Kidney function evaluation method and system and storage medium

Abstract

The invention relates to the technical field of medical evaluation, and discloses a kidney function evaluation method and system and a storage medium. The method comprises the following steps: acquiring medical detection data of a detected object, extracting sample data, and determining key indexes related to renal functions; inputting the sample data into a machine learning model, outputting a predicted value of the key index, and defining the detection object of which the predicted value does not meet a preset range as a risk object; the method comprises the following steps: acquiring and identifying target areas of various medical image data, extracting key areas from the target areas, calculating an ability score of each key area, and screening effective areas from the key areas based on the ability scores; acquiring a target kidney image of the risk object, dividing the target kidney image into a plurality of image blocks, identifying a key area of each image block, defining the key area as a first area, matching the first area with the effective area, and evaluating the kidney lesion degree of the risk object based on a matching score. The accuracy of kidney function evaluation is improved.

Description

Technical Field

[0001] The present application relates to the field of medical assessment technology, and in particular to a method, system, and storage medium for assessing renal function. Background Art

[0002] As a crucial excretory and regulatory organ in the human body, the kidney's functional status plays a vital role in maintaining a stable internal environment. Accurately assessing kidney function is of vital clinical significance for early diagnosis, disease monitoring, treatment efficacy assessment, and prognosis of kidney disease.

[0003] Existing methods for assessing renal function, such as the Chinese patent application with publication number CN118280566A, propose a method for assessing renal function for non-medical diagnosis. First, the test data of renal patients are collected, and feature and label columns are added to the data. The characteristic values ​​and label values ​​of the renal patient data are processed, and the characteristic values ​​and label values ​​are converted into data types. A neural network model is then used to train the model based on the processed data. Finally, based on the patient's real-time test data, the trained model is used to predict the type and degree of lesions. For example, the Chinese patent application with publication number CN114159026A proposes a method for evaluating renal microcirculation function, which includes: determination of renal microcirculation function parameters, preprocessing of renal microcirculation function data, and visualization of renal microcirculation function. By constructing a three-dimensional visualization model, a renal microcirculation function bubble diagram, and wavelet transform spectrum analysis, the renal microcirculation function is deeply analyzed from a holistic level.

[0004] However, although the above-mentioned existing technologies have solved the drawback of traditional renal function assessment that over-relies on the doctor's experience and judgment, there are still problems of insufficient assessment accuracy and low efficiency. Therefore, a method for renal function assessment is needed to achieve accurate and efficient assessment of the patient's renal function and provide more reliable support for clinical diagnosis and treatment. Summary of the Invention

[0005] The present application provides a method, system and storage medium for evaluating kidney function, which are used to more accurately evaluate kidney function.

[0006] In a first aspect, the present application provides a method for assessing renal function, the method comprising: Step S1: obtaining medical examination data of multiple test subjects within a historical period, determining key indicators related to kidney function based on the medical examination data, and extracting sample data from the medical examination data; Step S2: Establishing a machine learning model, inputting the sample data into the machine learning model, outputting predicted values ​​of key indicators of the current detection object in the future time period, and defining the current detection object whose predicted value does not meet the preset range as a risk object; Step S3: acquiring medical imaging data containing different renal function states, identifying multiple target regions in the medical imaging data, labeling the target regions, extracting key regions from the target regions, calculating the ability score of each key region for distinguishing normal types from different lesion types, and screening out effective regions from all key regions based on the ability score; Step S4: Obtain the target kidney image of the risk subject, divide the target kidney image into multiple image blocks, identify the key area of ​​each image block, define it as the first area, match the first area with the existing valid area, calculate the matching score of the first area, and evaluate the degree of kidney lesions in the risk subject based on the matching score.

[0007] In conjunction with the first aspect, in a first implementation of the first aspect of the present application, extracting sample data from medical test data includes: The sample data includes positive sample data and negative sample data, and the medical examination data includes health examination data, medication data, and lifestyle data. The key indicator is determined from the health examination data, and the measurement value of the key indicator of the test subject in a historical time period is obtained. The change amount of the corresponding key indicator in the historical time period is calculated. If the change amount shows a downward trend and the change amount is greater than a first threshold, the time series data consisting of the measurement value of the key indicator of the test subject in the historical time period and the corresponding change amount is used as positive sample data; otherwise, it is used as negative sample data. The medication data is divided into positive sample data and negative sample data based on the medication effect, and it is determined whether the lifestyle data contains life characteristics that cause the key indicators to decline. The lifestyle data is divided into positive sample data and negative sample data based on the proportion of the life characteristics. The health check data, the medication data, and the positive sample data corresponding to the lifestyle data are correlated, and the negative sample data are correlated with each other to obtain the final positive sample data and negative sample data.

