Perioperative period management method based on doctor terminal and related equipment

Abstract

The invention provides a perioperative period management method based on a doctor side and related equipment, and is applied to the technical field of data processing. The method comprises the following steps: receiving target user evaluation information, a preset eye early warning model and a training sample set sent by a doctor terminal; processing the training sample set based on a preset processing rule to generate a target training set and a target verification set; processing a preset eye early warning model based on the target training set and the target verification set to generate a target eye early warning model; processing the target user evaluation information based on the target eye early warning model to generate eye high-risk label information of the target user; processing the eye high-risk label information of the target user to generate preoperative evaluation information; the target user is processed based on the preoperative evaluation information, and intraoperative early warning information is generated; and processing the preoperative evaluation information and the intraoperative early warning information to generate postoperative early warning information.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a perioperative management method and related equipment based on the doctor's end. Background Technology

[0002] There are six main problems with the preoperative business needs: First, patient information collection is inefficient and costly, and there is a significant lack of data. Second, high-risk factors identified by nurses during preoperative assessments are difficult to communicate to surgeons in a timely and comprehensive manner. Third, manual scoring of questionnaires is inefficient and difficult to use effectively in actual preoperative assessments. Fourth, there are numerous preoperative laboratory tests, paper reports are easily lost, and electronic reports are scattered across different systems, making them inconvenient to access. Fifth, after patients complete their appointments and leave the hospital, the only traditional way to contact them is by phone, which is time-consuming and labor-intensive for batch operations such as reminding them to order medication and notifying them of their arrival time on the day of surgery. Sixth, there are numerous items related to patient surgical information, resulting in a large workload and low efficiency in manual data entry. There are two main problems with the business needs in the intraoperative scenario. First, it is difficult to record various situations and key surgical information in a timely manner, which cannot serve as a source of information for subsequent rehabilitation and screening of high-risk patients. Second, it is inconvenient for doctors to operate information collection systems during surgery, and it is difficult to collect information using traditional information recording (such as paper record sheets). In addition, it is also difficult for doctors to accept information transmitted by traditional paper information media. There are two main problems with the business needs in the postoperative scenario. First, after patients finish surgery and leave the hospital, it is time-consuming and laborious to make mass follow-up calls manually. At the same time, if patients do not check the follow-up questionnaire pushed by the APP in time, it is difficult to remind them one by one. Second, due to the heavy workload of doctors, it is difficult to track the real-time situation of high-risk patients and make timely responses and treatments.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this application is to provide a perioperative management method and related equipment based on the physician's perspective, which at least partially overcomes the problems existing in the prior art. It generates assessment reports by analyzing the multimodal information of the target user, such as personal information, purpose of medical visit, and work attributes. This information is used to train an eye early warning model, which can identify eye disease characteristics and generate high-risk label information. Preoperatively, the system uses this information to generate a detailed assessment, including eye image analysis and feature maps, to identify abnormalities. Intraoperatively, the system processes the physician's voice information to generate intraoperative warnings to assist the physician in performing the surgery. Postoperatively, the system combines the preoperative assessment and intraoperative warnings to generate updated postoperative warning information. In addition, the system also processes physician consultation and answer information to improve patient satisfaction. The entire process aims to improve the accuracy and efficiency of eye disease diagnosis, reduce surgical risks, and provide patients with personalized postoperative management recommendations.

[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0006] According to one aspect of this application, a perioperative management method based on a doctor's end is provided, comprising: receiving target user assessment information, a preset eye warning model, and a training sample set sent by a doctor's end, wherein the target user assessment information is generated based on multimodal target user information, target user's purpose of medical treatment information, and target user's work attribute information sent by a client, and the doctor's end is matched with the processing priority of the target user; processing the training sample set according to preset processing rules to generate a target training set and a target validation set; processing the preset eye warning model based on the target training set and the target validation set to generate a target eye warning model; processing the target user assessment information based on the target eye warning model to generate high-risk eye label information for the target user; processing the high-risk eye label information for the target user to generate preoperative assessment information; processing the target user based on the preoperative assessment information to generate intraoperative warning information; and processing the preoperative assessment information and the intraoperative warning information to generate postoperative warning information.

[0007] Another aspect of this application discloses a perioperative management device based on a doctor's end, characterized by comprising: an acquisition module for receiving target user assessment information, a preset eye warning model, and a training sample set sent by the doctor's end, wherein the target user assessment information is generated based on multimodal target user information, target user's purpose of medical treatment information, and target user's work attribute information sent by the client, and the processing priority of the doctor's end is matched with the target user; a processing module for processing the training sample set according to preset processing rules to generate a target training set and a target validation set; processing the preset eye warning model based on the target training set and the target validation set to generate a target eye warning model; processing the target user assessment information based on the target eye warning model to generate high-risk eye label information for the target user; processing the high-risk eye label information for the target user to generate preoperative assessment information; processing the target user based on the preoperative assessment information to generate intraoperative warning information; and processing the preoperative assessment information and intraoperative warning information to generate postoperative warning information.

[0008] According to another aspect of this application, an electronic device is characterized by comprising: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described perioperative management method based on the doctor's end via executing the executable instructions.

[0009] According to another aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a second processor, implements the above-described perioperative management method based on the doctor's end.

[0010] According to another aspect of this application, a computer program product is provided, comprising a computer program, characterized in that, when the computer program is executed by a third processor, it implements the above-described perioperative management method based on the doctor's end.

[0011] This application provides a perioperative management method and related equipment based on the doctor's perspective. The server generates an assessment report by analyzing the target user's multimodal information, such as personal information, purpose of medical visit, and work attributes. This information is used to train an eye early warning model, which can identify eye disease characteristics and generate high-risk label information. Preoperatively, the system uses this information to generate a detailed assessment, including eye image analysis and feature maps, to identify abnormalities. During surgery, the system processes the doctor's voice information to generate intraoperative warnings to assist the doctor in performing the surgery. Postoperatively, the system combines the preoperative assessment and intraoperative warnings to generate updated postoperative warning information. In addition, the system also processes the doctor's consultation and answer information to improve patient satisfaction. The entire process aims to improve the accuracy and efficiency of eye disease diagnosis, reduce surgical risks, and provide patients with personalized postoperative management recommendations.

[0012] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0013] Figure 1 A flowchart illustrating a perioperative management method based on a doctor's end according to an embodiment of this application is shown; Figure 2 A schematic diagram of a perioperative management device based on a doctor's end, provided in one embodiment of this application, is shown. Detailed Implementation

[0014] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0015] The following is combined with Figure 1 This application describes a doctor-based perioperative management method according to exemplary embodiments thereof. It should be noted that the following application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way. Rather, the embodiments of this application can be applied to any applicable scenario.

[0016] In one embodiment, this application also proposes a perioperative management method and related equipment based on the doctor's end. Figure 1 A schematic diagram illustrating a perioperative management method based on a physician's end, according to an embodiment of this application, is shown. Figure 1 As shown, this method is applied to a server and includes: S101 receives target user assessment information, preset eye early warning models, and training sample sets sent by the doctor.

[0017] In one implementation, the target user assessment information is generated based on multimodal target user information sent by the client, the target user's purpose of medical treatment information, and the target user's work attribute information. The doctor's processing priority is matched with the target user's. Multimodal target user information refers to various types of data uploaded by the user to the client via a mobile app. For example, user Xiao Li fills in the text information in the app: "Recently, I've noticed a significant decline in my vision; things are blurry, especially difficult when driving at night. There's no family history of eye diseases." Simultaneously, he records a voice description: "I work in front of a computer for at least 8 hours every day, and my eyes are often dry and tired. This recent decline in vision is very worrying." He also uploads photos of his eye appearance and a recent vision test report. The client sends this multimodal information, including text, voice, and images, to the server. Xiao Li selects "Diagnosis and Treatment of Eye Diseases" as his purpose of medical treatment in the app, indicating that he hopes to receive a comprehensive eye examination and an effective treatment plan due to his vision problems; this information is also transmitted to the server. Xiao Li fills in his work attributes in the app: his occupation is "programmer," his work environment is "indoor office, long hours using a computer," and his workload is "high, often working overtime." This information is also sent to the server.

