Ophthalmology nursing grade automatic evaluation system based on multi-source data fusion

Abstract

The invention discloses an ophthalmology nursing grade automatic evaluation system based on multi-source data fusion, and belongs to the technical field of ophthalmology nursing, and the system comprises a multi-source data collection module which is used for collecting ophthalmology patient multi-source data; the data processing and fusing module is used for preprocessing and fusing the multi-source data of the ophthalmology patient to form ophthalmology patient fused data; and the intelligent analysis and evaluation module is used for constructing an ophthalmology patient risk evaluation model, analyzing and identifying the ophthalmology patient fusion data according to the ophthalmology patient risk evaluation model, determining an ophthalmology patient risk evaluation result, and automatically dividing nursing grades according to the ophthalmology patient risk evaluation result. The problems that the objectivity and accuracy of ophthalmology nursing grade evaluation are low and the medical quality and efficiency are reduced due to the fact that automatic evaluation of the ophthalmology nursing grade cannot be achieved in the prior art are solved. According to the invention, the automatic evaluation of the ophthalmic nursing grade can be realized, the objectivity and accuracy of the evaluation of the ophthalmic nursing grade can be improved, and the medical quality and efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ophthalmic nursing, in particular to an ophthalmic nursing level automatic evaluation system based on multi-source data fusion. BACKGROUND

[0002] Ophthalmic nursing is a professional nursing service provided for patients with eye diseases and visual impairment, including eye examination, disease nursing, psychological support, health education and postoperative rehabilitation guidance, etc. Ophthalmic nursing has the following importance: 1) ensuring eye health: by timely discovering and handling eye problems, preventing disease deterioration and complications; 2) improving quality of life: improving patients' visual function, reducing eye discomfort, and improving patients' quality of life and happiness; 3) promoting medical effect: enhancing patients' treatment compliance, improving treatment effect, and reducing waste of medical resources. In ophthalmic nursing, the patient's eye is nursed according to the nursing level, and the nursing level evaluation is particularly important.

[0003] Traditional nursing evaluation mainly relies on manual experience, which is affected by the level of medical staff, work load and subjective judgment, and the evaluation result may have deviation, which cannot realize automatic evaluation of ophthalmic nursing level, resulting in low objectivity and accuracy of ophthalmic nursing level evaluation, and reducing medical quality and efficiency. SUMMARY

[0004] The purpose of the present application is to provide an ophthalmic nursing level automatic evaluation system based on multi-source data fusion, which can realize automatic evaluation of ophthalmic nursing level, improve the objectivity and accuracy of ophthalmic nursing level evaluation, improve medical quality and efficiency, and solve the problems raised in the above background.

[0005] To achieve the above purpose, the present application provides the following technical scheme: The ophthalmic nursing level automatic evaluation system based on multi-source data fusion comprises: A multi-source data acquisition module for acquiring ophthalmic patient multi-source data based on a multi-source data interface; A data processing and fusion module for pre-processing and fusing ophthalmic patient multi-source data based on natural language processing to form ophthalmic patient fusion data; An intelligent analysis and evaluation module for constructing an ophthalmic patient risk assessment model based on artificial intelligence, analyzing and identifying ophthalmic patient fusion data according to the ophthalmic patient risk assessment model, determining an ophthalmic patient risk assessment result, and automatically dividing a nursing level according to the ophthalmic patient risk assessment result.

[0006] Preferably, the multi-source data acquisition module comprises: An image data acquisition unit is configured to acquire image data of the ophthalmic patient according to color fundus photography, optical coherence tomography images, OCT angiography, perimetry, and anterior segment photography of the ophthalmic patient. A clinical data acquisition unit is configured to acquire clinical data of the ophthalmic patient according to medical history, diagnosis, surgical records, medication history, lifestyle, and quality of life scale of the ophthalmic patient. A monitoring data acquisition unit is configured to acquire monitoring data of the ophthalmic patient according to intraocular pressure, visual acuity, refraction, diopter, corneal curvature, and dynamic visual acuity changes of the ophthalmic patient. The ophthalmic patient multi-source data is formed according to the ophthalmic patient image data, the ophthalmic patient clinical data, and the ophthalmic patient monitoring data.

[0007] Preferably, the data processing fusion module comprises: A multi-source data cleaning unit is configured to clean the ophthalmic patient multi-source data based on natural language processing, remove noise in the ophthalmic patient multi-source data, identify missing values and abnormal values in the ophthalmic patient multi-source data, and process the identified missing values and abnormal values. A multi-source data standardization unit is configured to process the ophthalmic patient multi-source data based on Z-Score standardization, convert the ophthalmic patient multi-source data into a normal distribution with a mean of 0 and a standard deviation of 1, unify the data format, crop, enhance, and normalize the image, and form standardized ophthalmic patient multi-source data.

