Ophthalmic postoperative eye medicine box intelligent auxiliary management system

Abstract

The application relates to the field of intelligent eye medicine boxes, and particularly relates to an intelligent auxiliary management system for an ophthalmic postoperative eye medicine box, which comprises the following steps: an administration data acquisition module extracts data features based on historical use information, and creates a feature knowledge graph based on the data features; a feature data optimization module is used for optimizing and selecting feature data based on the feature knowledge graph through a grey wolf optimization algorithm according to current user use information; a prediction model construction module is used for training a CNN model by taking the feature data after optimization and selection as input and taking the use length of the eye medicine box as output, so that an eye medicine box use length prediction model is obtained; and an application module is used for obtaining a prediction result according to the eye medicine box use length prediction model, and performing eye medicine box management based on the prediction result. The application predicts the use length of the eye medicine box by optimizing model training parameters, manages the eye medicine box according to the use length of the eye medicine box, and constructs an efficient and reliable intelligent auxiliary management system for the eye medicine box.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent eye medicine boxes, in particular to an intelligent auxiliary management system for eye medicine boxes after ophthalmic surgery. BACKGROUND

[0002] The medication management after ophthalmic surgery is a key link to ensure the recovery of patients. However, many elderly patients face complex medication regimens after discharge, and it is difficult for them to accurately follow the doctor's advice. In particular, for patients such as postoperative patients of cataract, glaucoma, etc. who need to use multiple eye drops and eye ointments frequently, the time and frequency of medication are strictly required, which can easily lead to medication errors. In addition, patients who need long-term medication also face similar challenges.

[0003] Although the hospital provides detailed medication instructions and labels, there are still many difficulties in the actual implementation process due to the decline of vision, memory loss and limited understanding of complex medication regimens of the elderly.

[0004] In order to solve the above problems, intelligent eye medicine boxes have emerged. The existing intelligent eye medicine boxes can store multiple eye drops and have a reminding function, such as a splicing type eye drop use intelligent reminding box with Chinese patent number CN220721761U, which includes a controller, a box combination, a splicing structure, and a conduction structure. The box combination includes several box bodies, the top of each box body is provided with a top cover which can be turned over, the controller is connected with each box body through the conduction structure in sequence and can control the opening and closing state of the top cover; the controller includes an operation module, a speaker module, and an indicator light module. The operation module is used to set the use time of eye drops, the indicator light module is arranged on the box body, and the controller can independently control the indicator light module on the box body to flash and emit a reminder sound through the speaker module and make the top cover in an open state; the splicing structure includes a first splicing surface and a second splicing surface, the controller and the box combination are arranged by splicing the first splicing surface and the second splicing surface, which facilitates installation and disassembly and can remind the patient to use eye drops in time.

[0005] Although the intelligent eye medicine box solves the problem of medication accuracy to some extent, the existing scheme only has one-sided reminders and lacks monitoring and feedback of the actual medication process. The whole-process collaboration effect is poor, and the traditional follow-up method cannot monitor the patient's medication in real time, which makes it difficult for doctors to discover and correct the patient's medication deviation in time. Therefore, there is an urgent need to design an intelligent eye medicine box management system. SUMMARY

[0006] The present application aims to provide an intelligent auxiliary management system for eye medicine boxes after ophthalmic surgery. By optimizing the model training parameters first, and then predicting the use time of the eye medicine box based on the optimized parameters, the management of the eye medicine box is carried out according to the use time of the eye medicine box, so as to construct an efficient and reliable intelligent auxiliary management system for eye medicine boxes.

[0007] To achieve the above object, the application adopts the following technical scheme:

[0008] The intelligent auxiliary management system for ophthalmic postoperative eye medicine box comprises:

[0009] The medicine data acquisition module is used for acquiring historical user use information and performing data feature extraction based on the historical use information, and creating a feature knowledge graph based on the data features;

[0010] The feature data optimization module is used for optimizing and selecting the feature data based on the feature knowledge graph through the grey wolf optimization algorithm according to the current user use information;

[0011] The prediction model construction module is used for taking the optimized and selected feature data as input and the use duration of the eye medicine box as output, training a CNN model, and obtaining an eye medicine box use duration prediction model;

[0012] The application module is used for obtaining a prediction result according to the eye medicine box use duration prediction model and performing eye medicine box management based on the prediction result.

