Intelligent detection system for screening intima-media membrane in cerebral apoplexy

Abstract

The invention belongs to the technical field of medical imaging and diagnosis, and discloses a cerebral apoplexy screening intima-media membrane intelligent detection system which comprises a data acquisition module used for acquiring two-dimensional B ultrasonic image data, color Doppler image data and blood flow velocity waveform data; the data processing module is used for processing the two-dimensional B-mode ultrasound image data and the color Doppler image data to obtain carotid artery imaging data and extracting a carotid artery boundary and a carotid artery intima-media membrane area from the carotid artery imaging data; processing the carotid intima-media area to obtain carotid intima-media thickness data; the blood flow analysis module is used for extracting blood flow velocity and direction information from the color Doppler image data and identifying the blood flow velocity and direction information to obtain abnormal blood flow data; processing the blood flow velocity waveform data to obtain hemodynamic parameter data; through fusion analysis of multi-dimensional data, the cerebral apoplexy risk of the patient can be evaluated more comprehensively, and the treatment effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging and diagnosis technology, and more specifically, to an intelligent detection system for intima-media membrane of cerebral stroke screening. Background Art

[0002] Patent publication number CN1 15666405A9 discloses a stroke examination system (500) for examining a subject for signs of stroke. The stroke examination system (500) comprises: an acquisition unit (101) for obtaining personal information related to the brain disease of the subject; a determination unit (102) for determining the priority of each of a plurality of examination items related to stroke based on the obtained personal information; a plurality of examination units (103), each of the plurality of examination units (103) performing an examination of each of the plurality of examination items, and the plurality of examination units (103) performing the examination of the plurality of examination items in a descending order of the determined priorities; and a diagnosis unit (104) for outputting the diagnosis information related to the signs of stroke of the subject based on the examination results.

[0003] The existing intima-media detection system lacks an effective edge detection method, which may lead to inaccurate positioning of the carotid artery boundary, thus affecting the diagnosis of the disease and the formulation of treatment plans; the failure to use active contour models and threshold segmentation algorithms may lead to inefficient boundary detection, requiring more manual intervention, and increasing diagnosis time and costs;

[0004] The lack of comprehensive analysis of carotid intima-media thickness data, abnormal blood flow data, and hemodynamic parameter data sets may lead to an incomplete assessment of patients' stroke risk and ignore the combined impact of multiple factors. Due to the lack of effective risk prediction and intervention measures, the quality of life of patients will be affected. Failure to consider optimizing the weight coefficient results in the model being unable to fully utilize the information in the data set, thereby reducing the accuracy of the prediction, leading to delays in patient diagnosis and treatment, and affecting the treatment effect.

[0005] In view of this, the present invention proposes an intelligent detection system of the intima-media membrane for stroke screening to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an intelligent detection system for intima-media membrane for stroke screening, comprising:

[0007] A data acquisition module, used for acquiring two-dimensional B-ultrasound image data, color Doppler image data and blood flow velocity waveform data;

[0008] A data processing module is used to extract the carotid artery boundary and the carotid intima-media region from the two-dimensional B-ultrasound image data; and process the extracted carotid intima-media region to obtain carotid intima-media thickness data;

[0009] A blood flow analysis module is used to extract blood flow velocity and blood flow direction from color Doppler image data, identify blood flow velocity and blood flow direction to obtain abnormal blood flow data, and process blood flow velocity waveform data to obtain a hemodynamic parameter data set;

[0010] The model training module is used to fuse the carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data sets to obtain a comprehensive feature data set, build a stroke risk prediction model, and input the comprehensive feature data set into the stroke risk prediction model to predict the patient's stroke risk;

[0011] The risk warning module is used to compare the predicted patient's stroke risk with the preset patient's stroke risk threshold to determine whether the patient has a stroke risk;

[0012] The prevention and control module formulates corresponding control plans and preventive measures for patients to reduce the risk if there is a risk of stroke. If there is no risk of stroke, a health maintenance plan is formulated to maintain a healthy state. Each module is connected by wired and / or wireless means to achieve data transmission between modules.

[0013] Furthermore, the two-dimensional B-ultrasound image data includes ultrasonic pixel data, carotid artery structure data and arterial wall plaque feature data; the color Doppler image data includes blood flow velocity and blood flow direction; the blood flow velocity waveform data includes systolic peak velocity, end-diastolic velocity, pulse wave velocity and pulse index.

[0014] Furthermore, the method for extracting the carotid artery boundary and the carotid intima-media region from the two-dimensional B-ultrasound image data comprises:

[0015] The two-dimensional B-ultrasound image data is preprocessed and the image noise is removed by a Gaussian filter. The Gaussian filter expression is: S (x,y)=I(x,y)*G σ (x, y); where I(x, y) is the gray value of the original two-dimensional B-ultrasound image at point (x, y); G σ is a Gaussian kernel with standard deviation σ;

[0016] The contrast is enhanced by a linear transformation formula, which is: S′ (x,y)=a·I S (x,y)+b; where I S′(x, y) is the gray value of the two-dimensional B-ultrasound image at point (x, y) after linear transformation; a and b are parameters for adjusting the contrast;

[0017] The Sobel operator is used to calculate the gradient amplitude and direction of the two-dimensional B-ultrasound image, and the non-maximum suppression and double threshold technology are used to determine the edge of the two-dimensional B-ultrasound image; the active contour model is used to accurately locate the carotid artery boundary; the active contour model is:

[0018] Where E is the energy function of the model; C(s) is the active contour curve; C′(s) is the first-order derivative of the active contour curve; C"(s) is the second-order derivative of the active contour curve; s is the curve parameter; α and β are weight parameters that control the smoothness and rigidity of the active contour; E te To attract the external energy of the contour to move toward the edge;

[0019] A threshold segmentation algorithm is applied on the carotid artery boundary between the intima and the media to separate the carotid artery intima-media region from the entire carotid artery region of the two-dimensional B-ultrasound image; the threshold segmentation algorithm is: Among them, Q is the selected segmentation threshold; I B (x, y) is the gray value of the segmented 2D B-ultrasound image at point (x, y);

[0020] Through the threshold segmentation algorithm, I B The (x,y) value of 0 is the entire carotid artery area, I B The (x,y) value of 1 is the carotid intima-media region.

[0021] Furthermore, the method of processing the carotid intima-media region to obtain carotid intima-media thickness data includes:

[0022] Find the starting and ending positions of the carotid artery on the two-dimensional B-ultrasound image; the carotid artery is divided into the internal and external carotid arteries at the carotid bifurcation; use the ultrasonic equipment measurement tool to measure the carotid intima-media thickness from the inner wall to the outer wall of the carotid artery; select the proximal and distal ends of the carotid bifurcation for three measurements each, and then take the average value of the carotid intima-media thickness as the carotid intima-media thickness.

