High-precision three-dimensional cornea vertex positioning method and system based on image analysis

Abstract

The invention provides a high-precision three-dimensional cornea vertex positioning method and system based on image analysis, and belongs to the technical field of image processing, and the method specifically comprises the following steps: determining a positioning processing result of a three-dimensional cornea vertex by using a positioning processing method, so as to determine the three-dimensional cornea vertex in different users; on the basis of the deviation condition of positioning processing results among different positioning processing methods, multiple positioning demand features in cornea image features are determined, and available matching combinations in user combinations are determined according to positioning processing data of the multiple positioning demand features; and determining an optimization processing strategy of the three-dimensional cornea vertex of the user according to the positioning processing data of the three-dimensional cornea vertex when it is determined that optimization processing of the positioning processing strategy needs to be carried out by combining the positioning processing data of the user and the multi-positioning demand feature data of different positioning verification processing methods. And the positioning processing accuracy of the cornea vertex is improved.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a high-precision three-dimensional corneal vertex localization method and system based on image analysis. Background Technology

[0002] The corneal apex is the point on the anterior surface of the cornea with the greatest curvature and most prominent features. It is usually close to the visual axis and is a key reference point in ophthalmic surgeries (such as femtosecond laser flap creation and ICL implantation). It is also the core benchmark for refraction and corneal topography analysis. Its positioning accuracy directly determines the safety and effectiveness of the surgery.

[0003] When performing corneal vertex localization, auxiliary camera devices are often positioned in a fixed location. However, due to differences in the image characteristics of the user's cornea, the measurement results of the corneal vertex may not be accurate. Therefore, determining a method for identifying and processing the corneal vertex of auxiliary camera devices at multiple locations under different image characteristics, and selecting the optimal setting position of the auxiliary camera device under different corneal image characteristics to ensure the accuracy of the localization results, has become an urgent technical problem to be solved.

[0004] Therefore, there is an urgent need for a high-precision three-dimensional corneal vertex localization method and system based on image analysis. Summary of the Invention

[0005] To achieve the objectives of this invention, the following technical solution is adopted: Specifically, this application provides a high-precision three-dimensional corneal vertex localization method based on image analysis, which includes: S1 uses multi-dimensional image acquisition data and image analysis model to perform three-dimensional corneal vertex localization processing. Based on the verification processing data of localization processing results of different users, and combined with the extraction results of corneal image features of users, the localization processing method of the auxiliary camera device for the three-dimensional corneal vertex of the user is determined. S2 uses the aforementioned positioning processing method to determine the positioning processing result of the three-dimensional corneal vertex, and based on the deviation of the positioning processing results between different positioning processing methods in different users, determines the multiple positioning requirement features in the corneal image features. S3 determines the available matching combinations in the user combination based on the positioning processing data of multiple positioning requirement characteristics, and combines the positioning processing data of users in different positioning verification processing methods and the multiple positioning requirement characteristic data to determine when the positioning processing strategy needs to be optimized. Based on the positioning processing data of the three-dimensional corneal vertex, S3 determines the optimization processing strategy of the user's three-dimensional corneal vertex.

[0006] The beneficial effects of this invention are as follows: In this application, a method for locating the auxiliary camera device at the 3D corneal vertex of a user is determined. The most reliable and efficient auxiliary camera device positioning strategy is intelligently selected for each new user to improve positioning accuracy, shorten calibration time, and ensure the clinical reliability of the results. By analyzing the corneal image features and successful positioning data of historical users, users with highly similar corneal morphological features are grouped into the same "user group." Furthermore, based on the number of reliable positioning samples accumulated within different groups, the stability and reliability of the positioning results for this characteristic group are judged. Finally, based on the "experience level" of the characteristic group to which the new user belongs, the auxiliary camera device positioning method is dynamically determined—whether to use a mature, pre-set standard position scheme or to activate a more complex and flexible multi-position verification scheme—thereby achieving personalized and precise positioning support for different characteristic user groups.

[0007] In this application, based on the localization data of the 3D corneal vertex, an optimized processing strategy for the user's 3D corneal vertex is determined. First, an initial screening is conducted based on the "new position" of each combination. High-risk combinations (i.e., combinations with a high degree of fit using the new position) undergo a full upgrade, utilizing multiple positions for corneal vertex localization. For medium-risk combinations, further consideration is given to their internal validation practices (the proportion of other localization processes) and combination size, dynamically deciding whether to upgrade all combinations, upgrade some samples, or maintain the status quo. Ultimately, this achieves: comprehensive validation of combinations with reliable methods that can be switched to the new position (high fit coefficient); and careful validation and incremental optimization of potential problems (medium fit coefficient).

[0008] Furthermore, the multidimensional image is at least two images from different sources that contain the corneal region.

[0009] Furthermore, the localization processing data based on the corneal image features is determined according to the number of users undergoing localization processing based on the corneal image features.

[0010] Furthermore, the method for determining the positioning processing method of the auxiliary camera device for the user's three-dimensional corneal vertex is as follows: Based on the location processing data of different users, users whose corneal image features meet the similarity requirements are grouped into the same user group; Based on the distribution data of the user combinations, determine the number of users for location processing in different user combinations; Based on the extraction results of the user's corneal image features and the number of users in different user combinations for localization processing, the localization processing method of the auxiliary camera device for the user's three-dimensional corneal vertex is determined.

[0011] Further, determine whether the positioning processing strategy needs to be optimized, specifically including: Based on the available matching combination data in the user combination, determine the number of available matching combinations in the user combination; Based on the positioning data of users under different positioning verification processing methods, it is determined that in different user combinations, two positions are used to determine the location of the auxiliary camera device for three-dimensional corneal vertex and the positioning process, and this is used as another positioning process. Based on the number of available matching combinations, other location process data in different user combinations, and the multi-location demand characteristic data, it is determined whether the location processing strategy needs to be optimized.

[0012] In a second aspect, the present invention provides a computer system comprising: a memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-described high-precision three-dimensional corneal vertex localization method based on image analysis when running the computer program.

[0013] Other features and advantages will be set forth in the following description, and the objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0015] The above and other features and advantages of the present invention will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart of a high-precision three-dimensional corneal vertex localization method based on image analysis; Figure 2 This is a flowchart illustrating the method for determining the positioning processing method of the auxiliary camera device for the user's three-dimensional corneal vertex; Figure 3 This is a flowchart illustrating the method for determining multiple localization requirement features in corneal image features; Figure 4 This is a flowchart illustrating the method for determining whether an optimization of the positioning processing strategy is needed. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0018] Example 1 like Figure 1 As shown, this application provides a high-precision three-dimensional corneal vertex localization method based on image analysis, specifically including: S1 uses multi-dimensional image acquisition data and image analysis model to perform three-dimensional corneal vertex localization processing. Based on the verification processing data of localization processing results of different users, and combined with the extraction results of corneal image features of users, the localization processing method of the auxiliary camera device for the three-dimensional corneal vertex of the user is determined. S2 uses the aforementioned positioning processing method to determine the positioning processing result of the three-dimensional corneal vertex, and based on the deviation of the positioning processing results between different positioning processing methods in different users, determines the multiple positioning requirement features in the corneal image features. S3 acquires positioning processing data with multiple positioning requirements in real time, and combines it with the positioning processing data of users under different positioning verification processing methods. When it is determined that the positioning processing strategy needs to be optimized, the optimization processing strategy for the user's three-dimensional corneal vertex is determined based on the positioning processing data of the three-dimensional corneal vertex.

[0019] Furthermore, the multidimensional image is at least two images from different sources that contain the corneal region.

[0020] Furthermore, the localization processing data based on the corneal image features is determined according to the number of users undergoing localization processing based on the corneal image features.

