Disease state early warning system based on mobile nursing

Abstract

The invention discloses an illness state early warning system based on mobile nursing, and relates to the technical field of medical nursing, the illness state early warning system comprises a data acquisition module, a cloud memory, an illness state analysis module and a mobile nursing platform; and the data acquisition module is used for acquiring patient information. According to the illness state early warning system based on mobile nursing, the system obtains comprehensive patient information through the data acquisition module, and the cloud memory is used for caching, so that the integrity and traceability of data are ensured; a disease state analysis module adopts a disease diagnosis model combining a support vector machine and a neural network to compare and analyze vital sign features and image features of the patient with standard data and a typical model in a medical knowledge base, so that subjectivity caused by manual judgment only depending on doctors is avoided, the diagnosis accuracy is improved, and the diagnosis efficiency is improved. Therefore, the problems that disease diagnosis in an existing medical system is high in subjectivity, complex change trends are difficult to find, and timeliness is insufficient are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical care, and in particular to a disease early warning system based on mobile care. Background Art

[0002] In the existing medical system, disease diagnosis mainly relies on doctors to make manual judgments on medical data collected in real time. This diagnostic method has many limitations. It is highly subjective: the personal experience of doctors plays a dominant role in the diagnosis process. Different doctors may have different judgments on the condition of the same patient, and there is a lack of unified and objective diagnostic standards. This makes the diagnosis results affected by the doctor's subjective factors to a certain extent, which may lead to misdiagnosis or missed diagnosis. It is difficult to find complex change trends: when faced with a large amount of medical data, it is difficult for doctors to quickly and accurately find complex change trends between multiple medical data. For example, the patient's vital signs data, such as heart rate, blood pressure, body temperature, and blood oxygen saturation, may contain important information about the development of the disease through subtle changes and their interrelationships over a period of time. However, these complex change patterns can easily be overlooked by manual observation and analysis, thus missing the opportunity for early warning of the disease. It causes a lack of timeliness. The medical data collected in real time needs to be manually analyzed by doctors. This process may take a lot of time, resulting in the inability to respond to changes in the patient's condition in a timely manner. Especially when the patient's condition changes rapidly, untimely diagnosis may delay the best treatment time and pose a serious threat to the patient's health. Summary of the invention

[0003] In order to solve the above technical problems, the present invention is implemented through the following technical solutions: a disease early warning system based on mobile nursing, including a data acquisition module, a cloud storage, a disease analysis module and a mobile nursing platform;

[0004] Data collection module: used to obtain patient information, including patient identity information and physiological sign information;

[0005] Cloud storage: used to receive and cache patient information;

[0006] Condition analysis module: obtains patient information from cloud storage, analyzes the patient's vital signs, determines the patient's physical condition, obtains analysis results, and sends the analysis results to the mobile nursing platform;

[0007] The mobile nursing platform includes: a human-computer interaction unit: used to communicate with the platform through the human-computer interaction interface and perform operations; a display unit: used to display the analysis results of the patient's physical condition; and an early warning unit: used to alarm when the patient's physical condition is abnormal;

[0008] Wherein, the disease analysis module includes an image processing unit, and the image processing unit includes an image smoothing subunit, an image segmentation subunit and a feature extraction subunit;

[0009] The disease analysis module has a built-in medical knowledge base, which includes standard vital sign ranges, standard change patterns and typical image feature models corresponding to various diseases, which are divided into standard data and typical models; by adopting a disease diagnosis model composed of a support vector machine and a neural network, the vital sign features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base.

[0010] Preferably, the data acquisition module includes a patient identity information acquisition module and a physiological sign monitoring module:

[0011] The patient identity information acquisition module establishes a connection with the hospital information system HIS through a network interface protocol to acquire the patient identity information;

[0012] The physiological sign monitoring module is connected to the physiological sign monitoring equipment, which includes an electrocardiogram monitor, a sphygmomanometer, a thermometer, and a pulse oximeter; the physiological sign monitoring equipment establishes a communication link with the physiological sign monitoring module through a wired USB, RS232 or wireless Bluetooth, Wi-Fi communication protocol to obtain physiological sign information;

[0013] The processed physiological sign information is combined with the previously acquired and cached patient identity information to form patient information in the form of a data packet, which is then transmitted to the cloud storage for caching for subsequent use by the condition analysis module.

[0014] Preferably, the condition analysis module establishes a connection with a cloud storage, sends a data acquisition request to the cloud storage according to the patient's patient number or hospitalization number, obtains the patient's vital signs data and image data, the image data is a medical image, or image data related to the condition; receives the patient's vital signs data transmitted from the cloud storage, including real-time monitoring values ​​of heart rate, blood pressure, body temperature, and blood oxygen saturation, as well as historical record data, performs preliminary sorting of the vital signs data, checks the integrity and accuracy of the data, removes data points with obvious errors or omissions, and sorts them in chronological order for subsequent time series analysis.

[0015] Preferably, after the disease analysis module acquires the patient's image data, it transmits it to the image processing unit for image processing; including: using the image smoothing subunit to remove noise interference in the image by using the Gaussian filtering algorithm to smooth the image data, thereby improving the clarity and quality of the image and reducing the impact of noise on subsequent image analysis; and then the image segmentation subunit performs a segmentation operation on the smoothed image; including: using threshold segmentation and region growing methods to divide different tissue or organ regions in the medical image, and separating the target region from the background, so as to more accurately extract feature information, the target region includes the lesion region or a specific anatomical structure region related to vital sign analysis;

[0016] The feature extraction subunit is used to calculate feature parameters based on vital signs data, including: the mean value, variance, maximum value, minimum value and rate of change of heart rate within a specific time period; the mean value and fluctuation range of systolic and diastolic blood pressure; the average body temperature, peak value and duration of body temperature; the average value of blood oxygen saturation and the duration below the safety threshold; at the same time, the autoregressive moving average model ARIMA is used to analyze the change trend of vital signs data over time, determine whether it presents stable, rising, falling or periodic trend characteristics, and obtain vital signs characteristics;

[0017] The feature extraction subunit extracts key features from the image data processed by the image processing unit to obtain image features; and combines the image features with the vital sign features, analyzes the correlation between the two, and establishes a correlation model between the two, so as to more comprehensively evaluate the patient's condition.

