Intelligent diagnosis and treatment decision system for eye optometry based on knowledge graph

Abstract

The application discloses an eye vision and light intelligent diagnosis and treatment decision system based on a knowledge graph, performs knowledge extraction and formalization conversion on rule texts in the field of eye vision and light, generates computable constraint conditions and constructs a multi-dimensional physiological constraint correlation set, detects a constraint conflict state, allocates a penalty weight based on a constraint priority and constructs a comprehensive evaluation function, executes numerical approximation and boundary truncation along a parameter adjustment direction vector, and outputs a steady-state target solution vector. The application realizes a mapping relationship from discrete texts to continuous computable structures of field knowledge, drives continuous variable reasoning through the knowledge graph, realizes automatic optimization of parameters, avoids parameter jump to cause discomfort, and ensures safety and robustness of output parameters under physiological fluctuation conditions.

Description

Technical Field

[0001] This invention relates to the fields of knowledge engineering and intelligent information processing technology, specifically to an intelligent ophthalmic diagnosis and treatment decision system based on knowledge graphs. Background Technology

[0002] Traditional knowledge graphs, centered on discrete entities and semantic relationships, excel at handling symbolic knowledge reasoning and question answering. However, when faced with physical variables with continuous domains, existing knowledge graphs lack an underlying mechanism to directly map domain text rules into computable numerical constraints. This prevents them from directly driving automated optimization in continuous parameter spaces. In professional fields such as optometry, parameters such as spherical power, cylindrical power, astigmatic axis, and binocular vision function are all multidimensional continuous variables, and there are highly coupled and mutually restrictive relationships among various physiological indicators. With the cross-integration of optometry and computer information technology, intelligent decision support systems are increasingly widely used in optometry. Intelligent decision support systems are mainly used to assist optometrists or ophthalmologists in processing multidimensional eye examination data such as patients' refractive power, axial length, corneal curvature, and binocular vision function, and then formulate vision correction plans or prescription glasses for patients. These systems usually rely on an underlying medical knowledge base to perform logical reasoning and comprehensive analysis on the input clinical data to improve the efficiency of clinical diagnosis and treatment.

[0003] In current routine optometry practice, existing decision support systems mostly adopt an architecture based on static rule trees or threshold condition judgments. After receiving the patient's domain parameter data, the system performs layer-by-layer matching according to pre-set fixed threshold conditions. Based on the patient's monocular myopia degree, astigmatism axis, age, and other indicators, the system retrieves the corresponding standard prescription template from the static rule base and outputs preliminary prescription guidance suggestions through hard-coded logical judgment statements, which are then submitted to the doctor for secondary manual review. However, in optometry prescription decision-making scenarios that require gradual intervention, such as large-scale refractive power changes, binocular vision reconstruction, or contact lens fitting, existing decision support systems have significant limitations in handling the complex logical mapping of multidimensional physiological parameters. Due to the high coupling and mutual constraints among the various physiological indicators of the optometry system, the static rule engine on which the existing system relies is limited by the discrete threshold judgment mechanism, making it difficult to perform automated multi-objective trade-offs and dynamic optimization in the coupled space of multidimensional continuous variables such as spherical power, cylindrical power, and binocular fusion zone.

[0004] More fundamentally, existing knowledge bases mostly employ discrete symbolic representations or static rule tree structures. The integration of domain rules with continuous variable numerical optimization algorithms is not yet tight enough. The technical means to transform domain text rules into computable continuous constraints and integrate them into the parameter optimization process still need improvement. This invention aims to provide a technical solution that embeds optometry domain rules into the numerical optimization process in the form of structured constraint templates. By constructing a computational knowledge graph, whose nodes are state space nodes carrying continuous domains and whose edges are variable constraint mapping relationships carrying computable constraints, this invention assists optometrists in balancing medical safety bottom lines and patient adaptability in a multivariate coupled space, reducing the risk of parameters exceeding clinical tolerance ranges. Summary of the Invention

[0005] To address the technical problem of the difficulty in formally representing and participating in continuous variable reasoning of the aforementioned domain knowledge, this invention provides an intelligent optometry diagnosis and treatment decision-making system based on knowledge graphs, comprising:

[0006] The data access module is used to acquire parameter data in the field of optometry, perform standardization processing, and output a standardized parameter matrix.

[0007] The knowledge graph construction module, which is communicatively connected to the data access module, is used to perform knowledge extraction and formal transformation on rule texts in the optometry field, generate computable constraints, and construct a multidimensional physiological constraint association set in memory, which includes a set of state space nodes and a set of variable constraint mapping relationships. The standardized parameter matrix is ​​mapped to the corresponding state space nodes. A single variable constraint mapping relationship in the set of variable constraint mapping relationships connects the state space nodes and is configured with the computable constraints.

[0008] The conflict detection and parameter optimization module is communicatively connected to the knowledge graph construction module. It is used to allocate penalty weights based on constraint priority when the current value of the state space node makes the computable constraint condition not valid, construct a comprehensive evaluation function based on the allocated penalty weights, perform numerical approximation on the variable value of the state space node along the parameter adjustment direction vector, and output the steady-state target solution vector.

[0009] Furthermore, when configuring the computable constraints, the knowledge graph construction module is specifically used to: call a pre-set rule parsing template library to convert the threshold expressions in the optometry domain rule library into a set of numerical calculation conditions; configure the set of numerical calculation conditions as the computable constraints, and classify them into hard constraints for clinical safety and soft constraints for comfort.

[0010] Furthermore, the computable constraints are specifically clinical medical inequalities. When configuring the clinical medical inequalities, the knowledge graph construction module is specifically used to: call a preset rule parsing template library to convert the threshold expressions in the optometry clinical rule library into a set of numerical calculation conditions; configure the set of numerical calculation conditions as the clinical medical inequalities and classify them into hard constraints for clinical safety and soft constraints for comfort.

[0011] Furthermore, when allocating penalty weights, the conflict detection and parameter optimization module is specifically used to: substitute the current continuous variable values ​​of the state space nodes into the computable constraints to perform verification calculations, determine constraint conflicts and construct non-negative penalty terms; extract the constraint label categories carried by the constraint conflicts, match and extract the corresponding penalty weights in the constraint priority weight mapping table, and assign values.

[0012] Furthermore, when performing numerical approximation, the conflict detection and parameter optimization module (300) is specifically used to: summarize the penalty term and the penalty weight to construct the comprehensive evaluation function; calculate the parameter deviation gradient of the comprehensive evaluation function with respect to the multidimensional state vector, form the parameter adjustment direction vector and perform reverse numerical approximation; if the approximated value exceeds the lower or upper bound of physiological safety, the projection gradient method is used to truncate it to the corresponding lower or upper bound of physiological safety and remap it back to the multidimensional physiological constraint association set.

[0013] Furthermore, the system also includes a dynamic tolerance module for the adaptation period of wearing glasses, which is communicatively connected to the conflict detection and parameter optimization module. This module is used to calculate the parameter difference between the steady-state target solution vector and the initial value of the standardized parameter matrix, divide the time slice sequence according to the parameter difference, insert a temporary transition period tolerance mapping relationship configured with parameter smoothing constraint equations in the early stage of the time slice sequence to trigger the recalculation of the transition solution vector, and sequentially remove the temporary transition period tolerance mapping relationship as time progresses.

