Astigmatism axial position marker and marking method

Abstract

The invention provides an astigmatism axial position marker and a marking method, and relates to the technical field of medical instruments. The astigmatism axial position marker comprises a base, a control panel, a base and a display panel, the control panel is arranged in the front of the center of the upper end face of the base, the base is arranged in the rear of the center of the upper end face of the base, the display panel is arranged in the center of the front end face of the base, and a body is arranged on the upper end face of the base. A storage groove is formed in the upper center of the rear end face of the body. Facial feature points are extracted through image processing, the distance between the two eyes of a patient is calculated, the positions of the two flexible eyeshades are adjusted through the adjusting structure, the flexible eyeshades are made to be rapidly matched with the distance between the two eyes of the patient, and therefore it is ensured that the two monitoring marking structures can accurately detect and mark the cornea or a target area, and the detection accuracy is improved. Marking content can comprise specific reference marks (such as 0 degree, 180 degrees and 270 degrees) of corneal limbus or the axial position of an intraocular lens to be implanted and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of medical devices, and in particular to an astigmatism axis marker and a marking method. Background Art

[0002] Cataract is the most common blinding eye disease in the world. Phacoemulsification cataract extraction combined with intraocular lens implantation is the main treatment for cataract. Intraocular lens is an optical lens that replaces the turbid lens. The operation requires the use of phacoemulsification technology to remove the nucleus and cortex of the turbid lens, retain the lens capsule, and then implant the intraocular lens in the lens capsule. The main methods for correcting astigmatism in cataract patients include wearing glasses after surgery, corneal excimer laser surgery, corneal limbal loosening incision on the steep axis of the cornea, and the use of astigmatism-correcting intraocular lenses. Currently, laser marking is used before surgery.

[0003] There are still some problems in the use of existing astigmatism axis markers. The accuracy of laser marking of both eyes is crucial because any slight deviation may lead to poor surgical results or complications. The current laser markers and detection lenses cannot be adjusted according to the distance between the patient's eyes. The deviation between the detection lens, marking laser head and eyeball can easily cause errors. Therefore, those skilled in the art provide an astigmatism axis marker and a marking method to solve the problems raised in the above background technology. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the deficiencies in the prior art, the present invention provides an astigmatism axis marker and a marking method, which solve the problem that the accuracy of laser marking of both eyes is crucial, because any slight deviation may lead to poor surgical results or complications, and the current laser markers and detection lenses cannot be adjusted according to the distance between the patient's eyes, and the deviation between the detection lens, the marking laser head and the eyeball can easily cause errors.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an astigmatism axis marker, comprising a base, a control panel, a pedestal and a display panel, the control panel being arranged at the front center of the upper end surface of the base, the pedestal being arranged at the rear center of the upper end surface of the base, the display panel being arranged at the center of the front end surface of the base, the upper end surface of the base being provided with a main body, a storage slot being arranged at the upper center of the rear end surface of the main body, an adjustment structure being arranged inside the storage slot, grooves being arranged at the rear center of both inner side walls of the storage slot, elastic connecting wires being arranged inside the two grooves, connecting members being fixedly connected to the other ends of the two elastic connecting wires, monitoring marking structures being arranged at the center of the rear end surfaces of the two connecting members, flexible eye masks being arranged at the rear end surfaces of the two connecting members outside the two monitoring marking structures, stereo cameras being arranged at the rear center of both side walls of the two flexible eye masks, at the upper center of the rear end surface of the main body and at the lower center of the rear end surface of the pedestal;

[0008] The adjustment structure includes a base plate, two guide rods, two guide tubes, a first screw rod, a micro servo motor, two sliders and a second screw rod. The micro servo motor is arranged at the front center of an inner wall of a storage tank. The second screw rod is fixedly connected to the output end of the micro servo motor. The first screw rod is fixedly connected to the other end of the second screw rod. The other end of the first screw rod is rotatably connected to the other inner wall of the storage tank. The two sliders are respectively threadedly sleeved on the outside of the first screw rod and the second screw rod, and are both slidably connected to the inside of the storage tank. The two connecting members are respectively fixedly connected to the rear end surfaces of the two sliders. The base plate is fixedly connected to the rear end surface of the main body of the lower end of the storage tank. The two guide rods are respectively arranged at the center of the two side walls of the base plate. The two guide tubes are respectively sleeved on the outside of the two guide rods, and the other ends are respectively fixedly connected to the center of the upper end surfaces of the two flexible eye masks.