[0008] Preferably, the data processing fusion module further comprises: A data extraction fusion unit is configured to extract features from the ophthalmic patient multi-source data based on principal component analysis, extract feature vectors related to automatic evaluation of the ophthalmic care level from the ophthalmic patient multi-source data, and weight and fuse the extracted feature vectors to form ophthalmic patient fusion data. The pre-trained deep learning model is used to extract image features from the ophthalmic patient multi-source data, automatically extract lesion features including microaneurysms, hemorrhage, and RNFL thickness, encode non-image features of the ophthalmic patient multi-source data, and convert structured data of intraocular pressure, visual acuity, refraction, diopter, and corneal curvature and text data of medical history, diagnosis, surgical records, and medication history into numerical vectors.

[0009] Preferably, the intelligent analysis and evaluation module comprises: An evaluation model construction unit is configured to construct an ophthalmic patient risk evaluation model based on artificial intelligence. According to the automatic evaluation requirements of the ophthalmic care level, the ophthalmic patient risk evaluation historical data is collected, and the collected ophthalmic patient risk evaluation historical data is divided, wherein the ophthalmic patient risk evaluation historical data is divided into a training set and a test set according to a ratio of 7:3. The deep learning model is trained by using the training set, so that the deep learning model learns the ophthalmic patient risk assessment behavior from the training set and accurately assesses the eye risk of the ophthalmic patient, and determines the ophthalmic patient risk assessment model; The ophthalmic patient risk assessment model is tested and evaluated by using the test set, to determine whether the ophthalmic patient risk assessment model can accurately assess the eye risk of the ophthalmic patient, and determine the model test evaluation result; When the ophthalmic patient risk assessment model cannot accurately assess the eye risk of the ophthalmic patient, the parameters of the ophthalmic patient risk assessment model are adjusted and optimized until the ophthalmic patient risk assessment model can accurately assess the eye risk of the ophthalmic patient, and the optimal ophthalmic patient risk assessment model is determined.

[0010] Preferably, the deep learning model is trained by using the training set, so that the deep learning model learns the ophthalmic patient risk assessment behavior from the training set and accurately assesses the eye risk of the ophthalmic patient, and determines the ophthalmic patient risk assessment model, comprising: The training set is divided into four types of glaucoma, cataract, diabetic retinopathy, and maculopathy, to obtain four sub-training sets; The intraocular pressure dynamic curve, RNFL thickness, and visual field defect range in the glaucoma sub-training set are enhanced in feature weight, the optic nerve fiber layer defect progression rate is taken as a core evaluation index, the deep learning model is trained, and a glaucoma sub-model is obtained; The blood glucose control level, glycosylated hemoglobin, and microaneurysm number change in the diabetic retinopathy sub-training set are analyzed, the analysis result is taken as an incremental feature, the bleeding point addition speed is taken as the basis for dividing the emergency nursing level, the deep learning model is trained, and a diabetic retinopathy sub-model is obtained; The lens opacity degree, visual acuity decline rate, and postoperative complication risk in the cataract sub-training set are enhanced in feature weight, the deep learning model is trained, and a cataract sub-model is obtained; The macular foveal retinal thickness, outer retinal integrity, and macular hemorrhage range in the maculopathy sub-training set are enhanced in feature weight, and the nursing level mapping accuracy is taken as a core evaluation index, the deep learning model is trained, and a maculopathy sub-model is obtained; The data features of the four sub-training sets are learned, and a disease type automatic recognition mechanism is established; The disease type automatic recognition mechanism is added in front of the glaucoma sub-model, the diabetic retinopathy sub-model, the cataract sub-model, and the maculopathy sub-model, and an ophthalmic patient risk assessment model is established.

[0011] Preferably, the intelligent analysis and evaluation module further comprises: an eye risk assessment unit for accurately assessing the eye risk condition of an ophthalmic patient; deploying the ophthalmic patient risk assessment model in an actual ophthalmic patient risk assessment environment for accurately assessing the eye risk condition of an ophthalmic patient; wherein the ophthalmic patient fusion data is input as input data into the ophthalmic patient risk assessment model, and the ophthalmic patient risk assessment model is used to analyze and identify the ophthalmic patient fusion data to automatically and accurately assess the eye risk condition of the ophthalmic patient and determine the ophthalmic patient risk assessment result.

[0012] Preferably, the intelligent analysis and evaluation module further comprises: a nursing level division unit for automatically dividing the nursing level according to the ophthalmic patient risk assessment result; embedding clinical guidelines and expert knowledge based on the grading rule base, mapping the ophthalmic patient risk assessment result to a specific nursing level, and automatically dividing different levels according to the patient's nursing needs, including level I, level II, level III, and level IV; wherein level I requires annual routine review, level II requires 3-6 month specialist follow-up, level III requires emergency treatment within 1 month, and level IV requires emergency treatment within 48 hours; wherein the ophthalmic patient risk assessment result automatically triggers an alarm and notifies the doctor or nursing team for emergency treatment for patients in level IV.

[0013] Preferably, the intelligent analysis and evaluation module further comprises: a visual display unit for displaying the ophthalmic nursing assessment report in a visual form; forming the ophthalmic nursing assessment report of the patient according to the ophthalmic patient risk assessment result and the nursing level of the patient, and displaying the ophthalmic nursing assessment report of the patient in a visual form in real time to provide data-driven decision support for ophthalmologists and community nursing staff.