[0014] Furthermore, when dividing the time slice sequence, the dynamic tolerance module for the adaptation period of wearing glasses is specifically used to: subtract the values ​​of each item in the steady-state target solution vector from the initial value of the standardized parameter matrix to generate a parameter difference vector; input the parameter difference vector into the physiological tolerance mapping table to match the total duration parameter of the adaptation period and the number of slices parameter, and divide the time slice sequence accordingly.

[0015] Furthermore, the system also includes an eye fatigue tolerance verification module, which is communicatively connected to the dynamic tolerance module for the adaptation period of wearing glasses. This module is used to inject fatigue offset components into the transition solution vector and the steady-state target solution vector, substitute them into the calculable constraints for verification, and increase the corresponding penalty weight when the verification fails and feed it back to the conflict detection and parameter optimization module for re-solving.

[0016] Furthermore, the system also includes a multi-stage prescription generation module, which is communicatively connected to the dynamic tolerance module for the adaptation period of wearing glasses and the eye fatigue tolerance verification module, respectively. It is used to assemble the solution vectors that have passed the verification in the order of time slices to generate a staged set of optometric prescription parameters and output them.

[0017] Furthermore, when generating a phased optometry prescription parameter set, the multi-stage prescription generation module is specifically used to: extract time slice numbers and perform sorting operations on the transitional solution vector and the steady-state target solution vector; reverse map the values ​​of the solution vectors to the corresponding prescription dimensions, and perform step quantization and rounding based on the clinical manufacturing step standard value; calculate the recommended fitting time node in combination with the preset time slice physical duration benchmark, and assemble it with the quantized and rounded prescription parameters to form the phased optometry prescription parameter set.

[0018] The present invention, by adopting the above technical solution, can bring the following beneficial effects:

[0019] 1. This invention performs knowledge extraction and formal transformation on the rule text in the optometry field, transforming discrete domain knowledge into computable constraints. Based on these computable constraints, a multidimensional physiological constraint association set containing the mapping relationship between state space nodes and variable constraints is constructed, realizing the formal transformation of domain rules into continuous numerical constraints, and providing a low-level constraint representation for automated numerical optimization in continuous variable space.

[0020] 2. This invention detects constraint conflict states based on a multidimensional physiological constraint association set, allocates penalty weights according to constraint priority to construct a comprehensive evaluation function, performs directional numerical approximation of variable values ​​of nodes in the state space along the parameter adjustment direction vector, and performs forced boundary truncation to strictly satisfy the safety boundary of hard constraints. This achieves the effect of automatically optimizing in the parameter space and alleviating multi-objective constraint conflicts, ensuring the reliability of the final output solution vector.

[0021] 3. This invention divides the time slice sequence by calculating the parameter difference between the steady-state target solution vector and the initial value of the standardized parameter matrix. In the early stage of the time slice sequence, a temporary transition period tolerance mapping relationship configured with parameter smoothing constraint equations is inserted into the multidimensional physiological constraint association set to trigger the recalculation of the transition solution vector. As the slice progresses, the temporary transition period tolerance mapping relationship is removed sequentially. This realizes the construction of a time-series dynamic tolerance mechanism that matches the visual adaptation cycle in the underlying data structure. For scenarios requiring gradual adaptation of large-span refractive power changes or binocular vision function reconstruction, it can avoid eye discomfort caused by a one-time parameter jump and output a gradual fitting prescription.

[0022] 4. This invention injects fatigue offset components, which characterize fluctuations in physiological regulatory force, into the transitional solution vector and the steady-state target solution vector. The test solution vector is then resubmitted into the calculable constraints for verification. If the verification fails, the penalty weight of the relevant variable constraint mapping relationship is increased and feedback is given to re-solve the problem. This achieves the effect of applying fault tolerance verification before prescription finalization to check the fault tolerance capability of the solution vector under physiological fluctuation conditions, eliminating parameter combinations that are prone to exceeding the safety boundary, and ensuring the safety status of the final output optometric prescription. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the architecture of the intelligent optometry decision-making system of the present invention;

[0024] Figure 2 This is a flowchart of the intelligent decision-making method for optometry based on physiological constraint association set and dynamic tolerance constraint of the present invention.

[0025] Figure 3 This is a schematic diagram of the data mapping mechanism for state space nodes in this invention.

[0026] Figure 4 This is a schematic diagram illustrating the transformation of the variable constraint mapping relationship and medical rules in this invention;

[0027] Figure 5 Here is a logic diagram for configuring the conflict detection and prescription evaluation functions of this invention;

[0028] Figure 6 This is the control flow graph of the prescription parameter optimization algorithm of the present invention;

[0029] Figure 7 A logic diagram is inserted to illustrate the timing slice division and feedforward temporary transition period tolerance mapping relationship of this invention;

[0030] Figure 8 This is a diagram illustrating the constraint connection release and state restoration rules that evolve over time according to the present invention.

[0031] Figure 9 This is a logic diagram of the closed-loop verification mechanism for the steady-state verification and fatigue fault tolerance verification of the prescription of this invention;

[0032] Figure 10 This is the logic for assembling and parsing the multi-stage progressive prescription data of this invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Example 1

[0035] Please refer to the appendix. Figure 1 This invention provides an intelligent ophthalmic diagnosis and treatment decision system based on knowledge graphs, which is used to explain the composition relationship and data interaction logic of the various functional modules of this invention from the overall architecture level.

[0036] The above system includes: a data access module 100, a knowledge graph construction module 200, a conflict detection and parameter optimization module 300, a dynamic tolerance module for the adaptation period of wearing glasses 400, an eye fatigue tolerance verification module 500, and a multi-stage prescription generation module 600.

[0037] The data access module 100 is used to receive clinical examination data, including subjective refraction parameters, corneal topography data, and binocular vision function examination data. The data access module 100 performs format conversion and standardization processing on the received clinical examination data, extracts numerical features, and outputs a standardized parameter matrix to the knowledge graph construction module 200. By accessing multi-dimensional clinical examination data and performing standardization processing, a unified data source is provided for the subsequent construction of the underlying data structure, thereby improving the compatibility of the optometry intelligent decision-making system with multi-source clinical data.

[0038] The knowledge graph construction module 200 is communicatively connected to the data access module 100. The knowledge graph construction module 200 is used to construct a multidimensional physiological constraint association set in memory. The multidimensional physiological constraint association set includes a set of state space nodes and a set of variable constraint mapping relationships. Each state space node in the set of state space nodes maps to a variable in the clinical examination data and has a corresponding continuous domain. A single variable constraint mapping relationship in the set of variable constraint mapping relationships connects two or more state space nodes, and a single variable constraint mapping relationship is configured with a clinical medical inequality. The knowledge graph construction module 200 maps the standardized parameter matrix output by the data access module 100 to the corresponding state space nodes. By constructing a multidimensional physiological constraint association set containing state space nodes and variable constraint mapping relationships, the technical limitation of traditional discrete rule engines in directly handling multidimensional continuous variable coupling relationships is overcome, and the underlying algebraic expression of optometry medical rules is realized.

[0039] The conflict detection and parameter optimization module 300 is communicatively connected to the knowledge graph construction module 200. The conflict detection and parameter optimization module 300 is used to detect the connectivity and constraint conflict state within the multidimensional physiological constraint association set. When the current value of the state space node makes the clinical medical inequality not true, the conflict detection and parameter optimization module 300 assigns penalty weights to the corresponding variable constraint mapping relationship according to the preset medical priority. The conflict detection and parameter optimization module 300 constructs a prescription comprehensive evaluation function based on the assigned penalty weights, and performs directional numerical approximation on the variable values ​​of each state space node along the prescription parameter adjustment direction vector in the parameter space. When the approximation calculation ends, the conflict detection and parameter optimization module 300 outputs a steady-state target solution vector that meets the convergence condition. By introducing a numerical directional approximation mechanism, the technical problem of system decision stagnation caused by multi-objective parameter conflicts in optometry is solved, and the dynamic trade-off and automated optimization of quantitative prescription parameters are realized.

