A strabismus surgery risk data assessment method and system

Abstract

The application relates to the technical field of medical data processing, and discloses a strabismus operation risk data evaluation method and system. The method comprises the following steps: acquiring patient state data and a plurality of candidate operation parameter sets; performing quality processing on the patient state data, and extracting strabismus stability features based on the quality processing result; performing feature processing on the patient state data based on the quality processing result and the strabismus stability features, and generating a severity feature vector; inputting the strabismus stability features, the severity feature vector and the plurality of candidate operation parameter sets into a preset fusion model to obtain comprehensive risk results corresponding to the plurality of candidate operation parameter sets. The scheme solves the technical problems of single evaluation dimension of an operation scheme and incapability of quantitative comparison.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to a method and system for assessing the risk data of strabismus surgery. Background Technology

[0002] Strabismus is an ophthalmological condition caused by abnormal central control function or imbalance of extraocular muscle strength, resulting in the inability of both eyes to simultaneously fixate on the same target. Strabismus often occurs in childhood and can lead to amblyopia, impaired binocular vision, compensatory head posture, and psychosocial impairments, negatively impacting the patient's visual function and quality of life. Surgery is an important treatment for correcting eye alignment and improving binocular vision, but recurrence, overcorrection, undercorrection, muscle adhesions, and limited eye movement are still possible post-surgery. Currently, the development of strabismus surgical plans relies heavily on the surgeon's clinical experience and traditional surgical scales; different surgeons may make different assessments and surgical decisions for the same patient.

[0003] In recent years, artificial intelligence technology has been gradually applied to the field of strabismus diagnosis and treatment. For example, some studies have used machine learning methods to screen for risk factors related to postoperative recurrence of strabismus and to establish nomogram prediction models; others have used multi-task learning models to simultaneously predict the extraocular muscles requiring surgery and the corresponding amount of surgery. While these technologies have made some progress in assisting with the assessment of strabismus surgery, the following shortcomings still exist.

[0004] First, current strabismus surgery risk assessments are relatively singular in their dimensions, lacking a mechanism for simultaneously assessing multiple risks. Existing methods typically predict single outcomes such as postoperative recurrence or secondary surgery, for example, by using regression models, random forest models, or survival analysis models to screen relevant risk factors and output corresponding recurrence or secondary surgery risks. These methods struggle to simultaneously generate multiple risk information such as overcorrection risk, undercorrection risk, recurrence risk, muscle adhesion risk, and eye movement restriction risk, thus failing to comprehensively characterize the potential postoperative risks associated with the surgical plan.

[0005] Secondly, existing technologies lack a mechanism for quantitatively comparing multiple candidate surgical options for the same patient. Current surgical assistance design methods typically output suggested surgical approaches, target extraocular muscles, or suggested surgical amounts directly based on the patient's preoperative data. For example, multi-task learning models predict extraocular muscle selection and surgical dosage in horizontal strabismus surgery, or machine learning models predict surgical correction based on preoperative features. These methods primarily generate single surgical recommendations and do not conduct separate risk assessments for multiple candidate surgical parameter sets for the same patient, making it difficult to quantitatively compare the risk differences corresponding to different candidate surgical parameter sets.

[0006] Secondly, the quality information of clinical data is not fully utilized. Patient status data may contain missing data, abnormal values, and inconsistencies between different records. Existing methods typically only fill in missing data, without uniformly testing the completeness, validity, and consistency of data for different clinical characteristics. Furthermore, they do not use the data quality information corresponding to different clinical characteristics in the subsequent generation of patient status characteristics. This results in clinical data of varying quality potentially participating in risk assessment in the same way, affecting patient status representation and risk prediction results.

[0007] In addition, some deep learning-based strabismus assessment models directly output prediction results based on patients' preoperative data, which makes it difficult to reflect the relationship between different patient status characteristics and specific surgical parameters. The formation process of the model output results is not intuitive enough, which is not conducive to analyzing the impact of different patient status factors and surgical planning factors on risk assessment results.

[0008] Therefore, there is a need for a technical solution that can comprehensively process patient status data, including data quality, strabismus stability, and severity, in order to conduct risk assessment of candidate surgical parameters. Summary of the Invention

[0009] In view of this, the present invention proposes a method and system for assessing the risk data of strabismus surgery.

[0010] One aspect of this invention provides a method for assessing the risk data of strabismus surgery, specifically including the following steps: S1, acquire patient status data and a set of candidate surgical parameters; S2, perform quality processing on the patient status data, and extract strabismus stability features based on the quality processing results; S3, Based on the quality processing results and the strabismus stability characteristics, perform feature processing on the patient status data to generate a severity feature vector; S4. Input the strabismus stability features, the severity feature vector, and each of the candidate surgical parameter sets into a preset fusion model to obtain the comprehensive risk result corresponding to each of the candidate surgical parameter sets.

[0011] In some implementations, in step S1, the patient status data consists of several types of clinical features, including strabismus type identification features, strabismus angle measurement features, binocular vision function parameter features, eye movement function grading features, compensatory head position angle features, previous treatment record features, extraocular muscle imaging measurement parameter features, ocular and systemic comorbidity information features, and time-series measurement features consisting of strabismus degree measurement data collected at multiple preoperative time points. The candidate surgical parameter set consists of several types of surgical planning features, including surgical eye-specific features, target extraocular muscle identification features, surgical method type features, surgical operation value features, and combined surgical instruction features.

[0012] In some implementations, step S2 includes: Pre-defined tests are performed on various clinical features, including integrity testing, value range testing, and data consistency testing. Missing data obtained from the integrity detection are filled in, and abnormal data obtained from the value range detection are masked to obtain quality-processed patient status data. Based on the test results, the data quality values ​​corresponding to each of the clinical features are determined to form a data quality vector; The patient status data after quality processing and the data quality vector are used as the quality processing result.

[0013] In some implementations, step S2 further includes: Based on the temporal measurement characteristics in the patient status data after quality processing, the central tendency, dispersion, range of change, and trend of change of strabismus over measurement time are determined. The strabismus stability features are generated based on the central tendency, the degree of dispersion, the range of variation, and the trend of variation.