[0014] Preferably, when the ophthalmic patient risk assessment model cannot accurately assess the eye risk condition of the ophthalmic patient, the ophthalmic patient risk assessment model is adjusted and optimized, including: obtaining the predicted nursing level of the ophthalmic patient risk assessment model after a preset time, obtaining the inconsistent predicted nursing level to be corrected from the predicted nursing level and the actual nursing level, and obtaining the sample data corresponding to the predicted nursing level to be corrected as the alternative correction sample; based on the expert score and nursing effect test of the alternative correction sample, the score of each alternative correction sample is calculated; Select the candidate correction sample with a score greater than a preset score as a target correction sample; Based on the dimension normalization value of the target correction sample, the dimension weight of each temperature feature in the target correction sample is calculated; Based on the target correction sample and its corresponding dimension weight, and the corresponding actual nursing level, the ophthalmic patient risk assessment model is trained and optimized.

[0015] Compared with the prior art, the beneficial effects of the present application are: The present application collects ophthalmic patient image data, ophthalmic patient clinical data and ophthalmic patient monitoring data through a multi-source data interface, forms ophthalmic patient multi-source data, pre-processes and fuses the ophthalmic patient multi-source data based on natural language processing, extracts feature vectors related to automatic evaluation of ophthalmic nursing level from the ophthalmic patient multi-source data, and performs weighted fusion on the extracted feature vectors to form ophthalmic patient fusion data, constructs an ophthalmic patient risk assessment model based on artificial intelligence, analyzes and identifies the ophthalmic patient fusion data according to the ophthalmic patient risk assessment model, automatically and accurately evaluates the ophthalmic patient's eye risk, determines the ophthalmic patient risk assessment result, and automatically divides the nursing level according to the ophthalmic patient risk assessment result. It can realize the automatic evaluation of ophthalmic nursing level, improve the objectivity and accuracy of ophthalmic nursing level evaluation, and improve the medical quality and efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0016] Fig. 1 The module diagram of the ophthalmic nursing level automatic evaluation system of the present application; Fig. 2 The flowchart of the ophthalmic nursing level automatic evaluation system of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] In order to solve the problem that the existing technology cannot realize automatic evaluation of ophthalmic nursing level, resulting in low objectivity and accuracy of ophthalmic nursing level evaluation, and reducing medical quality and efficiency, please refer to Figs. 1-2 The technical solutions provided in the present embodiment are as follows: The ophthalmic nursing level automatic evaluation system based on multi-source data fusion comprises a multi-source data acquisition module, a data processing and fusion module, and an intelligent analysis and evaluation module.

[0019] The multi-source data acquisition module is configured to acquire multi-source data of the ophthalmic patient based on a multi-source data interface.

[0020] In this embodiment, the multi-source data acquisition module includes: The image data acquisition unit is configured to acquire image data of the ophthalmic patient according to color fundus photography, optical coherence tomography images, OCT angiography, perimetry, and anterior segment photography of the ophthalmic patient. It should be noted that the image data of the ophthalmic patient can provide structural and functional direct evidence and is the core of disease diagnosis, such as retinal thickness, cup-disc ratio, vascular abnormalities, hemorrhage and exudation, etc.

[0021] The clinical data acquisition unit is configured to acquire clinical data of the ophthalmic patient according to medical history, diagnosis, surgical records, medication history, lifestyle, and quality of life scale of the ophthalmic patient. It should be noted that the clinical data of the ophthalmic patient can provide medical history background and clinical context, such as diabetes duration, hypertension control, previous laser treatment history, and provide risk factors and subjective feeling information.

[0022] The monitoring data acquisition unit is configured to acquire monitoring data of the ophthalmic patient according to intraocular pressure, visual acuity, refraction, corneal curvature, and dynamic visual acuity changes of the ophthalmic patient. It should be noted that the monitoring data of the ophthalmic patient can provide key physiological parameters and functional indicators, as well as continuous and real-time disease condition change trends.

[0023] The multi-source data of the ophthalmic patient is formed according to the image data of the ophthalmic patient, the clinical data of the ophthalmic patient, and the monitoring data of the ophthalmic patient.

[0024] The data processing and fusion module is configured to preprocess and fuse the multi-source data of the ophthalmic patient based on natural language processing to form fusion data of the ophthalmic patient.

[0025] Specifically, the image data of the ophthalmic patient, the clinical data of the ophthalmic patient, and the monitoring data of the ophthalmic patient are acquired through the multi-source data interface to form the multi-source data of the ophthalmic patient, which can provide data support for subsequent accurate assessment of the eye risk of the ophthalmic patient.

[0026] In this embodiment, the data processing and fusion module includes: The multi-source data cleaning unit is configured to clean the multi-source data of the ophthalmic patient based on natural language processing to remove noise in the multi-source data of the ophthalmic patient, which can reduce the interference of noise data on the accurate assessment of the eye risk of the ophthalmic patient, identify missing values and abnormal values in the multi-source data of the ophthalmic patient, and process the identified missing values and abnormal values. The multi-source data standardization unit is used to process multi-source data of ophthalmology patients based on Z-Score standardization. It converts the multi-source data of ophthalmology patients into a normal distribution with a mean of 0 and a standard deviation of 1, unifies the data format, and performs cropping, enhancement and normalization of images to form standardized multi-source data of ophthalmology patients.

