Strabismus measurement system

Abstract

A strabismus measurement system. The method mainly solves the problem that an existing strabismus measurement means is difficult to be applied to strabismus screening work in schools, communities and remote areas due to the fact that a method of increasing measurement equipment and procedures is often adopted for improving accuracy. The system is characterized by comprising an image acquisition unit (1) which comprises a camera and a near-infrared light source and is used for continuously shooting images of an eye area; the cover plate unit (2) is driven by the linear driver (3) to move back and forth and is used for shielding eyeballs to different degrees; the data processing unit is used for processing (1) the acquired image data, and accurately positioning pupils, cornea reflecting spots and reference objects; the data processing unit comprises pupil size calculation based on a target distance measurement algorithm, strabismus disease capture based on an eye position deviation characteristic value, obstacle detection and strabismus degree calculation, the system puts the measurement process into the algorithm as much as possible for automatic processing, simplification is achieved, deviation caused by head shaking during shooting is avoided, and the shooting accuracy is improved. The robustness and the accuracy of the system are improved; and the result is consistent with synoptophore inspection.

Description

Technical Field

[0001] The invention relates to the field of vision detection, and in particular to a strabismus measurement system. Background Art

[0002] Strabismus is a common ophthalmic disease. The cover uncover test and alternate cover test are the commonly used methods for measuring strabismus in ophthalmology clinics.

[0003] The existing intelligent strabismus detection methods fail to obtain the degree of strabismus, the video processing scheme for diagnosing strabismus does not take into account the adverse effects of head shaking on the results, the equipment used is not economical and is not suitable for use in schools, communities and remote areas, and the pupil size is estimated to participate in the calculation of the strabismus degree. Summary of the invention

[0004] In order to overcome the shortcomings of the background technology, the present invention provides a strabismus measurement system, which mainly solves the problem that the existing strabismus measurement means are not suitable for application in schools, communities and remote areas.

[0005] The technical solution of the present invention is:

[0006] Strabismus measurement system, including

[0007] An image acquisition unit, including a camera and a near-infrared light source, for continuously capturing images of an eye region;

[0008] The cover unit is driven by a linear actuator to move forward and backward to block the eyeball to different degrees;

[0009] The data processing unit is used to statistically analyze the data obtained by the data processing unit to determine the current video eye state, automatically select the video clip used to calculate the strabismus degree, that is, the eye movement moment after covering; and use the reference object to convert the pixel distance into the actual distance, eliminate the obstacles that may cause calculation errors such as blinking and incomplete detection, and calculate the strabismus degree in real time.

[0010] The data processing unit processes the collected image data through the following process;

[0011] S1. Recruit strabismus patients and normal subjects to create a strabismus-specific dataset. The subjects wear red round standard stickers and use the camera and near-infrared light source in the image acquisition unit (1) to take 350 photos. The pupils, corneal light reflection points and reference objects are manually annotated and then scaled to a uniform size of 128×128.

[0012] S2, strabismus-specific dataset is divided into training set, validation set and test set in the ratio of 70%, 20% and 10%. After data enhancement, the training data is trained using the UNet neural network model. The image is segmented according to the trained weights. The tracking targets include pupil, corneal light spot and red circular sticker reference. The geometric center of the three is calculated as the pupil position P. x , P y Light reflection point position L x , L y , reference position P x , P y .

[0013] The data processing unit also includes calculation of pupil size based on target ranging algorithm, capture of strabismus based on eye position offset characteristic value, obstacle detection and calculation of strabismus degree, which is used to convert pixel distance into actual distance, judge the eye state of the current video and select segments that can show strabismus, exclude blinking, incomplete obstacle detection, and calculate strabismus degree in real time.

[0014] The data processing unit calculates pupil size based on the target distance measurement algorithm, captures strabismus based on eye position deviation feature values, detects obstacles and calculates strabismus degree through the following process to process data:

[0015] S1: Cut the subject’s cover test video into images frame by frame, and use the UNet neural network model to identify the pupil and reference objects. Count the segmented images frame by frame, and set the number of pupil pixels as the pupil area P. R , the number of reference pixel points is the reference area R e R, P R If it is 0, it means that no pupil is found in the video. Each frame of the video has a status value. R If it is 0, it is set to state 1; P R If it is not 0, it means that the pupil is detected in this frame of the video, and the eye state of this frame is not covered by the masker, so it is set to state 2; because the red sticker is worn on the skin near the right eye according to the experimental requirements, the reference area Re in a single-eye video R If it is 0, it is set to the left eye; if it is not 0, it is set to the right eye.