[0014] In some implementations, step S3 includes: Based on the strabismus type identification characteristics, determine the strabismus angle conversion parameters, binocular vision function conversion parameters, stability conversion parameters, and comorbidity conversion parameters; Based on the strabismus angle conversion parameter, the binocular vision function conversion parameter, the stability conversion parameter, and the comorbidity conversion parameter, the strabismus stability feature, the strabismus angle measurement feature, the binocular vision function parameter feature, and the ocular and systemic comorbidity information feature in the quality-processed patient status data are mapped to obtain initial scores including initial strabismus angle score, initial binocular vision function score, initial stability score, and initial comorbidity score. The preset weights corresponding to each initial score are adjusted according to the data quality vector to determine the target weights corresponding to each initial score; The initial scores are weighted and fused according to their corresponding target weights to generate the severity feature vector.

[0015] In some implementations, step S4 includes: The strabismus stability feature, the severity feature vector and each of the candidate surgical parameter sets are interactively encoded to obtain the candidate interactive features corresponding to each of the candidate surgical parameter sets. Each of the candidate interaction features is input into a preset fusion model, which includes a first layer base model set and a second layer meta-model. Based on each preset base model in the first layer base model set and each candidate interaction feature, the recurrence risk probability, complication risk probability, candidate surgery score and feature importance ranking results corresponding to each candidate surgical parameter set are obtained; Based on the recurrence risk probability, the complication risk probability, the candidate surgery score, the feature importance ranking result, and the second hierarchical model, the comprehensive risk result corresponding to each candidate surgery parameter set is obtained.

[0016] In some implementations, step S4 further includes: The recurrence risk prediction results, complication risk prediction results, feature importance sequence, and surgical necessity benchmark score are normalized to obtain normalized recurrence risk value, normalized complication risk value, normalized feature importance value, and normalized surgical necessity value. The normalized recurrence risk value, the normalized complication risk value, the normalized feature importance value, and the normalized surgical necessity value are concatenated to form a meta-model input feature vector; The input feature vector of the meta-model is input into the second layer meta-model to obtain the comprehensive risk result.

[0017] In some implementations, in step S4, the preset fusion model includes a first-layer base model set and a second-layer meta-model. The first-layer base model set includes an XGBoost model, a LightGBM model, a random forest model, and a logistic regression model. The second-layer meta-model includes a logistic regression meta-model. The comprehensive risk result includes the risk level and risk ranking result corresponding to each of the candidate surgical parameter sets.

[0018] In another aspect of this invention, a strabismus surgery risk data assessment system is also provided, comprising: The acquisition unit is configured to acquire patient status data and a set of candidate surgical parameters. The quality processing unit is configured to perform quality processing on the patient status data and extract strabismus stability features based on the quality processing results. The feature processing unit is configured to perform feature processing on the patient state data based on the quality processing results and the strabismus stability features to generate a severity feature vector. The risk assessment unit is configured to input the strabismus stability features, the severity feature vector, and each of the candidate surgical parameter sets into a preset fusion model to obtain the comprehensive risk result corresponding to each of the candidate surgical parameter sets.

[0019] In another aspect of the present invention, a computer device is provided, comprising: at least one processor; and a memory storing a computer program executable on the processor, the computer program performing the steps of the method described above when executed by the processor.

[0020] In another aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method steps.

[0021] The present invention has at least the following beneficial technical effects: by acquiring patient status data and several candidate surgical parameter sets, after processing the patient status data for completeness, validity and consistency, strabismus stability features are extracted and severity feature vectors are generated. Then, each candidate surgical parameter set is interactively encoded with the patient status data and input into a preset fusion model. This enables the output of comprehensive risk assessment results, including recurrence risk, overcorrection risk, undercorrection risk and complication risk, for multiple candidate surgical plans for the same patient. This solves the technical problem that the prior art cannot perform multi-dimensional risk quantification and comparison of multiple candidate surgical plans under the same framework, and realizes the quantitative ranking and differentiated assessment of the risk levels of multiple plans. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart of an embodiment of the strabismus surgery risk data assessment method provided by the present invention; Figure 2 A schematic diagram of an embodiment of the strabismus surgery risk data assessment system provided by the present invention; Figure 3 A schematic diagram of the structure of an embodiment of the computer device provided by the present invention; Figure 4 This is a schematic diagram of an embodiment of the computer-readable storage medium provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to specific examples and the accompanying drawings.

[0025] It should be noted that all uses of "first" and "second" in the embodiments of the present invention are for the purpose of distinguishing two entities or parameters with the same name but different names. It is clear that "first" and "second" are only for the convenience of expression and should not be construed as limiting the embodiments of the present invention. Subsequent embodiments will not explain this in detail.

[0026] Based on the above objectives, a first aspect of the present invention proposes a method for assessing strabismus surgery risk data. For example... Figure 1 As shown, it includes the following steps: S1, acquire patient status data and a set of candidate surgical parameters; S2, perform quality processing on patient status data, and extract strabismus stability features based on the quality processing results; S3, based on the quality processing results and strabismus stability characteristics, perform feature processing on the patient status data to generate a severity feature vector; S4. Input the strabismus stability features, severity feature vectors and each candidate surgical parameter set into the preset fusion model to obtain the comprehensive risk result corresponding to each candidate surgical parameter set.

[0027] In some embodiments, in step S1, the patient status data consists of several types of clinical features, including strabismus type identification features, strabismus angle measurement features, binocular vision function parameter features, eye movement function grading features, compensatory head position angle features, previous treatment record features, extraocular muscle imaging measurement parameter features, ocular and systemic comorbidity information features, and time-series measurement features consisting of strabismus degree measurement data collected at multiple preoperative time points.

[0028] It should be noted that the strabismus type identification feature is used to indicate the type of strabismus to which the target object belongs. This strabismus type may include, but is not limited to, concomitant strabismus, paralytic strabismus, or restrictive strabismus. For example, the strabismus type identification feature can be stored in a coded form, such as "00" representing concomitant esotropia, "01" representing concomitant exotropia, "10" representing paralytic strabismus, and "11" representing restrictive strabismus.

[0029] The strabismus angle measurement characteristics include the horizontal, vertical, and rotational strabismus of the target object under far-field (e.g., 5m) and near-field (e.g., 33cm) fixation conditions, with the unit being prism diopters (Δ). The above strabismus data can be obtained by measuring with a synoptophore or the alternating prism occlusion method.

