Heart failure patient health management method and system

Abstract

The invention provides a heart failure patient health management method and system, and the method comprises the steps: obtaining a physiological monitoring data set and a daily activity record data set uploaded by a patient end, carrying out the multi-modal feature fusion processing of the physiological monitoring data set and the daily activity record data set, and generating a comprehensive risk assessment feature set; and inputting the comprehensive risk assessment feature set into a pre-trained capacity overload prediction model, performing weight matching processing on the comprehensive risk assessment feature set through the capacity overload prediction model, and generating a capacity state assessment result and an abnormal early warning level, and generating a personalized intervention strategy set based on the capacity state evaluation result and the abnormal early warning level, feeding back the personalized intervention strategy set to the medical monitoring terminal, performing visual display on the personalized intervention strategy set, and feeding back the personalized intervention strategy set to the patient side for personalized intervention. According to the method, the timeliness of abnormal early warning and the accuracy of an intervention scheme can be improved, and the volume auxiliary management level of the heart failure patient is further improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a health management method and system for heart failure patients. Background Art

[0002] Volume management for heart failure patients is a key component of chronic heart failure treatment. Its core goal is to prevent acute deterioration caused by volume overload by monitoring fluid balance. Existing technologies typically rely on threshold alarms based on physiological indicators such as body weight and heart rate collected by wearable devices, or on rough estimates of dietary excretion recorded manually by patients. These methods, due to their single data dimension and lack of objectivity, struggle to accurately reflect the complex relationship between fluid retention and metabolic imbalance. Furthermore, traditional predictive models are unable to adjust the contribution of key features based on individual patient differences and the stage of the disease, resulting in limited warning sensitivity and specificity. Furthermore, existing systems often only provide isolated risk scores or warning signals, lacking a systematic mechanism to link assessment results with clinical intervention strategies in real time. This makes it difficult for medical staff to formulate personalized treatment plans in a timely manner, exacerbating the risk of delayed volume management. Summary of the Invention

[0003] The present invention provides a health management method and system for patients with heart failure.

[0004] In a first aspect, the present invention provides a method for health management of patients with heart failure, comprising the following steps: Obtaining a physiological monitoring data set and a daily activity record data set uploaded by the patient through the patient terminal. The physiological monitoring data set includes continuously collected body mass data, fluid balance parameters, and circulatory system indicators. The daily activity record data set includes dietary intake image information and fecal feature image information; Performing multimodal feature fusion processing on the physiological monitoring dataset and the daily activity record dataset to generate a comprehensive risk assessment feature set; The comprehensive risk assessment feature set is input into a pre-trained volume overload prediction model, and the comprehensive risk assessment feature set is weighted and matched by the volume overload prediction model to generate a volume status assessment result and an abnormality warning level; the volume status assessment result is used to indicate the degree of deviation between the patient's current fluid balance state and a preset clinical indicator, and the abnormality warning level is used to indicate the urgency of the fluid imbalance event; generating a personalized intervention strategy set based on the capacity status assessment result and the abnormal warning level; Feeding back the personalized intervention strategy set to a medical monitoring terminal triggers the medical monitoring terminal to perform a visual display of the personalized intervention strategy set.

[0005] In a second aspect, the present invention provides a computer system comprising: a memory storing a computer program; A processor is used to load the computer program to implement the health management method for heart failure patients as described above.

[0006] The health management method for heart failure patients provided by the present invention, by integrating physiological monitoring data uploaded by the patient with daily activity image data, constructs a comprehensive risk assessment feature set reflecting the trend of fluid retention, metabolic abnormalities, and intake-excretion imbalance, and uses a weight-matched volume overload prediction model to perform nonlinear weighted analysis on multimodal features to generate quantitative risk indicators and a set of configurable intervention parameters. Ultimately, real-time feedback on the strategy and clinical decision support are achieved through an interactive visualization protocol of the medical monitoring terminal. This method enhances the comprehensiveness of volume assessment through the cross-modal complementarity of heterogeneous data, uses a weight mechanism to adaptively capture key features of disease evolution, and breaks through the limitations of traditional static models in responding to individual differences and acute events. At the same time, the intervention strategy generation logic based on deviation calculation converts complex clinical indicators into executable parameters, opening up a systematic link from data collection to clinical decision-making, significantly improving the timeliness of abnormal warnings and the accuracy of intervention plans, and realizing the objectivity, personalization, and systematization of volume-assisted management of heart failure patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a flow chart of a health management method for heart failure patients provided by an embodiment of the present invention.

[0008] Figure 2 This is an architectural diagram of a health management method for heart failure patients provided by an embodiment of the present invention.

[0009] Figure 3 It is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0011] See also Figure 1 and Figure 2 , Figure 1 This is a flow chart of a method for health management of heart failure patients provided by an embodiment of the present invention. Figure 2This is an architectural diagram of a heart failure patient health management method provided by an embodiment of the present invention; the heart failure patient health management method can be executed by a computer system, and the heart failure patient health management method can include the following steps: Step S100: Acquire the physiological monitoring data set and daily activity record data set uploaded by the patient through the patient terminal. The physiological monitoring data set includes continuously collected body weight data, fluid balance parameters and circulatory system indicators, and the daily activity record data set includes dietary intake image information and excrement characteristic image information.

[0012] In this step, the physiological monitoring dataset is a set of data used to reflect the patient's physiological status. Among them, body mass data is a record of the patient's body weight. Continuously collecting body mass data can intuitively reflect the changes in the patient's body weight over time, such as the trend of weight gain or loss over a period of time. Fluid balance parameters measure the balance of fluid intake and output in the patient's body, such as the difference between fluid intake and output. Circulatory system indicators describe the function of the patient's circulatory system, including heart rate, blood pressure, and blood oxygen saturation. These indicators reflect the working status of the heart and blood vessels. The daily activity record dataset mainly records the patient's daily diet and excretion. Dietary intake image information is an image record of the patient's diet. By analyzing these images, we can understand the type and amount of the patient's diet. Excreta feature image information is an image of the patient's excreta. Analysis of these images can obtain characteristics such as excreta type, color, and texture, thereby inferring the patient's body metabolism and excretion function.

[0013] Patients can upload physiological monitoring data sets and daily activity record data sets through the patient terminal in a variety of ways. For weight data, an electronic scale with data transmission capabilities can be used. Patients measure their weight at a fixed time each day, and the electronic scale uploads the measurement data to the patient's mobile terminal via wireless communication methods such as Bluetooth or Wi-Fi. Fluid balance parameters can be collected by patients recording daily data such as water intake and urine output and manually entering them into the relevant mobile terminal application. Circulatory system indicators can be collected through wearable devices worn by patients. For example, smart bracelets can monitor heart rate and blood oxygen saturation in real time, and sphygmomanometers can measure blood pressure and transmit this data to the mobile terminal. Dietary intake image information can be obtained by patients using their mobile terminal's camera to take photos of each meal and upload them to the application. Excreta characteristic image information can also be obtained by patients taking photos of their excreta using their mobile terminal and uploading them.

[0014] As an implementation method, step S100, obtaining a physiological monitoring data set and a daily activity record data set uploaded by a patient, may specifically include the following steps S110 to S130: Step S110: Continuously collect body weight data and circulatory system indicators through the wearable device worn by the patient, convert the body weight data into daily body weight change rate, and decompose the circulatory system indicators into heart rate variability parameters, blood pressure fluctuation coefficients, and blood oxygen saturation trend values.

[0015] In this step, wearable devices refer to devices that can be worn on the patient's body and used to collect physiological data, such as smart watches, smart bracelets, etc. The continuous collection of body mass data and circulatory system indicators is to understand the patient's physiological status in real time and dynamically. The daily body mass change rate refers to the ratio of the change in body mass per day relative to the body mass of the previous day. It can more accurately reflect the changing trend of the patient's body mass. The heart rate variability parameter refers to the change in the difference between successive heartbeat cycles, which reflects the regulatory function of the autonomic nervous system on the heart. The blood pressure fluctuation coefficient is an indicator to measure the degree of fluctuation of blood pressure over a period of time. It is obtained by calculating statistics such as the standard deviation of blood pressure. The blood oxygen saturation trend value refers to the changing trend of blood oxygen saturation over time. The trend value is obtained by analyzing the blood oxygen saturation data over a period of time, such as fitting a trend line using methods such as linear regression.

[0016] Wearable devices use built-in sensors to collect body mass data and circulatory system indicators. For example, body mass data can be measured using a pressure sensor, heart rate (a circulatory system indicator) can be measured using a photoelectric sensor to detect changes in light absorption by hemoglobin in the blood, blood pressure can be measured using a pressure sensor to detect changes in pressure within the blood vessel wall, and blood oxygen saturation can be measured by detecting the absorption ratio of light of different wavelengths. To convert body mass data into daily body mass change rates, the wearable device or mobile terminal application automatically records daily body mass data and calculates the daily body mass change rate according to the above formula. To decompose circulatory system indicators into heart rate variability parameters, blood pressure fluctuation coefficient, and blood oxygen saturation trend values, heart rate variability parameters can be calculated by performing time domain or frequency domain analysis on heart rate data over a period of time, for example, using metrics such as the standard deviation of the RR interval. The blood pressure fluctuation coefficient can be calculated by calculating the standard deviation of blood pressure measurements over a period of time. The blood oxygen saturation trend value can be linearly fitted to the blood oxygen saturation data over a period of time using the least squares method, with the slope of the trend line being used as the trend value.

[0017] Step S120: The patient uses a mobile terminal to capture dietary intake image information and fecal feature image information, and performs illumination correction and perspective transformation processing on the dietary intake image information to obtain a standardized dietary image; and performs color space conversion and texture enhancement processing on the fecal feature image information to obtain an enhanced fecal image.

[0018] In this step, the mobile terminal usually refers to devices such as smart phones or tablets, and the patient can use its camera to capture dietary intake image information and fecal feature image information. Lighting correction and perspective transformation processing are important steps for preprocessing dietary intake image information, with the aim of eliminating the effects of uneven lighting and tilted shooting angle on image analysis. Lighting correction can make the brightness of the image more uniform, and perspective transformation can convert the tilted image into a top-down plane image under the orthographic projection perspective, so as to more accurately analyze the type and intake of food. Color space conversion and texture enhancement processing are steps for preprocessing fecal feature image information. Color space conversion can convert the image from one color space to another color space that is more suitable for analysis, such as from RGB color space to HSV color space. Texture enhancement processing can highlight the texture features in the fecal image, making the details in the image clearer, which is convenient for subsequent analysis of fecal types and features.