[0013] The principle and advantages of the scheme are as follows: in actual application, the historical user use information can reflect the individual differences of users, the historical user use information is acquired, data feature extraction is performed based on the historical use information, and a feature knowledge graph is created based on the data features, which provides a comprehensive historical medicine data basis and makes individualized analysis possible, the feature knowledge graph can intuitively show the relationship between different features, and potential patterns can be found conveniently. The grey wolf optimization algorithm is a meta-heuristic optimization algorithm based on the hunting behavior of grey wolves in nature, which is used to find the optimal solution in a complex search space, the grey wolf optimization algorithm optimizes and selects the feature data to determine which features are most critical for predicting the use duration of the eye medicine box, and the actual use condition of the current user is adjusted in real time to ensure that the selected features always match the actual situation, improve the accuracy and efficiency of feature selection, reduce unnecessary computational overhead, and the dynamic adjustment mechanism enables the system to adapt to the specific needs of different users, enhancing the flexibility and applicability of the system; the convolutional neural network is a deep learning model that is particularly good at processing data with grid structure, taking the optimized feature data as input and the use duration of the eye medicine box as output, training a CNN model to predict the future use duration, using the powerful pattern recognition capability of CNN to improve the prediction accuracy, and the model can capture complex nonlinear relationships, improving the understanding and prediction ability of individual differences; based on the prediction result generated by the trained prediction model, corresponding management strategies are developed, such as reminding patients to take medicine on time and automatically supplementing drug inventory, realizing accurate management and individualized service for the use of the eye medicine box, and improving the medication compliance of patients.

[0014] Preferably, as an improvement, the medication data acquisition module comprises:

[0015] a knowledge representation submodule for defining entities and entity relationships;

[0016] a graph creation submodule for generating corresponding nodes for each entity type, assigning attributes, establishing connections according to the relationships between entities, and adding detailed attribute information to nodes and edges;

[0017] a knowledge reasoning submodule for automatically inferring new knowledge according to logical rules;

[0018] an effect evaluation submodule for evaluating the coherence, coverage, and effectiveness of the graph.

[0019] Technical effect: not only can the data features that help understand the patient behavior patterns be effectively extracted, but also the structured knowledge graph can help better understand and manage the eye medicine box usage.

[0020] Preferably, as an improvement, the feature data optimization module comprises:

[0021] an initialization submodule for randomly generating a set of candidate solutions, each solution representing a feature selection scheme;

[0022] a target definition submodule for determining a fitness function based on the prediction model benefit and determining boundary conditions;

[0023] a position update submodule for gray wolf hunting based on fitness, retaining the current optimal solution obtained by hunting, and updating the position of the wolf pack according to the positions of the current best alpha, beta, and delta wolves in each iteration until the termination condition is met.

[0024] Technical effect: the data optimization screening problem can be simulated as the hunting behavior of gray wolf populations in nature, and the optimal solution can be found through the stages of tracking, surrounding, and attacking based on the social hierarchy and hunting strategies of gray wolf populations.

[0025] Preferably, as an improvement, the fitness function comprises:

[0026] y = w1A ccuracy -w2C f +w3S tability -w4T imeC

[0027] where w1, w2, w3, and w4 represent weight coefficients, A ccuracy represents the classification accuracy, C f represents the feature subset size penalty, S tability represents the measure of model stability, and T imeC represents the time cost.

[0028]

[0029] C f = p * |S|

[0030] S tability = Mean(A ccuracy_i ) - k * Std(A ccuracy_i )

[0031] wherein A p represents the number of samples correctly predicted as positive class, A n represents the number of samples correctly predicted as negative class, f p represents the number of samples wrongly predicted as positive class, f n represents the number of samples wrongly predicted as negative class; p represents the penalty coefficient, |S| represents the size of feature subset, Mean() represents the mean function, Std() represents the standard deviation function, and k represents the adjustment parameter.

[0032] Technical effects: based on the above fitness function, the model performance can be improved.