[0030] Binocular vision function parameters include simultaneous vision, stereopsis acuity, fusion range, and suppression status. Simultaneous vision, characterizing the ability of both eyes to simultaneously perceive the same target, can be obtained through synoptophore testing. Stereopsis acuity can be measured using the Titmus stereopsis chart or a synoptophore. Fusion range can be determined using a synoptophore fusion range test. Suppression status can be assessed using the Worth four-dot lamp test or the Bagolini linear lens test, and its value can be "no suppression," "monocular suppression," or "alternating suppression."

[0031] The grading characteristics of eye movement function include the degree of limitation in eye movement of the target subject in each diagnostic eye position. Eye positions include nine fixation directions: right-superior, right-lateral, right-inferior, frontal, top-lateral, bottom-lateral, left-superior, left-lateral, and left-inferior. The degree of limitation in eye movement can be quantified using a grading scale from -4 to 0, where 0 indicates normal movement, -1 indicates mild limitation, -2 indicates moderate limitation, -3 indicates severe limitation, and -4 indicates complete inability to reach the eye position. This data can be obtained through Hess screen or synoptophore examination.

[0032] Compensatory head tilt characteristics include the head tilt angle adopted by the target object to overcome diplopia or obtain best visual acuity. Specifically, it can include the amount of head tilt in the horizontal, vertical, and rotational directions, measured in degrees. The compensatory head tilt angle can be obtained by measuring a protractor or a three-dimensional motion capture system.

[0033] Previous treatment records include whether the target individual has received treatments such as corrective lenses, amblyopia training, botulinum toxin injections, or strabismus surgery, as well as the timing and frequency of each treatment. This data can be obtained through a medical record system.

[0034] The imaging measurement parameters of extraocular muscles include the muscle thickness, muscle attachment location, and orbital bony structure parameters of each extraocular muscle in the target subject. Among these, the muscle attachment location can be characterized by positional parameters such as the distance from the muscle attachment point to the limbus. These parameters can be obtained by computed tomography (CT) or magnetic resonance imaging (MRI), and the unit is millimeters.

[0035] Information on ocular and systemic complications includes whether the target subject has ocular complications such as amblyopia, anisometropia, or cataracts, as well as systemic complications such as neuropathy or thyroid eye disease. This information on complications can be obtained through clinical examination and medical records and stored in the form of classification tags or binary identifiers.

[0036] The temporal measurement features include strabismus measurement data of the target subject at multiple preoperative time points. These multiple preoperative time points can be determined based on clinical follow-up, such as corresponding to the initial consultation, follow-up consultation, and final preoperative assessment. Alternatively, other quantities and time intervals of measurement data can be used depending on the actual application scenario, such as at 6 months, 3 months, and 1 week preoperatively. Strabismus measurements of the target subject are collected at multiple preoperative time points under both near and far distances, and at various diagnostic eye positions, forming a strabismus time series. This temporal measurement feature is the foundational data source for subsequent extraction of strabismus stability features.

[0037] In some embodiments, in step S1, at least one candidate surgical parameter set is obtained, and each candidate surgical parameter set corresponds to an available surgical plan. Each candidate surgical parameter set consists of multiple types of surgical planning features. These multiple types of surgical planning features include surgical eye-specific features, target extraocular muscle identification features, surgical method type features, surgical operation value features, and combined surgical instruction features.

[0038] It should be noted that the surgical eye-specific feature indicates whether the surgery is performed on the left or right eye. The target extraocular muscle identifier feature indicates the extraocular muscle targeted by the surgery, such as the medial rectus, lateral rectus, superior rectus, inferior rectus, or oblique muscle. The surgical procedure type feature indicates the type of surgical operation, which may include muscle retraction, muscle shortening, muscle folding, or muscle transposition. The surgical operation amount feature indicates the specific surgical operation amount, such as the amount of muscle retraction or shortening, in millimeters. The combined surgical indication feature indicates whether the candidate surgical plan involves combined surgical operations on multiple extraocular muscles.

[0039] In some embodiments, the patient status data undergoes quality processing in step S2, which can be implemented in the following ways.

[0040] Pre-defined tests are performed on various clinical features, including completeness checks, value range checks, and data consistency checks. Completeness checks determine if any clinical feature is missing. If a feature is missing, a corresponding missing value is generated. For example, if a patient's binocular vision function parameters do not record stereoscopic acuity, the missing value for that feature is marked as "1" (indicating missing); otherwise, it is marked as "0" (indicating completeness). Value range checks determine if the values ​​of various clinical features are within a pre-defined physiological or measurement range. For example, the physiological range for horizontal strabismus is typically -90Δ to +90Δ. If a patient's recorded value is 120Δ, the value is considered abnormal, and an abnormality marker is generated. Data consistency checks determine whether different categories of clinical features conform to pre-defined medical logical constraints, i.e., whether there are contradictions between them. For example, if the strabismus type identifier indicates comitant esotropia, but the oculomotor function grading feature shows a difference in strabismus between left and right gaze exceeding 5Δ, then the two are considered inconsistent, and the corresponding consistency coefficient is reduced. For example, if the strabismus type identifier indicates paralytic strabismus, but the compensatory head position angle is recorded as 0°, then the data is judged as suspicious, and the consistency coefficient is reduced accordingly. The consistency coefficient is obtained by traversing multiple preset logical rule bases and calculating the ratio of the number of rules that pass the verification to the total number of rules.

[0041] Missing data obtained from integrity checks are imputed, and abnormal data obtained from range checks are masked to obtain quality-processed patient status data. Multiple imputation or mean substitution methods can be used for missing data imputation. Optionally, to distinguish them from true measurements, missing markers corresponding to various clinical features are retained after imputation, enabling subsequent processing steps to differentiate between true measurements and imputed estimates. Data marked as abnormal are not used as valid data in subsequent feature processing. For example, their values ​​can be set to null or assigned preset default values, while the abnormal markers are retained for subsequent weight adjustments.