[0040] The dynamic tolerance module 400 for the adaptation period of contact lenses communicates with the conflict detection and parameter optimization module 300. The dynamic tolerance module 400 receives the steady-state target solution vector and calculates the parameter difference between the steady-state target solution vector and the initial value of the standardized parameter matrix. Based on the parameter difference, the dynamic tolerance module 400 divides the time slice sequence of the adaptation period. In the early stage of the time slice sequence, the dynamic tolerance module 400 inserts a temporary transition period tolerance mapping relationship configured with parameter smoothing constraint equations into the multidimensional physiological constraint correlation set. By limiting the range of variable changes in the state space nodes, the dynamic tolerance module 400 of the lens adaptation period triggers the conflict detection and parameter optimization module 300 to recalculate the transitional solution vector under the current time slice. As the time slice sequence progresses, the dynamic tolerance module 400 of the lens adaptation period removes the temporary transitional tolerance mapping relationship from the multidimensional physiological constraint association set in turn, so that the multidimensional physiological constraint association set is restored to the initial architecture. A progressive tolerance constraint matching the visual adaptation cycle is constructed in the system, avoiding lens discomfort caused by a single final prescription and improving the physiological compatibility of the optometry decision scheme.

[0041] The eye fatigue tolerance verification module 500 communicates with the dynamic tolerance module 400 during the lens adaptation period. The eye fatigue tolerance verification module 500 obtains the transitional solution vector and the steady-state target solution vector output by the dynamic tolerance module 400 during the lens adaptation period. The eye fatigue tolerance verification module 500 injects fatigue offset components into the transitional solution vector and the steady-state target solution vector respectively. The fatigue offset components are used to characterize the fluctuation of physiological accommodation force. After injecting the fatigue offset components, the eye fatigue tolerance verification module 500 resubstitutes the transitional solution vector and the steady-state target solution vector after injecting the fatigue offset components into the clinical medical inequality for verification. If the clinical medical inequality does not hold, the eye fatigue tolerance verification module 500 increases the penalty weight of the relevant variable constraint mapping relationship and feeds back the variable constraint mapping relationship information after increasing the penalty weight to the conflict detection and parameter optimization module 300 to perform re-solution. By applying fatigue tolerance countermeasure verification, the robustness of prescription parameters under physiological fluctuation conditions is checked before prescription output, reducing the risk of the optometry prescription breaking through the medical safety boundary in the actual eye use fluctuation environment.

[0042] The multi-stage prescription generation module 600 is communicatively connected to the dynamic tolerance module 400 for the adaptation period of wearing glasses and the eye fatigue tolerance verification module 500. The multi-stage prescription generation module 600 collects all solution vectors that have passed the verification by the eye fatigue tolerance verification module 500 in the time slice sequence. The multi-stage prescription generation module 600 assembles the collected solution vectors according to the time slice order to generate a staged optometry prescription parameter set. The multi-stage prescription generation module 600 outputs the staged optometry prescription parameter set to an external memory or display terminal. By assembling the multi-stage parameter set, a progressive fitting prescription with a clear time evolution path is output, providing optometrists with complete staged diagnosis and treatment guidance.

[0043] Example 2

[0044] Please refer to the appendix. Figure 2 Based on the aforementioned knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system, this paper presents an intelligent ophthalmic decision-making method based on physiological constraint association sets and temporal dynamic constraints. This method is used to illustrate the execution order of each step and the calling relationship between modules from an overall process perspective. The method includes the following steps:

[0045] In step S100, the data access module 100 receives the patient's clinical examination data, performs format conversion and standardization processing on the received clinical examination data, extracts the numerical features after standardization processing, and outputs a standardized parameter matrix. By converting multi-source clinical examination data into a standardized parameter matrix, a unified data format is provided for subsequent model mapping, reducing the computational overhead of the system in processing heterogeneous clinical data.

[0046] In step S200, the knowledge graph construction module 200 receives the standardized parameter matrix output by the data access module 100. The knowledge graph construction module 200 constructs a multidimensional physiological constraint association set in memory, which includes a set of state space nodes and a set of variable constraint mapping relationships. The knowledge graph construction module 200 maps the standardized parameter matrix to the corresponding state space nodes in the set of state space nodes. By mapping clinical data to the multidimensional physiological constraint association set, a coupling constraint architecture between multidimensional optometry variables is established, overcoming the technical defect that static rule engines cannot handle continuous variables.

[0047] In step S300, the conflict detection and parameter optimization module 300 detects the constraint conflict state within the multidimensional physiological constraint association set. When the current variable value of a node in the state space does not satisfy the clinical medical inequality within the variable constraint mapping relationship, the conflict detection and parameter optimization module 300 assigns penalty weights to the variable constraint mapping relationship that generates the conflict based on the preset medical priority, and constructs a prescription comprehensive evaluation function based on the assigned penalty weights. The conflict detection and parameter optimization module 300 performs numerical approximation operation on the variable value of the node in the state space along the prescription parameter adjustment direction vector, and forcibly truncates the variable value that violates the hard constraint equation to the corresponding physiological safety boundary in the approximation calculation step. When the iteration ends, the conflict detection and parameter optimization module 300 outputs a steady-state target solution vector that satisfies the convergence condition. By eliminating constraint conflicts through numerical directional approximation operation, the automatic optimization of prescription parameters is realized in the parameter space, ensuring the medical reliability of the output steady-state target solution vector.

[0048] In step S400, the dynamic tolerance module 400 for the adaptation period of glasses receives the steady-state target solution vector output by the conflict detection and parameter optimization module 300. The dynamic tolerance module 400 for the adaptation period of glasses calculates the parameter difference between the steady-state target solution vector and the initial value of the standardized parameter matrix. The dynamic tolerance module 400 for the adaptation period of glasses divides the time slice sequence of the adaptation period of glasses according to the parameter difference. In the early stage of the time slice sequence, the dynamic tolerance module 400 for the adaptation period of glasses inserts a temporary transition period tolerance mapping relationship configured with parameter smoothing constraint equations into the multidimensional physiological constraint association set. The dynamic tolerance module 400 for the adaptation period of glasses triggers the conflict detection and parameter optimization module 300 to recalculate the transition solution vector under the current time slice sequence. As the time slice sequence progresses, the dynamic tolerance module 400 for the adaptation period of glasses sequentially removes the temporary transition period tolerance mapping relationship in the multidimensional physiological constraint association set. Through the operation of inserting and removing the temporary transition period tolerance mapping relationship, a progressive tolerance constraint mechanism matching the visual adaptation cycle is constructed inside the system, avoiding physiological discomfort caused by a one-time parameter jump.

