Automatic parameter optimization method for laser vision correction surgery

Abstract

The invention relates to the technical field of laser vision correction, and particularly discloses an automatic parameter optimization method for a laser vision correction surgery, which comprises the following steps: acquiring and preprocessing multi-dimensional eye data to obtain standardized data; an individual eye feature coupling model is constructed, and parameter nonlinear correlation is mined through an improved neural network; a multi-objective optimization function is constructed based on the coupling matrix, and initial operation parameters are generated through improved particle swarm optimization; dynamic data are collected in the operation, and parameters are dynamically adjusted through a model prediction control algorithm; and verifying the parameters after operation, and updating the model for re-optimization if the requirements are not met. Through multi-dimensional data support, multi-objective optimization and balance correction precision and safety, intraoperative dynamic adjustment and a closed-loop mechanism of postoperative verification, parameter optimization pertinence and operation comprehensive effect and reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of laser vision correction technology, specifically to an automatic parameter optimization method for laser vision correction surgery. Background Technology

[0002] Laser vision correction surgery is currently the mainstream medical treatment for refractive errors. Its core principle is to precisely ablate corneal tissue with a laser, adjusting the corneal refractive power to correct vision. The rationality of surgical parameters directly determines the corrective effect, postoperative visual quality, and surgical safety. Therefore, the optimized design of surgical parameters is a crucial step in laser vision correction surgery.

[0003] Current laser vision correction surgery parameter design largely relies on the surgeon's clinical experience, manually setting parameters based on the patient's preoperative baseline refractive data. This approach has several significant drawbacks: First, individual eye structures vary considerably, and relying solely on baseline refractive data cannot fully reflect the intrinsic relationships between various eye parameters, easily leading to insufficiently targeted parameter design and affecting correction accuracy. Second, manually setting parameters makes it difficult to simultaneously consider multiple objectives such as refractive correction effect, corneal structural safety, and postoperative visual quality, often resulting in compromises. For example, excessive pursuit of correction effect may lead to insufficient residual corneal thickness, causing postoperative complications. Third, dynamic factors such as micro-movements of the patient's eyeball and changes in corneal tissue ablation characteristics during surgery cannot be perceived and responded to in real time. Static preoperative parameter design cannot adapt to dynamic changes during surgery, potentially leading to significant deviations between actual ablation effect and expectations. Fourth, existing parameter optimization methods employ relatively simple algorithm models, failing to effectively explore the nonlinear coupling relationships of eye parameters, and thus requiring improvement in optimization accuracy and efficiency.

[0004] Therefore, there is an urgent need for an automatic optimization method for laser vision correction surgery parameters that can comprehensively integrate multi-dimensional individual eye data, accurately mine the intrinsic correlation of parameters, take into account the needs of multi-objective optimization, and respond in real time to dynamic changes during surgery, so as to overcome the shortcomings of existing technologies and improve the safety, accuracy and reliability of the surgery. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide an automatic parameter optimization method for laser vision correction surgery. By comprehensively collecting and deeply mining multi-dimensional eye data, an individual-specific eye feature coupling model is constructed. Combined with an improved optimization algorithm, multi-objective initial optimization of surgical parameters is achieved. Through intraoperative dynamic adjustment and postoperative verification optimization, the entire process of surgical parameters is precisely optimized, thereby improving the correction accuracy, postoperative visual quality and surgical safety of laser vision correction surgery.