[0021] Specifically, such as Figure 2 As shown, the method for determining the positioning processing method of the auxiliary camera device for the user's three-dimensional corneal vertex is as follows: In scenarios involving 3D corneal vertex localization using multi-source images (at least two), the most reliable and efficient auxiliary camera positioning strategy is intelligently selected for each new user to improve positioning accuracy, shorten calibration time, and ensure the clinical reliability of the results. Its core logic is "group characteristic similarity guiding individual positioning strategy." That is, by analyzing the corneal image features and successful positioning data of historical users, users with highly similar corneal morphological features are grouped into the same "user group." Then, based on the number of reliable positioning samples accumulated within different groups, the stability and reliability of the positioning results for that characteristic group are judged. Finally, based on the "experience level" of the new user's characteristic group, the auxiliary camera positioning method used is dynamically determined—whether to use a mature, pre-set standard positioning scheme or to activate a more complex and flexible multi-position verification scheme—thereby achieving personalized and precise positioning support for different characteristic user groups.

[0022] The specific method for positioning using a fundus camera and auxiliary imaging device is as follows: In the coaxial region of the illumination and imaging light paths of the fundus camera, a miniature coaxial structured light projection module is integrated. This module can project a set of known patterns (such as concentric rings, sine stripes, or pseudo-random spots) onto the anterior surface of the cornea.

[0023] High-precision binocular optical calibration of the integrated system: Using a calibration board, the internal and external parameters of the fundus camera's main imaging sensor and the auxiliary positioning camera located outside the camera and having a fixed spatial relationship with the main sensor are simultaneously calibrated, and the precise spatial transformation relationship between the two is established.

[0024] Synchronous acquisition and preprocessing of multi-view images.

[0025] The system is triggered to simultaneously acquire two images: the structured light reflection image I_main of the anterior corneal surface acquired by the main sensor of the fundus camera, and the oblique angle image I_slave of the frontal view of the eyeball (including the iris and sclera areas) acquired by the auxiliary positioning camera.

[0026] Image I_main is processed to suppress ring halo and restore reflective highlights; image I_slave is processed to segment the iris region and extract feature points (such as the limbus and iris texture).

[0027] Preliminary reconstruction of the three-dimensional anterior corneal surface based on monocular structured light.

[0028] In image I_main, the distortion feature points of the structured light pattern (taking concentric rings as an example) are extracted to obtain its sub-pixel image coordinates.

[0029] Based on the principle of phase measurement profilometry or a known projection geometry model, and combined with the calibration parameters in step S12, the two-dimensional distortion feature points in image I_main are mapped to a three-dimensional point cloud P_structured in the fundus camera coordinate system.

[0030] A robust surface fitting algorithm (such as RANSAC combined with B-spline surface fitting) is used to process the point cloud P_structured, filter out outliers, and generate a preliminary three-dimensional model M_initial of the anterior corneal surface.

[0031] Global pose optimization and refinement based on multi-view geometry.

[0032] Feature association: Project the feature points (such as the intersection of loops) on the model M_initial onto the auxiliary camera image I_slave, and match them with the corresponding features extracted from I_slave (such as the corneal limbus ellipse) to establish a multi-view feature correspondence.

[0033] Bundling adjustment: Using the parameters calibrated in step S12 as initial values, the pose of the fundus camera, the pose of the auxiliary camera, and the coordinates of the three-dimensional points on the corneal surface are jointly optimized to minimize the reprojection error of all feature points on the two images, resulting in the optimized high-precision three-dimensional corneal anterior surface model M_refined.

[0034] 3D corneal vertex calculation and output.

[0035] On the refined model M_refined, the vertex search region is defined with reference to the 3D estimated point of the pupil center calculated from the I_slave image.

[0036] Calculate the principal curvature of all points in the region, and determine the point with the maximum average curvature or the point with the maximum Gaussian curvature as the three-dimensional corneal vertex V_corneal.

[0037] Output the 3D coordinates of vertex V_corneal in the global coordinate system (usually with the auxiliary camera or a specific reference point as the origin), and calculate its offset relative to the view axis.

[0038] S11 uses location processing data from different users to group users whose corneal image features are similar enough to be grouped into the same user group. Based on the corneal image feature vectors of all historical users in the database, the feature similarity (such as cosine similarity) between any two users is calculated. Users with a similarity coefficient greater than a preset threshold (0.85) are grouped into the same user group.

[0039] A "user group" is a group of users who share highly similar corneal image features (such as curvature distribution, vertex region texture, and morphological contour). It represents a type of user with "family similarity" in corneal morphology.

[0040] This forms the basis for "experience reuse." The similarity in corneal morphology often implies commonalities in the representation of its three-dimensional vertices in images, its response to illumination and angle, and even the optimal auxiliary camera angle. Clustering users by features is equivalent to archiving historical positioning experience by "corneal type." This avoids erroneously applying successful experiences applicable to one type of cornea to another with a vastly different morphology, providing a scientific basis for subsequent group-based strategy decisions.

[0041] Specific example: The database contains 100 historical users. After feature extraction and cluster analysis, five user groups were formed: Group A (feature "large curvature flat cornea", containing 35 people), Group B ("standard spherical cornea", containing 40 people), Group C ("small curvature steep cornea", containing 15 people), Group D ("irregular astigmatic cornea", containing 8 people), and Group E ("postoperative morphological cornea", containing 2 people). The features of the new user U_new were extracted, and the feature similarity coefficient with Group B was 0.92, while that with Group C was 0.78. Therefore, the user was assigned to Group B.

[0042] S12 determines the number of users in different user combinations for location processing based on the distribution data of the user combinations; For each user group identified in step S11, count the number of users within that group who have successfully completed 3D corneal vertex localization in the past, i.e., the "number of users who have completed localization processing".

[0043] "Number of users processed for localization" specifically refers to the total number of historical users within a given user group who have used the system and successfully obtained 3D corneal vertex localization results. This number is a core quantitative indicator for measuring the maturity and reliability of the localization scheme for this specific user group.

[0044] Simply grouping users is insufficient for decision-making; the "empirical value" of each group must be quantified. The more users successfully located within a group, the more thoroughly the system's pre-set positions and parameter optimizations for auxiliary imaging devices targeting that corneal morphology have been validated, resulting in higher reliability and universality. Conversely, for groups with small sample sizes, the "optimal" positioning strategy may not have been fully explored or validated. This step transforms qualitative feature classification into comparable quantitative empirical indicators.

[0045] Specific examples (continuous): Count the number of people who successfully positioned themselves in the above 5 combinations: combination A (35 people), combination B (40 people), combination C (15 people), combination D (8 people), and combination E (2 people).

[0046] S13 determines the positioning processing method of the auxiliary camera device for the three-dimensional corneal vertex of the user based on the extraction results of the corneal image features of the user and the number of users in different user combinations for positioning processing.

[0047] Specifically, if the similarity coefficient of corneal image features between a user and other users is greater than a preset similarity coefficient threshold, then the user and other users will be grouped into the same user group.

[0048] Furthermore, based on the extraction results of the user's corneal image features and the number of users undergoing localization processing in different user combinations, the localization processing method for the auxiliary camera device at the three-dimensional corneal vertex of the user is determined, specifically including: S131 Based on the extraction results of the user's corneal image features, determine the user group in which the user's corneal image features are located, and use it as a matching user group. Determine whether the number of users in the matching user group for positioning processing is less than a preset user number threshold. If so, the reliability of the user's three-dimensional corneal vertex positioning processing result is difficult to determine, so proceed to step S132. If not, determine that the positioning processing method of the user's three-dimensional corneal vertex auxiliary camera device is to determine the three-dimensional corneal vertex auxiliary camera device according to a preset position, and use the auxiliary camera device and the main camera device for positioning processing. The primary assessment is whether the team has sufficient experience. Identify the matching user combination of U_new (i.e., combination B). Determine whether the number of users for this combination to be located (40 people) is less than the preset user number threshold (20 people).

