An artificial intelligence-based kidney stone extracorporeal lithotripsy detection system and method

Abstract

The application discloses an artificial intelligence-based kidney stone extracorporeal lithotripsy detection system and method, and belongs to the technical field of kidney stone extracorporeal lithotripsy detection. In order to improve the detection accuracy of the ultrasonic image of the kidney stone extracorporeal lithotripsy, the application comprises the following steps: obtaining the B-ultrasonic image of the kidney of a patient with kidney stones, performing data processing, adopting an HRNe model to combine mean filtering and Sobel operator processing technology to construct a kidney stone recognition model of the ultrasonic image, evaluating the trained kidney stone recognition model of the ultrasonic image by using a verification set, calculating the accuracy, sensitivity and specificity, evaluating the accuracy and reliability of the trained kidney stone recognition model of the ultrasonic image in the kidney stone recognition task, further optimizing the trained kidney stone recognition model of the ultrasonic image, obtaining the stone features output by the optimized kidney stone recognition model of the ultrasonic image, calculating the kidney stone extracorporeal lithotripsy energy, and planning a kidney stone extracorporeal lithotripsy scheme. The application provides a more accurate scheme.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of kidney stone extracorporeal lithotripsy detection, and particularly relates to an ultrasonic image-based kidney stone extracorporeal lithotripsy detection system and method based on artificial intelligence. BACKGROUND

[0002] ESWL is a non-invasive treatment method that uses high-energy shock waves to focus on and act on stones in vitro, causing them to gradually peel off and break from the surface, and ultimately be excreted out of the body with urine. This technology has been widely used in the treatment of kidney stones and has become one of the preferred methods for urolithiasis. B-ultrasound image acquisition: During treatment, B-ultrasound equipment is used to collect images of the kidney and stones. These images provide key information such as the location, size, and shape of the stones, which is crucial for guiding the positioning and emission of shock waves.

[0003] Although artificial intelligence has shown great potential in the diagnosis and treatment of kidney stone extracorporeal lithotripsy based on B-ultrasound images, there are still some defects in the existing technology;

[0004] 1. Image quality limitations: The quality of B-ultrasound images is affected by various factors such as patient position, abdominal wall fat thickness, stone composition and size. For smaller or deeper stones, B-ultrasound images may not be clear enough, making it difficult for AI algorithms to accurately identify and analyze.

[0005] 2. Algorithm accuracy: Although AI algorithms perform well in image analysis, there is still a possibility of misjudgment and omission. This may be due to limitations of the algorithm itself, insufficient training data, or differences in image quality.

[0006] 3. Operation complexity: Although AI algorithms can simplify the interpretation of B-ultrasound images and the positioning of shock waves, the entire diagnosis and treatment system still requires professional operators to monitor and adjust. In addition, the operation of AI algorithms also requires high-performance computing devices and professional technical support.

[0007] 4. Technology popularization: Currently, not all medical institutions have the conditions and ability to introduce artificial intelligence-based B-ultrasound image-based kidney stone extracorporeal lithotripsy diagnosis and treatment systems. This limits the popularization and application range of the technology.

[0008] In summary, artificial intelligence has broad application prospects in the diagnosis and treatment of kidney stone extracorporeal lithotripsy based on B-ultrasound images, but still needs to overcome some technical challenges and limitations. SUMMARY

[0009] The problem to be solved by the present application is to improve the accuracy of ultrasonic image-based kidney stone extracorporeal lithotripsy detection, and to propose an ultrasonic image-based kidney stone extracorporeal lithotripsy detection system and method based on artificial intelligence.

[0010] To achieve the above object, the present application is realized by the following technical solutions:

[0011] A kidney stone extracorporeal lithotripsy detection method based on an artificial intelligence ultrasonic image, comprising the following steps:

[0012] S1. Obtain a kidney B-ultrasound image of a patient with kidney stones, classify the kidney B-ultrasound image through an image labeling tool, and construct a data set;

[0013] S2. Use a data processing unit to process the data set obtained in step S1, perform data cleaning, image interpolation, and standardization processing to obtain a data set after data processing, and divide the data set into a training set and a validation set;

[0014] S3. The artificial intelligence algorithm unit adopts an HRNe model combined with mean filtering and Sobel operator processing technology to construct a kidney stone recognition model of an ultrasonic image, trains the constructed kidney stone recognition model of the ultrasonic image using the training set of step S2, obtains a trained kidney stone recognition model of the ultrasonic image, and outputs a kidney stone recognition result of the ultrasonic image;

[0015] S4. The model optimization unit evaluates the trained kidney stone recognition model of the ultrasonic image obtained in step S3 using the validation set, calculates the accuracy, sensitivity, and specificity, evaluates the accuracy and reliability of the trained kidney stone recognition model of the ultrasonic image in the kidney stone recognition task, further optimizes the trained kidney stone recognition model of the ultrasonic image, and obtains an optimized kidney stone recognition model of the ultrasonic image;

[0016] S5. The result interaction unit calculates the kidney stone extracorporeal lithotripsy energy based on the stone characteristics output by the optimized kidney stone recognition model of the ultrasonic image in step S4, and plans a kidney stone extracorporeal lithotripsy scheme.