[0049] In step S500, the eye fatigue tolerance verification module 500 obtains the transitional solution vector and the steady-state target solution vector. The eye fatigue tolerance verification module 500 injects a fatigue offset component, which is used to characterize the fluctuation of physiological regulatory force, into the transitional solution vector and the steady-state target solution vector. The eye fatigue tolerance verification module 500 substitutes the transitional solution vector and the steady-state target solution vector after injecting the fatigue offset component into the clinical medical inequality for verification. If the clinical medical inequality does not hold, the eye fatigue tolerance verification module 500 increases the penalty weight of the relevant variable constraint mapping relationship and feeds back the variable constraint mapping relationship information after increasing the penalty weight to the conflict detection and parameter optimization module 300 to solve again until the transitional solution vector and the steady-state target solution vector after injecting the fatigue offset component both satisfy the clinical medical inequality. By introducing the fatigue offset component to perform equation adversarial verification and form a closed-loop control flow, the parameter combination with poor anti-fatigue robustness is eliminated before the prescription is finalized, and the safety of the steady-state target solution vector in actual eye use scenarios is improved.

[0050] In step S600, the multi-stage prescription generation module 600 collects all the transitional solution vectors and steady-state target solution vectors that have passed the eye fatigue tolerance verification module 500. The multi-stage prescription generation module 600 assembles the collected transitional solution vectors and steady-state target solution vectors in the order of the time slice sequence to generate a staged optometry prescription parameter set. The multi-stage prescription generation module 600 outputs the staged optometry prescription parameter set. By assembling the solution vectors in the order of the time slice sequence, a progressive fitting prescription with time-series guidance is generated, which improves the systematic intervention process of optometry diagnosis and treatment.

[0051] Example 3

[0052] Please refer to the appendix. Figure 3 and attached Figure 4 Based on Embodiment 2, in step S200 above, the data mapping mechanism of state space nodes and the mathematical transformation mechanism of variable constraint mapping relationship with medical rules may include the following steps:

[0053] In step S210, the knowledge graph construction module 200 establishes a set of state space nodes in memory. The set of state space nodes contains multiple state space nodes. The knowledge graph construction module 200 defines optometry physiological variables as continuous state space nodes. State space nodes are different from discrete entities in traditional knowledge graphs. The knowledge graph construction module 200 configures state variables and continuous domains for each state space node. By configuring state space nodes with continuous domains, the underlying mathematical structure is provided for the optometry intelligent decision-making system to perform quantitative parameter optimization.

[0054] In step S220, the data access module 100 outputs the standardized parameter matrix to the knowledge graph construction module 200. Specifically, the data access module 100 establishes underlying data communication with the clinical end's comprehensive optometry instrument, computer optometry instrument, and corneal topography instrument through the Medical Digital Imaging and Communication Standard (DICOM) interface and the Health Level 7 (HL7) protocol. The data access module 100 parses the original messages transmitted by the above devices, filters invalid device identifiers, and extracts objective measured values ​​of spherical power, cylindrical power, astigmatism axis, pupillary distance, and accommodation amplitude. The data access module 100 writes the extracted objective measured values ​​into a two-dimensional array of preset size according to a unified floating-point number format to generate the standardized parameter matrix. The knowledge graph construction module 200 calls a preset variable mapping index table, which records the correspondence between the index numbers of each data dimension in the standardized parameter matrix and the nodes in the state space. The knowledge graph construction module 200 extracts the multidimensional ophthalmic measured values ​​from the standardized parameter matrix and assigns the multidimensional ophthalmic measured values ​​one by one to the corresponding continuous ophthalmic parameter variable values ​​in the state space nodes according to the variable mapping index table. Through standard protocol parsing and one-by-one assignment operations, the initial state mapping of external objective medical equipment hardware data to the multidimensional physiological constraint association set inside the system is completed, ensuring the deployability and data interoperability of this system in the actual clinical information system (HIS).

[0055] Step S230: After the assignment is completed, the knowledge graph construction module 200 calls the preset physiological limit database. The knowledge graph construction module 200 retrieves the corresponding safe range in the preset physiological limit database according to the variable type of the continuous parameter variable value of the eye. The knowledge graph construction module 200 writes the retrieved safe range into the lower physiological safety bound and the upper physiological safety bound respectively. For the state space node of the binocular fusion zone width type, the knowledge graph construction module 200 initializes the lower physiological safety bound and the upper physiological safety bound to the limit disparity angle to maintain binocular single vision function. By calling the preset physiological limit database to initialize the lower physiological safety bound and the upper physiological safety bound, the search space for subsequent algorithm solutions is limited, ensuring that the final output prescription parameters do not exceed the physiological tolerance limit of the human eye.

[0056] In step S240, the knowledge graph construction module 200 establishes a set of variable constraint mapping relationships in memory. This set contains multiple variable constraint mapping relationships. The knowledge graph construction module 200 reads a preset ophthalmology clinical rule base. The rules in this base are stored in a structured format, with each rule containing a rule identifier, applicable conditions, constraint type, associated physiological variable identifier, and threshold expression field. The knowledge graph construction module 200 calls a preset rule parsing template library, matching the corresponding parsing template based on the rule identifier, and converts the threshold expression in the structured rule into a numerical calculation condition set. This rule parsing template library is pre-configured by system developers based on ophthalmology clinical guidelines and includes at least one of the following: anisometropia rule template, binocular fusion rule template, and accommodative amplitude rule template. Each template defines the index mapping relationship and inequality operators for associated state space nodes. The knowledge graph construction module 200 configures the converted numerical calculation condition set as clinical medical inequalities within the variable constraint mapping relationships and classifies these inequalities into hard constraints for clinical safety and soft constraints for comfort. The hard constraints are those relating to the bottom line of physiological safety, including at least one of the binocular equivalent spherical aberration limit and the lower limit of accommodative amplitude; the soft constraints are those relating to visual comfort, including binocular fusion reserve ratio constraints. This invention employs a variable constraint mapping relationship, enabling a single connecting edge to simultaneously encompass all relevant state space nodes, thereby achieving a precise expression of complex joint medical constraints at the underlying data structure level.

[0057] In step S250, the knowledge graph construction module 200 constructs a first variable constraint mapping relationship based on the clinical rules for anisometropia and binocular fusion safety. The knowledge graph construction module 200 selects a first state space node, a second state space node, a third state space node, a fourth state space node, and a fifth state space node from the state space node set. The first state space node maps to the right eye spherical power variable, the second state space node maps to the right eye cylindrical power variable, the third state space node maps to the left eye spherical power variable, the fourth state space node maps to the left eye cylindrical power variable, and the fifth state space node maps to the maximum tolerance variable for binocular equivalent spherical aberration. The knowledge graph construction module 200 establishes a network connection between the first variable constraint mapping relationship and the first, second, third, fourth, and fifth state space nodes. By connecting five cross-eye state space nodes simultaneously through a single variable constraint mapping relationship, a high-dimensional parameter coupling architecture is constructed, which accurately reflects the physical mechanism of mutual constraints among multiple physiological indicators in the optometry system.