[0018] The target eye warning model was used to process Xiao Li's multimodal target user information, medical purpose information, and work attribute information. Analysis of the multimodal information revealed potential eye physiological conditions such as visual fatigue and myopia progression, as well as real-time environmental information related to prolonged indoor office work. First eye impact factors related to vision problems were generated based on the medical purpose, such as a lens accommodation ability impact factor of 0.75 (indicating a significant impact on eye condition); second eye impact factors were generated based on work attributes, such as a vision impact factor of 0.8 (due to prolonged close-range eye use). These impact factors were combined to obtain eye characteristic impact factors (e.g., a final vision-related impact factor of 0.78) and corresponding weight information (vision weight 0.45, etc.). Based on this information, an eye warning message for Xiao Li was generated, such as "User Xiao Li's eye health condition is at risk, with a high probability of myopia progression. It is recommended to reduce continuous eye use time and undergo a detailed eye examination as soon as possible, including refraction and intraocular pressure measurement." This eye warning message is part of the target user assessment information.

[0019] Suppose that Dr. Zhang, an ophthalmologist at the hospital, specializes in handling myopia-related issues, and his schedule matches Xiao Li's situation (e.g., Dr. Zhang has suitable outpatient hours and surgery dates scheduled this week). Based on the severity and urgency of the problem indicated by Xiao Li's eye warning information (e.g., significant vision loss, though not urgent, requires timely treatment and can be marked as "routine-medium risk"), the system determines that Dr. Zhang is the doctor whose treatment priority matches Xiao Li's, and assigns him to Xiao Li's treatment process. The doctor's end where Dr. Zhang is located is the doctor's end whose treatment priority matches Xiao Li's.

[0020] The doctor's app collects multimodal target user information about Xiao Li, including information on the purpose of his medical visit, his work attributes, and generated eye warning information, integrating this into a target user assessment. For example, the doctor's app displays Xiao Li's information as follows: multimodal information (description of vision loss, eye photos, vision test report, etc.), purpose of medical visit (diagnosis and treatment of eye diseases), work attributes (programmer, high eye strain), and eye warning information (risk of myopia progression, recommended examinations). Dr. Zhang can view this detailed target user assessment information on the doctor's app, allowing for a more precise treatment plan for Xiao Li, such as arranging further professional eye examinations and providing eye care advice based on his work habits. Simultaneously, the doctor's app can feed back some content from this assessment information (such as preliminary diagnostic suggestions and examination arrangements) to the client, enabling Xiao Li to stay informed about subsequent treatment plans and make appropriate preparations.

[0021] A pre-defined eye warning model is a model built based on machine learning or deep learning algorithms, designed to predict the risk of eye diseases or assess and warn of eye health conditions based on various input data. The model can employ neural network models, such as convolutional neural networks (CNNs) to process eye image data (e.g., fundus images), recurrent neural networks (RNNs) or long short-term memory networks (LSTMs) to process time-series related eye data (e.g., changes in intraocular pressure over time), and fully connected layers to integrate data features from different modalities. For example, the model's input layer receives multimodal data, including image data, text data (e.g., medical history descriptions), and numerical data (e.g., intraocular pressure, visual acuity values). Convolutional layers extract features from the images, RNN layers analyze the time-series data, and then fully connected layers fuse the extracted features. Finally, the output layer outputs the warning result, such as the probability of eye disease occurrence and risk level.

[0022] Alternatively, a model based on decision tree algorithms can be built. By learning from a large amount of sample data, a decision tree is constructed, and branch decisions are made based on different feature values ​​(such as age, gender, family history of glaucoma, eye examination indicators, etc.) to ultimately determine the eye health status or disease risk. For example, when a user is over 60 years old, has a family history of glaucoma, and has high intraocular pressure, the decision tree model may classify them as a high-risk group for eye diseases and provide corresponding warnings.

[0023] Examples of model functions are as follows: Disease Risk Prediction Function: Based on the user's multimodal information, the model can predict the risk of developing various eye diseases, such as cataracts, glaucoma, and macular degeneration. For example, for a user who works outdoors for extended periods, is over 50 years old, and has high myopia, the model can comprehensively consider these factors to predict a higher risk of developing cataracts and macular degeneration, and issue early warnings to the user and doctor, recommending a more detailed eye examination. Disease Progression Monitoring Function: For users already suffering from eye diseases, the model can monitor the trend of disease progression based on continuously input eye data (such as regularly checked visual acuity, intraocular pressure, fundus images, etc.) and predict whether the condition will worsen. For example, a glaucoma user regularly uploads intraocular pressure measurements and fundus images. The model can analyze these data changes and, if it finds poor intraocular pressure control and a worsening trend in optic nerve damage, promptly remind the doctor to adjust the treatment plan.

[0024] The training sample set is a dataset used to train the eye early warning model. It contains a large amount of eye-related information from different individuals, along with corresponding annotations (such as whether they have a certain eye disease, the severity of the disease, etc.). Specifically: Hospital electronic medical record data: Obtaining users' eye diagnosis records, treatment processes, examination reports, etc., from partner hospitals. For example, collecting medical records of ophthalmology outpatient users, including basic information (age, gender, occupation, etc.), chief symptoms (such as decreased vision, eye pain, distorted vision, etc.), eye examination results (visual acuity, intraocular pressure, refraction results, fundus examination description, etc.), and final diagnostic conclusions (such as confirmed diabetic retinopathy, retinal detachment, etc.). This data can cover users of different age groups and different types of eye diseases, providing a rich and diverse sample for model training. Clinical research data: Participating in or collecting data from ophthalmology-related clinical research projects. For example, in a study on the therapeutic effect of a new type of ophthalmic drug, detailed ophthalmic data of participating users before and after medication were recorded, including changes in ocular physiological indicators, adverse drug reactions, and treatment effect evaluation information. These rigorously controlled and monitored data provide high-quality samples for model training, helping to improve the model's predictive ability regarding drug treatment responses, etc. Public health database data: Utilizing publicly available eye health-related databases, such as eye disease statistics released by ophthalmology research institutions or government health departments, and large-scale population eye health screening data, can supplement the diversity of training samples. For example, including eye health status information from different regions and ethnic groups, this makes the trained model more widely applicable.

[0025] Each sample contains multiple features, such as the user's personal characteristics (age, gender, race, work environment, etc.), ocular physiological characteristics (visual acuity, intraocular pressure, axial length, corneal curvature, degree of lens opacity, etc.), medical history characteristics (whether there are systemic diseases such as diabetes, hypertension, history of eye surgery, family history of eye diseases, etc.), and lifestyle characteristics (smoking and drinking habits, eye habits such as daily use of electronic devices, duration of outdoor activities, etc.). For example, a sample might be described as follows: [Age: 45 years old, Gender: Female, Ethnicity: Asian, Work environment: long-term computer use in an indoor office, Visual acuity: left eye 0.6, right eye 0.8, Intraocular pressure: left eye 18 mmHg, right eye 20 mmHg, Axial length: left eye 23.5 mm, right eye 23.8 mm, Corneal curvature: left eye 43D, right eye 44D, Lens opacity: mild, 5-year history of diabetes, no family history of eye diseases, more than 8 hours of daily use of electronic devices, less than 1 hour of outdoor activity], with the corresponding labeling result being "high risk of early diabetic retinopathy".

[0026] Examples of surgical scheduling and user management related data are as follows: Surgical appointment rules and procedures: The hospital's surgical appointment process (such as how far in advance users need to make an appointment, what documents are required at the time of appointment, etc.), surgical priority rules (such as priority for emergency surgery, priority for high-risk users, etc.), and relevant regulations for surgical cancellation and rescheduling. For example, users of routine eye surgery need to make an appointment 3-7 days in advance and need to provide a recent eye examination report at the time of appointment; users with emergency eye trauma can be scheduled for surgery at any time; for users with high-risk eye diseases (such as during an acute angle-closure glaucoma attack), once diagnosed, surgery should be scheduled within 24 hours. These rules and procedures are closely related to the preoperative appointment information generated by the eye early warning model to ensure that users can receive surgical treatment in a timely manner according to a reasonable process. User information management system data: Basic user information (name, age, contact information, etc.), medical record information (diagnostic records, treatment history, examination reports, etc.), and user medical process records (number of outpatient visits, length of hospitalization, surgical records, etc.) stored in the hospital's internal user information management system. By interacting with the system, the eye early warning system can obtain the user's historical medical data, gain a more comprehensive understanding of the user's eye health status, and at the same time, it can feed back eye early warning information and postoperative follow-up results to the hospital information management system, realizing the integration and sharing of user medical information, which facilitates doctors to make comprehensive diagnosis and treatment decisions.