[0042] Based on the test results, data quality values ​​corresponding to various clinical features are determined to form a data quality vector. For any clinical feature, its completeness score, validity score, and consistency coefficient are combined to obtain the corresponding data quality value. The completeness score is determined based on the completeness test results: it is 1 when all measurement items of the feature are completely recorded, 0 when some are missing, and 0 when all are missing. The validity score is determined based on the value range test results: it is the ratio of the number of measurement items in the feature that do not exceed the preset physiological or measurement range to the total number of items. The completeness score, validity score, and consistency coefficient are multiplied to obtain the data quality value of the clinical feature. When the completeness score or validity score is 0, the data quality value is directly set to 0. When the consistency coefficient is lower than the preset threshold, the data quality value of the feature is directly set to 0. The data quality values ​​corresponding to various clinical features are determined in the above manner, and the data quality values ​​of various clinical features are arranged in a preset order to form a data quality vector. The preset order corresponds to the arrangement order of various clinical features in the patient status data, ensuring that each data quality value in the data quality vector establishes a definite mapping relationship with its corresponding clinical feature. This data quality vector is used to correct the weights of each feature dimension in subsequent steps. The quality-processed patient status data and the data quality vector are combined as the quality processing result and output to subsequent steps.

[0043] In some embodiments, step S2 extracts strabismus stability features based on patient status data, which can be achieved in the following ways.

[0044] Specifically, the temporal measurement features in the quality-processed patient status data are sorted according to the acquisition time corresponding to each strabismus measurement data, resulting in a strabismus measurement sequence arranged in chronological order. Based on the sorted strabismus measurement sequence, the central tendency, dispersion, range of variation, and trend of strabismus with measurement time are determined. The central tendency represents the overall level of strabismus measurement data corresponding to multiple preoperative time points. The dispersion represents the fluctuation of each strabismus measurement data relative to the overall level. The range of variation represents the difference between the maximum and minimum measured values ​​in the strabismus measurement sequence. The trend of variation represents the direction and rate of change of strabismus with increasing measurement time, whether it increases, decreases, or remains stable.

[0045] Central tendency can be determined based on the average or median of strabismus measurements at multiple preoperative time points. For strabismus measurement sequences without obvious outliers, the average value can be used to characterize central tendency. For strabismus measurement sequences with large fluctuations in the measurement data but not yet identified as outliers, the median can be used to characterize central tendency to reduce the impact of individual deviations on central tendency.

[0046] The degree of dispersion can be determined based on the standard deviation, mean absolute deviation, or other statistics used to reflect the dispersion of the measured values ​​in the strabismus measurement sequence. When using the coefficient of variation (i.e., the ratio of the standard deviation to the mean) to characterize the degree of dispersion, it is not used when the value corresponding to the central tendency is close to zero, to avoid abnormally amplifying the degree of dispersion due to an excessively small denominator. In this case, the standard deviation or mean absolute deviation can be used as the degree of dispersion.

[0047] The range of variation can be obtained by determining the maximum and minimum measurements in the strabismus measurement sequence and calculating the difference between them. The range of variation reflects the maximum change in strabismus between multiple preoperative time points and can be used in conjunction with the degree of dispersion to distinguish between continuous small fluctuations and large changes at individual time points.

[0048] The trend of strabismus over measurement time can be determined by establishing a regression relationship between measurement time and corresponding strabismus measurement values. For example, by using the measurement time corresponding to each preoperative time point as the independent variable and the corresponding strabismus measurement value as the dependent variable, a trend parameter can be obtained to characterize the direction and rate of change of strabismus over time. A positive trend parameter indicates that the strabismus increases over time; a negative trend parameter indicates that the strabismus decreases over time; and a trend parameter close to a preset stable range indicates that the strabismus remains relatively stable across multiple preoperative time points.

[0049] When the strabismus angle measurement features include horizontal strabismus, vertical strabismus, and rotational strabismus, the central tendency, dispersion, range of variation, and trend of change can be determined for the strabismus measurement sequences in different directions, and the calculation results corresponding to different directions can be combined in a preset order. For strabismus directions not involved in the patient status data, corresponding invalidation labels can be set to prevent subsequent steps from performing feature transformation on the data in that direction.

[0050] For example, when the target subject has distance horizontal strabismus measurement data and near horizontal strabismus measurement data at multiple preoperative time points, distance horizontal strabismus measurement sequences and near horizontal strabismus measurement sequences can be constructed separately. The central tendency, dispersion, range of change and trend of change of the two sets of sequences can be calculated respectively. The difference between distance horizontal strabismus and near horizontal strabismus at the same preoperative time point can also be calculated to supplement the characterization of the changes in strabismus status of patients at different examination distances.

[0051] Strabismus stability features are generated based on the central tendency, dispersion, range of variation, and trend of strabismus diopter. Specifically, the calculation results can be combined in a preset order to form feature data describing the stability of strabismus status. Strabismus stability features include the central tendency reflecting the overall level of strabismus diopter, the dispersion and range of variation reflecting measurement fluctuations, and the trend of change reflecting the direction and speed of changes in the condition.

[0052] In some embodiments, when the number of valid preoperative time points in the time-series measurement features is insufficient to calculate the trend of change, the calculable central tendency, dispersion, and range of change are determined based on the existing valid measurement data, and a time point insufficiency flag is set. The time point insufficiency flag, as part of the strabismus stability features, is used to enable subsequent steps to identify that the strabismus stability features do not contain valid trends of change, avoiding the use of data to fill in non-existent temporal variation patterns.

[0053] In some embodiments, in step S3, the patient status data is processed based on the quality processing results and strabismus stability characteristics to obtain a severity feature vector, which can be achieved in the following ways.

[0054] Based on the identifying characteristics of strabismus types, conversion parameters for strabismus angle, binocular vision function, stability, and comorbidities are determined for each strabismus type. Each conversion parameter specifies the mapping method for the corresponding clinical features and their corresponding preset weights. These parameters can be determined based on historical case data and validated using preset clinical rules. For comitant strabismus, conversion parameters corresponding to strabismus angle and binocular vision function can be configured. For paralytic or restrictive strabismus, the corresponding conversion parameters can be adjusted based on the degree of eye movement restriction, compensatory head posture, or extraocular muscle imaging measurements. This avoids using the same feature conversion standards for different strabismus types.

[0055] Based on strabismus angle conversion parameters, binocular vision function conversion parameters, stability conversion parameters, and comorbidity conversion parameters, the strabismus angle measurement characteristics, binocular vision function parameter characteristics, strabismus stability characteristics, and ocular and systemic comorbidity information characteristics are mapped to obtain initial scores, including initial strabismus angle scores, initial binocular vision function scores, initial stability scores, and initial comorbidity scores. For example, each feature value (or feature vector) is nonlinearly transformed to convert it into a score value within the range of 0 to 100. The mapping can be implemented using piecewise linear functions, sigmoid functions, or lookup tables.