[0033] Preferably, as an improvement, the boundary condition comprises:

[0034] s min ≤ s z ≤ s max

[0035] wherein s min and s max represent the minimum and maximum number of features of the feature set, and s z represents the actual number of features.

[0036] Technical effects: avoid too few or too many features leading to underfitting or overfitting.

[0037] Preferably, as an improvement, in the position updating submodule, the position coefficient vector and the distance coefficient vector are updated through chaotic mapping, the chaotic variable is initialized at the beginning of the algorithm, and a new chaotic value is generated through chaotic mapping at each iteration. The generated chaotic sequence is used to replace the random number generator, and the model is updated as follows:

[0038] C n = 2z i+1

[0039] A n = 2a * z i+1 -a

[0040] z i+1 = tau * z i (1 - z i )

[0041] wherein C n represents a position coefficient vector, A n represents a distance coefficient vector, τ represents a degree of controlling chaos, a represents a convergence factor, z i+1 represents a chaotic value.

[0042] Technical effects: By integrating the chaos mapping into the position updating process of the grey wolf optimization algorithm, the original algorithm advantages are maintained, and the ability to solve complex optimization problems is further improved, especially suitable for feature selection in high-dimensional optimization tasks. This not only enhances the exploration ability of the algorithm, but also provides a more effective solution for identifying abnormal medication types in the intelligent auxiliary management system of the postoperative eye medicine box.

[0043] Preferably, as an improvement, the prediction model construction module comprises:

[0044] The architecture building submodule defines the input layer size according to the data dimension, extracts feature patterns in the data in the convolution layer, reduces the spatial size of the feature map through average pooling in the pooling layer, maps the learned feature map to the specific output in the full connection layer, and outputs the specific duration in the output layer.

[0045] The training and verification submodule uses MSE as the loss function and Adam as the optimizer to train the model until convergence.

[0046] Technical effects: The key factors affecting user recovery can be deeply explored, so as to accurately predict the eye medicine box usage duration.

[0047] Preferably, as an improvement, it further comprises a follow-up recommendation module for comparing the target usage duration of the eye medicine box with the predicted usage duration and recommending follow-up when the predicted usage duration exceeds the threshold range.

[0048] Technical effects: When the predicted usage duration exceeds the threshold, it has a relatively high risk, and the follow-up recommendation is convenient for risk prevention.

[0049] Preferably, as an improvement, it further comprises a compliance analysis module for analyzing the medication compliance of the user, and the analysis model comprises:

[0050] Y c = θ d P d + θ l P l + θ c P c

[0051] wherein θ d , θ l , θ cP is a coefficient for adjustment d P represents long-term adherence rate l P represents frequency adherence rate c P represents time window adherence rate.

[0052] Technical effects: It is convenient to formulate corresponding intervention strategies as a basis.

[0053] Preferably, as an improvement, it also includes a warning module for warning when the adherence or the predicted eye drop box usage time exceeds a threshold.

[0054] Technical effects: It is convenient for the management end to adjust the reminder strategy of the intelligent eye drop box and guide the adherence. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 Figure 1 is a structural schematic diagram of an ophthalmic postoperative eye drop box intelligent auxiliary management system. DETAILED DESCRIPTION

[0056] The following will be further described in detail through specific embodiments:

[0057] The embodiments are basically as shown in the accompanying drawings. Figure 1 The ophthalmic postoperative eye drop box intelligent auxiliary management system includes:

[0058] A medication data acquisition module is configured to acquire historical user usage information, which includes user basic information, eye drop box reminder information, and medication information. The user basic information includes but is not limited to age, gender, diagnosis result, and medication target; the eye drop box reminder data includes but is not limited to reminder time and frequency; and the medication data includes but is not limited to medication time and dosage.

[0059] Data feature extraction is performed. First, the original data is cleaned and standardized to ensure the consistency and accuracy of the data, for example, all time formats are unified to "YYYY-MM-DD HH:MM", and missing values are filled or deleted; then, further feature extraction is performed, such as extracting time-related features (on-time rate, medication interval consistency), dose-related features (dose accuracy), and adherence-related features (adherence score).