[0008] In combination with the first aspect, in a second implementation of the first aspect of the present application, before identifying the multiple target areas in the medical image data, the method includes: The medical imaging data is divided into multiple sub-regions, and a target image sequence of each sub-region is obtained, wherein the target image sequence is composed of first images of the sub-region taken at multiple angles. A segmentation model is established, and the target image sequence of each sub-region is input into the segmentation model as training data. The segmentation model outputs the type probability of each pixel in each first image. If the type probability is greater than a second threshold, it is preliminarily determined that the pixel belongs to the corresponding region type, and the region type includes a kidney region type and a non-kidney region type. The number of first images of each pixel in the target image sequence that belongs to the region type is counted, which is defined as a first number. If the first number is greater than a third threshold, it is determined that the pixel belongs to the region type. All pixels belonging to the kidney region type constitute a kidney region, and the target region is identified based on the kidney region.

[0009] In combination with the first aspect, in a third implementation of the first aspect of the present application, extracting the key area from the target area includes: An image of the target area is defined as a second image, Gaussian smoothing is performed on the second image, and the difference between Gaussian smoothed images of different scales is calculated to obtain a Gaussian difference image. Key points are screened out from the target area based on the Gaussian difference image. A preset surrounding area is divided into multiple area blocks with each key point as the center. For each area block, gradient histograms in multiple directions are calculated. The sum of all gradient amplitudes is calculated based on the gradient histogram and defined as a first eigenvalue. A multidimensional feature vector representing the area block is generated based on the first eigenvalues ​​in all directions corresponding to each area block. The key area is constructed based on the key points and the corresponding multidimensional feature vectors.

[0010] In combination with the first aspect, in a fourth implementation of the first aspect of the present application, selecting key points from the target area based on the Gaussian difference image includes: In the Gaussian difference image, each pixel is compared with other pixels in a preset surrounding range to determine whether each pixel is an extreme point. If so, it is defined as a candidate point, and the edge points in the target area and the points in the Gaussian difference image that are less than a third threshold are defined as unstable points. The unstable points are deleted from the candidate points, and the remaining candidate points are defined as key points.

[0011] In combination with the first aspect, in a fifth implementation of the first aspect of the present application, calculating the ability score of each key area for distinguishing between normal types and different lesion types includes: Obtain the number of times the key area appears in the medical imaging data corresponding to each category and the total number of times it appears in all medical imaging data, which are defined as the first number and the second number respectively, define the ratio of the first number to the second number as the first probability that the key area belongs to each category, define the ratio of the number of images of the medical imaging data of each category to the number of images of all medical imaging data as the category probability of the medical imaging data of each category, calculate for each category the ratio of the first probability that the key area belongs to the category to the corresponding category probability, take the logarithm to obtain the first numerical value, and then add the product of the first probabilities corresponding to all analogies and the first numerical value to obtain the ability score of the key area.

[0012] In combination with the first aspect, in a sixth implementation of the first aspect of the present application, screening out valid areas from all the key areas based on the capability scores includes: Calculate the effective value of the key area, where the effective value is the difference between twice the product of the first value and the first probability of the corresponding category and the ability score of the corresponding category. If the effective value is greater than a fourth threshold, define the corresponding key area as a valid area.

[0013] In conjunction with the first aspect, in a seventh implementation of the first aspect of the present application, calculating the matching score of the first region includes: Calculate the similarity between the first region and each valid region in the database. If the similarity is greater than a fifth threshold, the match is deemed successful. Count the total number of the first regions and the number of successful matches between each first region and the valid region of each target category, which are defined as the third number and the fourth number. Define the ratio of the third number to the fourth number as the matching score of the first region matching the corresponding target category, where the target category is different lesion categories.

[0014] In a second aspect, the present application provides a system for evaluating renal function, the system comprising: a collection module, configured to obtain medical test data of multiple test subjects within a historical period, determine key indicators related to kidney function based on the medical test data, and extract sample data from the medical test data; A prediction module is used to establish a machine learning model, input the sample data into the machine learning model, output the predicted values ​​of the key indicators of the current detection object in the future time period, and define the current detection object whose predicted value does not meet the preset range as a risk object; a screening module configured to acquire medical imaging data containing different renal function states, identify multiple target regions in the medical imaging data, label the target regions, extract key regions from the target regions, calculate an ability score for each key region in distinguishing between normal types and different types of lesions, and screen effective regions from all key regions based on the ability scores; An evaluation module is used to obtain a target kidney image of the risk subject, divide the target kidney image into multiple image blocks, identify a key area of ​​each image block, define it as a first area, match the first area with an existing valid area, calculate a matching score for the first area, and evaluate the degree of kidney lesions in the risk subject based on the matching score.

