Renal injury risk assessment system based on super-resolution and multi-mode ultrasound

Abstract

The invention relates to the technical field of biomedicine, and discloses a kidney injury risk assessment system based on super-resolution and multi-mode ultrasound. Comprising the following modules and components: a system deployment and initialization module, a multi-modal ultrasonic data acquisition and preprocessing module, a multi-modal feature fusion and intelligent risk assessment module, a risk early warning and clinical intervention connection module and a system dynamic iterative optimization module. The system deployment and initialization module comprises a multi-mode ultrasonic device, a super-resolution ultrasonic probe, a data synchronization interface, a body position positioning clamp, a kidney injury feature database, an RIRI index threshold initialization tool, an LSTM model parameter configuration terminal and a medical staff training manual. The invention aims to solve the problems that the existing kidney injury risk assessment is easy to omit early injury signals, the assessment comprehensiveness is insufficient, the existing assessment mostly depends on the subjective experience of doctors for qualitative judgment, a unified quantitative assessment system is lacked, and a standardized assessment conclusion is difficult to form.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, and in particular to a kidney injury risk assessment system based on super-resolution and multimodal ultrasound. Background Technology

[0002] Kidney injury risk assessment is a medical assessment method that uses the collection of morphological, functional, blood perfusion and clinical data of kidney tissue, combined with standardized assessment logic and technical means, to determine the probability of an individual developing kidney injury, the current degree of injury and the future trend of progression. Its core value lies in providing a scientific basis for clinical identification of high-risk groups, development of personalized intervention plans and monitoring of disease changes, helping to achieve early detection and precise management of kidney injury and reduce the risk of kidney function deterioration. Current kidney injury risk assessments cannot simultaneously cover multiple dimensions of information such as kidney tissue morphology, mechanical properties, and blood perfusion, leading to the easy omission of early injury signals and insufficient comprehensiveness of the assessment. At the same time, current assessments rely heavily on doctors' subjective experience for qualitative judgments, lack a unified quantitative assessment system, have highly subjective feature weighting, and have poor consistency in results among different assessors, making it difficult to form standardized assessment conclusions. Summary of the Invention

[0003] The purpose of this invention is to provide a kidney injury risk assessment system based on super-resolution and multimodal ultrasound, in order to solve the problems in the existing kidney injury risk assessment in the background art, which cannot simultaneously cover multiple dimensions of information such as kidney tissue morphology, mechanical properties, and blood perfusion, resulting in the easy omission of early injury signals and insufficient comprehensiveness of assessment. At the same time, existing assessments mostly rely on the subjective experience of doctors for qualitative judgment, lack a unified quantitative assessment system, have strong subjectivity in feature weight allocation, poor consistency of results among different assessors, and make it difficult to form standardized assessment conclusions.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A kidney injury risk assessment system based on super-resolution and multimodal ultrasound includes the following modules and components: The system includes a deployment and initialization module, a multimodal ultrasound data acquisition and preprocessing module, a multimodal feature fusion and intelligent risk assessment module, a risk warning and clinical intervention linkage module, and a system dynamic iterative optimization module. The system deployment and initialization module includes: multimodal ultrasound equipment, super-resolution ultrasound probe, data synchronization interface, body positioning fixture, kidney injury feature database, RIRI index threshold initialization tool, LSTM model parameter configuration terminal, and medical staff training manual. The multimodal ultrasound data acquisition and preprocessing module includes: a multimodal ultrasound data acquisition terminal, a super-resolution image reconstruction algorithm unit, a motion artifact removal subunit, an outlier processing subunit, a multimodal data registration subsystem, and a normalization mapping unit; The multimodal feature fusion and intelligent risk assessment module includes: a lesion morphology feature extraction unit, a functional feature extraction unit, a multimodal fusion algorithm terminal, a kidney injury risk index (RIRI) calculation unit, an LSTM risk prediction model training platform, and a risk level classification terminal; The risk warning and clinical intervention linkage module includes: a clinical terminal warning unit, a risk report automatic generation subsystem, a doctor intervention interface, and a personalized intervention plan recommendation subunit; The system's dynamic iterative optimization module includes: a patient data archiving unit, a model parameter iteration tool, a feature database update terminal, and an optimization effect verification subunit.

