A system, method, and storage medium for intraocular lens clinical decision support that fuses patient cognitive augmentation with multi-source information integration

Abstract

A system, method and storage medium for intraocular lens clinical decision assistance by fusing patient cognition enhancement and multi-source information integration, comprising: a patient cognition enhancement module for improving the patient's cognition level of cataract disease and intraocular lens type through interactive information push technology, and inputting the cognition-enhanced patient to the subjective data collection link, so that the collected subjective data reflects the real needs of the patient after cognition enhancement; an objective eye condition analysis module for outputting an objective intraocular lens type recommendation result based on multi-modal preoperative examination data through an objective eye condition analysis model; a subjective will analysis module for outputting a subjective intraocular lens type adaptation result based on the collected subjective data through an integrated strategy combining a rule engine and a neural network model; and a comprehensive decision module for integrating the objective and subjective intraocular lens adaptation results and generating an intraocular lens recommendation report according to a decision logic.

Description

Technical Field

[0001] This invention relates to the field of medical assistive technology, and more specifically, to a system, method, and storage medium for intraocular lens clinical decision support that integrates patient cognitive enhancement and multi-source information integration. Background Technology

[0002] Intraocular lens (IOL) implantation is a crucial step in cataract surgery. With advancements in optical design, the variety of IOLs has increased significantly, including monofocal IOLs, astigmatic IOLs, and presbyopia-correcting IOLs (such as depth-of-field extension IOLs, bifocal IOLs, trifocal IOLs, and adjustable IOLs), providing patients with diverse options, each with its own advantages and disadvantages. Selecting the most suitable IOL for each patient, achieving precise and personalized decision-making, is a core challenge in clinical practice.

[0003] Currently, the selection of intraocular lenses (IOLs) primarily relies on the clinical physician's experience and judgment, combined with the patient's ocular biometric parameters and subjective preferences. With the development of artificial intelligence (AI) technology, using multimodal data fusion and intelligent algorithms to assist clinical decision-making is an important technological direction for improving the accuracy and efficiency of IOL selection. In the process of using information technology to assist in clinical decision-making regarding IOLs, the following technical needs and shortcomings exist:

[0004] First, there is the need to ensure the quality of patients' subjective information. Patients often lack sufficient understanding of cataracts and the indications, advantages, disadvantages, and postoperative risks of various intraocular lenses, which may lead to subjective information they provide that does not accurately reflect their true needs, affecting the accuracy of decision-making. The decision-making process remains "doctor-led" rather than "shared between doctor and patient."

[0005] Secondly, there is a need for refined analysis of objective eye examination data. Objective eye examination data includes parameters such as corneal curvature, total corneal astigmatism value and astigmatism type, as well as multi-dimensional parameters such as fundus condition, axial length, anterior chamber depth, and higher-order aberrations. This requires the integration of multi-source objective data for comprehensive analysis and decision-making. Astigmatism-correcting intraocular lenses (IOLs) are primarily decided based on a binary approach using parameters such as corneal curvature, total corneal astigmatism value, and astigmatism type. Presbyopia-correcting IOLs require a more comprehensive assessment, incorporating multiple dimensions including fundus condition, axial length, anterior chamber depth, and higher-order aberrations. A single analytical model is insufficient to simultaneously meet the specialized decision-making needs of both types of IOLs.

[0006] Third, there are the requirements for the safety and evolution of subjective intention analysis. Decision-making systems based on fixed rules lack the ability to continuously optimize from real-world data; decision-making systems based on a single machine learning model have technical risks such as unstable output and lack of safety constraints when conflicting with clinical expert consensus.

[0007] Fourth, the interpretability and safety requirements of comprehensive decision-making. Clinical decision-making requires traceable technological safeguards to ensure that the system can automatically execute a safety-first decision-making strategy when objective contraindications exist. Summary of the Invention

[0008] The present invention aims to overcome at least one of the defects (deficiencies) of the prior art mentioned above, and provides a system, method and storage medium for intraocular lens clinical decision support that integrates patient cognitive enhancement and multi-source information integration, in order to solve the technical problems of low patient participation, insufficient quality assurance of subjective information, insufficient refined analysis of objective data, insufficient security and evolvability of subjective intention analysis, and insufficient interpretability and security of comprehensive decision-making in intraocular lens clinical decision-making.

[0009] On one hand, the technical solution adopted by this invention is a system for clinical decision support of intraocular lenses that integrates patient cognitive enhancement and multi-source information integration, the system comprising: Patient cognition enhancement module: Connected to the subjective will analysis module, it is used to improve patients' cognitive level of cataract disease and intraocular lens type through interactive information push technology, and input the cognitively enhanced patients into the subjective data collection stage; the output of the patient cognition enhancement module is technically related to the subjective data collection stage, so that the collected subjective data reflects the patient's real needs after cognitive enhancement; The objective ocular condition analysis module, connected to the comprehensive decision-making module, is used to output objective intraocular lens type recommendations with interpretable indicators based on the collected multimodal preoperative examination data of patients and through an objective ocular condition analysis model trained with expert annotations. The subjective intention analysis module, connected to the patient cognition enhancement module, is used to output subjective intraocular lens type matching results based on the subjective data collected in the subjective data collection stage, through an integrated strategy combining a rule engine and a neural network model. The integrated decision-making module, connected to the objective eye condition analysis module and the subjective intention analysis module, integrates the objective intraocular lens type recommendation results with the subjective intraocular lens type matching results, and generates and outputs the final intraocular lens recommendation report according to a preset decision-making logic that prioritizes clinical safety.

[0010] The system described in this invention effectively addresses the industry pain points of traditional intraocular lens (IOL) clinical decision-making, such as high subjectivity, low data accuracy, and insufficient decision-making safety and iterative capabilities, through the coordinated operation of its various functional modules. The system relies on the interconnected design of the patient cognition enhancement module and the subjective data collection stage. Interactive science popularization pushes enhance patients' preoperative understanding, avoiding subjective biases caused by insufficient patient knowledge, and significantly improving the authenticity and effectiveness of subjectively collected data, providing reliable data support for subsequent decisions. Furthermore, the objective ocular condition analysis module uses an expert-annotated training model to analyze multimodal preoperative data, outputting recommendation results with interpretable indicators. This overcomes the shortcomings of traditional manual analysis, such as insufficient refinement and lack of basis for decision-making, improving the accuracy and credibility of objective ocular assessments. Simultaneously, the subjective intention analysis module combines a rule engine and neural network integration strategy, ensuring that clinical decisions align with expert consensus, are safe and stable, and endowing the system with continuous iterative optimization capabilities. Finally, the comprehensive decision-making module, with clinical safety as its core, integrates both subjective and objective analysis results to form a standardized and traceable decision-making system, effectively avoiding clinical decision-making risks and achieving personalized, precise, safe, and intelligent assisted decision-making for IOL selection.

[0011] Preferably, in the patient cognition enhancement module, the interactive information push technology includes: an interactive question-and-answer interface deployed with a natural language processing model, which provides patients with interactive question-and-answer services about the definition of cataracts, risk factors, diagnostic methods, treatment methods, and indications and contraindications of various intraocular lenses before the preoperative consultation, thereby improving patients' cognitive level of cataract disease and related knowledge about intraocular lenses.