[0018] Preferably, the disease analysis module has a built-in medical knowledge base, which includes standard vital sign ranges, standard change patterns and typical image feature models corresponding to various diseases, which are divided into standard data and typical models; by adopting a disease diagnosis model formed by combining a support vector machine with a neural network, the vital sign features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base; the support vector machine is first used to perform a preliminary classification of the vital sign data to screen out suspected patients; then, for suspected patients, the neural network is used to analyze the vital sign time series data and image data to determine the disease type, probability of illness and severity of the suspected patient, and obtain a comparison result; based on the comparison result, combined with clinical experience and medical expert rules, a comparison is performed. Make a judgment; during the judgment, the disease diagnosis results are revised and improved in combination with the patient's age, gender, and previous medical history; generate the final analysis results, which include the patient's physical condition assessment, the name and probability of the possible disease, risk warnings, and recommended measures for the current condition. Among them, the physical condition assessment includes healthy, sub-healthy, sick and the severity of the disease; the risk warning is the risk of complications that may be caused by the disease; the recommended measures include further examination items, treatment suggestions, and nursing precautions; the generated analysis results are encapsulated in JSON data format and sent to the mobile nursing platform through the secure and reliable network communication protocol HTTPS to ensure that the results can be delivered accurately so that medical staff can obtain them in time and take corresponding actions;

[0019] The analysis results as well as the data, algorithm parameters, and model usage involved in the entire analysis process are recorded and stored in detail locally in the disease analysis module to facilitate subsequent result tracing, data analysis optimization, and medical quality monitoring.

[0020] Preferably, the process of smoothing the image data is as follows:

[0021] Determine the filter parameters: Convolution kernel size: select the Gaussian kernel size based on the image resolution and noise level; Standard deviation σ: The standard deviation controls the shape of the Gaussian distribution, which in turn affects the degree of filtering; a smaller standard deviation will make the filtered image closer to the original image, but the noise removal effect may be weaker; a larger standard deviation can better remove noise, but may over-blur image details; generally, the standard deviation needs to be selected between 0.5-3 based on the characteristics of the noise and the image content;

[0022] Construct Gaussian kernel: According to the selected convolution kernel size and standard deviation, the value of each element in the convolution kernel is calculated by Gaussian function;

[0023] Image boundary processing:

[0024] Before filtering the image, when the processing of the image boundary pixels exceeds the image range, the zero-value pixels outside the image boundary are filled by the zero-filling method, and the image boundary pixels are expanded outward in a mirrored manner by the mirroring method;

[0025] Starting from the upper left pixel of the image, the Gaussian kernel is convolved with each pixel of the image and its neighborhood; for each pixel, its neighborhood pixels are multiplied by the corresponding Gaussian kernel weights, and then these products are added to obtain the filtered pixel value.

[0026] Preferably, the process of segmenting the smoothed image is as follows:

[0027] Threshold segmentation:

[0028] Manual threshold selection: If there is a significant difference in grayscale values ​​between the target area and the background area of ​​the image, and this difference is relatively fixed, then manually select a threshold; this threshold is determined by observing the grayscale histogram of the image, which shows the frequency of occurrence of different grayscale values ​​in the image; select a grayscale value so that most of the pixels in the target area have grayscale values ​​below or above this threshold, while the pixels in the background area have grayscale values ​​above or below this threshold;

[0029] Automatic threshold selection method:

[0030] OTSU method: Calculate the grayscale histogram of the image to get the number of pixels at each grayscale level; preset the grayscale range of the image; for each possible threshold, divide the image pixels into two categories; calculate the probability of the two categories of pixels; calculate the average grayscale value of the two categories of pixels;

[0031] Select the threshold that maximizes the inter-class variance as the optimal threshold, and use this threshold to segment the image into the target area and the background area;

[0032] Binarization processing: Binarize each pixel in the image according to the selected manual or automatic threshold; if the gray value of the pixel is less than or equal to the threshold, it is assigned a value of 0, indicating that it belongs to the target area; if the gray value of the pixel is greater than the threshold, it is assigned a value of 1, indicating that it belongs to the background area; in this way, a binary image is obtained, which preliminarily separates the target area from the background.

[0033] Preferably, during the segmentation operation on the smoothed image, the following process is also included:

[0034] Region growing method:

[0035] Seed point selection: seed points can be selected manually or automatically. In manual selection, doctors or professionals select one or more pixels as seed points in the target area based on their own experience and observation of the image. The automatic seed point selection method includes: selecting pixels with significantly different gray values ​​from the surrounding area as seed points based on the gray value statistics of the image; using prior knowledge, including: the approximate location or shape of the known lesion area, and selecting seed points within this range;

[0036] Growth criteria determination: Based on the similarity of pixels, define the criteria for regional growth; similarity is measured by grayscale value difference and texture features; including: setting a grayscale difference threshold T, when the grayscale difference between a pixel and the seed point is less than T, the pixel is considered similar to the seed point and merged into the growth region;

[0037] Region growing process: starting from the seed point, check whether its neighboring pixels meet the growth criteria; if the neighboring pixels meet the criteria, add them to the growth region, and use this pixel as a new seed point, and continue to check its neighboring pixels; repeat this process until no new pixels meet the growth criteria; grow a complete region containing the target area, and separate the target area from the background; through the region growing method, target areas with irregular shapes and uneven grayscale can be handled more flexibly, and combined with the threshold segmentation method, the accuracy of image segmentation can be improved.

[0038] Preferably, the process of generating the final analysis results is:

[0039] Preliminary integration based on comparison results: receiving the results of the support vector machine's preliminary classification of vital sign data and the neural network's analysis of the suspected patient's vital sign time series data and image data; the support vector machine's results are preliminary judgments on the classification of patients into different disease categories or normal ranges, and the neural network's results include the suspected patient's disease type prediction, disease probability estimation, and severity score information; integrating the results and recording them in a structured data format; creating a data object that contains a list of disease types predicted by the neural network, each disease type corresponds to a disease probability value and severity level, and records the key feature values ​​and classification basis in the support vector machine classification process;