[0027] S102, the training sample set is processed based on preset processing rules to generate a target training set and a target validation set.

[0028] In one implementation, the training sample set is processed to generate a target feature library, which includes both discontinuous and continuous variable data. A large number of data samples related to eye health are collected from various data sources (such as hospital electronic medical record systems, ophthalmic examination equipment databases, and clinical research data). Each sample contains multiple features, such as the user's age, gender, family history of eye diseases (discontinuous variable data), intraocular pressure measurement, visual acuity, and axial length (continuous variable data). The collected data is cleaned to remove erroneous data and records with excessive missing values ​​to ensure data quality. For example, if a user's age in a sample is incorrectly recorded as negative or a crucial eye examination value is severely missing, the sample is corrected or deleted.

[0029] For non-continuous variable data such as text-based medical history descriptions, feature extraction is performed using natural language processing techniques. For example, features such as whether the user has diabetes (yes / no) or has a history of eye trauma (yes / no) can be extracted from the user's medical history description. For numerical eye examination data, it is directly extracted as continuous variable data. For example, multiple intraocular pressure measurements and visual acuity values ​​at different time points can be extracted. These extracted features are then classified into non-continuous and continuous variable data, forming a target feature library.

[0030] The training sample set is processed based on non-continuous variable data to generate an initial training set and an initial validation set. For non-continuous variable data, such as gender (male / female) and family medical history (yes / no), one-hot encoding is used to convert them into numerical form. For example, gender can be encoded as [1,0] for male and [0,1] for female. The encoded non-continuous variable data is combined with other relevant information of the samples (such as sample identifiers) and randomly divided into the initial training set and the initial validation set according to a certain ratio (e.g., 70%:30%). For example, if there are 1000 samples, 700 samples are used as the initial training set and 300 samples are used as the initial validation set.

[0031] The initial training and validation sets are processed based on continuous variable data to generate target training and target validation sets. The target training set includes several ocular physiological reference intervals, with different intervals corresponding to different user information for the same disease type. For continuous variable data (such as intraocular pressure and visual acuity), ocular physiological reference intervals are defined based on medical expertise and data analysis results. For example, the normal range for intraocular pressure might be 10-21 mmHg. Samples with intraocular pressure within this range are grouped together, while samples with pressure above 21 mmHg are further subdivided according to different degrees (e.g., 22-30 mmHg for mild elevation, 31-40 mmHg for moderate elevation, etc.).

[0032] The continuous variable data in the initial training and validation sets are mapped according to predefined ocular physiological reference intervals, and the samples are reclassified into the corresponding interval categories, thereby generating the target training and target validation sets. For example, a sample with an intraocular pressure of 25 mmHg in the initial training set, after interval mapping, is classified into the sample category corresponding to the mild intraocular pressure elevation interval. Finally, the target training and target validation sets containing samples from different ocular physiological reference intervals are obtained, so that the model can learn the feature patterns of eye diseases under different physiological states.

[0033] First, generate the target feature library: Non-continuous variable data: age (55, 40, 60, 35, 50), gender (male, female, female, male, female), family medical history (yes, no, yes, no, yes). Continuous variable data: intraocular pressure (18, 22, 25, 16, 30), visual acuity (left: 0.6, 1.0, 0.4, 1.2, 0.3; right: 0.8, 1.0, 0.5, 1.2, 0.4). Assume that gender is encoded using one-hot encoding, male is [1,0], female is [0,1], and family medical history is [1] if yes, [0] if no. After random partitioning (assuming a 70%:30% partition), the initial training set may contain samples 1, 3, and 4, and the initial validation set may contain samples 2 and 5.

[0034] For intraocular pressure, the ranges are divided into normal (10-21 mmHg) and slightly elevated (22-30 mmHg). Sample 1 falls within the normal range, samples 3 and 5 within the slightly elevated range, and sample 4 within the normal range. For visual acuity (taking the left eye as an example), ranges can be defined based on visual acuity, such as 0.3-0.5 for low vision, 0.6-1.0 for normal vision, and above 1.0 for good vision. Samples 1 and 2 fall within the normal vision range, samples 3 and 5 within the low vision range, and sample 4 within the good vision range. Based on these range mappings, a target training set and a target validation set are generated. The target training set contains sample information corresponding to different ocular physiological reference ranges, facilitating the model's learning of the relationship between ocular features and disease types under different conditions.

[0035] S103, The preset eye warning model is processed based on the target training set and the target validation set to generate the target eye warning model.

[0036] In one implementation, the target training set is processed to generate eye feature information and correlation co-occurrence frequency, and the eye feature information and correlation co-occurrence frequency are further processed to generate correlation matrix information.

[0037] When generating the correlation matrix information, the association analysis between systemic disease features and ocular features is further incorporated to construct a correlation sub-matrix between systemic and ocular features. Based on the correlation matrix information and the correlation sub-matrix between systemic and ocular features, a prior knowledge graph for characterizing ocular diseases is generated. This prior knowledge graph is then used to process a pre-defined ocular early warning model, generating a trained ocular early warning model. The trained ocular early warning model is then processed using a target validation set to generate validation results. If the data samples in the validation results containing labeling information are ocular feature information characterizing ocular diseases, then the trained ocular early warning model is used as the target ocular early warning model. Assume a small training dataset containing 5 samples (n=5), each sample having 3 ocular features (intraocular pressure, visual acuity, and axial length). The method also includes a formula for calculating the covariance matrix, which is: ; ; ; Among them, the elements in the covariance matrix C Let represent the covariance between the i-th eye feature and the j-th eye feature. This represents the mean of the i-th eye feature. This represents the i-th eye feature value of the k-th sample. This represents the j-th eye feature value of the k-th sample.

[0038] Specifically, calculate the mean ( and ): For intraocular pressure (IOP) (assuming it's the first ocular feature, i=1), the IOP values ​​for the five samples are 15, 18, 20, 16, and 17. What is the mean IOP? for: .

[0039] For visual acuity (assuming it's the second ocular feature, j=2), the visual acuity values ​​of the five samples are 0.8, 1.0, 0.6, 0.9, and 1.2, respectively. Therefore, the mean visual acuity is... : .

[0040] For axial length (assuming it's the 3rd eye feature, j=3), the axial length values ​​for the 5 samples are 23, 24, 25, 23, and 24, respectively. Therefore, the mean axial length is... : ; Calculate the covariance matrix ( ): Calculate the covariance between intraocular pressure (i=1) and visual acuity (j=2). : First calculate ( () The value of ) For sample 1: (15-17.2)(0.8-0.9) = 0.22; For sample 2: (18-17.2)(1.0-0.9) = 0.08; For sample 3: (20-17.2)(0.6-0.9) = -0.84; For sample 4: (16-17.2)(0.9-0.9) = 0; For sample 5: (17-17.2)(1.2-0.9) = -0.06; Then calculate the covariance. : ; The same method can be used to calculate the covariance between intraocular pressure (i=1) and axial length (j=3). The covariance matrix C is obtained by considering factors such as visual acuity (i=2) and axial length (j=3).

[0041] The method also includes a formula for calculating the standard deviation vector, which is as follows: ; ; in, The standard deviation of the i-th eye feature is represented by... The standard deviation of the j-th eye feature; Specifically, calculate the standard deviation vector ( and ), calculate the standard deviation of intraocular pressure (i=1) : First calculate Value: For sample 1: For sample 2: ; For sample 3: For sample 4: ; For sample 5: ; Then calculate the standard deviation. : ; The same method can be used to calculate the standard deviation of visual acuity (i=2). and the standard deviation of axial length (j=3) Then, we obtain the standard deviation vector.