[0012] The patient cognition enhancement module utilizes interactive information push technology and an interactive question-and-answer interface equipped with a natural language processing model to educate patients about cataracts and the compatibility of various intraocular lenses before surgery. The two-way question-and-answer format makes the information easier to understand. Patients can also fully understand the advantages and disadvantages of intraocular lenses and their applicable scope before filling out the questionnaire, significantly reducing the demand bias caused by cognitive blind spots, ensuring the authenticity and reliability of subjective feedback, and providing high-quality input data for subsequent subjective and objective integrated decision-making.

[0013] Preferably, the output of the patient cognition enhancement module is technically related to the subjective data collection stage, including: the content of the interactive information push is associated with the items of the structured questionnaire provided in the subjective data collection stage, so that the patient can make judgments based on the knowledge reserves obtained by the interactive information push when answering the structured questionnaire.

[0014] By linking interactive content push with structured questionnaire items, a technological link is established between cognitive enhancement and subjective data collection. Patients first learn relevant professional knowledge about the questionnaire, and when answering, they can objectively assess their own needs based on their existing knowledge of intraocular lens indications and visual impact, avoiding misjudgments and vague answers due to a lack of knowledge. This effectively improves the authenticity and matching degree of questionnaire data, eliminates the problem of subjective information distortion, and allows the subsequent subjective intention analysis module to receive input that matches the patient's true needs, ensuring the accuracy of overall clinical decision-making from the source.

[0015] Preferably, the objective ocular condition analysis module further includes a decision-making submodule for astigmatism-correcting intraocular lenses (IOLs) and a decision-making submodule for presbyopia-correcting IOLs. This allows for refined analysis of the selection of different types of IOLs, adapting to different clinical recommendation guidelines and avoiding assessment confusion caused by uniform modeling. It achieves dimensional and refined analysis of multimodal ocular data, outputting targeted and interpretable assessment indicators, significantly improving the accuracy of IOL selection for the two types, and providing a layered and reliable objective basis for subsequent comprehensive decision-making.

[0016] More preferably, the presbyopia-correcting intraocular lens decision submodule includes: The fundus image screening unit is used to analyze fundus images using a deep learning-based image analysis model to identify whether there are contraindications for implantation of presbyopia-correcting intraocular lenses and output preliminary screening results. The ocular bioparameter assessment unit, when the initial screening results allow, uses a machine learning classifier to output multi-level recommendation results and feature importance analysis for various types of presbyopia-correcting intraocular lenses (such as depth-of-field extension intraocular lenses, bifocal intraocular lenses, trifocal intraocular lenses, adjustable intraocular lenses, etc.).

[0017] In the presbyopia-correcting intraocular lens (IOL) decision-making submodule, a two-stage hierarchical evaluation architecture is adopted. First, deep learning is used to screen fundus images to quickly identify fundus contraindications and proactively intercept cases unsuitable for presbyopia-correcting IOLs, thus strengthening the first line of defense for clinical safety. When there are no contraindications in the fundus, multi-dimensional ocular bio-optical parameters are integrated, and machine learning is used to output graded recommendations with accompanying feature importance. This addresses the shortcomings of traditional evaluations, such as lack of quantitative evidence and inability to be interpreted, balancing screening safety with evaluation refinement, and providing traceable and highly reliable objective evaluation support for comprehensive decision-making.

[0018] More preferably, the astigmatism-correcting intraocular lens decision submodule is a first sub-machine learning model trained based on astigmatism-related parameters such as corneal curvature, total corneal astigmatism value, and astigmatism type. It is used to output a prediction result of whether to recommend implantation of an astigmatism-correcting intraocular lens and to output the ranking of the importance of features affecting the decision.

[0019] In the astigmatism-correcting intraocular lens (IOL) decision-making submodule, a dedicated first sub-machine learning model is trained based on core parameters such as corneal curvature, total corneal astigmatism value, and astigmatism type. This model is specifically matched with the clinical evaluation logic for astigmatism-correcting IOLs. The model can directly output implantation recommendations and provide a ranking of the importance of each astigmatism parameter, making the decision-making process visual and interpretable. It quantifies the impact of each parameter on the decision, improves the accuracy of astigmatism-correcting IOL selection, and provides objective and quantitative evaluation criteria for subsequent comprehensive decision-making and the integration of astigmatism functionality.

[0020] Preferably, the subjective intention analysis module includes: The rule engine unit is used to evaluate the answers to the structured questionnaire based on preset expert knowledge and preset decision logic, and output the first subjective fit tendency and its reasoning process; wherein, the structured questionnaire covers at least one dimension of ophthalmic surgery history, occupational characteristics, lifestyle habits, visual needs, psychological traits and economic budget. The neural network model unit is a student neural network model that learns the decision logic of the rule engine through model compression technology, and is used to output a second subjective adaptation tendency; the student neural network model has continuous learning ability and can iteratively optimize the rule weights by absorbing patient feedback and clinical experience data from the real world. An integrated decision-making unit is used to use the first subjective adaptation tendency as the subjective intraocular lens type adaptation result when the second subjective adaptation tendency is inconsistent with the first subjective adaptation tendency.

[0021] The subjective intention analysis module employs a dual-unit architecture combining a rule engine and a distilled neural network. The rule engine leverages the traceable adaptation tendencies of multi-dimensional questionnaires and clinical expert logical outputs to ensure that decisions align with clinical safety standards; a lightweight student model performs rapid parallel inferences. Furthermore, when the conclusions of the two systems conflict, the rule engine's results are forcibly adopted, mitigating the black-box risk of pure neural networks. Simultaneously, the neural network continuously replicates expert decision-making logic, balancing real-time evaluation efficiency with long-term iterative optimization capabilities, and ensuring the safety, interpretability, and scalability of subjective evaluations.

[0022] Preferably, the preset decision logic includes: When the objective intraocular lens type recommendation result indicates that there is a contraindication to implantation of a presbyopic intraocular lens, the subjective intraocular lens type matching result is disregarded, and a monofocal intraocular lens is directly recommended; When the objective intraocular lens type recommendation result allows for the implantation of a presbyopic intraocular lens, the final presbyopic intraocular lens type is determined by combining the subjective intraocular lens type matching result; After determining the type of monofocal or presbyopic intraocular lens, based on the prediction result of whether to recommend implantation of an astigmatic correction intraocular lens in the objective intraocular lens type recommendation results, an independent decision is made on whether to superimpose astigmatic correction function on the monofocal or presbyopic intraocular lens type.

[0023] The hierarchical decision-making logic described in this invention prioritizes objective ocular safety conditions, directly blocking presbyopic intraocular lenses (IOLs) when contraindications exist, thus avoiding ocular risks associated with improper implantation. Once ocular conditions are met, the type of IOL is matched based on the patient's subjective needs, balancing treatment safety with individual preferences. Furthermore, after selecting a monofocal or presbyopic IOL, astigmatism correction function is independently determined, achieving hierarchical functional selection. This standardizes and makes the entire process traceable, balancing clinical safety baselines with personalized needs, and the resulting recommended solutions are both medically rigorous and patient-appropriate.

