Ultrasonic image quality evaluation system based on artificial intelligence

Abstract

The invention discloses an ultrasonic image quality evaluation system based on artificial intelligence. The system comprises a data acquisition module, a data preprocessing module, an ultrasonic image feature information processing module, an ultrasonic image quality evaluation model construction module and an ultrasonic image quality evaluation module. The invention belongs to the technical field of ultrasonic images, and particularly relates to an artificial intelligence-based ultrasonic image quality evaluation system. According to the scheme, gradients of ultrasonic images are calculated along four directions, a combined fusion gradient and a differential fusion gradient are obtained, a compensation coefficient and a correction coefficient are introduced for feature fusion, and a feature matrix is constructed; obtaining a feature vector by using snakelike scanning, and performing feature dimension reduction; the initial position of an individual is determined based on Fuch mapping and an opposite position, the individual transfer direction during position updating is calculated based on an individual position transfer probability and a roulette selection method, position updating is performed based on an adaptive disturbance regulation factor, and an optimal parameter position is found to construct a model.

Description

Technical Field

[0001] The present invention belongs to the technical field of ultrasonic imaging, and specifically refers to an ultrasonic image quality evaluation system based on artificial intelligence. Background Art

[0002] The ultrasonic image quality evaluation system uses artificial intelligence technology to evaluate the quality of the collected ultrasonic images, so as to obtain more accurate and reliable image quality level information. However, the existing image feature extraction methods have technical problems such as being unable to capture the key information in the images, having high computational complexity, resulting in information loss, insufficient generalization ability, and overfitting; when the existing ultrasonic image quality evaluation model searches for parameters, there are problems such as uneven initialization positions, being unable to fully cover the entire search space, lacking global search ability, and the search algorithm being restricted by various local optimal positions, resulting in the inability to find the global optimal position and a slow convergence speed. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides an ultrasonic image quality evaluation system based on artificial intelligence. Aiming at the technical problems of the existing image feature extraction methods, such as being unable to capture the key information in the images, having high computational complexity, resulting in information loss, insufficient generalization ability, and overfitting, this solution calculates the gradients of the ultrasonic image in four directions to obtain the combined fusion gradient and the differential fusion gradient, introduces a compensation coefficient and a correction coefficient for feature fusion, constructs a feature matrix, then uses a serpentine scan to obtain feature vectors, and performs feature dimensionality reduction, which helps to identify the edges and textures in the ultrasonic image, capture the details and changes in the ultrasonic image more comprehensively, extract more representative features, helps to reduce the complexity of the data, improve the computational efficiency, and retain key information at the same time; aiming at the problems of the existing ultrasonic image quality evaluation model when searching for parameters, such as uneven initialization positions, being unable to fully cover the entire search space, lacking global search ability, and the search algorithm being restricted by various local optimal positions, resulting in the inability to find the global optimal position and a slow convergence speed, this solution uses the Fuch mapping method to initialize the individual positions, improves the randomness and diversity of the individual initial positions, increases the coverage rate of the search space, generates opposite positions to better maintain the diversity of the individuals, and prompts the search algorithm to explore globally. Based on the individual position transfer probability and the roulette wheel selection method, the individual transfer direction is calculated when updating the position, which is beneficial to retaining and spreading excellent genetic information. Based on the adaptive perturbation adjustment factor, the position is updated, enabling the individuals to better balance the global exploration and local mining capabilities during the search process, helping to improve the convergence speed of the algorithm and avoid falling into local optimal solutions, so that the constructed ultrasonic quality evaluation model can more accurately evaluate the quality of ultrasonic images and improve the accuracy and robustness of the model.