[0042] The method also includes a formula for calculating the correlation matrix, which is as follows: ; Among them, the elements in the correlation matrix R The correlation coefficient represents the relationship between the i-th eye feature and the j-th eye feature.

[0043] Specifically, calculate the correlation matrix ( ): Calculate the correlation coefficient between intraocular pressure (i=1) and visual acuity (j=2). : The same method can be used to calculate the correlation coefficient between intraocular pressure (i=1) and axial length (j=3). The correlation coefficient between visual acuity (i=2) and axial length (j=3) was also observed. Finally, the correlation matrix R is obtained.

[0044] The ocular feature information extracted from the training set may include specific values ​​for each sample, such as intraocular pressure (IOP), visual acuity, and axial length, as well as the distribution of these features across different samples. For example, this involves counting the number of samples with IOP in different ranges (e.g., 10-15 mmHg, 15-20 mmHg), visual acuity in different visual acuity levels (e.g., 0.1-0.5, 0.5-1.0), and axial length in different length intervals (e.g., 20-22 mm, 22-24 mm). This statistical information helps to understand the overall situation of ocular features. The frequency of co-occurrence of different ocular feature values ​​is also calculated. For example, the frequency of samples with IOP between 15-20 mmHg and visual acuity between 0.5-1.0. By analyzing the co-occurrence frequency of correlations, a preliminary judgment can be made as to whether there is a correlation trend between different ocular features. If it is found that samples with higher IOP and lower visual acuity have a higher co-occurrence frequency, this may suggest a negative correlation between IOP and visual acuity, requiring further precise measurement using correlation coefficients.

[0045] A correlation matrix is ​​constructed based on the correlation coefficients calculated above. Examples include matrices composed of correlation coefficients between intraocular pressure and visual acuity, intraocular pressure and axial length, and visual acuity and axial length. The correlation matrix visually demonstrates the relationships between ocular features. If the absolute value of the correlation coefficient between two features in the correlation matrix is ​​close to 1, it indicates a high correlation (negative correlation); if the correlation coefficient is close to 0, it indicates a weak correlation. When generating the correlation matrix, further association analysis between systemic disease features (such as blood glucose and blood pressure) and ocular features is incorporated. For example, co-occurrence data of glycated hemoglobin levels and fundus hemorrhage incidence in diabetic patients in the training sample set are used to construct a 'systemic-ocular' feature correlation sub-matrix as a supplementary dimension to the prior knowledge graph. A prior knowledge graph is generated based on the correlation matrix information and the systemic-ocular feature correlation sub-matrix. A graph structure is constructed with ocular features as nodes and the correlations between features as edge weights. For example, if intraocular pressure (IOP) and visual acuity are highly negatively correlated, then there would be an edge between the IOP node and the visual acuity node in the graph, with the edge weight determined by the correlation coefficient (e.g., a weight of -0.8). Prior knowledge graphs can more intuitively present the complex network of relationships between eye features, helping to understand potential factors associated with eye diseases. For instance, the graph can reveal nodes related to IOP (such as axial length, corneal thickness, etc.), and these nodes, connected by edges, form a subgraph. This subgraph might represent a set of factors related to IOP-related diseases such as glaucoma. When building eye early warning models, prior knowledge graphs can provide important reference information, allowing the model to learn the relationship patterns between these features, thereby better predicting the risk of eye diseases.

[0046] The generated prior knowledge graph is used to process the pre-defined eye warning model. For example, if the pre-defined eye warning model is a neural network model, the eye feature relationships in the prior knowledge graph can be used to guide the structural design of the neural network, such as determining which feature nodes should be tightly connected (strongly correlated features) and which can be relatively loosely connected (weakly correlated features). Simultaneously, during training, the model parameters are adjusted based on the feature relationships in the prior knowledge graph, enabling the model to better capture the relationship between eye features and diseases. The trained eye warning model is validated using a target validation set. Samples from the target validation set are input into the trained model to obtain prediction results. For example, for a validation set sample containing features such as intraocular pressure, visual acuity, and axial length, the model predicts whether the sample has a certain eye disease (such as glaucoma) and the probability of having it. Then, the prediction results are compared with the actual annotations (whether or not the sample has glaucoma) in the validation set to generate validation results. If the model can accurately identify most of the eye features that characterize eye diseases (such as glaucoma) in the validation results (i.e., samples predicted to have glaucoma actually do have glaucoma, and samples predicted not to have glaucoma actually do not have glaucoma), then the model performance is good and the trained model can be used as the target eye warning model; otherwise, the model parameters need to be further adjusted or the training method improved, and the model needs to be retrained and validated.

[0047] S104, Based on the target eye early warning model, the target user assessment information is processed to generate high-risk eye tag information for the target user.

[0048] In one implementation, multimodal target user information is processed to generate real-time ocular physiological status information and real-time environmental information for the target user. For example, user Xiao Wang uploads multimodal information to the system via a mobile app. The text message mentions, "Recently, my vision has become increasingly blurry, feeling like there's a layer of fog. My eyes are prone to dryness and occasional stinging. I haven't had eye surgery before, but I have a history of hypertension." The voice message describes, "I work outdoors for long periods every day in very strong sunlight. Even though I wear sunglasses, my eyes still feel uncomfortable. My workload is very heavy, and I often have to rush to meet deadlines." The system also uploads a photo of the appearance of the eyes (showing some redness and swelling) and a recent intraocular pressure (IOP) report image (IOP value slightly higher than normal). The system uses natural language processing technology to extract keywords from the text and voice information, such as "blurred vision," "dry eyes," "sting sensation," and "high intraocular pressure." Combined with data from the IOP report, the system analyzes and concludes that Xiao Wang may have abnormally high intraocular pressure, dry and damaged ocular surface, and decreased vision, among other ocular physiological problems. Image analysis of photos showing the eyes revealed redness and swelling, confirming inflammation. Based on this, it was determined that Xiao Wang's current eye condition was poor, potentially indicating an eye disease such as early symptoms of glaucoma or vision problems caused by ocular surface inflammation. The voice messages mentioning "long hours of outdoor work" and "strong sunlight," along with the user's designated work location in the app, suggested that Xiao Wang's real-time environment involved strong outdoor sunlight, high work intensity, and excessive eye strain. Such environmental factors can negatively impact eye health, such as exacerbating eye fatigue and increasing the risk of UV damage.

[0049] The system processes the target user's medical purpose information to generate a primary ocular impact factor. Xiao Wang selected "eye discomfort examination and diagnosis" as his medical purpose on the app, stating in his description that recent vision loss and severe eye discomfort symptoms were significantly impacting his work and life, and he hoped to determine the cause and seek treatment through examination. The system generates the primary ocular impact factor based on this medical purpose, identifying ocular characteristics related to vision and eye comfort that are significantly affected. For example, the impact factor for lens transparency (affecting vision) is set at 0.7 (indicating a high degree of impact on eye condition), because vision loss may be related to lens disease; the impact factor for tear film stability (affecting dry eyes and stinging sensations) is set at 0.8, because eye discomfort symptoms are clearly related to abnormal tear film function. These impact factors will serve as important references in subsequent comprehensive assessments of ocular risks.

[0050] The system processes the target user's work attribute information to generate a second eye-related influencing factor. For example, Xiao Wang's work attribute information shows his occupation as "construction worker," his work environment as "outdoor construction site, strong sunlight, and a lot of dust," and his work intensity as "high intensity, 10-12 hours per day." Based on these work attributes, the system generates corresponding influencing factors. Prolonged exposure to strong outdoor sunlight increases the risk of eye diseases such as cataracts and macular degeneration, so the influencing factor for retinal health is set at 0.8; dusty environments easily lead to problems such as foreign body sensation and inflammation in the eyes, so the influencing factor for ocular surface cleanliness and defense function is set at 0.7; excessive eye strain from high-intensity work can cause eye fatigue and decreased vision, so the influencing factor for visual accommodation ability is set at 0.6.