[0019] When patients use the camera of a mobile terminal to capture dietary intake image information and fecal feature image information, they should try to ensure that the images are clear and complete. For the illumination correction and perspective transformation processing of dietary intake image information, illumination correction can use an adaptive brightness compensation algorithm, for example, by calculating the local brightness statistics of the image, adjusting the brightness of different areas to make the overall brightness of the image more uniform. Perspective transformation can first detect the outline of the tableware and the boundary of the food area in the dietary intake image information, determine the tilt angle of the shooting perspective based on the outline of the tableware, and then perform a three-dimensional projection transformation on the dietary intake image information based on the tilt angle to generate a top-down plane image under the orthographic projection perspective. For the color space conversion of fecal feature image information, the color space conversion formula can be used to convert the image from the RGB color space to the HSV color space. Texture enhancement processing can use methods such as histogram equalization to highlight texture features by enhancing the contrast of the image.

[0020] As an embodiment, in step S120, illumination correction and perspective transformation are performed on the dietary intake image information to obtain a standardized dietary image, which may specifically include the following steps S121 to S123: Step S121: Detect the tableware outline and food area boundary in the food intake image information, determine the shooting perspective tilt angle according to the tableware outline, and perform a three-dimensional projection transformation on the food intake image information based on the tilt angle to generate a top-down plane image under the orthographic projection perspective.

[0021] In this step, detecting the cutlery outline and food area boundaries forms the basis for subsequent processing. Algorithms such as image segmentation can be used to separate the cutlery and food from the image and determine their boundaries. The tilt angle of the camera refers to the degree of inclination of the camera relative to the orthographic projection perspective when capturing the image. This tilt angle can be calculated based on the geometric shape and positional information of the cutlery outline. Three-dimensional projection transformation is the process of converting a two-dimensional image shot at an angle into a top-down planar image under the orthographic projection perspective. This transformation eliminates the impact of the tilt of the camera on image analysis, making the shape and size of the food more realistic.

[0022] Edge detection algorithms, such as the Canny edge detection algorithm, can be used to detect the outlines of tableware and the boundaries of food areas in dietary intake images. This algorithm detects edges by calculating the gradient values of pixels in the image. Then, contour extraction algorithms, such as the findContours function in OpenCV, can be used to extract the outlines of the tableware and food. When determining the tilt angle of the shooting perspective based on the outline of the tableware, the tableware can be assumed to be a standard geometric shape, such as a round plate. By calculating the ratio of the major and minor axes of the elliptical shape of the plate in the image, as well as the center position of the ellipse, the tilt angle of the shooting perspective can be calculated using geometric relationships. When performing a three-dimensional projection transformation on the dietary intake image information based on the tilt angle, a perspective transformation matrix can be used. This matrix is calculated using information such as the known tilt angle and image size. The image can then be transformed using the warpPerspective function in OpenCV to generate a top-down planar image from an orthographic perspective.

[0023] Step S122: extracting a pixel distribution histogram of each food area in the top-view plane image, identifying overexposed areas and shadowed areas based on the pixel distribution histogram, and performing illumination equalization processing on the overexposed areas and shadowed areas using an adaptive brightness compensation algorithm.

[0024] In this step, the pixel distribution histogram is a statistical chart that describes the distribution of pixel grayscale values or color values in the image. By analyzing the pixel distribution histogram, the brightness and color characteristics of the image can be understood. Overexposed areas refer to areas in the image where the brightness is too high. The pixel values in these areas are close to or reach the maximum value, resulting in loss of image details. Shadowed areas refer to areas in the image where the brightness is too low. The pixel values in these areas are close to or reach the minimum value, which also affects the observation of image details. The adaptive brightness compensation algorithm is an algorithm that can perform adaptive adjustments based on the brightness of local areas of the image. This algorithm can perform lighting equalization on overexposed areas and shadowed areas to make the brightness of the image more uniform.

[0025] To extract the pixel distribution histogram of each food region in a top-down image, the image can be segmented into distinct food regions and then the pixel distribution histogram for each food region can be calculated. The histogram can be calculated using the calcHist function in OpenCV. To identify overexposed and shadowed areas based on the pixel distribution histogram, a brightness threshold can be set. When the number of pixels with certain grayscale or color values in the histogram exceeds a certain percentage, the area is considered overexposed; when the number of pixels with certain grayscale or color values falls below a certain percentage, the area is considered shadowed. To achieve illumination equalization for overexposed and shadowed areas using an adaptive brightness compensation algorithm, a local histogram equalization algorithm, such as CLAHE (Contrast Limited Adaptive Histogram Equalization), can be used. This algorithm divides the image into multiple small blocks, performs histogram equalization on each block, and then merges the processed blocks into a complete image using methods such as bilinear interpolation, thereby achieving illumination equalization.

[0026] Step S123: The pre-trained food volume estimation model is used to perform stereoscopic reconstruction on the top-view plane image after illumination equalization to generate intake estimation parameters; the stereoscopic reconstruction processing includes estimating the actual volume of the food based on the prior knowledge of the tableware size, and converting it into a liquid equivalent value in combination with the food density parameter.

[0027] In this step, the pre-trained food volume estimation model is trained using a large amount of food image data. This model can estimate the volume of food based on the food features in the image. Stereo reconstruction processing converts two-dimensional image information into three-dimensional food volume information. By combining prior knowledge of tableware size and food density parameters, the actual volume and liquid equivalent value of the food can be more accurately estimated. The intake estimation parameter is used to quantify the total amount of liquid intake of a patient during a single meal. It is obtained by converting the actual volume of the food into a liquid equivalent value.

[0028] Pre-trained food volume estimation models can utilize deep learning models, such as convolutional neural networks (CNNs). During training, the model is trained using a large amount of image data with labeled food volumes. By optimizing the model's parameters, the model accurately predicts the volume of food based on the images. When performing stereo reconstruction on top-view images after illumination equalization, the actual volume of the food is first estimated based on prior knowledge of tableware dimensions, such as the plate's diameter and height, combined with the relative position and size of the food and tableware in the image. The actual volume of the food is then converted to a liquid equivalent value based on the density parameters of the food. For example, for liquid foods, the liquid equivalent value can be directly calculated based on the volume; for solid foods, the corresponding liquid equivalent value can be calculated based on information such as the water content and the density parameter.

[0029] Step S130: align the timestamps and unify the data types of the daily body mass change rate, heart rate variability parameters, blood pressure fluctuation coefficient, blood oxygen saturation trend value, standardized diet images, and enhanced excrement images to generate a physiological monitoring data set and a daily activity record data set.

[0030] In this step, timestamp alignment refers to aligning data collected from different data sources with different timestamps to ensure temporal consistency for subsequent analysis and processing. Data type unification refers to converting data of different data types into a unified data type, such as converting image data into a digital matrix or text data into a numerical code, to facilitate computer processing and analysis. Physiological monitoring datasets and daily activity recording datasets are processed datasets for subsequent analysis.

[0031] When aligning the timestamps of daily body weight change rate, heart rate variability parameters, blood pressure fluctuation coefficients, blood oxygen saturation trend values, standardized dietary images, and enhanced fecal images, the timestamp of one data source can be used as a benchmark to match and align data from other data sources based on the timestamp. For example, using the acquisition time of body weight data as a benchmark, heart rate, blood pressure, and other data can be aligned based on the acquisition time. When unifying data types, image data can be converted into a digital matrix, such as an RGB image into a three-dimensional digital matrix where each element represents the color value of a pixel. Text data can be converted into a numerical code using methods such as one-hot encoding. After timestamp alignment and data type unification, these data are integrated to generate the physiological monitoring dataset and the daily activity recording dataset.

[0032] Step S200: Perform multimodal feature fusion processing on the physiological monitoring dataset and the daily activity record dataset to generate a comprehensive risk assessment feature set.

[0033] In this step, multimodal feature fusion combines features from data derived from different modalities (such as physiological monitoring data and daily activity recording data). This integration provides a more comprehensive understanding of the patient's health status. The comprehensive risk assessment feature set is a set of features used to assess a patient's heart failure volume management risk. It incorporates multiple features and more accurately reflects the patient's risk level.

[0034] When performing multimodal feature fusion processing on physiological monitoring datasets and daily activity recording datasets, various methods can be used, such as fusion methods based on deep learning or fusion methods based on traditional machine learning. Deep learning-based fusion methods can use multimodal convolutional neural networks (MCNNs), which can simultaneously process data from different modalities and fuse features from different modalities through a shared hidden layer. Traditional machine learning-based fusion methods can first extract features from different modal data separately and then concatenate or weightedly combine these features. The resulting comprehensive risk assessment feature set can comprehensively consider multiple factors such as the patient's physiological state, diet, and excretion, providing more comprehensive information for subsequent risk assessment.

[0035] As an embodiment, the comprehensive risk assessment feature set includes fluid retention trend features, metabolic abnormality association features, and intake-excretion imbalance features. Based on this, step S200 performs multimodal feature fusion processing on the physiological monitoring dataset and the daily activity record dataset to generate a comprehensive risk assessment feature set. Specifically, the following steps S210-S270 may be included: Step S210: Perform visual feature analysis on the dietary intake image information to obtain food type recognition results and intake estimation parameters; the food type recognition results are used to distinguish high-sodium food categories, high-water food categories and low-osmotic pressure food categories, and the intake estimation parameters are used to quantify the patient's total liquid intake in a single meal.

[0036] In this step, visual feature parsing analyzes and processes dietary intake image information, extracting visual features to generate food type identification results and intake estimation parameters. Food type identification results determine the food's category. By identifying high-sodium, high-water, and low-osmotic pressure food categories, we can understand the patient's dietary structure and its potential impact on fluid balance. Intake estimation parameters are calculated from information such as the volume and density of food in the image and are used to quantify the patient's total fluid intake during a single meal.

[0037] Deep learning models, such as convolutional neural networks (CNNs), can be used to analyze and process dietary intake image information using visual features. First, the dietary intake image information is input into a pre-trained CNN model, which has been trained on a large number of food images and is able to recognize different types of food. The model's classification layer outputs the food type recognition results, classifying foods into high-sodium food categories, high-water food categories, and low-osmotic pressure food categories. For intake estimation parameters, the total amount of liquid intake for a single meal can be calculated by combining the food volume estimation results and food density parameters obtained in the previous steps. For example, for high-water foods, the liquid intake can be calculated based on their volume and water content; for other foods, the corresponding liquid equivalent value can be estimated based on their component analysis and density information.

[0038] Step S220: Perform morphological analysis on the excretion feature image information to obtain excretion type classification results and excretion volume estimation parameters; the excretion type classification results are used to distinguish urine excretion categories, sweat excretion categories, and fecal excretion categories, and the excretion volume estimation parameters are used to quantify the total amount of liquid excreted by the patient in a single excretion.