[0004] An ultrasound image quality assessment system based on artificial intelligence provided by the present invention includes a data acquisition module, a data preprocessing module, an ultrasound image feature information processing module, a module for constructing an ultrasound image quality assessment model, and an ultrasound image quality assessment module;

[0005] The data acquisition module acquires ultrasound images and quality assessment levels, and the quality assessment levels include excellent, good, average, poor, and invalid;

[0006] The data preprocessing module preprocesses the acquired ultrasound images. The preprocessing includes denoising, enhancement, and smoothing, and uses the risk assessment level as a label;

[0007] The ultrasound image feature information processing module calculates the gradients of the ultrasound image in four directions to obtain a combined fusion gradient and a differential fusion gradient, introduces a compensation coefficient and a correction coefficient for feature fusion, constructs a feature matrix, then uses a serpentine scan to obtain feature vectors, and performs feature dimensionality reduction;

[0008] The module for constructing an ultrasound image quality assessment model determines the initial position of an individual based on the Fuch mapping and the opposite position, calculates the individual transfer direction during position update based on the individual position transfer probability and the roulette wheel selection method, performs position update based on the adaptive perturbation adjustment factor, and finds the optimal parameter position to construct the model;

[0009] The ultrasound image quality assessment module obtains the quality level of the ultrasound image based on the label output by the ultrasound image quality assessment model.

[0010] Further, the ultrasound image feature information processing module specifically includes the following:

[0011] Calculate the gradients. Convert the preprocessed ultrasound image into a grayscale image to obtain a matrix C of size m×n. Obtain the gradient information of each pixel point of the image by calculating the change in the grayscale value of the ultrasound image. Calculate the right gradient, left gradient, upper gradient, and lower gradient of each pixel point (x, y) in the matrix C. The used formulas are as follows:

[0012] A rx =|f(x + 1, y) - f(x, y)|;

[0013] A lx =|f(x, y) - f(x - 1, y)|;

[0014] A uy =|f(x, y + 1) - f(x, y)|;

[0015] A dy =|f(x, y) - f(x, y - 1)|;

[0016] In the formula, Arx 、A lx 、A uy and A dy are the right gradient, left gradient, upper gradient, and lower gradient of the pixel point (x, y) in matrix C respectively. x and y are the row index and column index of matrix C respectively, f(x, y) is the grayscale value of the pixel point (x, y) in matrix C, and f(x + 1, y), f(x - 1, y), f(x, y + 1), and f(x, y - 1) are the grayscale values of the pixel points to the right, left, above, and below the pixel point (x, y) respectively;

[0017] Calculate the combined fusion gradient A fu , add the gradient values in the four directions, and the formula used is as follows:

[0018] A fu = A rx + A lx + A uy + A dy ;

[0019] Calculate the differential fusion gradient A su , obtain the differential fusion gradient by calculating the difference of the gradients, and the formula used is as follows:

[0020] A su = (A rx - A lx ) + (A uy - A dy );

[0021] Feature fusion, based on the compensation coefficient and correction coefficient, perform weighted fusion on the pixel values in matrix C with the combined fusion gradient and the differential fusion gradient, and the formula used is as follows:

[0022] c(x, y) = c0(x, y) + g f × a fu (x, y) - g s × a su (x, y);

[0023] In the formula, c(x, y) is the feature value after the fusion processing of c0(x, y), c0(x, y) is the pixel value of the pixel point (x, y) in matrix C, g f is the compensation coefficient of the combined fusion gradient, g s is the correction coefficient of the differential fusion gradient, a fu (x, y) is the combined fusion matrix of the pixel point (x, y), a su (x, y) is the differential fusion matrix of the pixel point (x, y);

[0024] Construct a feature matrix, and go to the step of calculating the gradient for iteration until the eigenvalues after fusion processing are calculated for all pixel points in matrix C. The matrix C with size m×n is transformed into the first feature matrix G with size (m - 2)×(n - 2). Use the adjacent value filling algorithm to transform matrix G into the second feature matrix Q with size m×n;

[0025] Calculate the feature vector. Use a module with size z×z to partition the second feature matrix Q, calculate the histogram of each block, and connect the histograms of each block in the connection order of serpentine scanning to obtain the feature vector of the preprocessed ultrasonic image;

[0026] Feature dimensionality reduction. Perform dimensionality reduction processing on the feature vector, and construct a feature dataset based on the feature vector after dimensionality reduction.