[0058] In step S260, the knowledge graph construction module 200 generates a clinical medical inequality for anisometropia within the first variable constraint mapping relationship. This inequality is used to limit the binocular equivalent spherical difference to not exceeding the physiological safety threshold. The knowledge graph construction module 200 extracts the age and heterophoria values ​​from the clinical examination data. Based on these values, the module performs a matching query in a preset physiological tolerance mapping table to obtain the specific value of the individual's maximum tolerance for binocular equivalent spherical difference. The module assigns this specific value to the fifth state space node. The expression of the clinical medical inequality for anisometropia configured by the knowledge graph construction module 200 is as follows:

[0059]

[0060] In the formula, This represents the physiological tolerance deviation value within the first variable constraint mapping relationship; This represents the maximum tolerance threshold for binocular equivalent spherical lens difference corresponding to the fifth state space node; This represents the continuous variable value of the right eye lens power corresponding to the first state space node; This represents the continuous variable value of the right eye cylindrical lens power corresponding to the node in the second state space; This represents the continuous variable value of the left eye's lens power corresponding to the node in the third state space. This represents the continuous variable value of the left eye cylindrical lens power corresponding to the fourth state space node;

[0061] In step S270, the knowledge graph construction module 200 substitutes the continuous variable values ​​of each state space node acquired in real time into the refractive anisometropia clinical medical inequality for calculation. When the calculation result meets the boundary judgment, the knowledge graph construction module 200 determines that the first variable constraint mapping relationship is in a connected state. When the calculation result does not meet the boundary judgment, the knowledge graph construction module 200 determines that the first variable constraint mapping relationship has a constraint conflict and outputs a conflict anomaly signal to the conflict detection and parameter optimization module 300. By constructing the refractive anisometropia clinical medical inequality and performing numerical judgment, the qualitative medical constraints are transformed into computer-executable quantitative verification logic, realizing the function of directly calculating the binocular disparity tolerance of the knowledge graph, and providing a clear boundary for subsequent optimization algorithms.

[0062] Example 4

[0063] Please refer to the appendix. Figure 5 To be continued Figure 8 Based on Embodiment 2, step S300 specifically involves conflict detection and cost penalty function configuration logic and prescription parameter numerical approximation optimization control flow, including the following steps:

[0064] In step S310, the conflict detection and parameter optimization module 300 traverses the set of variable constraint mapping relationships within the multidimensional physiological constraint association set. The conflict detection and parameter optimization module 300 extracts the clinical medical inequalities within the variable constraint mapping relationships one by one. The conflict detection and parameter optimization module 300 obtains the continuous variable values ​​in each state space node at the current time. The conflict detection and parameter optimization module 300 substitutes the obtained continuous variable values ​​into the clinical medical inequalities to perform verification calculations. When the verification calculation result of the clinical medical inequalities does not meet the boundary condition of less than or equal to zero, the conflict detection and parameter optimization module 300 determines that a medical constraint conflict has occurred, thus realizing the location of the medical constraint conflict under the multidimensional parameter combination state.

[0065] Step S320: The conflict detection and parameter optimization module 300 constructs a penalty term for the variable constraint mapping relationship where medical constraint conflicts occur. The mathematical expression of the penalty term configured by the conflict detection and parameter optimization module 300 is as follows:

[0066]

[0067] In the formula, Indicates the first The penalty term value for the variable constraint mapping relationship; Indicates the first The numerical results of the verification calculation of clinical medical inequalities within the variable constraint mapping relationship; by constructing a one-sided penalty term, the verification calculation deviation of clinical medical inequalities is transformed into a non-negative numerical penalty index, providing a continuously differentiable mathematical basis for the subsequent global optimization algorithm of the conflict detection and parameter optimization module 300.

[0068] In step S330, the conflict detection and parameter optimization module 300 extracts the medical label categories carried by the variable constraint mapping relationship in which medical constraint conflict occurs. The conflict detection and parameter optimization module 300 inputs the medical label categories into a preset medical priority weight mapping table to perform a matching query. The preset medical priority weight mapping table records the penalty weight coefficients corresponding to different medical label categories. The preset medical priority weight mapping table is configured with weight grading rules. The weight grading rules set the penalty weight coefficients representing physiological safety bottom line labels to high-order parameters and set the penalty weight coefficients representing visual clarity labels to low-order parameters. The conflict detection and parameter optimization module 300 extracts the penalty weight coefficients output by the preset medical priority weight mapping table and assigns the penalty weight coefficients to the variable constraint mapping relationship in which medical constraint conflict occurs. By extracting medical label categories and matching penalty weight coefficients in the medical priority weight mapping table configured with weight grading rules, priority protection of core medical safety bottom lines such as binocular vision function is ensured in the case of multi-target conflict.

[0069] Step S340: The conflict detection and parameter optimization module 300 summarizes the penalty term values ​​and penalty weight coefficients of all variable constraint mapping relationships in the variable constraint mapping relationship set. The conflict detection and parameter optimization module 300 uses the summarized penalty term values ​​and penalty weight coefficients to construct a prescription comprehensive evaluation function. The expression of the prescription comprehensive evaluation function configured by the conflict detection and parameter optimization module 300 is as follows:

[0070]

[0071] In the formula, Represents the current multidimensional state vector; This represents a multidimensional state vector obtained by mapping the initial values ​​of the normalized parameter matrix; Represents the continuous function value of the prescription comprehensive evaluation function; This represents the total number of multidimensional physiological constraint association set variable constraint mapping relationships; The index number representing the variable constraint mapping relationship; Indicates the first The penalty weight coefficients corresponding to the variable constraint mapping relationships. Indicates the numerical value of the penalty item; This represents the fidelity weighting coefficient, used to balance the degree of constraint satisfaction with the extent to which prescription parameters deviate from the initial values ​​of the objective examination; in this embodiment... The value is dynamically configured based on the patient's age characteristics. Specifically: when the patient is in the adolescent development period (e.g., ≤18 years old), the value is set to... The value is set to a low weight (e.g., 0.1-0.5) to ensure the system prioritizes meeting the safety hard constraints of medical myopia prevention progress; when the patient's age is in the stable adult stage (e.g., >18 years old) and is a follow-up visit patient, the value is set to... The value of is given a high weight (e.g., 5.0-10.0) so that the prescription optimization process is as close as possible to the patient's original refractive habits (i.e., reducing susceptibility to refractive errors). (Deviation), avoiding changes to the patient's original visual habits that could cause discomfort. The above method transforms the medical multi-objective conflict problem into a clinical parameter numerical optimization problem;

[0072] In step S350, the conflict detection and parameter optimization module 300 extracts the current continuous variable values ​​of all state space nodes in the multidimensional physiological constraint association set into a multidimensional state vector, and the conflict detection and parameter optimization module 300 initializes the approximation step counter. Optimization step size parameters (The default initial value is) Convergence threshold (Default value is) ) and maximum number of iterations (Default value is) (times), wherein each dimension of the multidimensional state vector is normalized according to its corresponding clinical manufacturing step standard value, so that the gradient updates of each dimension of spherical power, cylindrical power, astigmatism axis and pupillary distance are on the same order of magnitude. By combining discrete node variables into a multidimensional state vector and initializing the algorithm hyperparameters, a data foundation is established for the operation of parameter approximation in the prescription optimization process;

[0073] In step S360, the conflict detection and parameter optimization module 300, based on the constructed prescription comprehensive evaluation function, calculates the clinical parameter deviation gradient of the prescription comprehensive evaluation function with respect to the multidimensional state vector using analytical differentiation. The conflict detection and parameter optimization module 300 then merges the clinical parameter deviation gradient results of each dimension to form the prescription parameter adjustment direction vector under the current approximation step. The prescription parameter adjustment direction vector characterizes the severity and direction of deterioration of visual discomfort and violation of medical rules caused by the current prescription parameter combination, and determines the approximation trajectory of the prescription comprehensive evaluation function in the parameter space;

[0074] In step S370, the conflict detection and parameter optimization module 300 performs a numerical approximation operation on the multidimensional state vector along the reverse of the prescription parameter adjustment direction vector, using a fine-tuning step size, to simulate the adjustment process of adding or subtracting lenses in clinical optometry; the numerical approximation formula configured in the conflict detection and parameter optimization module 300 is as follows:

[0075]

[0076] In the formula, Represents the multidimensional state vector for the next approximation step; This represents the multidimensional state vector of the current approximation step. This represents the optimization step size parameter.