[0049] Keyword Explanation: "Matching User Group" refers to the historical user group to which a new user is grouped based on their similar characteristics, and it is the primary reference for the system to recommend positioning strategies for them.

[0050] This is the most direct and prioritized decision-making path. If the user's group already has a large number of successful cases (≥threshold), it means that the system is already very proficient in locating this type of cornea. Directly using the preset auxiliary camera device positions (relative positions and angles of the main and auxiliary cameras) optimized for this group for positioning is the most reliable and fastest choice. This reflects full trust in and utilization of mature experience.

[0051] The number of users in group B is 40, which is not less than the threshold of 20. Therefore, the condition "if" (i.e., the number < 20) is not met. The system directly determines: for U_new, "the auxiliary camera device for the three-dimensional corneal vertex is determined according to the preset position, and the positioning processing is performed using the auxiliary camera device and the main camera device". This means that the system will call the optimal camera position parameters preset for the "standard spherical cornea" group.

[0052] S132 determines the user combination for location processing based on the location processing data in different user combinations, and uses it as the location processing combination. It then determines whether the number of the location processing combinations is greater than a preset threshold for the number of location processing combinations. If so, the reliability of the location processing is relatively high. Therefore, the location processing method for determining the auxiliary camera device of the three-dimensional corneal vertex of the user is to determine the auxiliary camera device of the three-dimensional corneal vertex according to the preset position, and to perform location processing using the auxiliary camera device and the main camera device. If not, proceed to the next step. (Hypothetical scenario: If U_new is assigned to combination E, and its number of users is 2, which is less than the threshold of 20, then proceed to S132) Step S132: Secondary judgment – ​​Is the overall system experience extensive? When a user's combination lacks experience, broaden the scope and count the number of "location processing combinations" in the entire system where at least one user has been successfully located.

[0053] Even if a specific combination of samples is small, if the system as a whole has achieved success on many different types of corneas (a large number of localization processing combinations), it indicates that the system's localization algorithm and framework itself have strong robustness and adaptability. In this case, although there is a lack of precise presets for this specific subgroup, using the system's general preset location scheme still has a high probability of success. This is a trust based on the overall capabilities of the system.

[0054] Specific example (hypothetical): If U_new belongs to group E (2 people), the system enters S132. Assume that among the 5 groups, A, B, C, and D are all location processing groups, then the number of location processing groups is 5. Determine if it is greater than a preset threshold (e.g., 5). If 5 is not greater than 5 (usually judged as "equal to", not "greater than"), the condition is not met, and proceed to S133.

[0055] S133 Based on the number of users in different user combinations, determine whether the number of user combinations in the user combination whose number of users for positioning processing is not less than a preset number threshold (greater than the preset number of users threshold) is greater than a preset combination number threshold. If yes, determine that the positioning processing method of the auxiliary camera device for the three-dimensional corneal vertex of the user is to determine the auxiliary camera device for the three-dimensional corneal vertex according to the preset position, and use the auxiliary camera device and the main camera device for positioning processing. If no, proceed to step S134. Count how many user combinations have a number of users that is not less than the preset user number threshold (20 people), and determine whether their number is greater than the preset combination number threshold (3).

[0056] This step assesses whether the system has accumulated sufficient experience with most common corneal types. If the system already has mature solutions for multiple (more than the threshold) corneal types, it can be inferred that the system also possesses a certain methodology and adjustment capability for "how to find a suitable position for a newly emerging subtype". In this case, even when faced with a completely new or rare combination, using a standard preset solution and making minor adjustments may be more efficient than blindly trying multiple positions.

[0057] Specific example (hypothesis): Among the 5 groups, only group A (35 people) and group B (40 people) meet the condition of "number of located users ≥ 20", which is 2. Determine if 2 is greater than the preset threshold 3? 2 is not greater than 3, so the condition is not met. Proceed to the final step S134.

[0058] S134 takes user groups whose number of users in the positioning process is less than a preset user number threshold as verification deviation user groups, and determines the positioning processing method of the auxiliary camera device for the three-dimensional corneal vertex of the user based on the verification deviation user group data.

[0059] Furthermore, it is determined whether the number of the verification deviation user combinations is greater than a preset threshold for the number of deviation user combinations. If so, the first number of verification deviation user combinations with the fewest users is obtained in real time and used as the analysis user combination. If the user's matching user combination belongs to the analysis user combination, the positioning processing method for the auxiliary camera device of the user's three-dimensional corneal vertex is to determine the auxiliary camera device of the three-dimensional corneal vertex using two set positions, and to perform positioning processing using the auxiliary camera device and the main camera device. If not, the verification deviation user combination with the fewest users is obtained in real time and used as the analysis user combination. If the user's matching user combination belongs to the analysis user combination, the auxiliary camera device of the three-dimensional corneal vertex is determined using two set positions, and to perform positioning processing using the auxiliary camera device and the main camera device.

[0060] Final ruling – a special strategy for the “validation bias” group: Combinations with fewer than a preset user threshold (20 people) are labeled as validation bias combinations. This type of group has a scarce sample size, making the location experience unreliable.

[0061] Determine the size of the biased group: If the number of user combinations with biases is large (greater than the preset bias threshold, such as 2), it indicates that the system faces a large number of "experience blind spots".

[0062] Accurately identify the scarcest group: From all validation bias combinations, identify the first few combinations (e.g., 3) with the fewest users in real time as the user combinations for analysis. These are the most uncertain groups in the system's perception and require the most careful handling. Individualized decision: Check whether the matching user combination of the new user U_new belongs to this list of "scarce" user combinations for analysis.

[0063] If so: This indicates that the user belongs to the most uncertain "extreme" group in the system's perception. The most conservative yet comprehensive strategy must be adopted: use multiple settings to determine the auxiliary camera device (e.g., conduct trial shots and positioning calculations at several possible angles, and comprehensively evaluate the results) to cope with its high degree of uncertainty.

[0064] If the number of user combinations with verification deviations is not greater than the preset deviation threshold, such as 2, it means that although the user belongs to the group with insufficient experience, it is not the most extreme case. The user combination with the fewest verification deviations is used as the analysis user combination. If the user combination matched by the user belongs to the analysis user combination, the auxiliary camera device of the three-dimensional corneal vertex is determined by two set positions, and the auxiliary camera device and the main camera device are used for positioning processing.

[0065] Specific examples (assuming a final outcome): The user combinations that were found to be biased are: C (15 people), D (8 people), and E (2 people). The number is 3, which is greater than the bias threshold of 2.

[0066] The three combinations with the fewest users are: E (2 people), D (8 people), and C (15 people). Therefore, the user combinations to be analyzed are {E, D, C}.

[0067] User U_new (assuming it belongs to combination E) is matched with user combination E, which belongs to the analysis user combination list. For U_new, the system adopts the approach of "using two set positions to determine the auxiliary camera device for three-dimensional corneal vertex, and using the auxiliary camera device and the main camera device for positioning processing". The system will not rely on a single preset camera position, but will initiate a multi-angle acquisition and calculation process to cope with the high specificity of its corneal morphology (postoperative morphology) and the extreme lack of historical experience.

[0068] Furthermore, the positioning processing method divides the processing according to the position of the auxiliary camera device.

[0069] Specifically, such as Figure 3 As shown, the method for determining the multiple localization requirement features in the corneal image features is as follows: By reverse-engineering the discrepancies in historical corneal localization results, common corneal morphological features that cause the failure of standard localization methods (preset auxiliary camera positions) are analyzed and identified, and these features are defined as "multiple localization requirement features." The core logic is "tracing the cause from the effect, extracting preventative knowledge from failures." Specifically, by comparing the results obtained by the same user using different localization methods (preset location vs. multiple locations), and using authoritative verification results as a benchmark, typical cases where "the standard method yields an error while the multiple location method yields a correct result" are identified. Furthermore, through cluster analysis of a large number of such cases, highly similar morphological features in the corneal images of the users involved are identified. Finally, these features are established as "multiple localization requirement features." The core value lies in the fact that once the system detects that a new user's cornea possesses these features, it can bypass potentially failed preset location attempts and directly initiate a more reliable multiple location localization scheme, thereby achieving an intelligent leap from "post-event remediation" to "pre-event prediction."