[0039] In this step, morphological analysis and processing analyze and process the characteristic image information of the excreta, extracting its morphological features to obtain the excreta type classification results and excretion volume estimation parameters. The excreta type classification result determines the category of the excreta. By distinguishing between urine excretion, sweat excretion, and fecal excretion, we can understand the patient's excretion pattern and body metabolic status. The excretion volume estimation parameters are calculated by calculating information such as the volume and density of the excreta in the image and are used to quantify the total amount of fluid excreted by the patient in a single excretion.

[0040] Image processing and machine learning algorithms can be used when performing morphological analysis on the characteristic image information of excrement. First, the characteristic image information of excrement is preprocessed, such as grayscale and binarization, and then the contour of the excrement is extracted using a contour extraction algorithm. Based on the contour shape, color and other characteristics of the excrement, the excrement is classified using a classification algorithm to obtain the excrement type classification result. For the excrement volume estimation parameter, the volume of the excrement can be estimated by combining the known container size or reference object information, and then the total amount of liquid discharged can be calculated based on the density parameters of different types of excrement. For example, for urine, the volume can be calculated based on the scale of the container or the height of the urine in the image and the cross-sectional area of the container, and then the liquid discharge volume can be estimated based on the density of the urine.

[0041] Step S230: Time-series alignment processing is performed on the body weight data, fluid balance parameters, and circulatory system indicators to generate a physiological state time series feature sequence; the physiological state time series feature sequence is used to characterize the patient's body weight fluctuation trend, fluid retention rate, and circulatory load change cycle.

[0042] In this step, time series alignment aligns weight data, fluid balance parameters, and circulatory system indicators in chronological order, ensuring temporal consistency and enabling better analysis of changes in a patient's physiological state over time. The physiological state time series feature sequence is a chronologically ordered set of feature sequences that contains information such as the patient's weight fluctuation trends, fluid retention rates, and circulatory load change cycles, providing a more intuitive reflection of the dynamic changes in a patient's physiological state.

[0043] When performing time series alignment on body weight data, fluid balance parameters, and circulatory system indicators, the timestamp of a particular data point can be used as a benchmark to match and align other data points based on the timestamp. For example, the acquisition time of the body weight data can be used as a benchmark to align fluid balance parameters and circulatory system indicators based on the acquisition time. When generating a physiological state time series feature sequence, the aligned data can be further processed and analyzed. To analyze body weight fluctuation trends, the rate of change of body weight can be calculated or a time-dependent curve of body weight can be fitted. To analyze fluid retention rate, the amount of fluid retained over a period of time can be calculated based on fluid balance parameters, and its rate of change can also be calculated. To analyze the periodicity of circulatory load changes, the periodicity of circulatory system indicators (such as heart rate and blood pressure) can be analyzed and their periodicity determined through methods such as spectral analysis.

[0044] Step S240: Perform cross-modal association analysis on the food type recognition results, intake estimation parameters, excretion type classification results, excretion estimation parameters, and physiological state time series feature sequences to generate fluid retention trend features, metabolic abnormality association features, and intake-excretion imbalance features. The cross-modal association analysis includes: In this step, cross-modal association analysis is the process of performing association analysis on features in data from different modalities (such as dietary intake, excretion, and physiological status data). By analyzing the relationships between these features, potential risk factors and health issues can be discovered. The fluid retention trend feature reflects the changes in fluid retention in the patient's body over time. It is closely related to factors such as food intake, excretion, and physiological status. The metabolic abnormality association feature reflects whether the patient's metabolic function is abnormal. It is determined by analyzing the relationship between factors such as food type and excretion and physiological status. The intake-excretion imbalance feature reflects the balance between the patient's dietary intake and excretion.

[0045] When performing cross-modal association analysis on food type identification results, intake estimation parameters, excretion type classification results, excretion estimation parameters, and physiological state time series feature sequences, association rule mining or machine learning algorithms can be used. For example, the Apriori algorithm can be used to mine association rules between different features to identify potential relationships between food type, intake, excretion, and physiological state. By analyzing these association rules, fluid retention trend features, metabolic abnormality association features, and intake-excretion imbalance features can be generated. For example, if a strong association is found between high-sodium food intake and increased body weight and accelerated fluid retention, this can be included as part of the fluid retention trend feature.

[0046] Step S250: Determine the positive impact weight of the high sodium intake event on the fluid retention rate based on the first correlation coefficient between the food type recognition result and the fluid balance parameter.

[0047] In this step, the first correlation coefficient is a statistical indicator that measures the correlation between food type identification results and fluid balance parameters. By calculating this coefficient, we can understand the degree to which different food types affect fluid balance. A high sodium intake event refers to a patient's consumption of a high-sodium food. The positive impact weight is a quantitative indicator of the impact of a high sodium intake event on the fluid retention rate. By determining this weight, we can more accurately assess the impact of high sodium intake on a patient's fluid balance.

[0048] When calculating the first correlation coefficient between food type identification results and fluid balance parameters, statistical methods such as the Pearson correlation coefficient can be used. First, the food type identification results are coded, for example, high-sodium food categories are coded as 1 and other food categories are coded as 0. Then, a correlation analysis is performed between the coded food type identification results and the fluid balance parameters (such as fluid retention) to calculate the first correlation coefficient. When determining the weight of the positive impact of high sodium intake events on fluid retention rate based on the first correlation coefficient, the correlation coefficient can be adjusted based on its magnitude. For example, if the correlation coefficient is positive and large, indicating a strong positive correlation between high sodium intake and fluid retention rate, the positive impact weight can be set to a larger value; if the correlation coefficient is small, the positive impact weight can be set to a smaller value.

[0049] Step S260: Determine a reverse adjustment factor of the abnormal excretion volume event on the circulatory load change based on the second correlation coefficient between the excretion type classification result and the circulatory system indicator.

[0050] In this step, the second correlation coefficient is a statistical indicator that measures the correlation between the excreta type classification results and circulatory system indicators. By calculating this coefficient, we can understand the degree of impact of different excreta types on the circulatory system. Abnormal excretion events refer to abnormalities in a patient's excretion volume. The reverse regulation factor is a quantitative indicator of the regulatory effect of abnormal excretion events on changes in circulatory load. Determining this factor allows for a more accurate assessment of the impact of excretion on a patient's circulatory system.

[0051] When calculating the second correlation coefficient between the excretion type classification results and circulatory system indicators, statistical methods such as the Pearson correlation coefficient can also be used. The excretion type classification results are encoded, for example, the urine excretion category is encoded as 1 and other excretion categories are encoded as 0. Then, the circulatory system indicators (such as heart rate, blood pressure, etc.) are correlated with the encoded excretion type classification results to calculate the second correlation coefficient. When determining the reverse adjustment factor of abnormal excretion events on circulatory load changes based on the second correlation coefficient, adjustments can be made based on the size and positive and negative signs of the correlation coefficient. If the correlation coefficient is negative and large, it means that there is a strong negative correlation between abnormal excretion and increased circulatory load, and the reverse adjustment factor can be set to a larger value; if the correlation coefficient is small, the reverse adjustment factor is set to a smaller value.

[0052] Step S270: Based on the positive impact weight and the negative adjustment factor, compensation calculation is performed on the intake estimation parameter and the excretion estimation parameter to generate an intake-excretion imbalance feature.

[0053] In this step, compensation calculation adjusts the estimated intake and excretion parameters based on the positive influence weights and negative adjustment factors. This compensation calculation more accurately reflects the patient's intake-excretion balance. The intake-excretion imbalance characteristic reflects the balance between a patient's dietary intake and excretion. It comprehensively considers food intake, excretion, and the effects of high sodium intake and abnormal excretion on fluid balance.

[0054] Based on the positive influence weight and the negative adjustment factor, a weighted summation method can be used to compensate for the intake estimation parameters and the excretion estimation parameters. First, multiply the intake estimation parameter by the positive influence weight and the excretion estimation parameter by the negative adjustment factor. Then, subtract the two to obtain the compensated difference, which is the intake-excretion imbalance characteristic. For example, if the positive influence weight is 0.8, the intake estimation parameter is 500ml, the negative adjustment factor is 0.6, and the excretion estimation parameter is 300ml, then the intake-excretion imbalance characteristic is 500*0.8-300*0.6=400-180=220ml.

[0055] Step S300: Input the comprehensive risk assessment feature set into the pre-trained volume overload prediction model, perform weight matching processing on the comprehensive risk assessment feature set through the volume overload prediction model, and generate a volume status assessment result and an abnormal warning level; the volume status assessment result is used to indicate the degree of deviation between the patient's current fluid balance status and the preset clinical indicators, and the abnormal warning level is used to characterize the urgency of the fluid imbalance event.

[0056] In this step, the pre-trained volume overload prediction model is a model obtained by training with a large amount of clinical data. The model can predict the patient's volume status and abnormal warning level based on the input comprehensive risk assessment feature set. Weight matching processing means that the model assigns corresponding weights to different features according to their importance, and performs weighted summation and other processing on the features to obtain the final prediction result. The volume status assessment result is an assessment of the patient's current fluid balance status. By comparing with preset clinical indicators, the degree of deviation of the patient's fluid balance status can be understood. The abnormal warning level is divided according to the patient's risk level, which is used to prompt medical staff to take appropriate intervention measures.

[0057] When the comprehensive risk assessment feature set is input into the pre-trained volume overload prediction model, the model will perform weight matching on the comprehensive risk assessment feature set according to its internal weight calculation rules. For example, the model assigns different weights to the fluid retention trend feature, metabolic abnormality association feature, and intake-excretion imbalance feature based on the relevance and importance of the features. By performing weighted summation and other calculations on these features, a comprehensive risk score is obtained. Then, based on the preset risk level thresholds, the comprehensive risk score is compared with these thresholds to determine the abnormal warning level. For example, if the comprehensive risk score exceeds the high-risk critical value, the abnormal warning level is set to high risk. The volume status assessment result can be generated based on the deviation between the comprehensive risk score and the nearest risk level threshold. For example, the percentage of deviation can be calculated to indicate the degree of deviation of the patient's current fluid balance status from the preset clinical indicators.

[0058] As an embodiment, step S300 inputs the comprehensive risk assessment feature set into a pre-trained capacity overload prediction model, performs weight matching processing on the comprehensive risk assessment feature set using the capacity overload prediction model, and generates a capacity status assessment result and an abnormal warning level. Specifically, the following steps S310 to S350 may be included: Step S310: Configure a feature weight allocation module and a risk threshold determination module in the volume overload prediction model; the feature weight allocation module is used to adjust the contribution weight of each feature in the volume status assessment based on the real-time change amplitude of the fluid retention trend feature, metabolic abnormality association feature, and intake-excretion imbalance feature; the risk threshold determination module is used to determine the matching degree of the current comprehensive risk assessment feature set and each warning level based on the fluid imbalance critical value corresponding to different warning levels in historical clinical data.