[0024] On the other hand, the present invention also provides a method for clinical decision support of intraocular lenses based on the above-described system, which integrates patient cognitive enhancement and multi-source information integration, the method comprising the following steps: Step S1: Obtain multimodal preoperative examination data of the target patient, which includes at least ocular biometric parameters, wavefront aberration parameters, and fundus images; at the same time, provide the target patient with popular science information on cataract disease and intraocular lens types through interactive information push technology to improve the target patient's cognitive level. Step S2: After the cognitive enhancement in step S1, the target patient's answers to the structured questionnaire are collected as the subjective intention data; Step S3: Input the multimodal preoperative examination data into an objective ocular condition analysis model trained with expert-annotated data for processing and output of an objective intraocular lens type recommendation result; Step S4: Based on the subjective intention data, the data is processed using an integrated strategy combining a rule engine and a neural network model to obtain the subjective intraocular lens type matching result; Step S5: Integrate the objective intraocular lens type recommendation results output in step S3 with the subjective intraocular lens type matching results output in step S4, and then generate and output the final intraocular lens recommendation report for the target patient according to the preset decision logic that prioritizes clinical safety.

[0025] This method constructs a complete closed-loop intelligent recommendation process for intraocular lenses (IOLs). First, it simultaneously collects multimodal ocular examination data and conducts patient awareness education, addressing the issue of distorted subjective claims caused by insufficient patient understanding at the source, ensuring the authenticity and validity of subsequent questionnaire data. Next, subjective questionnaire data is collected after cognitive enhancement, achieving temporal linkage between cognitive enhancement and subjective information collection. Then, it employs layered operations through independently trained objective analysis models, rule engines, and neural network-integrated subjective analysis models. The objective layer refines the analysis of ocular physiological conditions, while the subjective layer relies on expert logic to ensure the safety and stability of the assessment. Finally, using clinical safety as the core criterion, it integrates the two types of analysis results to generate a standardized and complete recommendation report. This makes the entire process clear in sequence and logically progressive. It improves selection accuracy through pre-operative education and dual-dimensional data modeling, while relying on rigid safety decision-making logic to safeguard medical risks. The entire decision-making process is traceable and explainable, balancing clinical treatment standardization, individualized patient needs, and continuous system iteration capabilities. This significantly reduces the workload of physicians' manual assessments and improves the efficiency and safety of preoperative IOL selection for cataract surgery.

[0026] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing computer program instructions, which are loaded and run by a processor to form a user interaction terminal (such as a WeChat mini-program). When the computer program instructions are executed by the processor, a method for clinical decision support of intraocular lenses based on the above-described system, integrating patient cognitive enhancement and multi-source information integration, is implemented.

[0027] This storage medium stores program instructions that can generate user-interactive interfaces. Once the processor runs these instructions, the aforementioned complete clinical decision support method can be implemented. No additional dedicated software installation is required for patients or hospitals, resulting in low deployment barriers, ease of use, and wider applicability. The mini-program platform can smoothly support the entire workflow of interactive science popularization, questionnaire completion, data uploading, intelligent computing, and report output. The medium solidifies the entire set of standardized algorithm logic, ensuring stable and consistent program operation. It can fully reproduce all core steps, including cognitive enhancement, multimodal objective analysis, and subjective-objective fusion safe decision-making, guaranteeing consistent output results across different terminals and reducing the deployment and usage costs of the intelligent assistance system.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By introducing a patient cognition enhancement module, a technical chain of "cognition enhancement - data collection - intention analysis" was established, which enables the subjective data collection process to obtain high-quality input data that reflects the patient's real needs and meets the technical requirements for ensuring the quality of the patient's subjective information. 2. By constructing a classification decision architecture, the objective analysis of astigmatism-correcting intraocular lenses and presbyopia-correcting intraocular lenses is decoupled into sub-modules, which are adapted to different clinical recommendation guidelines and machine learning strategies respectively, thereby improving the accuracy of objective analysis and meeting the technical requirements for refined analysis of objective data. 3. By designing a dual-engine integration strategy that combines a rule engine and a neural network model, and setting a safe rollback mechanism in case of inconsistency, the stability and evolvability of subjective intention analysis are unified, thus meeting the technical requirements of security and evolvability of subjective intention analysis. 4. By formulating an explicit decision-making logic that prioritizes clinical safety and taking objective contraindications as an insurmountable decision boundary, an interpretable and safety-first recommendation output mechanism is formed, which meets the technical requirements of interpretability and safety in comprehensive decision-making. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the intraocular lens clinical decision support system that integrates patient cognitive enhancement and multi-source information integration provided by the present invention.

[0030] Figure 2 This is a flowchart illustrating the objective eye analysis module provided by the present invention.

[0031] Figure 3 This is a schematic diagram of the comprehensive decision-making process provided by the present invention.

[0032] Figure 4 This is a schematic diagram of the process of the intraocular lens clinical decision support method that integrates patient cognitive enhancement and multi-source information integration provided by the present invention. Detailed Implementation

[0033] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. To better illustrate the following embodiments, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0034] Example 1 like Figure 1 As shown, this embodiment provides a system for clinical decision support using intraocular lenses that integrates patient cognitive enhancement and multi-source information integration. The system includes: Patient cognition enhancement module: Connected to the subjective will analysis module, it is used to improve patients' cognitive level of cataract disease and intraocular lens type through interactive information push technology, and input the cognitively enhanced patients into the subjective data collection stage; the output of the patient cognition enhancement module is technically related to the subjective data collection stage, so that the collected subjective data reflects the patient's real needs after cognitive enhancement; Preferably, in the patient cognition enhancement module, the interactive information push technology includes: an interactive question-and-answer interface deployed with a natural language processing model, providing patients with interactive question-and-answer services regarding the definition of cataracts, risk factors, diagnostic methods, treatment methods, and indications and contraindications for various intraocular lenses before the preoperative consultation, thereby improving patients' cognitive level regarding cataracts and related intraocular lenses. Furthermore, the technical connection between the output of the patient cognition enhancement module and the subjective data collection stage also includes: the content of the interactive information push being associated with the items of the structured questionnaire provided in the subjective data collection stage, so that patients can make judgments based on the knowledge gained from the interactive information push when answering the structured questionnaire.

[0035] This system utilizes interactive information push technology and an interactive question-and-answer interface equipped with a natural language processing model to educate patients about cataracts and various intraocular lens (IOL) compatibility before surgery. The two-way question-and-answer format makes the information easier to understand. Patients can also fully understand the advantages and disadvantages of IOLs and their applicable scope before filling out the questionnaire, significantly reducing the need bias caused by cognitive blind spots and ensuring the authenticity and reliability of subjective feedback. This provides high-quality input data for subsequent subjective-objective integrated decision-making. Furthermore, the system uses a linked design that binds interactive push content with structured questionnaire items to establish a technical link between cognitive enhancement and subjective data collection. Patients first learn the relevant professional knowledge of the questionnaire, and when answering, they can objectively assess their own needs based on the information they have already mastered about IOL indications and visual impact, avoiding misjudgments and vague answers due to a lack of knowledge. This effectively improves the authenticity and matching degree of questionnaire data, eliminates the problem of subjective information distortion, and allows the subsequent subjective intention analysis module to obtain input that matches the patient's true needs, ensuring the accuracy of overall clinical decision-making from the source.