[0015] A third aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, enable the computer to execute the above-mentioned method for assessing renal function.

[0016] Compared with the prior art, the present invention has the following beneficial effects: By leveraging the historical medical data of multiple subjects, the present invention can capture the changing trends of renal function in different individuals at different time points. By analyzing the data, the most relevant indicators for renal function (such as creatinine level and glomerular filtration rate) are identified. By analyzing changes in key indicators, renal dysfunction can be detected early, providing opportunities for timely intervention. A machine learning model can automatically predict future trends in key indicators, reducing the subjectivity and error of manual predictions. By predicting key indicators, subjects at risk for future renal dysfunction can be identified. Key region extraction can capture subtle local feature changes in images, such as the edges and texture of lesion areas. By calculating the ability scores of key regions, the key regions most useful for distinguishing normal from lesioned tissue are identified, defined as effective regions. Effective regions contain the most discriminative features and can better distinguish normal from lesioned tissue. A target kidney image of the subject at risk is obtained and divided into multiple image blocks. This allows for local assessment of different kidney regions, enabling more accurate lesion detection. By comparing the feature vectors of key regions, more precise matching is achieved, improving lesion detection accuracy. A matching score is calculated for each image block, quantifying the extent of lesions and providing more reliable data support for clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 This is a schematic diagram of an embodiment of a method for evaluating renal function in an embodiment of the present application; Figure 2 A flow chart for generating key areas in an embodiment of the present application; Figure 3 A flowchart of the screening of key points in the embodiments of this application; Figure 4 This is a schematic diagram of an embodiment of a system for evaluating renal function in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The embodiments of the present application provide a method, system and storage medium for assessing renal function. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0020] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the method for evaluating renal function in the embodiments of the present application includes: Step S1: Obtain medical examination data of multiple test subjects within a historical time period, determine key indicators related to kidney function based on the medical examination data, and extract sample data from the medical examination data.

[0021] Specifically, medical examination data on kidney function of multiple test subjects over the past three years are obtained from hospitals, clinics and other medical institutions. The medical examination data include health check data, medication data and lifestyle data. Key indicators include blood creatinine, glomerular filtration rate, blood urea nitrogen, urine protein, etc. Sample data are extracted from the medical examination data. The sample data include positive sample data and negative sample data. The extracted positive sample data is marked as 1, and the negative sample data is marked as 0.

[0022] Step S2: Establish a machine learning model, input sample data into the machine learning model, output the predicted values ​​of key indicators of the current detection object in the future time period, and define the current detection object whose predicted value does not meet the preset range as a risk object.

[0023] Specifically, a machine learning model is constructed based on the gradient boosting algorithm, and sample data is input into the machine learning model for training. The machine learning model learns the relationship between different medical test data, and the medical test data of the current test object is obtained and input into the trained machine learning model. The machine learning model obtains the predicted values ​​of key indicators and outputs the predicted values ​​of the key indicators of the current test object in the future time period, such as the predicted values ​​in the next year. The standard range of the key indicators is set. If the predicted values ​​of the key indicators are not within the standard range or significantly exceed the standard range, the current test object is judged as a risk object, where the risk object is a person with abnormal kidney function risk. However, it needs further confirmation as to which aspect of the kidney function of the risk object is abnormal or has a pathological risk.

[0024] Step S3: Obtain medical imaging data containing different renal function states, identify multiple target areas in the medical imaging data, and mark the target areas, extract key areas from the target areas, calculate the ability score of each key area for distinguishing normal types and different lesion types, and screen out effective areas from all key areas based on the ability score.

[0025] Specifically, various types of medical imaging data are collected, including normal kidney tissue images and different types of lesion images, such as CT or MRI images of kidney tumors, kidney cysts, kidney stones, glomerular sclerosis, and tubular damage. Experienced radiologists or pathologists manually annotate different target areas in the images, such as normal kidney tissue areas (renal tubules, glomeruli, etc.), tumor areas, or cyst areas, and mark the boundaries and category labels of each target area, such as "normal", "kidney tumor", "kidney cyst", etc.