[0006] To solve the above-mentioned technical problems, the technical solution provided by the present invention is: an automatic parameter optimization method for laser vision correction surgery, comprising the following steps:

[0007] Step 1: Multi-dimensional ocular data acquisition and preprocessing: Collect multi-dimensional ocular data of the patient before surgery, including basic refractive data, corneal structure data, ocular surface morphology data and ocular physiological function data; perform noise reduction, outlier removal and standardization on the raw data to obtain standardized ocular data;

[0008] Step 2: Construct an individual eye feature coupling model: Extract eye feature vectors based on standardized eye data, mine the nonlinear correlation of eye parameters in different dimensions through an improved neural network model, and output an individual eye feature coupling matrix; the improved neural network model adopts a structure combining convolutional layers, attention mechanisms and fully connected layers;

[0009] Step 3: Initial Surgical Parameter Generation: A multi-objective optimization function is constructed based on the individual ocular feature coupling matrix. This function is solved using an improved particle swarm optimization algorithm to generate an initial surgical parameter set that satisfies safety constraints. The multi-objective optimization function is:

[0010] minF(θ)=[f1(θ),f2(θ),f3(θ)] T ;

[0011] Where θ is the surgical parameter vector, f1(θ) is the refractive correction error function, f2(θ) is the corneal structural risk function, and f3(θ) is the visual quality loss function;

[0012] The improved particle swarm optimization algorithm introduces dynamic inertia weights, the expression of which is:

[0013] Step 4: Real-time data acquisition and dynamic optimization during surgery: Acquire dynamic data of corneal ablation during surgery, and calculate the deviation value δ(k) = |X real (k)-X pred (k)|, if the deviation exceeds a preset threshold, the surgical parameters are dynamically adjusted in real time using a model predictive control algorithm; the model predictive control algorithm is based on a real-time optimization subfunction. Parameter optimization is achieved; where λ is the safety weight coefficient, with a value ranging from 0.6 to 0.8.

[0014] Step 5: Parameter Optimization, Verification, and Output: Collect real-time eye data post-surgery and calculate the real-time post-operative correction error ε = |D post -D targetIf ε≤0.5D and the corneal structure indicators meet the safety requirements, the final optimized parameters are output; otherwise, the model is updated and re-optimized until the parameters that meet the requirements are obtained.

[0015] Further, in step 1, the basic refractive data includes spherical power, cylindrical power, axis, and accommodative amplitude; the corneal structural data includes central corneal thickness, peripheral corneal thickness, corneal stroma thickness, and corneal endothelial cell density; the ocular surface morphology data includes corneal curvature, corneal topography characteristic parameters, and tear film breakup time; and the ocular physiological function data includes intraocular pressure, pupil diameter for photopic / scotopic vision, and axial length.

[0016] Furthermore, in step 1, the standardization process employs the min-max standardization algorithm, the formula of which is: Where x represents the original data, x ′ For standardized data, x min x max These are the minimum and maximum values ​​of the data in this dimension, respectively.

[0017] Furthermore, in step 2, the ocular feature vector includes basic features and derived features. The derived features are the coupling calculation results of the basic features, including the ratio of corneal thickness to refractive power, the matching coefficient of corneal curvature to pupil diameter, and the difference between corneal stromal thickness and estimated ablation depth.

[0018] Further, in step 2, the input layer of the improved neural network model receives eye feature vectors; the convolutional layer uses three parallel convolutional kernels to perform multi-scale feature extraction on the eye feature vectors to obtain local correlation features of different dimensional parameters; the attention mechanism assigns weights to the feature maps output by the convolutional layer through a spatial attention module, strengthening the weights of key features and weakening the interference of secondary features; the fully connected layer maps the features processed by the attention mechanism to the feature coupling space and outputs an individual eye feature coupling matrix, where the elements in the eye feature coupling matrix represent the coupling correlation strength between different eye parameters.

[0019] Furthermore, in step 3, f1(θ) = |D pred (θ)-D target |, where D pred (θ) represents the postoperative refractive error predicted based on parameter θ, D target The target refractive power.

[0020] Furthermore, in step 3, f2(θ) = max(0,T) safe -H resid (θ))+max(0,σ pred (θ)-σ safe), where T safe H is the safe threshold for residual corneal thickness. resid (θ) represents the remaining corneal thickness after surgery, σ pred (θ) represents the predicted postoperative corneal stress, σ safe This is the corneal stress safety threshold.

