) represents the proportion of incomplete blinks; CEA represents the corneal exposed area. This is a reference value for normal corneal exposure area; The standard coefficient of variation for the height of the Tears River; , , These are the model correction coefficients; Based on the TDBI value, the health status of the ocular surface is divided into four levels: If TDBI ≥ 0.8, it is considered normal; if 0.5 ≤ TDBI < 0.8, it is considered mildly abnormal; if 0.3 ≤ TDBI < 0.5, it is considered moderately abnormal; and if TDBI < 0.3, it is considered severely abnormal.

10. An ocular surface monitoring system based on video analysis, characterized in that, The system is used to implement the method as described in any one of claims 1-9, the system comprising: The image acquisition module is used to acquire video sequences of the ocular surface and perform preprocessing. The anatomical structure recognition module uses a geometric constraint cascade localization strategy to identify the pupil, corneal boundary, and eyelid margin. The tear pancreas measurement module performs adaptive multi-point measurements based on the eyelid margin position to obtain spatial distribution information of tear pancreas height; The corneal exposure area calculation module calculates the corneal exposure area based on the corneal boundary and eyelid margin position. The blinking behavior analysis module detects blinking events based on the temporal changes in palpebral fissure height in continuous frame images, and calculates the blinking frequency and the proportion of incomplete blinks. The comprehensive assessment module integrates multiple indicators based on the tear dynamic balance model, calculates the tear dynamic balance index, and conducts a health classification assessment.

Images