[0070] S21 determines the location processing process with inconsistent location processing results based on the deviation between the location processing methods among the users, and identifies it as an inconsistent process. For each historical user, the "location processing results" (i.e., three-dimensional coordinates) obtained from all the different location processing methods used are compared. If the deviation between the results of different methods for the same user exceeds the allowable tolerance range, the location event is determined to be an "inconsistent process".

[0071] A “contradictory process” specifically refers to a complete location event occurring on the same individual where the output results are significantly contradictory due to the use of different location techniques (mainly “preset location” and “other location” methods). It indicates that the standard process may have encountered a problem in this particular case.

[0072] This is the starting point for identifying the problem. Inconsistent results act like a "crack," revealing discrepancies between different methods for this specific corneal morphology. Capturing these "cracks" provides the data foundation for subsequent root cause analysis, aiming to shift the focus of analysis from a large number of success stories to those questionable cases that warrant further investigation.

[0073] Specific example: User U123 uses the positions of two auxiliary cameras. The first position yields vertex coordinates P1, and the second position yields vertex coordinates P2. Calculations show that the spatial distance deviation between P1 and P2 is 0.15mm, exceeding the system tolerance of 0.05mm. Therefore, U123's positioning is marked as an "inconsistent process."

[0074] S22 uses the location processing method corresponding to the location processing result that is consistent with the verification processing result during the inconsistency process as the matching verification method; For each "inconsistent process," a "verification processing result" (gold standard V) is introduced as arbitration. The deviations of P1 and P2 from V are calculated separately. The positioning method corresponding to the result with the smaller deviation from V is determined as the "matching verification method" for that inconsistent process.

[0075] The "matching verification method" is the positioning method that is proven to be more accurate and reliable in a specific case by independent and authoritative evidence (the gold standard) when positioning discrepancies occur. It answers the question, "When method A and method B conflict, which one should be believed this time?"

[0076] It should be noted that the gold standard is as follows: On the 3D model M_front, the vertex V_candidate is identified. V_candidate is then projected onto the corneal cross-sectional image to verify whether it lies within the tangent region of the vertex of the corneal anterior surface segmentation curve. Simultaneously, it is checked whether the normal direction at this point deviates from the estimated visual axis direction within a threshold. If the verification passes, V_candidate is output as the final 3D corneal vertex coordinates.

[0077] This is a crucial step in establishing causal relationships. Simply knowing the results are inconsistent is insufficient; it's essential to determine which side is correct. By introducing a gold standard for adjudication, the "inconsistent process" can be clearly categorized into two types: one where the pre-defined location method is incorrect while the multi-location method is correct (indicating a flaw in the pre-defined location); and the other where the opposite is true. Only the first type provides valuable samples for identifying "multi-location requirement characteristics," as it directly demonstrates that the standard method fails for certain corneas.

[0078] User U123's verification results show that P2 (the second position) passed the test as an accurate positioning result. Therefore, the result for the second position is closer to the gold standard. Thus, in this inconsistency process, the "matching verification method" was determined to be the "second position".

[0079] S23 determines the multiple localization requirement features in the corneal image features based on the inconsistent processes and matching verification method data among different users.

[0080] It is understandable that, based on inconsistent processes and matching verification method data across different users, the multiple localization requirement features in the corneal image features are determined, specifically including: S231 uses the inconsistent process of matching verification method that does not belong to the determination of the three-dimensional corneal vertex auxiliary camera device according to the preset position as the identification matching process, and judges whether there is no identification matching process in different users. If so, it is determined that all user combinations do not belong to the multiple positioning requirement feature, that is, it is not necessary to switch the auxiliary camera device from the current preset position to other positions. If not, proceed to step S232. Traverse all inconsistent processes and mark the process where the "matching verification method" is "the auxiliary setting device does not use the preset position method, that is, the second position method" (i.e. "does not belong to the preset position...") as "identification matching process".

[0081] This is the core filter of the entire analysis. It precisely filters out the types of cases we are most interested in—cases where the standard method (pre-set position) fails, while the alternative (second position) succeeds. These cases provide direct evidence that "certain corneal characteristics render the pre-set position method unsuitable." If there are no such cases historically, it indicates that the pre-set position method is highly universal and there is no need to define new requirements.

[0082] Specific example (continuous): There are 10 cases marked as "inconsistent process" in the system. Upon verification, the "matching verification method" of 6 of these cases is "the position determined by the second position". These 6 processes are marked as "identification matching process". Since the identification matching process exists, the condition "none exist" is not met, and proceed to step S232.

[0083] S232 takes the user group with the identification and matching process as the potential matching user group, and determines whether the proportion of the potential matching user group in the user group is greater than the preset potential matching group proportion threshold. If so, the user group with the location processing result obtained at the preset location is inconsistent with the verification processing result is taken as the basis, and the corneal image features with similarity coefficients greater than the preset similarity thresholds of different users in the user group are taken as the multiple positioning requirement features. If not, proceed to step S233. Mark all user combinations in which at least one user has participated in the "identification and matching process" as "potentially matched user combinations". Calculate the proportion of these combinations to the total number of user combinations.

[0084] This step assesses whether the "preset location failure" problem is a widespread phenomenon across multiple groups or limited to only a few. If the percentage is high (above the threshold), it indicates a common problem, potentially affecting multiple corneal types where standard methods are inapplicable. In this case, common features should be extracted from all combinations that have encountered the problem (i.e., "user combinations with inconsistent positioning results at the preset location compared to the verification results") to define a broader range of "multiple positioning requirement features."

[0085] Assume there are 5 user combinations (A, B, C, D, E). Among them, users in combination B (keratoconus) and combination C (postoperative cornea) have undergone the "identification and matching process". Therefore, the potential matching user combinations are {B, C}, with a quantity of 2, accounting for 2 / 5 = 40%. Assuming the preset threshold is 40%, the percentage equals the threshold (not triggering "greater than"), the condition is not met, and proceed to step S233.

[0086] S233 determines the fit coefficient of the potential matching user combination based on the proportion of users in the potential matching user combination who have an identification matching process, and determines whether there is a potential matching user combination whose fit coefficient is greater than a preset fit coefficient threshold. If so, the user combination with inconsistent positioning processing results and verification processing results at a preset location is used as the basis, and corneal image features with similarity coefficients greater than a preset similarity thresholds to users different in the user combination are used as multiple positioning requirement features. If not, the user combination with inconsistent positioning processing results and verification processing results at a preset location and with an identification matching process is used as the basis, and corneal image features with similarity coefficients greater than a preset similarity thresholds to users different in the user combination are used as multiple positioning requirement features.

[0087] When a problem exists but is not widespread, a more detailed analysis of its concentration within the group is needed.

[0088] Calculate the fit coefficient: For each potential matching user combination, calculate its fit coefficient = (number of users within the combination who have undergone the "identification and matching process") / (total number of users within the combination). This coefficient reflects the degree of "incompatibility" or "rejection" of the standard method within the combination.

[0089] Hierarchical decision-making: If the fit coefficient of a certain combination exceeds the preset fit coefficient threshold (60%), it indicates that the preset location method is highly likely to fail within this specific combination. In this case, common corneal features of all users should be extracted based on this combination (requiring a feature similarity of >0.88 between any two users) as "multiple localization requirement features". This defines a strongly correlated, high-risk feature set.

[0090] If the fit coefficients of all combinations do not exceed the threshold, it indicates that the problem exists but is not concentrated. In this case, a more conservative approach should be taken, extracting features only from specific users who "have both experienced inconsistencies in the preset location results and actually participated in the identification and matching process." This defines a feature set with slightly weaker relevance but solid evidence.