[0027] Furthermore, the module for constructing the ultrasonic image quality evaluation model specifically includes the following content:

[0028] Initialize the individual position. Use the individual position to represent the parameter position, and initialize the individual position based on the Fuch mapping. The formula used is as follows:

[0029] ;

[0030] In the formula, K i，j is the initial position of the i-th individual in the j-th dimension, i is the individual index, j is the dimension index of the individual search space, UB j and LB j are the upper and lower limits of the individual search space in the j-th dimension respectively, and h i，j is the chaotic variable of the i-th individual in the j-th dimension using the Fuch mapping;

[0031] Generate the opposite position. The formula used is as follows:

[0032] ;

[0033] In the formula, is the opposite position of K i，j ;

[0034] Determine the initial position of the individual. Take the performance of the ultrasonic image quality evaluation model established based on the parameter position as the fitness value of the corresponding individual position. Sort the initial position and the opposite position from largest to smallest according to the size of the fitness value, and select the first N K individual positions as the final initial positions of the individuals;

[0035] Train the ultrasonic image quality evaluation model. Use python to import the sklearn library and call the SVM function based on the current parameter position to train the ultrasonic image quality evaluation model based on the feature dataset;

[0036] Calculate the individual position transfer probability and select the individual position K according to the roulette wheel selection method. b At iteration (t), when updating the individual position, transfer in the direction of K b at iteration (t). The formula used is as follows:

[0037] ;

[0038] ;

[0039] In the formula, P i，a at iteration (t) is the probability that the i-th individual transfers in the direction of the a-th individual position at the t-th iteration. a is the individual index not equal to i, t is the iteration index, K i at iteration (t) is the position of the i-th individual at the t-th iteration, K a at iteration (t) is the position of the a-th individual at the t-th iteration, J(K a at iteration (t)) is the relative fitness value of the a-th individual at the t-th iteration, d(K a at iteration (t)) is the fitness value of the position of the i-th individual at the t-th iteration, d worst at iteration (t) and d best at iteration (t) are the lowest fitness value and the highest fitness value at the t-th iteration respectively. ε is a very small value to avoid the denominator being zero, ||K i at iteration (t) - K a at iteration (t)|| is the Euclidean distance between K i at iteration (t) and K a at iteration (t);

[0040] Position update: Update the position based on the adaptive perturbation adjustment factor. The formula used is as follows:

[0041] ;

[0042] ;

[0043] In the formula, δ i at iteration (t) is the adaptive perturbation adjustment factor of the i-th individual at the t-th iteration, r1 is the first random number, λ is the position update step size, K best at iteration (t) is the global optimal position at the t-th iteration. The global optimal position is the individual position with the highest fitness value, ||K best at iteration (t) - K i at iteration (t)|| is the Euclidean distance between K best at iteration (t) and K i at iteration (t), K i at iteration (t + 1) is the position of the i-th individual at the (t + 1)-th iteration, r2 is the second random number;

[0044] The model parameters are determined. The fitness value evaluation threshold μ and the maximum number of iterations T are preset. The fitness value of the individual and the global optimal position are updated. When the fitness value corresponding to the global optimal position is higher than the fitness value evaluation threshold μ, the parameter position corresponding to the global optimal position is the optimal parameter position of the model, and an ultrasonic image quality evaluation model is constructed based on the optimal parameter position; otherwise, if the maximum number of iterations T is reached, the individual position is re-initialized; otherwise, the position update continues.

[0045] Further, the ultrasonic image quality evaluation module collects ultrasonic images to be evaluated. After preprocessing, the images are input into the ultrasonic image quality evaluation model, and the quality level of the ultrasonic images is obtained based on the labels output by the ultrasonic image quality evaluation model.

[0046] The beneficial effects of the present invention using the above scheme are as follows:

[0047] (1) Aiming at the technical problems existing in the existing image feature extraction methods, such as being unable to capture key information in images, having high computational complexity, resulting in information loss, insufficient generalization ability, and overfitting, this scheme calculates the gradients of ultrasonic images in four directions to obtain the combined fusion gradient and the differential fusion gradient, introduces a compensation coefficient and a correction coefficient for feature fusion, constructs a feature matrix, then uses serpentine scanning to obtain feature vectors, and performs feature dimensionality reduction, which helps to identify the edges and textures in ultrasonic images, capture the details and changes in ultrasonic images more comprehensively, extract more representative features, helps to reduce the complexity of data, improve the computational efficiency, and retain key information at the same time.