[0060] The feature knowledge graph is created based on data features, specifically including: a knowledge representation submodule, used to define entities; in this embodiment, entities include users, drugs, medication events, health status, etc., and entity relationships include medication, belonging, influence, etc.; a graph creation submodule, used to generate corresponding nodes for each entity type and assign attributes (such as user ID, drug name, etc.), establish connections based on the relationships between entities, such as "user A used drug B", and add detailed attribute information (such as medication time, dosage, etc.) to nodes and edges; a knowledge reasoning submodule, used to automatically infer new knowledge "if a user fails to take medication on time three times consecutively, then mark it" based on logical rules; and an effectiveness evaluation submodule, used to evaluate the graph's coherence, coverage, and effectiveness. The coherence check helps ensure that the information in the knowledge graph is consistent, the coverage evaluation helps measure how much useful information the graph covers, and the application effectiveness test helps test the graph's effectiveness in assisting decision-making.

[0061] The feature data optimization module optimizes feature data based on current user usage information and the Gray Wolf optimization algorithm using a feature knowledge graph. For users of eye drop boxes with individual differences, such as age, medication habits, and medical conditions, the duration of use varies, affecting the patient's recovery progress and making it difficult to accurately predict the return time of the smart eye drop box. Therefore, accurate prediction of user eye drop box usage time can serve as a basis for the allocation and management of eye drop boxes, thereby improving management efficiency. However, in addressing the issue of eye drop box usage time, obtaining effective data for model training is a key issue affecting the accuracy of the prediction results.

[0062] The feature data optimization module includes an initialization submodule, a target definition submodule, and a position update submodule. The initialization submodule randomly generates a set of candidate solutions, each solution representing a feature selection scheme. The target definition submodule determines the fitness function and boundary conditions based on the prediction model's return. The fitness function includes:

[0063] y = w1A ccuracy -w2C f +w3S tability -w4T imeC

[0064] Where w1, w2, w3, and w4 represent weighting coefficients, and A ccuracy C represents the classification accuracy. f S represents the penalty for the size of the feature subset. tability A metric representing model stability, T helps ensure performance stability under different data splits. imeC Indicates time cost;

[0065]

[0066] C f = p * |S|

[0067] S tability = Mean(A ccuracy_i ) - k * Std(A ccuracy_i )

[0068] where A p represents the number of samples correctly predicted as positive class, A n represents the number of samples correctly predicted as negative class, f p represents the number of samples incorrectly predicted as positive class, f n represents the number of samples incorrectly predicted as negative class; p represents the penalty coefficient, |S| represents the size of the feature subset, Mean() represents the mean function, Std() represents the standard deviation function, and k represents the adjustment parameter.

[0069] In order to avoid too few or too many features leading to underfitting or overfitting, the boundary conditions are set as follows:

[0070] s min ≤ s z ≤ s max

[0071] where s min and s max represent the minimum and maximum number of features of the feature set, and s z represents the actual number of features.

[0072] The position updating submodule performs grey wolf hunting based on fitness, and retains the current optimal solution obtained by hunting. In each iteration, the position of the wolf pack is updated according to the positions of the current best alpha, beta and delta wolves, until the termination condition is met.

[0073] The distance calculation formula between the wolf pack and the optimal three grey wolves alpha, beta and gamma is:

[0074] D α = |C1*X α -X|;

[0075] D β = |C2*X β -X|;

[0076] D δ = |C3*X δ -X|;

[0077] D α , D β , D δ represent the distances between other omega wolves and the optimal three grey wolves alpha, beta and delta; X α , Xβ and X δ respectively represent the positions of alpha, beta and delta wolves, X represents the position of the current gray wolf individual; C1, C2 and C3 are position coefficient vectors, C1, C2 and C3 are random numbers between 0 and 2.

[0078] The position updating formula of the gray wolf is:

[0079] X 1＝ X α -A1D α ;

[0080] X 2＝ X β -A2D β ;

[0081] X 3＝ X δ -A3D δ ;

[0082]

[0083] wherein A1, A2 and A3 are distance coefficient vectors, and R1, R2 and R3 are gray wolf position dynamic modification weights.

