Risk assessment method and system for noninvasive dynamic monitoring of urinary and male reproductive system diseases

Abstract

The invention provides a risk assessment method and system for noninvasive dynamic monitoring of diseases of a urinary and male reproductive system, and relates to the technical field of urology andrology. By collecting multi-dimensional physiological data and combining an intelligent algorithm to deeply analyze the data, potential disease signals can be automatically identified in a short time, and the risk assessment method and system can be applied to the field of non-invasive dynamic monitoring of diseases of the urinary and male reproductive system. And personalized disease risk prediction is provided based on historical data, the physiological state of the patient is analyzed and synthesized in real time, a monitoring strategy is dynamically adjusted, and a personalized disease prediction result is provided according to the health condition of the patient and medical record data. Compared with a traditional technology which depends on a single data source, the method has the advantages that through comprehensive analysis of multi-dimensional data, the defect that the traditional method lacks dynamic monitoring is overcome, and limitation caused by excessive dependence of the single data source can be avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urology and male medicine, and in particular to a risk assessment method and system for non-invasive dynamic monitoring of urological and male reproductive system diseases. BACKGROUND

[0002] With the continuous development of medical technology, the diagnosis methods for urological and male reproductive system diseases have gradually diversified. Traditional detection methods include urine analysis, blood chemistry testing, imaging examination (such as ultrasound, CT, MRI, etc.), and tissue biopsy, etc.

[0003] However, the existing diagnosis technology has limitations. For example, urine analysis and blood chemistry testing can perform preliminary screening, but they are difficult to provide sufficient pathological information for comprehensive diagnosis; imaging examination (such as ultrasound) can provide certain disease indications, but it is greatly affected by equipment, patient body type and other factors, and is prone to misdiagnosis or missed diagnosis; while tissue biopsy is a relatively reliable diagnosis method, it has risks and discomfort due to its invasive operation, and the sampling representation is limited, which is difficult to fully reflect the lesion. More importantly, the existing traditional detection methods mostly provide static data, lacking comprehensive monitoring and long-term evaluation of the dynamic changes of the disease, which makes it difficult to early warn and predict the progress of the disease. In actual clinical practice, doctors often rely on the symptoms and examination results of patients for judgment, but especially in the early stage, patients often have no obvious symptoms, leading to rapid development of the disease without awareness, missing the best treatment opportunity. Although the existing non-invasive monitoring technology reduces the risk of invasive operation to some extent, it also has problems such as insufficient dynamic data collection, discontinuous monitoring, and difficulty in effectively predicting disease progression.

[0004] Therefore, how to break through the limitations of existing technology through multi-modal data fusion and intelligent algorithms to realize real-time and accurate dynamic monitoring and risk assessment of urological and male reproductive system diseases has become a key challenge for technology development. SUMMARY

[0005] To solve the above problems, the present application provides a non-invasive dynamic monitoring and risk assessment method and system that can overcome the limitations of existing technology, aiming to provide accurate and real-time dynamic monitoring of urological and male reproductive system diseases by combining advanced sensing technology and intelligent algorithms.

[0006] To achieve the above purpose, the technical solution adopted by the present application is:

[0007] In the first aspect of the present application, a risk assessment method for non-invasive dynamic monitoring of urological and male reproductive system diseases is provided, characterized by comprising the following steps:

[0008] The urine composition data in the urine is measured in real time by a portable urine analyzer, the ultrasonic image data of the urinary system and the male reproductive system is obtained by a portable ultrasonic device, and then the urine composition data and the ultrasonic image data are processed by a Z-score standardization method;

[0009] The urine composition data is processed by PCA dimension reduction to obtain a urine composition feature vector, the ultrasonic image data is processed by CNN feature extraction to obtain an ultrasonic image feature vector, and the urine composition feature vector and the ultrasonic image feature vector are fused by a weighted fusion strategy to obtain a fusion feature vector;

[0010] The fusion feature vector is extracted by a pre-trained deep learning model, and the disease features with the highest diagnostic value are extracted, the deep learning model finally outputs a disease risk score, and the disease risk score represents the risk probability of a patient suffering from a specific disease;

[0011] Patient personalized data vectors are introduced to calibrate the risk probability, and a multi-task learning framework is used to output personalized probabilities of all target diseases;

[0012] Based on the personalized probabilities of all target diseases, a personalized health intervention plan is generated, and the health data of the patient is collected in real time by an Internet of Things device to dynamically adjust the personalized health intervention plan.