[0009] Preferably, an electric telescopic rod is provided at the center of the upper end surface of the base plate, and a lifting block is provided at the output end of the electric telescopic rod.

[0010] Preferably, the control panel also includes a visual positioning adjustment system, an automatic identification of astigmatism axis marking system and a continuous learning system, and the monitoring marking structure includes a detection camera and a laser marking head.

[0011] A method for marking an astigmatism axis comprises the following steps:

[0012] S1. Visual positioning: When the patient sits at the rear end of the main body, the patient's face is recognized by four stereo cameras;

[0013] S1.1. User preparation: Let the user sit within the capture range of the camera, maintain a natural expression and posture, and activate four stereo cameras at the same time to capture multi-angle images of the user's face, and transmit the captured image data to the image processing unit;

[0014] S1.2. Image processing: denoising, correcting and enhancing the captured images, extracting facial feature points from the images using the image processing algorithm deep learning model, performing stereo matching on images from different cameras, calculating the parallax of the feature points, and thus deriving depth information;

[0015] In deep learning, facial feature point detection is usually regarded as a regression problem. The goal of the model is to predict the coordinates (x, y) of each key point. The following are some basic mathematical expressions:

[0016] Input: a facial image I;

[0017] Output: A set of keypoint coordinates {(x1, y1), (x2, y2), ..., (x n ,y n )}, where n is the number of key points;

[0018] Model predictions:

[0019]

[0020] Among them, f θ is the deep learning model, θ is the parameter of the model, are the predicted key point coordinates;

[0021] Loss function, in order to train the model, you need to define a loss function to measure the difference between the predicted value and the true value. Commonly used loss functions include mean square error and smooth L1 loss. Mean square error:

[0022]

[0023] Smoothed L1 loss:

[0024]

[0025] Among them, Y i are the real key point coordinates, are the predicted key point coordinates;

[0026] S1.3. 3D modeling, converting the depth information of feature points into 3D point cloud data, using 3D reconstruction algorithm to convert the point cloud data into a 3D mesh model, cleaning, smoothing and texture mapping the generated 3D model to improve the realism and accuracy of the model;

[0027] S1.4. Spacing adjustment: the spacing between the patient's two eyes is calculated through the above steps, and the micro-servo motor is controlled to start, the micro-servo motor drives the second screw to rotate, and the second screw drives the first screw to rotate synchronously, so that the two sliders move to the center position or to both sides at the same time, and the positions of the two flexible eye masks are adjusted. Then, the height of the lifting block is adjusted by starting the electric telescopic rod to extend and retract, so as to quickly adapt to the spacing between the patient's two eyes, so that the two monitoring marking structures can detect and mark the cornea more accurately;

[0028] S2. Monitoring markers: The corneal position is detected through two monitoring marker structures, and a large amount of patient eye data is collected, including corneal topography, refractive data and OCT images. The collected data is cleaned, standardized and enhanced to ensure data quality.

[0029] S2.1. Feature extraction: extract key features from the preprocessed data, such as corneal curvature, astigmatism and axis position;

[0030] S2.2. Model selection: Select an appropriate machine learning model for structured data and use the labeled dataset to train the model so that it can identify the axis of astigmatism.