[0059] In this step, the feature weight allocation module is an important module in the volume overload prediction model. It can dynamically adjust the contribution weight of each feature in the volume status assessment according to the real-time changes in the features. For example, if the change amplitude of the fluid retention trend feature is large, it means that the feature has a greater impact on the patient's volume status, and its weight in the volume status assessment will be increased. The risk threshold determination module is a module that determines the matching degree of the current comprehensive risk assessment feature set and each warning level based on the fluid imbalance critical value corresponding to different warning levels in historical clinical data. This module can more accurately judge the patient's risk level.

[0060] When configuring the feature weight assignment module and the risk threshold determination module in the volume overload prediction model, software programming can be used for implementation. The feature weight assignment module can use an adaptive weight adjustment algorithm, such as a gradient descent-based algorithm, to dynamically adjust the weight according to the real-time change amplitude of the feature. The risk threshold determination module can store the critical values of fluid imbalance in historical clinical data in a database. When a comprehensive risk assessment feature set is input, it is compared with these critical values to determine the degree of match. For example, the comprehensive risk score calculated from the comprehensive risk assessment feature set is compared with the critical values corresponding to different warning levels. If the score exceeds the high-risk critical value, it is considered to match the high-risk level.

[0061] Step S320: The feature weight assignment module assigns a first weight to the fluid retention trend feature, a second weight to the metabolic abnormality association feature, and a third weight to the intake-excretion imbalance feature; the sum of the first weight, the second weight, and the third weight is a preset normalized value.

[0062] In this step, the first, second, and third weights are the weights assigned by the feature weight assignment module to the fluid retention trend feature, metabolic abnormality association feature, and intake-excretion imbalance feature, respectively. The preset normalization value is to ensure that the sum of the feature weights is a fixed value, typically set to 1, to make the weights of each feature comparable.

[0063] When assigning weights to each feature through the feature weight assignment module, the module comprehensively considers factors such as the real-time changes in the fluid retention trend feature, the metabolic abnormality-related feature, and the intake-excretion imbalance feature. For example, if the change in the fluid retention trend feature is large and has a greater impact on the patient's volume status, a larger first weight is assigned; if the change in the metabolic abnormality-related feature is relatively small, a smaller second weight is assigned. At the same time, it is necessary to ensure that the sum of the first, second, and third weights is a preset normalized value. A normalization algorithm can be used, such as calculating the original weights of each feature and dividing them by the total weight to obtain the normalized weight.

[0064] Step S330: Based on the first weight, the second weight, and the third weight, weighted fusion processing is performed on the fluid retention trend feature, the metabolic abnormality association feature, and the intake-excretion imbalance feature to generate a comprehensive risk score.

[0065] In this step, weighted fusion processing is the process of weighted summing the fluid retention trend characteristics, metabolic abnormality association characteristics, and intake-excretion imbalance characteristics according to their respective weights. Through weighted fusion processing, multiple characteristics can be combined into a comprehensive risk score, which can more comprehensively reflect the patient's risk level.

[0066] When weighted fusion processing is performed on the fluid retention trend feature, metabolic abnormality association feature, and intake-excretion imbalance feature based on the first, second, and third weights, the following formula can be used: Comprehensive Risk Score = Fluid Retention Trend Feature * First Weight + Metabolic Abnormality Association Feature * Second Weight + Intake-Excretion Imbalance Feature * Third Weight. For example, if the value of the fluid retention trend feature is 0.8, the first weight is 0.5, the value of the metabolic abnormality association feature is 0.6, the second weight is 0.3, the value of the intake-excretion imbalance feature is 0.7, and the third weight is 0.2, then the comprehensive risk score = 0.8 * 0.5 + 0.6 * 0.3 + 0.7 * 0.2 = 0.4 + 0.18 + 0.14 = 0.72.

[0067] Step S340: Input the comprehensive risk score into the risk threshold determination module, and determine the abnormal warning level by comparing the comprehensive risk score with multiple preset risk level thresholds; wherein the multiple risk level thresholds include low risk critical value, medium risk critical value and high risk critical value.

[0068] In this step, the risk threshold determination module determines the patient's abnormality warning level based on multiple preset risk thresholds. Low, medium, and high risk thresholds represent the boundaries of different risk levels, determined based on historical clinical data and medical research. By comparing the comprehensive risk score with these thresholds, the patient's abnormality warning level can be determined.

[0069] When the comprehensive risk score is input into the risk threshold determination module, the module compares the comprehensive risk score with the low risk threshold, medium risk threshold, and high risk threshold. If the comprehensive risk score is lower than the low risk threshold, the abnormal warning level is set to low risk; if the comprehensive risk score is between the low risk threshold and the medium risk threshold, the abnormal warning level is set to medium risk; if the comprehensive risk score is higher than the medium risk threshold and lower than the high risk threshold, the abnormal warning level is set to high risk; if the comprehensive risk score is higher than the high risk threshold, the abnormal warning level is set to very high risk.

[0070] Step S350: Generate a volume status assessment result based on the deviation between the comprehensive risk score and the nearest risk level threshold; the volume status assessment result includes the probability of fluid overload, the compensatory regulatory capacity index, and the acute event risk coefficient.

[0071] In this step, the deviation refers to the difference between the comprehensive risk score and the nearest risk level threshold. Calculating the deviation provides an understanding of how close the patient's risk level is to their current risk level. The volume status assessment is a comprehensive evaluation of the patient's current fluid balance, including indicators such as the probability of fluid overload, compensatory regulatory capacity index, and acute event risk factor. These indicators provide a more detailed picture of the patient's health status.

[0072] When generating volume status assessment results based on the deviation between the comprehensive risk score and the nearest risk level threshold, a pre-established mapping relationship or model can be used. For example, a relationship model can be established using regression analysis or other methods to establish the relationship between the deviation and the probability of fluid overload, the compensatory regulatory capacity index, and the risk factor for acute events. Based on the magnitude of the deviation, the corresponding probability of fluid overload, compensatory regulatory capacity index, and risk factor for acute events are calculated using this model. A larger deviation indicates a higher risk level for the patient, resulting in a higher probability of fluid overload and risk factor for acute events, and a lower compensatory regulatory capacity index.

[0073] Step S400: Generate a personalized intervention strategy set based on the capacity status assessment result and the abnormal warning level.

[0074] In this step, the personalized intervention strategy set is a collection of intervention measures tailored to the patient's specific circumstances (such as volume status assessment results and abnormality warning level). These measures enable more targeted treatment and management. By combining volume status assessment results and abnormality warning level, different intervention strategies can be developed for patients with different risk levels.

[0075] When generating a personalized set of intervention strategies based on volume status assessment results and abnormality warning levels, a rule engine or machine learning algorithm can be used. The rule engine can generate corresponding intervention strategies based on the volume status assessment results and abnormality warning levels according to preset rules. For example, if the abnormality warning level is high risk and the probability of fluid overload is high, the rule engine can generate intervention strategies such as increasing the dose of diuretics and strictly limiting fluid intake. Machine learning algorithms can establish a mapping relationship between volume status assessment results, abnormality warning levels, and intervention strategies by learning from a large amount of clinical data, and predict appropriate intervention strategies based on the input volume status assessment results and abnormality warning levels.

[0076] As an embodiment, the personalized intervention strategy set includes a medication adjustment plan, a dietary intervention plan, and a monitoring frequency optimization plan. Based on this, step S400 generates a personalized intervention strategy set based on the volume status assessment results and the abnormal warning level, which may specifically include the following steps S410-S440: Step S410: Filtering basic intervention templates from a preset strategy library according to the abnormal warning level; the preset strategy library stores medication adjustment rules, dietary restriction rules, and monitoring frequency rules corresponding to different warning levels.

[0077] In this step, the preset policy library is a database that stores intervention rules corresponding to different warning levels, including rules for medication adjustment, dietary restrictions, and monitoring frequency. The basic intervention template is a set of intervention rules selected from the preset policy library that corresponds to the current abnormal warning level. It provides the foundation for the subsequent generation of specific intervention plans.

[0078] When selecting basic intervention templates from the preset policy library based on the abnormality warning level, database query statements can be used. For example, if the abnormality warning level is high risk, the preset policy library is searched for medication adjustment rules, dietary restriction rules, and monitoring frequency rules corresponding to the high-risk warning level, and these rules are combined into a basic intervention template. The preset policy library can be stored in a relational or non-relational database, such as MySQL or MongoDB.

[0079] Step S420: Adjust the diuretic dosage in the basic intervention template according to the probability of fluid overload to generate a drug adjustment plan; the drug adjustment plan includes a dose escalation gradient, a dosing time interval, and a drug incompatibility prompt.

[0080] In this step, the probability of fluid overload is a key indicator in the volume status assessment, reflecting the likelihood of fluid overload in the patient. Adjusting the diuretic dose in the basic intervention template based on the probability of fluid overload can more accurately control the patient's fluid balance. The medication adjustment plan, based on the basic intervention template and adjusted according to the probability of fluid overload, generates a specific medication regimen. It includes information such as the dose escalation gradient, dosing interval, and drug incompatibilities, guiding healthcare professionals in the proper use of medications. When adjusting the diuretic dose in the basic intervention template based on the probability of fluid overload, a dose adjustment model can be established. This model can adjust the diuretic dose according to specific rules based on the probability of fluid overload. For example, if the probability of fluid overload is high, the diuretic dose is appropriately increased; if the probability of fluid overload is low, the diuretic dose is maintained or reduced. The dose escalation gradient refers to the magnitude of each diuretic dose increase, the dosing interval refers to the time interval between doses, and the drug incompatibilities remind healthcare professionals of drug interactions and other issues that require attention when using medications. Medication adjustment plans incorporating this information were generated by adjusting the basic intervention template.

[0081] Step S430: Optimize the upper limit of daily fluid intake in the basic intervention template according to the compensatory regulatory ability index and generate a dietary intervention plan; the dietary intervention plan includes the intake ratio of different food categories, sodium salt restriction standards and water supplementation recommendations.

[0082] In this step, the compensatory regulatory capacity index is an indicator in the volume status assessment results, which reflects the patient's body's ability to compensate for changes in body fluids. By optimizing the upper limit of daily fluid intake in the basic intervention template according to the compensatory regulatory capacity index, fluid intake can be reasonably controlled according to the patient's actual situation. The dietary intervention plan is a specific dietary plan obtained after optimization based on the compensatory regulatory capacity index on the basis of the basic intervention template. It contains information such as the intake ratio of different food categories, sodium salt restriction standards, and water supplementation recommendations, which can guide patients to eat reasonably.