[0091] Specific examples (final decision): Calculate the matching coefficient for group B (keratoconus): Group B consists of 20 people, of whom 3 have gone through the "identification and matching process", so the matching coefficient = 3 / 20 = 15%.

[0092] Calculate the fit coefficient of combination C (postoperative cornea): There are 10 people in combination C, of ​​whom 7 have gone through the "identification and matching process", and the fit coefficient = 7 / 10 = 70%.

[0093] The fit coefficient of combination C (70%) is greater than the preset threshold (60%), so the condition is met.

[0094] Final decision: Adopt the first refined strategy. Based on combination C (postoperative cornea), analyze the corneal image features of all 10 users within this combination, and identify a subset of common features where the similarity between any two users is higher than 0.88 (e.g., "linear scar at the superior limbus", "abrupt curvature gradient at the apex exceeds a certain threshold"), etc. These common features are formally defined as "multiple localization requirement features".

[0095] This embodiment of the "Multiple Positioning Requirement Feature Determination Method" constructs a "Reverse Diagnosis and Knowledge Discovery Based on Evidence Chain" framework, the core value of which is: This represents a fundamental shift from "empirical avoidance" to "feature-based prediction": traditional methods might only know that "postoperative corneal localization is difficult," but not the specific reasons for the difficulty. This method, through rigorous causal analysis (inconsistency -> verification -> matching), precisely pinpoints the specific morphological features that cause standard methods to fail. This allows the system to directly detect these features in new user images, thus making the optimal strategy selection before the initial localization, transforming a passive approach into a proactive one.

[0096] A data-driven, interpretable expert knowledge base was established: the final output of "multi-location requirement features" is no longer a vague experience description, but a quantifiable set of image features calculated based on the common similarity of the group. It is transformed into explicit rules that the algorithm can recognize and execute, which greatly reduces the system's dependence on the operator's personal experience.

[0097] This enhances the overall system performance and robustness, forming a self-evolutionary closed loop: by applying this feature library, the system can effectively avoid ineffective preset position attempts for high-risk corneas, saving time and directly adopting a more successful multi-position approach, thus improving the success rate and efficiency of initial localization. Simultaneously, newly added case data will continuously feed back into this analysis process, continuously validating and enriching the "multi-positioning requirement feature" library, enabling the system to possess self-learning and evolutionary capabilities, and adapt to constantly emerging new and complex corneal morphologies.

[0098] Specifically, the available matching combination is a user combination with an adaptation coefficient greater than a preset adaptation coefficient threshold, wherein the adaptation coefficient is the proportion of users in the user combination who have undergone the identification and matching process.

[0099] Specifically, such as Figure 4 As shown, determining whether the positioning processing strategy needs optimization includes: Based on the identification of "available matching combinations" (user groups with high fit coefficients, i.e., those prone to failure of preset methods) and "multiple positioning needs characteristics" (corneal morphological characteristics that easily lead to the failure of preset methods), the effectiveness and risks of existing positioning strategies are comprehensively evaluated, and intelligent decisions are made on whether a comprehensive optimization of the preset position strategy of the auxiliary camera device in the entire system is necessary. Its core logic is "quantifying the scale of risk, assessing the sufficiency of evidence, and judging the urgency of optimization." That is, by analyzing the number of high-risk groups, the high-risk range based on feature identification, and the scale and frequency of cases validated using multi-position methods, a comprehensive judgment is made on: 1) whether the potential benefits of optimization (switching preset positions) are large enough; 2) whether existing data is sufficient to support a reliable optimization decision; and 3) if data is insufficient, whether optimization should be proactively initiated to collect key data. Ultimately, a prudent, data-driven decision is made between "maintaining the status quo" and "initiating systemic optimization."

[0100] S31 uses the available matching combination data in the user combination to determine the number of available matching combinations in the user combination; Count the number of user combinations that are marked as "available matching combinations" among all current user combinations.

[0101] The number of available matching combinations directly reflects the number of user groups in the system that have been confirmed by empirical data as having a "high risk of failure with the preset location method." This number is the most direct indicator of the "vulnerability" of the current preset location strategy.

[0102] This is the primary basis for optimization decisions. If this number is 0, it means there is no conclusive evidence that any user's pre-defined method poses a systemic risk, and blind optimization may bring unnecessary costs and uncertainties. Therefore, a number of 0 is a strong reason for "no optimization needed," and the decision-making process ends here. Conversely, it indicates the existence of a "problem group" that needs attention.

[0103] Specific example: Based on the premise, the only available matching combination is combination C, so the quantity is 1. Since there is an available matching combination (quantity > 0), proceed to the next step of judgment.

[0104] S32 determines, based on the user's positioning processing data under different positioning verification processing methods, the positioning processing process of determining the auxiliary camera device for the three-dimensional corneal vertex using two set positions in different user combinations, and uses it as other positioning processes; Statistics show that in each user group, the historically actual positioning process of "determining and locating the auxiliary camera device for three-dimensional corneal vertex at two set positions" (i.e., the multi-position method) was used, and these processes were marked as "other positioning processes".

[0105] "Other positioning processes" refer to the verification or exploratory operations that the system takes when facing positioning difficulties or uncertainties, which differ from the standard single preset location. They represent the additional efforts and historical experience that the system has already put in to solve the positioning problem.

[0106] This data is crucial for assessing whether alternative solutions exist to partially solve the problem and whether the verification data is sufficient. If many combinations frequently use the multi-location method, it indicates that the existing preset locations may no longer be fully trusted, and the system has undergone "silent optimization" in practice. At the same time, the data generated by these processes is also a valuable reference for evaluating whether new preset locations are better.

[0107] Specific examples (continuous): Statistics on the occurrence of "other positioning processes" for each combination: Combination A (Standard cornea): In 100 localization attempts, the multi-position method was used twice.

[0108] Combination B (keratoconus): In 15 out of 80 localization attempts, the multi-position method was used.

[0109] Combination C (postoperative cornea): In 40 out of 50 localization attempts, the multi-position method was used.

[0110] Combinations D and E: rarely used (assuming each is used less than 5 times).

[0111] S33 determines whether to optimize the positioning processing strategy based on the number of available matching combinations, other positioning process data in different user combinations, and the multi-positioning demand characteristic data.

[0112] It should be noted that if no matching combination is available, then it is determined that there is no need to switch the setting position of the existing auxiliary camera device, and therefore no optimization of the positioning processing strategy is required.

[0113] Additionally, it's understandable that if available matching combinations exist, the following situations also apply: Scenario 1: If the number of available matching combinations is greater than the preset threshold for the number of available matching combinations, then by switching the setting position of the existing auxiliary camera device, the positioning optimization needs of many existing user combinations can be met. Therefore, it is determined that the positioning processing strategy needs to be optimized to improve the reliability of positioning processing using the auxiliary camera device and the main camera device in different user combinations by using the switched setting position.

[0114] If the number of available matching combinations is greater than 2 (large-scale risk), it can be directly determined that optimization of the positioning processing strategy is required.

[0115] When the number of high-risk groups (with available matching combinations) is large (>2), it indicates that the preset location strategy has significant shortcomings on multiple corneal types. In this case, the potential benefits of comprehensive optimization (finding and switching to a new, more universally applicable preset location or set of preset locations) are substantial, potentially benefiting multiple user groups at once and significantly improving overall positioning reliability and efficiency. This is a decision based on the principle that "the problem is widespread and must be addressed."

[0116] Case 2: If the number of available matching combinations is not greater than the preset threshold for the number of available matching combinations, then the multi-location requirement feature data is obtained, and the user combination belonging to multiple location requirement features is taken as the location optimization combination. If the proportion of the location optimization combination in the user combination is greater than the preset combination proportion threshold, then the number of user combinations that can be identified by the reliability of location processing after switching is large, so it is determined that no location processing strategy optimization is needed. Currently, the number of available matching combinations is 1 (≤2), proceed to this branch.