[0026] Based on the scale-invariant feature transform (SIFT) algorithm, key points such as tumor edges or normal structures of renal tubules and glomeruli are extracted from medical imaging data. Key regions are formed based on the key points, and the ability score of each key region for distinguishing normal kidney tissue from different types of lesions is calculated. The larger the ability score, the more useful the key region is for distinguishing different categories. The matching value is calculated based on the ability score. The matching value is used to measure the ability of the key region to distinguish the target category, that is, the different types of lesions. The larger the matching value, the more effective the key region is for identifying various types of lesions.

[0027] Step S4: obtain the target kidney image of the risk subject, divide the target kidney image into multiple image blocks, identify the key area of ​​each image block, define it as the first area, match the first area with the existing valid area, calculate the matching score of the first area, and evaluate the degree of kidney lesions in the risk subject based on the matching score.

[0028] Specifically, a target kidney image of a risk subject to be assessed is obtained, with an image resolution of 512×512 pixels. The image is divided into rectangular image blocks of a fixed size (e.g., 64×64 pixels). The same key region extraction method as in step S3 is used to extract key regions in each image block. For example, in an image block, key points of the renal tubule edge are extracted and defined as a first region. The identified first region is matched with valid regions in a pre-constructed database, and the cosine similarity between the feature vectors of the first region and the valid regions in the database is calculated. If the similarity is greater than a preset threshold, the match is considered successful. The number of valid regions of various different types of lesions matched in each small image block is counted, and a matching score is calculated. The degree of renal lesions is classified into different levels based on the matching score. For example, a matching score greater than 2.0 is defined as a high-probability lesion, a matching score greater than 1 and less than or equal to 2.0 is defined as a suspected region, and a matching score less than or equal to 1 is defined as a normal kidney region. The present invention converts complex renal images into intuitive lesion probability maps, allowing doctors to quickly locate abnormal regions, significantly improving the early detection rate and diagnostic efficiency of kidney diseases (e.g., renal cancer and cysts).

[0029] In a specific embodiment, extracting sample data from medical test data includes the following steps: The sample data includes positive sample data and negative sample data. The medical examination data includes health examination data, medication data, and lifestyle data. Key indicators are determined from the health examination data, and the measured values ​​of the key indicators of the test subject in a historical time period are obtained. The change in the corresponding key indicators in the historical time period is calculated. If the change shows a downward trend and the change is greater than a first threshold, the time series data consisting of the measured values ​​of the key indicators of the test subject in the historical time period and the corresponding change is used as positive sample data; otherwise, it is used as negative sample data. Based on the medication effect, the medication data is divided into positive sample data and negative sample data. It is determined whether there are life characteristics in the lifestyle data that lead to a decline in key indicators. The lifestyle data is divided into positive sample data and negative sample data based on the proportion of life characteristics. The health check data, medication data and the positive sample data corresponding to the lifestyle data are correlated, and the negative sample data are correlated to obtain the final positive sample data and negative sample data.

[0030] Specifically, health check data includes name, disease, and various indicator test data; medication data includes drug name, drug dosage, medication time, and medication frequency; lifestyle data includes eating habits, exercise habits, drinking habits, and other life data. Based on the health check data, key indicators related to kidney function, such as glomerular filtration rate, blood creatinine, urine protein, etc., are determined. Each type of data in the medical test data is analyzed and sample data is extracted. The measurement values ​​of multiple key indicators, such as glomerular filtration rate, are obtained from multiple physical examinations within three years. The measurement values ​​of each key indicator at different time points are arranged in chronological order to construct time series data. The change of each key indicator between adjacent time points is calculated, and the trend of the change is analyzed. If the change shows a downward trend and is judged to be greater than a first threshold, the first threshold is set to 5 mL / min / 1.73 m², for example, the time series data of the key indicator is marked as positive sample data; otherwise, it is defined as negative sample data. By analyzing the change of key indicators, the changing trend of the patient's kidney function can be more accurately assessed, which helps to identify patients with rapidly declining eGFR, that is, people at high risk of chronic kidney disease.