[0005] As a further improvement to this technical solution: the multimodal ultrasound device integrates B-mode ultrasound, elastography, and blood flow Doppler functions, providing multi-dimensional raw ultrasound data sources; the super-resolution ultrasound probe adopts micro-nano focusing technology and ultrasound signal enhancement algorithms, with an element spacing ≤50μm, combined with an ultrasound echo signal noise reduction algorithm, achieving a detection resolution ≤0.1mm, capturing small lesions of early kidney injury; the data synchronization interface conforms to the HL7FHIR medical standard, realizing real-time synchronous transmission and format unification of multimodal ultrasound data; the body positioning clamp features an adjustable ergonomic design to fix the patient's lumbar position; The database describes the characteristics of kidney injury, storing over 100,000 clinically annotated data points, including multimodal ultrasound data and clinical outcomes of normal kidney tissue and kidney injury at different stages. The RIRI index threshold initialization tool, based on clinical evidence-based data, statistically analyzes the prognostic data of 5,000 patients with kidney injury at different stages, setting risk level thresholds (extremely high, high, intermediate, and low) through ROC curve analysis. The LSTM model parameter configuration includes a terminal initialization model learning rate of 0.001, 128 hidden layer nodes, and 500 iterations. The training manual for medical personnel standardizes equipment operation procedures and data quality assessment criteria.

[0006] As a further improvement to this technical solution: the multimodal ultrasound data acquisition terminal receives the raw ultrasound signal from the super-resolution probe and converts it into digital images and quantized data; the super-resolution image reconstruction algorithm unit is based on a generative adversarial network (GAN) architecture, and the formula is: The motion artifact removal subunit identifies and removes artifact regions through image grayscale continuity analysis; the outlier processing subunit uses the 3σ principle to remove extreme abnormal data in elastic modulus and blood perfusion; the multimodal data registration subsystem extracts anatomical landmarks such as the renal hilum and renal pelvis based on a spatial coordinate calibration algorithm, establishes coordinate mapping relationships, with a mapping error ≤0.05mm, and unifies the multimodal data spatial coordinate system; the standardized mapping unit uses the Z-score normalization method, with the formula: .

[0007] As a further improvement to this technical solution: the lesion morphology feature extraction unit extracts the irregularity, area, and standard deviation of density distribution of lesion edges using edge detection and region growing algorithms; the functional feature extraction unit calculates the rate of renal tissue elastic modulus exceeding the standard and the rate of abnormal blood perfusion; the multimodal fusion algorithm terminal (entropy weight TOPSIS method): The first step is to calculate the information entropy and weights, using the following formula: ,in Let m be the entropy of the j-th feature, and m be the number of samples. The normalized value of the j-th feature of the i-th sample , Let be the weight of the j-th feature, and n be the total number of features, n=5; The second step is to calculate the overall score, using the following formula: ,in The distance from the sample to the positive ideal solution. The distance from the sample to the negative ideal solution; The formula for calculating the Kidney Injury Risk Index (RIRI) is as follows: ; The LSTM risk prediction model training platform is based on a long short-term memory network architecture, and the formula is: The formula for the time-series window adaptive adjustment mechanism is: ; The risk level classification terminal classifies risks into four levels: extremely high, high, medium, and low, based on RIRI scores and progression probability.

[0008] As a further improvement to this technical solution: the clinical terminal early warning unit activates ultrasound equipment audio-visual warnings, doctor's mobile APP push warnings, and hospital HIS system pop-up warnings based on risk level; the automatic risk report generation subsystem integrates super-resolution ultrasound images, lesion feature annotation maps, RIRI index calculation process, progression probability prediction results, and risk level determination criteria to generate standardized clinical reports; the doctor intervention interface allows doctors to view detailed patient data, modify risk levels (with reasons provided), and enter intervention plans; the personalized intervention plan recommendation subunit recommends corresponding intervention directions based on risk level and patient basic information.

[0009] As a further improvement to this technical solution: the patient data archiving unit archives patient ultrasound data, risk assessment results, clinical intervention plans, and follow-up outcomes monthly; the model parameter iteration tool aims to maximize early warning accuracy and adjusts parameters using a gradient descent method, with the formula: The feature database update terminal adds ultrasound feature data of rare kidney injury types every quarter; the optimization effect verification subunit verifies system performance through early warning accuracy, risk stratification accuracy, and detection rate of small lesions.

[0010] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention overcomes the resolution limitations of traditional equipment by using super-resolution ultrasound technology, enabling the capture of minute pathological changes in early kidney injury. Combined with multimodal data acquisition and fusion technology, it comprehensively covers multiple dimensions of information such as kidney tissue morphology, mechanical properties, and blood perfusion, avoiding the one-sidedness of assessment caused by a single data dimension. At the same time, relying on standardized data processing procedures and quantitative assessment systems, it replaces the traditional qualitative judgment based on subjective experience, effectively reducing the bias between different assessors and making the assessment conclusions more consistent and reliable, providing strong support for the timely identification of early kidney injury.

[0011] 2. This invention uses a dynamic prediction model to anticipate the future progression of kidney injury, reserving sufficient time for clinical intervention. Combined with personalized intervention recommendations and a multi-channel early warning mechanism, it improves the efficiency and targeting of medical intervention. The closed-loop iterative mechanism of the system can continuously optimize model parameters and feature database based on clinical follow-up data, constantly adapting to different types of kidney injury with different causes and progression rates. In long-term application, it can continuously improve assessment performance, providing stable and reliable technical support for the precise management and monitoring of kidney injury.