[0023] Furthermore, the method of extracting blood flow velocity and blood flow direction from color Doppler image data, identifying blood flow velocity and blood flow direction data to obtain abnormal blood flow data, and processing blood flow velocity waveform data to obtain a hemodynamic parameter data set includes:

[0024] The color Doppler image noise is removed by Gaussian filter, the blood flow area in the color Doppler image is extracted by threshold segmentation algorithm, the color features are extracted from the blood flow area, and converted into quantifiable values; the color features are represented by RGB color space, red and blue are used to represent the blood flow direction features, and the brightness or saturation of the color is represented as the blood flow velocity features; the mean and standard deviation of each feature are calculated, and the 3σ method is used to detect abnormal blood flow data in the blood flow data:

[0025] The mean of each feature is: Where R(i″) is each data in each feature; n′ is the total number of features; calculate the standard deviation Among them, μ is the mean of each feature data;

[0026] The normal blood flow threshold range is preset to (μ-3σ′, μ+3σ′); check whether each characteristic data falls within the normal blood flow threshold range. If a certain characteristic data exceeds the normal blood flow threshold range, it is determined to be abnormal blood flow data;

[0027] Calculate the systolic peak velocity, end-diastolic velocity, average blood flow velocity, pulse wave velocity and pulse index in the blood flow velocity waveform data; the systolic peak velocity is: PSV = max(V(t)); where PSV is the systolic peak velocity; V(t) is the blood flow velocity at time t; max(V(t)) is the maximum value of the blood flow velocity in one cardiac cycle; t is the time required to calculate the blood flow velocity in one cardiac cycle;

[0028] End-diastolic velocity: EDV = V(t ED ), where EDV is the end-diastolic velocity; t ED It is the time point at the end of diastole of a cardiac cycle;

[0029] The average value of the blood flow velocity in one cardiac cycle is obtained according to the peak systolic velocity and the end-diastolic velocity. The calculation formula of the average velocity is: Where, T′ is the duration of a cardiac cycle; It is the integral of blood flow velocity in one cardiac cycle;

[0030] The pulse wave speed is: Where D is the distance the pulse wave travels; Δt is the time difference for the pulse wave to travel from one location to another;

[0031] The pulse index is calculated by the average of the peak systolic velocity, the end-diastolic velocity and the blood flow velocity in one cardiac cycle. The pulse index is: The calculated peak systolic velocity, end-diastolic velocity, average blood flow velocity, pulse wave velocity and pulse index were combined to obtain a hemodynamic parameter data set.

[0032] Furthermore, the method of fusing the carotid intima-media thickness data, the abnormal blood flow data and the hemodynamic parameter data set to obtain a comprehensive feature data set includes:

[0033] Identify outliers in carotid intima-media thickness data, abnormal blood flow data, and hemodynamic parameter data sets; preset a carotid intima-media thickness data threshold, and mark data exceeding the carotid intima-media thickness data threshold as outliers; calculate the standard deviation of abnormal blood flow data, and mark data that deviates from the mean value by more than the standard deviation as outliers; identify outliers in hemodynamic parameter data using a density clustering algorithm;

[0034] The identified outliers are eliminated to obtain the processed intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set; the obtained intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set are subjected to standard deviation normalization and converted into a standard normal distribution to eliminate the dimensional influence between the data, and finally the normalized intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set are obtained;

[0035] The carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data set are fused through a weighted formula to form a comprehensive feature data set Z, where the carotid intima-media thickness data is recorded as L, the abnormal blood flow data is recorded as F, and the hemodynamic parameter data set is recorded as H, and the weight coefficient of each feature data set in the weighted formula is designed;

[0036] The weighted formula is: Among them, ω L is the weight coefficient of the carotid intima-media thickness data; ω F is the weight coefficient of abnormal blood flow data; ω H is the weight coefficient of the hemodynamic parameter data set.

[0037] Furthermore, the method for designing the weight coefficient of each feature data set in the weighted formula includes:

[0038] A random initial population is generated, and each individual in the population contains a set of randomly generated weight coefficients [ω L ,ω F ,ω H ], satisfying the constraint that the sum of weight coefficients is 1;

[0039] The initialization population is: Among them, i is each individual in the population; N is the number of individuals in the population;

[0040] The performance of each individual is evaluated by the fitness function, and the fitness function is: FKS = α′·KU(Z); where KU(Z) is an indicator for measuring the inter-class discrimination of the comprehensive feature data set Z; α′ is the weight factor of KU(Z) in the fitness function;

[0041] The stroke risk prediction model constructed by default predicts that the patients' stroke risk is divided into two categories: high-risk category C1 and low-risk category C2;

[0042] The inter-class distinction is: Among them, D B (C1, C2) is the distance between the high-risk category C1 and the low-risk category C2; D W (C1) is the average distance from the internal data points of the high-risk category C1 to its category center point; D W (C2) is the average distance from the internal data points of the low-risk category C2 to its category center point;

[0043] Calculate the sum of all individual fitness values ​​in the population: Among them, FKS i is the fitness value of the i-th individual;

[0044] For each generation of individuals in the population, m individuals are randomly selected and their probabilities of being selected for reproduction are calculated using a probability model; the probability model is: Among them, FP J is the probability of the Jth individual being selected;

[0045] The method for randomly selecting m individuals includes: presetting a random selection probability threshold r in the interval [0,1], selecting individuals according to the random selection probability threshold r; calculating the probability of each individual being selected and the probability of all individuals before the individual being selected, and finding an individual whose selection probability is greater than or equal to the random selection probability threshold r;

[0046] Assume that the two parent individuals are and Perform crossover operations on the selected individuals to generate new individuals and Mutate the weight coefficient of each new individual: ω′ L =ω L +Δω L ; Among them, ω′ L is the weight coefficient after mutation; Δω L is the weight coefficient ω L Adjustment factor; ω′ F =ω F +Δω F ; Among them, ω′ Fis the weight coefficient after mutation; Δω F is the weight coefficient ω F Adjustment factor; ω′ H =1-ω′ L -ω′ F ; Among them, ω′ H is the weight coefficient after mutation;

[0047] Construct an adjustment factor restriction function to limit the range of the weight coefficient adjustment factor; this can prevent excessive random changes from destroying the excellent characteristics of the individual;

[0048] The adjustment factor limiting function is: Δω(o) = Δω0·e- λ·o ; Among them, Δω(o) is the adjustment factor of the weight coefficient when the number of random selections is O; Δω0 is the initial weight coefficient adjustment factor; λ is a positive attenuation coefficient that controls the speed at which the weight coefficient adjustment factor decreases; O is the current number of random selections;

[0049] Define the upper limit of λ as λ max , the lower limit of λ is λ min , limit λ∈(λ min ,λ max ), thereby further limiting the range of weight coefficient adjustment factors;

[0050] A larger λ value means that the weight coefficient adjustment factor decreases rapidly with the number of random selections, thereby shifting from exploration to utilization more quickly; because a larger λ value makes the mutation decrease faster, random exploration is reduced and the refined search for the current optimal solution is increased; a smaller λ value means that the weight coefficient adjustment factor decreases more slowly with the number of random selections, maintaining a longer exploration time, which helps the algorithm to conduct a wider search in the solution space and increase the chance of discovering new solutions;

[0051] Repeat the selection, crossover and mutation process to generate a new generation of population until the maximum number of iterations is reached or the fitness value no longer increases. Select the individual with the highest fitness from the final population as the optimal solution; apply the selected optimal weight coefficient to the weighted formula to fuse and obtain a comprehensive feature data set.