[0077] In particular, when the multidimensional state vector It includes continuous variables characterizing the cylindrical lens astigmatism axis. To avoid gradient direction calculation errors caused by axis periodicity (0° and 180° are physically equivalent) in the optimization algorithm, a periodic boundary mapping calculation with a modulus of 180 is performed on the astigmatic axis variable after each numerical approximation, i.e.:

[0078]

[0079] The conflict detection and parameter optimization module 300 compares the values ​​in the multidimensional state vector of the next approximation step with the physiological safety lower bound and physiological safety upper bound configured in the state space node. If the values ​​in the multidimensional state vector exceed the physiological safety lower bound or physiological safety upper bound, the projection gradient method is used to directly truncate the excess values ​​to the corresponding physiological safety lower bound or physiological safety upper bound. The conflict detection and parameter optimization module 300 remaps the truncated multidimensional state vector of the next approximation step back to the multidimensional physiological constraint association set. By introducing a physiological boundary forced truncation mechanism, the clinical safety baseline is guaranteed.

[0080] In step S380, the conflict detection and parameter optimization module 300 recalculates the prescription comprehensive evaluation function value using the updated multidimensional state vector. The conflict detection and parameter optimization module 300 compares the prescription comprehensive evaluation function value with the preset convergence threshold. When the prescription comprehensive evaluation function value is less than or equal to the preset convergence threshold, the conflict detection and parameter optimization module 300 stops the calculation and outputs the current multidimensional state vector as the steady-state target solution vector to the dynamic tolerance module 400 during the lens adaptation period, ensuring that the steady-state target solution vector meets the safe error range of the ophthalmic medical rules.

[0081] In step S390, when the value of the prescription comprehensive evaluation function is greater than the preset convergence threshold, the conflict detection and parameter optimization module 300 increases the value of the approximation step counter. The conflict detection and parameter optimization module 300 controls the system to return to step S360 and uses the updated multidimensional state vector to perform a new round of rate of change calculation and numerical approximation operation. By configuring the closed-loop judgment logic of the return calculation, the continuous operation of the optimization iteration process is ensured until the multidimensional state vector reaches a medical equilibrium state in the multidimensional physiological constraint association set.

[0082] Furthermore, step S400 specifically involves the insertion logic of the temporal slice partitioning and feedforward temporary transition tolerance mapping relationship, as well as the constraint connection release and state restoration rules that evolve over time, including the following steps:

[0083] In step S410, the dynamic tolerance module 400 for the adaptation period of wearing glasses obtains the steady-state target solution vector output by the conflict detection and parameter optimization module 300. The dynamic tolerance module 400 for the adaptation period of wearing glasses extracts the initial value of the standardized parameter matrix output by the data access module 100. The dynamic tolerance module 400 for the adaptation period of wearing glasses performs numerical difference processing on each value in the steady-state target solution vector and each value in the initial value of the standardized parameter matrix to generate a parameter difference vector. This quantifies the span between the prescription parameters in the steady-state target solution vector and the patient's current initial physiological state, providing an objective data basis for the subsequent division of the adaptation period.

[0084] In step S420, the dynamic tolerance module 400 of the lens adaptation period inputs the parameter difference vector into a preset physiological tolerance mapping table. This physiological tolerance mapping table is pre-constructed by system developers based on clinical ophthalmological progressive fitting guidelines. Its recorded fields include: age range, maximum monocular spherical power change range, total adaptation period duration, and number of slides. For example: for patients under 18 years of age, if the monocular spherical power change is ≤1.50 diopters, the adaptation period is set to 7 days and 2 slides are used; if the change is between 1.50 and 3.00 diopters, the adaptation period is set to 14 days and 4 slides are used; if the change is >3.00 diopters, the system triggers manual review and blocking. When the parameter difference vector involves simultaneous changes in multiple dimensions (such as spherical and cylindrical lenses), the dimension with the largest weighted change is used as the index for mapping table matching. The dynamic tolerance module 400 for the adaptation period of wearing glasses performs a matching query in the preset physiological tolerance mapping table, extracts the total duration parameter and the number of slices parameter that match the parameter difference vector, and divides the time axis of the adaptation period of wearing glasses into a time slice sequence containing multiple time slices based on the total duration parameter and the number of slices parameter. This ensures the objective clinical basis for the division of time slices and eliminates the subjective error caused by uniformly setting the adaptation period.

[0085] In step S430, in the early time slice of the time slice sequence, the dynamic tolerance module 400 of the lens adaptation period generates a feedforward temporary transition period tolerance mapping relationship in memory. The dynamic tolerance module 400 of the lens adaptation period configures a parameter smoothing constraint equation within the feedforward temporary transition period tolerance mapping relationship. The parameter smoothing constraint equation is used to limit the absolute value of the difference between the state variable value under the current time slice and the initial value of the standardized parameter matrix to be less than or equal to the dynamic tolerance threshold. The dynamic tolerance module 400 of the lens adaptation period determines the dynamic tolerance threshold according to the slice number of the time slice sequence. By configuring the parameter smoothing constraint equation containing the dynamic tolerance threshold, a progressive tolerance mechanism that restricts the mutation of prescription parameters is established in the model.

[0086] In step S440, the dynamic tolerance module 400 of the lens adaptation period inserts the feedforward temporary transition period tolerance mapping relationship into the multidimensional physiological constraint association set. The dynamic tolerance module 400 of the lens adaptation period establishes a connection between the feedforward temporary transition period tolerance mapping relationship and the state space nodes in the parameter difference vector whose change amplitude exceeds the physiological tolerance threshold. After inserting the feedforward temporary transition period tolerance mapping relationship, the dynamic tolerance module 400 of the lens adaptation period sends a trigger control command to the conflict detection and parameter optimization module 300. After receiving the trigger control command, the conflict detection and parameter optimization module 300 re-executes the numerical approximation operation and outputs the transition solution vector under the current time slice. This ensures that the optometry intelligent decision-making system can output a transition period prescription that conforms to the principle of gradual medicine and reduces the risk of eye discomfort caused by prescription parameter jumps.

[0087] In step S450, the dynamic tolerance module 400 for the adaptation period of wearing glasses extracts the next adjacent time slice from the time slice sequence of the current time slice, extracts the slice number corresponding to the next adjacent time slice, extracts the monotonically increasing mapping table configured inside the preset physiological fitness function, performs a query in the monotonically increasing mapping table according to the slice number corresponding to the next adjacent time slice to obtain the corresponding step size coefficient, and multiplies the obtained step size coefficient with the basic tolerance parameter to calculate the relaxation tolerance threshold, which provides a numerical basis for subsequent adjustment of the multidimensional physiological constraint association set.

[0088] In step S460, the dynamic tolerance module 400 for the adaptation period of glasses obtains the parameter difference vector. The parameter difference vector is the difference between each value in the steady-state target solution vector and each value in the initial value of the standardized parameter matrix. The dynamic tolerance module 400 for the adaptation period of glasses performs a numerical comparison between the relaxed tolerance threshold and the absolute value of the value in the parameter difference vector. When the dynamic tolerance module 400 for the adaptation period of glasses determines that the relaxed tolerance threshold is greater than or equal to the absolute value of the value in the parameter difference vector, the dynamic tolerance module 400 for the adaptation period of glasses generates a constraint connection release instruction, thereby avoiding the system from performing invalid time slice calculations.