[0117] First, obtain the "location optimization combination" (a combination with multiple location requirement characteristics). In this example, it is combination C. Calculate its proportion among all user combinations: 1 (combination C) / 5 (total combinations) = 20%.

[0118] If the proportion of location optimization combination is greater than 30%, it is determined that no location processing strategy optimization is needed.

[0119] If the system has already identified a significant proportion (>30%) of users through feature recognition that they have multiple location needs, this means the system is highly aware of these groups and has already adopted targeted strategies such as multiple location access. In this case, the marginal benefits of comprehensively optimizing preset locations may be limited, and it might even disrupt established response processes tailored to specific characteristics. Maintaining the status quo and focusing on the precise management of identified high-risk groups may be a better choice.

[0120] Case 3: If the proportion of the optimized positioning combination in the user combination is not greater than the preset combination proportion threshold, the proportion of the user combination with other positioning processes in the total user combination is determined by using other positioning process data in different user combinations. When the proportion of the user combination with other positioning processes in the total user combination is less than the preset user combination proportion threshold, the verification sufficiency is not high, so it is determined that the positioning processing strategy needs to be optimized. If the proportion of users with optimized location combinations is ≤ 30%, and the proportion of users with other location processes is < 50%, the strategy is to determine which location processing strategy needs to be optimized.

[0121] In this example, the location optimization combination ratio is 20% (≤30%). Next, the ratio of "user combinations with other location processes" is calculated: among combinations A, B, C, D, and E, A, B, C, D, and E have all had "other location processes", so the ratio is 5 / 5 = 100%, which is much greater than 50%. The condition of "less than 50%" is not met, so case 3 is not triggered.

[0122] Case 4: When the proportion of user groups with other positioning processes in all user groups is not less than the preset user group proportion threshold, the positioning processing matching factor is determined by the average of the proportion of user groups with other positioning processes in all user groups and the average of the proportions of other positioning processes in all positioning processing processes in user groups with other positioning processes. Based on the positioning processing matching factor, it is determined whether the positioning processing strategy needs to be optimized.

[0123] It should be noted that if the location processing matching factor is less than the preset matching factor threshold, it is determined that the location processing strategy needs to be optimized, thereby increasing the proportion of other location processes in some user combinations, so as to determine as soon as possible whether the setting position of the auxiliary camera device needs to be switched, so as to improve the accuracy of the location processing.

[0124] The proportion of users with optimized location combinations ≤ 30%, and the proportion of users with other location processes ≥ 50% If the current condition is met (20%≤30% and 100%≥50%), proceed to the final detailed calculation of case 4.

[0125] Calculate the location processing matching factor: Factor 1: Proportion of user combinations with other location processes = 100% = 1.0, Factor 2: For each combination with other location processes, calculate the "proportion of other location processes" (number of times the multi-location method is used in this combination / total number of location times), and then calculate the average of these proportions.

[0126] Combination A: 2 / 100 = 0.02, Combination B: 15 / 80 ≈ 0.19, Combination C: 40 / 50 = 0.80, Combination D: Assume 3 / 60 = 0.05, Combination E: Assume 1 / 70 ≈ 0.014, Average = (0.02+0.19+0.80+0.05+0.014) / 5 ≈ 0.215.

[0127] The matching factor for location processing = (factor1 + factor2) / 2 = (1.0 + 0.215) / 2 = 0.6075.

[0128] Decision: Determine if the factor is less than the preset matching factor threshold (0.4). 0.6075 is not less than 0.4. Therefore, the condition "if less, then optimization is needed" is not true. Logically, when the factor is not less than the threshold, it means the system has already undergone relatively extensive validation (factor 1 is high), and among those validated combinations, the multi-position method also occupies a certain usage proportion (factor 2 is not extremely low), indicating that the overall validation sufficiency is acceptable. At this point, the system may determine that a global, disruptive strategy optimization is not temporarily necessary, but continuous monitoring is still possible. In the branch description, if "if not," it indicates that the existing validation data does not strongly suggest that immediate optimization is required, therefore, optimization is not currently needed.

[0129] Specific example (final ruling): Based on the above, this example triggers condition 4, and the calculated location processing matching factor is 0.6075, which is greater than the threshold of 0.4. Therefore, the system ultimately determines that: there is currently no need to optimize the global location processing strategy (preset position of the auxiliary camera device).

[0130] This embodiment of the "optimization decision-making" method constructs a prudent optimization decision-making framework that "balances risk, evidence, and cost": This approach avoids blind and excessive technological iteration: instead of immediately demanding "optimization" every time a problem is discovered, it establishes a multi-layered evaluation standard (number of at-risk groups, scope of feature identification, and sufficiency of validation data). This effectively prevents the hasty initiation of costly and highly uncertain system-level parameter adjustments (such as recalibrating the preset positions of all devices) due to a few isolated cases or insufficient data, ensuring the robustness of technological changes.

[0131] This achieves optimal resource allocation driven by data: the decision-making logic treats "optimization" as an action requiring resource investment (time, computing power, and manpower). Through quantitative analysis, optimization is triggered only when one of the following conditions occurs: ① The risk scale is large (condition 1), and the benefits are clear; ② Validation data is severely insufficient (condition 3), requiring proactive exploration to obtain a basis for decision-making. In other cases (such as when risks have been accurately managed through feature identification, or validation data indicates that the problem is not urgent), the status quo or partial adjustments are chosen, thereby concentrating valuable R&D and operation resources on the most profitable or uncertainties that need to be clarified.

[0132] This fosters the co-evolution of system cognition and decision-making maturity: by introducing a comprehensive indicator such as the "location processing matching factor," the system is not only making binary decisions about "whether to optimize," but also assessing its own cognitive confidence in the effectiveness of the current strategy. A low factor indicates vague cognition and insufficient decision-making basis; in this case, proactive optimization (actually proactive exploration) is a necessary step to improve cognition. A high factor indicates relatively clear cognition, enabling more prudent decisions. This endows the system with advanced capabilities of self-cognition and prudent decision-making, marking its evolution from a simple "execution-feedback" cycle to a more complex "perception-evaluation-decision-learning" intelligent agent.

[0133] Specifically, the method for determining the optimization strategy for the user's three-dimensional corneal vertex is as follows: Based on the grouping of users with different corneal types (user combinations) and the understanding of their historical positioning performance, a refined strategy is intelligently formulated for each user combination regarding "whether and how" to upgrade the positioning method, in order to systematically improve the reliability and efficiency of 3D corneal vertex positioning. Its core logic is "tiered optimization based on historical risk." This method first performs an initial screening based on the "fit coefficient" (preset historical failure rate of the method) of each combination, implementing a full upgrade for high-risk combinations; for medium-risk combinations, it further considers their internal validation practices (the proportion of other positioning processes) and combination size to dynamically decide whether to upgrade all combinations, upgrade some samples, or maintain the status quo. Ultimately, this achieves: decisively eliminating clearly unreliable methods (high fit coefficient), carefully verifying and progressively optimizing potential problems (medium fit coefficient), and maintaining stability for reliable methods (low fit coefficient), thereby maximizing the overall performance of the positioning system while controlling upgrade costs and risks.

[0134] S41 uses the positioning data of the three-dimensional corneal vertex to determine the adaptation coefficient of the user combination; The above steps include the following situations: Case 1: If the adaptation coefficient of the user group is greater than the preset adaptation coefficient threshold, then the optimization processing strategy for the three-dimensional corneal vertex of the user is determined to be to use two set positions for all users in the user group to determine and locate the auxiliary camera device for the three-dimensional corneal vertex, that is, to use two set positions to determine the auxiliary camera device for the three-dimensional corneal vertex, and to use the auxiliary camera device and the main camera device for positioning processing.