[0027] In this embodiment, the data processing and fusion module further includes: The data extraction and fusion unit is used to extract features from multi-source data of ophthalmology patients based on principal component analysis. It extracts feature vectors related to the automatic assessment of ophthalmology care level from the multi-source data of ophthalmology patients, and performs weighted fusion of the extracted feature vectors to form fused data of ophthalmology patients. Among them, a pre-trained deep learning model is used to extract image features from multi-source data of ophthalmology patients, automatically extract lesion features, including microaneurysms, hemorrhages and RNFL thickness, encode non-image features from multi-source data of ophthalmology patients, and convert structured data of intraocular pressure, visual acuity, refraction, refractive power and corneal curvature, as well as text data of medical history, diagnosis, surgical records and medication history into numerical vectors.

[0028] Specifically, based on natural language processing, multi-source data of ophthalmology patients is preprocessed and fused. Feature vectors related to the automatic assessment of ophthalmology care level are extracted from the multi-source data of ophthalmology patients, and the extracted feature vectors are weighted and fused to form fused data of ophthalmology patients, which can better and more accurately assess the ophthalmic risk of ophthalmology patients.

[0029] The intelligent analysis and assessment module is used to build an ophthalmology patient risk assessment model based on artificial intelligence. It analyzes and identifies the fused data of ophthalmology patients according to the ophthalmology patient risk assessment model, determines the ophthalmology patient risk assessment results, and automatically classifies the nursing level according to the ophthalmology patient risk assessment results.

[0030] In this embodiment, the intelligent analysis and evaluation module includes: The assessment model building unit is used to build an ophthalmic patient risk assessment model based on artificial intelligence. Based on the need for automatic assessment of ophthalmic care levels, historical data on risk assessment of ophthalmic patients were collected and divided into training and testing sets in a 7:3 ratio. The deep learning model is trained using a training set, enabling it to autonomously learn the risk assessment behavior of ophthalmology patients from the training set and accurately assess the ophthalmology risk status of patients, thus determining the risk assessment model for ophthalmology patients. The test set is used to test and evaluate the ophthalmic patient risk assessment model to determine whether the ophthalmic patient risk assessment model can accurately assess the eye risk of the ophthalmic patient, and to determine the model test evaluation result; When the ophthalmic patient risk assessment model cannot accurately assess the eye risk of the ophthalmic patient, the parameters of the ophthalmic patient risk assessment model are adjusted and optimized until the ophthalmic patient risk assessment model can accurately assess the eye risk of the ophthalmic patient, and the optimal ophthalmic patient risk assessment model is determined.

[0031] In one embodiment, the training set is used to train the deep learning model, so that the deep learning model can learn the ophthalmic patient risk assessment behavior from the training set and accurately assess the eye risk of the ophthalmic patient, and determine the ophthalmic patient risk assessment model, including: The training set is divided into four types of glaucoma, cataract, diabetic retinopathy, and maculopathy, and four sub-training sets are obtained; The intraocular pressure dynamic curve, RNFL thickness, and visual field defect range in the glaucoma sub-training set are enhanced in feature weight, the optic nerve fiber layer defect progression rate is used as a core evaluation index to train the deep learning model, and a glaucoma sub-model is obtained; The blood glucose control level, glycosylated hemoglobin, and microaneurysm number change in the diabetic retinopathy sub-training set are analyzed, the correlation analysis result is used as an incremental feature, the hemorrhagic point addition speed is used as the basis for dividing the emergency care level, the deep learning model is trained, and a diabetic retinopathy sub-model is obtained; The lens opacity, visual acuity decline rate, and postoperative complication risk in the cataract sub-training set are enhanced in feature weight, and the deep learning model is trained to obtain a cataract sub-model; The macular foveal retinal thickness, outer retinal integrity, and macular hemorrhage range in the maculopathy sub-training set are enhanced in feature weight, and the deep learning model is trained with the nursing level mapping accuracy as the core evaluation index to obtain a maculopathy sub-model; The data features of the four types of sub-training sets are learned to establish a disease type automatic recognition mechanism; The disease type automatic recognition mechanism is added in front of the glaucoma sub-model, the diabetic retinopathy sub-model, the cataract sub-model, and the maculopathy sub-model to establish an ophthalmic patient risk assessment model.

[0032] In this embodiment, a disease type automatic recognition mechanism is added in front of the glaucoma sub-model, the diabetic retinopathy sub-model, the cataract sub-model and the maculopathy sub-model to establish an ophthalmic patient risk assessment model. In use, when image data identifies microaneurysms and retinal hemorrhage (typical features of diabetic retinopathy) and clinical data is labeled with a history of diabetes, the diabetic retinopathy sub-model is automatically called for evaluation.