[0040] Adjustments based on clinical experience and expert rules: Based on the different susceptibility of different age groups to diseases, adjustments are made based on the patient's age information, including: children are more susceptible to infectious diseases, while the elderly are more susceptible to cardiovascular diseases and chronic degenerative diseases; if the patient is a child and the neural network predicts a risk of cardiovascular disease, but the vital signs in the support vector machine classification are only slightly abnormal, the final diagnostic weight of cardiovascular disease may be reduced, and more attention may be paid to whether there are abnormalities in indicators related to infectious diseases; based on gender differences in diseases, adjustments are made based on the patient's gender factor, among which women are more susceptible to autoimmune diseases, and men are more susceptible to certain urinary system diseases; if the patient is a female patient and the image features show that there are some organs related to the immune system Minor abnormalities, combined with clinical experience, will indicate the necessity of further screening for autoimmune diseases in the final results, even if the neural network does not list it as a high-probability disease; adjustments are made based on the analysis of the patient's past medical history. If the patient has a history of hypertension and the current vital signs data show large fluctuations in blood pressure, even if other disease models have certain indications, the risk of hypertension-related complications will be given priority, and recommendations for blood pressure control and cardiovascular system examinations will be emphasized in the final results; for patients with a history of tumors, if new suspected nodules appear in the image data, tumor recurrence will be highly suspected, and the requirement for tumor reexamination will be highlighted in the final results regardless of the neural network's predicted probability for other diseases; adjustment information is obtained after the adjustment;

[0041] Conduct assessment and generate results: Evaluate the adjustment information to determine the patient's physical condition assessment. If all indicators are within the normal range and there are no obvious abnormalities, it is judged to be healthy; if some indicators are slightly abnormal but do not meet the disease diagnosis criteria, it is judged to be sub-healthy; if there are clear disease characteristics and meet the diagnostic conditions, it is judged to be ill and graded according to the severity of the disease, including: mild, moderate, and severe; generate the final analysis result document, including: the patient's physical condition assessment results, including: healthy, sub-healthy, ill and the severity of the disease; for the list of disease names, sort them from high to low in terms of probability, and attach the probability value of each disease; for the risk warning part, explain the risk of complications that may be caused by each disease; for recommended measures: give further examination items, including: for suspected diseases, list the specific project names and examination purposes of laboratory tests and imaging examinations that need to be performed; for treatment suggestions, provide preliminary treatment plans or treatment directions based on the type and severity of the disease; for nursing precautions, include suggestions on diet, rest, and exercise;

[0042] Preferably, the vital sign features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base, and the process is as follows:

[0043] For the patient's heart rate characteristics, its average value is compared with the heart rate range corresponding to various diseases in the medical knowledge base; similarly, the vital sign characteristics of blood pressure, body temperature, and blood oxygen saturation are compared to determine whether each vital sign exceeds the normal range and the possible corresponding disease range;

[0044] Using the time series data of the patient's vital signs, extract the features related to the change pattern; match the change pattern of the patient's vital signs with the standard change pattern in the medical knowledge base; perform feature vector distance calculation: first select the distance measurement method: for the patient's image feature vector and the typical image feature model, select the distance measurement method of Euclidean distance, Manhattan distance, and cosine similarity to measure the similarity; then compare the features one by one: calculate the distance between the patient's image features and the corresponding typical image feature model respectively; if the distance is small, it means that the patient's image features are more similar to the typical model;

[0045] Using a pre-trained convolutional neural network (CNN) model for disease image recognition, the patient's image data is input into the model for feature matching; based on the feature patterns of different disease images that the model has learned, the probability or score of the image matching different disease categories can be directly output; and the results of multiple image feature comparison methods or multiple different image recognition models are fused; a weighted average or voting mechanism is used, taking into account the feature vector distance calculation results and the probability of the model output, to obtain the final image feature comparison and analysis results, and determine which disease's typical image features the image features are most similar to;

[0046] The comparative analysis results of vital signs features and image features are integrated to generate comparative analysis results, and the results of comparative analysis are recorded; including: the comparison of each vital sign feature with the standard range and change pattern, the matching of each image feature with the typical model, and the disease possibility ranking information after comprehensive consideration of vital signs and image features.

[0047] The present invention provides a disease early warning system based on mobile nursing, which has the following beneficial effects:

[0048] 1. The disease warning system based on mobile care obtains comprehensive patient information, including identity information and various physiological signs information, through the data acquisition module, and uses cloud storage for caching to ensure the integrity and traceability of the data; and adopts a disease diagnosis model that combines support vector machines and neural networks through the disease analysis module to compare and analyze the patient's vital signs and image features with standard data and typical models in the medical knowledge base, avoiding the subjectivity brought by relying solely on the doctor's manual judgment and improving the accuracy of diagnosis; among them, the image processing unit performs smoothing, segmentation and feature extraction operations on the image data, which can more accurately obtain the key information in the image, provide strong support for disease analysis, and further improve the accuracy of diagnosis, thereby solving the problems of strong subjectivity in disease diagnosis, difficulty in discovering complex change trends, and lack of timeliness in the current medical system.

[0049] 2. The disease warning system based on mobile care generates detailed comparative analysis results by comprehensively comparing vital sign characteristics and image characteristics, including the comparison of each vital sign characteristic with the standard range and change pattern, the matching of each image feature with the typical model, and the disease possibility ranking information after comprehensive consideration, etc., providing comprehensive data support for medical staff to help them make more scientific and reasonable medical decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The present invention is a flowchart of a disease early warning system based on mobile nursing. DETAILED DESCRIPTION

[0051] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are provided for the purpose of illustration and description, and are not intended to be exhaustive or to limit the present invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention, and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

[0052] like Figure 1 As shown, the present invention provides a technical solution: a disease early warning system based on mobile nursing, including a data acquisition module, a cloud storage, a disease analysis module and a mobile nursing platform;

[0053] Data collection module: used to obtain patient information, including patient identity information and physiological sign information;

[0054] Cloud storage: used to receive and cache patient information;

[0055] Condition analysis module: obtains patient information from cloud storage, analyzes the patient's vital signs, determines the patient's physical condition, obtains analysis results, and sends the analysis results to the mobile nursing platform;

[0056] The mobile nursing platform includes: a human-computer interaction unit: used to communicate with the platform through the human-computer interaction interface and perform operations; a display unit: used to display the analysis results of the patient's physical condition; and an early warning unit: used to alarm when the patient's physical condition is abnormal;

[0057] Wherein, the disease analysis module includes an image processing unit, and the image processing unit includes an image smoothing subunit, an image segmentation subunit and a feature extraction subunit;

[0058] The disease analysis module has a built-in medical knowledge base, which includes standard vital sign ranges, standard change patterns and typical image feature models corresponding to various diseases, which are divided into standard data and typical models; by adopting a disease diagnosis model composed of a support vector machine and a neural network, the vital sign features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base.