[0089] In step S470, the dynamic tolerance module 400 during the adaptation period of wearing glasses issues a constraint connection release command to the multidimensional physiological constraint association set. Based on the constraint connection release command, the dynamic tolerance module 400 during the adaptation period of wearing glasses locates the feedforward temporary transition period tolerance mapping relationship of the internal connection of the multidimensional physiological constraint association set. The dynamic tolerance module 400 during the adaptation period of wearing glasses releases the network connection relationship between the feedforward temporary transition period tolerance mapping relationship and the state space node, and deletes the feedforward temporary transition period tolerance mapping relationship after the network connection relationship is released from the multidimensional physiological constraint association set. By releasing the network connection relationship and deleting the temporary constraint, the transition constraint imposed inside the system is removed, and the underlying data structure is restored to a steady state.

[0090] In step S480, after the feedforward temporary transition period tolerance mapping relationship is deleted, the dynamic tolerance module 400 of the glasses-wearing adaptation period determines that the physiological adaptation period has ended. The dynamic tolerance module 400 of the glasses-wearing adaptation period stops sending trigger control commands to the conflict detection and parameter optimization module 300. The dynamic tolerance module 400 of the glasses-wearing adaptation period directly outputs the initially obtained steady-state target solution vector as the stage target under the final time slice, ensuring that the system accurately outputs the final target prescription after the adaptation period ends, avoiding redundant iterative calculations.

[0091] Example 5

[0092] Please refer to the appendix. Figure 9 and attached Figure 10 In step S500 above, the specific closed-loop mechanism for prescription steady-state verification and fatigue tolerance verification includes the following steps:

[0093] Step S510: The eye fatigue tolerance verification module 500 acquires the transitional solution vector and steady-state target solution vector output by the dynamic tolerance module 400 during the lens adaptation period. The eye fatigue tolerance verification module 500 extracts the age value and accommodation amplitude check value from the standardized parameter matrix. The eye fatigue tolerance verification module 500 inputs the age value and accommodation amplitude check value into a preset accommodation fluctuation mapping table, extracts the corresponding physiological accommodation fluctuation parameter benchmark, and generates a fatigue offset component based on the physiological accommodation fluctuation parameter benchmark. The fatigue offset component characterizes the refractive power fluctuation value under ciliary muscle fatigue state, providing a data basis for verifying the anti-fatigue robustness and stability of the prescription parameters; Specific implementation The extraction logic of the physiological regulation fluctuation parameter benchmark and the generation logic of the fatigue offset component are as follows: When the age value is greater than or equal to 40 years old (i.e., there is a presbyopia trend), and the objective regulation amplitude test value is less than the normal lower limit for the same age group (e.g., regulation amplitude AMP < 3.00D), the patient is judged to have weak ciliary muscle compensation ability. At this time, the extracted physiological regulation fluctuation parameter benchmark is a negative fluctuation, and the generated fatigue offset component is a positive lens compensation offset value of +0.50D to +0.75D superimposed on the spherical lens power state variable (Sph); when the age value is less than 18 years old, and there is regulation overlimit or regulation tension, the generated fatigue offset component is a negative regulation sluggish fluctuation value with the regulation force state variable introduced into -1.00D. Through the above specific numerical mapping, an extreme test scenario that truly reflects ciliary muscle fatigue or spasm is constructed in the multidimensional parameter space.

[0094] In step S520, the eye fatigue tolerance verification module 500 injects the fatigue offset component into the transition solution vector and the steady-state target solution vector. Specifically, the eye fatigue tolerance verification module 500 extracts the values ​​of the mapping spherical power prescription variable and the accommodation force state variable from the transition solution vector and the steady-state target solution vector, and performs a numerical addition operation with the fatigue offset component and the extracted values ​​to generate a test solution vector. This constructs a parameter test boundary with fatigue offset, realizing the dynamic testing of the static solution vector.

[0095] Step S530: The eye fatigue tolerance verification module 500 substitutes the test solution vector into the clinical medical inequality within the multidimensional physiological constraint association set for verification calculation. The eye fatigue tolerance verification module 500 reads the verification calculation result of the clinical medical inequality and determines whether the verification calculation result satisfies the requirements. The boundary conditions were examined to see if prescription parameters would breach the medical safety baseline when faced with fluctuations in physiological values.

[0096] Step S540, when the verification calculation result does not meet the requirements When the boundary conditions are met, the eye fatigue tolerance verification module 500 determines that the test solution vector has failed verification. The eye fatigue tolerance verification module 500 locates the variable constraint mapping relationship that does not meet the boundary condition verification calculation result. The eye fatigue tolerance verification module 500 extracts the current penalty weight coefficient of the located variable constraint mapping relationship. The eye fatigue tolerance verification module 500 multiplies the extracted penalty weight coefficient with the preset amplification factor to obtain the updated penalty weight coefficient. When the number of times feedback is triggered for the same transition solution vector reaches the preset maximum recalculation threshold, it is determined that the prescription cannot meet the anti-fatigue requirements under the current constraints. The system triggers the safety degradation mechanism and outputs a conservative target prescription with a safety margin to avoid falling into a calculation dead loop. If the maximum recalculation threshold is not reached, the eye fatigue tolerance verification module 500 feeds back the updated penalty weight coefficient to the conflict detection and parameter optimization module 300, thus constructing a closed-loop feedback adjustment logic between the reverse verification module and the forward solution module.

[0097] In step S550, the conflict detection and parameter optimization module 300 receives the updated penalty weight coefficients. The conflict detection and parameter optimization module 300 replaces the original penalty weight coefficients corresponding to the multidimensional physiological constraint association set with the updated penalty weight coefficients. The conflict detection and parameter optimization module 300 reconstructs the prescription comprehensive evaluation function based on the replaced multidimensional physiological constraint association set and re-executes the numerical approximation operation, forcing the conflict detection and parameter optimization module 300 to actively avoid parameter solution paths with poor anti-fatigue stability in the multidimensional parameter space, thereby improving the medical reliability of prescription parameters.

[0098] Step S560, when the verification calculation result satisfies When the boundary conditions are met, the eye fatigue fault tolerance verification module 500 determines that the test solution vector has passed the verification. The eye fatigue fault tolerance verification module 500 marks the transition solution vector and the steady-state target solution vector corresponding to the verified test solution vector as valid states. The eye fatigue fault tolerance verification module 500 outputs the transition solution vector and the steady-state target solution vector in the valid state to the multi-stage prescription generation module 600, ensuring the safe state of the final output parameters of the optometry intelligent decision system.

[0099] Furthermore, step S600 above specifically involves the assembly and parsing output process of multi-stage progressive prescription data, which includes the following steps:

[0100] In step S610, the multi-stage prescription generation module 600 receives the transitional solution vector and the steady-state target solution vector in an effective state. The multi-stage prescription generation module 600 extracts the time slice sequence number corresponding to the transitional solution vector and the steady-state target solution vector. The multi-stage prescription generation module 600 performs a sorting operation on the transitional solution vector and the steady-state target solution vector in ascending order of the time slice sequence number to generate a solution vector time series, which provides time series data support for progressive fitting.