[0033] The beneficial effects of the above design scheme are: by dividing the training set into four types of glaucoma, cataract, diabetic retinopathy and maculopathy to obtain four types of sub-training sets, four disease-specific sub-models are designed to enable the model to adapt to the specificity of different ophthalmic diseases. The sub-models specifically strengthen the core feature weights of each disease, avoiding the dilution of key information caused by the sharing of feature weights in the general model. Each sub-model sets core evaluation indicators that fit the disease diagnosis and treatment logic. The automatic recognition mechanism ensures that these exclusive indicators only take effect in the corresponding disease evaluation, so that the evaluation result directly interfaces with the clinical care decision. The design of the disease type automatic recognition mechanism can accurately match the patient's disease type, ensure the on-demand calling of the sub-model, and focus the evaluation on the core risk points of the disease, significantly reducing the probability of misjudgment and omission. At the same time, the sub-model is trained only for the core features of a single disease, and the parameter size is much smaller than that of the general model. The disease type automatic recognition mechanism can be called on demand to avoid the waste of computing power caused by the operation of the full parameters of the general model. Each sub-model is embedded with disease-specific emergency risk trigger logic, and the automatic recognition mechanism ensures that the core warning indicators of high-risk diseases are calculated first to avoid the delay of the general model in giving early warnings due to considering all diseases. The automatic recognition mechanism, combined with the exclusive design of the sub-model, can adapt to multiple scene requirements and provide more practical value for ophthalmic nursing level assessment.

[0034] In this embodiment, the intelligent analysis and evaluation module further includes: An eye risk assessment unit for accurately assessing the eye risk of the ophthalmic patient. The constructed ophthalmic patient risk assessment model is deployed in an actual ophthalmic patient risk assessment environment for accurately assessing the eye risk of the ophthalmic patient. The ophthalmic patient fusion data is input into the ophthalmic patient risk assessment model as input data, and the ophthalmic patient risk assessment model is used to analyze and identify the ophthalmic patient fusion data to accurately assess the eye risk of the ophthalmic patient and determine the ophthalmic patient risk assessment result.

[0035] In this embodiment, the intelligent analysis and evaluation module further includes: The nursing level division unit is configured to automatically divide the nursing level according to the ophthalmic patient risk assessment result. The ophthalmic patient risk assessment result is mapped to a specific nursing level based on the hierarchical rule base embedded with clinical guidelines and expert knowledge, and different levels are automatically divided according to the patient's nursing needs, including level I, level II, level III, and level IV. Among them, level I needs annual routine review; level II needs 3-6 month specialist follow-up; level III needs emergency treatment within 1 month; and level IV needs emergency treatment within 48 hours. Among them, according to the ophthalmic patient risk assessment result, the patient of level IV automatically triggers an alarm and notifies the doctor or nursing team for emergency treatment.

[0036] In this embodiment, the intelligent analysis and evaluation module further comprises: The visualization display unit is configured to display the ophthalmic nursing evaluation report in a visual form. According to the ophthalmic patient risk assessment result and the nursing level of the patient, the ophthalmic nursing evaluation report of the patient is formed, and the ophthalmic nursing evaluation report of the patient is displayed in a visual form in real time, providing data-driven decision support for ophthalmologists and community nursing personnel.

[0037] In summary, through the multi-source data interface, the ophthalmic patient image data, the ophthalmic patient clinical data, and the ophthalmic patient monitoring data are collected to form the ophthalmic patient multi-source data. The ophthalmic patient multi-source data is preprocessed and fused based on natural language processing. The feature vectors related to the automatic evaluation of the ophthalmic nursing level are extracted from the ophthalmic patient multi-source data, and the extracted feature vectors are weighted and fused to form the ophthalmic patient fusion data. An ophthalmic patient risk assessment model is constructed based on artificial intelligence. The ophthalmic patient fusion data is analyzed and recognized according to the ophthalmic patient risk assessment model, the ophthalmic patient's eye risk situation is accurately evaluated automatically, the ophthalmic patient risk assessment result is determined, the nursing level is automatically divided according to the ophthalmic patient risk assessment result, the automatic evaluation of the ophthalmic nursing level can be realized, the objectivity and accuracy of the ophthalmic nursing level evaluation can be improved, and the medical quality and efficiency can be improved.

[0038] The ophthalmic nursing level automatic evaluation system based on multi-source data fusion can be applied to the following scenarios: 1) ophthalmic nursing level evaluation scene: dynamically monitor the patient's state, adjust the nursing level in real time, and for ophthalmic diseases (such as glaucoma, cataract, and fundus lesions), through the analysis of the patient's visual function, intraocular pressure and other indicators, the nursing needs are scientifically evaluated; 2) assist nursing decision-making scene: provide data-driven nursing suggestions for medical staff, reduce the influence of subjective judgment, and in postoperative care, through the analysis of the patient's recovery, the nursing plan is timely adjusted; 3) personalized nursing scene: according to the patient's age, medical history, psychological state and other factors, personalized nursing plan is provided, and for patients with visual impairment, self-care ability is evaluated to develop targeted nursing plan.