[0012] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a system diagram of a kidney injury risk assessment system based on super-resolution and multimodal ultrasound. Detailed Implementation

[0014] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0015] Please see Figure 1 In this embodiment of the invention, a kidney injury risk assessment system based on super-resolution and multimodal ultrasound includes the following modules and components: The system includes a deployment and initialization module, a multimodal ultrasound data acquisition and preprocessing module, a multimodal feature fusion and intelligent risk assessment module, a risk warning and clinical intervention linkage module, and a system dynamic iterative optimization module. The system deployment and initialization module includes: multimodal ultrasound equipment, super-resolution ultrasound probe, data synchronization interface, body positioning fixture, kidney injury feature database, RIRI index threshold initialization tool, LSTM model parameter configuration terminal, and medical staff training manual. The multimodal ultrasound data acquisition and preprocessing module includes: a multimodal ultrasound data acquisition terminal, a super-resolution image reconstruction algorithm unit, a motion artifact removal subunit, an outlier processing subunit, a multimodal data registration subsystem, and a normalization mapping unit; The multimodal feature fusion and intelligent risk assessment module includes: a lesion morphology feature extraction unit, a functional feature extraction unit, a multimodal fusion algorithm terminal, a kidney injury risk index (RIRI) calculation unit, an LSTM risk prediction model training platform, and a risk level classification terminal; The risk warning and clinical intervention linkage module includes: a clinical terminal warning unit, an automatic risk report generation subsystem, a physician intervention interface, and a personalized intervention plan recommendation subunit; The system dynamic iterative optimization module includes: a patient data archiving unit, a model parameter iteration tool, a feature database update terminal, and an optimization effect verification subunit; Specifically, the core modules of the system are: system deployment and initialization module, multimodal ultrasound data acquisition and preprocessing module, multimodal feature fusion and intelligent risk assessment module, risk warning and clinical intervention linkage module, and system dynamic iterative optimization module. Each module is connected according to the process of deployment, acquisition, assessment, intervention and iteration to form a complete assessment link. The system deployment and initialization module includes: multimodal ultrasound equipment, super-resolution ultrasound probe, data synchronization interface, body positioning fixture, kidney injury feature database, RIRI index threshold initialization tool, LSTM model parameter configuration terminal, and medical staff training manual, covering all dimensions of initialization needs for equipment, data, parameters, and personnel training. The multimodal ultrasound data acquisition and preprocessing module includes: a multimodal ultrasound data acquisition terminal, a super-resolution image reconstruction algorithm unit, a motion artifact removal subunit, an outlier processing subunit, a multimodal data registration subsystem, and a standardization mapping unit, realizing the entire process from raw data acquisition to standardized data output; The multimodal feature fusion and intelligent risk assessment module includes: a lesion morphology feature extraction unit, a functional feature extraction unit, a multimodal fusion algorithm terminal, a kidney injury risk index (RIRI) calculation unit, an LSTM risk prediction model training platform, and a risk level classification terminal. It is the core decision-making link of the system and completes feature processing and risk quantification. The risk warning and clinical intervention linkage module includes: a clinical terminal warning unit, an automatic risk report generation subsystem, a physician intervention interface, a personalized intervention plan recommendation subunit, assessment results, and the implementation path of clinical intervention; The system dynamic iterative optimization module includes: a patient data archiving unit, a model parameter iteration tool, a feature database update terminal, and an optimization effect verification subunit, enabling continuous iteration and upgrading of system performance; The multimodal ultrasound equipment integrates B-mode ultrasound, elastography, and blood flow Doppler functions, providing multi-dimensional raw ultrasound data sources; the super-resolution ultrasound probe employs micro-nano focusing technology and ultrasound signal enhancement algorithms, with an element spacing ≤50μm, combined with an ultrasound echo signal noise reduction algorithm, achieving a detection resolution ≤0.1mm, capturing small lesions in early kidney injury; the data synchronization interface conforms to the HL7FHIR medical standard, enabling real-time synchronous transmission and format uniformity of multimodal ultrasound data; the adjustable ergonomic positioning clamp stabilizes the patient's lumbar position; kidney injury characteristic database: It stores over 100,000 clinically labeled data, including multimodal ultrasound data and clinical outcomes of normal kidney tissue and different stages of kidney injury; RIRI index threshold initialization tool: based on clinical evidence-based data, it statistically analyzes the prognostic data of 5,000 patients with kidney injury at different stages, and sets risk level thresholds (extremely high, high, intermediate, and low) through ROC curve analysis; LSTM model parameter configuration: terminal initialization model learning rate 0.001, number of hidden layer nodes 128, number of iterations 500; medical staff training manual standardizes equipment operation procedures and data quality judgment standards; Specifically, the multimodal ultrasound equipment integrates three major functions: B-mode ultrasound, elastography, and blood flow Doppler. B-mode ultrasound is used to acquire morphological images of kidney tissue, elastography is used to acquire kidney tissue stiffness data, and blood flow Doppler is used to acquire images of blood flow perfusion in the kidney, together providing a multidimensional raw ultrasound data source. Super-resolution ultrasound probe: It adopts micro-nano focusing technology and ultrasound signal enhancement algorithm, with a pole spacing of ≤50μm and a pole arrangement density that is 3 times higher than that of traditional probes. Combined with ultrasound echo signal noise reduction algorithm (based on wavelet transform noise reduction), the detection resolution is ≤0.1mm, which can capture small lesions of early kidney injury (such as renal tubular dilatation and glomerular sclerosis with a diameter of <1mm). Data synchronization interface: Complies with the HL7FHIR medical data exchange standard, supports DICOM format data transmission, and realizes real-time synchronous transmission and format unification of multimodal ultrasound data (B-ultrasound images, elastic modulus data, blood perfusion data), with a synchronization delay of ≤100ms; Body positioning clamp: Adjustable ergonomic design, the waist fixation area is made of soft silicone material, the adjustment range is suitable for patients with a height of 1.5-1.9m and a weight of 40-120kg. The snap-on fixation method ensures that the patient's waist position is fixed during the examination, reducing the error of body position change; Kidney Injury Feature Database: Stores over 100,000 clinically annotated data cases, with data sources covering the nephrology departments of 20 top-tier hospitals nationwide. It includes multimodal ultrasound data of normal kidney tissue and early, intermediate, and late-stage kidney injury, and is also associated with patient clinical outcomes (such as changes in kidney function and treatment effects). The data is stored in categories according to etiology, injury stage, ultrasound features, and clinical outcome. RIRI Index Threshold Initialization Tool: Based on clinical evidence-based data, the prognostic data of 5000 patients with kidney injury at different stages (diabetic nephropathy, hypertensive nephropathy, drug-induced kidney injury, etc.) were statistically analyzed. Through ROC curve analysis (taking the score corresponding to the maximum value of the Youden index as the cutoff value), risk level cutoff values ​​were set: extremely high risk (≥80 points), high risk (60-79 points), intermediate risk (40-59 points), and low risk (<40 points). LSTM model parameter configuration terminal: Initialize the core parameters of the model, learning rate 0.001 (adaptive learning rate strategy, dynamically adjusted during training), number of hidden layer nodes 128 (including input layer, 3 hidden layers, and output layer), number of iterations 500, and configure batch size as 32 and activation function as ReLU; Training manual for medical staff: It includes equipment operation procedures (probe placement angle, amount of coupling agent applied, data acquisition time), data quality judgment standards (image clarity requirements, allowable range of artifacts), and common troubleshooting guidelines. The manual uses a combination of text and illustrations to facilitate quick learning for ultrasound medical staff.