[0016] S2, based on the pupil area, the mean value of the pupil area is "μ", the standard deviation is "σ", and the pupil area of ​​a certain frame is "x", calculate the z score, Z = (x-μ) / σ, and filter out abnormal data with a Z score result greater than a threshold of 2. Abnormal data is mostly caused by incomplete pupil detection in a certain frame of the video due to blinking or the cover plate not being completely removed from the eye. Therefore, the eye state of the frame where the abnormal data is located is changed from state 1 where the pupil is detected to fault state 3;

[0017] S3, calculate the difference between the pupil position in a certain frame of the video in state 1 and the pupil position in the next 5 frames and set it as the characteristic value B, and iterate each frame of the video, select the frame with the largest absolute value of B as the eye movement moment. It is observed that the eye movement caused by occlusion generally lasts for 5 frames in the video, so the eye movement moment delayed by 5 frames is set as state 2 - eye movement state.

[0018] S4. The diameter of the red circular sticker in the statistical image is A pixels. The actual distance corresponding to each pixel in the video is 8÷A, in millimeters. The difference between the pupil position and the corneal reflection point position is calculated in real time. According to the corneal reflection point method, the pupil distance per millimeter to the corneal reflection point is calculated as the strabismus degree G3. The strabismus degree of the subject is calculated during the motor cover process (P r -L x )×H3×8+A(horizontal squint), (P r -L r )×H3×8+A(vertical squint), the unit is degree.

[0019] The linear drive comprises a motor and a driven wheel, wherein the motor and the driven wheel are connected via a belt, a mounting plate is mounted on the belt, and the cover plate is mounted on the mounting plate.

[0020] The beneficial effects of the present invention are as follows: the present invention provides a strabismus measurement system, which converts pixel distance into strabismus degree without adding pupil measurement equipment; simultaneously tracks the pupil and corneal light reflection points, and calculates the real-time relative distance, thereby reducing the influence of head shaking on the results and improving the robustness of the system; and the results are consistent with those of the synoptophore examination. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram for identifying corneal light reflection points.

[0022] Figure 2 Schematic diagram of reference objects for identifying stickers.

[0023] Figure 3 Synoptophore examination and Bland-Altman diagram (horizontal strabismus) of the strabismus automatic detection system were performed for the strabismus subjects.

[0024] Figure 4 Synoptophore examination and Bland-Altman diagram (vertical strabismus) of the strabismus automatic detection system were performed for the strabismus subjects.

[0025] Figure 5 It is a logic block diagram of the present invention.

[0026] Figure 6 A schematic diagram of this system.

[0027] Figure 7 Schematic diagram of a linear drive.

[0028] Figure 8 Schematic diagram of eye deviation characteristics. DETAILED DESCRIPTION

[0029] The present invention is further described below in conjunction with the accompanying drawings: as shown in the figure, 1. a strabismus measurement system includes an image acquisition unit 1, including a camera and a near-infrared light source, for continuously capturing images of the eye area; a cover unit 2, driven by a linear driver 3 to move forward and backward, for blocking the eyeball to different degrees; a data processing unit, for statistically analyzing the data obtained by the data processing unit to determine the current video eye state, and automatically selecting a video clip for calculating the strabismus degree, that is, the eye movement moment occurring after covering; and converting the pixel distance into the actual distance with the help of a reference object, and calculating the strabismus degree in real time after eliminating obstacles such as blinking and incomplete detection that may cause calculation errors.

[0030] The data processing unit processes the collected image data through the following process;

[0031] S1. Recruit strabismus patients and normal subjects to create a strabismus-specific dataset. The subjects wear red round standard stickers and use the camera and near-infrared light source in the image acquisition unit 1 to take 350 photos. The pupils, corneal light reflection points and reference objects are manually annotated and then scaled to a uniform size of 128×128.

[0032] S2, strabismus-specific dataset is divided into training set, validation set and test set in the ratio of 70%, 20% and 10%. After data enhancement, the training data is trained using the UNet neural network model. The image is segmented according to the trained weights. The tracking targets include pupil, corneal light spot and red circular sticker reference. The geometric center of the three is calculated as the pupil position P. x , P y Light reflection point position L x , L y , reference position P x , P y .

[0033] The data processing unit also includes calculation of pupil size based on target ranging algorithm, capture of strabismus based on eye position offset characteristic value, obstacle detection and calculation of strabismus degree, which is used to convert pixel distance into actual distance, judge the eye state of the current video and select segments that can show strabismus, exclude blinking, incomplete obstacle detection, and calculate strabismus degree in real time.