[0051] The first and second ocular influencing factors are processed to generate ocular feature influencing factors and corresponding weight information. The first and second ocular influencing factors are then combined. Taking lens transparency as an example, the influencing factor of 0.7 from the medical purpose information and the related influencing factor of 0.3 from the work attribute information (such as the potential impact of strong outdoor light on the lens) are weighted and averaged to obtain a final ocular feature influencing factor of 0.6 for lens transparency (calculated as: 0.7 × 0.6 + 0.3 × 0.3 = 0.6). The influencing factors for other ocular features (such as tear film stability, retinal health, ocular surface cleanliness, and visual accommodation ability) are calculated using the same method. Weights are determined based on the importance of ocular features to eye health and the degree of change in the current user's condition. For example, the weight of lens transparency is set at 0.3 (because it is closely related to vision and there may be problems at present), tear film stability at 0.2 (ocular surface symptoms are obvious but slightly less important than lens problems), retinal health at 0.25 (retinal risk increases in bright light), ocular surface cleanliness at 0.15 (dusty environment has an impact), and visual accommodation ability at 0.1 (currently has an impact but is slightly less important than other factors).

[0052] The system processes the target user's real-time eye physiological status, real-time environmental information, eye feature influencing factors, and corresponding weight information based on the target eye early warning model. This generates an eye risk warning value for the target user. The system then processes this value using a pre-defined eye risk warning table to generate a high-risk eye label, indicating that the target user's eye risk warning value exceeds the pre-defined value. In this example, the system inputs Xiao Wang's real-time eye physiological status (high intraocular pressure, ocular surface inflammation, decreased vision, etc.), real-time environmental information (strong outdoor light, high dust levels, high-intensity eye strain), eye feature influencing factors (e.g., lens transparency 0.6), and corresponding weight information (e.g., lens transparency weight 0.3) into the target eye early warning model. The model performs a comprehensive calculation based on a pre-defined algorithm and training data to determine Xiao Wang's eye risk warning value. For example, the model calculates a value of 0.75 (the specific calculation process is based on a complex internal algorithm; this is just an example). Assuming the preset eye risk warning value is 0.6, and Xiao Wang's eye risk warning value of 0.75 is greater than the preset value, the system generates a high-risk eye label for Xiao Wang based on the preset eye risk warning table, such as "High Risk - Suspected Eye Disease (Early Glaucoma? Vision Problems Caused by Ocular Surface Inflammation?) Requires Urgent Examination and Intervention." This label will prompt doctors and the medical system to pay close attention to Xiao Wang's condition, prioritizing detailed examinations (such as further intraocular pressure monitoring, fundus examination, ocular surface function examination, etc.) and corresponding treatment measures to prevent the eye condition from worsening. Through the above example, it is clear how the system integrates multiple aspects of information, processes them step by step, and generates an assessment and warning of the user's eye health status, thereby achieving effective management of perioperative eye risks.

[0053] S105 processes the high-risk ocular label information of the target user to generate preoperative assessment information.

[0054] In one implementation, the system processes the high-risk eye label information of a target user to generate eye image information that matches the target user's high-risk eye label information. Assume that the target user, Xiao Zhao, has a high-risk eye label information of "High Risk - Suspected Glaucoma - High Risk of Optic Nerve Damage". This label indicates that Xiao Zhao may have glaucoma and that there is a risk of optic nerve damage, requiring further eye image analysis to determine the severity of the condition and formulate a surgical plan. Generating matching eye image information: Based on Xiao Zhao's high-risk eye label information, the system retrieves matching eye image information from the hospital's image database. For example, it retrieves Xiao Zhao's recent fundus color photographs and optical coherence tomography (OCT) images, which can provide detailed information about Xiao Zhao's eye structures (such as optic nerve head morphology, retinal nerve fiber layer thickness, etc.), helping to further assess eye risk.

[0055] The system slices eye images to generate target eye slice images. Taking Xiao Zhao's fundus photograph as an example, the system slices it. The fundus photograph is a two-dimensional image, and the system divides it into multiple small blocks according to preset rules. Each small block is a target eye slice image. For example, the fundus photograph is divided into several rectangular slices of the same size. Each slice contains information about fundus tissue in a certain area, such as the optic nerve head region and the peripheral retina. These slices can display the structural features of different parts of the eye in more detail, facilitating subsequent analysis.

[0056] The system processes the target eye slice image information to generate feature maps at different levels. Feature maps at the same level include feature information from different channel dimensions. For each target eye slice image, the system uses a convolutional neural network (CNN). The convolutional layers of the CNN perform convolution operations on the slice image to generate feature maps at different levels. At the initial level, the feature maps may retain more of the original image details. As the level deepens, the feature maps become increasingly abstract, capturing higher-level image features such as the edges and textures of different tissue structures. For example, the first-level feature map may show the general direction and distribution of blood vessels in the fundus, while deeper-level feature maps can highlight morphological changes in the optic disc region and other glaucoma-related features. Feature maps at the same level contain feature information from different channel dimensions. Taking a three-layer convolutional neural network as an example, the first layer feature map may have three channels, corresponding to the red, green, and blue channels of the image, respectively. Each channel extracts different features through the convolution operation of the convolution kernel. For example, the red channel may emphasize the contrast of retinal blood vessels, the green channel may be more sensitive to changes in the retinal pigment epithelium, and the blue channel helps to highlight the boundary of the optic nerve head, etc. As the number of layers increases, the number of channels may increase depending on the network structure design. The feature information of each channel dimension provides different perspectives for the subsequent generation of eye abnormality labels.

[0057] The system processes feature maps of the same channel dimension separately to generate eye abnormality markers. For each level of feature map of the same channel dimension, the system further processes them. For example, it calculates the statistical information (such as mean, variance, etc.) of the area surrounding each pixel in the feature map to determine if there is an abnormality in that area. If the statistical value of a certain area differs significantly from the statistical value of normal eye image features, the system marks that area as a potential abnormal area and generates an eye abnormality marker. Taking the optic disc region as an example, if the pixel value distribution of the optic disc region in a feature map of a certain channel dimension is significantly abnormal compared to normal samples (such as reduced average brightness, increased texture complexity, etc.), an eye abnormality marker is generated in that area. The system comprehensively considers eye abnormality markers from multiple channel dimensions. Different channels may reflect eye abnormalities from different perspectives. By integrating these markers, the location and type of eye abnormalities can be determined more comprehensively and accurately. For example, one channel may identify changes in the morphology of the optic disc, while another channel may highlight areas of thinning of the retinal nerve fiber layer. Combining these markers allows for a more precise assessment of the degree and extent of glaucoma-related lesions in Xiao Zhao's eyes.

[0058] The system processes ocular abnormality markers to generate preoperative assessment information, which characterizes the target user's ocular risk. Based on the generated ocular abnormality markers, the system conducts a detailed assessment of Xiao Zhao's ocular risk, generating preoperative assessment information. This assessment information includes a detailed description of Xiao Zhao's ocular condition, such as an increased cup-to-disc ratio of the optic nerve head and multiple thinning of the retinal nerve fiber layer, as well as predicted surgical risks and prognoses based on these abnormalities. For example, the preoperative assessment information might indicate that Xiao Zhao's glaucoma is in the intermediate stage, and surgical intervention can effectively prevent further deterioration. However, due to some degree of optic nerve damage, postoperative visual recovery may be limited, requiring close monitoring of intraocular pressure and optic nerve function changes. The preoperative assessment information not only characterizes Xiao Zhao's ocular risk but also provides crucial information for doctors to develop surgical plans. Based on the specific location and extent of ocular structural abnormalities in the assessment information, doctors select appropriate surgical methods (such as trabeculectomy, glaucoma drainage valve implantation, etc.) and determine key areas and precautions for the surgical procedure. Meanwhile, the prognostic predictions in the assessment information also help doctors communicate effectively with Xiao Zhao and his family, enabling them to have reasonable expectations about the surgical outcome and subsequent recovery. Through the above examples, we can clearly see the complete process from high-risk ocular label information to the generation of preoperative assessment information, and the important role of each step in the diagnosis and surgical planning of ocular diseases.