[0036] Specifically, in the patient cognition enhancement module described in this embodiment, the patient asks questions freely, covering the basic definition of cataracts, risk factors, diagnostic methods, treatments, and various intraocular lenses (IOLs), including monofocal IOLs, astigmatic correction IOLs, and presbyopia correction IOLs (such as depth-of-field extension IOLs, bifocal IOLs, trifocal IOLs, etc.), their indications, advantages, disadvantages, and potential risks (postoperative adverse visual phenomena, such as glare and halo). Furthermore, the technical association includes: the content of the interactive information push is linked to the items of the structured questionnaire, enabling the patient to make judgments based on the knowledge gained from the interactive information push when answering the structured questionnaire. For example, if the interactive Q&A explains to the patient that "bifocal IOLs can provide vision at both near and far distances to a certain extent, but may be accompanied by visual interference phenomena such as halos and glare," then the patient will make a more accurate self-assessment based on the aforementioned knowledge for subsequent structured questionnaire questions regarding "your need for night driving" and "your acceptance of halo phenomena." The patient cognition enhancement module dynamically adjusts the content and depth of subsequent information pushes based on the patient's feedback data during the interaction process to achieve personalized cognition enhancement. For example, if a patient shows confusion about the concept of "astigmatism correction" during the interaction, the module will automatically add push content about the principles of astigmatism and the working principle of astigmatism-correcting intraocular lenses; if the patient already shows a high level of understanding of a certain type of intraocular lens, relevant redundant information will be reduced to improve interaction efficiency.

[0037] Thus, through the above modules, patients have a basic understanding of the disease and intraocular lenses before the preoperative consultation, enabling them to more accurately express their true needs when providing subjective information later, and to communicate more proactively with doctors during the decision-making process, effectively making up for the technical shortcomings of the existing technology that create a cognitive gap for patients.

[0038] The objective ocular condition analysis module, connected to the comprehensive decision-making module, is used to output objective intraocular lens type recommendations with interpretable indicators based on the collected multimodal preoperative examination data of patients and through an objective ocular condition analysis model trained with expert annotations. This module requires the prior acquisition of relevant data, including multimodal preoperative examination data of the target patient. This preoperative multimodal examination data includes at least ocular biometric parameters, wavefront aberration parameters, and fundus images. The acquired multimodal preoperative examination data is then analyzed. For cataract patients' preoperative ocular biometric and aberration examination reports, this embodiment utilizes Optical Character Recognition (OCR) technology to identify important parameters in the examination reports. Through expert annotation, an objective ocular condition analysis model is trained, outputting an objective intraocular lens type recommendation result containing interpretability indicators. Unlike existing technologies that process ocular data as a whole, this embodiment also makes refined improvements to this module, such as… Figure 2 As shown, the objective eye condition analysis module is further divided into an astigmatism correction intraocular lens decision submodule and a presbyopia correction intraocular lens decision submodule to perform refined analysis on the selection of different types of intraocular lenses.

[0039] By separating astigmatism correction and presbyopia correction intraocular lenses into independent sub-modules for separate calculation, and adapting to the different clinical evaluation criteria for the two types of intraocular lenses, the system avoids the evaluation confusion caused by unified modeling. It achieves multi-dimensional and refined analysis of ocular multimodal data, outputs targeted and interpretable evaluation indicators, significantly improves the accuracy of intraocular lens selection, and provides a hierarchical and reliable objective basis for subsequent comprehensive decision-making.

[0040] More preferably, the astigmatism-correcting intraocular lens decision submodule is a first sub-machine learning model trained based on astigmatism-related parameters such as corneal curvature, total corneal astigmatism value, and astigmatism type. It is used to output a prediction result on whether to recommend implantation of an astigmatism-correcting intraocular lens and to output the ranking of the importance of features affecting the decision.

[0041] In the astigmatism-correcting intraocular lens (IOL) decision-making submodule, a first sub-machine learning model, trained based on core parameters such as the patient's corneal curvature, total corneal astigmatism value, and astigmatism type (with-the-rule / out-of-the-rule / oblique-axis), is specifically matched with the clinical evaluation logic for astigmatism-correcting IOLs. This model can directly output an implantation recommendation conclusion, i.e., a prediction of whether to recommend implantation of an astigmatism-correcting IOL. It also provides a ranking of the importance of each astigmatism parameter, making the decision-making process visual and interpretable, quantifying the impact of each parameter on the decision, improving the accuracy of astigmatism-correcting IOL selection, and providing objective and quantitative evaluation criteria for subsequent comprehensive decision-making and the integration of astigmatism functionality.

[0042] Specifically, in this embodiment, an ensemble learning method based on a gradient boosting framework (such as random forest, gradient boosting tree, or a combination thereof) is used to construct the first sub-machine learning model. The model input features include preprocessed standardized parameters such as corneal curvature, total corneal astigmatism value, and astigmatic axis. The model outputs predicted probabilities and feature contribution rankings. In internal validation, this sub-module demonstrated excellent discriminative ability, and its prediction results showed a high degree of consistency with clinical expert consensus.

[0043] The first sub-machine learning model was constructed using the LightGBM algorithm. The training dataset contained preoperative examination data from 3798 eyes, labeled by three senior refractive cataract specialists as "Recommended for Toric IOL implantation" and "Not Recommended for Toric IOL implantation" (binary labels). The model input features included corneal curvature, total corneal astigmatism value, and astigmatism type. On the independent test set, the model achieved an AUC of 98.9% and a recall of 97.9%, indicating a very high recognition rate for patients requiring astigmatism-correcting intraocular lenses. Feature importance analysis showed that the model's decisions were highly dependent on total corneal astigmatism value and astigmatism type, consistent with core recommendations in clinical guidelines. On an independent external validation set containing 800 eyes, the AUC reached 98.2%, confirming the module's good generalization ability.

[0044] More preferably, the presbyopia-correcting intraocular lens decision submodule includes: The fundus image screening unit is used to analyze fundus images using a deep learning-based image analysis model to identify whether there are contraindications for implantation of presbyopia-correcting intraocular lenses and output preliminary screening results. Specifically, the unit outputs a spatial attention heatmap corresponding to the initial screening results, highlighting the image areas of interest to the model (such as macular edema, drusen, etc.) to assist physicians in reviewing and judging, thereby providing doctors with an intuitive and interpretable basis for their judgments and avoiding the "black box" output problem in existing technologies.

[0045] The ocular bioparameter assessment unit is used to, when the initial screening results allow, take the collected patient's age and related bio-optical parameters such as axial length, anterior chamber depth, higher-order aberrations, Kappa angle, and pupil size, and output multi-level recommendation results and feature importance analysis through a machine learning classifier.