[0031] Based on the medication effect, medication data is divided into positive sample data and negative sample data. The changes in key indicators (such as eGFR, blood creatinine, blood pressure, etc.) after medication are analyzed. Positive sample data refers to the absence of significant improvement or deterioration in key indicators after medication, while negative sample data refers to the presence of an improvement trend or a stable trend. During health check-up visits, lifestyle data is extracted from the lifestyle questionnaire completed by the patient. The lifestyle data is analyzed to identify lifestyle characteristics that lead to a decline in key indicators. For example, a high-salt diet may cause high blood pressure and increase the burden on the kidneys; a high-protein diet may increase the metabolic burden on the kidneys and accelerate the decline in renal function. A threshold for the proportion of lifestyle characteristics is set (for example, 50%). If the proportion of lifestyle characteristics is higher than 50%, it means that unhealthy lifestyle habits have a greater impact on key indicators. The lifestyle data is defined as positive sample data; otherwise, it is defined as negative sample data. The health check data, medication data, and lifestyle data marked as positive sample data and negative sample data are correlated separately to comprehensively analyze the impact of different types of data on kidney function.

[0032] When the predicted value of each key indicator is subsequently compared with the corresponding standard range, if the predicted values ​​of all key indicators are within the standard range, the subject is judged to be normal. The current test subject's medication data is defined as reference data, such as effective drug types, dosages, and duration of medication. The current test subject's lifestyle data is defined as reference data, such as healthy eating habits and exercise habits. Based on this reference data, combined with clinical guidelines and expert opinions, a personalized reference opinion is formed.

[0033] In a specific embodiment, the method further includes the following steps before identifying multiple target areas in medical image data: The medical imaging data is divided into multiple sub-regions, and a target image sequence of each sub-region is obtained. The target image sequence consists of first images of the sub-region taken at multiple angles. A segmentation model is established, and the target image sequence of each sub-region is input into the segmentation model as training data. The segmentation model outputs the type probability of each pixel in each first image. If the type probability is greater than a second threshold, it is preliminarily determined that the pixel belongs to the corresponding region type. The region type includes kidney region type and non-kidney region type. The number of first images of each pixel in the target image sequence that belongs to the region type is counted, which is defined as the first number. If the first number is greater than a third threshold, it is determined that the pixel belongs to the region type. All pixels belonging to the kidney region type constitute the kidney region, and the target region is identified based on the kidney region.

[0034] Specifically, according to the anatomical structure of the kidney, the medical imaging data is divided into multiple sub-regions, such as the upper region, the middle region and the lower region. It is assumed that each sub-region is photographed from 8 angles to obtain a target image sequence of each sub-region. A segmentation model is established based on a convolutional neural network. The target image sequence is used as the input of the segmentation model. A second threshold is set to 0.6. If the probability that a pixel point belongs to the kidney region type is greater than the second threshold, the pixel point is preliminarily determined to belong to the kidney region type. Assuming that a pixel point is judged to be the kidney region in 6 out of 8 angles, the first number is 6, and the third threshold is set to 4. If the first number is greater than the third threshold, it means that the pixel point belongs to the kidney region type. By acquiring image data from multiple angles, the morphology and characteristics of the kidney can be captured more comprehensively, the accuracy of kidney region recognition can be improved, and misjudgment can be reduced.

[0035] In a specific embodiment, extracting the key area from the target area specifically includes the following steps: An image of the target area is defined as a second image, and Gaussian smoothing is performed on the second image. The difference between Gaussian smoothed images of different scales is calculated to obtain a Gaussian difference image. Key points are screened out from the target area based on the Gaussian difference image. The preset surrounding area is divided into multiple area blocks with each key point as the center. For each area block, gradient histograms in multiple directions are calculated. The sum of all gradient amplitudes is calculated based on the gradient histogram and defined as the first eigenvalue. A multidimensional feature vector representing the area block is generated based on the first eigenvalues ​​of all directions corresponding to each area block. A key area is constructed based on the key points and the corresponding multidimensional feature vectors.

[0036] Specifically, if Figure 2As shown in FIG, a flowchart for generating key regions is shown. The target region segmented from the medical imaging data, such as the renal tumor region, renal cyst region, renal tubule region, normal kidney region, etc., is defined as the second image. The second image is convolved with Gaussian filters of different standard deviations to generate multiple Gaussian smoothed images of different scales. The difference between the Gaussian smoothed images of different scales is calculated to obtain a Gaussian difference image. Key points are screened out from the Gaussian difference image. These key points are feature points that are invariant to scale, rotation, and illumination changes. For a key point, the surrounding area, for example, 16×16 pixels, is divided into 16 4×4 pixel blocks. For each area, The domain block is constructed, and the gradient direction and amplitude of its 16 pixel points are calculated, and an 8-dimensional gradient histogram is constructed. The 8 gradient amplitudes in the gradient histogram of the area block are summed to obtain a first eigenvalue. The first eigenvalues ​​of the 16 area blocks are combined to generate a 128-dimensional feature vector. The key points and their multi-dimensional feature vectors are combined to construct a key area. The key area includes the position, scale, direction and corresponding alignment feature vector of each key point in the area. The key area is used to capture important information in the image, such as edges, corners, texture changes, etc., and can be used to detect and locate the area of ​​lesions in kidney images, such as capturing the boundary features of tumors.