[0059] S106, based on preoperative assessment information, processes the target user and generates intraoperative early warning information.

[0060] In one implementation, the target surgeon's intraoperative voice information is acquired, whereby the intraoperative voice information is used to characterize intraoperative record information and preoperative assessment information related to the target user. During eye surgery, the target surgeon, Dr. Li, uses a smart surgical device equipped with voice recognition capabilities. During the surgery, Dr. Li verbally dictates various surgery-related information, which is collected by the device in real time. For example, Dr. Li might say, "Patient Xiao Wang's preoperative eye examination shows an irregular optic nerve head morphology and a slightly large cup-to-disc ratio, requiring special attention to intraocular pressure changes during surgery" (involving preoperative assessment information), and "Now we begin the trabeculectomy, first making a small incision at the limbus" (intraoperative record information). The smart surgical device transmits the collected voice information to the hospital's information processing system, which receives and prepares to process it.

[0061] The system processes the intraoperative speech of the target surgeon to generate intraoperative risk information for the target user. It uses speech recognition technology to convert the speech into text, and then natural language processing (NLP) to analyze the text. From the example speech, the system extracts key information such as "irregular optic nerve head morphology," "slightly large cup-to-disc ratio," and "trabeculectomy." Based on a pre-established knowledge base of ocular surgical risks, this key information is associated with different ocular conditions and surgical procedures. For example, abnormal optic nerve head morphology and an increased cup-to-disc ratio may indicate a more complex glaucoma condition and a higher risk of intraocular pressure fluctuations during surgery; trabeculectomy itself also carries certain risks, such as hemorrhage and bleb failure. The system comprehensively considers the extracted key information, the type of surgery, and other relevant preoperative information for the patient (such as the patient's age, medical history, and other ocular physiological indicators) to generate intraoperative risk information for the target user, Xiao Wang. For example, the risk information may indicate that Xiao Wang faces a higher risk of uncontrolled intraocular pressure during surgery, potential anterior chamber hemorrhage affecting the surgical field, and poor healing at the trabeculectomy site leading to poor surgical outcomes.

[0062] The system processes intraoperative risk information for the target user, generating intraoperative reminders and abnormal information. Intraoperative reminders are used to inform the target surgeon of subsequent procedures and surgical completion status. Based on the surgical procedure and preoperative planning, the system generates intraoperative reminders based on intraoperative risk information. For example, when a trabeculectomy reaches a specific step, the system reminds Dr. Li, "The next step requires careful separation of the trabecular tissue, taking care to avoid damaging surrounding tissues and maintaining stable intraocular pressure" (reminding the surgeon of subsequent procedures). As the surgery nears completion, the system, based on the preset surgical time and steps, reminds Dr. Li, "The surgery is about to be completed. Please check the surgical site for any remaining tissue and prepare for suturing." When a key step is completed, such as successfully establishing a filtration channel, the system indicates, "The key steps of the trabeculectomy have been completed, and the surgery is progressing smoothly" (indicating surgical completion). During the surgery, if the system detects discrepancies between the actual situation and expectations, it generates intraoperative abnormal information. For example, if the intraocular pressure monitoring device shows that Xiao Wang's intraocular pressure suddenly rises above the safe range, the system will generate an abnormal message: "The patient's intraocular pressure is abnormally high, which may be due to the surgical procedure or the patient's own eye reaction. Immediate measures should be taken to lower the intraocular pressure." Or, if abnormal bleeding occurs in the surgical field, the system will prompt: "There is a lot of bleeding at the surgical site, which exceeds the normal range. Please pay attention to hemostasis."

[0063] The system processes intraoperative reminders and abnormal information to generate intraoperative warnings. It integrates these warnings and generates warnings based on pre-defined rules. For example, if intraoperative intraocular pressure rises abnormally and is accompanied by significant bleeding, the warning might be: "Red Warning: The patient's intraocular pressure is out of control and bleeding is severe, which may affect the surgical outcome and the patient's eye function. Please take immediate emergency measures, such as adjusting the surgical procedure, using hemostatic drugs, controlling intraocular pressure, etc., and closely monitor the patient's eye condition." If only a single reminder, such as a surgical step reminder, is issued, the warning might be: "Blue Warning: Please proceed to the next step according to the standard surgical procedure to ensure the successful completion of the surgery." Intraoperative warnings are promptly pushed to Dr. Li's surgical equipment display screen in a prominent manner (such as pop-ups, color indicators, etc.) to ensure Dr. Li can see and react immediately. Simultaneously, the information is also transmitted to the operating room monitoring system so that other medical staff can understand the surgical progress and risks, and provide assistance when necessary.

[0064] S107 processes preoperative assessment information and intraoperative early warning information to generate postoperative early warning information.

[0065] In one implementation, the preoperative assessment information for patient Xiao Zhang showed that he had moderate glaucoma, some damage to the optic nerve, persistently high intraocular pressure, and ocular physiological structures such as the optic disc cup-to-disc ratio of 0.7 (normal range 0.3-0.5), and thinning of the retinal nerve fiber layer. The planned surgery was trabeculectomy. It was anticipated that the risk of bleeding during the surgery might increase due to the fragility of the ocular tissues, and that postoperative intraocular pressure control would be difficult. During the surgery, the system generated intraoperative warning information. For example, when the surgery reached the point of separating the trabecular tissue, the intraocular pressure suddenly increased beyond the normal range, reaching 35 mmHg (normal intraocular pressure 10-21 mmHg), and a small amount of bleeding appeared in the surgical field. The warning message indicated "Red Warning: Intraoperative intraocular pressure has risen sharply and is accompanied by bleeding, which may affect the surgical outcome. Please take immediate measures to control intraocular pressure and stop bleeding, and closely monitor the damage to ocular tissues." After the doctor took timely measures, the intraocular pressure was controlled. However, in the later stages of the surgery, it was found that the formation of the filtering bleb was not ideal. The warning information was updated to "Orange warning: poor formation of the filtering bleb may affect the postoperative aqueous humor drainage effect. It is necessary to closely monitor changes in intraocular pressure after surgery and take further intervention measures if necessary."

[0066] The system comprehensively analyzed preoperative assessment information and intraoperative early warning information. Because Xiao Zhang had preoperative difficulties in controlling intraocular pressure, and intraoperative fluctuations and abnormalities related to the filtering bleb occurred, the risk of further deterioration of glaucoma due to uncontrolled intraocular pressure postoperatively was significantly increased. Simultaneously, intraoperative hemorrhage and suboptimal filtering bleb function could exacerbate ocular inflammation, affecting optic nerve function recovery and even posing a risk of blindness. Furthermore, due to the fragility of ocular tissues, the risk of postoperative infection was relatively high. Based on these risk factors, the key postoperative monitoring indicators were determined to be intraocular pressure, ocular inflammation markers (such as white blood cell count and C-reactive protein), visual acuity, and optic nerve function (such as visual field testing). An intraocular pressure warning threshold of 25 mmHg was set (intraocular pressure fluctuations are significant in the early postoperative period, but a sustained value above this may indicate failure of intraocular pressure control). An early warning was triggered when ocular inflammation markers exceeded a certain percentage of the normal range, and timely alerts were also issued when visual acuity decreased beyond a certain level or the area of ​​visual field defects expanded. For Xiao Zhang's case, the generated postoperative warning information is as follows: "Postoperatively, patient Xiao Zhang needs to closely monitor intraocular pressure, measuring it every 2 hours. If the intraocular pressure exceeds 25 mmHg twice consecutively, or if symptoms such as eye pain, sudden decrease in vision, worsening eye redness and swelling, or increased discharge occur, it indicates possible uncontrolled intraocular pressure or eye infection, requiring immediate medical attention. Visual acuity and visual field tests should be performed daily for one week postoperatively to observe changes in optic nerve function. If vision decreases by more than two lines or visual field defects expand, timely treatment measures should be taken. Simultaneously, anti-inflammatory drugs should be used as prescribed to prevent the inflammatory response from worsening eye damage. Given the unsatisfactory formation of the filtering bleb during surgery, regular follow-up examinations are required within one month postoperatively. Depending on the intraocular pressure, further surgical intervention may be necessary to improve aqueous humor drainage."