[0135] The first-level strategy decision is made based on the "fit coefficient" of each user combination.

[0136] The "fit coefficient" directly quantifies the "incompatibility" or "failure risk" of a user group to a standard preset location method. The higher the coefficient, the less reliable the preset method is in that group, representing the proportion of users in the matching user group who require a matching process.

[0137] This is the most efficient decision-making entry point. Prioritizing limited optimization resources to the groups whose problems have been proven most severe by data aligns with the principle of "solving the principal contradiction." By setting two thresholds, high and low, portfolios can be quickly categorized into three types: high risk (>0.6), medium risk (≥0.1 and ≤0.6), and low risk (<0.1), with different optimization tones preset for each category.

[0138] For a matching coefficient > 0.6 (high-risk combination), the future location of all users in this combination will be determined by the auxiliary camera device and subsequent location processing using "two set locations".

[0139] When historical data shows that new locations can be reliably processed for positioning, and the accuracy of positioning at the preset locations is low (>60%), continuing to use this method for these users would be extremely irresponsible. Immediately implementing a more reliable, though slightly more expensive, multi-location method for the entire group can fundamentally eliminate positioning failures for these users, ensuring their medical safety and measurement accuracy. This is a "necessary and urgent" optimization.

[0140] For example, the fit coefficient for combination C (post-operative cornea) is 0.7, which is greater than 0.6. Therefore, the system immediately determines that the optimization strategy for all users belonging to combination C in the future will be "to perform positioning processing using the two set positions".

[0141] Case 2: If the adaptation coefficient of the user combination is not greater than the preset adaptation coefficient threshold, or if the adaptation coefficient of the user combination is less than the preset coefficient threshold, then the optimization strategy for the three-dimensional corneal vertex of the user in the user combination is determined to be no optimization.

[0142] Further determine whether it is less than the preset coefficient threshold (0.1): If the fit coefficient is < 0.1: the portfolio is considered a low-risk portfolio. Its optimization strategy is determined to be "no optimization required," and the existing preset position method continues to be used.

[0143] For groups with an extremely low failure rate (<10%) for the pre-defined method, this indicates that the existing method is highly effective. Forcibly upgrading to a multi-location method would only introduce unnecessary operational complexity and time costs, without any significant benefits. Maintaining the status quo is the optimal choice.

[0144] The fit coefficient of combination A (standard cornea) is 0.05, which is less than 0.1. Therefore, the strategy for combination A is "no optimization processing required".

[0145] Case 3: If the adaptation coefficient of the user combination is not less than the preset coefficient threshold, obtain the proportion of user combinations that do not require optimization processing in all user combinations, and determine whether the proportion of user combinations that do not require optimization processing in all user combinations is greater than the preset combination proportion threshold. If so, determine that the optimization processing strategy for the three-dimensional corneal vertex of the users in the remaining user combinations is to use two set positions for the determination and positioning processing of the auxiliary camera device for the three-dimensional corneal vertex for all users in the user combination, that is, to use two set positions for the determination of the auxiliary camera device for the three-dimensional corneal vertex, and use the auxiliary camera device and the main camera device for positioning processing. If not, proceed to step S42.

[0146] Fit coefficient ≥ 0.1 and ≤ 0.6 (medium-risk portfolio): When combinations B (0.4), D (0.25), and E (0.15) enter this branch, first, obtain the proportion of all combinations judged as "no optimization required" (i.e., combination A) in the total combinations: 1 / 5 = 20%. Determine whether this proportion is greater than the preset combination proportion threshold (50%).

[0147] If the proportion is > 50%, it indicates that most combinations in the system are low-risk and do not require optimization. In this case, to simplify management, a conservative strategy of "upgrading all combinations to a dual-position approach" can be adopted for all remaining medium-risk combinations (B, D, E).

[0148] When the "safe zone" dominates, a unified and strong intervention is taken for a few "problem zones," which has low management costs and can completely eliminate the uncertainty in these areas.

[0149] In this example, we determine that 20% is no greater than 50%, so the condition is not met. Therefore, we proceed to step S42 to conduct a more refined individualized analysis of the medium-risk portfolio.

[0150] S42 uses the positioning processing data of the three-dimensional corneal vertex in the user group to determine the proportion of other positioning processes in the user group, and uses it as the proportion of other processes; It should be noted that the above steps include the following: S421 Determine whether the proportion of other processes in the user combination is greater than the preset process proportion threshold. If so, the efficiency of the verification process is high. Therefore, the optimization strategy for the three-dimensional corneal vertex of the user in the user combination is determined to be no optimization process required. If not, proceed to step S422. For each medium-risk combination (B, D, E) that requires detailed analysis in step S41, case 3, analyze its internal validation practices.

[0151] Step S421: Assess the adequacy of internal validation: Obtain the "Other Process Ratio" for each combination (i.e., the proportion of times the dual-position method has been used historically). Determine if this ratio is greater than a preset process ratio threshold (30%).

[0152] The "Other Processes Ratio" reflects the frequency with which the system has actively explored or validated alternatives (dual-position method) in practice. A high ratio indicates that the system has accumulated some experience with multi-position methods within this system.

[0153] If a medium-risk portfolio has been using the dual-position method frequently (proportion > 30%), it means that the system has already "spontaneously" validated and optimized it extensively in practice. In this case, the current state of "no optimization needed" may already be a better state after adjustments, and there is no need to force a change.

[0154] The proportion of other processes in combination B (keratoconus) is 0.5 (i.e., 50%). Judgment: 0.5 is greater than 0.30, therefore the optimization strategy for the user's three-dimensional corneal vertex in the user combination is determined to be no optimization required. Combination E is 0.2, then proceed to step S422.

[0155] S422 obtains the proportion of the number of users in the user group to the total number of users, and determines whether the proportion of the number of users in the user group to the total number of users is greater than a preset user proportion threshold. If so, the optimization processing strategy for the three-dimensional corneal vertex of the users in the user group is determined to be that the auxiliary camera device for the three-dimensional corneal vertex is determined and the positioning processing is performed by using two set positions for all users in the user group. That is, the auxiliary camera device for the three-dimensional corneal vertex is determined by using two set positions, and the positioning processing is performed by using the auxiliary camera device and the main camera device. If not, proceed to the next step. Calculate the percentage of users in this group relative to the total number of users. Determine if this percentage exceeds a preset user percentage threshold (20%).

[0156] This step measures the “impact” of optimizing the portfolio. If a medium-risk portfolio has a large user base (>20%), then its location reliability issues will affect a large number of users. Even if its internal validation data is insufficient, there is a strong reason to upgrade all users to quickly eliminate large-scale potential risks, out of responsibility to the majority of users.

[0157] Specific example (continuous, for combination E): Assume a total of 1000 users, with 180 users in combination E, representing 18%. Judgment: 18% is not greater than 20%, so the condition is not met. Therefore, for combination E, proceed to step S43 for final decision.

[0158] S43 determines the optimization processing strategy for the three-dimensional corneal vertex of the user in the user group based on the adaptation coefficient of the user group and other process ratios.

[0159] It should be noted that the "other process ratio" refers to the proportion of other location processes in the location processing of the user combination.

[0160] It is understood that in the above steps, based on the adaptation coefficient of the user combination and other process ratios, the optimization demand factor of the user combination is determined. It is then determined whether the optimization demand factor of the user combination is greater than a preset demand factor threshold. If so, the optimization processing strategy for the three-dimensional corneal vertex of the users in the user combination is determined to be that for the first proportion of users in the user combination, two preset positions are used to determine and locate the auxiliary camera device for the three-dimensional corneal vertex. That is, two preset positions are used to determine the auxiliary camera device for the three-dimensional corneal vertex, and the auxiliary camera device and the main camera device are used for positioning processing. If not, the optimization processing strategy for the three-dimensional corneal vertex of the users in the user combination is determined to be that for the second proportion of users in the user combination, two preset positions are used to determine and locate the auxiliary camera device for the three-dimensional corneal vertex. That is, two preset positions are used to determine the auxiliary camera device for the three-dimensional corneal vertex, and the auxiliary camera device and the main camera device are used for positioning processing.