[0056] Taking strabismus angle measurement characteristics as an example, a segmented mapping rule can be preset: horizontal strabismus of 0~10Δ is mapped to 0~20 points, 10~20Δ to 21~50 points, 20~40Δ to 51~80 points, and above 40Δ to 81~100 points. The threshold of this segmented mapping rule differs for different types of strabismus. For example, the mapping threshold for paralytic strabismus is lower than that for concomitant strabismus, meaning that at the same degree, the initial angle score for paralytic strabismus is higher.

[0057] Taking binocular vision function parameters as an example, the mapping rules are as follows: if there is no stereopsis and there is monocular suppression, the mapping score is 81~100; if the stereopsis acuity is greater than 200 seconds arc and the fusion range is less than 3°, the mapping score is 51~80; if the stereopsis acuity is 60~200 seconds arc and the fusion range is 3°~5°, the mapping score is 21~50; if the stereopsis acuity is less than 60 seconds arc and the fusion range is greater than 5°, the mapping score is 0~20.

[0058] Taking the stability characteristic of strabismus as an example, the mapping rules are as follows: if the trend shows a regression slope greater than 2Δ / month and a large degree of dispersion (e.g., coefficient of variation greater than 30%), the mapping score is 81~100; if the trend shows a regression slope greater than 2Δ / month but a moderate degree of dispersion (e.g., coefficient of variation 15%-30%), or a trend shows a regression slope between -1 and +1Δ / month but a large degree of dispersion, the mapping score is 51~80; if the trend shows a regression slope between -1 and +1Δ / month and a small degree of dispersion (e.g., coefficient of variation less than 15%), but a large range of change (e.g., range greater than 15Δ) or a high level of central tendency (e.g., mean greater than 30Δ), the mapping score is 21~50; if the trend is stable, the degree of dispersion is small, the range of change is small (e.g., range less than 15Δ), and the level of central tendency is low (mean less than 30Δ), the mapping score is 0~20.

[0059] Taking the information characteristics of ocular and systemic complications as an example, if there are high-risk complications such as paralysis, limiting factors, or a history of previous surgery, the mapping score is 80-100. If there are no such high-risk factors, the mapping score is 0.

[0060] The preset weights for each initial score include: the preset weight for angle corresponding to the initial strabismus angle score, the preset weight for function corresponding to the initial score for binocular vision function, the preset weight for stability corresponding to the initial score for stability, and the preset weight for comorbidities corresponding to the initial score for comorbidities. The sum of these four preset weights is 100%. In a specific example, the preset weight for angle is 40%, the preset weight for function is 30%, the preset weight for stability is 20%, and the preset weight for comorbidities is 10%. These preset weights are baseline values ​​pre-set based on clinical big data, reflecting the basic contribution of each dimension to the assessment of strabismus severity.

[0061] When adjusting the preset weights corresponding to each initial score based on the data quality vector, the data quality values ​​of various clinical features are applied to the preset weights of their respective dimensions. For example, multiplying the data quality value corresponding to the strabismus angle measurement feature by the preset weight of the angle yields the target weight for the initial strabismus angle score; multiplying the data quality value corresponding to the binocular vision function parameter feature by the preset weight of the function yields the target weight for the initial binocular vision function score; multiplying the data quality value corresponding to the temporal measurement feature by the preset weight of stability yields the target weight for the initial stability score; and multiplying the data quality value corresponding to the ocular and systemic comorbidity information feature by the preset weight of comorbidities yields the target weight for the initial comorbidity score. This adjustment ensures that feature dimensions with higher data quality have a greater weight in the severity score, while the weight of feature dimensions with lower data quality is correspondingly reduced. The sum of the adjusted target weights can be normalized to 100%.

[0062] Each initial score is weighted and fused according to its corresponding target weight to generate a severity feature vector. The specific calculation method for the weighted fusion is as follows: Severity Feature Vector Value = Initial Angle Score × Angle Target Weight + Initial Functional Score × Functional Target Weight + Initial Stability Score × Stability Target Weight + Initial Comorbidity Score × Comorbidity Target Weight. The weighted fusion result is a value between 0 and 100, with a higher value indicating a higher degree of severity.

[0063] In some embodiments, in step S4, the strabismus stability features, severity feature vectors, and each candidate surgical parameter set are input into a preset fusion model to obtain the comprehensive risk result corresponding to each candidate surgical parameter set. This can be achieved in the following ways.

[0064] The strabismus stability features, severity feature vectors, and candidate surgical parameter sets are interactively encoded to obtain candidate interactive features corresponding to each candidate surgical parameter set. In one example, various clinical features in the patient status data are encoded into a patient feature sequence. Specifically, the quality-processed patient status data includes multiple clinical feature values ​​after missing value imputation and anomaly masking. For each clinical feature, its feature value is input into the corresponding feature encoding network, and a nonlinear transformation is performed through a multi-layer fully connected network to output a fixed-dimensional feature vector. The feature vectors corresponding to various clinical features are arranged in a preset order to form the patient feature sequence Z=[z1, z2, ..., z...]. K[K], where K is the number of clinical feature categories. When generating patient feature sequences, the corresponding data quality values, missing data markers, and anomaly markers are also input into the feature encoding network to ensure the encoding results reflect the data quality of each feature. Various surgical planning features in each candidate surgical parameter set are encoded into corresponding candidate surgical parameter feature vectors. For any candidate surgical parameter set, its included surgical eye-specific features, target extraocular muscle identifier features, surgical method type features, surgical operation value features, and combined surgical instruction features are concatenated to form the original surgical parameter feature vector. Categorical features such as surgical eye-specific features, target extraocular muscle identifiers, and surgical method types are encoded using one-hot encoding or embedding layers; numerical features such as surgical operation values ​​are directly used. This original surgical parameter feature vector is input into the surgical parameter encoding network, which performs a nonlinear transformation through a multi-layer fully connected network to output a fixed-dimensional candidate surgical parameter feature vector, denoted as h. o,j Query mapping, key mapping, and value mapping are performed on the candidate surgical parameter feature vectors and each patient feature unit, respectively: q j =W Q h o,j ; k k =W K z k ; v k =W V z k ; Among them, W Q W K and W V q represents the feature mapping parameters determined through model training. j and k k Having the same feature dimension v k This makes the scaling dot product operation mathematically computable.