[0031] S2.3. Diagnosis and suggestions: deploy the trained model to automatically analyze the eye data of new patients, identify the axis of astigmatism, and provide marking suggestions for the target axis and surgical incision location based on the model output for the doctor's reference;

[0032] S2.4. Laser marking, automatic identification of astigmatism axis marking system The laser system can be used to make precise marks on the cornea, marking the target axis, surgical incision location, and specific reference marks on the corneal limbus (0°, 180°, 270°);

[0033] S3. Prepare and perform surgery based on the marking results;

[0034] S4. Continuous learning. In a clinical setting, doctors’ labeling results and surgical effect data are collected, the collected data are cleaned and formatted to ensure data quality, and the collected data are used to fine-tune or retrain the existing model. Assuming the initial model M0, when new data D is received, new After that, update the model parameters θ:

[0035]

[0036] Among them, η is the learning rate and L is the loss function.

[0037] Preferably, the deep learning model described in step S1.2. above further includes a stochastic gradient descent module for training the deep learning model, including a model for facial feature point detection, the deep learning model f θ , where θ represents the parameters of the model, the input of the model is a facial image I, and the output is a set of predicted facial feature point coordinates Where n is the number of feature points:

[0038]

[0039] Using the SGD algorithm, we update the parameters by calculating the gradient of the loss function with respect to the model parameters. The formula for parameter update is as follows:

[0040]

[0041] Among them, θ new is the updated parameter, θ old is the current parameter, η is the learning rate, which controls the step size of each update, is the gradient of the loss function L with respect to the parameters θ.

[0042] (III) Beneficial effects

[0043] The present invention provides an astigmatism axis marker and a marking method, which have the following beneficial effects:

[0044] 1. In the present invention, the captured image is denoised, corrected and enhanced by image processing, the facial feature points are extracted from the image by using the image processing algorithm deep learning model, the distance between the patient's two eyes is calculated, and the micro-servo motor is controlled to start, the micro-servo motor drives the second screw to rotate, and the second screw drives the first screw to rotate synchronously, so that the two sliders move to the center position or to both sides at the same time, and the positions of the two flexible eye masks are adjusted to facilitate rapid adaptation to the distance between the patient's two eyes, thereby ensuring that the two monitoring marking structures can accurately detect and mark the cornea or target area, and the marking content may include specific reference marks of the corneal limbus (such as 0°, 180°, 270°) or the axis position of the artificial lens to be implanted, etc.

[0045] 2. In the present invention, the key features, corneal curvature, astigmatism degree and axis position, are extracted from the pre-processed data by automatically identifying the astigmatism axis marking system, and the trained model is deployed to automatically analyze the eye data of new patients, identify the astigmatism axis position, and provide marking suggestions for the target axis position and surgical incision position based on the model output for the doctor's reference. The automatic system can quickly process large amounts of data and shorten the diagnosis time. The automated process can reduce the workload of doctors, allowing them to have more time to focus on other important clinical tasks.

[0046] 3. In the present invention, the doctor's marking results and surgical effect data are collected through a continuous learning system, and the collected data are cleaned and formatted to ensure data quality. Removing errors and outliers can reduce noise in model training and improve the generalization ability of the model. Through continuous learning, the system can adapt to changes in different patients and surgical conditions and improve the robustness of the model.

[0047] 4. In the present invention, the face is recognized during the process of automatically capturing images, the distance between the two eyes is adjusted, and then the height of the lifting block is adjusted by starting the electric telescopic rod to extend and contract. The patient's chin is lifted by the lifting block, which is convenient for rapid recognition according to the patient's face shape. After recognition, it is adjusted according to the face shape, so that it is suitable for patients with different face shapes. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A perspective view of the present invention;

[0049] Figure 2 A three-dimensional diagram of another viewing angle of the present invention;

[0050] Figure 3 is a rear cross-sectional view of the present invention;

[0051] Figure 4 A top sectional view of the present invention;

[0052] Figure 5 It is a schematic diagram of the flow of the visual positioning adjustment system of the present invention;

[0053] Figure 6 The figure is a flow chart of the automatic astigmatism axis marking system of the present invention.