[0084] In the embodiment, in order to enhance the diversity of the search process and avoid premature convergence, the position coefficient vector and the distance coefficient vector are updated based on chaotic mapping. Chaotic sequences can more effectively explore the solution space due to their inherent randomness and ergodicity. Specifically, at the beginning of the algorithm, chaotic variables are initialized, and a chaotic homeomorphism is selected, and the Logistic mapping is preferred in the embodiment; in each iteration, a new chaotic value is generated through chaotic mapping, and the generation of the chaotic value is as follows:

[0085] z i+1 =τ*z i (1-z i )

[0086] wherein τ represents the degree of controlling chaos.

[0087] The generated chaotic sequence is used to replace the random number generator to update the position coefficient vector and the distance coefficient vector:

[0088] A n =2a*z i+1 -a

[0089] z i+1 =τ*z i (1-z i )

[0090] wherein C n represents the position coefficient vector, and A ndenotes the distance coefficient vector, a denotes the convergence factor. The position update of the grey wolf is performed based on the updated position coefficient vector and the distance coefficient vector.

[0091] The prediction model construction module is configured to take the selected feature data as input and the use duration of the eye medicine box as output, train a CNN model, and obtain an eye medicine box use duration prediction model; the prediction model construction module includes an architecture building submodule and a training and verification submodule. The architecture building submodule defines the input layer size according to the data dimension, extracts the feature pattern in the data in the convolution layer, reduces the spatial size of the feature mapping through average pooling in the pooling layer, maps the feature mapping learned by the previous layer to a specific output in the full connection layer, and outputs a specific duration in the output layer; the training and verification submodule takes MSE as the loss function, Adam as the optimizer, and trains the model until convergence.

[0092] The revisit recommendation module is further included for comparing the target use duration of the eye medicine box with the predicted use duration and recommending a revisit when the predicted use duration exceeds a threshold range. The target use duration is a target recovery duration obtained from the diagnosis information.

[0093] For eye medicine box users with individual differences, the application requirements of the eye medicine box are different. Some users can take medicine on time and in the right amount through the reminders of the eye medicine box, some users can take medicine on time and in the right amount without the reminders of the eye medicine box, and some users cannot take medicine on time or in the right amount even with the reminders of the eye medicine box. Therefore, analyzing the abnormal types of the users' medication can serve as a basis for the distribution and management of the eye medicine box, thereby improving the management efficiency of the eye medicine box.

[0094] Considering the complexity of postoperative medication, such as time-sensitive medication (hormones, antibiotics, and eye pressure reducing drugs) after ophthalmic surgery, which needs to be strictly adjusted in stages (such as decreasing the frequency of medication in the first to fourth weeks after surgery); patients with chronic eye diseases such as glaucoma need long-term or even lifelong medication, and missing or incorrect use can easily lead to disease deterioration; the elderly have problems such as visual decline, memory confusion, and inconvenience in operation. The compliance analysis module is further included in the embodiment for analyzing the medication compliance of the user to facilitate the development of corresponding intervention strategies as a basis. The analysis model includes:

[0095] Y c = θ d P d + θ l P l + θ c P c

[0096] wherein θ d , θ l , θ c are adjustment coefficients, P d denotes the long-term compliance rate, and Pl denotes the frequency compliance rate, P c denotes the time window compliance rate.

[0097]

[0098] wherein P d_s denotes the actual covered days, P d_z denotes the total days; P l_s denotes the actual compliance frequency, P l_z denotes the should-compliance frequency, P c_s denotes the compliance frequency within the specified time window, P c_z denotes the should-compliance frequency within the specified time window.

[0099] The early warning module is further included for early warning when the compliance or the predicted eye drop box use duration exceeds a threshold value. Through early warning, the eye drop box reminding strategy can be adjusted in time, such as increasing the reminding frequency; when the compliance exceeds the threshold range, timely reminding and intervention are performed to improve the use efficiency of the eye drop box.

[0100] The above is only an embodiment of the present application, and the specific technical solutions and / or common knowledge of the scheme are not described in detail. It should be noted that for those skilled in the art, without departing from the technical solutions of the present application, a number of modifications and improvements can be made, which should also be considered as the protection scope of the present application, and these will not affect the effect and practicality of the present application. The protection scope of the present application should be subject to the content of its claims, and the specific implementation mode and the like in the specification can be used to explain the content of the claims.