[0021] Furthermore, in step 3, Where SA(θ) is the postoperative spherical aberration prediction value, SA pre Preoperative ball aberration measurement value, CS pre This represents the measured value of preoperative contrast sensitivity, CS pred (θ) is the predicted value of postoperative contrast sensitivity.

[0022] Furthermore, in step 4, the dynamic data of corneal ablation includes real-time corneal ablation depth, corneal tissue removal amount, laser spot positioning accuracy, and patient eyeball micro-movement amplitude; the preset thresholds are ablation depth deviation ≥ 5 μm and spot positioning deviation ≥ 0.1 mm.

[0023] Furthermore, in step 4, the surgical parameters that are dynamically adjusted in real time include laser energy density, cutting speed, and spot compensation amount.

[0024] The advantages of this invention compared to the prior art are:

[0025] This invention collects multi-dimensional eye data and obtains eye feature vectors containing derived features through preprocessing and feature extraction. Combined with an improved neural network model, it constructs an individual eye feature coupling matrix, which can accurately explore the nonlinear correlation of individual eye parameters, provide exclusive individual data support for parameter optimization, and improve the pertinence of parameter design.

[0026] This invention constructs a multi-objective optimization function that covers refractive correction error, corneal structural risk, and visual quality loss. It combines an improved particle swarm optimization algorithm to solve the problem, which can simultaneously balance correction accuracy, surgical safety, and postoperative visual quality. This avoids the problem of neglecting one aspect for another caused by single-objective optimization in existing technologies and improves the overall surgical effect.

[0027] This invention introduces an intraoperative real-time data acquisition and dynamic optimization mechanism. Through model predictive control algorithms, it responds in real time to the dynamic deviation of corneal ablation during surgery, and makes precise adjustments to surgical parameters. This effectively makes up for the shortcomings of static preoperative parameter design in adapting to dynamic changes during surgery, and further improves the optimization accuracy of surgical parameters.

[0028] This invention forms a closed-loop optimization process of "preoperative data collection - model optimization - intraoperative adjustment - postoperative verification - model update" through postoperative parameter verification and model update mechanism, ensuring that the final output surgical parameters can meet clinical requirements and improving the reliability and safety of the surgery. Attached Figure Description

[0029] Figure 1 This is a flowchart of an automatic parameter optimization method for laser vision correction surgery according to the present invention. Detailed Implementation

[0030] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0031] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0032] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0033] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0034] The following detailed description, in conjunction with the accompanying drawings, provides a method for automatically optimizing parameters in laser vision correction surgery according to the present invention.

[0035] Combined with appendix Figure 1 The specific implementation process of the automatic parameter optimization method for laser vision correction surgery of the present invention is as follows:

[0036] An automatic parameter optimization method for laser vision correction surgery, specifically including the following steps:

[0037] Step 1: Multi-dimensional eye data acquisition and preprocessing

[0038] First, preoperative multidimensional ocular data was collected from the patient. This comprehensive data covered basic refractive error data, corneal structural data, ocular surface morphology data, and ocular physiological function data to ensure data integrity and completeness, providing data support for subsequent accurate modeling. Basic refractive error data included spherical power, cylindrical power, axis, and accommodative amplitude, reflecting the patient's core refractive error status. Corneal structural data included central corneal thickness, peripheral corneal thickness, corneal stroma thickness, and corneal endothelial cell density, used to assess the ablation capability and structural stability of the corneal tissue. Ocular surface morphology data included corneal curvature, corneal topography characteristic parameters, and tear film breakup time, used to characterize the degree of corneal irregularity and ocular surface health. Ocular physiological function data included intraocular pressure, pupil diameter in photopic / scotopic vision, and axial length, used to reflect the impact of the overall ocular physiological state on the surgery.