[0052] Furthermore, the training method of the stroke risk prediction model includes:

[0053] The data set is divided into a training set and a test set to build a stroke risk prediction model; the input data of the model is a historical comprehensive feature data set; the output label is the patient's stroke risk; the stroke risk prediction model is a random forest regressor model;

[0054] Initialize the stroke risk prediction model and set the number and depth hyperparameters of the tree; use the training set to train the stroke risk prediction model, use the k-fold cross-validation method to adjust the model's hyperparameters, and tune the initially set parameters;

[0055] The performance of the model is evaluated using the test set data, and the determination coefficient is used to evaluate the difference between the predicted results and the true labels; the determination coefficient is in, The sum of squares of the difference between the patient's stroke risk predicted by the model and the patient's actual stroke risk; is the sum of squares of the difference between the actual patient's stroke risk and the average stroke risk of all patients; i is the actual stroke risk of the i-th patient; The risk of stroke for the i-th patient predicted by the model; is the average risk of stroke for all patients; n is the total number of patients;

[0056] According to the model performance feedback, the number of trees and the depth hyperparameters are adjusted to tune the model, and the model is retrained using the adjusted hyperparameters. When the training reaches the predetermined number of iterations, it is stopped to obtain the final trained stroke risk prediction model. The trained stroke risk prediction model is used to predict the current comprehensive feature data set to predict the patient's stroke risk.

[0057] Furthermore, the method of adjusting the number of trees and the depth hyperparameters to tune the model includes:

[0058] Define an objective function to evaluate the performance of the model, θ is a vector containing hyperparameters θ = (T, D) where T is the number of trees; D is the depth of the tree;

[0059] The possible value set of the number of trees is T ax ={T1, T2, ..., T a}; where a is the total number of possible values ​​of the number of trees; the possible value set of the depth of the tree is D by ={D1, D2, ..., D b}; where b is the total number of possible values ​​of the depth of the tree;

[0060] For each iteration i′, from the set of possible values ​​of the number of trees T ax The possible values ​​of the sum tree depth are D by Randomly select T i′ , and D i′ ;

[0061] The objective function is defined as the accuracy f(θ) of the model, which is:

[0062] Among them, TK is the number of patients correctly predicted by the model to have stroke; TN is the number of patients correctly predicted by the model not to have stroke; FK is the number of patients incorrectly predicted by the model to have stroke; FN is the number of patients incorrectly predicted by the model not to have stroke;

[0063] Record the accuracy of the model in each iteration and keep the hyperparameter configuration with the highest accuracy; repeat the above process until the preset number of iterations is reached; select the hyperparameter configuration θ that maximizes the accuracy f(θ) from all iterations * :θ * = arg max θ∈SE f(θ); where SE is the space of all possible hyperparameter configurations, given by T ax ×D by composition.

[0064] Furthermore, the method of comparing the predicted patient's stroke risk with a preset patient's stroke risk threshold to determine whether the patient has a stroke risk includes:

[0065] If the predicted patient's stroke risk is greater than or equal to the preset patient's stroke risk threshold, the patient is determined to be at risk of stroke;

[0066] If the predicted patient's stroke risk is less than the preset patient's stroke risk threshold, the patient is judged to have no stroke risk.

[0067] The technical effects and advantages of the intelligent detection system for intima-media membrane of stroke screening of the present invention are as follows:

[0068] The present invention improves the image quality by performing Gaussian filtering and denoising on two-dimensional B-ultrasound images; enhances the contrast by linear transformation, making different tissues and structures in the image more clearly visible; more accurately determines the edge of the image based on edge detection, and can accurately locate the boundary of the carotid artery; automatically adapts to the shape and size of the carotid artery through the active contour model and threshold segmentation algorithm, and improves the accuracy of boundary detection; improves the automation of image processing and the accuracy of diagnosis, and is helpful for early detection of carotid artery disease; calculates the mean and standard deviation of each feature, and uses the 3σ method to detect abnormal blood flow data in blood flow data, which can effectively identify abnormal blood flow and is helpful for early detection of potential vascular diseases; combines the calculated hemodynamic parameters into a data set, and provides comprehensive data support for subsequent data analysis, disease diagnosis and risk assessment;

[0069] By comprehensively analyzing the carotid intima-media thickness, abnormal blood flow and hemodynamic parameters and fusing them to obtain a comprehensive feature data set, the accuracy of the stroke risk prediction model is improved; by optimizing the weight coefficient, the accuracy of the stroke risk prediction model can be further improved, which is beneficial for the fitness function to evaluate individual performance and ensure that the selected weight coefficient can maximize the predictive performance of the model; the automatic adjustment of the weight coefficient reduces manual intervention and improves the efficiency of model construction; the introduction of the probability model and the adjustment factor restriction function provides the flexibility of the algorithm, and the algorithm parameters can be adjusted according to the specific situation; the randomness and iteration of the algorithm improve the robustness of the model to different data sets; the automated weight coefficient design reduces labor costs and improves the utilization efficiency of medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a schematic diagram of the structure of an intelligent detection system for intima-media membrane for stroke screening according to the present invention;

[0071] Figure 2 A schematic diagram of a process flow of an intelligent detection method for intima-media membrane for stroke screening according to the present invention; DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] Embodiment 1

[0074] See also Figure 1 As shown, the intima-media membrane intelligent detection system for stroke screening described in this embodiment includes:

[0075] A data acquisition module, used for acquiring two-dimensional B-ultrasound image data, color Doppler image data and blood flow velocity waveform data;

[0076] A data processing module is used to extract the carotid artery boundary and the carotid intima-media region from the two-dimensional B-ultrasound image data; and process the extracted carotid intima-media region to obtain carotid intima-media thickness data;

[0077] A blood flow analysis module is used to extract blood flow velocity and blood flow direction from color Doppler image data, identify blood flow velocity and blood flow direction to obtain abnormal blood flow data, and process blood flow velocity waveform data to obtain a hemodynamic parameter data set;

[0078] The model training module is used to fuse the carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data sets to obtain a comprehensive feature data set, build a stroke risk prediction model, and input the comprehensive feature data set into the stroke risk prediction model to predict the patient's stroke risk;

[0079] The risk warning module is used to compare the predicted patient's stroke risk with the preset patient's stroke risk threshold to determine whether the patient has a stroke risk;

[0080] The prevention and control module formulates corresponding control plans and preventive measures for patients to reduce the risk if there is a risk of stroke. If there is no risk of stroke, a health maintenance plan is formulated to maintain a healthy state. Each module is connected by wired and / or wireless means to achieve data transmission between modules.

