Internet hospital-based intracranial aneurysm rupture risk assessment and management system

Abstract

The invention relates to the technical field of medical treatment, in particular to an intracranial aneurysm rupture risk assessment and management system based on an internet hospital, which comprises a patient end module, a patient end module, a management end module and a management end module, and the patient end module uploads medical history data and performs automatic filing and labeling processing on basic health data through a system standardization questionnaire and a multi-modal data acquisition mechanism; according to the method, the consistency and comparability of the health data are realized through vectorization coding and standardization processing; a blind deconvolution technology is adopted to improve the definition of a cerebrovascular image, and risk assessment is optimized in combination with the image and clinical features; privacy is protected through a self-adaptive watermarking technology, and data security and compliance are ensured in combination with double encryption and de-identification processing; the system timely identifies risks and pushes personalized health intervention through dynamic monitoring and intelligent early warning, and the accuracy and initiative of chronic disease management are improved. In addition, the block chain technology ensures data traceability and tampering prevention, and the doctor-patient trust is enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the medical technology field, in particular to an intracranial aneurysm rupture risk assessment and management system based on an Internet hospital. BACKGROUND

[0002] With the rapid development of medical imaging technology, early diagnosis and accurate assessment of cerebrovascular diseases increasingly rely on high-quality image data and comprehensive health information. However, cerebrovascular image data processing faces challenges such as image blurring, noise interference, and edge recognition difficulties, and traditional risk assessment methods often rely only on limited clinical data and single imaging features, making it difficult to comprehensively and accurately assess individualized rupture risk. In addition, cerebrovascular diseases have high individual differences, and simply relying on traditional static assessment methods often cannot respond to the dynamic changes in patients' health status in a timely manner, resulting in imperfect individualized treatment and intervention plans.

[0003] The existing patient risk assessment system has many shortcomings and cannot meet the actual needs of data privacy protection, health trend monitoring, and accurate risk prediction in clinical practice. The encryption protection measures of traditional health information systems are relatively single, lacking comprehensive mechanisms such as double encryption, dynamic key generation, de-identification, and blockchain log recording, and there is a risk of data leakage and non-traceability, which does not meet the increasingly stringent data compliance requirements. The identification and processing of sensitive text areas in images are not intelligent enough, making it difficult to achieve automated and accurate covering, increasing the risk of patient privacy leakage. The existing system's processing of time series data such as patient blood pressure, weight, and activity level is rough, lacking smoothing filtering, trend modeling, and abnormal warning mechanisms, making it difficult to detect abnormal fluctuations in health status in a timely manner, affecting the effectiveness of early intervention for chronic diseases. Currently, single models or rule-based scoring systems are used, which do not fully integrate image features, clinical features, and dynamic health indicators, resulting in insufficient prediction accuracy and a lack of dynamic weight adjustment and feature importance enhancement mechanisms, leading to insensitive identification of key high-risk factors and easy misdiagnosis or misdiagnosis of high-risk patients.

[0004] In summary, developing an intracranial aneurysm rupture risk assessment and management system based on an Internet hospital is still a key problem that needs to be solved in the medical field. SUMMARY