[0065] Calculate the scaled dot product between the query vector and each key vector to obtain the initial association value. Obtain the data quality value corresponding to the k-th basic clinical feature in the data quality vector, correct the initial association value based on the data quality value, and normalize the corrected association value to obtain the association weight.

[0066] Based on the association weights corresponding to various basic clinical features, the value vectors are weighted and aggregated to obtain the patient aggregated features: ; in, This is used to characterize the patient status information participating in risk assessment under the j-th candidate surgical parameter set. This represents the association weight.

[0067] The candidate surgical parameter feature vector h o,j Patient aggregation characteristics c j Strabismus stability characteristics V sta and severity feature vector V sev By concatenating the parameters, we obtain the interaction features corresponding to the j-th candidate surgical parameter set: H j =[h o,j , c j V sta V sev ] The above processing is performed on each candidate surgical parameter set to obtain the interaction features corresponding to each candidate surgical parameter set. This allows multiple candidate plans for the same patient to form corresponding interaction features, thereby enabling risk comparison among multiple plans.

[0068] In this way, the data quality vector generated in step S2 participates in the interactive encoding in step S4. By adjusting the association weights, the data quality value controls the contribution of various clinical features in the generation of patient aggregated features, forming a complete data processing closed loop from data quality detection to feature encoding.

[0069] Each candidate interaction feature is input into a preset fusion model, which includes a first-layer base model set and a second-layer meta-model.

[0070] It should be noted that the preset fusion model is a pre-trained machine learning model. XGBoost (ExtremeGradient Boosting), LightGBM (Light Gradient Boosting Machine), Random Forest, and Logistic Regression models serve as the first-layer base models. Each base model independently receives candidate interaction features as input and outputs its corresponding intermediate prediction results. The second-layer logistic regression meta-model receives the intermediate prediction results from each base model as input and outputs a comprehensive risk result.

[0071] Based on each preset base model in the first-layer base model set and each candidate interaction feature, the recurrence risk probability, complication risk probability, candidate surgical score, and feature importance ranking results corresponding to each candidate surgical parameter set are obtained.

[0072] The XGBoost model is used to perform nonlinear feature interaction mining on candidate interactive features and output recurrence risk prediction results. XGBoost is an ensemble learning algorithm based on gradient boosting decision trees. By building decision trees layer by layer and fitting the residuals of the previous tree, it can effectively capture the nonlinear interaction relationships between features. In the strabismus surgery risk assessment scenario, the recurrence risk prediction results are represented by probability values ​​between 0 and 1, with higher values ​​indicating a greater likelihood of postoperative recurrence.

[0073] The LightGBM model is used to perform high-dimensional feature operations on candidate interaction features and output complication risk prediction results. LightGBM is an efficient ensemble learning algorithm based on the gradient boosting framework. It employs a histogram-based decision tree algorithm and one-sided gradient sampling technique, which can efficiently process high-dimensional feature data. In the strabismus surgery risk assessment scenario, the complication risk prediction results are represented by probability values ​​between 0 and 1. The higher the value, the greater the likelihood of postoperative complications (including but not limited to muscle adhesion and limited eye movement).

[0074] Random forest models are used to combat overfitting of candidate interaction features, outputting a feature importance sequence. Random forest is an ensemble learning algorithm based on decision trees, which reduces the risk of overfitting by constructing multiple decision trees and employing a random subspace method. In the strabismus surgery risk assessment scenario, the feature importance sequence indicates the degree of contribution of each dimension of the candidate interaction features to the prediction result, and is sorted from highest to lowest contribution to enable interpretability analysis of the model.

[0075] Logistic regression models are used to determine the linear baseline for candidate interaction features and output a baseline score for surgical necessity. Logistic regression is a generalized linear model that outputs a probability value between 0 and 1 through linear weighting and a logistic function mapping. In the strabismus surgery risk assessment scenario, the baseline score for surgical necessity is represented by a value between 0 and 1, with a higher value indicating a greater necessity for performing the surgery.

[0076] The outputs of each base model serve as intermediate predictions, which are then fed into the second-level meta-model for fusion. Based on the recurrence risk probability, complication risk probability, candidate surgical scores, feature importance ranking, and the second-level meta-model, a comprehensive risk result corresponding to each candidate surgical parameter set is obtained. In one example, the second-level meta-model outputs a comprehensive risk value in the range of 0 to 1 by applying linear weighting and nonlinear mapping to the input risk probabilities. Since each base model predicts the risk of the same candidate procedure from different perspectives, and the second-level meta-model determines the weights of each base model's output in the fusion process through training, the generated comprehensive risk value integrates the judgments of multiple base models. The lower the comprehensive risk value, the lower the comprehensive risk corresponding to the candidate surgical parameter set; the higher the comprehensive risk value, the higher the comprehensive risk corresponding to the candidate surgical parameter set. It should be noted that the comprehensive risk value is a comprehensive evaluation index obtained by fusing the intermediate prediction results output by each base model and does not represent the actual probability of occurrence of a specific complication.

[0077] Based on the comprehensive risk value corresponding to each candidate surgical parameter set, the risk level and risk ranking result for each candidate surgical parameter set are determined. The risk level is determined based on the numerical range of the comprehensive risk value. The boundary values ​​of the numerical range are predetermined based on the risk value distribution of the training samples, and different threshold settings can be used for different application scenarios. The risk level can be divided into three levels: low risk, medium risk, and high risk, or it can be set to two or five levels according to clinical needs. All candidate surgical parameter sets are ranked according to the comprehensive risk value corresponding to each candidate surgical parameter set to obtain the risk ranking result. The ranking method is as follows: the lower the comprehensive risk value of the candidate surgical parameter set, the higher its risk ranking, indicating that the comprehensive risk of the plan is lower and it can be given priority in clinical decision-making; the higher the comprehensive risk value of the candidate surgical parameter set, the lower its risk ranking, indicating that the comprehensive risk of the plan is higher and it should be considered with caution. The feature importance ranking result, the probability of each risk, the comprehensive risk value, the risk level, and the risk ranking result are used as the comprehensive risk result to provide risk comparison information between different candidate surgical parameter sets, and are not directly used to determine the final surgical plan.