[0081] The two-dimensional B-ultrasound image data includes ultrasonic pixel data, carotid artery structure data and arterial wall plaque feature data; the color Doppler image data includes blood flow velocity and blood flow direction; the blood flow velocity waveform data includes systolic peak velocity, end-diastolic velocity, pulse wave velocity and pulse index.

[0082] The two-dimensional B-ultrasound image data is acquired through a B-ultrasound sensor; the color Doppler image data is acquired through a color Doppler ultrasound sensor; and the blood flow velocity waveform data is acquired through ultrasound Doppler technology.

[0083] The method for extracting the carotid artery boundary and the carotid intima-media region from the two-dimensional B-ultrasound image data comprises:

[0084] The two-dimensional B-ultrasound image data is preprocessed and the image noise is removed by a Gaussian filter. The Gaussian filter expression is: S (x, y) = I (x, y) * G σ (x, y); where I(x, y) is the gray value of the original two-dimensional B-ultrasound image at point (x, y); G σ is a Gaussian kernel with standard deviation σ;

[0085] The contrast is enhanced by a linear transformation formula, which is: S′ (x, y) = a·I S (x, y) + b; where I S′ (x, y) is the gray value of the two-dimensional B-ultrasound image at point (x, y) after linear transformation; a and b are parameters for adjusting the contrast;

[0086] The Sobel operator is used to calculate the gradient amplitude and direction of the two-dimensional B-ultrasound image, and the non-maximum suppression and double threshold technology are used to determine the edge of the two-dimensional B-ultrasound image; the active contour model is used to accurately locate the carotid artery boundary; the active contour model is:

[0087] Where E is the energy function of the model; C(s) is the active contour curve; C′(s) is the first-order derivative of the active contour curve; C″(s) is the second-order derivative of the active contour curve; s is the curve parameter; α and β are weight parameters that control the smoothness and rigidity of the active contour; E te To attract the external energy of the contour to move toward the edge;

[0088] A threshold segmentation algorithm is applied on the carotid artery boundary between the intima and the media to separate the carotid artery intima-media region from the entire carotid artery region of the two-dimensional B-ultrasound image; the threshold segmentation algorithm is: Among them, Q is the selected segmentation threshold; I B (x, y) is the gray value of the segmented 2D B-ultrasound image at point (x, y);

[0089] Through the threshold segmentation algorithm, I B The (x, y) value of 0 is the entire carotid artery area, B The (x, y) value of 1 is the carotid intima-media region.

[0090] For example, suppose there is a two-dimensional B-ultrasound image that contains an image of the carotid artery; each pixel in the image has a grayscale value, which is usually between 0 and 255; the grayscale value of the carotid intima-media region is usually different from that of the surrounding tissue; first, a segmentation threshold Q needs to be selected; this threshold is determined based on the grayscale histogram of the image and the grayscale characteristics of the area we want to segment;

[0091] For example, if it is observed that the grayscale value of the carotid intima-media region is usually between 120 and 150, a segmentation threshold Q of 135 will be selected. Next, each pixel in the image is traversed and the threshold segmentation algorithm is applied. If the grayscale value of a pixel is 140, then according to the segmentation threshold Q of 135, this pixel will be marked as 1 because its grayscale value is greater than the threshold and belongs to the carotid intima-media region; on the contrary, if the grayscale value of a pixel is 130, it will be marked as 0 because its grayscale value is less than or equal to the threshold and belongs to the entire carotid region;

[0092] The method for processing the carotid intima-media region to obtain carotid intima-media thickness data comprises:

[0093] Find the starting and ending positions of the carotid artery on the two-dimensional B-ultrasound image; the carotid artery is divided into the internal and external carotid arteries at the carotid bifurcation; use the ultrasonic equipment measurement tool to measure the carotid intima-media thickness from the inner wall to the outer wall of the carotid artery; select the proximal and distal ends of the carotid bifurcation for three measurements each, and then take the average value of the carotid intima-media thickness as the carotid intima-media thickness.

[0094] The method of extracting blood flow velocity and blood flow direction from color Doppler image data, identifying blood flow velocity and blood flow direction data to obtain abnormal blood flow data, and processing blood flow velocity waveform data to obtain a hemodynamic parameter data set includes:

[0095] The color Doppler image noise is removed by Gaussian filter, the blood flow area in the color Doppler image is extracted by threshold segmentation algorithm, the color features are extracted from the blood flow area, and converted into quantifiable values; the color features are represented by RGB color space, red and blue are used to represent the blood flow direction features, and the brightness or saturation of the color is represented as the blood flow velocity features; the mean and standard deviation of each feature are calculated, and the 3σ method is used to detect abnormal blood flow data in the blood flow data:

[0096] The mean of each feature is: Where R(i″) is each data in each feature; n′ is the total number of features; calculate the standard deviation Among them, μ is the mean of each feature data;

[0097] The normal blood flow threshold range is preset to (μ-3σ′, μ+3σ′); each characteristic data is checked to see whether it falls within the normal blood flow threshold range. If a certain characteristic data exceeds the normal blood flow threshold range, it is determined to be abnormal blood flow data;

[0098] For example, suppose there is a set of blood flow velocity measurement values ​​V = {70, 75, 72, 78, 74, 80, 69, 73, 76, 77}; calculate the mean μ of this set of data, μ = (70+75+72+78+74+80+69+73+76+77) / 10 = 74.5; next calculate the standard deviation σ′ as 3.5; use the 3σ detection method to detect abnormal blood flow data in the blood flow data, and the normal blood flow threshold range is (63.5, 85.5); now, check whether each feature data falls within the normal blood flow threshold range; in this example, all measurement values ​​are within the range of 63.5 to 85.5, so there is no abnormal blood flow data; if there is an additional measurement value of 90, then this value will exceed the upper limit of the normal blood flow threshold range of 85.5, and therefore can be determined as abnormal blood flow data.