[0016] The multimodal ultrasound data acquisition terminal receives the raw ultrasound signal from the super-resolution probe and converts it into digital images and quantized data; the super-resolution image reconstruction algorithm unit is based on a generative adversarial network (GAN) architecture, and the formula is: The motion artifact removal subunit identifies and removes artifact regions through image grayscale continuity analysis; the outlier processing subunit uses the 3σ principle to remove extreme abnormal data in elastic modulus and blood perfusion; the multimodal data registration subsystem extracts anatomical landmarks such as the renal hilum and renal pelvis based on a spatial coordinate calibration algorithm, establishes coordinate mapping relationships, with a mapping error ≤0.05mm, and unifies the spatial coordinate system of multimodal data; the standardized mapping unit uses the Z-score normalization method, with the formula: ; Specifically, the multimodal ultrasound data acquisition terminal uses a high-performance FPGA chip to receive the original ultrasound analog signal transmitted by the super-resolution probe and convert it into digital images and quantized data through analog-to-digital conversion (sampling rate ≥100MHz). The image resolution is 1024×768 pixels and the quantized data accuracy is 16 bits. Super-resolution image reconstruction algorithm unit: Based on the Generative Adversarial Network (GAN) architecture, it consists of a generator (containing 5 convolutional layers and 2 deconvolutional layers) and a discriminator (containing 4 convolutional layers and 1 fully connected layer), with the following formula: Formula annotation: This is a high-quality image after super-resolution reconstruction. The image is a low-resolution raw image acquired by the probe. The optimal parameters (including the weight matrix and bias terms of the generator and discriminator) are obtained during model training. During the reconstruction process, the discriminator is used to determine the similarity between the generated image and the high-resolution labeled image, and the final image is generated through iterative optimization. Motion artifact removal subunit: Through image grayscale value continuity analysis, a grayscale value difference threshold of 30% is set between adjacent frames. When the grayscale value difference of a certain area exceeds the threshold, it is determined to be an artifact area. Neighborhood interpolation is used to repair the artifact area and identify and remove the artifact area. Outlier processing subunit: Using the 3σ principle (based on normal distribution characteristics), the mean μ and standard deviation σ of elastic modulus and blood perfusion data are first calculated. Data exceeding the range of μ±3σ are identified as extreme outliers. The outliers are replaced with adjacent normal data using linear interpolation to eliminate extreme outliers in elastic modulus and blood perfusion. Multimodal data registration subsystem: Based on the spatial coordinate calibration algorithm, it extracts anatomical landmarks such as renal hilum, renal pelvis, and lower pole of the kidney from three types of images (no less than 3 landmarks are extracted from each type of image). The coordinate mapping relationship is established by the least squares method, with a mapping error ≤0.05mm. This achieves the unification of the spatial coordinate system of B-ultrasound images, elastic modulus distribution maps, and blood perfusion maps, and unifies the spatial coordinate system of multimodal data. Standardized mapping unit: The Z-score normalization method is used, and the formula is: Formula annotation: X represents the original data (such as lesion area, elastic modulus, perfusion volume). This is the global mean of this type of data. The global standard deviation for this type of data is used to transform data of different dimensions into standardized data with a mean of 0 and a standard deviation of 1 through normalization, thereby eliminating the difference in dimensions.