[0013] Preferably, the urine composition data includes pH value, protein content and red blood cell count, wherein the urine composition data is normalized to convert the original data into standardized values, and the processing of each data includes subtracting the mean value of the data and dividing by the standard deviation, thereby obtaining the standardized data.

[0014] Preferably, the specific steps of the PCA dimension reduction processing include:

[0015] The urine composition data is projected into a new coordinate system by linear transformation, the data with the largest variance in the urine composition data is retained, a regularization term is introduced to control the correlation between features, and the data with the highest diagnostic value is ensured to be retained in the dimension reduction process.

[0016] Preferably, the specific steps of the CNN feature extraction processing include:

[0017] The ultrasonic image data is convolved at different scales by a multi-scale CNN structure, and low-level and high-level image features are gradually extracted, and finally a high-dimensional feature vector containing key image information is extracted.

[0018] Preferably, the disease risk score is a length of a vector, each dimension of which represents the probability of occurrence of a certain disease, and doctors obtain real-time health warning information of patients through the disease risk score and make personalized treatment and intervention measures for patients.

[0019] As preferred, the personalized material vector includes static items, semi-static items and dynamic physiological items;

[0020] The static items include age, gender and previously diagnosed diseases;

[0021] The semi-static items include smoking history, drinking history and occupational exposure level;

[0022] The dynamic physiological items include the average heart rate, the number of steps and the sleep duration summarized by the wearable device within 24 hours.

[0023] As preferred, the personalized material vector includes static items, semi-static items and dynamic physiological items;

[0024] The static items include age, gender and previously diagnosed diseases;

[0025] The semi-static items include smoking history, drinking history and occupational exposure level;

[0026] The dynamic physiological items include the average heart rate, the number of steps and the sleep duration summarized by the wearable device within 24 hours.

[0027] In the second aspect of the present application, a non-invasive dynamic monitoring risk assessment system for urinary and male reproductive system diseases is also provided, comprising:

[0028] A data acquisition and processing module, which measures the urine composition data in urine in real time through a portable urine analyzer, acquires ultrasonic image data of the urinary system and the male reproductive system through a portable ultrasonic device, and then processes the urine composition data and the ultrasonic image data through a Z-score standardization method;

[0029] A feature extraction and fusion module, which is used for obtaining a urine composition feature vector by performing PCA dimension reduction processing on the urine composition data, obtaining an ultrasonic image feature vector by performing CNN feature extraction on the ultrasonic image data, and fusing the urine composition feature vector and the ultrasonic image feature vector to obtain a fusion feature vector through a weighted fusion strategy;

[0030] A specific disease risk assessment module, which extracts the most diagnostic disease features by performing feature extraction on the fusion feature vector through a pre-trained deep learning model, and finally outputs a disease risk score, which represents the risk probability of a patient developing a specific disease;

[0031] A personalized disease risk assessment and prediction module is used to introduce a patient personalized data vector, calibrate the risk probability, and output personalized probabilities of all target diseases through a multi-task learning framework.

[0032] A health management intervention module is used to generate a personalized health intervention plan based on the personalized probabilities of all target diseases, and dynamically adjust the personalized health intervention plan by collecting real-time health data of the patient through Internet of Things devices.

[0033] The present application has the following advantages: the present application can automatically identify potential disease signals in a short time by collecting multi-dimensional physiological data (such as urine composition, ultrasound images, and physical data) and conducting deep analysis of these data using intelligent algorithms, and providing personalized disease risk prediction based on historical data. The core innovation of the present application is multi-modal data fusion and intelligent adaptive analysis, which enables the system to analyze and integrate the physiological state of the patient in real time, dynamically adjust the monitoring strategy, and provide personalized disease prediction results based on the health status and medical history of the patient. Compared with traditional technologies that rely on a single data source, the present application can not only solve the problem of lack of dynamic monitoring in traditional methods, but also avoid the limitations caused by excessive dependence on a single data source. More importantly, the present application introduces intelligent algorithms, which enable the system to monitor the health status of the patient based on a small amount of non-invasive data, greatly improving the accuracy and timeliness of disease prediction, and effectively avoiding the misdiagnosis problem caused by excessive reliance on clinical experience in traditional methods. Finally, the present application realizes precise and dynamic risk assessment of urinary and male reproductive system diseases, and provides a convenient, reliable, and personalized health monitoring and management platform for clinicians. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a risk assessment method flowchart of the non-invasive dynamic monitoring of urinary and male reproductive system diseases of the present application.