[0067] Based on the postoperative warning information, the nurses developed a detailed care plan, including timely measurement of Xiao Zhang's intraocular pressure, close monitoring of eye symptoms, and ensuring the timely and accurate administration of anti-inflammatory medications. The doctors, based on the warning information, prepared treatment plans for potential complications, such as preparing intraocular pressure-lowering medications and developing anti-infection treatment plans. Medical staff provided detailed postoperative care education to Xiao Zhang and his family based on the warning information, informing them how to identify abnormal symptoms, the importance of timely follow-up appointments, and how to perform simple self-monitoring of vision and intraocular pressure at home (e.g., using a home tonometer). Understanding the risks of his condition, Xiao Zhang was able to better cooperate with medical staff during postoperative recovery and improve his self-management abilities. During Xiao Zhang's postoperative recovery, the system continuously updated and adjusted the postoperative warning information based on actual monitoring data and changes in his condition. For example, if Xiao Zhang's intraocular pressure was well controlled and there were no obvious abnormal symptoms in the first few days after surgery, the system might appropriately extend the intraocular pressure monitoring interval. Conversely, if intraocular pressure fluctuations approached the warning threshold, the system would promptly issue a stricter warning, reminding medical staff to strengthen observation and treatment.

[0068] Optionally, in another embodiment based on the method described above in this application, after processing the preoperative assessment information and the intraoperative early warning information to generate postoperative early warning information, the method further includes: Obtain consultation and answer information from the target doctor, whereby the consultation and answer information from the target doctor represents the answer information to the consultation information of the target user; The consultation and answer information of the target doctor is processed to generate other consultation and answer information, which are the answers from other doctors that match the consultation information of the target user. The system processes other consultation and response information and the consultation and response information of the target doctor to generate consultation and response information difference values; If the difference between the consultation and answer information is lower than the preset threshold, then the target consultation and answer information of the target doctor will be generated. The target doctor's consultation and response information is processed to generate the target doctor's medical condition consultation information; The system generates answers to questions about a patient's condition based on the patient's consultation with the target doctor. The postoperative warning information is processed based on the patient's medical condition information to generate updated postoperative warning information.

[0069] In one implementation, during his recovery period after eye surgery, patient Xiao Li submitted a consultation message to the system via a mobile app: "It's been a week since my surgery, and my vision is still a bit blurry, and I feel like there's something in my eye. Is this normal?" The system forwarded this consultation message to the target doctor, Dr. Zhang. Based on his professional knowledge and clinical experience, Dr. Zhang replied on the doctor's app: "Some blurriness and a feeling of something in your eye are normal to some extent within a week after surgery. However, if the symptoms persist or worsen, you need to come to the hospital for a follow-up examination. During this period, pay attention to keeping your eyes clean, avoid rubbing your eyes, and continue to use eye drops as prescribed." The system then retrieved Dr. Zhang's consultation response.

[0070] The system searches its database for other doctors' answers to similar inquiries (such as blurred vision and foreign body sensation after eye surgery). For example, if the search finds Dr. Wang's previous answer to a similar question: "Blurred vision and foreign body sensation after surgery may be due to incomplete healing of surgical trauma or eye inflammation. It is recommended to observe for a few days. If the foreign body sensation is severe or vision does not improve, you can come to the hospital for a follow-up examination. Also, pay attention to rest and avoid prolonged eye use," the system extracts Dr. Wang's answer as additional information for other inquiries.

[0071] The system analyzes the responses from Dr. Zhang and Dr. Wang using natural language processing technology. The difference value can be calculated based on multiple dimensions, such as keyword matching and semantic similarity. For example, the similarity of key content regarding symptom duration assessment and coping strategies in the two responses can be compared. If both responses agree that "persistent symptoms or worsening require a follow-up visit" (keywords match perfectly, high similarity), but Dr. Zhang emphasizes maintaining eye hygiene and taking medication as prescribed, while Dr. Wang focuses more on rest and avoiding prolonged eye strain (some keywords differ, moderate similarity), the difference value is calculated by combining these factors. Assuming that after a series of calculations, this difference value is 0.3 (out of 1.0, indicating a small difference).

[0072] The system generates target consultation and answer information along with patient medical consultation information. Specifically, a preset threshold of 0.5 is used. Since the calculated difference value of 0.3 is lower than the preset threshold, the system considers Dr. Zhang's answer to have high credibility and consistency. Therefore, the system generates Dr. Zhang's target consultation and answer information: "Some blurred vision and a foreign body sensation in the eyes within a week after surgery are normal to some extent. However, if the symptoms persist or worsen, a follow-up examination at the hospital is necessary. During this period, it is important to keep the eyes clean, avoid rubbing them, and continue using eye drops as prescribed." The system further processes the target consultation and answer information, extracting key information to generate the target doctor's patient medical consultation information. For example, it extracts key content such as "symptom status one week after surgery," "criteria for judging persistent or worsening symptoms," and "coping measures (eye cleaning, medication, avoiding eye rubbing, etc.)" to form patient medical consultation information for subsequent comparison and analysis with the patient's medical consultation information.

[0073] After receiving Dr. Zhang's answer, Xiao Li responded to the medical advice on the app based on her own experience: "I followed the doctor's instructions, and the foreign body sensation in my eye seems to have lessened slightly these past two days, but my vision is still not very clear, and things are still a bit blurry." The system updated its post-operative warning information based on Xiao Li's response. Originally, the post-operative warning might have only required Xiao Li to have a routine check-up two weeks after surgery, but now, due to the persistent blurred vision, the system adjusted the warning, advising Xiao Li to closely monitor changes in vision. If her vision does not improve within the next three days, she should contact her doctor as soon as possible to arrange a follow-up examination. At the same time, the system may add some reminders regarding precautions for vision recovery, such as avoiding strong light stimulation and appropriately supplementing eye nutrition, generating updated post-operative warning information to better guide Xiao Li's post-operative recovery process.

[0074] This application aims to assess and warn of patients' eye health risks through multimodal data analysis and machine learning models. The system first collects detailed information about target users, including personal information, purpose of medical visit, and work attributes. This information is then used to generate assessment reports that are aligned with doctors' treatment priorities, ensuring timely responses in emergencies. The training sample set is processed using preset rules to generate ocular and validation sets for training. These sets contain both discontinuous and continuous variable data, such as ocular physiological reference intervals, which are crucial for building accurate warning models. During model training, the system uses ocular feature information and co-occurrence frequencies to generate a correlation matrix, thereby constructing a prior knowledge graph. This knowledge graph is used to train the ocular warning model, and its performance is evaluated using the validation set. If the model can accurately identify ocular disease characteristics, it will be considered the target ocular warning model. Next, the system uses the target ocular warning model to process the target users' assessment information, generating high-risk label information. This information is further used to generate preoperative assessment information, including ocular image processing and feature map analysis, to identify potential ocular abnormalities.

[0075] During surgery, the system acquires and processes the doctor's voice information in real time to generate intraoperative risk information. This information is then used to generate intraoperative warnings, including alerts and abnormal information, to assist the doctor in the operation. Finally, the system also processes preoperative assessment and intraoperative warning information to generate postoperative warning information. In addition, the system collects the doctor's consultation answers and compares them with those of other doctors to generate a consultation answer difference value. If the difference value is below a preset threshold, the system generates target consultation answer information from the target doctor, which helps improve the patient's understanding of the postoperative condition and satisfaction. The entire system is designed to improve the accuracy and efficiency of diagnosing eye diseases, reduce surgical risks, and provide patients with personalized postoperative management advice. In this way, medical professionals can better understand and manage the patient's health status, thereby improving treatment outcomes and patient satisfaction.