[0046] Specifically, the refined assessment unit for ocular biometric parameters, when the initial screening results allow, outputs multi-level recommendation results through a second sub-machine learning model, along with an index showing the relative contribution of each input feature to the decision result, providing deeper interpretability. In one embodiment, the second sub-machine learning model is constructed using an ensemble learning method (such as a combination of gradient boosting trees, support vector machines, or neural network classifiers). After feature selection, the input features include biometric parameters such as axial length, anterior chamber depth, corneal higher-order aberrations, and pupil diameter. The model outputs at least two levels of recommendation strength. On the validation set, this unit demonstrates excellent predictive performance, with its recommendation level matching expert decisions exceeding a preset threshold.

[0047] In this embodiment, the XGBoost algorithm was used to construct the second sub-machine learning model. The training dataset contained preoperative examination data from 3798 eyes, labeled with three levels of tags by three senior refractive cataract experts based on OCT screening results and biooptical parameters: "Not recommended for presbyopic IOL implantation," "Acceptable for presbyopic IOL implantation," and "Ideal conditions for presbyopic IOL implantation." The model input features, after feature selection, included: axial length, anterior chamber depth, higher-order aberrations of the whole eye, mesoscopic pupil size, and kappa angle. On the independent test set, the model achieved a macro-average AUC of 98.9% and a macro-average recall of 94.7%. Feature importance analysis showed that mesoscopic pupil size, higher-order aberrations of the whole eye, axial length, and kappa angle were the most critical features driving the model's decisions. On an independent external validation set containing 800 eyes, the model achieved a macro-average AUC of 99.9%, validating the module's good generalization ability.

[0048] Therefore, in the presbyopia-correcting intraocular lens (IOL) decision-making submodule, a two-stage hierarchical evaluation architecture is adopted. First, deep learning is used to screen fundus images to quickly identify fundus contraindications and intercept cases that are not suitable for presbyopia-correcting IOLs in advance, thus building a solid first line of defense for clinical safety. When there are no contraindications in the fundus, multi-dimensional ocular bio-optical parameters are integrated, and machine learning is used to output graded recommendations with feature importance. This addresses the shortcomings of traditional evaluations, such as lack of quantitative evidence and inability to be interpreted, and balances screening safety with evaluation refinement, providing traceable and highly reliable objective evaluation support for comprehensive decision-making.

[0049] The subjective intention analysis module, connected to the patient cognition enhancement module, is used to output subjective intraocular lens type matching results based on the subjective data collected in the subjective data collection stage, through an integrated strategy combining a rule engine and a neural network model. Preferably, the subjective intention analysis module includes: The rule engine unit is used to evaluate the answers to the structured questionnaire based on preset expert knowledge and preset decision logic, and output the first subjective fit tendency and its reasoning process; wherein, the structured questionnaire covers at least one dimension of ophthalmic surgery history, occupational characteristics, lifestyle habits, visual needs, psychological traits and economic budget. Specifically, within the rule engine unit, a multi-dimensional structured questionnaire was designed based on the clinical decision-making logic of cataract experts. This questionnaire covers multiple dimensions, including ophthalmic surgical history, occupational characteristics, lifestyle habits, visual needs, psychological traits, and economic budget. The rule engine employs a multi-step reasoning logic: first, it performs a safety rejection screening to exclude intraocular lens types with clear contraindications; second, it performs need-product matching based on the patient's primary eye-use scenarios; and finally, it fine-tunes the results by incorporating the patient's psychological traits, generating a first subjective fit preference result and its clear reasoning process.

[0050] The neural network model unit is a student neural network model that learns the decision logic of the rule engine through model compression technology, and is used to output a second subjective adaptation tendency; wherein, the model compression technology includes at least one of knowledge distillation, network pruning or quantization.

[0051] Specifically, in the neural network model unit, an auxiliary decision-making model trained using the decision logic of the rule engine unit is simulated through knowledge distillation technology. This model learns the input-output mapping relationship of the rule engine, enabling rapid and batch subjective intention assessment and outputting a second subjective fit tendency result. In one embodiment, a lightweight neural network is trained using a knowledge distillation framework. The output of the rule engine is used as a guiding signal to train a student neural network model containing several network layers. The input of the student neural network model is consistent with the rule engine, and the output is the fit probability distribution of various intraocular lenses. The compressed model maintains consistency with the teacher model while significantly improving inference speed, making it suitable for high-concurrency clinical scenarios. On the test set, the student neural network model maintains a high degree of consistency with the rule engine output, and the prediction deviation for each intraocular lens type is controlled within a clinically acceptable range.

[0052] An integrated decision-making unit is used to use the first subjective adaptation tendency as the subjective intraocular lens type adaptation result when the second subjective adaptation tendency is inconsistent with the first subjective adaptation tendency.

[0053] Specifically, in practical applications, a rule engine and a neural network model output first and second subjective fitting propensity results, respectively. When the first and second subjective fitting propensity results are inconsistent, the result output by the rule engine unit, representing expert consensus, is taken as the final result to ensure the stability and safety of clinical decision-making. Simultaneously, the neural network model unit continuously receives feedback from real-world clinical data, constantly training and iteratively optimizing its decision-making logic. With the accumulation of data and the evolution of the model, this student neural network model possesses continuous learning capabilities, enabling iterative optimization by absorbing real-world patient feedback and clinical experience data. It is expected to gradually integrate expert knowledge and real-world evidence, achieving dynamic optimization of the rule engine's decision-making logic, thereby providing patients with more accurate and adaptive subjective preference fitting results while ensuring safety.

[0054] Therefore, in the subjective intention analysis module, a dual-unit architecture combining a rule engine and a distilled neural network is adopted by integrating decision-making strategies. The rule engine relies on the traceable adaptation tendency of multi-dimensional questionnaires and clinical expert logic outputs to ensure that decisions conform to clinical safety standards; a lightweight student model performs rapid parallel inference. Furthermore, when the conclusions of the two conflict, the rule engine result is forcibly adopted to avoid the black box risk of pure neural networks. At the same time, the neural network continuously replicates the expert decision-making logic, balancing real-time evaluation efficiency and long-term iterative optimization capabilities, and taking into account the safety, interpretability, and scalability of subjective evaluation. This gives the system the potential for continuous iterative optimization from real-world data, while ensuring that the decision baseline does not deviate from expert knowledge under any circumstances, effectively meeting the technical requirements of safety and evolvability in subjective intention analysis.

[0055] The comprehensive decision-making module is connected to the objective eye condition analysis module and the subjective will analysis module, such as... Figure 3 As shown, the objective intraocular lens type recommendation results are integrated with the subjective intraocular lens type matching results, and a final intraocular lens recommendation report is generated and output according to a preset decision logic that prioritizes clinical safety.