[0037] In a specific embodiment, selecting key points from a target area based on a Gaussian difference image specifically includes the following steps: In the Gaussian difference image, each pixel is compared with other pixels in a preset surrounding range to determine whether each pixel is an extreme point. If so, it is defined as a candidate point, and the edge points in the target area and the points in the Gaussian difference image that are less than the third threshold are defined as unstable points. The unstable points are deleted from the candidate points, and the remaining candidate points are defined as key points.

[0038] Specifically, if Figure 3 The figure shows the flowchart of key point screening. In the Gaussian difference image, the values ​​of each pixel are compared with those of its 26 neighboring pixels (including pixels of adjacent scales). If the pixel is an extreme value (maximum or minimum value), it is marked as a candidate point for the key point. The edge detection algorithm (such as Canny edge detection) is used to identify edge points in the image. Points on the edge are easily affected by noise and have poor stability. Such feature points should be excluded. A third threshold (such as 0.03) is set, and points with Gaussian difference values ​​less than the threshold are defined as low-contrast points. Low-contrast points are sensitive to changes in illumination and have poor stability. Unstable points are deleted from the candidate points, and stable key points are retained. The remaining candidate points are defined as key points, such as key points at the edge of renal tumors, texture feature points inside tumors, etc. By excluding edge points and low-contrast points, unstable key point candidates can be removed and the stability of key points can be improved.

[0039] In a specific embodiment, calculating the ability score of each key region for distinguishing normal types and different types of lesions specifically includes the following steps: Obtain the number of times the key area appears in the medical imaging data corresponding to each category and the total number of times it appears in all medical imaging data, which are defined as the first number and the second number respectively, define the ratio of the first number to the second number as the first probability that the key area belongs to each category, and define the ratio of the number of images of the medical imaging data of each category to the number of images of all medical imaging data as the category probability of the medical imaging data of each category, calculate for each category the ratio of the first probability that the key area belongs to the category to the corresponding category probability, take the logarithm to obtain the first numerical value, and then add the product of the first probabilities corresponding to all analogies and the first numerical value to obtain the ability score of the key area.

[0040] Specifically, assuming that the key area The number of occurrences in different categories is: normal kidney tissue: 5 times; kidney tumor: 30 times; kidney cyst: 10 times; total number of occurrences: 50 times; the first probability of the key area belonging to each category is calculated based on the first formula, which is: ,in, For key areas The first probability of belonging to category X, For key areas The number of times it appears in the medical imaging data of category X, is the total number of data categories of all medical imaging data. Assuming that a key area is calculated, , ; The category probability of each category in all medical imaging data is calculated based on the second formula, which is: ,in, is the number of images of medical imaging data of category X, is the total number of images of all categories. Assume that there are 200 images in total, of which 50 are normal kidney tissue images, 100 are kidney tumor images, 30 are kidney cyst images, and 20 are other lesions images. , , then the capability score of each key area is calculated based on the third formula, which is: It is the sum of the first probability and the first value product of all categories. By calculating the ability score, the contribution of each key area to distinguishing different categories can be quantified. The higher the ability score, the more effective the key area is in distinguishing normal and diseased tissues.

[0041] In a specific embodiment, screening out effective areas from all key areas based on capability scores specifically includes the following steps: Calculate the effective value of the key area, where the effective value is the difference between twice the product of the first value and the first probability of the corresponding category and the ability score of the corresponding category. If the effective value is greater than the fourth threshold, the corresponding key area is defined as a valid area.

[0042] Specifically, the most effective important regions for identifying different lesion types (such as renal tumors, renal cysts, etc.) are screened out from all key regions, and the effective value of each key region is calculated based on the fourth formula, which is: , assuming the critical area For the kidney tumor category (X=kidney tumor), the first value is 0.0792, the first probability is 0.6, and the ability score is 0.03291. The key area is calculated. The valid values ​​are , all key areas and all lesion categories are processed cyclically, and effective areas that are effective for each lesion category are screened out. By calculating the effective value, the key areas that are most effective in identifying specific lesion types can be accurately screened out. By analyzing the effective value, the contribution of the key areas to specific lesion types can be explained, which can effectively improve the accuracy and efficiency of kidney disease diagnosis and provide more reliable support for clinical diagnosis and treatment.