[0054] Among them, 1. base; 2. control panel; 3. display panel; 4. main body; 5. base; 6. stereo camera; 7. flexible eye mask; 8. storage slot; 9. adjustment structure; 901. base plate; 902. guide rod; 903. guide tube; 904. lifting block; 905. electric telescopic rod; 906. first screw rod; 907. micro servo motor; 908. slider; 909. second screw rod; 10. monitoring mark structure; 11. groove; 12. elastic connecting line; 13. connecting piece. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Embodiment 1:

[0057] like Figure 1-6 As shown, an embodiment of the present invention provides an astigmatism axis marker, including a base 1, a control panel 2, a base 5 and a display panel 3, wherein the control panel 2 is arranged at the front center of the upper end surface of the base 1, the base 5 is arranged at the rear center of the upper end surface of the base 1, the display panel 3 is arranged at the center of the front end surface of the base 5, the upper end surface of the base 5 is provided with a main body 4, the center of the rear end surface of the main body 4 is provided with a storage groove 8, an adjustment structure 9 is arranged inside the storage groove 8, grooves 11 are opened at the rear center of the two inner side walls of the storage groove 8, elastic connecting wires 12 are arranged inside the two grooves 11, and the other ends of the two elastic connecting wires 12 are fixedly connected with connecting wires A connector 13 is connected to the two control panels 2 and the display panel 3 through two elastic connecting lines 12. A monitoring mark structure 10 is provided at the center of the rear end surface of the connector 13. Detection and marking are performed through the two monitoring mark structures 10. Flexible eye masks 7 are provided on the rear end surfaces of the two connectors 13 outside the two monitoring mark structures 10. The two flexible eye masks 7 reduce the patient's discomfort. Stereo cameras 6 are provided at the rear centers of the two side walls of the two flexible eye masks 7, at the upper center of the rear end surface of the main body 4, and at the lower center of the rear end surface of the base 5. The facial features of the patient are extracted through the four stereo cameras 6.

[0058] The adjustment structure 9 includes a base plate 901, two guide rods 902, two guide tubes 903, a first screw rod 906, a micro servo motor 907, two sliders 908 and a second screw rod 909. The micro servo motor 907 is arranged at the front center of an inner side wall of the storage tank 8. The second screw rod 909 is fixedly connected to the output end of the micro servo motor 907. The first screw rod 906 is fixedly connected to the other end of the second screw rod 909. The other end of the first screw rod 906 is rotatably connected to the other inner side wall of the storage tank 8. The two sliders 908 are respectively threadedly sleeved on the outside of the first screw rod 906 and the second screw rod 909, and are both slidably connected to the inside of the storage tank 8. The two connecting members 13 are respectively fixedly connected On the rear end surfaces of the two sliders 908, the base plate 901 is fixedly connected to the rear end surface of the main body 4 at the lower end of the storage tank 8, and the two guide rods 902 are respectively arranged at the center of the two side walls of the base plate 901, and the two guide tubes 903 are respectively sleeved on the outside of the two guide rods 902, and the other ends are respectively fixedly connected to the center of the upper end surfaces of the two flexible eye masks 7. By controlling the micro servo motor 907 to start, the micro servo motor 907 drives the second screw rod 909 to rotate, and the second screw rod 909 drives the first screw rod 906 to rotate synchronously, so that the two sliders 908 move to the center position or to both sides at the same time, and adjust the position of the two flexible eye masks to facilitate rapid adaptation to the distance between the patient's eyes.

[0059] An electric telescopic rod 905 is provided at the center of the upper end surface of the base plate 901, and a lifting block 904 is provided at the output end of the electric telescopic rod 905. The electric telescopic rod 905 is extended and retracted to adjust the height of the lifting block 904, and the patient's chin is lifted by the lifting block.

[0060] The control panel 2 also includes a visual positioning adjustment system, an automatic identification of astigmatism axis marking system and a continuous learning system, and the monitoring marking structure 10 includes a detection camera and a laser marking head.