[0056] Preferably, the integrated decision-making module follows the following preset decision-making logic: When the objective intraocular lens type recommendation result indicates a contraindication for implantation of a presbyopic intraocular lens, the subjective intraocular lens type matching result is disregarded, and a monofocal intraocular lens is directly recommended. When the objective intraocular lens type recommendation result allows for the implantation of a presbyopic intraocular lens, the final presbyopic intraocular lens type is determined by combining the subjective intraocular lens type matching result; wherein, the presbyopic intraocular lens includes at least one or more of bifocal intraocular lenses, trifocal intraocular lenses, depth-of-field extension intraocular lenses, or regional refractive intraocular lenses.

[0057] After determining the type of monofocal or presbyopic intraocular lens, based on the prediction result of whether to recommend implantation of an astigmatic correction intraocular lens in the objective intraocular lens type recommendation results, an independent decision is made on whether to superimpose astigmatic correction function on the monofocal or presbyopic intraocular lens type.

[0058] Specifically, the safety-first principle applies: when the objective ocular condition analysis module outputs a contraindication warning from the presbyopia correction intraocular lens decision submodule, or the fundus image screening unit outputs a contraindication warning, the system directly recommends a monofocal intraocular lens regardless of the subjective preference analysis module's output. This strategy ensures that patients with clear contraindications are not recommended for presbyopia correction intraocular lens implantation, maximizing clinical safety.

[0059] The system adopts a comprehensive decision-making principle that combines objective conditions and subjective preferences: When objective eye conditions permit (i.e., the recommended level of the presbyopia-correcting intraocular lens output by the presbyopia-correcting intraocular lens decision module reaches the preset threshold), the system further combines the results of subjective preference analysis to make a decision. When the subjective preference analysis module recommends a monofocal intraocular lens, the system ultimately recommends a monofocal intraocular lens, reflecting full respect for the patient's personal choice; When the subjective preference analysis module recommends presbyopia-correcting intraocular lenses such as depth-extending, bifocal, or trifocal lenses, the system adopts the subjective preference recommendation and recommends the corresponding type of presbyopia-correcting intraocular lens, reflecting the system's commitment to meeting patients' individual needs while maintaining safety.

[0060] Astigmatism correction function superposition principle: After determining the type of monofocal or presbyopic intraocular lens (IOL), the system further integrates the output of the astigmatism correction IOL decision submodule. If this submodule recommends implanting an astigmatism correction IOL, the final recommendation report will include a suggestion to superimpose astigmatism correction function on the determined monofocal or presbyopic IOL.

[0061] The final recommendation report includes the following core contents: the recommended type of intraocular lens, detailed reasons for the recommendation (based on the interpretability indicators output by the objective module and the reasoning process of the subjective module), the recommendation confidence level, and related precautions.

[0062] Thus, in this embodiment, through the above design, the system forms a complete, safe, and interpretable personalized intraocular lens recommendation solution, effectively meeting the technical requirements of comprehensive decision-making interpretability and safety, forming a standardized and traceable decision-making system, effectively avoiding clinical decision-making risks, and realizing personalized, precise, safe, and intelligent assisted decision-making for intraocular lens selection.

[0063] Example 2 like Figure 4 As shown, this embodiment provides a method for clinical decision support of intraocular lenses based on the system described in Embodiment 1, which integrates patient cognitive enhancement and multi-source information integration. The method includes the following steps: Step S1: Obtain multimodal preoperative examination data of the target patient, which includes at least ocular biometric parameters, wavefront aberration parameters, and fundus images; at the same time, provide the target patient with popular science information on cataract disease and intraocular lens types through interactive information push technology to improve the target patient's cognitive level. Preferably, in this embodiment, by deploying a large language model locally and pre-setting the role of "Cataract Specialist Medical Assistant", patients can freely ask questions through front-end interfaces such as mini-programs. The content covers the basic definition of cataracts, risk factors, diagnostic methods, treatment methods, and the indications, advantages, disadvantages, and potential risks (postoperative adverse visual phenomena, such as glare and halo) of various intraocular lenses (monofocal intraocular lenses, astigmatism-correcting intraocular lenses, depth-of-field-extending intraocular lenses, bifocal intraocular lenses, trifocal intraocular lenses, etc.).

[0064] Furthermore, the interactive information push content is linked to the questions in the structured questionnaire, enabling patients to make judgments based on the knowledge gained from the interactive information push when answering the structured questionnaire. For example, if the interactive Q&A explains to the patient that "depth-extending intraocular lenses can provide intermediate vision to a certain extent, but may be accompanied by a slight halo phenomenon," then in subsequent structured questionnaire questions about "your need for night driving" and "your acceptance of halo phenomena," the patient will make a more accurate self-assessment based on the aforementioned knowledge. Thus, based on the patient's feedback data during the interaction, the content and depth of subsequent information pushes are dynamically adjusted to achieve personalized cognitive enhancement. For example, if the patient shows confusion about the concept of "astigmatism correction" during the interaction, the system will automatically add content about the principles of astigmatism and the working principle of astigmatism-correcting intraocular lenses; if the patient already shows a high level of understanding of a certain type of intraocular lens, then relevant redundant information will be reduced to improve interaction efficiency.

[0065] Step S2: After the cognitive enhancement in step S1, the target patient's answers to the structured questionnaire are collected as the subjective intention data, so that they can more accurately express their true needs and communicate more proactively with doctors in the decision-making process, effectively making up for the technical defects of the patient cognitive gap in the existing technology.

[0066] Step S3: Input the multimodal preoperative examination data into an objective ocular condition analysis model trained with expert-annotated data for processing and output of an objective intraocular lens type recommendation result; Preferably, the objective intraocular lens type recommendation results include at least recommendations for astigmatism-correcting intraocular lenses and presbyopia-correcting intraocular lenses; specifically: In this embodiment, for recommendations regarding astigmatism-correcting intraocular lenses (IOLs), a first sub-machine learning model trained based on core parameters such as the patient's corneal curvature, total corneal astigmatism value, and astigmatism type (with-the-rule / out-of-the-rule / oblique-axis) is used to specifically match the clinical evaluation logic for astigmatism-correcting IOLs. This model can directly output an implantation recommendation conclusion, i.e., a prediction of whether or not to recommend implantation of an astigmatism-correcting IOL, while also providing a ranking of the feature importance of each astigmatism parameter, thus giving a recommendation for astigmatism-correcting IOLs. This visualizes and interprets the decision-making process, overcoming the limitations of traditional manual rough estimation of astigmatism, quantifying the influence of each parameter on the decision, improving the accuracy of astigmatism-correcting IOL selection, and providing an objective and quantitative evaluation basis for subsequent comprehensive decision-making and the integration of astigmatism function.

[0067] Regarding recommendations for presbyopic intraocular lenses (IOLs), this embodiment first employs a deep learning-based image analysis model to analyze fundus images to identify any contraindications for IOL implantation, outputting preliminary screening results. These preliminary screening results are presented in the form of a spatial attention heatmap, highlighting the image areas of interest (such as macular edema, drusen, etc.) to assist physicians in their review and judgment. This provides doctors with an intuitive and interpretable basis for their decisions, avoiding the "black box" output problem of existing technologies. Next, if the preliminary screening results indicate the absence of contraindications for IOL implantation, a detailed assessment of ocular bioparameters is performed, outputting multi-level recommendation results and feature importance analysis: three-level labels: "Presbyopic IOL implantation not recommended," "Presbyopic IOL implantation acceptable," and "Ideal conditions for presbyopic IOL implantation," along with feature importance, to obtain recommendations for presbyopic IOLs.