[0099] Calculate the systolic peak velocity, end-diastolic velocity, average blood flow velocity, pulse wave velocity and pulse index in the blood flow velocity waveform data; the systolic peak velocity is: PSV = max(V(t)); where PSV is the systolic peak velocity; V(t) is the blood flow velocity at time t; max(V(t)) is the maximum value of the blood flow velocity in one cardiac cycle; t is the time required to calculate the blood flow velocity in one cardiac cycle;

[0100] End-diastolic velocity: EDV = V(t ED ), where EDV is the end-diastolic velocity; t ED It is the time point at the end of diastole of a cardiac cycle;

[0101] The average value of the blood flow velocity in one cardiac cycle is obtained according to the peak systolic velocity and the end-diastolic velocity. The calculation formula of the average velocity is: Where, T′ is the duration of a cardiac cycle; It is the integral of blood flow velocity in one cardiac cycle;

[0102] The pulse wave speed is: Where D is the distance the pulse wave travels; Δt is the time difference for the pulse wave to travel from one location to another;

[0103] The pulse index is calculated by the average of the peak systolic velocity, the end-diastolic velocity and the blood flow velocity in one cardiac cycle. The pulse index is: The calculated peak systolic velocity, end-diastolic velocity, average blood flow velocity, pulse wave velocity and pulse index were combined to obtain a hemodynamic parameter data set.

[0104] The method for fusing the carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data set to obtain a comprehensive feature data set includes:

[0105] Identify outliers in carotid intima-media thickness data, abnormal blood flow data, and hemodynamic parameter data sets; preset a carotid intima-media thickness data threshold, and mark data exceeding the carotid intima-media thickness data threshold as outliers; calculate the standard deviation of abnormal blood flow data, and mark data that deviates from the mean value by more than the standard deviation as outliers; identify outliers in hemodynamic parameter data using a density clustering algorithm;

[0106] The identified outliers are eliminated to obtain the processed intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set; the obtained intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set are subjected to standard deviation normalization and converted into a standard normal distribution to eliminate the dimensional influence between the data, and finally the normalized intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set are obtained;

[0107] The carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data set are fused through a weighted formula to form a comprehensive feature data set Z, where the carotid intima-media thickness data is recorded as L, the abnormal blood flow data is recorded as F, and the hemodynamic parameter data set is recorded as H, and the weight coefficient of each feature data set in the weighted formula is designed;

[0108] The weighted formula is: Among them, ω L is the weight coefficient of the carotid intima-media thickness data; ω F is the weight coefficient of abnormal blood flow data; ω H is the weight coefficient of the hemodynamic parameter data set.

[0109] The method for designing the weight coefficient of each feature data set in the weighted formula includes:

[0110] A random initial population is generated, and each individual in the population contains a set of randomly generated weight coefficients [ω L ,ω F ,ω H ], satisfying the constraint that the sum of weight coefficients is 1;

[0111] The initialization population is: Among them, i is each individual in the population; N is the number of individuals in the population;

[0112] The performance of each individual is evaluated by the fitness function, and the fitness function is: FKS = α′·KU(Z); where KU(Z) is an indicator for measuring the inter-class discrimination of the comprehensive feature data set Z; α′ is the weight factor of KU(Z) in the fitness function;

[0113] The stroke risk prediction model constructed by default predicts that the patients' stroke risk is divided into two categories: high-risk category C1 and low-risk category C2;

[0114] The inter-class distinction is: Among them, D B (C1, C2) is the distance between the high-risk category C1 and the low-risk category C2; D W(C1) is the average distance from the internal data points of the high-risk category C1 to its category center point; D W (C2) is the average distance from the internal data points of the low-risk category C2 to its category center point;

[0115] Calculate the sum of all individual fitness values ​​in the population: Among them, FKS i is the fitness value of the lth individual;

[0116] For each generation of individuals in the population, m individuals are randomly selected and their probabilities of being selected for reproduction are calculated using a probability model; the probability model is: Among them, FPJ is the probability of the Jth individual being selected;

[0117] The method for randomly selecting m individuals includes: presetting a random selection probability threshold r in the interval [0, 1], selecting individuals according to the random selection probability threshold r; calculating the probability of each individual being selected and the probability of all individuals before the individual being selected, and finding an individual whose selection probability is greater than or equal to the random selection probability threshold r;

[0118] Assume that the two parent individuals are and Perform crossover operations on the selected individuals to generate new individuals and Mutate the weight coefficient of each new individual: ω′ L =ω L +Δω L ; Among them, ω′ L is the weight coefficient after mutation; Δω L is the weight coefficient ω L Adjustment factor; ω′ F =ω F +Δω F ; Among them, ω′ F is the weight coefficient after mutation; Δω F is the weight coefficient ω F Adjustment factor; ω′ H =1-ω′ L -ω′ F ; Among them, ω′ H is the weight coefficient after mutation;

[0119] Construct an adjustment factor restriction function to limit the range of the weight coefficient adjustment factor; this can prevent excessive random changes from destroying the excellent characteristics of the individual;

[0120] The adjustment factor limiting function is: Δω(o)=Δω0·e -λ·o; Among them, Δω(o) is the adjustment factor of the weight coefficient when the number of random selections is O; Δω0 is the initial weight coefficient adjustment factor; λ is a positive attenuation coefficient that controls the speed at which the weight coefficient adjustment factor decreases; O is the current number of random selections;

[0121] Define the upper limit of λ as λ max , the lower limit of λ is λ min , limit λ∈(λ min ,λ max ), thereby further limiting the range of weight coefficient adjustment factors;

[0122] For example, assuming that Δω0 is 0.1, λ is 0.01, and the number of random selections is 10, and assuming that in the 10th iteration, the current weight coefficient adjustment factor Δω(o)=0.1·e -0.01·10 ≈0.09; define the upper limit of λ max is 0.2, the lower limit of λ min is 0.05. According to the above calculation, 0.09 is between the upper limit of 0.20.2 and the lower limit of 0.050.05, so it is effective. Δω(10) is used to adjust the weight coefficient to ensure that the adjusted weight coefficient does not deviate too far from its excellent characteristics.

[0123] A larger λ value means that the weight coefficient adjustment factor decreases rapidly with the number of random selections, thereby shifting from exploration to utilization more quickly; because a larger λ value makes the mutation decrease faster, random exploration is reduced and the refined search for the current optimal solution is increased; a smaller λ value means that the weight coefficient adjustment factor decreases more slowly with the number of random selections, maintaining a longer exploration time, which helps the algorithm to conduct a wider search in the solution space and increase the chance of discovering new solutions;

[0124] Repeat the selection, crossover and mutation process to generate a new generation of population until the maximum number of iterations is reached or the fitness value no longer increases. Select the individual with the highest fitness from the final population as the optimal solution; apply the selected optimal weight coefficient to the weighted formula to fuse and obtain a comprehensive feature data set.