[0005] The purpose of the present application is to solve the problem of insufficient data privacy protection, limited data processing capacity, and lack of dynamic monitoring and feedback mechanisms in existing intelligent systems.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions: The application provides an intracranial aneurysm rupture risk assessment and management system based on an Internet hospital, and the system comprises a risk assessment module, which extracts and analyzes patient medical history data, generates a risk level assessment and a visualized assessment report, and comprises the following steps: standardized medical history data and multi-modal medical image data of a patient are acquired, and the multi-modal medical image data is preprocessed, the preprocessing comprising voxel homogenization processing by using a trilinear interpolation method, and smoothing noise and highlighting brain blood vessel edge features by using a Laplacian enhancement algorithm; based on the preprocessed image data, an aneurysm region is extracted and multi-scale semantic segmentation is performed, and a tumor surface complexity index H and a convexity defect index CDI are calculated, the tumor surface complexity H is calculated as follows: , wherein, represents the area of the i-th local surface segment, represents the total surface area of the aneurysm, represents the proportion of the i-th segment area in the total area; the convexity defect index CDI is calculated as follows: , , wherein, represents the three-dimensional convex hull volume of the tumor, represents the true volume of the tumor itself; the tumor surface complexity H and the convexity defect index CDI are formed into image feature encoding, and are fused with clinical feature encoding formed by clinical risk factor encoding to form a unified feature matrix; based on the unified feature matrix, a feature modeling is performed by using an ensemble learning model, wherein the model optimization target is to minimize a weighted loss function, and the importance score of each feature is calculated based on information gain, the feature matrix is weighted to form a standardized input matrix; based on the standardized input matrix, an aneurysm rupture probability is calculated by using a multi-modal neural network, and a final risk value is obtained by combining a risk adjustment term.

[0007] Advantages Compared with the known prior art, the technical scheme provided by the application has the following advantages: ​The application combines imaging features (such as tumor surface complexity H and convexity defect index CDI) and clinical risk factors by a multi-modal data fusion technology, overcomes the limitation of traditional methods relying on a single data source. An integrated learning model is used, and a feature importance weighting mechanism is introduced, so that the influence of key risk factors can be strengthened, thereby a more accurate and personalized risk assessment model is constructed, and the reliability of the rupture risk prediction is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 A flowchart of the intracranial aneurysm rupture risk assessment and management system based on the Internet hospital of the application. DETAILED DESCRIPTION

[0009] In order for those skilled in the art to better understand the application scheme, the technical solutions in the embodiments of the application will be described clearly and completely below in conjunction with the drawings in the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the application.

[0010] It should be noted that the terms "first", "second", etc. in the specification and claims of the application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but includes other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0011] The application will be described in further detail below in conjunction with the drawings: Embodiment: As Figure 1 shown, the application provides an intracranial aneurysm rupture risk assessment and management system based on an Internet hospital, characterized in that it comprises: A patient end module, the patient uploads medical history data through the system standardized questionnaire and multi-modal data acquisition mechanism for automatic archiving and tagging of basic health data; Further, the operation process of the patient end module includes: the system generates a unique ID and establishes a health record, normalizes the data uploaded by the patient through vector coding, wherein, represents an age interval, represents a smoking index of the patient, represents a hypertension weight coefficient, represents a symptom index, recovering a clear image from a blurred image by blind deconvolution image restoration, wherein, represents an observed blurred image, represents a clear image to be restored, represents a blur kernel caused by motion blur, represents a two-dimensional convolution operation, represents the gradient of an image of edge information, represents the overall energy of the kernel, restores the clear details of the blurred MRA image, by estimating the optimal blur kernel, the system can deduce the image as close to the real clear state as possible from the input blurred image, enhance the edge information of the cerebral vascular image, introduce the image gradient information in the deconvolution process, and focus on protecting the details of the vascular profile and small branches, covering sensitive information in the patient's medical image by an adaptive watermarking method, using a PP-OCRv3 model to locate the text area in the image, and judging whether the sensitive information is contained,

[0012] wherein, represents a weight matrix of the same size as the image, used to control the transparency of the watermark, represents the identified privacy area, the weight of the identified area is set to 0.9, and the weight of other areas is 0, keeping the original image unchanged, Based on the above, the synthesis protection image formula after covering sensitive information in the image is:

[0013] wherein, represents the original image, represents a Gaussian noise image used as a watermark background, represents the final output image after protection, represents a weight matrix, represents the weight of a non-private area, Multi-modal data fusion formula: wherein, represents behavior time series data, represents static structure data, represents image data / image features, , , indicates a dynamic weight, indicates a fusion vector; by respectively collecting the dynamic behavior time series data, static clinical structure data (such as medical history, family history, genetic information) and medical image feature data (such as cerebral vascular morphological parameters) of the patient, and combining the dynamically adjusted weight factor, different types of data are given flexible importance scores; finally, a unified fusion vector is generated to comprehensively and accurately represent the current comprehensive health status of the patient, and to provide a reliable basis for subsequent risk assessment and intervention decision, Specifically, in the patient end module, the user completes real-name registration and health information initialization, the system automatically generates a unique ID and establishes a personal health record, the uploaded basic health information of the patient is vectorized and normalized to realize consistent standardization of different feature dimensions, at the same time, the cerebral vascular image data MRA is recovered from the blurred image to restore clear image details through blind deconvolution image restoration technology, the system uses adaptive watermark overlay method to automatically locate the sensitive text area in the medical image, and realizes the covering protection of the privacy information through the setting of the area weight matrix. For multi-source health data, the system realizes multi-modal data fusion based on a dynamic weight mechanism to generate a unified fusion vector.

[0014] The risk assessment module extracts and analyzes the patient's medical history data through an integrated deep learning algorithm to generate a risk level assessment and a visualized assessment report. Further, the operation process of the risk assessment module includes: the system performs preprocessing operation on the image, and performs voxel homogenization processing on the MRA image data in DICOM format by using a trilinear interpolation method to keep the resolution of each axis consistent, and the formula is: , Among them, indicates the gray value of the target point in the resampled image, indicates the interpolation weight, indicates the voxel value of the original image, which is used to eliminate the reading error caused by the difference in original resolution, The image is processed by Laplace enhancement to smooth the noise while retaining the edge of the blood vessel: Among them, indicates a two-dimensional Laplace operator, indicates a Gaussian kernel and the convolution result of the image , indicates an edge enhancement weight parameter, which is used to highlight the edge features of the cerebral blood vessels, the image is preliminarily denoised using a Gaussian kernel, and then the edge part of the image is emphasized through a Laplace operator to retain the detail structure of the cerebral blood vessels and aneurysm, by adjusting the edge enhancement weight parameter, the enhancement effect can be accurately controlled, and the enhanced image can more easily distinguish the small lesion area of the aneurysm, Based on the above highlighted edge features of cerebral blood vessels, the blood vessel structure is segmented and the aneurysm region is extracted, and the processed image is subjected to multi-scale semantic segmentation, wherein, for measuring the coincidence degree of the segmented region and the true label, denotes the binary cross-entropy loss, , denotes the loss function weighting coefficient, The system constructs a shape classification submodule, which calculates the complexity of the tumor surface by shape entropy and convex hull deviation factor: , wherein, denotes the area of the th local surface segment, denotes the total surface area of the aneurysm, denotes the proportion of the area of the th segment to the total area, , wherein, denotes the three-dimensional convex hull volume of the tumor, denotes the true volume of the tumor itself, The greater the volume difference between , the greater the surface depression of the tumor, The value range of , tends to 0, indicating that the tumor is smooth, indicating that the tumor surface has a depression.