[0059] Preferably, the data acquisition module includes a patient identity information acquisition module and a physiological sign monitoring module:

[0060] The patient identity information acquisition module establishes a connection with the hospital information system HIS through a network interface protocol to acquire the patient identity information;

[0061] The patient identity information acquisition module sends an identity information query request to the hospital information system, and the identity information query request includes a query instruction and an access key. After receiving the identity information query request, the hospital information system retrieves the patient's identity information record in its database according to the instruction in the identity information query request, and accurately locates and extracts it through the structured query language SQL. The retrieved patient identity information, including: name, gender, age, ID number, and hospitalization number, is encapsulated in the JSON data format and transmitted back to the patient identity information acquisition module through the established network connection. After receiving the patient identity information, the patient identity information acquisition module performs data parsing and verification. It checks the integrity, accuracy and format of the data to see if it meets the requirements of subsequent processing. If data errors or omissions are found, it promptly sends error feedback information to the hospital information system and requests resending or supplementing the data. The verified patient identity information is stored in a local temporary cache area, waiting to be transmitted to the cloud storage together with the physiological sign information.

[0062] The physiological sign monitoring module is connected to the physiological sign monitoring equipment, which includes an electrocardiogram monitor, a sphygmomanometer, a thermometer, and a pulse oximeter; the physiological sign monitoring equipment establishes a communication link with the physiological sign monitoring module through a wired USB, RS232 or wireless Bluetooth, Wi-Fi communication protocol to obtain physiological sign information;

[0063] The physiological sign monitoring module sends data collection instructions to the connected physiological sign monitoring device at a preset monitoring frequency: once every 5 minutes. The data collection instructions include the device identification code, measurement duration, and measurement accuracy to ensure that the device can accurately understand and execute the collection task; after receiving the data collection instruction, the physiological sign monitoring device starts the measurement program to collect the patient's corresponding physiological sign information, including: the ECG monitor measures the patient's ECG waveform and calculates the heart rate data, the sphygmomanometer measures the systolic and diastolic blood pressure, the thermometer measures the body temperature, and the pulse oximeter measures the blood oxygen saturation; after the collection is completed, the physiological sign monitoring device transmits the measured data back to the physiological sign monitoring module in the format of a digital signal; after receiving the physiological sign information from each device, the physiological sign monitoring module performs data integration and preliminary processing, including sorting and associating the data of different devices according to a unified timestamp, removing abnormal values, and unifying and converting the data units, including converting the blood pressure data from mmHg to kPa;

[0064] Through the above steps, the data acquisition module can effectively obtain the patient's identity information and physiological sign information, provide basic data support for the operation of the entire mobile care-based disease warning system, and ensure that the system can accurately analyze and warn the patient's condition;

[0065] The processed physiological sign information is combined with the previously acquired and cached patient identity information to form patient information in the form of a data packet, which is then transmitted to the cloud storage for caching for subsequent use by the condition analysis module.

[0066] Preferably, the condition analysis module establishes a connection with a cloud storage, sends a data acquisition request to the cloud storage according to the patient's patient number or hospitalization number, obtains the patient's vital signs data and image data, the image data is a medical image, or image data related to the condition; receives the patient's vital signs data transmitted from the cloud storage, including real-time monitoring values ​​of heart rate, blood pressure, body temperature, and blood oxygen saturation, as well as historical record data, performs preliminary sorting of the vital signs data, checks the integrity and accuracy of the data, removes data points with obvious errors or omissions, and sorts them in chronological order for subsequent time series analysis.

[0067] Preferably, after the disease analysis module acquires the patient's image data, it transmits it to the image processing unit for image processing; including: using the image smoothing subunit to remove noise interference in the image by using the Gaussian filtering algorithm to smooth the image data, thereby improving the clarity and quality of the image and reducing the impact of noise on subsequent image analysis; and then the image segmentation subunit performs a segmentation operation on the smoothed image; including: using threshold segmentation and region growing methods to divide different tissue or organ regions in the medical image, and separating the target region from the background, so as to more accurately extract feature information, the target region includes the lesion region or a specific anatomical structure region related to vital sign analysis;

[0068] The feature extraction subunit is used to calculate feature parameters based on vital signs data, including: the mean value, variance, maximum value, minimum value and rate of change of heart rate within a specific time period; the mean value and fluctuation range of systolic and diastolic blood pressure; the average body temperature, peak value and duration of body temperature; the average value of blood oxygen saturation and the duration below the safety threshold; at the same time, the autoregressive moving average model ARIMA is used to analyze the change trend of vital signs data over time, determine whether it presents stable, rising, falling or periodic trend characteristics, and obtain vital signs characteristics;

[0069] The feature extraction subunit extracts key features from the image data processed by the image processing unit to obtain image features; for example, the size, shape, and texture features of the lesion area are extracted from the medical image: such as grayscale co-occurrence matrix features and density information; and the image features are combined with the vital signs features to analyze the correlation between the two; for example, certain heart diseases may be reflected in specific features of both the vital signs data of the electrocardiogram and the image data of the cardiac ultrasound image, and a correlation model between the two is established to more comprehensively evaluate the patient's condition.

[0070] Preferably, the disease analysis module has a built-in medical knowledge base, which includes standard vital sign ranges, standard change patterns and typical image feature models corresponding to various diseases, which are divided into standard data and typical models; by adopting a disease diagnosis model formed by combining a support vector machine with a neural network, the vital sign features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base; the support vector machine is first used to perform a preliminary classification of the vital sign data to screen out suspected patients; then, for suspected patients, the neural network is used to analyze the vital sign time series data and image data to determine the disease type, probability of illness and severity of the suspected patient, and obtain a comparison result; based on the comparison result, combined with clinical experience and medical expert rules, a comparison is performed. Make a judgment; during the judgment, the disease diagnosis results are revised and improved in combination with the patient's age, gender, and previous medical history; generate the final analysis results, which include the patient's physical condition assessment, the name and probability of the possible disease, risk warnings, and recommended measures for the current condition. Among them, the physical condition assessment includes healthy, sub-healthy, sick and the severity of the disease; the risk warning is the risk of complications that may be caused by the disease; the recommended measures include further examination items, treatment suggestions, and nursing precautions; the generated analysis results are encapsulated in JSON data format and sent to the mobile nursing platform through the secure and reliable network communication protocol HTTPS to ensure that the results can be delivered accurately so that medical staff can obtain them in time and take corresponding actions;

[0071] The analysis results as well as the data, algorithm parameters, and model usage involved in the entire analysis process are recorded and stored in detail locally in the disease analysis module to facilitate subsequent result tracing, data analysis optimization, and medical quality monitoring.