[0043] In a specific embodiment, calculating the matching score of the first region specifically includes the following steps: Calculate the similarity between the first region and each valid region in the database. If the similarity is greater than the fifth threshold, the match is considered successful. Count the total number of first regions and the number of successful matches between each first region and the valid region of each target category, which are defined as the third number and the fourth number. The ratio of the third number to the fourth number is defined as the matching score of the first region matching the corresponding target category, where the target category is different lesion categories.

[0044] Specifically, the similarity between the feature vector of the first region and the feature vector of each valid region in the database is calculated, and the matching score of the first region to each target category is calculated based on the number of matches. The matching score represents the proportion of the first region that matches the valid region of the target category. The matching score can be used to quantify the degree of matching between the first region and the valid regions of different lesion categories. Different matching scores can be calculated for different categories to achieve the distinction between multiple lesion types.

[0045] The above describes the method for evaluating renal function in the embodiment of the present application. The following describes the system for evaluating renal function in the embodiment of the present application. Figure 4 One embodiment of the system for evaluating renal function in the present application includes: a collection module, configured to obtain medical test data of multiple test subjects within a historical period, determine key indicators related to kidney function based on the medical test data, and extract sample data from the medical test data; The prediction module is used to establish a machine learning model, input sample data into the machine learning model, output the predicted values ​​of key indicators of the current detection object in the future time period, and define the current detection object whose predicted value does not meet the preset range as a risk object; a screening module for acquiring medical imaging data containing different renal function states, identifying multiple target regions in the medical imaging data, labeling the target regions, extracting key regions from the target regions, calculating the ability score of each key region for distinguishing normal types from different lesion types, and screening effective regions from all key regions based on the ability score; An evaluation module is used to obtain a target kidney image of a risk subject, divide the target kidney image into multiple image blocks, identify a key area of ​​each image block, define it as a first area, match the first area with an existing valid area, calculate a matching score for the first area, and evaluate the degree of kidney lesions in the risk subject based on the matching score.

[0046] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to execute the steps of the method for assessing renal function.

[0047] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0048] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0049] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for evaluating renal function, characterized in that: The method comprises the following steps: Step S1: obtaining medical examination data of multiple test subjects within a historical period, determining key indicators related to kidney function based on the medical examination data, and extracting sample data from the medical examination data; Step S2: Establishing a machine learning model, inputting the sample data into the machine learning model, outputting predicted values ​​of key indicators of the current detection object in the future time period, and defining the current detection object whose predicted value does not meet the preset range as a risk object; Step S3: acquiring medical imaging data containing different renal function states, identifying multiple target regions in the medical imaging data, labeling the target regions, extracting key regions from the target regions, calculating the ability score of each key region for distinguishing normal types from different lesion types, and screening out effective regions from all key regions based on the ability score; Step S4: Obtain the target kidney image of the risk subject, divide the target kidney image into multiple image blocks, identify the key area of ​​each image block, define it as the first area, match the first area with the existing valid area, calculate the matching score of the first area, and evaluate the degree of kidney lesions in the risk subject based on the matching score.

2. The method according to claim 1, characterized in that Extracting sample data from medical test data includes the following steps: The sample data includes positive sample data and negative sample data, and the medical examination data includes health examination data, medication data, and lifestyle data. The key indicator is determined from the health examination data, and the measurement value of the key indicator of the test subject in a historical time period is obtained. The change amount of the corresponding key indicator in the historical time period is calculated. If the change amount shows a downward trend and the change amount is greater than a first threshold, the time series data consisting of the measurement value of the key indicator of the test subject in the historical time period and the corresponding change amount is used as positive sample data; otherwise, it is used as negative sample data. The medication data is divided into positive sample data and negative sample data based on the medication effect, and it is determined whether the lifestyle data contains life characteristics that cause the key indicators to decline. The lifestyle data is divided into positive sample data and negative sample data based on the proportion of the life characteristics. The health check data, the medication data, and the positive sample data corresponding to the lifestyle data are correlated, and the negative sample data are correlated with each other to obtain the final positive sample data and negative sample data.