[0039] In one embodiment, when the ophthalmic patient risk assessment model cannot accurately assess the eye risk of the ophthalmic patient, the ophthalmic patient risk assessment model is adjusted and optimized, including: obtaining the predicted nursing level of the ophthalmic patient risk assessment model after a preset time, obtaining the to-be-corrected predicted nursing level inconsistent with the actual nursing level from the predicted nursing level; obtaining the sample data corresponding to the to-be-corrected predicted nursing level as the alternative correction sample; Based on the expert score and nursing effect test of the alternative correction sample, the score of each alternative correction sample is calculated ;

[0040] wherein,

[0041] wherein, represents the expert score weight of the alternative correction sample, represents the nursing effect test weight of the alternative correction sample, represents a natural constant, and the value is 2.72, , and respectively represent the confirmation degree of the alternative correction sample by three different experts, represents the standard deviation of the confirmation degree of the alternative correction sample by three different experts, represents the mean of the confirmation degree of the alternative correction sample by three different experts, represents the first calibration coefficient, and the value is 2.3, represents the second calibration coefficient, and the value is 1.8, represents the third calibration coefficient, and the value is 1.5, represents the predicted nursing level of the alternative correction sample, represents the actual nursing level of the alternative correction sample, represents the nursing improvement rate, represents the value of the disease core efficacy index of the alternative revision sample after nursing, represents the value of the disease core efficacy index of the alternative revision sample before nursing, represents the index normal reference value of the alternative revision sample; selecting the alternative revision sample with a score greater than a preset score as a target revision sample; based on the dimension normalization value of the target revision sample, the dimension weight of each dimension in the target revision sample is calculated;

[0042] wherein, represents the dimension weight of the target revision sample in the jth dimension, represents the expert score dimension, represents the nursing effect test dimension, represents the number of target revision samples, represents the normalized value of the ith target revision sample in the jth dimension; based on the target revision sample and its corresponding dimension weight, and the corresponding actual nursing level, the ophthalmic patient risk assessment model is trained and optimized again.

[0043] In this embodiment, based on the target revision sample and its corresponding dimension weight, and the corresponding actual nursing level, the ophthalmic patient risk assessment model is trained and optimized again, which is to integrate the dimension weight into the feature vector, strengthen the influence of high weight dimension, reconstruct the sample feature vector, and take the actual nursing level as the result. The ophthalmic patient risk assessment model is trained and optimized again.

[0044] In this embodiment, the value of the confirmation degree of the expert to the alternative revision sample is 1-5, 5 means considering confirming the need for revision, and 1 means denying the need for revision.

[0045] In this embodiment, is used to represent the score coefficient of variation, which is used to reflect the expert difference. The purpose of the first calibration coefficient is to make the score of the target revision sample greater than the score of the alternative revision sample. = 0, = , = 1, approximately equal to 0.1 .

[0046] In this embodiment, the disease core efficacy index is, for example, the BCVA visual acuity improvement letter number of maculopathy, and the intraocular pressure reduction amplitude of glaucoma.

[0047] In this embodiment, wherein , Let represent the maximum and minimum values ​​of the j-th dimension, respectively. This represents the value of the j-th dimension of the target correction sample.

[0048] In this embodiment, the formula for calculating the nursing improvement rate is designed to ensure that the nursing improvement rate is within the range of [0,1].

[0049] In this embodiment, This represents the second calibration factor, with a value of 1.8. This represents the third calibration factor, with a value of 1.5, to ensure... =2 and When =0.8, =0.9.

[0050] In this embodiment, This represents the information entropy in the j-th dimension.

[0051] In this embodiment, for example , , They are 4.5, 4.2, and 4.3 respectively; =0.4, =0.6, =2, =4, =0.8, =0.5, =0.8, calculated as follows =1.72 + 0.60 = 2.32. When the pre-designed preset score is 2.00, the current sample is determined as the target correction sample. Subsequently, if in the expert scoring dimension, =0.80+0.90+0.70+0.85+0.95=4.20, in the dimension of nursing effectiveness evaluation, =0.70+0.72+0.68+0.75+0.65=3.50, then The approximate values ​​are: 0.80 / 4.20 ≈ 0.1905, 0.90 / 4.20 ≈ 0.2143, 0.70 / 4.20 ≈ 0.1667, 0.85 / 4.20 ≈ 0.2024, and 0.95 / 4.20 ≈ 0.2262. Given 0.70 / 3.50 = 0.2000, 0.72 / 3.50 ≈ 0.2057, 0.68 / 3.50 ≈ 0.1943, 0.75 / 3.50 ≈ 0.2143, 0.65 / 3.50 ≈ 0.1857, and lnS = ln5 ≈ 1.6094, then in the expert rating dimension... = [(0.1905 x ln 0.1905) + (0.2143 x ln 0.2143) + (0.1667 x ln 0.1667) + (0.2024 x ln 0.2024) + (0.2262 x ln 0.2262)] = -1.6094 (-1.503) = 0.943, then = 0.066, then ; in the nursing effect test dimension = -1.6094 [(0.2000 x ln 0.2000) + (0.2057 x ln 0.2057) + (0.1943 x ln 0.1943) + (0.2143 x ln 0.2143) + (0.1857 x ln 0.1857)] = -1.6094 (-1.507) = 0.936, then = 0.064, then .