[0125] The training method of the stroke risk prediction model comprises:

[0126] The data set is divided into a training set and a test set to build a stroke risk prediction model; the input data of the model is a historical comprehensive feature data set; the output label is the patient's stroke risk; the stroke risk prediction model is a random forest regressor model;

[0127] Initialize the stroke risk prediction model and set the number and depth hyperparameters of the tree; use the training set to train the stroke risk prediction model, use the k-fold cross-validation method to adjust the model's hyperparameters, and tune the initially set parameters;

[0128] The performance of the model is evaluated using the test set data, and the determination coefficient is used to evaluate the difference between the predicted results and the true labels; the determination coefficient is in, The sum of squares of the difference between the patient's stroke risk predicted by the model and the patient's actual stroke risk; is the sum of squares of the difference between the actual patient's stroke risk and the average stroke risk of all patients; yi is the actual stroke risk of the ith patient; The risk of stroke for the i-th patient predicted by the model; is the average risk of stroke for all patients; n is the total number of patients;

[0129] According to the model performance feedback, the number of trees and the depth hyperparameters are adjusted to tune the model, and the model is retrained using the adjusted hyperparameters. When the training reaches the predetermined number of iterations, it is stopped to obtain the final trained stroke risk prediction model. The trained stroke risk prediction model is used to predict the current comprehensive feature data set to predict the patient's stroke risk.

[0130] The method of adjusting the number and depth hyperparameters of the trees to optimize the model includes:

[0131] Define an objective function to evaluate the performance of the model, θ is a vector containing hyperparameters θ = (T, D) where T is the number of trees; D is the depth of the tree;

[0132] The possible value set of the number of trees is T ax ={T1, T2, ..., T a}; where a is the total number of possible values ​​of the number of trees; the possible value set of the depth of the tree is D by ={D1, D2, ..., D b}; where b is the total number of possible values ​​of the depth of the tree;

[0133] For each iteration i′, from the set of possible values ​​of the number of trees T ax The possible values ​​of the sum tree depth are D by Randomly select T i′ , and D i′ ;

[0134] The objective function is defined as the accuracy f(θ) of the model, which is:

[0135] Among them, TK is the number of patients correctly predicted by the model to have stroke; TN is the number of patients correctly predicted by the model not to have stroke; FK is the number of patients incorrectly predicted by the model to have stroke; FN is the number of patients incorrectly predicted by the model not to have stroke;

[0136] Record the accuracy of the model in each iteration and keep the hyperparameter configuration with the highest accuracy; repeat the above process until the preset number of iterations is reached; select the hyperparameter configuration θ* that maximizes the accuracy f(θ) from all iterations: θ*=arg max θ∈SE f(θ); where SE is the space of all possible hyperparameter configurations, given by T ax ×D by composition.

[0137] The method of comparing the predicted patient's stroke risk with a preset patient's stroke risk threshold to determine whether the patient has a stroke risk includes:

[0138] If the predicted patient's stroke risk is greater than or equal to the preset patient's stroke risk threshold, the patient is determined to be at risk of stroke;

[0139] If the predicted patient's stroke risk is less than the preset patient's stroke risk threshold, the patient is judged to have no stroke risk.

[0140] The stroke risk threshold of the preset patient is set by the staff. The stroke risk values ​​of different patients are collected through the stroke screening intima-media membrane intelligent detection terminal, and the average of the stroke risk values ​​of multiple patients is taken as the stroke risk threshold of the preset patient. Similarly, the segmentation threshold, normal blood flow threshold and random selection probability threshold are set.

[0141] In this embodiment, the image quality is improved by performing Gaussian filtering and denoising on the two-dimensional B-ultrasound image; the contrast is enhanced by linear transformation, so that different tissues and structures in the image are more clearly visible; the edge of the image is determined more accurately based on edge detection, and the boundary of the carotid artery can be accurately located; the shape and size of the carotid artery are automatically adapted through the active contour model and the threshold segmentation algorithm, so as to improve the accuracy of boundary detection; the degree of automation of image processing and the accuracy of diagnosis are improved, which is helpful for early detection of carotid artery disease; the mean and standard deviation of each feature are calculated, and the abnormal blood flow data in the blood flow data is detected by using the 3σ method, so as to effectively identify the abnormal blood flow, which is helpful for early detection of potential vascular diseases; the calculated hemodynamic parameters are combined into a data set, so as to provide comprehensive data support for subsequent data analysis, disease diagnosis and risk assessment;

[0142] By comprehensively analyzing the carotid intima-media thickness, abnormal blood flow and hemodynamic parameters and fusing them to obtain a comprehensive feature data set, the accuracy of the stroke risk prediction model is improved; by optimizing the weight coefficient, the accuracy of the stroke risk prediction model can be further improved, which is beneficial for the fitness function to evaluate individual performance and ensure that the selected weight coefficient can maximize the predictive performance of the model; the automatic adjustment of the weight coefficient reduces manual intervention and improves the efficiency of model construction; the introduction of the probability model and the adjustment factor restriction function provides the flexibility of the algorithm, and the algorithm parameters can be adjusted according to the specific situation; the randomness and iteration of the algorithm improve the robustness of the model to different data sets; the automated weight coefficient design reduces labor costs and improves the utilization efficiency of medical resources.

[0143] Embodiment 2

[0144] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an intelligent detection method for intima-media membrane of stroke screening, comprising:

[0145] S1, collecting two-dimensional B-ultrasound image data, color Doppler image data and blood flow velocity waveform data;

[0146] S2, extracting the carotid artery boundary and the carotid intima-media region from the two-dimensional B-ultrasound image data; processing the extracted carotid intima-media region to obtain carotid intima-media thickness data;

[0147] S3, extracting blood flow velocity and blood flow direction from the color Doppler image data, identifying the blood flow velocity and blood flow direction to obtain abnormal blood flow data, and processing the blood flow velocity waveform data to obtain a hemodynamic parameter data set;

[0148] S4. Fusing the carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data sets to obtain a comprehensive feature data set, constructing a stroke risk prediction model, and inputting the comprehensive feature data set into the stroke risk prediction model to predict the patient's stroke risk;

[0149] S5. Compare the predicted patient's stroke risk with the preset patient's stroke risk threshold to determine whether the patient has a stroke risk;

[0150] S6. If there is a risk of stroke, formulate corresponding control plans and preventive measures for patients to reduce the risk. If there is no risk of stroke, formulate a health maintenance plan to maintain a healthy state.

[0151] Since the electronic device introduced in this embodiment is an electronic device used to implement an intelligent detection system for the endometrium for stroke screening in the embodiment of this application, based on the intelligent detection system for the endometrium for stroke screening introduced in the embodiment of this application, a person skilled in the art can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application will not be described in detail here. As long as a person skilled in the art implements an electronic device used in an intelligent detection system for the endometrium for stroke screening in the embodiment of this application, it belongs to the scope of protection of this application.