[0048] (2) Aiming at the problems that when the existing ultrasonic image quality evaluation model searches for parameters, the initial positions are uneven, unable to fully cover the entire search space, and lack of global search ability, and the search algorithm is restricted by various local optimal positions, resulting in the inability to find the global optimal position and a slow convergence speed, this scheme uses the Fuch mapping method to initialize the individual positions, improves the randomness and diversity of the individual initial positions, increases the coverage rate of the search space, generates opposite positions to better maintain the diversity of individuals, and prompts the search algorithm to explore within the global range. The individual transfer direction during position update is calculated based on the individual position transfer probability and the roulette wheel selection method, which is beneficial to retaining and spreading excellent genetic information. The position update is performed based on the adaptive perturbation adjustment factor, enabling individuals to better balance the global exploration and local mining capabilities during the search process, helping to improve the convergence speed of the algorithm and avoid falling into local optimal solutions, and enabling the constructed ultrasonic quality evaluation model to more accurately evaluate the quality of ultrasonic images, improving the accuracy and robustness of the model. Description of the Drawings

[0049] Figure 1Schematic diagram of an ultrasonic image quality evaluation system based on artificial intelligence provided by the present invention;

[0050] Figure 2 Schematic diagram of the ultrasonic image feature information processing module;

[0051] Figure 3 Schematic diagram of the module for constructing an ultrasonic image quality evaluation model.

[0052] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. Detailed implementation manners

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0054] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention.

[0055] Embodiment 1, refer to Figure 1 An ultrasonic image quality evaluation system based on artificial intelligence provided by the present invention includes a data acquisition module, a data preprocessing module, an ultrasonic image feature information processing module, a module for constructing an ultrasonic image quality evaluation model, and an ultrasonic image quality evaluation module;

[0056] The data acquisition module acquires ultrasonic images and quality evaluation grades, and the quality evaluation grades include excellent, good, average, poor, and invalid;

[0057] The data preprocessing module preprocesses the acquired ultrasonic images. The preprocessing includes denoising, enhancement, and smoothing, and takes the risk assessment grade as a label;

[0058] The ultrasonic image feature information processing module calculates the gradients of the ultrasonic image in four directions to obtain a combined fusion gradient and a differential fusion gradient, introduces a compensation coefficient and a correction coefficient for feature fusion, constructs a feature matrix, then obtains a feature vector using serpentine scanning, and performs feature dimensionality reduction;

[0059] The module for constructing an ultrasonic image quality evaluation model determines the initial position of an individual based on the Fuch mapping and the opposite position, calculates the individual transfer direction during position update based on the individual position transfer probability and the roulette selection method, performs position update based on the adaptive perturbation adjustment factor, and finds the optimal parameter position to construct the model;

[0060] The ultrasonic image quality evaluation module obtains the quality level of the ultrasonic image based on the label output by the ultrasonic image quality evaluation model.

[0061] Example 2, refer to Figure 1 , this example is based on the above example. The data preprocessing module preprocesses the collected ultrasonic images, and the preprocessing includes denoising, enhancement, and smoothing. The risk assessment level is used as the label. Denoising uses a Gaussian filter to reduce the noise in the ultrasonic image. Enhancement uses the histogram equalization technique to enhance the contrast and details of the ultrasonic image. Smoothing uses a smoothing filter to smooth the edges and details of the ultrasonic image.