[0078] The feature importance ranking results serve as auxiliary information for model interpretability and are not included in the fusion input of the second-level meta-model. For example, for a candidate surgical procedure, the feature importance ranking output by the random forest model might show: "preoperative horizontal strabismus" ranked first, "binocular vision function" ranked second, and "surgical workload" ranked third. Based on this, doctors can understand the main risk drivers of the candidate procedure and pay special attention to these high-risk factors before surgery.

[0079] In some embodiments, when the data quality vector of a candidate surgical parameter set contains features below a preset threshold, a data quality warning is included in the risk assessment result. For example, "Note: The binocular vision function parameter feature data on which this scheme is based has low quality, and the corresponding risk prediction results are for reference only," to remind clinicians to carefully interpret prediction results with insufficient data quality.

[0080] In addition, the risk assessment results are output in a structured data format, which can be displayed on the screen of electronic devices or stored and archived as part of an electronic medical record system for clinicians to refer to when developing surgical plans.

[0081] In summary, by merging multiple risk probabilities into a comprehensive risk value, and further outputting risk levels and risk ranking results, the risk ranking results can intuitively reflect the risk differences between various candidate surgical parameter sets, enabling clinicians to compare and weigh multiple optional surgical options according to their risk levels. Since the risk predictions for each candidate option are based on the same patient status data and employ a unified fusion model, the evaluation results between different options are comparable, overcoming the technical limitations of traditional methods that rely on single, individual predictions and cannot provide quantitative comparisons between options.

[0082] In one specific embodiment, data from 2167 desensitized strabismus cases across three ophthalmology centers from 2018 to 2024 were used for model training and internal validation. Of these, 1517 cases served as the training set, 650 as the internal validation set, and 420 external multicenter cases served as the external test set. The samples covered concomitant strabismus, paralytic strabismus, and restrictive strabismus, and included strabismus degree, strabismus type, age of onset, duration of disease, binocular vision, best corrected visual acuity, refractive status, oculomotor function and AV signs, compensatory head posture, past treatment history, extraocular muscle imaging parameters, and information on ocular and systemic complications. Categorical features were encoded, and continuous features were normalized.

[0083] Model training was performed on an in-house server equipped with a 24-core processor and 64GB of memory. Python 3.9, Scikit-learn, XGBoost, and LightGBM were used for data preprocessing, training and inference of the first-layer base model set and the second-layer meta-model. After training, a lightweight inference version for outpatient terminals was obtained through model pruning and quantization compression.

[0084] In one severity calculation example, a 10-year-old patient with constant comitant exotropia had a strabismus angle of 45Δ, impaired binocular vision, a disease duration of 6 years, and amblyopia and mild compensatory head posture. Assuming the data quality across all dimensions met preset requirements, the sub-scores for strabismus angle, binocular vision function, stability, and comorbidities were 34, 18, 14, and 7, respectively, corresponding to a total severity score of 73. The total severity score, along with the sub-scores, was used to construct a severity feature vector, which served as input for subsequent risk assessment.

[0085] On an internal validation set of 650 cases, ROC (Receiver Operating Characteristic) analysis was performed with a strabismus degree of not less than 10Δ at 6 months post-surgery as the criterion for positive recurrence. The ROC-AUC (Receiver Operating Characteristic - Area Under Curve) for predicting recurrence risk using the pre-defined fusion model was 0.92. The recurrence risk prediction AUC of 0.74 using only the three basic features of age, strabismus degree, and visual acuity, and the recurrence risk prediction AUC of 0.83 using the ablation model without incorporating the strabismus type-corresponding transformation parameter and severity feature vector, was 420 cases. The recurrence risk prediction AUC of 0.89 on the external multicenter test set of 420 cases, the AUC of 0.87 on the complication prediction task on the internal validation set, and the Kappa (a statistical indicator for evaluating the consistency of two classification results) between the candidate surgery score and the surgical necessity label provided by senior ophthalmologists were 0.81.

[0086] For the aforementioned 10-year-old patient, the model outputs a recurrence risk probability of 62%, a complication risk probability of 31%, and a candidate surgical score of 84. It can also output key feature information ranked by importance. The risk assessment results can be recorded in structured data format, including patient baseline information, severity characteristics, various risk probabilities, candidate surgical scores, and feature importance ranking results, and can be saved, printed, or encrypted and archived.

[0087] In another example, a 7-year-old patient with intermittent exotropia had a strabismus angle of 15Δ, with largely preserved stereopsis and no amblyopia or compensatory head posture. After the aforementioned processing, the total severity score was 22, and the model output a candidate surgery score of 18, a recurrence risk probability of 11%, and a complication risk probability of 7%. In yet another example, a 27-year-old patient with restrictive strabismus had thyroid-associated ophthalmopathy, previous failed strabismus correction surgery, large-angle strabismus, and significant eye movement restriction. After the aforementioned processing, the total severity score was 88, and the model output a candidate surgery score of 92, a recurrence risk probability of 76%, and a complication risk probability of 58%, and also output important characteristics such as previous failed strabismus surgery, restrictive oculomotor disease, and severe visual impairment.

[0088] Based on the same inventive concept, according to another aspect of the present invention, such as Figure 2 As shown, embodiments of the present invention also provide a strabismus surgery risk data assessment system, characterized in that it includes: The acquisition unit 110 is configured to acquire patient status data and a set of candidate surgical parameters; The quality processing unit 120 is configured to perform quality processing on patient status data and extract strabismus stability features based on the quality processing results. The feature processing unit 130 is configured to perform feature processing on patient status data based on quality processing results and strabismus stability features to generate a severity feature vector. The risk assessment unit 140 is configured to input strabismus stability features, severity feature vectors and each candidate surgical parameter set into a preset fusion model to obtain the comprehensive risk result corresponding to each candidate surgical parameter set.

[0089] Based on the same inventive concept, according to another aspect of the present invention, such as Figure 3 As shown, an embodiment of the present invention also provides a computer device 30, which includes a processor 310 and a memory 320. The memory 320 stores a computer program 321 that can be run on the processor. When the processor 310 executes the program, it performs the steps of the method described above.