[0072] Preferably, the process of smoothing the image data is as follows:

[0073] Determine the filter parameters: Convolution kernel size: select the Gaussian kernel size based on the image resolution and noise level; Standard deviation σ: The standard deviation controls the shape of the Gaussian distribution, which in turn affects the degree of filtering; a smaller standard deviation will make the filtered image closer to the original image, but the noise removal effect may be weaker; a larger standard deviation can better remove noise, but may over-blur image details; generally, the standard deviation needs to be selected between 0.5-3 based on the characteristics of the noise and the image content;

[0074] Construct Gaussian kernel: According to the selected convolution kernel size and standard deviation, the Gaussian function is used to calculate the value of each element in the convolution kernel. For a 3x3 Gaussian kernel, the center pixel is taken as the origin, and the Gaussian function value corresponding to each position (x, y) is calculated as the weight value in the convolution kernel. The Gaussian function formula is:

[0075] ;

[0076] Where x and y are the coordinates relative to the center of the kernel; the sum of the calculated weight values ​​is 1 to ensure that the brightness of the filtered image remains unchanged;

[0077] Image boundary processing:

[0078] Before filtering the image, when the processing of the image boundary pixels exceeds the image range, the zero-filling method is used to fill zero-value pixels outside the image boundary. In this way, the convolution operation can be performed normally when calculating the filter value of the boundary pixels, but some unnatural effects may be produced at the boundary. At the same time, the image boundary pixels are expanded outward in a mirrored manner through the mirroring method. For example, the leftmost pixel column is copied in a mirrored manner outside the left boundary. This can make the filtering calculation of the boundary pixels more natural and reduce boundary artifacts;

[0079] Starting from the top left pixel of the image, convolve the Gaussian kernel with each pixel of the image and its neighborhood. For each pixel, multiply its neighborhood pixels with the corresponding Gaussian kernel weights, and then add these products to get the filtered pixel value. The neighborhood pixels are determined by the kernel size. For example, for a 3x3 kernel, for pixels in the image , the pixel value after filtering The calculation is as follows:

[0080] ;

[0081] in, is the weight value in the Gaussian kernel, is the neighborhood pixel value;

[0082] The entire image is traversed from left to right and from top to bottom, and such convolution operation is performed on each pixel to complete the smoothing of the entire image.

[0083] Preferably, the process of segmenting the smoothed image is as follows:

[0084] Threshold segmentation:

[0085] Manual threshold selection: If there is a significant difference in grayscale values ​​between the target area and the background area of ​​the image, and this difference is relatively fixed, for example, in some simple black-and-white medical images, the lesion area is obviously blacker or whiter than the surrounding tissue, then a threshold is manually selected; the threshold is determined by observing the grayscale histogram of the image, which shows the frequency of occurrence of different grayscale values ​​in the image; a grayscale value is selected so that most of the pixels in the target area have grayscale values ​​below or above this threshold, while the pixels in the background area have grayscale values ​​above or below this threshold;

[0086] Automatic threshold selection method:

[0087] OTSU method: Calculate the grayscale histogram of the image and get the number of pixels at each grayscale level; the grayscale range of the preset image is ;

[0088] in, is the number of gray levels (usually for 8-bit images ), the total number of pixels is , grayscale The number of pixels is ;

[0089] For each possible threshold , the image pixels are divided into two categories: (Grayscale value is less than or equal to )and (Grayscale value greater than );

[0090] Calculate the probability of these two types of pixels and ;

[0091] in, ;

[0092] ;

[0093] Calculate the average gray value of two types of pixels and ;

[0094] ;

[0095] Calculate the between-class variance:

[0096] ;

[0097] ,in, is the average gray value of the entire image:

[0098] ;

[0099] Choose the between-class variance The maximum threshold is taken as the optimal threshold, and this threshold is used to segment the image into the target area and the background area;

[0100] Binarization processing: Binarize each pixel in the image according to the selected manual or automatic threshold; if the gray value of the pixel is less than or equal to the threshold, it is assigned a value of 0, indicating that it belongs to the target area, such as the lesion area; if the gray value of the pixel is greater than the threshold, it is assigned a value of 1, indicating that it belongs to the background area; in this way, a binary image is obtained, which preliminarily separates the target area from the background.

[0101] Preferably, during the segmentation operation on the smoothed image, the following process is also included:

[0102] Region growing method:

[0103] Seed point selection: seed points can be selected manually or automatically. In manual selection, doctors or professionals select one or more pixels as seed points in the target area, such as the suspected lesion area, based on their own experience and observation of the image. The automatic seed point selection method includes: selecting pixels with significantly different gray values ​​from the surrounding area as seed points based on the gray value statistics of the image; for example, searching for the gray value area corresponding to the peak in the gray histogram, and selecting some pixels from this area as seed points; using prior knowledge, including: selecting seed points within the known approximate location or shape of the lesion area;

[0104] Growth criteria determination: Based on the similarity of pixels, define the criteria for regional growth; similarity is measured by grayscale value difference and texture features; including: setting a grayscale difference threshold T, when the grayscale difference between a pixel and the seed point is less than T, the pixel is considered similar to the seed point and merged into the growth region;

[0105] Region growing process: starting from the seed point, check whether its neighborhood pixels, 4-neighborhood or 8-neighborhood, meet the growth criteria; if the neighborhood pixel meets the criteria, add it to the growth region, and use this pixel as a new seed point, and continue to check its neighborhood pixels; repeat this process until no new pixels meet the growth criteria; in this way, a complete region containing the target area can be gradually grown, separating the target area from the background; through the region growing method, target areas with irregular shapes and uneven grayscale can be handled more flexibly, and combined with the threshold segmentation method, the accuracy of image segmentation can be improved.