[0062] Example 3, refer to Figure 1 and Figure 2 , this example is based on the above example. The ultrasonic image feature information processing module specifically includes the following content:

[0063] Calculate the gradient. Convert the preprocessed ultrasonic image into a grayscale image to obtain a matrix C of size m×n. Obtain the gradient information of each pixel point in the image by calculating the change in the grayscale value of the ultrasonic image. Calculate the right gradient, left gradient, upper gradient, and lower gradient of each pixel point (x, y) in the matrix C. The formulas used are as follows:

[0064] A rx =|f(x + 1, y) - f(x, y)|;

[0065] A lx =|f(x, y) - f(x - 1, y)|;

[0066] A uy =|f(x, y + 1) - f(x, y)|;

[0067] A dy =|f(x, y) - f(x, y - 1)|;

[0068] In the formula, A rx , A lx , A uy and A dyThey are the right gradient, left gradient, upper gradient, and lower gradient of the pixel point (x, y) in matrix C, where x and y are the row index and column index of matrix C respectively, f(x, y) is the gray value at the pixel point (x, y) in matrix C, and f(x + 1, y), f(x - 1, y), f(x, y + 1), and f(x, y - 1) are the gray values of the pixel points adjacent to the right, left, above, and below the pixel point (x, y) respectively;

[0069] Calculate the combined fusion gradient A fu , add the gradient values in the four directions. The combined fusion gradient can comprehensively consider the global feature information of the image, and the formula used is as follows:

[0070] A fu = A rx + A lx + A uy + A dy ;

[0071] Calculate the differential fusion gradient A su , obtain the differential fusion gradient by calculating the gradient difference. The differential fusion gradient can highlight the detail features of the image, and the formula used is as follows:

[0072] A su = (A rx - A lx ) + (A uy - A dy );

[0073] Feature fusion: Weightedly fuse the pixel values in matrix C with the combined fusion gradient and the differential fusion gradient based on the compensation coefficient and the correction coefficient. The formula used is as follows:

[0074] c(x, y) = c0(x, y) + g f × a fu (x, y) - g s × a su (x, y);

[0075] In the formula, c(x, y) is the feature value after the fusion processing of c0(x, y), c0(x, y) is the pixel value of the pixel point (x, y) in matrix C, g f is the compensation coefficient of the combined fusion gradient, g s is the correction coefficient of the differential fusion gradient, a fu (x, y) is the combined fusion matrix of the pixel point (x, y), and a su (x, y) is the differential fusion matrix of the pixel point (x, y);

[0076] Construct a feature matrix, and go to the step of calculating the gradient for iteration until the eigenvalues after fusion processing are calculated for all pixel points in matrix C. The matrix C with size m×n is transformed into the first feature matrix G with size (m - 2)×(n - 2). Then, use the adjacent value filling algorithm to transform matrix G into the second feature matrix Q with size m×n;

[0077] Calculate the feature vector. Use a module with size z×z to partition the second feature matrix Q, calculate the histogram of each block, and connect the histograms of each block in the connection order of serpentine scanning to obtain the feature vector of the preprocessed ultrasonic image;

[0078] Feature dimensionality reduction. Perform dimensionality reduction processing on the feature vector, and construct a feature data set based on the feature vector after dimensionality reduction.

[0079] By performing the above operations, for the technical problems existing in the existing image feature extraction methods, such as being unable to capture key information in the image, having high computational complexity, resulting in information loss, insufficient generalization ability, and overfitting, this solution calculates the gradients of the ultrasonic image in four directions to obtain the merged fusion gradient and the differential fusion gradient, introduces a compensation coefficient and a correction coefficient for feature fusion, constructs a feature matrix, then uses serpentine scanning to obtain the feature vector, and performs feature dimensionality reduction, which helps to identify the edges and textures in the ultrasonic image, capture the details and changes in the ultrasonic image more comprehensively, extract more representative features, helps to reduce the complexity of the data, improve the computational efficiency, and at the same time retain the key information.