[0152] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0153] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent intima-media membrane detection system for stroke screening, characterized in that: include: A data acquisition module, used for acquiring two-dimensional B-ultrasound image data, color Doppler image data and blood flow velocity waveform data; A data processing module is used to extract the carotid artery boundary and the carotid intima-media region from the two-dimensional B-ultrasound image data; and process the extracted carotid intima-media region to obtain carotid intima-media thickness data; A blood flow analysis module is used to extract blood flow velocity and blood flow direction from color Doppler image data, identify blood flow velocity and blood flow direction to obtain abnormal blood flow data, and process blood flow velocity waveform data to obtain a hemodynamic parameter data set; The model training module is used to fuse the carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data sets to obtain a comprehensive feature data set, build a stroke risk prediction model, and input the comprehensive feature data set into the stroke risk prediction model to predict the patient's stroke risk; The risk warning module is used to compare the predicted patient's stroke risk with the preset patient's stroke risk threshold to determine whether the patient has a stroke risk; The prevention and control module formulates corresponding control plans and preventive measures for patients to reduce the risk if there is a risk of stroke. If there is no risk of stroke, a health maintenance plan is formulated to maintain a healthy state. Each module is connected by wired and / or wireless means.

2. The intelligent detection system for intima-media membrane for stroke screening according to claim 1 is characterized in that: The two-dimensional B-ultrasound image data includes ultrasonic pixel data, carotid artery structure data and arterial wall plaque feature data; the color Doppler image data includes blood flow velocity and blood flow direction; the blood flow velocity waveform data includes systolic peak velocity, end-diastolic velocity, pulse wave velocity and pulse index.

3. The intelligent detection system for intima-media membrane for stroke screening according to claim 2 is characterized in that: The method for extracting the carotid artery boundary and the carotid intima-media region from the two-dimensional B-ultrasound image data comprises: The two-dimensional B-ultrasound image data is preprocessed and the image noise is removed by a Gaussian filter. The Gaussian filter expression is: S (x,y)=I(x,y)*G σ (x, y); where I(x, y) is the gray value of the original two-dimensional B-ultrasound image at point (x, y); G σ is a Gaussian kernel with standard deviation σ; The contrast is enhanced by a linear transformation formula, which is: S′ (x,y)=a·I S (x,y)+b; where I S′ (x, y) is the gray value of the two-dimensional B-ultrasound image at point (x, y) after linear transformation; a and b are parameters for adjusting the contrast; The Sobel operator is used to calculate the gradient amplitude and direction of the two-dimensional B-ultrasound image, and the non-maximum suppression and double threshold technology are used to determine the edge of the two-dimensional B-ultrasound image; the active contour model is used to accurately locate the carotid artery boundary; the active contour model is: Where E is the energy function of the model; C(s) is the active contour curve; C′(s) is the first-order derivative of the active contour curve; C"(s) is the second-order derivative of the active contour curve; s is the curve parameter; α and β are weight parameters that control the smoothness and rigidity of the active contour; E te To attract the external energy of the contour to move toward the edge; A threshold segmentation algorithm is applied on the carotid artery boundary between the intima and the media to separate the carotid artery intima-media region from the entire carotid artery region of the two-dimensional B-ultrasound image; the threshold segmentation algorithm is: Among them, is the selected segmentation threshold; I B (x, y) is the gray value of the segmented 2D B-ultrasound image at point (x, y); Through the threshold segmentation algorithm, I B The (x,y) value of 0 is the entire carotid artery area, I B The (x,y) value of 1 is the carotid intima-media region.

4. The intelligent intima-media membrane detection system for stroke screening according to claim 3 is characterized in that: The method for processing the carotid intima-media region to obtain carotid intima-media thickness data comprises: Find the starting and ending positions of the carotid artery on the two-dimensional B-ultrasound image; the carotid artery is divided into the internal and external carotid arteries at the carotid bifurcation; use the ultrasonic equipment measurement tool to measure the carotid intima-media thickness from the inner wall to the outer wall of the carotid artery; select the proximal and distal ends of the carotid bifurcation for three measurements each, and then take the average value of the carotid intima-media thickness as the carotid intima-media thickness.

5. The intelligent intima-media membrane detection system for stroke screening according to claim 4 is characterized in that: The method of extracting blood flow velocity and blood flow direction from color Doppler image data, identifying blood flow velocity and blood flow direction data to obtain abnormal blood flow data, and processing blood flow velocity waveform data to obtain a hemodynamic parameter data set includes: The color Doppler image noise is removed by Gaussian filter, the blood flow area in the color Doppler image is extracted by threshold segmentation algorithm, the color features are extracted from the blood flow area, and converted into quantifiable values; the color features are represented by RGB color space, red and blue are used to represent the blood flow direction features, and the brightness or saturation of the color is represented as the blood flow velocity features; the mean and standard deviation of each feature are calculated, and the 3σ method is used to detect abnormal blood flow data in the blood flow data: The mean of each feature is: Where R(i″) is each data in each feature; n′ is the total number of features; calculate the standard deviation Among them, μ is the mean of each feature data; The normal blood flow threshold range is preset to (μ3σ′,μ+3σ′); each characteristic data is checked to see whether it falls within the normal blood flow threshold range. If a certain characteristic data exceeds the normal blood flow threshold range, it is determined to be abnormal blood flow data; Calculate the systolic peak velocity, end-diastolic velocity, average blood flow velocity, pulse wave velocity and pulse index in the blood flow velocity waveform data; the systolic peak velocity is: PSV = max(V(t)); where PSV is the systolic peak velocity; V(t) is the blood flow velocity at time t; max(V(t)) is the maximum value of the blood flow velocity in one cardiac cycle; t is the time required to calculate the blood flow velocity in one cardiac cycle; End-diastolic velocity: EDV = V(t ED ), where EDV is the end-diastolic velocity; t ED It is the time point at the end of diastole of a cardiac cycle; The average value of the blood flow velocity in one cardiac cycle is obtained according to the peak systolic velocity and the end-diastolic velocity. The calculation formula of the average velocity is: Where, T′ is the duration of a cardiac cycle; It is the integral of blood flow velocity in one cardiac cycle; The pulse wave speed is: Where D is the distance the pulse wave travels; Δt is the time difference for the pulse wave to travel from one location to another; The pulse index is calculated by the average of the peak systolic velocity, the end-diastolic velocity and the blood flow velocity in one cardiac cycle. The pulse index is: The calculated peak systolic velocity, end-diastolic velocity, average blood flow velocity, pulse wave velocity and pulse index were combined to obtain a hemodynamic parameter data set.

6. The intelligent intima-media membrane detection system for stroke screening according to claim 5, characterized in that: The method for fusing the carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data set to obtain a comprehensive feature data set includes: Identify outliers in carotid intima-media thickness data, abnormal blood flow data, and hemodynamic parameter data sets; preset a carotid intima-media thickness data threshold, and mark data exceeding the carotid intima-media thickness data threshold as outliers; calculate the standard deviation of abnormal blood flow data, and mark data that deviates from the mean value by more than the standard deviation as outliers; identify outliers in hemodynamic parameter data using a density clustering algorithm; The identified outliers are eliminated to obtain the processed intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set; the obtained intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set are subjected to standard deviation normalization and converted into standard normal distribution, and finally the normalized intima-media thickness characteristic data set, abnormal blood flow characteristic data set and hemodynamic parameter characteristic data set are obtained; The carotid intima-media thickness data, abnormal blood flow data and hemodynamic parameter data set are fused through a weighted formula to form a comprehensive feature data set Z, where the carotid intima-media thickness data is recorded as L, the abnormal blood flow data is recorded as F, and the hemodynamic parameter data set is recorded as H, and the weight coefficient of each feature data set in the weighted formula is designed; The weighted formula is: Among them, ω L is the weight coefficient of the carotid intima-media thickness data; ω F is the weight coefficient of abnormal blood flow data; ω H is the weight coefficient of the hemodynamic parameter data set.