[0106] Preferably, the process of generating the final analysis results is:

[0107] Preliminary integration based on comparison results: receiving the results of the support vector machine's preliminary classification of vital sign data and the neural network's analysis of the suspected patient's vital sign time series data and image data; the support vector machine's results are preliminary judgments on the classification of patients into different disease categories or normal ranges, and the neural network's results include the suspected patient's disease type prediction, disease probability estimation, and severity score information; integrating the results and recording them in a structured data format; creating a data object that contains a list of disease types predicted by the neural network, each disease type corresponds to a disease probability value and severity level, and records the key feature values ​​and classification basis in the support vector machine classification process;

[0108] Adjustment based on clinical experience and expert rules: Based on the different susceptibility of different age groups to diseases, adjustments are made based on the patient's age information, including: children are more susceptible to infectious diseases, while the elderly are more susceptible to cardiovascular diseases and chronic degenerative diseases; if the patient is a child and the neural network predicts a risk of cardiovascular disease, but the vital signs in the support vector machine classification are only slightly abnormal, the final diagnostic weight of cardiovascular disease may be reduced, and more attention may be paid to whether there are abnormalities in indicators related to infectious diseases; based on gender differences in diseases, adjustments are made based on the patient's gender factor, such as women are more susceptible to autoimmune diseases, and men are more susceptible to certain urinary system diseases; if the patient is a female patient and the image features show slight abnormalities in some organs related to the immune system Micro-abnormalities, combined with clinical experience, will indicate the necessity of further screening for autoimmune diseases in the final results, even if the neural network does not list it as a high-probability disease; based on the analysis of the patient's past medical history, adjustments will be made. If the patient has a history of hypertension and the current vital signs data show large fluctuations in blood pressure, even if other disease models have certain indications, the risk of hypertension-related complications will be given priority, and the recommendations for blood pressure control and cardiovascular system examinations will be emphasized in the final results; for patients with a history of tumors, if new suspected nodules appear in the image data, tumor recurrence will be highly suspected, and the requirements for tumor reexamination will be highlighted in the final results regardless of the neural network's predicted probability for other diseases; adjustment information is obtained after the adjustment;

[0109] Evaluation and result generation: Evaluate the adjustment information to determine the patient's physical condition assessment. If all indicators are within the normal range and there are no obvious abnormalities, the patient is judged to be healthy; if some indicators are slightly abnormal but do not meet the disease diagnosis criteria, the patient is judged to be sub-healthy; if there are clear disease characteristics and the diagnosis conditions are met, the patient is judged to be ill and graded according to the severity of the disease, including: mild, moderate, and severe; generate the final analysis result document, including: the patient's physical condition assessment results, including: healthy, sub-healthy, ill and the severity of the disease; for the list of disease names, sort them from high to low according to the probability, and attach the probability value of each disease; for the risk warning part, explain the risk of complications that may be caused by each disease, for example, for patients For patients with diabetes, the possible risks of complications such as diabetic nephropathy and diabetic retinopathy are prompted, and the corresponding prevention and monitoring points are explained; for recommended measures: further examination items are given, including: for suspected diseases, the specific names and purposes of laboratory tests (such as blood biochemical tests, immunological tests, etc.) and imaging tests (such as enhanced CT, magnetic resonance imaging, etc.) that need to be performed are listed; for treatment suggestions, preliminary treatment plans or treatment directions are provided according to the type and severity of the disease, such as the selection of drug treatment types and the feasibility assessment of surgical treatment; for nursing precautions, including recommendations on diet, rest, and exercise, for example, for patients with heart disease, it is recommended to eat a low-salt and low-fat diet, avoid strenuous exercise, and ensure adequate sleep.

[0110] Preferably, the vital sign features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base, and the process is as follows:

[0111] For the patient's heart rate characteristics, compare its average value with the heart rate ranges corresponding to various diseases in the medical knowledge base; for example, if the patient's average heart rate is higher than the lower limit of the heart rate corresponding to the tachycardia disease in the knowledge base, then the patient may be at risk of tachycardia; similarly, perform range comparisons on the vital sign characteristics of blood pressure, body temperature, and blood oxygen saturation to determine whether each vital sign exceeds the normal range and the possible corresponding disease range;

[0112] Based on the relationship between vital signs, some diseases may manifest as multiple abnormal vital signs at the same time; for example, septic shock may cause the patient's heart rate to increase, blood pressure to drop, and body temperature to increase abnormally; by establishing a joint judgment rule for multiple vital signs, it is possible to analyze whether the patient's vital sign combination meets the typical manifestations of a specific disease; logical rules or simple statistical models, such as Bayesian networks, can be used to perform this joint analysis;

[0113] Use the time series data of the patient's vital signs to extract features related to the change pattern; for example, calculate the change slope and fluctuation frequency of the heart rate in the past few hours or days; for blood pressure, analyze whether its circadian rhythm changes are normal and whether there are abnormal fluctuation peaks;

[0114] Match the changing pattern of the patient's vital signs with the standard changing pattern in the medical knowledge base; for example, for patients with heart failure, their typical heart rate change pattern may be a sharp increase in heart rate when paroxysmal dyspnea occurs at night, followed by a gradual decrease; if the patient's heart rate change pattern is similar to this typical pattern, the possibility of heart failure increases; the dynamic time warping DTW algorithm can be used to make this comparison;

[0115] Perform feature vector distance calculation: First, select the distance measurement method: For the patient image feature vector and the typical image feature model, select the distance measurement method of Euclidean distance, Manhattan distance, and cosine similarity to measure the similarity; for example, when comparing the shape features of the lesion area, the Euclidean distance can be used to calculate the spatial difference between the patient lesion shape and the typical disease lesion shape; then perform feature comparison one by one: calculate the distance between the patient image features, such as lesion size, texture features and the corresponding typical image feature model; if the distance is small, it means that the patient's image features are more similar to the typical model; for example, for the image of the lung nodule, calculate the distance between the texture features of the patient nodule and the texture feature models of benign nodules and malignant nodules in the knowledge base. If the distance to the texture feature model of the malignant nodule is closer, then the possibility that the nodule is malignant is relatively high;

[0116] Using a pre-trained convolutional neural network (CNN) model for disease image recognition, the patient's image data is input into the model for feature matching; based on the model having learned the feature patterns of different disease images, the model can directly output the probability or score of the image matching different disease categories; for example, a lung X-ray image is input into a trained lung disease recognition CNN model, and the model will output the probability that the image is pneumonia, tuberculosis, and other different diseases; and the results of multiple image feature comparison methods or multiple different image recognition models are fused; a weighted average or voting mechanism is used, taking into account the feature vector distance calculation results and the probability of the model output, to obtain the final image feature comparison and analysis results, and determine which disease's typical image features the image features are most similar to;

[0117] The results of the comparative analysis of vital signs features and the results of the comparative analysis of image features are combined. For example, if the vital signs features show that the patient may have heart disease, and the image features (such as cardiac ultrasound images) also find abnormalities in the structure and function of the heart, then the diagnostic credibility of the heart disease will be greatly increased; generate comparative analysis results and record the results of the comparative analysis; including: the comparison of each vital sign feature with the standard range and change pattern, the matching of each image feature with the typical model, and the ranking information of the disease possibility after comprehensively considering the vital signs and image features; for example, create a table containing fields such as disease name, vital sign matching degree, image feature matching degree, and comprehensive diagnostic possibility to record the results of the comparative analysis and provide detailed data support for subsequent disease diagnosis and decision-making.