[0039] Since the raw data may contain noise and outliers, preprocessing is necessary. Preprocessing includes denoising, outlier removal, and standardization: Gaussian filtering is used to denoise the raw data, eliminating random noise; outliers are removed using the 3σ criterion to ensure data reliability; and the min-max standardization algorithm is used to standardize the data, mapping data of different dimensions and units to the [0, 1] interval to avoid the impact of differences in data units on subsequent model training. The formula for the min-max standardization algorithm is: Where x is the original data, x ′ For standardized data, x min x max These are the minimum and maximum values ​​for this dimension of data, respectively. After preprocessing, standardized eye data is obtained.

[0040] Step 2: Construct an individual eye feature coupling model

[0041] Based on the standardized eye data obtained in step 1, an eye feature vector is extracted. The eye feature vector includes basic features and derived features: the basic features are the original data of each dimension after standardization; the derived features are the coupling calculation results of the basic features, which are obtained by performing correlation operations on the basic features. They can more accurately reflect the intrinsic correlation of parameters in different dimensions, specifically including the ratio of corneal thickness to refractive power, the matching coefficient of corneal curvature and pupil diameter, and the difference between corneal stroma thickness and estimated ablation depth.

[0042] An improved neural network model is used to explore the nonlinear correlations between eye parameters of different dimensions, outputting an individual eye feature coupling matrix. The improved neural network model employs a structure combining convolutional layers, attention mechanisms, and fully connected layers, with the functions of each layer as follows:

[0043] Input layer: Receives the extracted eye feature vectors and transforms them into tensor form that the model can process;

[0044] Convolutional layer: Three parallel convolutional kernels (with kernel sizes of 1×3, 1×5, and 1×7) are used to extract features from the eye feature vector at multiple scales. By using convolutional kernels of different sizes, feature associations at different length scales are captured, and local association features with different dimensional parameters are obtained.

[0045] Attention mechanism: The spatial attention module assigns weights to the feature maps output by the convolutional layer, calculates the attention weight of each feature point (the larger the weight value, the higher the importance of the feature to subsequent modeling), strengthens the weight of key features (such as features corresponding to core parameters such as central corneal thickness and spherical power), weakens the interference of secondary features, and improves the model's ability to capture key information.

[0046] Fully connected layer: Maps the features processed by the attention mechanism to the feature coupling space, and achieves deep feature fusion through multi-layer fully connected operations, ultimately outputting an individual eye feature coupling matrix. The elements in the eye feature coupling matrix represent the coupling strength between different eye parameters; the larger the matrix element value, the stronger the intrinsic correlation between the corresponding two eye parameters.

[0047] Step 3: Initial Surgical Parameter Generation

[0048] Based on the individual eye feature coupling matrix obtained in step 2, and combined with the core objective of laser vision correction surgery, a multi-objective optimization function is constructed. The function is solved by an improved particle swarm optimization algorithm to generate an initial set of surgical parameters that meets the safety constraints.

[0049] The multi-objective optimization function is: minF(θ)=[f1(θ),f2(θ),f3(θ)] T ;

[0050] Where θ is the surgical parameter vector, including core surgical parameters such as laser energy density, cutting speed, spot size, and cutting depth; f1(θ) is the refractive correction error function, used to measure the deviation between the predicted refractive correction effect and the target effect corresponding to the surgical parameters; f2(θ) is the corneal structure risk function, used to assess the impact of surgical parameters on the safety of corneal structure; and f3(θ) is the visual quality loss function, used to measure the postoperative visual quality loss that may be caused by surgical parameters.

[0051] The specific expressions for each objective function are as follows:

[0052] Refractive error function: f1(θ)=|D pred (θ)-D target |;wherein, D pred(θ) represents the postoperative refractive error predicted based on parameter θ, obtained by combining the correlation between corneal ablation amount and refractive error using an individual ocular feature coupling model; D target The target refractive power is the ideal refractive power determined based on the patient's preoperative needs and ocular condition.