[0015] Further, the operation process of the risk assessment module includes: the system uniformly encodes various types of clinical risk factors collected into vector , the aneurysm geometric parameters extracted from the image are encoded into vector , and the two are fused to form a unified feature matrix: Based on the fusion, feature modeling is performed through an ensemble learning method, and the core optimization objective is to minimize the weighted loss function: , wherein, denotes the single-point loss between the predicted value and the true label , denotes the cumulative penalty term for the complexity of all weak learners, Based on the above modeling, the importance score of the feature is calculated by information gain: wherein denotes the a feature, a feature total number of times the feature is used to split nodes during training, loss value before splitting, loss value after splitting, then represents the optimization degree of this split, the feature Each time participating in decision splitting, it brings a certain "model better" (loss reduction), summarize all these improvement effects, and average it, which is its importance score , the higher the score, the more important the feature, The system acts on the original feature matrix to form the standardized input matrix: , wherein represents the Hadamard product, and the processing of high importance risk factors is strengthened through feature weighting processing, and the risk base value is obtained based on the basic model score of the patient features: , wherein represents the score of the classical statistical model, represents the number of features involved in the statistical model, represents the standardized value of the i-th feature variable, represents the weight or score of the feature in the corresponding scoring system, and the high-risk feature is strengthened, the important feature value is amplified, and the irrelevant or low importance feature is assigned a smaller weight, and the model generalization ability is improved, Through the multi-modal neural network, the image feature vector and the clinical feature vector are jointly input according to the fusion model, and the multi-modal neural network calculates the rupture probability: , wherein represents the image feature encoding, represents the clinical feature encoding, represents the multi-modal deep fusion network, represents the Sigmoid activation function, is the final output rupture risk probability, The system introduces a risk adjustment term , and the final prediction risk value is: According to the final risk value, the system divides the risk level: , wherein, represents the final risk value, represents the upper limit of low risk, represents the lower limit of high risk, and the division rule is: then it is judged as low risk, then it is judged as medium risk, then it is judged as high risk; the doctor can adjust the output of the system model to make the system more in line with actual clinical judgment, for example, the doctor may find that a certain case has special characteristics and needs to make appropriate corrections to the risk value, the way of dividing risk levels can make the doctor understand and explain the basis of the system's decision-making, provide more reliable basis for clinical decision-making, through the doctor's fine-tuning and the setting of risk threshold, ensure that the system output is within a safe range and will not make too extreme or inappropriate risk judgments, Specifically, the image data, medical history information and clinical characteristics are comprehensively analyzed to generate personalized risk assessment and visualization report. The system first performs voxel homogenization on the MRA image through trilinear interpolation to improve image quality, and uses Laplace enhancement to highlight the edges of blood vessels, assisting in the extraction of blood vessel and aneurysm regions. The system uses multi-scale semantic segmentation algorithm and morphological analysis to quantify the surface complexity of aneurysm. The system integrates learning to fuse image features and clinical data, optimize feature modeling, calculate risk score and predict rupture risk. The system divides the risk level (low, medium and high risk) according to the predicted value, generates a personalized assessment report to assist the doctor in making accurate decisions and improve the effectiveness and safety of risk management.

[0016] Further, the operation process of the health management module includes: standardize the health data of the patient and generate a feature matrix: wherein, represents the change of the patient's blood pressure curve over time, represents the patient's weight change curve, represents the patient's activity amount, represents the patient's dietary calorie intake record, represents the patient's sleep time and quality indicators, and the data is normalized: wherein represents the mean value of each health data, is the standard deviation, the weighted moving average is used to monitor the patient's health trend: wherein, represents the smoothing factor, represents the real-time detection of health status change trend, when the trend deviates from the set threshold abnormally, the system will trigger personalized health intervention or high-risk warning, Specifically, the system first collects and standardizes the patient's health data, including blood pressure curve, weight change, activity amount, dietary calorie intake, and sleep quality indicators, and generates a unified feature matrix. After normalization processing, the health data is consistent in dimension, facilitating subsequent analysis. The system uses a weighted moving average algorithm to monitor the patient's health status in real time. Through a smoothing factor, the system can detect the trend of the patient's health status and dynamically monitor the preset threshold. When the health trend is abnormal or deviates from the set threshold, the system will automatically trigger personalized health interventions or issue a high-risk warning, ensuring that patients can receive timely and targeted health management and risk warnings, improving the accuracy and proactivity of health management.

[0017] Further, the specific process of the data security module includes: According to the image data and health information uploaded by the patient, double encryption processing is performed: , wherein, is a one-time symmetric encryption key, is the image data of the patient, is the health information of the patient, represents the image and health data content encrypted with The symmetric key encrypted with the server-side public key , , and represent the final storage, and the uploaded health data set is de-normalized. The system generates an unalterable record for data access or operation behavior.