[0118] It should be further explained that, in the specific implementation process, comprehensive patient information, including identity information and multiple physiological signs information, is obtained through the data acquisition module, and is cached using cloud storage to ensure data integrity and traceability; and a disease diagnosis model combining support vector machines and neural networks is used in the disease analysis module to compare and analyze the patient's vital sign characteristics and image characteristics with standard data and typical models in the medical knowledge base, thereby avoiding the subjectivity caused by relying solely on the doctor's manual judgment and improving the accuracy of diagnosis; among them, the image processing unit performs smoothing, segmentation and feature extraction operations on the image data, which can more accurately obtain key information in the image, provide strong support for disease analysis, and further improve the accuracy of diagnosis; thereby solving the problems of strong subjectivity in disease diagnosis, difficulty in discovering complex change trends, and lack of timeliness in the current medical system;

[0119] It can collect the patient's physiological data in real time, including heart rate, blood pressure, body temperature, and blood oxygen saturation, and perform time series analysis on it, and can timely capture the changing trends of these data; by matching with the standard change patterns in the medical knowledge base, the system can detect potential changes in the patient's condition earlier, buying precious time for medical staff to take corresponding measures;

[0120] By comprehensively comparing vital sign features and image features, detailed comparative analysis results are generated, including the comparison of each vital sign feature with the standard range and change pattern, the matching of each image feature with the typical model, and the disease possibility ranking information after comprehensive consideration, etc., providing comprehensive data support for medical staff to help them make more scientific and reasonable medical decisions.

[0121] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. A disease early warning system based on mobile nursing, characterized in that: It includes data collection module, cloud storage, condition analysis module and mobile nursing platform; Data collection module: used to obtain patient information, including patient identity information and physiological sign information; Cloud storage: used to receive and cache patient information; Condition analysis module: obtains patient information from cloud storage, analyzes the patient's vital signs, determines the patient's physical condition, obtains analysis results, and sends the analysis results to the mobile nursing platform; The mobile nursing platform includes: a human-computer interaction unit: used to communicate with the platform through the human-computer interaction interface and perform operations; a display unit: used to display the analysis results of the patient's physical condition; and an early warning unit: used to alarm when the patient's physical condition is abnormal; Wherein, the disease analysis module includes an image processing unit, and the image processing unit includes an image smoothing subunit, an image segmentation subunit and a feature extraction subunit; The disease analysis module has a built-in medical knowledge base, which includes standard vital sign ranges, standard change patterns and typical image feature models corresponding to various diseases, which are divided into standard data and typical models; by adopting a disease diagnosis model composed of a support vector machine and a neural network, the vital sign features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base.

2. According to claim 1, a disease early warning system based on mobile nursing is characterized in that: The data acquisition module includes a patient identity information acquisition module and a physiological sign monitoring module: The patient identity information acquisition module establishes a connection with the hospital information system HIS through a network interface protocol to acquire the patient identity information; The physiological sign monitoring module is connected to the physiological sign monitoring equipment, which includes an electrocardiogram monitor, a sphygmomanometer, a thermometer, and a pulse oximeter; The physiological sign monitoring devices establish a communication link with the physiological sign monitoring module through wired USB, RS232 or wireless Bluetooth, Wi-Fi communication protocol to obtain physiological sign information; The physiological sign information is combined with the patient identity information to form patient information in the form of a data packet, which is then transmitted to the cloud storage for caching.

3. The disease early warning system based on mobile nursing according to claim 2 is characterized by: The condition analysis module establishes a connection with the cloud storage, sends a data acquisition request to the cloud storage according to the patient's patient number or hospitalization number, obtains the patient's vital signs data and image data, receives the patient's vital signs data transmitted from the cloud storage, including real-time monitoring values ​​of heart rate, blood pressure, body temperature, and blood oxygen saturation, as well as historical record data, performs preliminary sorting of the vital signs data, checks the integrity and accuracy of the data, eliminates obviously erroneous or missing data points, and sorts them in chronological order.

4. The disease early warning system based on mobile nursing according to claim 3 is characterized by: After the disease analysis module acquires the patient's image data, it transmits it to the image processing unit for image processing; including: removing noise interference in the image by using the Gaussian filtering algorithm through the image smoothing subunit to smooth the image data; then the image segmentation subunit then performs a segmentation operation on the smoothed image; including: dividing different tissue or organ regions in the medical image through threshold segmentation and region growing methods to separate the target region from the background; The feature extraction subunit is used to calculate feature parameters based on vital sign data, including: mean value, variance, maximum value, minimum value and rate of change of heart rate in a specific time period; at the same time, the autoregressive moving average model ARIMA is used to analyze the change trend of vital sign data over time, determine whether it presents stable, rising, falling or periodic trend characteristics, and obtain vital sign characteristics; The feature extraction subunit extracts key features from the image data processed by the image processing unit to obtain image features; combines the image features with the vital signs features to analyze the correlation between the two; and establishes a correlation model between the two.

5. The disease early warning system based on mobile nursing according to claim 4 is characterized in that: The disease analysis module has a built-in medical knowledge base, which includes standard vital sign ranges, standard change patterns and typical image feature models corresponding to various diseases, which are divided into standard data and typical models; by adopting a disease diagnosis model that combines a support vector machine with a neural network, the vital sign features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base; the support vector machine first performs a preliminary classification of the vital sign data to screen out suspected patients; then, for suspected patients, the neural network is used to analyze the vital sign time series data and image data to determine the disease type, probability and severity of the suspected patients, and obtain comparison results; based on the comparison results, judgments are made in combination with clinical experience and medical expert rules; during the judgment period, the disease diagnosis results are corrected and improved in combination with the patient's age, gender, and previous medical history; the final analysis results are generated; the generated analysis results are encapsulated in JSON data format and sent to the mobile nursing platform via the network communication protocol HTTPS.