[0017] The lesion morphology feature extraction unit extracts the irregularity, area, and standard deviation of density distribution of lesion edges using edge detection and region growing algorithms; the functional feature extraction unit calculates the rate of renal tissue elastic modulus exceeding the standard and the rate of abnormal blood perfusion; the multimodal fusion algorithm terminal (entropy weight TOPSIS method) is used. The first step is to calculate the information entropy and weights, using the following formula: ,in Let m be the entropy of the j-th feature, and m be the number of samples. The normalized value of the j-th feature of the i-th sample , Let be the weight of the j-th feature, and n be the total number of features, n=5; The second step is to calculate the overall score, using the following formula: ,in The distance from the sample to the positive ideal solution. The distance from the sample to the negative ideal solution; The formula for calculating the Kidney Injury Risk Index (RIRI) is as follows: ; LSTM risk prediction model training platform: based on Long Short-Term Memory network architecture, the formula is: The formula for the time-series window adaptive adjustment mechanism is: ; Risk level classification terminal: Based on RIRI score and progression probability, it is divided into four levels: extremely risky, high risky, medium risky, and low risk. Specifically, the lesion morphology feature extraction unit: uses the Canny edge detection algorithm (threshold range 0.1-0.3) to extract the lesion edge, and uses a region growing algorithm (growing threshold of 10) to determine the lesion range, and then calculates the lesion edge irregularity (actual number of points on the lesion edge / equivalent number of points on the circular edge) and the lesion area (converted to actual area by pixel count × pixel spacing, unit mm). 2 The three types of morphological features are extracted using edge detection algorithms and region growing algorithms: density distribution standard deviation (standard deviation of gray values ​​within the lesion area), ... Functional feature extraction unit: Set the threshold for abnormal elastic modulus to 30 kPa (clinically recognized critical value for abnormal renal tissue stiffness), calculate the rate of excessive elastic modulus (area of ​​region with elastic modulus ≥ 30 kPa / total area of ​​renal cortex); set the threshold for abnormal blood perfusion to 5 ml / min・100 g (lower limit of normal renal blood perfusion), calculate the rate of abnormal blood perfusion (area of ​​region with perfusion volume < 5 ml / min・100 g / total area of ​​renal cortex), and calculate two types of functional indicators of renal tissue; Multimodal fusion algorithm terminal (entropy weight TOPSIS method): Step 1: Calculate information entropy and weights. Information entropy formula: ,in Let be the entropy of the j-th feature (the smaller the entropy value, the higher the discrimination), and m be the number of samples (the number of matching samples taken from the kidney injury feature database). The normalized value of the j-th feature of the i-th sample (Sum and normalize all sample data under the j-th feature). Let j be the weight of the j-th feature (the sum of weights is 1), and n be the total number of features (n=5, including 3 morphological features and 2 functional features). Weights are automatically assigned using information entropy; features with higher discriminative power have larger weights. The second step is to calculate the comprehensive score using the formula: ,in Let be the Euclidean distance from the i-th sample to the positive ideal solution (the positive ideal solution is the maximum value of each feature). The Euclidean distance from the i-th sample to the negative ideal solution (the negative ideal solution is the minimum value of each feature) is given. The overall score ranges from 0 to 1, with the closer to 1 being the sample closer to the ideal state. Kidney Injury Risk Index (RIRI) Calculation Unit: Formula annotation: : Lesion feature matching degree (0-1), the similarity between the current lesion feature and the late-stage lesion feature in the database, calculated by the cosine similarity of the feature vectors; Blood perfusion abnormality rate (0-1), directly output by the functional feature extraction unit; The rate of excess elastic modulus (0-1) is directly output by the functional feature extraction unit. Clinical indicator correlation (0-1): The correlation strength between ultrasound features and patient renal function indicators (serum creatinine, blood urea nitrogen), calculated using the Pearson correlation coefficient; Medical history weight (0-1): 0.8-1.0 for those with a high-risk history of kidney injury (diabetes, hypertension, history of use of nephrotoxic drugs), and 0.1-0.3 for those without. LSTM Risk Prediction Model Training Platform: Based on a Long Short-Term Memory (LSTM) network architecture, the input layer receives a comprehensive feature vector, the hidden layer processes temporal information through gating units (input gate, forget gate, output gate), and the output layer outputs probability values ​​through a sigmoid activation function. Core formula: Formula annotation: The probability of kidney injury progression in the next t months (t=1,2,3) (0-1), a probability ≥0.7 indicates a high risk of progression; : These are the comprehensive feature vectors acquired during ultrasound examinations at times t-2, t-1, and t, respectively; W: Model weight matrix (including input-hidden layer, hidden layer-hidden layer, and hidden layer-output layer weights); b: Model bias term (including biases for each layer); The formula for the time-series window adaptive adjustment mechanism is: Formula annotation: The volatility of lesion characteristics (standard deviation / mean of characteristic values ​​from the last 3 examinations) reflects the rate of change of lesions. The greater the volatility, the larger the time window T (not exceeding 60 days) to ensure the predictive sensitivity for rapidly progressive lesions. Risk level classification terminal: Based on the dual dimensions of RIRI score and progression probability, the risk level is classified into four levels: extremely high (RIRI ≥ 80 or progression probability ≥ 0.7), high (60 ≤ RIRI < 80 and 0.5 ≤ progression probability < 0.7), medium (40 ≤ RIRI < 60 and 0.3 ≤ progression probability < 0.5), and low (RIRI < 40 and progression probability < 0.3).