[0068] This approach employs a two-stage, hierarchical assessment framework. First, deep learning is used to screen fundus images, rapidly identifying contraindications and proactively intercepting cases unsuitable for presbyopia-correcting intraocular lenses, thus strengthening the first line of defense for clinical safety. When no contraindications are found in the fundus images, multi-dimensional ocular bio-optical parameters are integrated, and machine learning is used to output tiered recommendations with accompanying feature importance. This addresses the shortcomings of traditional assessments, such as lack of quantitative evidence and inability to be interpreted, balancing screening safety with assessment precision. It provides traceable and highly reliable objective assessment support for comprehensive decision-making.

[0069] Step S4: Based on the subjective intention data, the data is processed using an integrated strategy combining a rule engine and a neural network model to obtain the subjective intraocular lens type matching result; Preferably, in this embodiment, based on the clinical decision-making logic of cataract experts, a multi-dimensional structured questionnaire is designed, covering multiple dimensions such as ophthalmic surgery history, occupational characteristics, lifestyle habits, visual needs, psychological traits, and economic budget. The rule engine employs a multi-step reasoning logic: first, a safety rejection screening is performed to exclude intraocular lens types with clear contraindications; second, demand-product matching is performed based on the patient's main eye usage scenarios; finally, the results are fine-tuned in conjunction with the patient's psychological traits to generate a first subjective fitting tendency result and its clear reasoning process. Next, a neural network model is used to record discrepancies with the output of the preset rule engine in the background, and these discrepancies are used for subsequent iterations of the neural network model, thereby achieving rapid, batch subjective willingness assessment and outputting a second subjective fitting tendency result. Finally, an integrated decision is made; when the second subjective fitting tendency is inconsistent with the first subjective fitting tendency, the first subjective fitting tendency is used as the subjective intraocular lens type fitting result.

[0070] Therefore, in this step, a rule engine, backed by expert logic, is used to conduct tiered screening and matching to obtain traceable and appropriate results, ensuring clinical safety. Furthermore, a neural network is used for rapid batch analysis, retaining conflicting data for continuous iteration. When the conclusions of the two approaches conflict, the rule engine's output is adopted, balancing evaluation efficiency, decision stability, and long-term self-optimization capabilities, providing a safe and reliable subjective basis for integrated subjective and objective decision-making.

[0071] Step S5: Integrate the objective intraocular lens type recommendation results output in step S3 with the subjective intraocular lens type matching results output in step S4, and then generate and output the final intraocular lens recommendation report for the target patient according to the preset decision logic that prioritizes clinical safety.

[0072] Specifically, in this embodiment, the principles for pre-setting the decision logic in the above steps include: Safety First Principle: When contraindications exist or are indicated, monofocal intraocular lenses (IOLs) are directly recommended regardless of subjective IOL type suitability. This strategy ensures that patients with clear contraindications are not recommended for presbyopia correction IOL implantation, maximizing clinical safety.

[0073] Decision-making principle combining objective conditions and subjective wishes: When objective ocular conditions permit (i.e., the recommended level of presbyopia-correcting intraocular lenses reaches the preset threshold), decisions can be further made by combining the results of subjective wish analysis, including: When a monofocal intraocular lens is recommended, it will ultimately be recommended, reflecting full respect for the patient's personal choice. When recommending presbyopia-correcting intraocular lenses such as depth-extending, bifocal, or trifocal lenses, subjective recommendations are adopted, and the corresponding type of presbyopia-correcting intraocular lens is recommended, reflecting the satisfaction of patients' individual needs on the basis of safety.

[0074] The principle of superimposed astigmatism correction function: After determining the type of monofocal or presbyopic intraocular lens (IOL) mentioned above, further integrate the recommendations for astigmatism-correcting IOLs. If astigmatism-correcting IOLs are recommended, the final recommendation report will include suggestions to superimpose astigmatism correction function on the determined monofocal or presbyopic IOLs.

[0075] Finally, a final recommendation report is generated and output, which includes at least the following core contents: the recommended type of intraocular lens, detailed reasons for the recommendation (based on the interpretability indicators output by the objective module and the reasoning process of the subjective module), the recommendation confidence level, and related precautions.

[0076] In summary, the method described in this embodiment constructs a complete closed-loop intelligent recommendation process for intraocular lenses (IOLs). First, it simultaneously collects multimodal eye examination data and conducts patient awareness education, addressing the issue of distorted subjective claims caused by insufficient patient understanding at the source, ensuring the authenticity and validity of subsequent questionnaire data. Next, subjective questionnaire data is collected after cognitive enhancement, achieving temporal linkage between cognitive enhancement and subjective information collection. Then, it employs layered operations through independently trained objective analysis models, rule engines, and neural network-integrated subjective analysis models. The objective layer refines the analysis of ocular physiological conditions, while the subjective layer relies on expert logic to ensure the safety and stability of the assessment. Finally, using clinical safety as the core criterion, it integrates the two types of analysis results to generate a standardized and complete recommendation report. This makes the entire process clear in sequence and logically progressive, improving selection accuracy through pre-operative education and dual-dimensional data modeling, while adhering to rigid safety decision-making logic to safeguard medical risks. The entire decision-making process is traceable and explainable, balancing clinical treatment standardization, individualized patient needs, and continuous system iteration capabilities. This significantly reduces the workload of physicians' manual assessments and improves the efficiency and safety of preoperative IOL selection for cataract surgery.

[0077] Example 3 This embodiment provides a non-transitory computer-readable storage medium storing computer program instructions. The computer program instructions are loaded and run by a processor to form a user interaction terminal (e.g., a WeChat mini-program). When the computer program instructions are executed by the processor, a method for clinical decision support of intraocular lenses based on the above-described system, as described in Embodiment 2, is implemented.

[0078] In this embodiment, the computer program instructions within the storage medium are a mini-program code package adapted to the WeChat operating environment. After the processor reads the program instructions, the WeChat client parses and compiles them, loading and generating an interactive user-interactive front-end program. During the mini-program's operation, all processing steps described in Embodiment 2—a method for clinical decision support for intraocular lenses based on the aforementioned system that integrates patient cognitive enhancement and multi-source information—are completed sequentially. Furthermore, the non-transitory computer-readable storage media that can be used include, but are not limited to, persistent storage devices such as local hard drives, solid-state drives, USB flash drives, ROMs, cloud server disks, and memory cards.

[0079] Therefore, in this embodiment, the storage medium stores program instructions that can generate a user interaction terminal. After the processor runs these instructions, the aforementioned complete clinical decision support method can be implemented. The user terminal can be a WeChat mini-program, a mobile app, a web browser, etc. For example, relying on the WeChat ecosystem eliminates the need for patients and hospitals to install additional dedicated software, resulting in low deployment barriers, ease of use, and wider adoption. The mini-program platform can smoothly support the entire process of interactive science popularization, questionnaire completion, data uploading, intelligent computing, and report output. The medium solidifies the entire set of standardized algorithm logic, ensuring stable and unified program operation. It can completely reproduce all core steps such as cognitive enhancement, multimodal objective analysis, and subjective-objective fusion safe decision-making, ensuring consistent output results across different terminals. This facilitates lightweight implementation in primary ophthalmology clinics and significantly reduces the deployment and usage costs of the intelligent assistance system.