[0083] When optimizing the upper limit of daily fluid intake in the basic intervention template according to the compensatory regulatory ability index, an upper limit intake optimization model can be established. This model can adjust the upper limit of daily fluid intake according to certain rules based on the size of the compensatory regulatory ability index. For example, if the compensatory regulatory ability index is low, the upper limit of daily fluid intake should be appropriately lowered; if the compensatory regulatory ability index is high, the upper limit of daily fluid intake can be appropriately increased. The intake ratio of different food categories refers to the proportion of different categories of food such as high-sodium food, high-water food, and low-osmotic pressure food in the diet. The sodium salt restriction standard is a restriction on the patient's daily sodium salt intake, and the water supplementation recommendation is a reasonable water supplementation recommendation based on the patient's condition. By optimizing the basic intervention template, a dietary intervention plan containing this information is generated.

[0084] Step S440: Adjust the frequency of physiological parameter collection in the basic intervention template according to the risk factor of acute events, and generate a monitoring frequency optimization plan; the monitoring frequency optimization plan includes the body weight monitoring cycle, blood biochemical detection time window and symptom self-assessment trigger conditions.

[0085] In this step, the acute event risk factor is an indicator in the volume status assessment results, which reflects the patient's likelihood of experiencing an acute event (such as an acute heart failure attack). Adjusting the physiological parameter collection frequency in the basic intervention template based on the acute event risk factor can more timely monitor changes in the patient's physiological status. The monitoring frequency optimization plan is a specific monitoring plan obtained by adjusting the basic intervention template based on the acute event risk factor. It includes information such as the body weight monitoring cycle, blood biochemical testing time window, and symptom self-assessment trigger conditions, which can guide medical staff and patients to arrange monitoring work in a reasonable manner.

[0086] When adjusting the frequency of physiological parameter collection in the basic intervention template based on the acute event risk factor, a frequency adjustment model can be established. This model can adjust the frequency of physiological parameter collection according to specific rules based on the magnitude of the acute event risk factor. For example, if the acute event risk factor is high, the body weight monitoring cycle can be shortened, the blood biochemical testing window can be advanced, and the symptom self-assessment trigger conditions can be lowered. If the acute event risk factor is low, the body weight monitoring cycle can be appropriately extended, the blood biochemical testing window can be postponed, and the symptom self-assessment trigger conditions can be increased. The body weight monitoring cycle refers to the time interval between two body weight measurements, the blood biochemical testing window refers to the time range for blood biochemical testing, and the symptom self-assessment trigger conditions refer to the conditions under which patients need to self-assess when certain symptoms occur. By adjusting the basic intervention template, an optimized monitoring frequency plan that incorporates this information is generated.

[0087] Step S500: Feedback the personalized intervention strategy set to the medical monitoring terminal, triggering the medical monitoring terminal to visualize the personalized intervention strategy set.

[0088] In this step, the medical monitoring terminal is a device, such as a computer or tablet, used by medical staff to monitor patient health and manage intervention strategies. Feeding the personalized intervention strategy set to the medical monitoring terminal allows medical staff to promptly understand the patient's intervention plan. Visualization displays the personalized intervention strategy set on the medical monitoring terminal screen in intuitive graphical or tabular formats, making it easier for medical staff to view and operate.

[0089] Feedback of the personalized intervention strategy set to the medical monitoring terminal can be achieved through network communication technologies such as Wi-Fi and Bluetooth. The personalized intervention strategy set can be transmitted to the medical monitoring terminal in the form of a data file. After receiving the data, the medical monitoring terminal will parse and process it, and then display the drug adjustment plan, dietary intervention plan, and monitoring frequency optimization plan in a visual manner. For example, a table can be used to display information such as the dose escalation gradient and dosing time interval in the drug adjustment plan, and a chart can be used to display information such as the intake ratio of different food categories in the dietary intervention plan.

[0090] As an implementation method, step S500, feeding back the personalized intervention strategy set to the medical monitoring terminal, triggering the medical monitoring terminal to visualize the personalized intervention strategy set, may specifically include the following steps S510 to S540: Step S510: configuring a strategy execution dashboard, a risk trend chart, and an early warning prompt window in the interactive interface of the medical monitoring terminal; the strategy execution dashboard is used to display key execution indicators of the drug adjustment plan, dietary intervention plan, and monitoring frequency optimization plan by region.

[0091] In this step, the interactive interface of the medical monitoring terminal is where medical staff interact with the device. By configuring a policy execution dashboard, risk trend charts, and early warning prompt windows within this interface, a more comprehensive display of patient information and intervention strategies can be achieved. The policy execution dashboard is an area that displays key execution indicators within the personalized intervention strategy set. It displays important information about medication adjustment plans, dietary intervention plans, and monitoring frequency optimization plans in separate sections, making it easier for medical staff to view and operate.

[0092] Graphical user interface (GUI) development tools such as Qt and Tkinter can be used to configure the strategy execution dashboard, risk trend charts, and early warning prompt windows within the interactive interface of the medical monitoring terminal. The strategy execution dashboard can be designed with multiple areas to display key execution indicators, such as the dose escalation gradient and dosing interval for a medication adjustment plan; the intake ratio of different food groups and sodium restriction standards for a dietary intervention plan; and the weight monitoring period and blood biochemical testing time window for a monitoring frequency optimization plan. Each area can display information in a table or list format, and can be configured with different colors and fonts to highlight important information.

[0093] Step S520: Generate a fluid balance trend curve based on the historical change data of the fluid retention trend characteristics, and embed the fluid balance trend curve into the risk trend chart; the fluid balance trend curve includes fluctuation nodes corresponding to daily intake and excretion and impact marks corresponding to clinical intervention events.

[0094] In this step, the historical change data of the fluid retention trend characteristics records the changes in the patient's fluid retention over time. By analyzing and processing this data, a fluid balance trend curve can be generated. The fluid balance trend curve is a curve that reflects the changes in the patient's fluid balance status over time. It contains fluctuation nodes corresponding to daily intake and excretion and impact markers corresponding to clinical intervention events, which can intuitively display the changes in the patient's fluid balance. The risk trend chart is used to show the changes in the patient's risk level over time. Embedding the fluid balance trend curve in it can more comprehensively reflect the patient's health status.

[0095] To generate a fluid balance trend curve based on historical data on fluid retention trend characteristics, data visualization tools such as Matplotlib and Seaborn can be used. First, arrange the historical data on fluid retention trend characteristics in chronological order. Then, use an interpolation algorithm to smooth the data to generate a continuous curve. Mark the curve with fluctuation nodes corresponding to daily intake and excretion, for example, at time points when intake and excretion experience significant changes. Also, mark the curve with impact markers corresponding to clinical intervention events, such as special symbols at the time points of clinical intervention events such as diuretic dose adjustments and dietary interventions. Finally, embed the generated fluid balance trend curve into the risk trend chart and display it alongside other risk indicators.

[0096] Step S530: When the abnormal warning level reaches the high-risk critical value, an emergency contact instruction and a multidisciplinary consultation request are pushed through the warning prompt window; the emergency contact instruction is used to notify the responsible physician to start the remote consultation process, and the multidisciplinary consultation request is used to coordinate the cardiovascular specialist, nutrition department and pharmacy department to conduct a joint diagnosis and treatment evaluation.

[0097] In this step, if the abnormal warning level reaches the high-risk critical value, it means that the patient's condition is more serious and timely intervention measures are required. The warning prompt window is a window in the interactive interface of the medical monitoring terminal, which is used to display warning information. The emergency contact instruction is an instruction to notify the responsible physician to initiate the remote consultation process. Through this instruction, the responsible physician can be contacted in time for remote diagnosis and treatment of the patient. The multidisciplinary consultation request is a request to coordinate experts from multiple disciplines such as cardiovascular specialists, nutrition departments, and pharmacy departments to conduct joint diagnosis and treatment evaluations. Through multidisciplinary consultations, the patient's condition can be more comprehensively assessed and a more reasonable treatment plan can be formulated.

[0098] When the abnormal warning level reaches the high-risk threshold, the medical monitoring terminal automatically triggers a warning window, displaying emergency contact instructions and a multidisciplinary consultation request. The emergency contact instructions can be sent to the responsible physician via text message, email, or instant messaging, informing them to initiate the remote consultation process. The multidisciplinary consultation request can be sent through the hospital's internal information system to relevant departments such as the cardiovascular department, nutrition department, and pharmacy department to coordinate experts for joint diagnosis and treatment evaluation. The warning window also displays the patient's basic information, medical condition overview, and other relevant information, making it easier for physicians and experts to understand the patient's condition.

[0099] Step S540: Record the medical staff's correction operations and effect evaluation data on the personalized intervention strategy set in the medical monitoring terminal, and transmit the correction operations and effect evaluation data back to the cloud server to trigger incremental learning updates of the capacity overload prediction model.

[0100] In this step, healthcare professionals modify the personalized intervention strategy set by making adjustments to medication adjustments, dietary intervention plans, and monitoring frequency optimization plans based on the patient's specific circumstances. Effectiveness evaluation data provides healthcare professionals with an assessment of the effectiveness of implementing the personalized intervention strategy set, such as patient improvement and changes in physiological parameters. Transmitting these modification and effectiveness evaluation data back to the cloud server provides new training data for the capacity overload prediction model, triggering incremental learning updates. This allows the model to continuously learn and adapt to new situations, improving prediction accuracy.

[0101] When recording medical staff's correction operations and effect evaluation data for personalized intervention strategy sets on the medical monitoring terminal, a database management system such as SQLite or MySQL can be used. After medical staff perform correction operations and enter effect evaluation data on the medical monitoring terminal, the system stores this data in the database. The data in the database is then transmitted back to the cloud server via network communication technology. After receiving the data, the cloud server inputs it into the capacity overload prediction model as new training data, triggering incremental learning updates of the model. Incremental learning updates can use online learning algorithms, such as stochastic gradient descent, to adjust the model parameters so that the model can better adapt to new data and situations.

[0102] As an embodiment, step S540 records the medical staff's correction operations and effect evaluation data for the personalized intervention strategy set in the medical monitoring terminal, and transmits the correction operations and effect evaluation data back to the cloud server to trigger incremental learning and updating of the capacity overload prediction model. Specifically, the following steps S541 to S544 may be included: Step S541: Collect the actual adjustment records and clinical effect feedback data of the medical staff for the drug adjustment plan to generate a first incremental training sample; the first incremental training sample includes the mapping relationship between the original drug plan, the revised drug plan and the improvement range of the physiological parameters.

[0103] In this step, the actual adjustment record of the medical staff for the drug adjustment plan refers to the record of the medical staff's modification of parameters such as the dose escalation gradient and dosing time interval in the drug adjustment plan. Clinical effect feedback data is the evaluation information of the medical staff on the implementation effect of the drug adjustment plan, such as the patient's symptom improvement, physiological parameter changes, etc. The first incremental training sample is a sample generated for model training based on the actual adjustment record and clinical effect feedback data. It contains the mapping relationship between the original drug plan, the revised drug plan and the improvement range of physiological parameters. Through this sample, the model can learn the relationship between drug adjustment and clinical effect.