[0080] Example 4, refer to Figure 1 and Figure 3 , based on the above example, construct an ultrasonic image quality assessment model module, which specifically includes the following content:

[0081] Initialize the individual position. Use the individual position to represent the parameter position, and initialize the individual position based on the Fuch mapping. The formula used is as follows:

[0082] ;

[0083] In the formula, K i，j is the initial position of the i-th individual in the j-th dimension, i is the individual index, j is the dimension index of the individual search space, UB j and LB j are the upper and lower limits of the individual search space in the j-th dimension respectively, and h i，j is the chaotic variable of the i-th individual in the j-th dimension using the Fuch mapping;

[0084] Generate the opposite position. The formula used is as follows:

[0085] ;

[0086] In the formula, is the opposite position of K i，j ;

[0087] Determine the initial position of the individual, take the performance of the ultrasonic image quality evaluation model established based on the parameter position as the fitness value of the corresponding individual position, sort the initial position and the opposite position from large to small according to the size of the fitness value, and select the first N K individual positions as the final initial positions of the individuals;

[0088] Train the ultrasonic image quality evaluation model, use python to import the sklearn library, call the SVM function based on the current parameter position, and train the ultrasonic image quality evaluation model based on the feature data set;

[0089] Calculate the individual position transfer probability, select the individual position K b (t) according to the roulette wheel selection method. When updating the individual position, transfer in the direction where K b (t) is located. The formula used is as follows:

[0090] ;

[0091] ;

[0092] In the formula, P i，a (t) is the probability that the i-th individual transfers to the a-th individual position direction at the t-th iteration, a is the individual index not equal to i, t is the iteration index, K i (t) is the position of the i-th individual at the t-th iteration, K a (t) is the position of the a-th individual at the t-th iteration, J(K a (t)) is the relative fitness value of the a-th individual at the t-th iteration, d(K a (t)) is the fitness value of the i-th individual position at the t-th iteration, d worst (t) and d best (t) are the lowest fitness value and the highest fitness value at the t-th iteration respectively, ε is a very small value to avoid the denominator being zero, ||K i (t)-K a (t)|| is the Euclidean distance between K i (t) and K a (t);

[0093] Update the position, update the position based on the adaptive perturbation adjustment factor. The formula used is as follows:

[0094] ;

[0095] ;

[0096] where δ i (t) is the adaptive perturbation adjustment factor of the i-th individual at the t-th iteration, r1 is the first random number, λ is the position update step size, K best (t) is the global optimal position at the t-th iteration, and the global optimal position is the individual position with the highest fitness value, ||K best (t) - K i (t)|| is the Euclidean distance between K best (t) and K i (t), K i (t + 1) is the position of the i-th individual at the (t + 1)-th iteration, and r2 is the second random number;

[0097] For model parameter determination, the fitness value evaluation threshold μ and the maximum number of iterations T are preset, the fitness values of individuals and the global optimal position are updated. When the fitness value corresponding to the global optimal position is higher than the fitness value evaluation threshold μ, the parameter position corresponding to the global optimal position is the optimal parameter position of the model, and an ultrasonic image quality evaluation model is constructed based on the optimal parameter position; otherwise, if the maximum number of iterations T is reached, the individual positions are re-initialized; otherwise, the position update continues.

[0098] By performing the above operations, for the existing ultrasonic image quality evaluation model during parameter search, there are problems such as uneven initialization positions, inability to fully cover the entire search space, lack of global search ability, and the search algorithm being trapped by various local optimal positions, resulting in the inability to find the global optimal position and a slow convergence speed. This solution uses the Fuch mapping method to initialize the individual positions, improving the randomness and diversity of the individual initial positions, increasing the coverage rate of the search space, generating opposite positions to better maintain the diversity of individuals, and prompting the search algorithm to explore globally. Based on the individual position transfer probability and the roulette wheel selection method, the individual transfer direction during position update is calculated, which is beneficial to retaining and spreading excellent genetic information. Position update is performed based on the adaptive perturbation adjustment factor, enabling individuals to better balance the ability of global exploration and local mining during the search process, helping to improve the convergence speed of the algorithm and avoid falling into local optimal solutions, so that the constructed ultrasonic quality evaluation model can more accurately evaluate the quality of ultrasonic images and improve the accuracy and robustness of the model.

[0099] Example Five, refer to Figure 1 , based on the above example, the ultrasonic image quality evaluation module acquires the ultrasonic images to be evaluated, performs preprocessing, and then inputs them into the ultrasonic image quality evaluation model. Based on the labels output by the ultrasonic image quality evaluation model, the quality grade of the ultrasonic images is obtained.