[0035] Figure 2 is a risk assessment system framework diagram of the non-invasive dynamic monitoring of urinary and male reproductive system diseases of the present application. DETAILED DESCRIPTION

[0036] The approximate logic of the embodiment is as follows: Step one, through the portable urine analyzer and ultrasonic equipment, the core physiological data of the urinary system and male reproductive system are collected non-invasively, and these data are standardized to eliminate the influence of equipment differences and environmental factors, providing reliable input for subsequent analysis. Step two, further feature extraction and fusion of standardized urine composition data and ultrasonic image data, effectively extracting key features through PCA and CNN, and integrating the two types of heterogeneous data into a unified feature representation through a weighted fusion strategy. Step three, through the innovative joint perception neural network (JPNN), disease-related features are extracted from the fused feature vector, combining a multi-channel input structure and a dynamic feature interaction layer to ensure full use of the potential of multi-modal data, thereby improving disease recognition accuracy. Step four, through the personalized multi-task learning model (MTL), combined with the specific physiological characteristics of the patient, further refine the disease risk assessment, and provide personalized risk scores for different diseases. Finally, step five, combined with real-time health monitoring data and personalized risk scores, a precise health intervention plan is generated through a weighted fusion algorithm, and dynamically adjusted to respond to real-time health changes of the patient.

[0037] Referring to Figure 1 In the first aspect of the present application, a non-invasive dynamic monitoring risk assessment method for urinary and male reproductive system diseases is provided, characterized in that it comprises the following steps:

[0038] Real-time measurement of urine composition data in urine by portable urine analyzer, and acquisition of ultrasonic image data of urinary system and male reproductive system by portable ultrasonic equipment, followed by Z-score standardization method to process urine composition data and ultrasonic image data;

[0039] The core physiological data related to the urinary system and male reproductive system are collected non-invasively, and these data are standardized to provide high-quality input data for subsequent data analysis and disease assessment. Focusing on two most diagnostic data sources: urine composition data and ultrasonic image data, avoiding interference of redundant information, and ensuring the accuracy and consistency of data collection. Through portable urine analyzer and portable ultrasonic equipment, these two types of data can be efficiently collected and ensure that they can provide reliable basis for subsequent steps.

[0040] The data collection process starts with a portable urine analyzer, which uses ion-selective electrode technology to measure key components such as pH, protein content, and red blood cell count in urine in real time. For example, by measuring the concentration of hydrogen ions in urine to calculate the pH value, the instrument can automatically calculate the indicators and convert the results into digital signals.

[0041] These data are fed back to the operator in real time through the built-in system, providing accurate physiological data. Next, a portable ultrasound device is used to obtain ultrasound image data of the urinary system and male reproductive system. During this process, the high-frequency sound waves emitted by the ultrasound probe penetrate the human body tissue and generate images through the reflected echoes, showing the structure of the kidneys, bladder, prostate, and testicles. To ensure the accuracy of the ultrasound image data, the device can automatically optimize the frequency and probe depth according to the needs of different scanning areas, ensuring clear images and effective identification of lesions.

[0042] In the data standardization processing stage, it is necessary to standardize the urine composition data to eliminate the influence of device differences, sampling environment changes, and other factors on the data, thereby ensuring the consistency and comparability of data from different sources. For urine composition data, especially red blood cell count and PH value, the Z-score standardization method is used for processing.

[0043] Specifically, first, each item of urine composition data (e.g., red blood cell count and PH value) is normalized to convert the original data into standardized values. For each data, first calculate the mean and standard deviation of the data in multiple samples. Then, by subtracting the mean of the data from the original data and dividing by its standard deviation, the standardized data is obtained. This standardization process eliminates measurement differences caused by different devices or environmental factors, making the data consistent under different devices and sampling conditions.

[0044] Through this method, the data of red blood cell count and PH value are converted into standardized data with zero mean and unit standard deviation, which ensures the uniformity of urine composition data and ensures accurate comparison and processing in subsequent analysis.

[0045] For ultrasound image data, Gaussian filter denoising is used, which can effectively remove noise in the image and preserve important structural features. In addition, after histogram equalization processing, the contrast of the image is enhanced, making the key areas in the image clearer and more prominent. This processing method ensures the quality of the image and ensures accurate identification and extraction of important structural information in the subsequent analysis process, especially in the disease feature extraction stage.

[0046] The standardized urine composition data and ultrasound image data will be transmitted to the subsequent data analysis platform through a unified data interface, ensuring the accuracy and integrity of the data transmission process. All data will be integrated in a unified format, facilitating subsequent processing and analysis. Throughout the process, all devices (such as urine analyzers and ultrasound devices) will be automatically calibrated to ensure the accuracy of the measurements. The working parameters of the devices (such as frequency, sampling rate) will be automatically adjusted according to the individual characteristics of the patient (such as body type, age, gender), ensuring consistency and accuracy in each data collection.