[0080] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A system for clinical decision support using intraocular lenses that integrates patient cognitive enhancement and multi-source information integration, characterized in that, The system includes: Patient cognition enhancement module: Connected to the subjective will analysis module, it is used to improve patients' cognitive level of cataract disease and intraocular lens type through interactive information push technology, and input the cognitively enhanced patients into the subjective data collection stage; the output of the patient cognition enhancement module is technically related to the subjective data collection stage, so that the collected subjective data reflects the patient's real needs after cognitive enhancement; The objective ocular condition analysis module, connected to the comprehensive decision-making module, is used to output objective intraocular lens type recommendations with interpretable indicators based on the collected multimodal preoperative examination data of patients and through an objective ocular condition analysis model trained by experts. The subjective intention analysis module, connected to the patient cognition enhancement module, is used to output subjective intraocular lens type matching results based on the subjective data collected in the subjective data collection stage, through an integrated strategy combining a rule engine and a neural network model. The integrated decision-making module, connected to the objective eye condition analysis module and the subjective intention analysis module, integrates the objective intraocular lens type recommendation results with the subjective intraocular lens type matching results, and generates and outputs the final intraocular lens recommendation report according to a preset decision-making logic that prioritizes clinical safety.

2. The system for clinical decision support of intraocular lenses that integrates patient cognitive enhancement and multi-source information integration according to claim 1, characterized in that, In the patient cognition enhancement module, the interactive information push technology includes: an interactive question-and-answer interface deployed with a natural language processing model, which provides patients with interactive question-and-answer services about the definition of cataracts, risk factors, diagnostic methods, treatment methods, and indications and contraindications of various intraocular lenses before the preoperative consultation, thereby improving patients' cognitive level of cataract disease and related knowledge about intraocular lenses.

3. The system for clinical decision support of intraocular lenses that integrates patient cognitive enhancement and multi-source information integration according to claim 2, characterized in that, The output of the patient cognition enhancement module is technically related to the subjective data collection stage, including: the content of the interactive information push is associated with the items of the structured questionnaire provided in the subjective data collection stage, so that the patient can make judgments based on the knowledge reserves obtained by the interactive information push when answering the structured questionnaire.

4. The system for clinical decision support of intraocular lenses that integrates patient cognitive enhancement and multi-source information integration according to claim 3, characterized in that, The objective eye condition analysis module also includes an astigmatism-correcting intraocular lens decision submodule and a presbyopia-correcting intraocular lens decision submodule to perform refined analysis on the selection of different types of intraocular lenses.

5. The system for clinical decision support of intraocular lenses that integrates patient cognitive enhancement and multi-source information integration according to claim 4, characterized in that, The decision-making submodule for presbyopia correction intraocular lenses includes: The fundus image screening unit is used to analyze fundus images using a deep learning-based image analysis model to identify whether there are contraindications for implantation of presbyopia-correcting intraocular lenses and output preliminary screening results. The ocular bioparameter assessment unit, when the initial screening results permit, uses a machine learning classifier to output multi-level recommendation results and feature importance analysis for various types of presbyopic intraocular lenses based on the collected patient's relevant biooptical parameters.

6. The system for clinical decision support of intraocular lenses integrating patient cognitive enhancement and multi-source information integration as described in claim 5, characterized in that, The astigmatism-correcting intraocular lens decision submodule is a first sub-machine learning model trained based on astigmatism-related parameters. It is used to output the prediction result of whether to recommend implantation of astigmatism-correcting intraocular lenses and to output the ranking of the importance of features affecting the decision.

7. A system for clinical decision support of intraocular lenses that integrates patient cognitive enhancement and multi-source information integration according to claim 6, characterized in that, The subjective intention analysis module includes: The rule engine unit is used to evaluate the answers to the structured questionnaire based on preset expert knowledge and preset decision logic, and output the first subjective fit tendency and its reasoning process; wherein, the structured questionnaire covers at least one dimension of ophthalmic surgery history, occupational characteristics, lifestyle habits, visual needs, psychological traits and economic budget. The neural network model unit is a student neural network model that learns the decision logic of the rule engine through model compression technology, and is used to output a second subjective adaptation tendency; the student neural network model has continuous learning ability and can iteratively optimize the rule weights by absorbing patient feedback and clinical experience data from the real world. An integrated decision unit is used to use the first subjective adaptation tendency as the subjective intraocular lens type adaptation result when the second subjective adaptation tendency is inconsistent with the first subjective adaptation tendency.

8. The system for clinical decision support of intraocular lenses integrating patient cognitive enhancement and multi-source information integration according to claim 7, characterized in that, The preset decision logic includes: When the objective intraocular lens type recommendation result indicates that there is a contraindication to implantation of a presbyopic intraocular lens, the subjective intraocular lens type matching result is disregarded, and a monofocal intraocular lens is directly recommended; When the objective intraocular lens type recommendation result allows for the implantation of a presbyopic intraocular lens, the final presbyopic intraocular lens type is determined by combining the subjective intraocular lens type matching result; After determining the type of monofocal or presbyopic intraocular lens, based on the prediction result of whether to recommend implantation of an astigmatic correction intraocular lens in the objective intraocular lens type recommendation results, an independent decision is made on whether to superimpose astigmatic correction function on the monofocal or presbyopic intraocular lens type.

9. A method for clinical decision support using an intraocular lens based on the system described in any one of claims 1-8, integrating patient cognitive enhancement and multi-source information integration, characterized in that, The method includes the following steps: Step S1: Obtain multimodal preoperative examination data of the target patient, which includes at least ocular biometric parameters, wavefront aberration parameters, and fundus images; at the same time, provide the target patient with popular science information on cataract disease and intraocular lens types through interactive information push technology to improve the target patient's cognitive level. Step S2: After the cognitive enhancement in step S1, the target patient's answers to the structured questionnaire are collected as the subjective intention data; Step S3: Input the multimodal preoperative examination data into an objective ocular condition analysis model trained with expert-annotated data for processing and output of an objective intraocular lens type recommendation result; Step S4: Based on the subjective intention data, the data is processed using an integrated strategy combining a rule engine and a neural network model to obtain the subjective intraocular lens type matching result; Step S5: Integrate the objective intraocular lens type recommendation results output in Step S3 with the subjective intraocular lens type matching results output in Step S4, and then generate and output the final intraocular lens recommendation report for the target patient according to the preset decision logic that prioritizes clinical safety.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer program instructions, which are loaded and executed by a processor to form a user interaction terminal. When the computer program instructions are executed by the processor, they realize the method for clinical decision support of intraocular lenses based on the above-described system as described in claim 9, which integrates patient cognitive enhancement and multi-source information integration.

Images