[0100] It should be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0101] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0102] The above description of the present invention and its embodiments is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In summary, if those of ordinary skill in the art are inspired by this and, without departing from the gist of the present invention, design similar structural forms and embodiments to this technical solution without creative efforts, they shall fall within the protection scope of the present invention.

Claims

1. An ultrasound image quality assessment system based on artificial intelligence, characterized in that: It includes a data acquisition module, a data preprocessing module, an ultrasonic image feature information processing module, a module for constructing an ultrasonic image quality evaluation model, and an ultrasonic image quality evaluation module; The data acquisition module acquires ultrasonic images and quality evaluation grades; The data preprocessing module performs denoising, enhancement, and smoothing processing on the acquired ultrasonic images; The ultrasonic image feature information processing module calculates the gradients of the ultrasonic image in four directions to obtain the merged fusion gradient and the differential fusion gradient, introduces a compensation coefficient and a correction coefficient for feature fusion, constructs a feature matrix, then uses a serpentine scan to obtain feature vectors, and performs feature dimensionality reduction; The module for constructing an ultrasonic image quality evaluation model determines the initial position of an individual based on the Fuch mapping and the opposite position, calculates the individual transfer direction during position update based on the individual position transfer probability and the roulette selection method, performs position update based on the adaptive perturbation adjustment factor, and finds the optimal parameter position to construct the model; The ultrasonic image quality evaluation module obtains the quality grade of the ultrasonic image based on the label output by the ultrasonic image quality evaluation model.

2. The ultrasonic image quality evaluation system based on artificial intelligence according to claim 1, wherein: The ultrasonic image feature information processing module specifically includes the following: Calculate gradients. Convert the preprocessed ultrasonic image into a grayscale image to obtain a matrix C of size m×n. Calculate the right gradient, left gradient, upper gradient, and lower gradient of each pixel point (x, y) in matrix C. The formulas used are as follows: A rx = |f(x + 1, y) - f(x, y)|; A lx = |f(x, y) - f(x - 1, y)|; A uy = |f(x, y + 1) - f(x, y)|; A dy = |f(x, y) - f(x, y - 1)|; where A rx , A lx , A uy and A dy are the right gradient, left gradient, upper gradient and lower gradient of the pixel point (x, y) in matrix C respectively, x and y are the row index and column index of matrix C respectively, f(x, y) is the gray value at the pixel point (x, y) in matrix C, and f(x + 1, y), f(x - 1, y), f(x, y + 1) and f(x, y - 1) are the gray values of the pixel points adjacent to the right, left, upper and lower sides of the pixel point (x, y) respectively; Calculate the merged fusion gradient A fu , and the formula used is as follows: A fu =A rx +A lx +A uy +A dy ; Calculate the differential fusion gradient A su , and the formula used is as follows: A su = (A rx - A lx ) + (A uy - A dy ); Feature fusion. The formulas used are as follows: c(x, y) = c0(x, y) + g f × a fu (x, y) - g s × a su (x, y); Wherein, c(x, y) is the eigenvalue after the fusion processing of c0(x, y), c0(x, y) is the pixel value of the pixel point (x, y) in the matrix C, g f is the compensation coefficient for merging the fusion gradient, g s is the correction coefficient for the differential fusion gradient, a fu (x, y) is the merging fusion matrix of the pixel point (x, y), a su (x, y) is the differential fusion matrix of the pixel point (x, y); Construct a feature matrix. Go back to the step of calculating gradients for iteration until the fused feature values are calculated for all pixel points in matrix C. The matrix C of size m×n is transformed into a first feature matrix G of size (m - 2)×(n - 2). Use the adjacent value filling algorithm to transform matrix G into a second feature matrix Q of size m×n; Calculate feature vectors. Use a module of size z×z to block the second feature matrix Q, calculate the histogram of each block, and connect the histograms of each block in the connection order of the serpentine scan to obtain the feature vectors of the preprocessed ultrasonic image; Feature dimensionality reduction. Perform dimensionality reduction processing on the feature vectors and construct a feature dataset based on the dimensionality-reduced feature vectors.