[0047] Finally, the output of this step is the standardized urine composition data and ultrasound image data. After Z-score standardization of the urine composition data, the standardized red blood cell count and PH value are obtained ( and ), which will serve as input for subsequent analysis, ensuring data consistency and comparability. After denoising and enhancement processing of the ultrasound image, clear image data is output for subsequent disease pattern recognition and risk assessment.

[0048] The urine composition data is processed by PCA dimensionality reduction to obtain the urine composition feature vector, and the ultrasound image data is processed by CNN feature extraction to obtain the ultrasound image feature vector. The urine composition feature vector and the ultrasound image feature vector are fused by a weighted fusion strategy to obtain a fused feature vector.

[0049] The standardized urine composition data and ultrasound image data output in the first step are further processed and fused to ensure that the most valuable features are extracted from the two different modalities of data and converted into a unified feature representation, facilitating subsequent disease pattern recognition and dynamic risk assessment.

[0050] First, the urine composition data is processed. The urine composition data after Z-score standardization in the previous step has eliminated the effects of device differences and environmental changes, but further extraction of the most critical component features for disease risk prediction is still needed.

[0051] To better represent the potential relationships in the urine composition data, this embodiment uses Principal Component Analysis (PCA) to extract the most diagnostic features, with the goal of projecting the original data into a new coordinate system through linear transformation, retaining the components with the largest variance in the data. To avoid the loss of useful information due to correlation between features, a regularization term is introduced based on PCA, which controls the correlation between features to ensure that the most diagnostic components are retained during dimensionality reduction. The specific formula is:

[0052]

[0053] where, is the standardized urine composition data, is the PCA eigenvector matrix, is the dimension-reduced urine composition eigenvector, is the regularization coefficient, represents the L1 norm of (i.e., sparsity), by controlling the L1 norm, the regularization term can limit the correlation between features, ensuring that the most diagnostic components are preserved during dimension reduction. In this way, the most representative features in the urine composition data can be efficiently extracted, thereby improving the accuracy and robustness of subsequent analysis.

[0054] Next, the ultrasound image data is processed. The processing of image data mainly relies on convolutional neural networks (CNN), which can automatically extract spatial features from images, especially in identifying structural changes in organs such as kidneys, bladders, and prostates. In the patent scenario, ultrasound images may be disturbed by various factors such as image blur, noise, and resolution differences, so a multi-scale CNN structure is designed that can perform convolution operations on images at different scales to extract more levels of features. Specifically, the input image will go through multiple convolutional layers and pooling layers to gradually extract low-level and high-level image features, and finally obtain a high-dimensional feature vector containing key information of the image .

[0055] In order to enable CNN to efficiently process different ultrasound image data and ensure the effectiveness of feature extraction, residual connections are added to the network structure, which can solve the gradient vanishing problem commonly encountered in deep networks and improve the training effect and stability of the model. After training, the CNN outputs a high-dimensional feature vector that contains all the important structural information in the image for subsequent disease recognition and prediction.

[0056] After the urine composition data and ultrasound image data are respectively processed by PCA dimension reduction and CNN feature extraction, the next step is to effectively fuse these two heterogeneous data sources. In order to maximize the advantages of each data source, a weighted fusion strategy is designed to fuse the feature vectors of urine composition data and image data into a unified feature representation.

[0057] First, the urine composition data and the ultrasound image data are respectively reduced in dimension by PCA and extracted by CNN to obtain feature vectors, wherein the feature vectors of the urine composition data contain key information in the urine, such as PH value, number of red blood cells, etc., and the feature vectors of the ultrasound image data extract important information reflecting the structures of the urinary system and male reproductive system in the image. In order to fuse the data from the two sources, the embodiment of the application adopts a weighted summation method, and different weight coefficients are assigned to each feature vector according to the importance of each data source in the specific task.

[0058] In the weighted fusion process, the feature vectors of the urine composition data and the feature vectors of the ultrasound image data are respectively multiplied by their respective weight coefficients, which are determined by the task requirements or the importance of the data source. By manually adjusting these weight coefficients, the contribution of each data can be optimized in the fusion process, so that the final fused feature vector can maximize the retention and integration of key information from the two data sources. Specifically, this weighting strategy can help ensure that the fused feature vector makes full use of all important information in the urine composition and image data according to the needs of the actual application scenario.