[0061] A method for marking an astigmatism axis comprises the following steps:

[0062] S1. Visual positioning, when the patient sits at the rear end of the main body 4, the patient's face is recognized by four stereo cameras 6;

[0063] S1.1. User preparation: Let the user sit within the capture range of the camera, maintain a natural expression and posture, and activate four stereo cameras at the same time to capture multi-angle images of the user's face, and transmit the captured image data to the image processing unit;

[0064] S1.2. Image processing: denoising, correcting and enhancing the captured images, extracting facial feature points from the images using the image processing algorithm deep learning model, performing stereo matching on images from different cameras, calculating the parallax of the feature points, and thus deriving depth information;

[0065] In deep learning, facial feature point detection is usually regarded as a regression problem. The goal of the model is to predict the coordinates (x, y) of each key point. The following are some basic mathematical expressions:

[0066] Input: a facial image I;

[0067] Output: A set of keypoint coordinates {(x1, y1), (x2, y2), ..., (x n ,y n )}, where n is the number of key points;

[0068] Model predictions:

[0069]

[0070] Among them, f θ is the deep learning model, θ is the parameter of the model, are the predicted key point coordinates;

[0071] Loss function, in order to train the model, you need to define a loss function to measure the difference between the predicted value and the true value. Commonly used loss functions include mean square error and smooth L1 loss. Mean square error:

[0072]

[0073] Smoothed L1 loss:

[0074]

[0075] Among them, Y i are the real key point coordinates, are the predicted key point coordinates;

[0076] S1.3. 3D modeling, converting the depth information of feature points into 3D point cloud data, using 3D reconstruction algorithm to convert the point cloud data into a 3D mesh model, cleaning, smoothing and texture mapping the generated 3D model to improve the realism and accuracy of the model;

[0077] S1.4. Spacing adjustment: the spacing between the patient's two eyes and the target axis position or surgical incision position of the artificial lens to be implanted are calculated through the above steps. The micro servo motor 907 is controlled to start, and the micro servo motor 907 drives the second screw 909 to rotate. The second screw 909 drives the first screw 906 to rotate synchronously, so that the two sliders 908 move to the center position or both sides at the same time to adjust the positions of the two flexible eye masks. Then, the electric telescopic rod 905 is started to extend and retract to adjust the height of the lifting block 904, so as to quickly adapt to the spacing between the patient's two eyes, so that the two monitoring marking structures 10 can detect and mark the cornea more accurately.

[0078] S2. Monitoring markers: The corneal position is detected by two monitoring marker structures 10, and a large amount of patient eye data, including corneal topography, refractive data and OCT images, is collected, cleaned, standardized and enhanced to ensure data quality;

[0079] S2.1. Feature extraction: extract key features from the preprocessed data, such as corneal curvature, axis position of the proposed intraocular lens, etc.

[0080] S2.2. Model selection: Select an appropriate machine learning model for structured data and use the labeled dataset to train the model so that it can recognize and predict the target axis or surgical incision position;

[0081] S2.3. Diagnosis and recommendations: deploy trained models to automatically analyze the eye data of new patients, identify the axis of astigmatism, and provide marking recommendations for the target axis or surgical incision location based on the model output for the doctor's reference;

[0082] S2.4. Laser marking, automatic identification of astigmatism axis marking system. The laser system can be used to make precise marks on the cornea to mark the target axis, surgical incision location, or specific reference marks on the limbus (such as 0°, 180°, 270°).

[0083] S3. Prepare and perform surgery based on the marking results;

[0084] S4. Continuous learning. In a clinical setting, doctors’ labeling results and surgical effect data are collected, the collected data are cleaned and formatted to ensure data quality, and the collected data are used to fine-tune or retrain the existing model. Assuming the initial model M0, when new data D is received, new After that, update the model parameters θ:

[0085]

[0086] Among them, η is the learning rate, L is the loss function, and the continuous learning system collects the doctors' marking results and surgical effect data. The collected data is cleaned and formatted to ensure data quality. Removing errors and outliers can reduce noise in model training and improve the generalization ability of the model. Through continuous learning, the system can adapt to changes in different patients and surgical conditions and improve the robustness of the model.