3. An ultrasound image quality evaluation system based on artificial intelligence according to claim 1, characterized in that: The module for constructing an ultrasonic image quality evaluation model specifically includes the following: Initialize the individual position. Represent the parameter position with the individual position and initialize the individual position based on the Fuch mapping. The formula used is as follows: ； where K i，j is the initial position of the i-th individual in the j-th dimension, i is the individual index, j is the dimension index of the individual search space, UB j and LB j are the upper and lower bounds of the individual search space in the j-th dimension respectively, h i，j is the chaotic variable of the i-th individual in the j-th dimension using the Fuch mapping; Generate the opposite position. The formula used is as follows: ； Wherein, is the opposite position of K i，j ; Determine the initial position of the individual, take the performance of the ultrasonic image quality evaluation model established based on the parameter position as the fitness value of the corresponding individual position, sort the initial position and the opposing position from large to small according to the magnitude of the fitness value, and select the first N K individual positions as the final initial positions of the individuals; Train the ultrasonic image quality evaluation model. Use python to import the sklearn library and call the SVM function based on the current parameter position to train the ultrasonic image quality evaluation model based on the feature dataset; Calculate the individual position transfer probability, and select the individual position K according to the roulette wheel selection method b At time (t), when the individual position is updated, it transfers in the direction where K b is located at time (t). The formula used is as follows: ； ； Where P i，a (t) is the probability that the i-th individual moves to the position of the a-th individual at the t-th iteration, a is the individual index that is not equal to i, t is the iteration index, and K i (t) is the position of the i-th individual at the t-th iteration, K a (t) is the position of the ath individual at the tth iteration, J(K a (t)) is the relative fitness value of the ath individual at the tth iteration, d(K a (t)) is the fitness value of the i-th individual position at the t-th iteration, d worst (t) and d best (t) are the lowest and highest fitness values at the tth iteration, ε is the minimum value to avoid the denominator being zero, and ||K i (t)-K a (t)|| is K i (t) and K a (t) Euclidean distance between; Position update. Perform position update based on the adaptive perturbation adjustment factor. The formula used is as follows: ； ； where δ i (t) is the adaptive perturbation adjustment factor of the i-th individual at the t-th iteration, r1 is the first random number, λ is the position update step size, K best (t) is the global optimal position at the t-th iteration, and the global optimal position is the individual position with the highest fitness value, ||K best (t) - K i (t)|| is the Euclidean distance between K best (t) and K i (t), K i (t + 1) is the position of the i-th individual at the (t + 1)-th iteration, and r2 is the second random number; Model parameters are determined. The fitness value evaluation threshold μ and the maximum number of iterations T are preset. The fitness value of the individual and the global optimal position are updated. When the fitness value corresponding to the global optimal position is higher than the fitness value evaluation threshold μ, the parameter position corresponding to the global optimal position is the optimal parameter position of the model, and an ultrasonic image quality evaluation model is constructed based on the optimal parameter position; otherwise, if the maximum number of iterations T is reached, the individual position is re-initialized; otherwise, the position update continues.

4. An ultrasound image quality evaluation system based on artificial intelligence according to claim 1, characterized in that: The data acquisition module acquires ultrasonic images and quality evaluation grades, and the quality evaluation grades include excellent, good, average, poor, and invalid.

5. The ultrasonic image quality evaluation system based on artificial intelligence according to claim 1, wherein: The data preprocessing module preprocesses the acquired ultrasonic images. The preprocessing includes denoising, enhancement, and smoothing, and the risk assessment grade is used as a label.

6. The ultrasound image quality evaluation system based on artificial intelligence according to claim 1, wherein: The ultrasonic image quality evaluation module acquires the ultrasonic images to be evaluated. After preprocessing, they are input into the ultrasonic image quality evaluation model, and based on the label output by the ultrasonic image quality evaluation model, the quality grade of the ultrasonic images is obtained.

Images