[0104] When collecting medical staff's actual adjustment records and clinical effect feedback data for the drug adjustment plan, a dedicated input interface can be set up on the medical monitoring terminal for medical staff to enter relevant information. The system will store this information in the database and organize and analyze it. When generating the first incremental training sample, the original drug plan, the revised drug plan, and the improvement in physiological parameters are associated to form a sample record. For example, the original drug plan is a diuretic dose of 5mg once a day; the revised drug plan is a diuretic dose of 10mg twice a day; and the improvement in physiological parameters is a weight loss of 2kg. This information is combined into a sample record as the first incremental training sample.

[0105] Step S542: Collect the actual execution data of the dietary intervention plan by medical staff and the patient compliance assessment results to generate a second incremental training sample; the second incremental training sample includes a correlation matrix between theoretical intake limits, actual intake deviations and fluid balance fluctuations.

[0106] In this step, the actual execution data of the dietary intervention plan by medical staff refers to the actual dietary conditions of patients during the implementation of the dietary intervention plan, such as the actual intake of different food categories, etc. The patient compliance assessment results are the evaluation information of medical staff on whether the patient follows the dietary intervention plan. The second incremental training sample is a sample generated for model training based on the actual execution data and the patient compliance assessment results. It contains the correlation matrix between theoretical intake limits, actual intake deviations, and fluid balance fluctuations. Through this sample, the model can learn the relationship between dietary intervention and fluid balance.

[0107] When collecting the actual implementation data of the dietary intervention plan by medical staff and the results of patient compliance assessment, medical staff can record the patient's diet through the medical monitoring terminal and evaluate the patient's compliance. The system will store this information in the database and organize and analyze it. When generating the second incremental training sample, the deviation between the theoretical intake limit and the actual intake is calculated, and the patient's fluid balance fluctuation is recorded. This information is constructed into a correlation matrix as the second incremental training sample. For example, the theoretical intake limit is that the daily fluid intake does not exceed 1000ml, the actual intake deviation is more than 200ml, and the fluid balance fluctuation is an increase of 1kg in body weight. This information is constructed into a matrix to reflect the correlation between them.

[0108] Step S543: Collect the execution logs of the monitoring frequency optimization plan and the statistical results of abnormal event detection rates of medical staff to generate a third incremental training sample; the third incremental training sample includes the temporal correlation characteristics between the monitoring cycle adjustment parameters, the acute event warning delay time and the intervention success rate.

[0109] In this step, the execution log of the monitoring frequency optimization plan by medical staff refers to the specific operations and time information recorded by medical staff during the execution of the monitoring frequency optimization plan, such as the time of body weight monitoring, the time of blood biochemical testing, etc. The statistical results of the abnormal event detection rate are the results of statistics on the proportion of abnormal events detected during the monitoring process. The third incremental training sample is a sample generated for model training based on the execution log and the statistical results of the abnormal event detection rate. It contains the temporal correlation characteristics between the monitoring cycle adjustment parameters, the acute event warning delay time and the intervention success rate. Through this sample, the model can learn the relationship between monitoring frequency optimization and acute event warning and intervention effects.

[0110] When collecting the execution logs of the monitoring frequency optimization plan and the statistical results of the abnormal event detection rate of medical staff, the logging function can be set on the medical monitoring terminal to allow medical staff to record information during the execution process. At the same time, the system will collect statistics on the detection of abnormal events. When generating the third incremental training sample, extract information such as the monitoring cycle adjustment parameters, acute event warning delay time and intervention success rate, and analyze the temporal correlation features between them. For example, the monitoring cycle adjustment parameter is to adjust the body weight monitoring cycle from once a week to once a day, the acute event warning delay time is shortened by 1 day, and the intervention success rate is increased by 20%. This information is constructed into a temporal correlation feature as the third incremental training sample.

[0111] Step S544: Input the first incremental training sample, the second incremental training sample, and the third incremental training sample into the parameter optimization module of the capacity overload prediction model. By comparing the feature differences between the prediction strategy and the actual correction strategy, adjust the weight calculation rules in the feature weight allocation module and the level matching algorithm in the risk threshold determination module.

[0112] In this step, the parameter optimization module of the capacity overload prediction model is a module used to optimize and adjust the parameters of the model. By inputting the first incremental training sample, the second incremental training sample, and the third incremental training sample into this module, the model can learn new data and situations, thereby adjusting its own parameters. Comparing the feature differences between the prediction strategy and the actual correction strategy refers to comparing the differences between the intervention strategy predicted by the model and the intervention strategy actually corrected by medical staff. By analyzing these differences, the deficiencies of the model can be discovered, and then the weight calculation rules in the feature weight allocation module and the grade matching algorithm in the risk threshold determination module can be adjusted.

[0113] When the first, second, and third incremental training samples are input into the parameter optimization module of the volume overload prediction model, the module analyzes and processes these samples. First, the feature differences between the prediction strategy and the actual correction strategy are compared, and the difference indicators between the two, such as the mean square error, are calculated. Then, the weight calculation rules in the feature weight allocation module and the level matching algorithm in the risk threshold determination module are adjusted based on the difference indicators. For example, if it is found that the model's weight allocation for the fluid retention trend feature is unreasonable, the weight attenuation factor can be adjusted based on the difference indicator; if it is found that the model's judgment of the abnormal warning level is inaccurate, the boundary interval of the risk level threshold can be corrected based on the difference indicator.

[0114] As an embodiment, in step S544, by comparing the feature differences between the prediction strategy and the actual correction strategy, the weight calculation rules in the feature weight allocation module and the level matching algorithm in the risk threshold determination module are adjusted. Specifically, the following steps S5441 to S5444 may be included: Step S5441: Calculate a first difference coefficient between the first weight distribution value of the fluid retention trend feature in the original prediction strategy and the second weight distribution value in the revised strategy, and adjust the weight attenuation factor for the fluid retention trend feature in the feature weight distribution module according to the first difference coefficient.

[0115] In this step, the first weight assigned is the weight assigned to the fluid retention trend feature in the original prediction strategy, and the second weight assigned is the weight assigned to the fluid retention trend feature in the revised strategy. The first difference coefficient is an indicator that calculates the degree of difference between the first and second weight assigned values. This coefficient can be used to determine whether the model's weight assignment for the fluid retention trend feature is reasonable. The weight attenuation factor is a parameter used in the feature weight assignment module to adjust the weight of the fluid retention trend feature. Adjusting this factor based on the first difference coefficient can enable the model to more accurately assign weights.

[0116] To calculate the first difference coefficient between the first weight assigned to the fluid retention trend feature in the original prediction strategy and the second weight assigned to the feature in the revised strategy, use the following formula: First Difference Coefficient = |First Weight Assignment Value - Second Weight Assignment Value| / First Weight Assignment Value. For example, if the first weight assigned is 0.6 and the second weight assigned is 0.8, then the first difference coefficient = |0.6 - 0.8| / 0.6 = 0.33. When adjusting the weight attenuation factor for the fluid retention trend feature in the feature weight assignment module based on the first difference coefficient, if the first difference coefficient is large, it indicates that the model's weight assignment for the fluid retention trend feature is unreasonable, and the weight attenuation factor needs to be adjusted to make the weight assignment more consistent with actual conditions.

[0117] Step S5442: Calculate the second difference coefficient between the third weight allocation value of the metabolic abnormality associated feature in the original prediction strategy and the fourth weight allocation value in the revised strategy, and adjust the weight enhancement amplitude for the metabolic abnormality associated feature in the feature weight allocation module according to the second difference coefficient.

[0118] In this step, the third weight assignment value is the weight assigned to the metabolic abnormality-associated feature in the original prediction strategy, and the fourth weight assignment value is the weight assigned to the metabolic abnormality-associated feature in the revised strategy. The second difference coefficient is an indicator that calculates the degree of difference between the third and fourth weight assignment values. This coefficient can be used to understand whether the model's weight assignment of metabolic abnormality-associated features is reasonable. The weight enhancement amplitude is a parameter used in the feature weight assignment module to adjust the weight of the metabolic abnormality-associated feature. Adjusting this amplitude based on the second difference coefficient can enable the model to more accurately assign weights.

[0119] When calculating the second difference coefficient between the third weight assigned to a metabolic abnormality-associated feature in the original prediction strategy and the fourth weight assigned to it in the revised strategy, a formula similar to the first difference coefficient can be used: Second Difference Coefficient = |Third Weight Assignment Value - Fourth Weight Assignment Value| / Third Weight Assignment Value. For example, if the third weight assigned is 0.3 and the fourth weight assigned is 0.4, then the second difference coefficient = |0.3-0.4| / 0.3 = 0.33. When adjusting the weight boost for metabolic abnormality-associated features in the feature weight assignment module based on the second difference coefficient, a large second difference coefficient indicates that the model's weight assignment for metabolic abnormality-associated features is unreasonable, and the weight boost should be adjusted to make the weight assignment more realistic. For example, an adjustment rule can be set such that when the second difference coefficient exceeds a certain threshold (e.g., 0.2), the weight boost is increased or decreased by a certain percentage (e.g., 10%). If the second difference coefficient is 0.33 and the current weight enhancement amplitude is 0.1, then the weight enhancement amplitude can be adjusted to 0.1×(1+0.1)=0.11, so that the metabolic abnormality-related features can be more reasonably weighted in subsequent predictions, thereby improving the accuracy of model predictions.

[0120] Step S5443: Analyze the level deviation between the first level identification of the abnormal warning level in the original prediction result and the second level identification in the actual clinical evaluation, and correct the boundary interval of each risk level threshold in the risk threshold determination module according to the level deviation.

[0121] In this step, the first-level identifier is the identifier of the abnormal warning level in the original prediction result, such as low risk, medium risk, and high risk; the second-level identifier is the identifier of the abnormal warning level in the actual clinical evaluation. The level deviation is an indicator that measures the degree of difference between the first-level identifier and the second-level identifier. By calculating this deviation, the deviation of the model in determining the abnormal warning level can be detected. The boundary intervals of each risk level threshold in the risk threshold determination module are the boundaries that divide different risk levels. Correcting these boundary intervals based on the level deviation can make the model's determination of the abnormal warning level more accurate.

[0122] When analyzing the deviation between the first-level indicator in the original prediction results and the second-level indicator in the actual clinical evaluation, a level coding approach can be used. For example, low risk can be coded as 1, medium risk as 2, and high risk as 3. The level deviation can be calculated using the absolute value of the difference between the two coding values: level deviation = |first-level indicator coding value - second-level indicator coding value|. If the first-level indicator is medium risk (coded as 2) and the second-level indicator is high risk (coded as 3), the level deviation is |2-3| = 1.