[0087] In the deep learning model in the above step S1.2., a stochastic gradient descent module is also included for training the deep learning model, including a model for facial feature point detection. The deep learning model f θ , where θ represents the parameters of the model, the input of the model is a facial image I, and the output is a set of predicted facial feature point coordinates Where n is the number of feature points:

[0088]

[0089] Using the SGD algorithm, we update the parameters by calculating the gradient of the loss function with respect to the model parameters. The formula for parameter update is as follows:

[0090]

[0091] Among them, θ new is the updated parameter, θ old is the current parameter, η is the learning rate, which controls the step size of each update, is the gradient of the loss function L with respect to the parameters θ.

[0092] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An astigmatism axis marker, comprising a base (1), a control panel (2), a pedestal (5) and a display panel (3), wherein the control panel (2) is arranged at the front center of the upper end surface of the base (1), the pedestal (5) is arranged at the rear center of the upper end surface of the base (1), and the display panel (3) is arranged at the center of the front end surface of the pedestal (5), characterized in that: The upper end surface of the base (5) is provided with a main body (4), a storage groove (8) is provided at the center of the upper rear end surface of the main body (4), an adjustment structure (9) is provided inside the storage groove (8), grooves (11) are provided at the center of the two inner side walls of the storage groove (8), elastic connecting lines (12) are provided inside the two grooves (11), the other ends of the two elastic connecting lines (12) are fixedly connected with connecting pieces (13), monitoring marking structures (10) are provided at the center of the rear end surfaces of the two connecting pieces (13), flexible eye masks (7) are provided on the rear end surfaces of the two connecting pieces (13) outside the two monitoring marking structures (10), and stereo cameras (6) are provided at the center of the two side walls of the two flexible eye masks (7), at the center of the upper rear end surface of the main body (4) and at the center of the lower rear end surface of the base (5); The adjusting structure (9) comprises a base plate (901), two guide rods (902), two guide tubes (903), a first screw rod (906), a micro servo motor (907), two sliders (908) and a second screw rod (909), wherein the micro servo motor (907) is arranged at the front center of an inner side wall of the storage tank (8), the second screw rod (909) is fixedly connected to the output end of the micro servo motor (907), the first screw rod (906) is fixedly connected to the other end of the second screw rod (909), the other end of the first screw rod (906) is rotatably connected to the other inner side wall of the storage tank (8), and the two The sliders (908) are respectively threadedly sleeved on the outside of the first screw rod (906) and the second screw rod (909), and are both slidably connected to the inside of the storage tank (8). The two connecting members (13) are respectively fixedly connected to the rear end surfaces of the two sliders (908). The base plate (901) is fixedly connected to the rear end surface of the main body (4) at the lower end of the storage tank (8). The two guide rods (902) are respectively arranged at the center of the two side walls of the base plate (901). The two guide tubes (903) are respectively sleeved on the outside of the two guide rods (902), and the other ends are respectively fixedly connected to the center of the upper end surfaces of the two flexible eye masks (7).

2. The astigmatism axis marker according to claim 1, characterized in that: An electric telescopic rod (905) is provided at the center of the upper end surface of the base plate (901), and a lifting block (904) is provided at the output end of the electric telescopic rod (905).

3. The astigmatism axis marker according to claim 1, characterized in that: The control panel (2) also includes a visual positioning adjustment system, an automatic identification of astigmatism axis marking system and a continuous learning system, and the monitoring marking structure (10) includes a detection camera and a laser marking head.

4. A method for marking an astigmatism axis, using an astigmatism axis marker according to any one of claims 1 to 3, characterized in that: The following steps are involved: S1. Visual positioning: when the patient sits at the rear end of the main body (4), the patient's face is recognized by four stereo cameras (6); S1.

1. User preparation: Let the user sit within the capture range of the camera, maintain a natural expression and posture, and activate four stereo cameras at the same time to capture multi-angle images of the user's face, and transmit the captured image data to the image processing unit; S1.