[0101] In step S620, the multi-stage prescription generation module 600 calls a preset variable mapping index table. Based on the preset variable mapping index table, the multi-stage prescription generation module 600 reverse-maps the values ​​of each solution vector in the solution vector time series to the corresponding prescription dimensions. The multi-stage prescription generation module 600 extracts preset clinical manufacturing step standard values. The multi-stage prescription generation module 600 then uses these clinical manufacturing step standard values ​​(such as...) The refractive power is used to perform step-by-step quantization and rounding on the inversely mapped value to generate initial screening discrete optometry prescription parameters that meet clinical standards. Subsequently, the multi-stage prescription generation module 600 resubmits the initial screening discrete optometry prescription parameters into the clinical safety hard constraint equation in the multi-dimensional physiological constraint association set for secondary boundary verification. If the verification result exceeds the physiological safety boundary, the discrete step-by-step back mechanism is triggered. In the direction of reducing the absolute value of refractive power or reducing binocular disparity, the parameter that exceeds the error is shifted step by step to the next adjacent clinical manufacturing step standard value until the clinical safety hard constraint equation is satisfied again, thereby generating the final optometry prescription parameters. The algorithm solution vector in the continuous mathematical space is transformed into discrete prescription indicators that can be manufactured in the medical industry, ensuring the clinical physical executability of the system output results.

[0102] In step S630, the multi-stage prescription generation module 600 extracts the preset physical duration benchmark of the time slice, multiplies the preset physical duration benchmark of the time slice with the corresponding time slice number, calculates the recommended fitting time node corresponding to each optometry prescription parameter, and assembles the optometry prescription parameters and the recommended fitting time node to generate a staged optometry prescription parameter set, constructs a treatment plan with execution time nodes, and improves the clinical intervention process of optometry diagnosis and treatment.

[0103] In step S640, the multi-stage prescription generation module 600 sends the staged optometry prescription parameter set to the external display terminal. The external display terminal receives the staged optometry prescription parameter set and displays the parameter list on the display screen. The multi-stage prescription generation module 600 writes the staged optometry prescription parameter set into the local memory for archiving, realizing the output of decision results and data traceability, and supporting optometrists in prescription writing and subsequent review and comparison operations in actual clinical work.

[0104] It should be noted that traditional knowledge graphs mostly focus on semantic relationships within text. However, optometric decisions heavily rely on algebraic computations of continuous physiological data. Therefore, the 'knowledge graph' referred to in this invention is a computational knowledge graph, where nodes are state space nodes carrying continuous domains, and edges are constraint mapping relationships carrying clinical medical inequalities.

Claims

1. A knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system, characterized in that, include: The data access module (100) is used to acquire parameter data in the field of optometry and perform standardization processing to output a standardized parameter matrix; The knowledge graph construction module (200) is communicatively connected to the data access module (100) and is used to perform knowledge extraction and formal transformation on the rule text in the optometry field, generate computable constraints, and construct a multidimensional physiological constraint association set in memory, which includes a set of state space nodes and a set of variable constraint mapping relationships; the standardized parameter matrix is ​​mapped to the corresponding state space nodes; a single variable constraint mapping relationship in the set of variable constraint mapping relationships is connected to the state space nodes and is configured with the computable constraints; The conflict detection and parameter optimization module (300) is communicatively connected to the knowledge graph construction module (200). It is used to allocate penalty weights based on constraint priority when the current value of the state space node makes the computable constraint condition not valid, construct a comprehensive evaluation function based on the allocated penalty weights, perform numerical approximation on the variable value of the state space node along the parameter adjustment direction vector, and output the steady-state target solution vector.

2. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 1, characterized in that, When configuring the computable constraints, the knowledge graph construction module (200) is specifically used to: call a preset rule parsing template library to convert the threshold expression in the optometry domain rule library into a numerical calculation condition set; configure the numerical calculation condition set as the computable constraints and classify it into clinical safety hard constraints and comfort soft constraints.

3. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 1, characterized in that, The computable constraints are specifically clinical medical inequalities. When configuring the clinical medical inequalities, the knowledge graph construction module (200) is specifically used to: call a preset rule parsing template library to convert the threshold expressions in the optometry clinical rule library into a set of numerical calculation conditions; configure the set of numerical calculation conditions as the clinical medical inequalities and classify them into clinical safety hard constraints and comfort soft constraints.

4. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 1, characterized in that, When allocating penalty weights, the conflict detection and parameter optimization module (300) is specifically used to: substitute the current continuous variable values ​​of the state space nodes into the computable constraints to perform verification calculations, determine constraint conflicts and construct non-negative penalty terms; extract the constraint label categories carried by the constraint conflicts, match and extract the corresponding penalty weights in the constraint priority weight mapping table and assign values.

5. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 4, characterized in that, When performing numerical approximation, the conflict detection and parameter optimization module (300) is specifically used to: summarize the penalty terms and the penalty weights to construct the comprehensive evaluation function; calculate the parameter deviation gradient of the comprehensive evaluation function with respect to the multidimensional state vector, form the parameter adjustment direction vector, and perform reverse numerical approximation; If the approximated value exceeds the lower or upper bound of physiological safety, the projection gradient method is used to truncate it to the corresponding lower or upper bound of physiological safety, and then remap it back to the multidimensional physiological constraint association set.

6. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 1, characterized in that, The system also includes a dynamic tolerance module (400) for the adaptation period of wearing glasses, which is communicatively connected to the conflict detection and parameter optimization module (300). It is used to calculate the parameter difference between the steady-state target solution vector and the initial value of the standardized parameter matrix, divide the time slice sequence according to the parameter difference, insert a temporary transition period tolerance mapping relationship configured with parameter smoothing constraint equations in the early stage of the time slice sequence to trigger the recalculation of the transition solution vector, and remove the temporary transition period tolerance mapping relationship in turn as time goes by.

7. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 6, characterized in that, The dynamic tolerance module (400) for the adaptation period of wearing glasses is specifically used to: subtract the values ​​of each item in the steady-state target solution vector from the initial value of the standardized parameter matrix to generate a parameter difference vector; input the parameter difference vector into the physiological tolerance mapping table to match the total duration parameter of the adaptation period and the number of slices parameter, and divide the time slice sequence accordingly.

8. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 7, characterized in that, The system also includes an eye fatigue tolerance verification module (500), which is communicatively connected to the dynamic tolerance module (400) for the adaptation period of wearing glasses. It is used to inject fatigue offset components into the transition solution vector and the steady-state target solution vector, substitute them into the calculable constraints for verification, and increase the corresponding penalty weight when the verification fails and feed it back to the conflict detection and parameter optimization module (300) to solve again.

9. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 8, characterized in that, The system also includes a multi-stage prescription generation module (600), which is communicatively connected to the dynamic tolerance module (400) for the adaptation period of wearing glasses and the eye fatigue tolerance verification module (500), respectively. It is used to assemble the solution vectors that pass the verification in the order of time slices to generate a staged set of optometric prescription parameters and output them.

10. The knowledge graph-based intelligent ophthalmic diagnosis and treatment decision-making system according to claim 9, characterized in that, When generating a phased optometry prescription parameter set, the multi-stage prescription generation module (600) is specifically used to: extract time slice numbers and perform sorting operations on the transitional solution vector and the steady-state target solution vector; reverse map the values ​​of the solution vectors to the corresponding prescription dimensions, and perform step quantization and rounding based on the clinical manufacturing step standard value; calculate the recommended fitting time node in combination with the preset time slice physical duration benchmark, and assemble it with the quantized and rounded prescription parameters to form the phased optometry prescription parameter set.

Images