[0059] Finally, the output of this step is the fused feature vector which contains all the important features extracted from the urine composition data and the ultrasound image data. These features will be used as input in subsequent steps (such as disease pattern recognition, dynamic risk assessment and health intervention) to ensure accurate and real-time assessment of patient health risks.

[0060] The pre-trained deep learning model is used to extract the most diagnostic disease features from the fused feature vector, and the deep learning model finally outputs a disease risk score, which represents the probability of a patient developing a specific disease;

[0061] The most diagnostic disease features are extracted from the feature vector fused from the previous step, and a self-designed deep learning model is used for pattern learning. This process is crucial in this patent because it not only needs to extract potential disease signals from urine composition and ultrasound image data, but also needs to fully consider the relevance between multi-modal data and the diagnostic characteristics of different modal data. Traditional deep learning models, such as multi-layer perceptron (MLP) or convolutional neural network (CNN), are common and effective, but due to the special nature of the patent scene, using traditional models may not be able to fully utilize the feature correlation between the two types of data sources (urine composition data and ultrasound image data).

[0062] Therefore, a joint perception network (JPNN) is proposed, which is designed for the fusion and feature learning of multi-modal data, has an innovative design and structure, and can better extract and fuse key information from different data sources.

[0063] In this step, the joint perception network (JPNN) is used to learn the exclusive representation of two types of information (urine composition and ultrasound image) in the same vector space, and then dynamically fuse them to complete disease pattern learning. The specific structure includes:

[0064] Gated weighted input layer:

[0065] The network first generates a soft mask through a 32-unit gated layer (Kaiming initialization, Sigmoid activation) , and then performs adaptive re-weighting in the same vector space:

[0066]

[0067] By learning through a fully connected layer, it is ensured that the mask corresponds to the input one by one;

[0068] represents element-wise multiplication, ensuring that each feature is weighted according to its importance in different tasks.

[0069] Dynamic feature interaction layer:

[0070] On the weighted features , the network designs a dynamic feature interaction layer to dynamically adjust the degree of information interaction between different modal features.

[0071] This layer uses a dynamic weighting mechanism to weight and fuse features of different modalities into a final joint feature representation. Specifically, an adaptive weight coefficient is used, which automatically adjusts the contribution of each data modality to the final feature representation based on . The generation method of this weight coefficient is:

[0072]

[0073] where:

[0074] is a Sigmoid activation function to ensure .

[0075] and is the parameter obtained through training, which is used to learn the most appropriate weighting strategy. The resulting weighted joint features will be used for subsequent deep learning processing.

[0076] Joint learning and regularization:

[0077] After the fusion of the features , the network performs disease risk prediction through a deep learning module. To enhance the generalization ability of the model, an adaptive regularization mechanism is adopted, which dynamically adjusts the strength of regularization according to the contribution of different modal data.

[0078] Specifically, the regularization term is assigned to the features of urine composition and ultrasound image data, respectively using different regularization coefficients and . The final loss function includes cross-entropy loss and two regularization terms, as follows:

[0079]

[0080] where:

[0081] is the cross-entropy loss function, which measures the gap between the predicted risk and the actual disease occurrence.

[0082] and are the regularization coefficients of urine composition data and ultrasound image data, respectively, to prevent overfitting.

[0083] and are the network weights corresponding to the two modal data.

[0084] Through this adaptive regularization, the regularization strength can be dynamically adjusted according to the contribution of urine composition and ultrasound image in the task, thereby improving the accuracy of the model and preventing overfitting.

[0085] Disease risk score output:

[0086] After deep learning and feature fusion, the network outputs a disease risk score , which represents the probability of a patient developing a specific disease (such as kidney stones, prostate disease, etc.). This score integrates biomarkers in urine composition data and organ shape information in ultrasound images, ensuring the accuracy and comprehensiveness of disease risk prediction.

[0087]

[0088] where is a length vector, each dimension represents the probability of occurrence of a certain disease. Through this vector, doctors can obtain real-time health warnings for patients and develop personalized treatment and intervention measures for them.

[0089] Finally, the output of this step is a disease risk score vector .

[0090] Introduce the patient's personalized information vector, calibrate the risk probability, and output the personalized probability of all target diseases through a multi-task learning framework;

[0091] Based on the disease risk vector output in the previous step, introduce the patient's personalized information vector , make accurate calibration for each disease risk, and output the personalized probability of all target diseases at once with the help of a multi-task learning framework. The core task is to adjust the "group-level" prediction to "individual-level" conclusion.