2. Image processing: denoising, correcting and enhancing the captured images, extracting facial feature points from the images using the image processing algorithm deep learning model, performing stereo matching on images from different cameras, calculating the parallax of the feature points, and thus deriving depth information; In deep learning, facial feature point detection is usually regarded as a regression problem. The goal of the model is to predict the coordinates (x, y) of each key point. The following are some basic mathematical expressions: Input: a facial image I; Output: A set of keypoint coordinates {(x1, y1), (x2, y2), ..., (x n ,y n )}, where n is the number of key points; Model predictions: Among them, f θ is the deep learning model, θ is the parameter of the model, are the predicted key point coordinates; Loss function, in order to train the model, you need to define a loss function to measure the difference between the predicted value and the true value. Commonly used loss functions include mean square error and smooth L1 loss. Mean square error: Smoothed L1 loss: Among them, Y i are the real key point coordinates, are the predicted key point coordinates; S1.

3. 3D modeling, converting the depth information of feature points into 3D point cloud data, using 3D reconstruction algorithm to convert the point cloud data into a 3D mesh model, cleaning, smoothing and texture mapping the generated 3D model to improve the realism and accuracy of the model; S1.

4. Spacing adjustment: the spacing between the patient's two eyes, the target axis position of the artificial lens to be implanted, and the surgical incision position are calculated through the above steps. The micro servo motor (907) is controlled to start, and the micro servo motor (907) drives the second screw (909) to rotate. The second screw (909) drives the first screw (906) to rotate synchronously, so that the two sliders (908) move to the center position or to both sides at the same time, and the positions of the two flexible eye masks are adjusted. Then, the electric telescopic rod (905) is started to extend and retract, and the height of the lifting block (904) is adjusted to facilitate rapid adaptation of the spacing between the patient's two eyes, so that the two monitoring marking structures (10) are more accurate when detecting and marking the cornea; S2. Monitoring markers: using two monitoring marker structures (10) to detect the corneal position, collect a large amount of patient eye data, including corneal topography, refractive data and OCT images, and clean, standardize and enhance the collected data to ensure data quality; S2.

1. Feature extraction: extract key features from the preprocessed data, such as corneal curvature and the axis position of the artificial lens to be implanted; S2.

2. Model selection: Select an appropriate machine learning model for structured data and use the labeled dataset to train the model so that it can recognize and predict the target axis and surgical incision location. S2.

3. Diagnosis and suggestions: deploy the trained model to automatically analyze the eye data of new patients, identify the axis of astigmatism, and provide marking suggestions for the target axis and surgical incision location based on the model output for the doctor's reference; S2.

4. Laser marking, automatic identification of astigmatism axis marking system The laser system can be used to make precise marks on the cornea, marking the target axis, surgical incision location, and specific reference marks on the corneal limbus (0°, 180°, 270°); S3. Prepare and perform surgery based on the marking results; S4. Continuous learning. In a clinical setting, doctors’ labeling results and surgical effect data are collected, the collected data are cleaned and formatted to ensure data quality, and the collected data are used to fine-tune or retrain the existing model. Assuming the initial model M0, when new data D is received, new After that, update the model parameters θ: Among them, η is the learning rate and L is the loss function.

5. The astigmatism axis marking method according to claim 4, characterized in that: In the deep learning model described in step S1.

2. above, a stochastic gradient descent module is also included for training the deep learning model, including a model for facial feature point detection. The deep learning model f θ , where θ represents the parameters of the model, the input of the model is a facial image I, and the output is a set of predicted facial feature point coordinates Where n is the number of feature points: Using the SGD algorithm, we update the parameters by calculating the gradient of the loss function with respect to the model parameters. The formula for parameter update is as follows: Among them, θ new is the updated parameter, θ old is the current parameter, η is the learning rate, which controls the step size of each update, is the gradient of the loss function L with respect to the parameters θ.

Images