[0090] Based on the same inventive concept, according to another aspect of the present invention, such as Figure 4 As shown, embodiments of the present invention also provide a computer-readable storage medium 40, which stores a computer program 410 that, when executed by a processor, performs the methods described above.

[0091] Finally, it should be noted that those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium for the program can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. The above computer program embodiments can achieve the same or similar effects as any of the corresponding foregoing method embodiments.

[0092] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the functionality of various illustrative components, blocks, modules, circuits, and steps has been generally described. Whether this functionality is implemented as software or as hardware depends on the specific application and the design constraints imposed on the system as a whole. Those skilled in the art can implement the functionality in various ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the embodiments disclosed herein.

[0093] The above are exemplary embodiments disclosed in this invention. However, it should be noted that various changes and modifications can be made without departing from the scope of the embodiments of this invention as defined by the claims. The functions, steps, and / or actions of the methods according to the disclosed embodiments described herein do not need to be performed in any particular order. The sequence numbers of the disclosed embodiments of this invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. Furthermore, although the elements disclosed in the embodiments of this invention may be described or claimed individually, they may be understood as multiple unless explicitly limited to a singular number.

[0094] It should be understood that, as used herein, the singular form “a” is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, “and / or” refers to any and all possible combinations of one or more of the associated listed items.

[0095] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of the invention, technical features of the above embodiments or different embodiments can be combined, and many other variations of different aspects of the invention exist, which are not provided in the details for the sake of brevity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A method for assessing the risk data of strabismus surgery, characterized in that, Includes the following steps: S1, acquire patient status data and a set of candidate surgical parameters; S2, perform quality processing on the patient status data, and extract strabismus stability features based on the quality processing results; S3, Based on the quality processing results and the strabismus stability characteristics, perform feature processing on the patient status data to generate a severity feature vector; S4. Input the strabismus stability features, the severity feature vector, and each of the candidate surgical parameter sets into a preset fusion model to obtain the comprehensive risk result corresponding to each of the candidate surgical parameter sets.

2. The method for assessing strabismus surgery risk data according to claim 1, characterized in that, In step S1, the patient status data consists of several types of clinical features, including strabismus type identification features, strabismus angle measurement features, binocular vision function parameter features, eye movement function grading features, compensatory head position angle features, previous treatment record features, extraocular muscle imaging measurement parameter features, ocular and systemic comorbidity information features, and time-series measurement features consisting of strabismus degree measurement data collected at multiple preoperative time points. The candidate surgical parameter set consists of several types of surgical planning features, including surgical eye-specific features, target extraocular muscle identification features, surgical method type features, surgical operation value features, and combined surgical instruction features.

3. The method for assessing strabismus surgery risk data according to claim 2, characterized in that, Step S2 includes: Pre-defined tests are performed on various clinical features, including integrity testing, value range testing, and data consistency testing. Missing data obtained from the integrity detection are filled in, and abnormal data obtained from the value range detection are masked to obtain quality-processed patient status data. Based on the test results, the data quality values ​​corresponding to each of the clinical features are determined to form a data quality vector; The patient status data after quality processing and the data quality vector are used as the quality processing result.

4. The method for assessing strabismus surgery risk data according to claim 3, characterized in that, Step S2 also includes: Based on the temporal measurement characteristics in the patient status data after quality processing, the central tendency, dispersion, range of change, and trend of change of strabismus over measurement time are determined. The strabismus stability features are generated based on the central tendency, the degree of dispersion, the range of variation, and the trend of variation.

5. The method for assessing strabismus surgery risk data according to claim 3 or 4, characterized in that, Step S3 includes: Based on the strabismus type identification characteristics, determine the strabismus angle conversion parameters, binocular vision function conversion parameters, stability conversion parameters, and comorbidity conversion parameters; Based on the strabismus angle conversion parameter, the binocular vision function conversion parameter, the stability conversion parameter, and the comorbidity conversion parameter, the strabismus stability feature, the strabismus angle measurement feature, the binocular vision function parameter feature, and the ocular and systemic comorbidity information feature in the quality-processed patient status data are mapped to obtain initial scores including initial strabismus angle score, initial binocular vision function score, initial stability score, and initial comorbidity score. The preset weights corresponding to each initial score are adjusted according to the data quality vector to determine the target weights corresponding to each initial score; The initial scores are weighted and fused according to their corresponding target weights to generate the severity feature vector.

6. The method for assessing strabismus surgery risk data according to claim 2, characterized in that, Step S4 includes: The strabismus stability feature, the severity feature vector and each of the candidate surgical parameter sets are interactively encoded to obtain the candidate interactive features corresponding to each of the candidate surgical parameter sets. Each of the candidate interaction features is input into a preset fusion model, which includes a first layer base model set and a second layer meta-model. Based on each preset base model in the first layer base model set and each candidate interaction feature, the recurrence risk probability, complication risk probability, candidate surgery score and feature importance ranking results corresponding to each candidate surgical parameter set are obtained; Based on the recurrence risk probability, the complication risk probability, the candidate surgery score, the feature importance ranking result, and the second hierarchical model, the comprehensive risk result corresponding to each candidate surgery parameter set is obtained.

7. The method for assessing strabismus surgery risk data according to claim 6, characterized in that, In step S4, the first layer base model set includes the XGBoost model, the LightGBM model, the random forest model, and the logistic regression model; the second layer meta-model includes the logistic regression meta-model; and the comprehensive risk result includes the risk level and risk ranking result corresponding to each of the candidate surgical parameter sets.

8. A strabismus surgery risk data assessment system, characterized in that, include: The acquisition unit is configured to acquire patient status data and a set of candidate surgical parameters. The quality processing unit is configured to perform quality processing on the patient status data and extract strabismus stability features based on the quality processing results. The feature processing unit is configured to perform feature processing on the patient state data based on the quality processing results and the strabismus stability features to generate a severity feature vector. The risk assessment unit is configured to input the strabismus stability features, the severity feature vector, and each of the candidate surgical parameter sets into a preset fusion model to obtain the comprehensive risk result corresponding to each of the candidate surgical parameter sets.

9. A computer device, comprising: At least one processor; as well as A memory storing a computer program executable on the processor, characterized in that the processor executes the program by performing the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it performs the steps of the method as described in any one of claims 1 to 7.

Images