3. The method according to claim 1, characterized in that The steps of identifying the plurality of target regions in the medical image data include: The medical imaging data is divided into multiple sub-regions, and a target image sequence of each sub-region is obtained, wherein the target image sequence is composed of first images of the sub-region taken at multiple angles. A segmentation model is established, and the target image sequence of each sub-region is input into the segmentation model as training data. The segmentation model outputs the type probability of each pixel in each first image. If the type probability is greater than a second threshold, it is preliminarily determined that the pixel belongs to the corresponding region type, and the region type includes a kidney region type and a non-kidney region type. The number of first images of each pixel in the target image sequence that belongs to the region type is counted, which is defined as a first number. If the first number is greater than a third threshold, it is determined that the pixel belongs to the region type. All pixels belonging to the kidney region type constitute a kidney region, and the target region is identified based on the kidney region.

4. The method according to claim 1, wherein Extracting the key area from the target area comprises the following steps: An image of the target area is defined as a second image, Gaussian smoothing is performed on the second image, and the difference between Gaussian smoothed images of different scales is calculated to obtain a Gaussian difference image. Key points are screened out from the target area based on the Gaussian difference image. A preset surrounding area is divided into multiple area blocks with each key point as the center. For each area block, gradient histograms in multiple directions are calculated. The sum of all gradient amplitudes is calculated based on the gradient histogram and defined as a first eigenvalue. A multidimensional feature vector representing the area block is generated based on the first eigenvalues ​​in all directions corresponding to each area block. The key area is constructed based on the key points and the corresponding multidimensional feature vectors.

5. The method according to claim 4, characterized in that Screening out key points from the target area based on the Gaussian difference image includes the following steps: In the Gaussian difference image, each pixel is compared with other pixels in a preset surrounding range to determine whether each pixel is an extreme point. If so, it is defined as a candidate point, and the edge points in the target area and the points in the Gaussian difference image that are less than a third threshold are defined as unstable points. The unstable points are deleted from the candidate points, and the remaining candidate points are defined as key points.

6. The method according to claim 1, characterized in that Calculating the ability score of each key region to distinguish normal types and different types of lesions includes the following steps: Obtain the number of times the key area appears in the medical imaging data corresponding to each category and the total number of times it appears in all medical imaging data, which are defined as the first number and the second number respectively, define the ratio of the first number to the second number as the first probability that the key area belongs to each category, define the ratio of the number of images of the medical imaging data of each category to the number of images of all medical imaging data as the category probability of the medical imaging data of each category, calculate for each category the ratio of the first probability that the key area belongs to the category to the corresponding category probability, take the logarithm to obtain the first numerical value, and then add the product of the first probabilities corresponding to all analogies and the first numerical value to obtain the ability score of the key area.

7. The method according to claim 6, characterized in that Screening out effective areas from all the key areas based on the capability scores comprises the following steps: Calculate the effective value of the key area, where the effective value is the difference between twice the product of the first value and the first probability of the corresponding category and the ability score of the corresponding category. If the effective value is greater than a fourth threshold, define the corresponding key area as a valid area.

8. The method according to claim 7, characterized in that Calculating the matching score of the first region includes the following steps: Calculate the similarity between the first region and each valid region in the database. If the similarity is greater than a fifth threshold, the match is deemed successful. Count the total number of the first regions and the number of successful matches between each first region and the valid region of each target category, which are defined as the third number and the fourth number. Define the ratio of the third number to the fourth number as the matching score of the first region matching the corresponding target category, where the target category is different lesion categories.

9. A system for evaluating renal function, for implementing the method for evaluating renal function according to any one of claims 1 to 8, characterized in that: The system comprises: a collection module, configured to obtain medical test data of multiple test subjects within a historical period, determine key indicators related to kidney function based on the medical test data, and extract sample data from the medical test data; A prediction module is used to establish a machine learning model, input the sample data into the machine learning model, output the predicted values ​​of the key indicators of the current detection object in the future time period, and define the current detection object whose predicted value does not meet the preset range as a risk object; a screening module configured to acquire medical imaging data containing different renal function states, identify multiple target regions in the medical imaging data, label the target regions, extract key regions from the target regions, calculate an ability score for each key region in distinguishing between normal types and different types of lesions, and screen effective regions from all key regions based on the ability scores; An evaluation module is used to obtain a target kidney image of the risk subject, divide the target kidney image into multiple image blocks, identify a key area of ​​each image block, define it as a first area, match the first area with an existing valid area, calculate a matching score for the first area, and evaluate the degree of kidney lesions in the risk subject based on the matching score.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the method for assessing renal function according to any one of claims 1 to 8 is implemented.

Images