[0123] A correction rule can be established to modify the boundary intervals of each risk level threshold in the risk threshold determination module based on level deviation. When the level deviation is 1, it indicates a one-level deviation in the model's judgment. In this case, the corresponding risk level threshold boundary interval can be adjusted by a certain percentage (e.g., 5%). For example, if a low-risk diagnosis is misclassified as medium-risk, the boundary thresholds for low and medium risks can be increased by 5%. If a medium-risk diagnosis is misclassified as high-risk, the boundary thresholds for medium and high risks can be increased by 5%. By continuously modifying the boundary intervals based on level deviation, the model can better adapt to actual clinical situations and improve the accuracy of abnormal warning level determination.

[0124] Step S5444: Optimizing the logistic regression coefficient for calculating the probability of fluid overload in the volume overload prediction model based on the probability error between the predicted value of the probability of fluid overload and the actual clinical diagnosis result.

[0125] In this step, the predicted value of the probability of fluid overload is the probability of a patient's fluid overload calculated by the volume overload prediction model based on the input data. The actual clinical diagnosis is the actual condition of the patient determined by the physician through clinical examination and diagnosis. The probability error is the difference between the predicted value and the actual clinical diagnosis. This error can be used to assess the accuracy of the model in calculating the probability of fluid overload. The logistic regression coefficients are the parameters used in the volume overload prediction model to calculate the probability of fluid overload. Optimizing these coefficients based on the probability error can enable the model to more accurately calculate the probability of fluid overload.

[0126] When calculating the probability error between the predicted value of the probability of fluid overload and the actual clinical diagnosis result, statistical methods such as mean square error can be used. Assume that the predicted value is p 预测 The actual clinical diagnosis result is recorded as 1 if there is fluid overload, and 0 if there is no fluid overload, and recorded as p 实际 , then the probability error E can be expressed as E=(p 预测 -p 实际 ) 2For example, the model predicted that the probability of fluid overload in patients was 0.7, while the actual clinical diagnosis showed that the patient had fluid overload (p 实际 =1), then the probability error is (0.7-1) 2 =0.09.

[0127] The gradient descent algorithm can be used to optimize the logistic regression coefficients used to calculate the probability of fluid overload in volume overload prediction models based on the probability error. The gradient descent algorithm is an iterative optimization algorithm that gradually reduces the probability error by continuously adjusting the logistic regression coefficients. Specifically, the gradient of the probability error with respect to the logistic regression coefficients is first calculated, and then the logistic regression coefficients are updated based on the direction and magnitude of the gradient. The magnitude of each update is controlled by the learning rate, a pre-set parameter that determines the step size of each update. By iteratively updating the logistic regression coefficients multiple times, the probability error is minimized, thereby optimizing the model's calculation of the probability of fluid overload.

[0128] As an implementation manner, the method provided in the embodiment of the present invention may further include the following steps S600 to S800: Step S600: Configure a patient identity authentication module and a data encryption transmission channel in the medical monitoring terminal; the patient identity authentication module is used to verify the patient's identity information through biometric recognition or password, and the data encryption transmission channel is used to perform end-to-end encryption on the physiological monitoring data set and the daily activity record data set.

[0129] In this step, the patient identity verification module plays a crucial role in ensuring patient data security and privacy. Biometric recognition is a technology that uses biological characteristics (such as fingerprints, facial features, and irises) to identify individuals, offering high accuracy and security. Password verification requires patients to enter a pre-set password or passcode to verify their identity. The patient identity verification module ensures that only authorized patients or healthcare professionals can access relevant data, preventing data leaks and unauthorized access.

[0130] The encrypted data transmission channel is used to securely transmit physiological monitoring and daily activity data sets. End-to-end encryption ensures that data remains encrypted throughout its transmission from sender to receiver, ensuring that only the receiver possesses the decryption key. This prevents data from being eavesdropped or tampered with during transmission, ensuring data integrity and confidentiality.

[0131] When configuring a patient authentication module in a medical monitoring terminal, you can integrate a biometric sensor (such as a fingerprint sensor or facial recognition camera) or provide a password input interface. When a patient or healthcare provider needs to access data, the system will require authentication. If biometric authentication is used, the system collects the patient's biometric data and compares it to a pre-stored template. If password verification is used, the system will require a password for verification.

[0132] When configuring a data encryption transmission channel, you can use either a symmetric encryption algorithm (such as AES) or an asymmetric encryption algorithm (such as RSA). At the data sender, the physiological monitoring and daily activity recording datasets are encrypted using an encryption key, and then the encrypted data is transmitted over the network to the receiver. At the receiver, the data is decrypted using the decryption key. To ensure secure key transmission, a key exchange protocol (such as the Diffie-Hellman key exchange protocol) can be used to generate and exchange keys.

[0133] Step S700: When it is detected that an unauthorized device attempts to access the personalized intervention policy set, the data leakage protection mechanism is triggered. The data leakage protection mechanism includes immediately terminating data transmission, clearing temporary cache files, and sending security warning information to the supervision platform.

[0134] In this step, an unauthorized device is one that attempts to access the personalized intervention policy set without authorization. Detecting unauthorized device access can be achieved through network security monitoring technologies. For example, firewalls can monitor network traffic and identify abnormal access requests. Intrusion detection systems can detect abnormal behavior in the system and determine whether unauthorized devices are accessing the system.

[0135] Data leakage prevention mechanisms are a series of measures designed to prevent the leakage of sensitive data, such as personalized intervention policy sets. Immediately terminating data transmission prevents unauthorized devices from accessing further data; clearing temporary cache files ensures no sensitive data remains on the device; and sending security alerts to the regulatory platform promptly notifies relevant personnel to take measures, such as strengthening security measures and investigating the source of the intrusion.

[0136] When an unauthorized device attempts to access a personalized intervention policy set, the network security monitoring system issues an alert. Upon receiving the alert, the medical monitoring terminal immediately terminates any ongoing data transmission and ceases communication with the unauthorized device. Simultaneously, the system clears temporary cache files, which may contain data from a portion of the personalized intervention policy set. Clearing these files reduces the risk of data leakage. The system then sends a security alert to the monitoring platform. This alert may include information such as the unauthorized device's IP address, access time, and the accessed dataset, enabling further investigation and action.

[0137] Step S800: Set up a data access permission classification policy in the cloud server so that the responsible physician has full policy modification authority, the nursing staff has partial policy viewing authority, and the patient's family members only have warning status query authority.

[0138] In this step, the data access permission grading policy is set to ensure that different personnel have access and operational permissions to data such as the personalized intervention policy set that meet their responsibilities and needs. The responsible physician is the primary person responsible for patient treatment and management. They need to modify and adjust the personalized intervention policy set, so they are granted full policy modification permissions. Nursing staff are primarily responsible for implementing and observing patient treatment. They need to view some policy information but do not need to modify it. Therefore, they are granted partial policy viewing permissions. Family members are primarily concerned with the patient's alert status, so they are only granted alert status query permissions.

[0139] When setting up a hierarchical data access policy in a cloud server, a role-permission model can be used. First, define different roles, such as responsible physicians, nursing staff, and patient family members. Then, assign corresponding permissions to each role. The responsible physician's permissions include the ability to modify and view medication adjustment plans, dietary intervention plans, and monitoring frequency optimization plans. The nursing staff's permissions include the ability to view some policy information (such as medication instructions and dietary precautions). The patient family members' permissions only include the ability to query warning status (such as abnormal warning level and the risk of acute events).

[0140] When different personnel log in to the cloud server to access data, the system verifies their permissions based on their roles. If the responsible physician logs in, the system allows them to modify and view the entire policy. If a nurse logs in, the system displays only the policy information they are authorized to view. If a patient's family member logs in, the system only provides access to alert status information. This hierarchical data access policy effectively protects data security and privacy while ensuring that different personnel have access to the information they need.

[0141] It should be noted that the method provided in the embodiment of the present invention is only to provide reference results to help medical staff and patients, and is not a direct diagnostic and treatment result. That is, the core purpose of the solution provided in the embodiment of the present invention is not to directly diagnose heart failure or implement specific treatment behaviors. The present invention aims to build a comprehensive volume management system for patients with heart failure, which focuses on long-term and dynamic monitoring and management of the patient's volume status to prevent and reduce the risks associated with heart failure caused by volume imbalance. By collecting and analyzing the patient's daily physiological data and activity records, abnormal volume trends can be discovered in advance, providing medical staff with a scientific and objective basis for decision-making, and assisting in the formulation of personalized intervention strategies, thereby maintaining the patient's fluid balance, improving the patient's quality of life, and reducing the possibility of acute attacks of heart failure.

[0142] In addition, it can be understood that the various algorithms involved in the above-mentioned introduction of the embodiments of the present invention can be learned from the relevant content in the prior art. In order to save space, they will not be expanded too much in the embodiments of the present invention. In addition, when implementing the scheme of the present invention, those skilled in the art can supplement the details according to the common knowledge in this field. For example, according to the common knowledge in this field, normalization can be used to eliminate dimensional conflicts before feature fusion, interpolation can be used to eliminate dimensional differences, and historical data, experience or business scenario requirements can be combined to reasonably set thresholds, and models can be trained based on general model training methods, etc. The present invention will no longer provide redundant introductions to overly detailed implementation processes.

[0143] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 may be connected via a bus or other means. The processor 101 (or central processing unit (CPU)) is the computing and control core of the computer system, capable of parsing various instructions within the computer system and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, a mobile communication interface, etc.), which can be used to send and receive data under the control of the processor 101. The communication interface 102 can also be used for data transmission and interaction within the computer system. The memory 103 is a memory device in the computer system for storing programs and data. It is understood that the memory 103 here can include both the built-in memory of the computer system and, of course, the extended memory supported by the computer system. The memory 103 provides storage space, which stores the computer system's operating system, but this is not limited to this in the present invention.

[0144] In one embodiment, the processor 101 executes the health management method for heart failure patients provided in the above embodiment of the present invention by running the computer program in the memory 103 .

Claims

1. A method for health management of patients with heart failure, characterized in that: The following steps are involved: Obtaining a physiological monitoring data set and a daily activity record data set uploaded by the patient through the patient terminal. The physiological monitoring data set includes continuously collected body mass data, fluid balance parameters, and circulatory system indicators. The daily activity record data set includes dietary intake image information and fecal feature image information; Performing multimodal feature fusion processing on the physiological monitoring dataset and the daily activity record dataset to generate a comprehensive risk assessment feature set; The comprehensive risk assessment feature set is input into a pre-trained volume overload prediction model, and the comprehensive risk assessment feature set is weighted and matched by the volume overload prediction model to generate a volume status assessment result and an abnormality warning level; the volume status assessment result is used to indicate the degree of deviation between the patient's current fluid balance state and a preset clinical indicator, and the abnormality warning level is used to indicate the urgency of the fluid imbalance event; generating a personalized intervention strategy set based on the capacity status assessment result and the abnormal warning level; Feeding back the personalized intervention strategy set to a medical monitoring terminal triggers the medical monitoring terminal to perform a visual display of the personalized intervention strategy set.