Claims

1. An intelligent auxiliary management system for post-operative ophthalmic eye drop boxes, characterized in that, include: The medication data acquisition module is used to acquire historical user usage information, extract data features based on historical usage information, and create a feature knowledge graph based on the data features; The feature data optimization module is used to optimize and select feature data based on the current user usage information and the feature knowledge graph using the Grey Wolf optimization algorithm. The feature data optimization module includes: Initialize the submodule and randomly generate a set of candidate solutions, each solution representing a feature selection scheme; The objective definition submodule determines the fitness function and boundary conditions based on the prediction model's returns. The fitness function includes: in, , , , Indicates the weighting coefficient. Indicates classification accuracy. Indicates the penalty for the size of the feature subset. A measure of model stability Indicates time cost; in, This represents the number of samples correctly predicted as positive. This represents the number of samples that were correctly predicted as negative. This represents the number of samples that were incorrectly predicted as positive. This represents the number of samples that were incorrectly predicted as negative. Indicates the penalty coefficient. Indicates the size of the feature subset. () represents the average function. () represents the standard deviation function. Indicates the adjustment parameter; The boundary conditions include: in, and Let represent the minimum and maximum number of features in the feature set. Represents the actual number of features; The position update submodule performs gray wolf hunting based on fitness, retains the current optimal solution obtained from the hunt, and in each iteration, updates the position based on the current optimal solution. , , The wolf pack's position is updated based on the wolf's location until the termination condition is met; In the location update submodule, the location coefficient vector and distance coefficient vector are updated through chaotic mapping. At the beginning of the algorithm, chaotic variables are initialized. In each iteration, new chaotic values ​​are generated through chaotic mapping. The generated chaotic sequence is used to replace the random number generator, and the model is updated as follows: in, Represents the position coefficient vector. Represents the distance coefficient vector. This indicates the degree of control over chaos, and 'a' represents the convergence factor. Represents chaotic values; The prediction model building module is used to train a CNN model by taking the optimized feature data as input and the usage time of the eye drop box as output, and then obtaining the eye drop box usage time prediction model. The application module is used to obtain prediction results based on the eye drop box usage time prediction model, and to manage the eye drop box based on the prediction results.

2. The intelligent auxiliary management system for post-operative ophthalmic eye drop boxes according to claim 1, characterized in that, The medication data acquisition module includes: The knowledge representation submodule is used to define entities and entity relationships; The graph creation submodule is used to generate corresponding nodes for each entity type, assign attributes, establish connections based on the relationships between entities, and add detailed attribute information to nodes and edges; The knowledge reasoning submodule is used to automatically infer new knowledge based on logical rules. The effectiveness evaluation submodule is used to evaluate the coherence, coverage, and effectiveness of the map.

3. The intelligent auxiliary management system for postoperative ophthalmic eye drops according to claim 1, characterized in that, The prediction model construction module includes: The architecture is built into sub-modules. In the input layer, the size of the input layer is defined according to the data dimension. In the convolutional layer, the feature patterns in the data are extracted. In the pooling layer, the spatial size of the feature mapping is reduced by average pooling. In the fully connected layer, the features learned in the previous layers are mapped to the specific output. In the output layer, the specific duration is output. The training and validation submodule uses MSE as the loss function and Adam as the optimizer to train the model until it converges.

4. The intelligent auxiliary management system for postoperative ophthalmic eye drops according to claim 1, characterized in that: It also includes a follow-up recommendation module, which compares the target usage time of the eye drop box with the predicted usage time, and recommends a follow-up visit when the predicted usage time exceeds the threshold range.

5. The intelligent auxiliary management system for postoperative ophthalmic eye drops according to claim 1, characterized in that: It also includes a compliance analysis module for analyzing users' medication compliance. The analysis model includes: in, , , For adjustment coefficients, Indicates long-term compliance rate, Indicates the compliance rate based on frequency. This indicates the compliance rate within a time window.

6. The intelligent auxiliary management system for postoperative ophthalmic eye drops according to claim 5, characterized in that: It also includes an early warning module, which is used to issue an alert when compliance or predicted usage time of the eye drop box exceeds a threshold.

Images