[0092] Input:

[0093] - The basic risk probability output by the previous JPNN, arranged in the order of diseases;

[0094] - The patient's personalized information vector, with clear and traceable sources:

[0095] Static items: automatically pull age (integer years), gender (0 / 1), and previously diagnosed diseases (multiple hot encoding) from electronic medical records;

[0096] Semi-static items: collect smoking / drinking history (one-hot encoding) and occupational exposure level (5-level label) from pre-admission questionnaires;

[0097] Dynamic physiological items: 24-hour average heart rate, step count, and sleep duration (standardized and embedded into a uniform length vector) from wearable devices.

[0098] The model first concatenates and to form an input pair , and then extracts common features through a shared fully connected layer (128 units, ReLU). Each disease branch then contains a 32-unit perceptron layer, which outputs a calibrated probability normalized by Softmax. Personalized adjustment is completed through a set of learnable weight matrices , and the overall formula is as follows:

[0099]

[0100] ​ mapping personalized features to disease dimensions;

[0101] " denotes concatenation; " denotes concatenation;

[0102] are shared layer posterior to output layer weights and biases.

[0103] Training employs task-weighted cross-entropy with uniform regularization. Weight coefficients are automatically tuned by validation set to ensure common diseases are not overlooked due to sample imbalance.

[0104] The only output vector whose first dimension is the final risk probability of the first urology / andriatry disease for the current patient, combined with its personalized information.

[0105] Based on the personalized probabilities of all target diseases, generate a personalized health intervention plan for the patient, and dynamically adjust the personalized health intervention plan by collecting the patient's health data in real time through Internet of Things devices;

[0106] In this step, the goal is to provide personalized health intervention plans for each patient based on the disease risk scores generated in the previous step, and achieve health management through real-time feedback and dynamic adjustment. According to the disease risk scores generated in the previous step, the system will generate a personalized health intervention plan for each patient.

[0107] This process relies on accurate risk assessment and continuously adjusts intervention strategies by combining with real-time health data such as blood sugar, heart rate, weight, etc., so as to ensure that each patient can obtain the most suitable health management measures for themselves. In this way, the system not only reduces the occurrence of diseases, but also provides personalized preventive suggestions and treatment measures for patients.

[0108] First, the physiological state of each patient is continuously tracked through real-time data collection and health monitoring. Through Internet of Things devices such as smart watches, blood sugar monitors, smart blood pressure meters, etc., the health data of patients is collected in real time and transmitted to the health management platform. The system monitors the health changes of patients in real time, such as monitoring high heart rate or abnormal blood sugar level, and immediately reminds the patient to take appropriate action (such as adjusting diet, increasing exercise, etc.). These real-time data will be combined with the disease risk scores in step 4 to form a comprehensive health assessment, and the health management plan will be dynamically adjusted based on this.

[0109] The data uplink keeps two-way synchronization: one is the individualized disease risk vector The other is real-time physiological indicators aggregated through the hospital's IoT gateway. The latter is automatically pushed from multiple devices: a wrist-worn monitor provides 60-second average heart rate and step count; a continuous glucose sensor uploads a 5-minute blood glucose curve; a home blood pressure terminal writes systolic / diastolic blood pressure after the patient's morning and evening measurements; a Bluetooth scale records morning weight; and the sleep tracking provides a nighttime HRV overview from a mattress pressure sensor array. The gateway standardizes timestamps, units (mmHg, mg / dL, etc.), performs Z-score analysis, and then... Perform Min-Max compression to [0,1] to match... Units must be consistent; then write to the cache to ensure consistency. The indicators in relation to Synchronized updates.

[0110] The intervention engine relies on two sparse mapping matrices: and . The columns correspond to urological / andrological diseases, and the rows correspond to five major intervention categories (diet, exercise, sleep rhythm, water intake, and follow-up examination). The initial values ​​were marked as 0–1 by urologists and nutritionists using the Delphi method (e.g., "kidney stones" was weighted 0.85 for "dietary intervention"), and then fine-tuned using L1-regular linear regression on retrospective medical records. The establishment process is similar: first, the public health and sports medicine team provides physiological indicators → initial values ​​of intervention weights (such as high heart rate mainly pointing to exercise intervention 0.75), then real-time streaming data is used for correlation screening and iterative optimization is performed through group sparsity constraints.

[0111] The fusion formula remains unchanged. First, the two types of information are projected onto the same intervention space, and then Sigmoid compression is performed to (0,1):

[0112]

[0113] get The five dimensions represent the recommendation strengths for diet, exercise, sleep, hydration, and follow-up, respectively. For example, a 35-year-old male with a stone risk of 0.78, low nocturnal HRV, and a morning weight gain of 0.9 kg would have the following formula output: The system prioritizes diet and hydration, while moderate exercise, sleep monitoring, and follow-up are temporarily suspended.