[0018] Specifically, the patient uploads the MRA image and health information through the system, and the system performs double encryption processing on the uploaded data. The system generates a one-time symmetric encryption key for encrypting the patient's image data and health information, and then encrypts the symmetric key using the server-side public key. The final encrypted data and key are securely stored in the server to ensure data security during transmission and storage. The system de-identifies the uploaded health data set, removes sensitive personal information, and generates an anonymous data set to ensure patient privacy is effectively protected and meets relevant data privacy compliance requirements. The system generates an unalterable operation record for each data access or operation behavior, including user unique identifier, operation type, operation time, and user permission set information, further enhancing the data security of the system and the trust between doctors and patients.

[0019] A specific example of feasibility: assume Mr. Wang (45 years old, long-term smoker, with a history of hypertension) as an example, Mr. Wang first completes real-name registration through the patient end module, the system automatically generates a unique ID and initializes his personal health record, and then uploads basic health information (including age, smoking index, blood pressure control level, family history, etc.) and cerebral vascular MRA image data. The system first vectorizes and normalizes the uploaded health data, standardizes each feature dimension, ensures data consistency, and forms a normalized feature vector: For the uploaded MRA image, the system applies blind convolution restoration method to restore clear details: And after voxel homogenization processing (using three linear interpolation: Then perform Laplace enhancement: Highlight the edges of blood vessels and improve the visibility of small blood vessels and aneurysm structures. The image contains sensitive information such as name, hospital number, etc. The system automatically detects and covers sensitive areas through an adaptive watermark covering mechanism, Protecting privacy information from being leaked, after covering sensitive information, the system synthesizes a protected image: The system integrates Mr. Wang's multi-modal health data based on a dynamic weight fusion mechanism: Including static medical history, dynamic behavior data (such as blood pressure fluctuation curve in the past 3 months, exercise amount change) and cerebral vascular image features, forming a unified fusion vector The fusion vector is input into the integrated learning model, the system automatically quantifies key indicators such as aneurysm surface complexity, blood vessel wall irregularity, and local blood flow characteristics, calculates the rupture risk score by combining image features and health data, and divides it into medium and high risk levels according to the set threshold, , Where, represents the final risk value, represents the upper limit of low risk, represents the lower limit of high risk, the division rule is: If it is judged as low risk, If it is judged as medium risk, If it is judged as high risk, the system generates a personalized assessment report; Table 1

[0020] See Table 1, when Mr. Wang is in case A, it belongs to low risk, when Mr. Wang is in case B, it belongs to medium risk, when Mr. Wang belongs to case C, it belongs to high risk, when in case A, the system recommends maintaining a healthy lifestyle, such as controlling diet, quitting smoking and limiting alcohol, regular exercise, when in case B, the system recommends scheduling a review image (MRA, CTA) within 6-12 months, and strengthening blood pressure and blood lipid management, when in case C, the system recommends timely medical treatment, and perfecting high-resolution blood vessel wall imaging to assess whether intervention or surgery is needed, and if necessary, prevent rupture in advance, the system applies weighted moving average smoothing processing to Mr. Wang's continuous health monitoring data (such as blood pressure curve, weight, sleep quality): wherein, represents a smoothing factor, represents real-time detection of health status change trend, when the trend deviates from the set threshold, the system will trigger personalized health intervention or high-risk warning, real-time detection of its health status change trend, once the blood pressure is found to be abnormally high and exceeds the safety threshold, the system will automatically trigger a high-risk warning, and push personalized health guidance, such as drug adjustment suggestion and lifestyle intervention measures, in terms of data security, all data uploaded by Mr. Wang is protected by double encryption strategy, first, the data itself is encrypted using a one-time symmetric key: wherein, is a one-time generated symmetric encryption key, is the image data of the patient, is the health information of the patient, represents the image and health data content encrypted with the server-side public key, represents the symmetric key encrypted with the server-side public key , and represent the final storage, and the server-side public key is encrypted again to ensure the security of the data during transmission and storage, at the same time, each data access or processing behavior is recorded by the system, the uploaded health data set is de-normalized to improve the data trust and privacy compliance of doctors and patients, and the system generates an unalterable record of data access or operation behavior.