[0018] The clinical terminal early warning unit activates ultrasound equipment audio-visual warnings, doctor's mobile APP push warnings, and hospital HIS system pop-up warnings based on risk level; the risk report automatic generation subsystem integrates super-resolution ultrasound images, lesion feature annotation maps, RIRI index calculation process, progression probability prediction results, and risk level determination criteria to generate standardized clinical reports; the doctor intervention interface allows doctors to view detailed patient data, modify risk levels (with reasons required), and enter intervention plans; the personalized intervention plan recommendation subunit recommends corresponding intervention directions based on risk level and patient basic information. Specifically, the clinical terminal early warning unit triggers multi-channel early warnings based on risk level. For extremely high / high-risk patients, a three-level early warning is activated: a red audio-visual warning on the ultrasound equipment panel (sound intensity ≥60dB, light flashing frequency 2 times / second), a pop-up warning via the doctor's mobile APP (including patient name, risk level, and key ultrasound features), and a pop-up warning in the hospital's HIS system (linked to the patient's electronic medical record). The automatic risk report generation subsystem integrates super-resolution ultrasound images (including lesion annotation boxes), lesion feature annotation maps (annotation edge irregularity, area, etc.), RIRI index calculation process (detailed scores for each parameter), progression probability prediction results, and risk level determination criteria to generate a standardized PDF report that conforms to clinical standards. The report includes four modules: basic patient information, examination information, assessment results, and intervention recommendations. Doctor Intervention Interface: Supports doctors to log in to the system with an account and password (hierarchical access: chief physicians can modify risk levels, attending physicians can view data and enter intervention plans), view detailed patient data (original ultrasound images, feature calculation process, historical assessment records), and fill in the reason for modification when modifying the risk level (e.g., the patient has severe diabetes and is adjusted to critically ill). When entering intervention plans, doctors can choose a preset plan template or customize the input. Personalized intervention plan recommendation sub-unit: Based on risk level and patient basic information (etiology, age, comorbidities, liver and kidney function), the corresponding intervention directions are recommended as follows: For extremely high-risk patients, emergency nephrology consultation, renal biopsy, and intensive treatment are recommended; for high-risk patients, adjustment of medications for underlying diseases (such as antihypertensive drugs and hypoglycemic drugs) and follow-up ultrasound examination after 1 month are recommended; for intermediate-risk patients, follow-up ultrasound examination after 3 months and lifestyle intervention (low-salt diet, regular exercise) are recommended; for low-risk patients, routine follow-up ultrasound examination and health management follow-up are recommended after 6 months.