[0034] The data processing unit calculates pupil size based on the target distance measurement algorithm, captures strabismus based on eye position deviation feature values, detects obstacles and calculates strabismus degree through the following process to process data:

[0035] S1: Cut the subject’s cover test video into images frame by frame, and use the UNet neural network model to identify the pupil and reference objects. Count the segmented images frame by frame, and set the number of pupil pixels as the pupil area P. R , the number of reference object pixels is the reference object area Re R , P R If it is 0, it means that no pupil is found in the video. Each frame of the video has a status value. R If it is 0, it is set to state 1; P R If it is not 0, it means that the pupil is detected in this frame of the video, and the eye state of this frame is not covered by the masker, so it is set to state 2; because the red sticker is worn on the skin near the right eye according to the experimental requirements, the reference area Re in a single-eye video R If it is 0, it is set to the left eye; if it is not 0, it is set to the right eye.

[0036] S2, based on the pupil area, the mean value of the pupil area is "μ", the standard deviation is "σ", and the pupil area of ​​a certain frame is "x", calculate the z score, Z = (x-μ) / σ, and filter out abnormal data with a Z score result greater than a threshold of 2. Abnormal data is mostly caused by incomplete pupil detection in a certain frame of the video due to blinking or the cover plate not being completely removed from the eye. Therefore, the eye state of the frame where the abnormal data is located is changed from state 1 where the pupil is detected to fault state 3;

[0037] S3, calculate the difference between the pupil position in a certain frame of the video in state 1 and the pupil position in the next 5 frames and set it as the characteristic value B, and iterate each frame of the video, select the frame with the largest absolute value of B as the eye movement moment. It is observed that the eye movement caused by occlusion generally lasts for 5 frames in the video, so the eye movement moment delayed by 5 frames is set as state 2 - eye movement state.

[0038] S4. The diameter of the red circular sticker in the statistical image is A pixels. The actual distance corresponding to each pixel in the video is 8÷A, in millimeters. The difference between the pupil position and the corneal reflection point position is calculated in real time. According to the corneal reflection point method, the pupil distance per millimeter to the corneal reflection point is calculated as the strabismus degree H3. The strabismus degree of the subject is calculated during the motor cover process (P r -L r )×H3×8+A horizontal strabismus, (P y -L z )×H3×B+A(vertical squint), the unit is degree.

[0039] The linear drive comprises a motor 31 and a driven wheel 33 , wherein the motor and the driven wheel are connected via a belt 32 , a mounting plate 34 is mounted on the belt, and the cover plate is mounted on the mounting plate.

[0040] A more economical control module is used to drive the motor. After the motor is powered on, it is installed on the slide rail and drives the rubber synchronous belt through gears. The synchronous belt is fixed with a plastic cover plate. The control device sends a periodic pulse signal from the STC89C51 microcontroller to the ULN2003 stepper motor driver. According to different periodic signals, four buttons are used to issue instructions to the motor device:

[0041] Button 1: Move left first, pause for 5 seconds, then move right;

[0042] Button 2: Move right first, pause for 5 seconds, then move left;

[0043] Button three: move left;

[0044] Button 4: Move right;

[0045] Average covering time of the stepper motor driven cover (adult pupil distance 6 cm, cover plate moves from left eye to right eye): 0.25 seconds; Near infrared light:

[0046] BDQ-C8080KIRJK-1W; power: 1W; luminous angle: 60°; center wavelength: 850NM infrared camera RER-USBFHD01M; clarity: 2 million; fluency: 120FPS.

[0047] Reference Figure 1 , the average accuracy of pupil recognition for 10 strabismus subjects was 92%, the average recall was 87%, the average F1 score was 89%, and the average IOU was 81%;

[0048] Reference Figure 2 , the average accuracy of pupil recognition for 10 strabismus subjects was 87%, the average recall was 92%, the average F1 score was 89%, and the average IOU was 81%;

[0049] Strabismus capture based on eye position deviation feature value

[0050] The test video was automatically judged and manually labeled based on the feature value of the eye state, with the feature value on the ordinate, the frame on the abscissa, and the cover color on the manual label. In the test video, the automatic judgment and manual labeling of the uncovered state, covered state, and eye movement state were basically consistent.

[0051] Strabismus test results,

[0052] Ten strabismus subjects were examined with a synoptophore and the strabismus automatic detection system. The following is the Bland-Altman diagram (see Figure 3 , Figure 4), which compares two quantitative measurements by plotting the difference between the two measurements versus their mean. This visualization technique is particularly important in the field of medical research because it provides a clear and intuitive way to evaluate the agreement or disagreement between two measurement methods or instruments. The results of the present study and the synoptophore examination of 10 patients with strabismus showed that 90% of the data for horizontal strabismus were within the 95% confidence interval, and all the data for vertical strabismus were within the 95% confidence interval, indicating good agreement between the two methods.