[0053] Corneal structural risk function: f2(θ)=max(0,T) safe -H resid (θ))+max(0,σ pred (θ)-σ safe ), where T safe The safe threshold for residual corneal thickness is set at no less than 410 μm according to clinical standards; H resid (θ) represents the remaining corneal thickness after surgery, obtained by subtracting the estimated ablation depth based on parameter θ from the preoperative central corneal thickness; σ pred (θ) represents the predicted postoperative corneal stress, obtained through a corneal structural mechanics model combined with postoperative corneal morphology prediction; σ safe The corneal stress safety threshold is set based on corneal biomechanical experimental data; when H resid (θ)≥T safe And σ pred (θ)≤σ safe When f2(θ) = 0, it indicates that the corneal structure is safe and without risk.

[0054] 3) Visual quality loss function: in,

[0055] SA(θ) is the predicted value of postoperative spherical aberration, obtained by combining an ocular optical model with postoperative corneal morphology; SA pre Preoperative ball aberration measurement; CS pre Preoperative comparative sensitivity measurement value; CS pred (θ) is the postoperative contrast sensitivity prediction value, which reflects the patient's ability to distinguish objects of different spatial frequencies after surgery; the smaller the value of this function, the smaller the loss of visual quality after surgery and the better the visual effect.

[0056] The improved particle swarm optimization algorithm introduces dynamic inertia weights on the basis of the traditional particle swarm optimization algorithm to balance the algorithm's global search capability and local convergence capability. The expression for the dynamic inertia weights is:

[0057] Where w(k) is the inertia weight of the k-th iteration, w max For the maximum inertia weight, w min For the minimum inertia weight, k maxLet be the maximum number of iterations, and k be the current iteration number. As the number of iterations increases, the inertia weight decreases linearly, giving the algorithm a larger inertia weight in the early stages of iteration, enhancing global search capabilities and avoiding getting trapped in local optima; and a smaller inertia weight in the later stages of iteration, enhancing local convergence capabilities and improving optimization accuracy.

[0058] The algorithm is constrained as follows: laser energy density range of 100–300 mJ / cm². 2 The cutting speed range is 100–500 Hz, and the postoperative corneal residual thickness is ≥T. safe Postoperative corneal stress ≤σ safe By solving the multi-objective optimization function using an improved particle swarm optimization algorithm, multiple candidate initial surgical parameters were obtained from the Pareto optimal solution set. Combined with clinical experience, the initial surgical parameter set with the best overall performance was then selected.

[0059] Step 4: Real-time data acquisition and dynamic optimization during the surgical procedure

[0060] During laser vision correction surgery, real-time monitoring equipment is used to collect dynamic data on corneal ablation. This data includes real-time corneal ablation depth, amount of corneal tissue removed, laser spot positioning accuracy, and the amplitude of micro-movements of the patient's eyeball. This data is used to monitor the deviation between the surgical procedure and the expected results in real time.

[0061] Based on the individual ocular feature coupling model constructed in step 2, the theoretical data for corneal ablation at the corresponding surgical stage (i.e., predicted data X) is predicted. pred (k)), calculate the real-time acquired data X real The deviation between (k) and the predicted data: δ(k)=|X real (k)-X pred (k)|; Set a preset threshold, which is set according to the clinical surgical precision requirements as follows: cutting depth deviation ≥ 5 μm, spot positioning deviation ≥ 0.1 mm; If the deviation value δ(k) does not exceed the preset threshold, it indicates that the surgical process meets expectations and the initial surgical parameters are continued; If the deviation value exceeds the preset threshold, it indicates that dynamic deviation occurs during the operation, and the surgical parameters need to be dynamically adjusted in real time through the model prediction control algorithm to eliminate the deviation and ensure the surgical effect.