[0019] The patient data archiving unit archives patient ultrasound data, risk assessment results, clinical intervention plans, and follow-up outcomes monthly; the model parameter iteration tool aims to maximize early warning accuracy, using a gradient descent method to adjust parameters, with the formula as follows: The feature database update terminal adds ultrasound feature data of rare kidney injury types every quarter; the optimization effect verification subunit verifies system performance through early warning accuracy, risk stratification accuracy, and detection rate of small lesions. Specifically, the patient data archiving unit archives patients' ultrasound data (raw images, standardized data, feature data), risk assessment results (RIRI score, risk level, probability of progression), clinical intervention plans (medication adjustments, examination arrangements), and follow-up outcomes (changes in renal function indicators, damage progression / improvement) monthly. The data storage adopts a distributed database and supports multi-dimensional retrieval by patient ID, examination date, etiology, etc. Model parameter iteration tool: With the goal of maximizing early warning accuracy, the parameters are adjusted using stochastic gradient descent (SGD), as shown in the formula: Formula annotation: Parameters to be optimized (including the weights / biases of the LSTM model and the coefficients of each index of the RIRI index). : The optimized new parameters; : Old parameters before optimization; Learning rate (dynamically adjusted from 0.0001 to 0.001, adaptively adjusted according to the rate of decrease of the loss function); : The gradient of the loss function, which is defined as the warning error rate (false alarm rate and false alarm rate). Feature database update terminal: Add ultrasound feature data for rare kidney injury types (such as hereditary kidney disease and infectious kidney disease) every quarter. The new data must be labeled and reviewed by 3 nephrology experts to ensure the accuracy of the data. At the same time, update the parameters of the feature extraction algorithm (such as edge detection threshold and region growth threshold) to adapt to the new lesion types. The optimization effect verification sub-unit verifies system performance through three core indicators: early warning accuracy (number of true positive cases, number of true negative cases / total number of cases), risk stratification accuracy (number of cases matching clinical outcomes at each risk level / total number of cases at that level), and detection rate of small lesions (number of correctly detected lesions with a diameter <1mm / total number of cases of that type). The verification data are taken from real clinical cases in the past month to ensure that the system performance does not decline after optimization.

[0020] The method of use and working principle of this invention are as follows: Usage: Before use, system deployment and initialization must be completed, relevant equipment and databases must be set up, risk level thresholds and initial model parameters must be set, and operation training should be conducted for medical staff. In daily application, the patient's position is fixed by body positioning, and relevant data of kidney tissue are collected using multimodal ultrasound equipment. The system automatically preprocesses and standardizes the data, and then completes feature extraction, fusion and risk quantification assessment to generate risk level and progression prediction results. Medical staff can view detailed assessment basis through risk reports, formulate and enter clinical intervention plans based on recommended intervention directions, and trigger multi-channel warnings for extremely high or high-risk patients to prompt priority treatment. The system regularly archives patient assessment data, intervention plans and follow-up outcomes, and continuously iterates and optimizes model parameters and feature databases to ensure the stability and adaptability of assessment performance.

[0021] Working Principle: Operating within a closed-loop logic of acquisition, processing, assessment, intervention, and iteration, the system relies on multi-module collaboration to achieve accurate risk assessment of kidney injury. First, multimodal ultrasound equipment and super-resolution probes acquire data related to kidney tissue morphology, function, and blood flow. The preprocessing module performs image reconstruction, artifact removal, data registration, and standardization to eliminate data interference and dimensional differences. Next, the feature extraction unit extracts lesion morphology and functional features, and the fusion algorithm terminal performs multimodal feature fusion according to objective weights. Then, the risk index calculation and prediction model outputs quantitative risk values ​​and progression probabilities, classifying risk levels. Subsequently, the early warning and intervention module transforms the assessment results into clinically usable reports and intervention recommendations, bridging the gap between assessment and clinical operation. Finally, the dynamic iteration module uses clinical follow-up data as feedback to continuously optimize model parameters and the feature database, ensuring that the system's assessment capabilities continuously improve with clinical application.

[0022] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the description above. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, using the disclosed technical content, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A kidney injury risk assessment system based on super-resolution and multimodal ultrasound, characterized in that, Includes the following modules and components: The system includes a deployment and initialization module, a multimodal ultrasound data acquisition and preprocessing module, a multimodal feature fusion and intelligent risk assessment module, a risk warning and clinical intervention linkage module, and a system dynamic iterative optimization module. The system deployment and initialization module includes: multimodal ultrasound equipment, super-resolution ultrasound probe, data synchronization interface, body positioning fixture, kidney injury feature database, RIRI index threshold initialization tool, LSTM model parameter configuration terminal, and medical staff training manual. The multimodal ultrasound data acquisition and preprocessing module includes: a multimodal ultrasound data acquisition terminal, a super-resolution image reconstruction algorithm unit, a motion artifact removal subunit, an outlier processing subunit, a multimodal data registration subsystem, and a normalization mapping unit; The multimodal feature fusion and intelligent risk assessment module includes: a lesion morphology feature extraction unit, a functional feature extraction unit, a multimodal fusion algorithm terminal, a kidney injury risk index (RIRI) calculation unit, an LSTM risk prediction model training platform, and a risk level classification terminal; The risk warning and clinical intervention linkage module includes: a clinical terminal warning unit, a risk report automatic generation subsystem, a doctor intervention interface, and a personalized intervention plan recommendation subunit; The system's dynamic iterative optimization module includes: a patient data archiving unit, a model parameter iteration tool, a feature database update terminal, and an optimization effect verification subunit.