2. The method according to claim 1, characterized in that The comprehensive risk assessment feature set includes a fluid retention trend feature, a metabolic abnormality association feature, and an intake-excretion imbalance feature. The multimodal feature fusion processing of the physiological monitoring dataset and the daily activity record dataset is performed to generate a comprehensive risk assessment feature set, including: Performing visual feature analysis on the dietary intake image information to obtain food type recognition results and intake estimation parameters; the food type recognition results are used to distinguish high-sodium food categories, high-water food categories, and low-osmotic pressure food categories, and the intake estimation parameters are used to quantify the patient's total liquid intake during a single meal; Performing morphological analysis on the excretion characteristic image information to obtain an excretion type classification result and an excretion volume estimation parameter; the excretion type classification result is used to distinguish between urine excretion categories, sweat excretion categories, and fecal excretion categories, and the excretion volume estimation parameter is used to quantify the total amount of fluid excreted by the patient in a single excretion; Performing time series alignment processing on the body weight data, the body fluid balance parameters, and the circulatory system indicators to generate a physiological state time series feature sequence; the physiological state time series feature sequence is used to characterize the patient's body weight fluctuation trend, body fluid retention rate, and circulatory load change cycle; Performing cross-modal association analysis on the food type recognition result, the intake estimation parameter, the excretion type classification result, the excretion estimation parameter, and the physiological state time series feature sequence to generate the fluid retention trend feature, the metabolic abnormality association feature, and the intake-excretion imbalance feature; wherein the cross-modal association analysis includes: determining a positive impact weight of a high sodium intake event on a body fluid retention rate based on a first correlation coefficient between the food type recognition result and the body fluid balance parameter; determining a reverse adjustment factor of the abnormal excretion volume event on the circulatory load change based on a second correlation coefficient between the excretion type classification result and the circulatory system indicator; Based on the positive impact weight and the negative adjustment factor, a compensation calculation is performed on the intake estimation parameter and the excretion estimation parameter to generate the intake-excretion imbalance feature.

3. The method according to claim 2, characterized in that The step of inputting the comprehensive risk assessment feature set into a pre-trained capacity overload prediction model, performing weight matching processing on the comprehensive risk assessment feature set by the capacity overload prediction model, and generating a capacity status assessment result and an abnormal warning level includes: A feature weight allocation module and a risk threshold determination module are configured in the volume overload prediction model; the feature weight allocation module is used to adjust the contribution weight of each feature in the volume status assessment based on the real-time change amplitude of the fluid retention trend feature, the metabolic abnormality association feature, and the intake-excretion imbalance feature; the risk threshold determination module is used to determine the matching degree of the current comprehensive risk assessment feature set and each warning level based on the fluid imbalance critical value corresponding to different warning levels in historical clinical data; The feature weight assignment module assigns a first weight to the fluid retention trend feature, a second weight to the metabolic abnormality association feature, and a third weight to the intake-excretion imbalance feature; the sum of the first weight, the second weight, and the third weight is a preset normalized value; performing weighted fusion processing on the fluid retention trend feature, the metabolic abnormality association feature, and the intake-excretion imbalance feature according to the first weight, the second weight, and the third weight to generate a comprehensive risk score; Inputting the comprehensive risk score into the risk threshold determination module, and determining the abnormality warning level by comparing the comprehensive risk score with a plurality of preset risk level thresholds; wherein the plurality of risk level thresholds include a low risk critical value, a medium risk critical value, and a high risk critical value; The volume status assessment result is generated based on the deviation between the comprehensive risk score and the nearest risk level threshold; the volume status assessment result includes the probability of fluid overload, the compensatory regulatory capacity index and the risk coefficient of acute event occurrence.

4. The method according to claim 3, characterized in that The personalized intervention strategy set includes a medication adjustment plan, a diet intervention plan, and a monitoring frequency optimization plan. The personalized intervention strategy set is generated based on the volume status assessment result and the abnormal warning level, including: Filtering a basic intervention template from a preset strategy library according to the abnormal warning level; the preset strategy library stores medication adjustment rules, dietary restriction rules, and monitoring frequency rules corresponding to different warning levels; adjusting the diuretic dosage in the basic intervention template according to the probability of fluid overload to generate the drug adjustment plan; the drug adjustment plan includes a dose escalation gradient, a dosing time interval, and a drug incompatibility prompt; Optimizing the upper limit of daily fluid intake in the basic intervention template according to the compensatory regulatory capacity index to generate the dietary intervention plan; the dietary intervention plan includes intake ratios of different food categories, sodium salt restriction standards, and water supplementation recommendations; The physiological parameter collection frequency in the basic intervention template is adjusted according to the risk coefficient of the acute event to generate the monitoring frequency optimization plan; the monitoring frequency optimization plan includes the body weight monitoring cycle, the blood biochemical detection time window and the symptom self-assessment trigger condition.

5. The method according to claim 4, characterized in that Feeding back the personalized intervention strategy set to the medical monitoring terminal, and triggering the medical monitoring terminal to visually display the personalized intervention strategy set, includes: A strategy execution dashboard, a risk trend chart, and an early warning prompt window are configured in the interactive interface of the medical monitoring terminal; the strategy execution dashboard is used to display key execution indicators of the medication adjustment plan, the dietary intervention plan, and the monitoring frequency optimization plan in different areas; Generate a fluid balance trend curve based on the historical change data of the fluid retention trend characteristics, and embed the fluid balance trend curve into the risk trend chart; the fluid balance trend curve includes fluctuation nodes corresponding to daily intake and excretion and impact markers corresponding to clinical intervention events; When the abnormal warning level reaches the high-risk critical value, an emergency contact instruction and a multidisciplinary consultation request are pushed through the warning prompt window; the emergency contact instruction is used to notify the responsible physician to initiate a remote consultation process, and the multidisciplinary consultation request is used to coordinate the cardiovascular specialist, nutrition department, and pharmacy department to conduct a joint diagnosis and treatment evaluation; The medical monitoring terminal records the correction operations and effect evaluation data of the medical staff on the personalized intervention strategy set, and transmits the correction operations and the effect evaluation data back to the cloud server to trigger incremental learning updates of the capacity overload prediction model.

6. The method according to claim 5, characterized in that The step of recording the medical staff's correction operations and effect evaluation data on the personalized intervention strategy set in the medical monitoring terminal, and transmitting the correction operations and effect evaluation data back to the cloud server to trigger incremental learning and updating of the capacity overload prediction model, includes: Collecting actual adjustment records and clinical effect feedback data of medical staff on the drug adjustment plan to generate a first incremental training sample; the first incremental training sample includes a mapping relationship between the original drug plan, the revised drug plan, and the improvement range of the physiological parameters; Collecting actual implementation data of the dietary intervention plan by medical staff and patient compliance assessment results to generate a second incremental training sample; the second incremental training sample includes a correlation matrix between theoretical intake limits, actual intake deviations, and fluid balance fluctuations; Collecting the execution logs of the monitoring frequency optimization plan and the statistical results of abnormal event detection rates of medical staff to generate a third incremental training sample; the third incremental training sample includes the temporal correlation characteristics between the monitoring cycle adjustment parameter, the acute event warning delay time, and the intervention success rate; The first incremental training sample, the second incremental training sample, and the third incremental training sample are input into the parameter optimization module of the capacity overload prediction model. By comparing the feature differences between the prediction strategy and the actual correction strategy, the weight calculation rules in the feature weight allocation module and the level matching algorithm in the risk threshold determination module are adjusted.

7. The method according to claim 6, characterized in that The method of adjusting the weight calculation rules in the feature weight allocation module and the level matching algorithm in the risk threshold determination module by comparing the feature differences between the prediction strategy and the actual correction strategy includes: calculating a first difference coefficient between a first weight distribution value of the fluid retention trend feature in the original prediction strategy and a second weight distribution value in the revised strategy, and adjusting a weight attenuation factor for the fluid retention trend feature in the feature weight distribution module according to the first difference coefficient; Calculating a second difference coefficient between a third weight allocation value of the metabolic abnormality associated feature in the original prediction strategy and a fourth weight allocation value in the revised strategy, and adjusting a weight enhancement amplitude for the metabolic abnormality associated feature in the feature weight allocation module according to the second difference coefficient; Analyzing the level deviation between the first level identifier of the abnormal warning level in the original prediction result and the second level identifier in the actual clinical evaluation, and correcting the boundary interval of each risk level threshold in the risk threshold determination module according to the level deviation; The logistic regression coefficient for calculating the probability of fluid overload in the volume overload prediction model is optimized according to the probability error between the predicted value of the probability of fluid overload and the actual clinical diagnosis result.

8. The method according to claim 1, characterized in that The step of obtaining the physiological monitoring data set and daily activity record data set uploaded by the patient includes: Continuously collecting the body mass data and the circulatory system index through a wearable device worn by the patient, converting the body mass data into a daily body mass change rate, and decomposing the circulatory system index into a heart rate variability parameter, a blood pressure fluctuation coefficient, and a blood oxygen saturation trend value; The patient captures the dietary intake image information and the fecal characteristic image information through a mobile terminal, and performs illumination correction and perspective transformation processing on the dietary intake image information to obtain a standardized dietary image; and performs color space conversion and texture enhancement processing on the fecal characteristic image information to obtain an enhanced fecal image; The daily body mass change rate, the heart rate variability parameter, the blood pressure fluctuation coefficient, the blood oxygen saturation trend value, the standardized diet image, and the enhanced excrement image are timestamp aligned and data type unified to generate the physiological monitoring dataset and the daily activity recording dataset.

9. The method according to claim 8, characterized in that The step of performing illumination correction and perspective transformation on the dietary intake image information to obtain a standardized dietary image includes: detecting the outline of the tableware and the boundary of the food area in the food intake image information, determining the tilt angle of the shooting perspective according to the outline of the tableware, and performing a three-dimensional projection transformation on the food intake image information based on the tilt angle to generate a top-view plane image under an orthographic projection perspective; extracting a pixel distribution histogram of each food area in the top-view planar image, identifying overexposed areas and shadowed areas based on the pixel distribution histogram, and performing illumination equalization processing on the overexposed areas and the shadowed areas using an adaptive brightness compensation algorithm; The pre-trained food volume estimation model is used to perform stereoscopic reconstruction on the top-view plane image after illumination equalization to generate intake estimation parameters; the stereoscopic reconstruction processing includes inferring the actual volume of the food based on prior knowledge of the tableware size, and converting it into a liquid equivalent value based on the food density parameter.

10. A computer system, characterized in that: include: a memory storing a computer program; A processor, configured to load the computer program to implement the health management method for heart failure patients according to any one of claims 1 to 9.

Images