[0114] Intervention recommendations are driven by intensity thresholds: ≥ 0.85 triggers "immediate execution" actions (generate daily low purine meal plan or 8 daily hydration reminders), 0.60-0.85 generates "scheduled execution" tasks (3x30min brisk walks this week), 0.30-0.60 only observational cues, <0.30 enters background weekly reporting. All recommendations are packaged into JSON instructions after fine-tuning with patient contraindications and past completion rates, and delivered to the mobile app and wearable device with fields containing target values, reminder frequency and review dates.

[0115] Wherein the intensity threshold is usually initially set by an expert group (such as urologists, nutritionists, etc.) based on clinical experience.

[0116] The present application continuously monitors device data during the execution period and recalculates the risk score every five minutes If the specific gravity of urine returns to normal and the heart rate at night decreases, the hydration and exercise intensity coefficients automatically decrease; on the contrary, if the blood pressure rises again, the relevant intervention dimensions are increased. Such a closed-loop cycle not only traces the source of the matrix weight, but also ensures that the intervention strength is dynamically adaptive to the patient's state, meeting the dual requirements of reproducibility and clinical interpretability of the patent.

[0117] The output of this step is a health intervention plan generated based on personalized risk assessment and real-time health data. This plan includes detailed health recommendations (such as diet, exercise, and examination plans) and intervention measures. The health intervention plan is pushed to the patient in real time and feedback is provided through intelligent devices. The system dynamically updates the intervention plan to ensure timely adjustments to the treatment plan according to changes in the patient's health, thereby improving the accuracy and effectiveness of health intervention. Through this innovative health management system, patients can receive continuous and personalized health monitoring and intervention, reducing the incidence of disease and improving the quality of life.

[0118] Through the design of this step, the disease risk score is effectively combined with real-time health data to create a dynamic and personalized health management system. The system ensures that each patient's health management plan can be flexibly adjusted according to their physiological data and risk score through real-time feedback, weighted fusion algorithm, and health intervention model, thereby providing accurate and effective health intervention. This plan not only meets the needs of personalized medicine, but also significantly improves the patient's health management experience and reduces the risk of disease occurrence.

[0119] Please refer to Figure 2 In the second aspect of the present application, a risk assessment system for non-invasive dynamic monitoring of the urinary and male reproductive system diseases is also provided, comprising:

[0120] The data acquisition and processing module measures the urine composition data in urine in real time through a portable urine analyzer, acquires ultrasonic image data of the urinary system and male reproductive system through a portable ultrasonic device, and then processes the urine composition data and the ultrasonic image data through a Z-score standardization method;

[0121] The feature extraction and fusion module is configured to obtain a urine composition feature vector by performing PCA dimension reduction processing on the urine composition data, obtain an ultrasonic image feature vector by performing CNN feature extraction on the ultrasonic image data, and fuse the urine composition feature vector and the ultrasonic image feature vector to obtain a fusion feature vector through a weighted fusion strategy;

[0122] The specific disease risk assessment module is configured to extract disease features with the highest diagnostic value by performing feature extraction on the fusion feature vector through a pre-trained deep learning model, and finally output a disease risk score, which represents the risk probability of a patient suffering from a specific disease, through the deep learning model.

[0123] The personalized disease risk assessment and prediction module is configured to introduce a patient personalized information vector, calibrate the risk probability, and output personalized probabilities of all target diseases through a multi-task learning framework.

[0124] The health management intervention module is configured to generate a personalized health intervention plan based on the personalized probabilities of all target diseases, and dynamically adjust the personalized health intervention plan by collecting health data of the patient in real time through an Internet of Things device.

[0125] The above embodiments only describe the preferred embodiments of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by ordinary engineering technicians in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. A risk assessment method for non-invasive dynamic monitoring of diseases of the urinary and male reproductive systems, characterized in that, Includes the following steps: Urine composition data in urine is measured in real time using a portable urine analyzer, and ultrasound image data of the urinary system and male reproductive system are acquired using a portable ultrasound device. The urine composition data and ultrasound image data are then processed using the Z-score normalization method. The urine component data is processed by PCA dimensionality reduction to obtain a urine component feature vector, and the ultrasound image data is processed by CNN feature extraction to obtain an ultrasound image feature vector. The urine component feature vector and the ultrasound image feature vector are fused by a weighted fusion strategy to obtain a fused feature vector. The fused feature vector is processed by a pre-trained deep learning model to extract the most diagnostically valuable disease features. The deep learning model then outputs a disease risk score, which represents the probability that a patient will develop a specific disease. By introducing patient-specific data vectors, the risk probabilities are calibrated, and a multi-task learning framework is used to output the personalized probabilities of all target diseases. Based on the individualized probabilities of all target diseases, a personalized health intervention plan is generated, and the patient's health data is collected in real time through IoT devices to dynamically adjust the personalized health intervention plan.