7. The intelligent intima-media membrane detection system for stroke screening according to claim 6, characterized in that: The method for designing the weight coefficient of each feature data set in the weighted formula includes: A random initial population is generated, and each individual in the population contains a set of randomly generated weight coefficients [ω L ,ω F ,ω H ], satisfying the constraint that the sum of weight coefficients is 1; The initialization population is: Among them, i is each individual in the population; N is the number of individuals in the population; The performance of each individual is evaluated by the fitness function, and the fitness function is: FKS = α′·KU(Z); where KU(Z) is an indicator for measuring the inter-class discrimination of the comprehensive feature data set Z; α′ is the weight factor of KU(Z) in the fitness function; The stroke risk prediction model constructed by default predicts that the patients' stroke risk is divided into two categories: high-risk category C1 and low-risk category C2; The inter-class distinction is: Among them, D B (C1, C2) is the distance between the high-risk category C1 and the low-risk category C2; D W (C1) is the average distance from the internal data points of the high-risk category C1 to its category center point; D W (C2) is the average distance from the internal data points of the low-risk category C2 to its category center point; Calculate the sum of all individual fitness values ​​in the population: Among them, FKS i is the fitness value of the i-th individual; For each generation of individuals in the population, m individuals are randomly selected and their probabilities of being selected for reproduction are calculated using a probability model; the probability model is: Among them, FP J is the probability of the Jth individual being selected; The method for randomly selecting m individuals includes: presetting a random selection probability threshold r in the interval [0,1], selecting individuals according to the random selection probability threshold r; calculating the probability of each individual being selected and the probability of all individuals before the individual being selected, and finding an individual whose selection probability is greater than or equal to the random selection probability threshold r; Assume that the two parent individuals are and Perform crossover operations on the selected individuals to generate new individuals and Mutate the weight coefficient of each new individual: ω′ L =ω L +Δω L ; Among them, ω′ L is the weight coefficient after mutation; Δω L is the weight coefficient ω L Adjustment factor; ω′ F =ω F +Δω F ; Among them, ω′ F is the weight coefficient after mutation; Δω F is the weight coefficient ω F Adjustment factor; ω′ H =1-ω′ L -ω′ F ; Among them, ω′ H is the weight coefficient after mutation; Construct an adjustment factor restriction function to limit the range of the weight coefficient adjustment factor; the adjustment factor restriction function is: Δω(o) = Δω0·e -λ·o ; Among them, Δω(o) is the adjustment factor of the weight coefficient when the number of random selections is O; Δω0 is the initial weight coefficient adjustment factor; λ is a positive attenuation coefficient that controls the speed at which the weight coefficient adjustment factor decreases; o is the current number of random selections; Define the upper limit of λ as λ max , the lower limit of λ is λ min , limit λ∈(λ min ,λ max ), thereby further limiting the range of weight coefficient adjustment factors; Repeat the selection, crossover and mutation process to generate a new generation of population until the maximum number of iterations is reached or the fitness value no longer increases. Select the individual with the highest fitness from the final population as the optimal solution; apply the selected optimal weight coefficient to the weighted formula to fuse and obtain a comprehensive feature data set.

8. The intelligent intima-media membrane detection system for stroke screening according to claim 7, characterized in that: The training method of the stroke risk prediction model comprises: The data set is divided into a training set and a test set to build a stroke risk prediction model; the input data of the model is a historical comprehensive feature data set; the output label is the patient's stroke risk; the stroke risk prediction model is a random forest regressor model; Initialize the stroke risk prediction model and set the number and depth hyperparameters of the tree; use the training set to train the stroke risk prediction model, use the k-fold cross-validation method to adjust the model's hyperparameters, and tune the initially set parameters; The performance of the model is evaluated using the test set data, and the determination coefficient is used to evaluate the difference between the predicted results and the true labels; the determination coefficient is in, The sum of squares of the difference between the patient's stroke risk predicted by the model and the patient's actual stroke risk; is the sum of squares of the difference between the actual patient's stroke risk and the average stroke risk of all patients; i is the actual stroke risk of the i-th patient; The risk of stroke for the i-th patient predicted by the model; is the average risk of stroke for all patients; n is the total number of patients; According to the model performance feedback, the number of trees and the depth hyperparameters are adjusted to tune the model, and the model is retrained using the adjusted hyperparameters. When the training reaches the predetermined number of iterations, it is stopped to obtain the final trained stroke risk prediction model. The trained stroke risk prediction model is used to predict the current comprehensive feature data set to predict the patient's stroke risk.

9. The intelligent intima-media membrane detection system for stroke screening according to claim 8, characterized in that: The method of adjusting the number and depth hyperparameters of the trees to optimize the model includes: Define an objective function to evaluate the performance of the model, θ is a vector containing hyperparameters θ = (T, D) where T is the number of trees; D is the depth of the tree; The possible value set of the number of trees is T ax ={T1,T2,...,T a }; where a is the total number of possible values ​​of the number of trees; the possible value set of the depth of the tree is D by ={D1,D2,...,D b }; where b is the total number of possible values ​​of the depth of the tree; For each iteration i′, from the set of possible values ​​of the number of trees T ax The possible values ​​of the sum tree depth are D by Randomly select T i′ and D i′ ; The objective function is defined as the accuracy f(θ) of the model, which is: Among them, TK is the number of patients correctly predicted by the model to have stroke; TN is the number of patients correctly predicted by the model not to have stroke; FK is the number of patients incorrectly predicted by the model to have stroke; FN is the number of patients incorrectly predicted by the model not to have stroke; Record the accuracy of the model in each iteration and keep the hyperparameter configuration with the highest accuracy; repeat the above process until the preset number of iterations is reached; select the hyperparameter configuration θ that maximizes the accuracy f(θ) from all iterations * :θ * = arg max θ∈SE f(θ); where SE is the space of all possible hyperparameter configurations, given by T ax ×D by composition.

10. The intelligent intima-media membrane detection system for stroke screening according to claim 9, characterized in that: The method of comparing the predicted patient's stroke risk with a preset patient's stroke risk threshold to determine whether the patient has a stroke risk includes: If the predicted patient's stroke risk is greater than or equal to the preset patient's stroke risk threshold, the patient is determined to be at risk of stroke; If the predicted patient's stroke risk is less than the preset patient's stroke risk threshold, the patient is judged to have no stroke risk.

Images