[0062] Model predictive control algorithms optimize parameters based on real-time optimization sub-functions, which are: Wherein, Δθ is the adjustment amount of surgical parameters, J is the optimization target value, λ is the safety weight coefficient, with a value ranging from 0.6 to 0.8 (set according to the priority of surgical safety; the higher the priority, the larger the value of λ), and f2(θ+Δθ) is the corneal structural risk function value corresponding to the adjusted surgical parameters. This sub-function ensures that the surgical effect returns to the expected result by minimizing the deviation value δ(k), while strengthening the corneal structural safety constraint by introducing the safety weight coefficient λ, thus avoiding an increase in corneal structural risk due to parameter adjustment.

[0063] The surgical parameters that are dynamically adjusted in real time include laser energy density, cutting speed, and spot compensation amount: by adjusting the laser energy density and cutting speed, the changes in the cutting characteristics of corneal tissue are compensated, and by adjusting the spot compensation amount, the laser spot positioning deviation is corrected, so as to achieve precise dynamic adaptation of surgical parameters.

[0064] Step 5: Optimize Parameters, Verify, and Output

[0065] After the surgery, the patient's immediate postoperative eye data, including the immediate postoperative refractive error (D), was collected using a postoperative monitoring device. post Key indicators include residual corneal thickness and corneal stress. The immediate postoperative correction error is calculated as: ε = |D post -D target |; and verify whether the corneal structural parameters (remaining corneal thickness, corneal stress) meet the safety requirements (remaining corneal thickness ≥ T) safe Corneal stress ≤ σ safe ).

[0066] If ε≤0.5D and the corneal structure indicators meet the safety requirements, it indicates that the optimized surgical parameters have achieved the expected results. The final optimized parameters (including the initial surgical parameters and the intraoperative dynamic adjustment parameters) are output as the final parameter record for this surgery. If ε>0.5D or the corneal structure indicators do not meet the safety requirements, it indicates that the optimized parameters have not achieved the expected results. The immediate postoperative ocular data needs to be fed back to the individual ocular feature coupling model in step 2 to update and optimize the model parameters. Then, the process from step 3 to step 5 is repeated to re-optimize and verify the parameters until the parameters that meet the requirements are obtained.

[0067] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for automatically optimizing parameters in laser vision correction surgery, characterized in that, Includes the following steps: Step 1: Multi-dimensional ocular data acquisition and preprocessing: Collect multi-dimensional ocular data of the patient before surgery, including basic refractive data, corneal structure data, ocular surface morphology data and ocular physiological function data; perform noise reduction, outlier removal and standardization on the raw data to obtain standardized ocular data; Step 2: Construct an individual eye feature coupling model: Extract eye feature vectors based on standardized eye data, mine the nonlinear correlation of eye parameters in different dimensions through an improved neural network model, and output an individual eye feature coupling matrix; the improved neural network model adopts a structure combining convolutional layers, attention mechanisms and fully connected layers; Step 3: Initial Surgical Parameter Generation: A multi-objective optimization function is constructed based on the individual ocular feature coupling matrix. This function is solved using an improved particle swarm optimization algorithm to generate an initial surgical parameter set that satisfies safety constraints. The multi-objective optimization function is: minF(θ)=[f1(θ),f2(θ),f3(θ)] T ; Where θ is the surgical parameter vector, f1(θ) is the refractive correction error function, f2(θ) is the corneal structural risk function, and f3(θ) is the visual quality loss function; The improved particle swarm optimization algorithm introduces dynamic inertia weights, the expression of which is: Step 4: Real-time data acquisition and dynamic optimization during surgery: Acquire dynamic data of corneal ablation during surgery, and calculate the deviation value δ(k) = |X real (k)-X pred (k)|, if the deviation exceeds a preset threshold, the surgical parameters are dynamically adjusted in real time using a model predictive control algorithm; the model predictive control algorithm is based on a real-time optimization subfunction. Parameter optimization is achieved; where λ is the safety weight coefficient, with a value ranging from 0.6 to 0.