6. The disease early warning system based on mobile nursing according to claim 5 is characterized by: The process of smoothing image data is as follows: Determine the filter parameters: Convolution kernel size: select the Gaussian kernel size according to the image resolution and noise level; Standard deviation σ: select the standard deviation between 0.5-3 according to the characteristics of the noise and the image content; Construct Gaussian kernel: According to the selected convolution kernel size and standard deviation, the value of each element in the convolution kernel is calculated by Gaussian function; Image boundary processing: Before filtering the image, when the processing of the image boundary pixels exceeds the image range, the zero-value pixels outside the image boundary are filled by the zero-filling method, and the image boundary pixels are expanded outward in a mirrored manner by the mirroring method; Starting from the upper left pixel of the image, the Gaussian kernel is convolved with each pixel of the image and its neighborhood; for each pixel, its neighborhood pixels are multiplied by the corresponding Gaussian kernel weights, and then these products are added to obtain the filtered pixel value.

7. The disease early warning system based on mobile nursing according to claim 6 is characterized by: The process of segmenting the smoothed image is as follows: If there is a significant difference in grayscale values ​​between the target area and the background area of ​​the image, and this difference is relatively fixed, a threshold is manually selected; the threshold is determined by observing the grayscale histogram of the image; a grayscale value is selected so that most of the pixel grayscale values ​​in the target area are lower or higher than this threshold, while the pixel grayscale values ​​in the background area are higher or lower than this threshold; Automatic threshold selection method: OTSU method: calculate the grayscale histogram of the image to obtain the number of pixels at each grayscale level; preset the grayscale range of the image; for each possible threshold, divide the image pixels into two categories; calculate the probability of these two categories of pixels; calculate the average grayscale value of the two categories of pixels, select the threshold that maximizes the inter-class variance as the optimal threshold, and use this threshold to segment the image into the target area and the background area; Binarization: Binarize each pixel in the image according to the selected manual or automatic threshold; if the gray value of the pixel is less than or equal to the threshold, it is assigned a value of 0, indicating that it belongs to the target area; If the grayscale value of a pixel is greater than the threshold, it is assigned a value of 1.

8. The disease early warning system based on mobile nursing according to claim 7 is characterized in that: The segmentation operation of the smoothed image also includes the following processes: Region growing method: Seed point selection: seed points can be selected manually or automatically. When manually selecting, doctors or professionals select one or more pixels as seed points in the target area based on their own experience and observation of the image; The method of automatically selecting seed points includes: selecting pixel points whose grayscale values ​​are significantly different from those of the surrounding areas as seed points based on the grayscale value statistics of the image; utilizing prior knowledge, including: the approximate location or shape of the known lesion area, and selecting seed points within this range; Growth criteria determination: Based on the similarity of pixels, define the criteria for regional growth; similarity is measured by grayscale value difference and texture features; including: setting a grayscale difference threshold T, when the grayscale difference between a pixel and the seed point is less than T, the pixel is considered similar to the seed point and merged into the growth region; Region growing process: starting from the seed point, check whether its neighboring pixels meet the growth criteria; if the neighboring pixels meet the criteria, add them to the growth region, and use this pixel as a new seed point to continue checking its neighboring pixels; repeat this process until no new pixels meet the growth criteria; grow a complete region containing the target area and separate the target area from the background.

9. The disease early warning system based on mobile nursing according to claim 8 is characterized by: The process of generating the final analysis results: Preliminary integration based on comparison results: receiving the results of the preliminary classification of vital signs data by the support vector machine and the analysis of the vital signs time series data and image data of the suspected patients by the neural network; Integrate the results; Adjustment based on clinical experience and expert rules: Based on the different susceptibility of different age groups to the disease, adjustments are made based on the patient's age information; Based on the gender differences of the disease, the patient's gender factor is considered for adjustment; the adjustment is based on the analysis of the patient's past medical history; and the adjustment information is obtained after the adjustment; Conduct assessment and generate results: Evaluate the adjustment information to determine the patient's physical condition assessment. If all indicators are within the normal range and there are no obvious abnormalities, the patient is judged to be healthy; if some indicators are slightly abnormal but do not meet the disease diagnosis criteria, the patient is judged to be sub-healthy; if there are clear disease characteristics and the diagnosis conditions are met, the patient is judged to be ill and graded according to the severity of the disease, including: mild, moderate, and severe; generate the final analysis result document, including: the patient's physical condition assessment results, including: healthy, sub-healthy, ill and the severity of the disease; for the list of disease names, sort them from high to low in terms of probability, and attach the probability value of each disease; for the risk warning part, explain the risk of complications that may be caused by each disease, and for the recommended measures: give further examination items; for treatment suggestions, provide a preliminary treatment plan or treatment direction based on the type and severity of the disease, and for nursing precautions, include suggestions on diet, rest, and exercise.

10. The disease early warning system based on mobile nursing according to claim 9 is characterized in that: The vital signs features and image features extracted from the patient information are compared and analyzed with the standard data and typical models in the medical knowledge base. The process is as follows: for the patient's heart rate feature, its average value is compared with the heart rate range corresponding to various diseases in the medical knowledge base; similarly, the vital signs features of blood pressure, body temperature, and blood oxygen saturation are compared to determine whether each vital sign exceeds the normal range and the possible corresponding disease range; the time series data of the patient's vital signs are used to extract the features related to the change pattern; the change pattern of the patient's vital signs is matched with the standard change pattern in the medical knowledge base; Perform feature vector distance calculation: First, select the distance measurement method: For the patient image feature vector and the typical image feature model, select the distance measurement method of Euclidean distance, Manhattan distance, and cosine similarity to measure the similarity; then perform feature comparison one by one: calculate the distance between the patient image feature and the corresponding typical image feature model respectively; if the distance is small, it means that the patient's image feature is more similar to the typical model; Using a pre-trained convolutional neural network (CNN) model for disease image recognition, the patient's image data is input into the model for feature matching; the probability or score of the image matching different disease categories is output; and the results of multiple image feature comparison methods or multiple different image recognition models are fused; a weighted average or voting mechanism is used, taking into account the feature vector distance calculation results and the probability of the model output, to obtain the final image feature comparison and analysis results, and determine which disease's typical image features the image features are most similar to; The comparative analysis results of vital signs features and image features are integrated to generate comparative analysis results, and the results of comparative analysis are recorded; including: the comparison of each vital sign feature with the standard range and change pattern, the matching of each image feature with the typical model, and the disease possibility ranking information after comprehensive consideration of vital signs and image features.

Images