[0021] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements will not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. The intracranial aneurysm rupture risk assessment and management system based on the Internet hospital is characterized by: The system includes a risk assessment module: Acquiring standardized medical history data and multimodal medical imaging data of the patient, and preprocessing the multimodal medical imaging data, wherein the preprocessing includes voxel normalization using a trilinear interpolation method and a Laplace enhancement algorithm to smooth noise and highlight cerebral vascular edge features; Based on the preprocessed image data, the aneurysm area is extracted and multi-scale semantic segmentation is performed to calculate the aneurysm surface complexity index H and convexity defect index CDI. The tumor surface complexity H and convexity defect index CDI are used to form an image feature code, which is then fused with the clinical feature code encoded by the clinical risk factor to form a unified feature matrix; Based on the unified feature matrix, feature modeling is performed using an ensemble learning model, wherein the model optimization objective is to minimize a weighted loss function, and the importance score of each feature is calculated based on the information gain, and the feature matrix is ​​weighted to form a standardized input matrix; Based on the standardized input matrix, the probability of aneurysm rupture is calculated by a multimodal neural network, and the final risk value is obtained by combining the risk adjustment item.

2. The intracranial aneurysm rupture risk assessment and management system based on the Internet hospital according to claim 1 is characterized in that: Calculating the probability of rupture The formula is: ,in Represents image feature coding, Indicates clinical feature codes, represents the multimodal deep fusion network, represents the Sigmoid activation function, is the final output rupture risk probability.

3. The intracranial aneurysm rupture risk assessment and management system based on the Internet hospital according to claim 2 is characterized in that: Final predicted risk value The formula is: ； in, is a risk adjustment item.

4. The intracranial aneurysm rupture risk assessment and management system based on the Internet hospital according to claim 3 is characterized in that: According to the final risk value , the system divides risk levels: ， in, represents the final risk value, Indicates the upper limit of low risk, Indicates the lower limit of high risk, division rules: It is judged as low risk. It is judged as medium risk. It is judged as high risk.

5. The intracranial aneurysm rupture risk assessment and management system based on the Internet hospital according to claim 1 is characterized in that: The system also includes a patient-side module, where patients upload their medical history data and automatically archive and label their basic health data through the system's standardized questionnaire and multimodal data collection mechanism: Generate a unique ID and establish a health record, and normalize the data uploaded by the patient through vectorized coding. Recover clear images from blurred images through blind area convolution image restoration, mask sensitive information in patient medical images through adaptive watermark overlay method, and use PP-OCRv3 model to locate text areas in images and determine whether they contain sensitive information. After covering the sensitive information in the image, the protected image is synthesized using the multimodal data fusion formula: ,in, Represents behavioral time series data, Represents static structure data, Represents image data / image features, 、 、 represents the dynamic weight, Represents the fusion vector.

6. The intracranial aneurysm rupture risk assessment and management system based on the Internet hospital according to claim 1 is characterized in that: The system also includes a health management module, which continuously uploads patients' daily health data and pushes health management suggestions and review reminders through the intelligent reminder system: Normalize the patient's health data and generate a feature matrix: ,in, Indicates that the patient's blood pressure curve changes over time. Shows the patient's weight change curve, Indicates the patient's activity level, Indicates the patient's dietary calorie intake record, Represents the patient's sleep time and quality indicators, and normalizes the data: ,in represents the mean of various health data, is the standard deviation, and a weighted moving average is used to monitor patient health trends.

7. The intracranial aneurysm rupture risk assessment and management system based on the Internet hospital according to claim 1 is characterized in that: The system also includes a data security module that uses blockchain and tokenization technology to protect data privacy: Double encryption is performed based on the image data and health information uploaded by the patient: ,in, is a one-time generated symmetric encryption key, For the patient's image data, For patients' health information, Indicates Encrypted image and health data content, Represents the symmetric key encrypted with the server's public key , and Indicates final storage, de-standardizes uploaded health data sets, and the system generates tamper-proof records of data access or operation behaviors.

Images