[0053] The embodiments described with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention. The embodiments should not be regarded as limiting the present invention, but any improvements made based on the spirit of the present invention should be within the scope of protection of the present invention.

Claims

1. A squint measurement system, characterized in that: include An image acquisition unit (1), comprising a camera and a near-infrared light source, for continuously capturing images of the eye region; The cover unit (2) is driven by a linear drive (3) to move forward and backward, and is used to cover the eyeball to different degrees; The data processing unit is used to statistically analyze the data obtained by the data processing unit to determine the current video eye state, automatically select the video clip used to calculate the strabismus degree; and convert the pixel distance into the actual distance with the help of a reference object to calculate the strabismus degree in real time.

2. The strabismus measurement system according to claim 1, characterized in that: The data processing unit processes the collected image data through the following process; S1. Recruit strabismus patients and normal subjects to create a strabismus-specific dataset. The subjects wear red round standard stickers and use the camera and near-infrared light source in the image acquisition unit (1) to take 350 photos. The pupils, corneal light reflection points and reference objects are manually annotated and then scaled to a uniform size of 128×128. S2, strabismus-specific dataset is divided into training set, validation set and test set in the ratio of 70%, 20% and 10%. After data enhancement, the training data is trained using the UNet neural network model. The image is segmented according to the trained weights, and the tracking targets include pupil, corneal light spot and red circular sticker reference. The geometric center of the three is calculated as the pupil position P x , P y Light reflection point position L x , L y , reference position P x , P y .

3. The strabismus measurement system according to claim 1, characterized in that: The data processing unit also includes calculation of pupil size based on target ranging algorithm, capture of strabismus based on eye position offset characteristic value, obstacle detection and calculation of strabismus degree, which is used to convert pixel distance into actual distance, judge the eye state of the current video and select segments that can show strabismus, exclude blinking, incomplete obstacle detection, and calculate strabismus degree in real time.

4. The strabismus measurement system according to claim 3, characterized in that: The data processing unit calculates pupil size based on the target distance measurement algorithm, captures strabismus based on eye position deviation feature values, detects obstacles and calculates strabismus degree through the following process to process data: S1: Cut the subject’s cover test video into images frame by frame, and use the UNet neural network model to identify the pupil and reference objects. Count the segmented images frame by frame, and set the number of pupil pixels as the pupil area P. R , the number of reference object pixels is the reference object area Re R , P R If it is 0, it means that no pupil is found in the video. Each frame of the video has a status value. R If it is 0, it is set to state 1; P R If it is not 0, it means that the pupil is detected in this frame of the video, and the eye state of this frame is not covered by the masker, so it is set to state 2; because the red sticker is worn on the skin near the right eye according to the experimental requirements, the reference area Re in a single-eye video R If it is 0, it is set to the left eye; if it is not 0, it is set to the right eye. S2, based on the pupil area, the mean value of the pupil area is "μ", the standard deviation is "σ", and the pupil area of ​​a certain frame is "x", calculate the z score, Z = (xF) / σ, and filter out abnormal data with a Z score result greater than a threshold of 2. Abnormal data is mostly caused by incomplete pupil detection in a certain frame of the video due to blinking or the cover plate not being completely removed from the eye. Therefore, the eye state of the frame where the abnormal data is located is changed from state 1 where the pupil is detected to fault state 3; S3, calculate the difference between the pupil position in a certain frame of the video in state 1 and the pupil position in the next 5 frames and set it as the characteristic value B, and iterate each frame of the video, select the frame with the largest absolute value of B as the eye movement moment. It is observed that the eye movement caused by occlusion generally lasts for 5 frames in the video, so the eye movement moment delayed by 5 frames is set as state 2 - eye movement state. S4. The diameter of the red circular sticker in the statistical image is A pixels. The actual distance corresponding to each pixel in the video is 8÷A, in millimeters. The difference between the pupil position and the corneal reflection point position is calculated in real time. According to the corneal reflection point method, the distance from the pupil to the corneal reflection point per millimeter is calculated as the strabismus degree H0. The strabismus degree of the subject is calculated during the motor cover process. The unit is degree.

5. The strabismus measurement system according to claim 1, characterized in that: The linear drive comprises a motor (31) and a driven wheel (33), wherein the motor and the driven wheel are connected via a belt (32), a mounting plate (34) is mounted on the belt, and the cover plate is mounted on the mounting plate.

Images