2. The kidney injury risk assessment system based on super-resolution and multimodal ultrasound according to claim 1, characterized in that, The multimodal ultrasound device integrates B-mode ultrasound, elastography, and blood flow Doppler functions, providing multi-dimensional raw ultrasound data sources. The super-resolution ultrasound probe employs micro-nano focusing technology and ultrasound signal enhancement algorithms, with an element spacing ≤50μm. Combined with an ultrasound echo signal noise reduction algorithm, the detection resolution is ≤0.1mm, capturing small lesions in early kidney injury. The data synchronization interface conforms to the HL7FHIR medical standard, enabling real-time synchronous transmission and format uniformity of multimodal ultrasound data. The adjustable ergonomic positioning clamp stabilizes the patient's lumbar position. The kidney injury characteristic data... The database stores over 100,000 clinically labeled data, including multimodal ultrasound data and clinical outcomes of normal kidney tissue and kidney injury at different stages; the RIRI index threshold initialization tool: based on clinical evidence-based data, it statistically analyzes the prognostic data of 5,000 patients with kidney injury at different stages, and sets risk level thresholds (extremely high, high, intermediate, and low) through ROC curve analysis; the LSTM model parameter configuration terminal initializes the model learning rate to 0.001, the number of hidden layer nodes to 128, and the number of iterations to 500; the medical staff training manual standardizes equipment operation procedures and data quality judgment criteria.

3. A kidney injury risk assessment system based on super-resolution and multimodal ultrasound according to claim 2, characterized in that, The multimodal ultrasound data acquisition terminal receives the raw ultrasound signal from the super-resolution probe and converts it into digital images and quantized data; the super-resolution image reconstruction algorithm unit is based on a generative adversarial network (GAN) architecture, and the formula is: The motion artifact removal subunit identifies and removes artifact regions through image grayscale continuity analysis; the outlier processing subunit uses the 3σ principle to remove extreme abnormal data in elastic modulus and blood perfusion; the multimodal data registration subsystem extracts anatomical landmarks such as the renal hilum and renal pelvis based on a spatial coordinate calibration algorithm, establishes coordinate mapping relationships, with a mapping error ≤0.05mm, and unifies the multimodal data spatial coordinate system; the standardized mapping unit uses the Z-score normalization method, with the formula: .

4. A kidney injury risk assessment system based on super-resolution and multimodal ultrasound according to claim 1, characterized in that, The lesion morphology feature extraction unit extracts the irregularity, area, and density distribution standard deviation of lesion edges using edge detection and region growing algorithms; the functional feature extraction unit calculates the rate of renal tissue elastic modulus exceeding the standard and the rate of abnormal blood perfusion; the multimodal fusion algorithm terminal (entropy weight TOPSIS method) is used. The first step is to calculate the information entropy and weights, using the following formula: ,in Let m be the entropy of the j-th feature, and m be the number of samples. The normalized value of the j-th feature of the i-th sample , Let be the weight of the j-th feature, and n be the total number of features, n=5; The second step is to calculate the overall score, using the following formula: ,in The distance from the sample to the positive ideal solution. The distance from the sample to the negative ideal solution; The formula for calculating the Kidney Injury Risk Index (RIRI) is as follows: ; The LSTM risk prediction model training platform is based on a long short-term memory network architecture, and the formula is: The formula for the time-series window adaptive adjustment mechanism is: ; The risk level classification terminal classifies risks into four levels: extremely high, high, medium, and low, based on RIRI scores and progression probability.

5. A kidney injury risk assessment system based on super-resolution and multimodal ultrasound according to claim 1, characterized in that, The clinical terminal early warning unit activates ultrasound equipment audio-visual warnings, doctor's mobile APP push warnings, and hospital HIS system pop-up warnings based on risk levels; the risk report automatic generation subsystem integrates super-resolution ultrasound images, lesion feature annotation maps, RIRI index calculation process, progression probability prediction results, and risk level determination criteria to generate standardized clinical reports; the doctor intervention interface allows doctors to view detailed patient data, modify risk levels (with reasons provided), and enter intervention plans. The personalized intervention recommendation subunit recommends corresponding intervention directions based on risk level and patient basic information.

6. A kidney injury risk assessment system based on super-resolution and multimodal ultrasound according to claim 1, characterized in that, The patient data archiving unit archives patient ultrasound data, risk assessment results, clinical intervention plans, and follow-up outcomes monthly; the model parameter iteration tool aims to maximize early warning accuracy and adjusts parameters using a gradient descent method, with the following formula: The feature database update terminal adds ultrasound feature data of rare kidney injury types every quarter. The optimization effect verification subunit verifies the system performance by measuring the accuracy of early warning, the accuracy of risk stratification, and the detection rate of minute lesions.

Images