[0076] In one implementation, such as Figure 2 As shown, this application also provides a perioperative management device based on the doctor's end, including: The acquisition module 201 is used to receive target user assessment information, preset eye early warning model and training sample set sent by the doctor terminal. The target user assessment information is generated based on multimodal target user information, target user's medical purpose information and target user's work attribute information sent by the client. The doctor terminal is matched with the processing priority of the target user. The processing module 202 is used to process the training sample set according to preset processing rules to generate a target training set and a target validation set; process the preset eye warning model based on the target training set and the target validation set to generate a target eye warning model; process the target user assessment information based on the target eye warning model to generate high-risk eye label information for the target user; process the high-risk eye label information for the target user to generate preoperative assessment information; process the target user based on the preoperative assessment information to generate intraoperative warning information; and process the preoperative assessment information and the intraoperative warning information to generate postoperative warning information.

[0077] The computer-readable storage medium provided in the above embodiments of this application and the perioperative management method based on the doctor's end provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application stored therein.

[0078] The computer program product provided in the above embodiments of this application and the perioperative management method based on the doctor's end provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application stored therein.

[0079] The various embodiments in this application are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for evaluating the physician-based perioperative management method, electronic device, electronic device, and readable storage medium are relatively simple in description because they are substantially similar to the physician-based perioperative management method embodiments described above. Relevant parts can be referred to the descriptions of the physician-based perioperative management method embodiments described above.

[0080] While this application discloses the above information, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of this application; therefore, the scope of protection of this application shall be determined by the scope defined in the claims.

Claims

1. A perioperative management method based on the doctor's end, characterized in that, include: The system receives target user assessment information, a preset eye warning model, and a training sample set sent from the doctor's end. The target user assessment information is generated based on multimodal target user information, target user's purpose of medical treatment information, and target user's work attribute information sent from the client. The doctor's end is matched with the processing priority of the target user. The training sample set is processed based on preset processing rules to generate the target training set and the target validation set; The preset eye warning model is processed based on the target training set and the target validation set to generate a target eye warning model; Based on the target eye early warning model, the target user assessment information is processed to generate high-risk eye tag information for the target user; Process the high-risk ocular label information of the target users to generate preoperative assessment information; Based on preoperative assessment information, target users are processed to generate intraoperative early warning information; The preoperative assessment information and intraoperative early warning information are processed to generate postoperative early warning information.

2. The method as described in claim 1, characterized in that, The training sample set is processed based on preset processing rules to generate a target training set and a target validation set, including: The training sample set is processed to generate a target feature library, which includes non-continuous variable data and continuous variable data. The training sample set is processed based on non-continuous variable data to generate an initial training set and an initial validation set; The initial training set and initial validation set are processed based on continuous variable data to generate the target training set and target validation set. The target training set includes several ocular physiological reference intervals, and different ocular physiological reference intervals correspond to different user information of the same disease type.

3. The method as described in claim 2, characterized in that, The pre-defined eye warning model is processed based on the target training set and the target validation set to generate a target eye warning model, including: The target training set is processed to generate eye feature information and correlation co-occurrence frequency; The eye feature information and related co-occurrence frequencies are processed to generate a correlation matrix. When generating the correlation matrix information, the association analysis between systemic disease characteristics and ocular characteristics is further incorporated to construct a sub-matrix of correlation between systemic and ocular characteristics; A prior knowledge graph for characterizing eye diseases is generated based on correlation matrix information and correlation sub-matrices of whole-body and ocular features. The pre-set eye warning model is processed based on the prior knowledge graph to generate a trained eye warning model. The trained eye warning model is processed based on the target validation set to generate validation results; If the data sample containing the identification information in the verification result is an ocular feature information characterizing an eye disease, then the trained ocular early warning model will be used as the target ocular early warning model. The method also includes a formula for calculating the covariance matrix, which is as follows: ； ； ； Among them, the elements in the covariance matrix C Let represent the covariance between the i-th eye feature and the j-th eye feature. This represents the mean of the i-th eye feature. This represents the i-th eye feature value of the k-th sample. This represents the j-th eye feature value of the k-th sample; The method also includes a formula for calculating the standard deviation vector, which is as follows: ； ； in, The standard deviation of the i-th eye feature is represented by... The standard deviation of the j-th eye feature; The method also includes a formula for calculating the correlation matrix, which is as follows: ； Among them, the elements in the correlation matrix R The correlation coefficient represents the relationship between the i-th eye feature and the j-th eye feature.

4. The method as described in claim 1, characterized in that, Based on the target eye early warning model, the assessment information of the target user is processed to generate high-risk eye tag information for the target user, including: The multimodal target user information is processed to generate real-time ocular physiological status information and real-time environmental information of the target user; The medical purpose information of target users is processed to generate the first ocular impact factor; The target user's work attribute information is processed to generate a second eye influence factor; The first and second ocular influence factors are processed to generate ocular feature influence factors and corresponding weight information. Based on the target eye early warning model, the real-time eye physiological status information of the target user, the real-time environmental information of the target user, the eye feature influencing factors and the weight information corresponding to the eye feature influencing factors are processed to generate the eye risk early warning value of the target user. Based on a preset eye risk warning table, the eye risk warning value of the target user is processed to generate the target user's high-risk eye label information. The target user's high-risk eye label information is used to indicate that the target user's eye risk warning value is greater than the preset eye risk warning value.

5. The method as described in claim 4, characterized in that, The high-risk ocular tagged information of the target users is processed to generate preoperative assessment information, including: The high-risk eye tag information of the target user is processed to generate eye image information that matches the high-risk eye tag information of the target user; The eye image information is sliced ​​to generate the target eye slice image information; The target eye slice image information is processed to generate feature maps at different levels. The feature maps at the same level include feature information of different channel dimensions. The feature maps of the same channel dimension are processed separately to generate eye anomaly markers; The abnormal eye markers are processed separately to generate preoperative assessment information, which is used to characterize the eye risk of the target user.

6. The method as described in claim 5, characterized in that, Based on preoperative assessment information, target users are processed to generate intraoperative early warning information, including: Acquire the target physician's voice information during surgery, whereby the voice information during surgery is used to characterize the intraoperative record information and preoperative assessment information of the target user; The voice information of the target doctor during the operation is processed to generate intraoperative risk information for the target user; The intraoperative risk information of the target user is processed to generate intraoperative reminder information and intraoperative abnormal information. The intraoperative reminder information is used to remind the target doctor of the subsequent operation process and the surgical completion information. Intraoperative reminders and abnormal information are processed to generate intraoperative early warning information.

7. The method as described in claim 6, characterized in that, After processing preoperative assessment information and intraoperative early warning information to generate postoperative early warning information, the following also includes: Obtain consultation and answer information from the target doctor, whereby the consultation and answer information from the target doctor represents the answer information to the consultation information of the target user; The consultation and answer information of the target doctor is processed to generate other consultation and answer information, which are the answers from other doctors that match the consultation information of the target user. The system processes other consultation and response information and the consultation and response information of the target doctor to generate consultation and response information difference values; If the difference between the consultation and answer information is lower than the preset threshold, then the target consultation and answer information of the target doctor will be generated. The target doctor's consultation and response information is processed to generate the target doctor's medical condition consultation information; The system generates answers to questions about a patient's condition based on the patient's consultation with the target doctor. The postoperative warning information is processed based on the patient's medical condition information to generate updated postoperative warning information.

8. A perioperative management device based on a doctor's end, characterized in that, The device includes: The acquisition module is used to receive target user assessment information, preset eye early warning model and training sample set sent by the doctor. The target user assessment information is generated based on the multimodal target user information, target user's purpose of medical treatment information and target user's work attribute information sent by the client. The doctor's processing priority is matched with the target user. The processing module is used to process the training sample set according to preset processing rules to generate a target training set and a target validation set; process the preset eye warning model based on the target training set and the target validation set to generate a target eye warning model; process the target user assessment information based on the target eye warning model to generate high-risk eye label information for the target user; process the high-risk eye label information for the target user to generate preoperative assessment information; process the target user based on the preoperative assessment information to generate intraoperative warning information; and process the preoperative assessment information and intraoperative warning information to generate postoperative warning information.

9. An electronic device, characterized in that, include: First processor; and memory for storing executable instructions of the first processor; The first processor is configured to execute the perioperative management method based on any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the perioperative management method based on the doctor's end as described in any one of claims 1 to 7.

Images