[0161] It should be noted that the first ratio is greater than the second ratio.

[0162] It is understood that the value range of the optimization requirement factor of the user combination is between 0 and 1. The larger the adaptation coefficient of the user combination and the smaller the proportion of other processes, the larger the optimization requirement factor of the user combination.

[0163] For combinations like combination E that enter this step, the optimization range is dynamically determined by calculating a comprehensive "optimization demand factor".

[0164] Calculate the optimization demand factor: Assume the formula is Optimization Demand Factor = Fit Coefficient * (1 - Proportion of Other Processes). This formula satisfies the requirement that "the larger the fit coefficient and the smaller the proportion of other processes, the larger the factor will be".

[0165] Combination E: Assuming the fit coefficient is 0.15 and the proportion of other processes is 0.2, then the factor = 0.15 * 0.8 = 0.12. Decision: Determine whether the factor is greater than the preset demand factor threshold (0.1).

[0166] If the factor > 0.1: Optimize using the "first proportion" (30%) of users. That is, randomly select 30% of users from this combination or select them according to certain rules, and upgrade to the dual-location method in future positioning.

[0167] If the factor is ≤ 0.1: use the "second proportion" (10%) of users for optimization.

[0168] This is the most refined trade-off strategy. Optimizing the demand factor integrates "risk intensity" (fit coefficient) and "cognitive blind spot" (1 - validation ratio). A high factor indicates clear risk and insufficient validation, requiring large-scale proactive optimization (30% sampling) to quickly acquire data and reduce risk. A low factor indicates relatively low risk or some validation, allowing for smaller-scale exploratory optimization (10% sampling) to continuously monitor and collect data at a more economical cost, avoiding resource waste from over-optimization.

[0169] Specific examples (final decision): The optimization demand factor for combination E is 0.12, which is greater than 0.1. Therefore, the first proportion (30%) of optimization is adopted for it. The strategy is: in combination E, 30% of users will be located using the dual-location method in the future, while the remaining 70% will still be located in the original way.

[0170] This embodiment of the "Method for Determining Optimization Processing Strategies for 3D Corneal Vertex" constructs a precise optimization decision-making system based on "risk stratification, evidence weighting, and proportional control." The method breaks away from the crude "either all or nothing" approach, classifying risks through adaptation coefficients and introducing the concept of "optimization proportion" for the medium-risk group. This ensures that the high-cost multi-location method is applied only to the users who need it most (all high-risk users and a sample of medium-risk users), while the majority of low-risk users still enjoy the efficient standard process. Overall, it achieves maximum system reliability gains with minimal cost improvement.

[0171] The innovation of "optimizing demand factors" lies in the fact that it is not only a measure of risk but also a measure of "decision-making confidence." For high-factor groups, due to limited historical validation and low decision-making confidence, a larger proportion of proactive optimization is used to "explore the situation." For low-factor groups, decision-making confidence is relatively high, so a smaller proportion of optimization is used for "long-term monitoring." This makes the system's optimization behavior itself a continuous learning and dynamic adjustment scientific experimental process, rather than a one-time subjective decision. The system can continuously collect key comparative data, constantly verify and adjust the "fit coefficient" and "optimization demand factors" of each combination, forming a closed loop of "evaluation -> optimization -> re-evaluation," driving the entire positioning system to autonomously evolve towards higher accuracy and stronger adaptability.

[0172] Example 2 In a second aspect, the present invention provides a computer system comprising: a memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-described high-precision three-dimensional corneal vertex localization method based on image analysis when running the computer program.

[0173] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0174] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0175] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A high-precision three-dimensional corneal vertex localization method based on image analysis, characterized in that, Specifically, it includes: The localization process of the three-dimensional corneal vertex is performed using multi-dimensional image acquisition data and image analysis models. Based on the verification data of localization processing results for different users and combined with the extraction results of corneal image features of users, the localization processing method of the auxiliary camera device for the three-dimensional corneal vertex of the user is determined. The positioning processing result of the three-dimensional corneal vertex is determined using the positioning processing method described above. Based on the deviation of the positioning processing results between different positioning processing methods in different users, the multiple positioning requirement features in the corneal image features are determined. Based on the location processing data of multiple location requirement characteristics, the available matching combinations in the user combination are determined. Combined with the location processing data of users under different location verification processing methods and the multiple location requirement characteristic data, when it is determined that the location processing strategy needs to be optimized, the optimization processing strategy of the user's three-dimensional corneal vertex is determined based on the location processing data of the three-dimensional corneal vertex.

2. The high-precision three-dimensional corneal vertex localization method based on image analysis as described in claim 1, characterized in that, The multidimensional image is at least two images from different sources that contain the corneal region.

3. The high-precision three-dimensional corneal vertex localization method based on image analysis as described in claim 1, characterized in that, The localization processing data based on the corneal image features is determined according to the number of users undergoing localization processing based on the corneal image features.

4. The high-precision three-dimensional corneal vertex localization method based on image analysis as described in claim 1, characterized in that, The method for determining the positioning processing method of the auxiliary camera device for the user's three-dimensional corneal vertex is as follows: Based on the location processing data of different users, users whose corneal image features meet the similarity requirements are grouped into the same user group; Based on the distribution data of the user combinations, determine the number of users for location processing in different user combinations; Based on the extraction results of the user's corneal image features and the number of users in different user combinations for localization processing, the localization processing method of the auxiliary camera device for the user's three-dimensional corneal vertex is determined.

5. The high-precision three-dimensional corneal vertex localization method based on image analysis as described in claim 4, characterized in that, If the similarity coefficient of corneal image features between a user and other users is greater than a preset similarity coefficient threshold, then the user and other users will be grouped into the same user group.

6. The high-precision three-dimensional corneal vertex localization method based on image analysis as described in claim 1, characterized in that, The positioning processing method divides the processing according to the position of the auxiliary camera device.

7. The high-precision three-dimensional corneal vertex localization method based on image analysis as described in claim 1, characterized in that, The method for determining the multiple localization requirement features in the corneal image features is as follows: Based on the deviation of the positioning processing results among the positioning processing methods for the users, the positioning processing process with inconsistent positioning results is identified and regarded as an inconsistent process. The location processing method corresponding to the location processing result that is consistent with the verification processing result during the inconsistency process is used as the matching verification method; Based on inconsistent processes and matching verification method data across different users, the multiple localization requirement features in the corneal image features are determined.

8. The high-precision three-dimensional corneal vertex localization method based on image analysis as described in claim 1, characterized in that, The available matching combination is a user combination with an adaptation coefficient greater than a preset adaptation coefficient threshold, wherein the adaptation coefficient is the proportion of users in the user combination who have an identification and matching process.

9. The high-precision three-dimensional corneal vertex localization method based on image analysis as described in claim 1, characterized in that, Determine whether the positioning processing strategy needs optimization, specifically including: Based on the available matching combination data in the user combination, determine the number of available matching combinations in the user combination; Based on the positioning data of users under different positioning verification processing methods, it is determined that in different user combinations, two positions are used to determine the location of the auxiliary camera device for three-dimensional corneal vertex and the positioning process, and this is used as another positioning process. Based on the number of available matching combinations, other location process data in different user combinations, and the multi-location demand characteristic data, it is determined whether the location processing strategy needs to be optimized.

10. A computer system, comprising: A memory and processor connected in communication, and a computer program stored in the memory and capable of running on the processor, characterized in that, when the processor runs the computer program, it executes a high-precision three-dimensional corneal vertex localization method based on image analysis as described in any one of claims 1-9.

Images