2. The risk assessment method for non-invasive dynamic monitoring of urinary and male reproductive system diseases according to claim 1, characterized in that, The urine composition data includes pH value, protein content, and red blood cell count. The urine composition data is normalized to convert the raw data into standardized values. The processing of each data item includes subtracting the mean of the data from the raw data and dividing by its standard deviation to obtain the standardized data.

3. The risk assessment method for non-invasive dynamic monitoring of urinary and male reproductive system diseases according to claim 1, characterized in that, The specific steps of the PCA dimensionality reduction process include: By linearly transforming the urine composition data and projecting it into a new coordinate system, the data with the largest variance in the urine composition data is retained. A regularization term is introduced to control the correlation between features, ensuring that the most diagnostically valuable data is retained during the dimensionality reduction process.

4. The risk assessment method for non-invasive dynamic monitoring of urinary and male reproductive system diseases according to claim 1, characterized in that, The specific steps of the CNN feature extraction process include: By using a multi-scale CNN structure to perform convolution operations on the ultrasound image data at different scales, low-level and high-level image features are gradually extracted, and finally a high-dimensional feature vector containing key image information is obtained.

5. The risk assessment method for non-invasive dynamic monitoring of urinary and male reproductive system diseases according to claim 1, characterized in that, The disease risk score is a length of The vector represents the probability of a certain disease occurring in each dimension. Doctors can obtain real-time health warning information for patients through the disease risk score and formulate personalized treatment and intervention measures for patients.

6. The risk assessment method for non-invasive dynamic monitoring of urinary and male reproductive system diseases according to claim 1, characterized in that, The personalized data vector includes static terms, semi-static terms, and dynamic physiological terms; The static items include age, gender, and previously diagnosed diseases; The semi-static items include smoking history, alcohol consumption history, and occupational exposure level; The dynamic physiological parameters include average heart rate, steps, and sleep duration summarized by the wearable device over 24 hours.

7. The risk assessment method for non-invasive dynamic monitoring of urinary and male reproductive system diseases according to claim 1, characterized in that, The personalized health intervention program includes health advice and intervention measures; The intervention is driven by a preset intensity threshold: When the intensity threshold is greater than or equal to 0.85, an immediate action is triggered, generating a daily low-purine diet or eight daily water intake reminders. When the intensity threshold is in the range of 0.60–0.85, a task plan is generated to require the patient to walk briskly for 30 minutes at least 3 times a week. When the intensity threshold is in the range of 0.30–0.60, only observational indications are given; When the intensity threshold is less than 0.30, only the data is saved; All intervention measures, after being fine-tuned based on patient contraindications and previous completion rates, are packaged into JSON instructions and sent to mobile applications and wearable devices. The fields include target values, reminder frequency, and review dates.

8. A risk assessment system for non-invasive dynamic monitoring of diseases of the urinary and male reproductive systems, characterized in that, include: The data acquisition and processing module measures urine composition data in real time using a portable urine analyzer and acquires ultrasound image data of the urinary system and male reproductive system using a portable ultrasound device. The urine composition data and ultrasound image data are then processed using the Z-score standardization method. The feature extraction and fusion module is used to obtain a urine component feature vector by performing PCA dimensionality reduction on the urine component data, to obtain an ultrasound image feature vector by performing CNN feature extraction on the ultrasound image data, and to fuse the urine component feature vector and the ultrasound image feature vector by performing a weighted fusion strategy to obtain a fused feature vector. The specific disease risk assessment module uses a pre-trained deep learning model to extract features from the fused feature vector, extracting the disease features with the most diagnostic value. The deep learning model ultimately outputs a disease risk score, which represents the probability that a patient will develop a specific disease. The personalized disease risk assessment and prediction module is used to introduce patient-specific data vectors, calibrate the risk probabilities, and output the personalized probabilities of all target diseases through a multi-task learning framework. The health management intervention module is used to generate personalized health intervention plans based on the personalized probabilities of all target diseases, and to dynamically adjust the personalized health intervention plans by collecting patients' health data in real time through IoT devices.

Images