8. Step 5: Parameter Optimization, Verification, and Output: Collect real-time eye data post-surgery and calculate the real-time post-operative correction error ε = |D post -D target If ε≤0.5D and the corneal structure indicators meet the safety requirements, then the final optimized parameters are output. Otherwise, update the model and re-optimize it until the required parameters are obtained.

2. The automatic parameter optimization method for laser vision correction surgery according to claim 1, characterized in that: In step 1, the basic refractive data includes spherical power, cylindrical power, axis, and accommodative amplitude; the corneal structural data includes central corneal thickness, peripheral corneal thickness, corneal stroma thickness, and corneal endothelial cell density; and the ocular surface morphology data includes corneal curvature, corneal topography characteristic parameters, and tear film breakup time. The ocular physiological function data include intraocular pressure, pupil diameter for photopic / scotopic vision, and axial length.

3. The automatic parameter optimization method for laser vision correction surgery according to claim 2, characterized in that: In step 1, the standardization process employs the min-max standardization algorithm, the formula of which is: Where x represents the original data, x ′ For standardized data, x min x max These are the minimum and maximum values ​​of the data in this dimension, respectively.

4. The automatic parameter optimization method for laser vision correction surgery according to claim 3, characterized in that: In step 2, the eye feature vector includes basic features and derived features. The derived features are the coupling calculation results of the basic features, including the ratio of corneal thickness to refractive power, the matching coefficient of corneal curvature and pupil diameter, and the difference between corneal stroma thickness and estimated ablation depth.

5. The automatic parameter optimization method for laser vision correction surgery according to claim 4, characterized in that: In step 2, the input layer of the improved neural network model receives eye feature vectors; the convolutional layer uses three parallel convolutional kernels to perform multi-scale feature extraction on the eye feature vectors, obtaining local correlation features of different dimensional parameters; the attention mechanism assigns weights to the feature maps output by the convolutional layer through a spatial attention module, strengthening the weights of key features and weakening the interference of secondary features; the fully connected layer maps the features processed by the attention mechanism to the feature coupling space, outputting an individual eye feature coupling matrix, where the elements in the eye feature coupling matrix represent the coupling correlation strength between different eye parameters.

6. The automatic parameter optimization method for laser vision correction surgery according to claim 5, characterized in that: In step 3, f1(θ) = |D pred (θ)-D target |, where D pred (θ) represents the postoperative refractive error predicted based on parameter θ, D target The target refractive power.

7. The automatic parameter optimization method for laser vision correction surgery according to claim 6, characterized in that: In step 3, f2(θ) = max(0,T) safe -H resid (θ))+max(0,σ pred (θ)-σ safe ), where T safe H is the safe threshold for residual corneal thickness. resid (θ) represents the remaining corneal thickness after surgery, σ pred (θ) represents the predicted postoperative corneal stress, σ safe This is the corneal stress safety threshold.

8. The automatic parameter optimization method for laser vision correction surgery according to claim 7, characterized in that: In step 3, Where SA(θ) is the postoperative spherical aberration prediction value, SA pre Preoperative ball aberration measurement value, CS pre This represents the measured value of preoperative contrast sensitivity, CS pred (θ) represents the postoperative contrast sensitivity prediction value.

9. The automatic parameter optimization method for laser vision correction surgery according to claim 8, characterized in that: In step 4, the dynamic data of corneal ablation includes real-time corneal ablation depth, amount of corneal tissue removed, laser spot positioning accuracy, and patient eyeball micro-movement amplitude; the preset thresholds are ablation depth deviation ≥ 5 μm and spot positioning deviation ≥ 0.1 mm.

10. The automatic parameter optimization method for laser vision correction surgery according to claim 9, characterized in that: In step 4, the surgical parameters that are dynamically adjusted in real time include laser energy density, cutting speed, and spot compensation.

Images