[0052] The beneficial effects of the above design scheme are: by quantifying and fusing the two core dimensions of expert consensus degree and nursing effect test, it is ensured that the selected target correction sample conforms to the clinical expert consensus, and the objective effectiveness is guaranteed, high-quality sample support is provided for model optimization, the difference of the nursing improvement rate formula is actually differentiated, the quantification of the nursing effect is more in line with the clinical practice, and it is ensured that the selected sample can truly reflect the model bias and nursing correction, the dimension weight is designed by using the entropy weight method, and the actual data distribution of the target correction sample is automatically generated based on the target correction sample, rather than manually setting a fixed weight, if the normalization value of the expert score dimension has high dispersion, the information entropy of the dimension is small, and the dimension weight is larger, which means that the dimension is more valuable for distinguishing high-quality correction samples, if the normalization value of the nursing effect test dimension tends to be the same, the information entropy is large, and the weight is smaller, avoiding invalid information from occupying optimization resources, the dynamic allocation logic automatically inclines to the dimension with high information value during the secondary training of the model, improves the pertinence of optimization, finally, high-quality samples are selected through double-dimension quantification to avoid interference from poor data; the dimension weight is dynamically allocated by using the entropy weight method, so that the optimization focuses on the core bias; through targeted secondary training, the gap between the model and the clinical practice is directly corrected; meanwhile, the long-term iteration ability and the clinical operability are taken into account, and finally the multiple goals of improving the model evaluation accuracy, making the clinical nursing decision more scientific, and reducing the medical risk are achieved, which perfectly adapts to the clinical application scene of the ophthalmic nursing grade automatic evaluation system.

[0053] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0054] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, and it is intended that the scope of the application be limited solely by the scope of the appended claims and the equivalents thereof.

Claims

1. An automatic assessment system for ophthalmic care levels based on multi-source data fusion, characterized in that, include: Multi-source data acquisition module, used to acquire multi-source data from ophthalmology patients based on a multi-source data interface; The data processing and fusion module is used to preprocess and fuse multi-source data of ophthalmology patients based on natural language processing to form fused data of ophthalmology patients; The intelligent analysis and assessment module is used to build an ophthalmology patient risk assessment model based on artificial intelligence. It analyzes and identifies the fused data of ophthalmology patients according to the ophthalmology patient risk assessment model, determines the risk assessment results of ophthalmology patients, and automatically classifies the nursing level according to the risk assessment results of ophthalmology patients.

2. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 1, characterized in that, The multi-source data acquisition module includes: The image data acquisition unit is used to acquire image data of ophthalmic patients based on color fundus photography, optical coherence tomography images, OCT angiography, visual field examination and anterior segment photography. The clinical data acquisition unit is used to acquire clinical data of ophthalmology patients based on their medical history, diagnosis, surgical records, medication history, lifestyle, and quality of life scales. The monitoring data acquisition unit is used to acquire monitoring data of ophthalmology patients based on their intraocular pressure, visual acuity, refraction, refractive error, corneal curvature, and dynamic changes in visual acuity. Among them, ophthalmology patient imaging data, ophthalmology patient clinical data, and ophthalmology patient monitoring data are used to form multi-source data of ophthalmology patients.

3. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 2, characterized in that, The data processing and fusion module includes: The multi-source data cleaning unit is used to clean multi-source data of ophthalmology patients based on natural language processing, remove noise from the multi-source data of ophthalmology patients, identify missing values ​​and outliers in the multi-source data of ophthalmology patients, and process the identified missing values ​​and outliers. The multi-source data standardization unit is used to process multi-source data of ophthalmology patients based on Z-Score standardization. It converts the multi-source data of ophthalmology patients into a normal distribution with a mean of 0 and a standard deviation of 1, unifies the data format, and performs cropping, enhancement and normalization of images to form standardized multi-source data of ophthalmology patients.

4. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 3, characterized in that, The data processing and fusion module further includes: The data extraction and fusion unit is used to extract features from multi-source data of ophthalmology patients based on principal component analysis. It extracts feature vectors related to the automatic assessment of ophthalmology care level from the multi-source data of ophthalmology patients, and performs weighted fusion of the extracted feature vectors to form fused data of ophthalmology patients. Among them, a pre-trained deep learning model is used to extract image features from multi-source data of ophthalmology patients, automatically extract lesion features, including microaneurysms, hemorrhages and RNFL thickness, encode non-image features from multi-source data of ophthalmology patients, and convert structured data of intraocular pressure, visual acuity, refraction, refractive power and corneal curvature, as well as text data of medical history, diagnosis, surgical records and medication history into numerical vectors.

5. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 4, characterized in that, The intelligent analysis and evaluation module includes: The assessment model building unit is used to build an ophthalmic patient risk assessment model based on artificial intelligence. Based on the need for automatic assessment of ophthalmic care levels, historical data on risk assessment of ophthalmic patients were collected and divided into training and testing sets in a 7:3 ratio. The deep learning model is trained using a training set, enabling it to autonomously learn the risk assessment behavior of ophthalmology patients from the training set and accurately assess the ophthalmology risk status of patients, thus determining the risk assessment model for ophthalmology patients. The ophthalmology patient risk assessment model was tested and evaluated using a test set to determine whether the model could accurately assess the ophthalmological risk of patients and to determine the model test evaluation results. When the ophthalmology patient risk assessment model cannot accurately assess the ophthalmological risk of ophthalmology patients, the parameters of the ophthalmology patient risk assessment model are adjusted and optimized until the ophthalmology patient risk assessment model can accurately assess the ophthalmological risk of ophthalmology patients, and the optimal ophthalmology patient risk assessment model is determined.

6. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 5, characterized in that, A training set is used to train a deep learning model, enabling the model to autonomously learn ophthalmology patient risk assessment behaviors from the training set and accurately assess the ophthalmology patient's ocular risk status. This establishes an ophthalmology patient risk assessment model, including: The training set was divided into four types according to glaucoma, cataract, diabetic retinopathy, and macular degeneration, resulting in four sub-training sets. The intraocular pressure dynamic curve, RNFL thickness, and visual field defect range in the sub-training set of glaucoma are enhanced with feature weights. The rate of optic nerve fiber layer defect progression is used as the core evaluation index to train the deep learning model and obtain the glaucoma sub-model. Correlation analysis was performed on changes in blood glucose control level, glycated hemoglobin, and number of microaneurysms in the sub-training set of diabetic retinopathy. The results of the correlation analysis were used as incremental features, and the rate of new bleeding points was used as the basis for classifying emergency care levels. The deep learning model was then trained to obtain the diabetic retinopathy sub-model. The feature weights of lens opacity, rate of vision loss, and risk of postoperative complications in the sub-training set of cataracts were enhanced, and the deep learning model was trained to obtain the cataract sub-model. The foveal retinal thickness, outer retinal integrity, and macular hemorrhage extent in the sub-training set of macular lesions were enhanced with feature weights, and the deep learning model was trained with the nursing grade mapping accuracy as the core evaluation index to obtain the macular lesion sub-model. By learning the data features of four types of sub-training sets, an automatic disease type identification mechanism is established; By adding an automatic disease type identification mechanism to the sub-models of glaucoma, diabetic retinopathy, cataract, and macular degeneration, a risk assessment model for ophthalmology patients can be established.

7. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 5, characterized in that, The intelligent analysis and evaluation module also includes: The ocular risk assessment unit is used to accurately assess the ocular risk status of ophthalmology patients. The constructed ophthalmology patient risk assessment model is deployed in a real ophthalmology patient risk assessment environment to accurately assess the ophthalmology risk status of ophthalmology patients. The process involves using fusion data from ophthalmology patients as input data and feeding it into an ophthalmology patient risk assessment model. The model analyzes and identifies the fusion data to automatically and accurately assess the ophthalmology patients' eye risks and determine the risk assessment results.

8. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 7, characterized in that, The intelligent analysis and evaluation module also includes: Nursing level classification unit, used to automatically classify nursing levels based on the risk assessment results of ophthalmology patients; Based on a hierarchical rule base embedded with clinical guidelines and expert knowledge, the risk assessment results of ophthalmology patients are mapped to specific nursing levels. Different levels are automatically divided according to the patient's nursing needs, including Level I, Level II, Level III and Level IV. At level I, a routine annual check-up is required; at level II, a specialist follow-up is required every 3-6 months; at level III, emergency treatment is required within 1 month; and at level IV, emergency treatment is required within 48 hours. Among them, based on the risk assessment results of ophthalmology patients, an alarm will be automatically triggered for patients of level IV and the doctor or nursing team will be notified to provide emergency treatment.

9. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 8, characterized in that, The intelligent analysis and evaluation module also includes: The visualization unit is used to present ophthalmic care assessment reports in a visual format; Based on the risk assessment results of ophthalmology patients and the nursing level classified by the patients, an ophthalmology nursing assessment report is generated for each patient and displayed in real time in a visual format, providing data-driven decision support for ophthalmologists and community nurses.

10. The automatic assessment system for ophthalmic care level based on multi-source data fusion according to claim 6, characterized in that, When the ophthalmology patient risk assessment model cannot accurately assess the ocular risk of ophthalmology patients, the parameters of the ophthalmology patient risk assessment model will be adjusted and optimized, including: Obtain the predicted care level of the ophthalmology patient risk assessment model after a preset time; extract the predicted care level that is inconsistent with the actual care level from the predicted care level; and obtain the sample data corresponding to the predicted care level to be corrected as the candidate correction sample. Based on the expert scores and nursing efficacy tests of the alternative corrected samples, the score of each alternative corrected sample was calculated. Select candidate correction samples with scores greater than the preset scores as target correction samples; Based on the dimension normalization values ​​of the target correction sample, the dimension weight of each temperature feature in the target correction sample is calculated. Based on the target modified samples and their corresponding dimensional weights, as well as the corresponding actual care levels, the risk assessment model for ophthalmology patients is trained and optimized a second time.

Images