[0017] Further, the specific implementation method of step S2 image interpolation comprises the following steps:

[0018] S2.1. For the image data in the data set obtained in step S1, first perform linear interpolation in the direction:

[0019]

[0020] Wherein, f is the interpolation function, R1 is the coordinate of the first pixel point on the image, is the 11 axis coordinate of Q , is the 21 axis coordinate of Q , is the axis coordinate of the first pixel point coordinate, and Q11 is the first interpolation point of R1 in the direction of 21 is the first interpolation point of R1 in the direction of is the first interpolation point of R1 in the direction of is the first interpolation point of R1 in the direction of

[0021]

[0022] is the first interpolation point of R1 in the direction of 12 is the first interpolation point of R1 in the direction of 22 is the first interpolation point of R1 in the direction of is the first interpolation point of R1 in the direction of is the first interpolation point of R1 in the direction of

[0023] S2.2. Then linear interpolation is performed in the direction of

[0024]

[0025] is the first interpolation point of R1 in the direction of

[0026] S2.3. The image interpolation result is obtained by combining step S2.1 and step S2.2 as follows:

[0027]

[0028] Further, the specific implementation method of step S3 includes the following steps:

[0029] S3.1. Constructing the HRNe model, including downsampling through two convolutional layers with a kernel size of 3x3 and a stride of 2, then repeatedly stacking through the Layer1 module, and adding the stacking result to finally obtain the fusion output of the 4 times down-sampling branch through the ReLU function; the fusion output of the 4 times down-sampling branch includes combining the output of the 8 times down-sampling branch after 2 times up-sampling, the output of the 16 times down-sampling branch after 4 times up-sampling, and the outputs of the three branches are activated again through the ReLU to form the final fusion output of the model;

[0030] The expression of the HRNe model is finally obtained as follows:

[0031] F out = H HRNet (F in )

[0032] ​​​​​​wherein, F out represents the feature map output by the HRNe model, H HRNet represents the processing procedure of the HRNet algorithm, F in represents the input convolutional image;

[0033] S3.2. The feature map output by the HRNe model obtained in step S3.1 is smoothed by using the mean filtering method. For each pixel point in the image, the filtered pixel value of the pixel point is calculated according to the following formula:

[0034]

[0035] wherein, M represents the length of the mean filtering window, N represents the width of the mean filtering window, is the pixel value of the pixel point in the feature map output by the HRNe model;

[0036] S3.3. The pixel value filtered in step S3.2 is subjected to edge processing in the image by using the Sobel operator. The template in the horizontal direction is set as:

[0037]

[0038] For the pixel point in the image, the gradient amplitude in the horizontal direction is:

[0039]

[0040] The template in the vertical direction is set as:

[0041]

[0042] The gradient amplitude in the vertical direction is:

[0043]

[0044] The total gradient amplitude is:

[0045]

[0046] Further, the calculation formula of the accuracy acc in step S4 is:

[0047]

[0048] Wherein, TP is the number of samples actually with stones and predicted by the model as having stones; TN is the number of samples actually without stones and predicted by the model as without stones; FP is the number of samples actually without stones but predicted by the model as having stones; and FN is the number of samples actually with stones but predicted by the model as without stones.

[0049] The calculation formula of sensitivity TPR is as follows:

[0050]

[0051] The calculation formula of specificity TNR is as follows:

[0052]

[0053] Further, the specific implementation method of step S5 includes the following steps:

[0054] S5.1. Collect the stone features output by the kidney stone recognition model of the optimized ultrasonic image in step S4, and combine the specific conditions of the patient, including stone size, position and shape;

[0055] S5.2. Apply SVM to build a classification model to comprehensively calculate the different features of the stone obtained in step S5.1, and divide the hyperplane in the sample space of the stone as follows:

[0056] w T +b=y i

[0057] Wherein, w=(w1, w2, …, w d ) is the parameter feature of the stone, b is the distance of the hyperplane, y i is the classification of the patient to select the stone breaking energy;

[0058] S5.3. The physician selects the parameters of the extracorporeal stone breaking device according to experience combined with the model parameters obtained in step S5.2, including shock wave intensity, frequency, focal point, and plans the extracorporeal stone breaking scheme of kidney stones.

[0059] An ultrasonic image kidney stone extracorporeal lithotripsy detection system based on artificial intelligence is realized by relying on the ultrasonic image kidney stone extracorporeal lithotripsy detection method based on artificial intelligence, and includes a data processing unit, an artificial intelligence algorithm unit, a model optimization unit and a result interaction unit.

[0060] The beneficial effects of the present application are as follows:

[0061] The kidney stone extracorporeal lithotripsy detection method based on the artificial intelligence of the ultrasonic image significantly improves the resolution and feature extraction accuracy of the ultrasonic image through the deep fusion image interpolation technology, the high-resolution network (HRNet) algorithm and the mean filter method, and then applies the support vector machine to select the setting of the lithotripsy energy of the stone, so that a more accurate and personalized solution is provided for the extracorporeal lithotripsy treatment of the kidney stone.

[0062] The kidney stone extracorporeal lithotripsy detection method based on the artificial intelligence of the ultrasonic image utilizes the deep learning and other frontier technologies to automatically analyze and extract features of the ultrasonic image. Through training and optimization of the algorithm model, the system can automatically identify and locate the kidney stone, and accurately measure and analyze key features such as the size, shape and position of the kidney stone. These key features will serve as an important basis for extracorporeal lithotripsy treatment decision-making, providing doctors with more accurate and reliable diagnostic results and treatment recommendations. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 A flowchart of the kidney stone extracorporeal lithotripsy detection method based on the artificial intelligence of the ultrasonic image is provided.

[0064] Figure 2 A structure block diagram of the kidney stone extracorporeal lithotripsy detection method based on the artificial intelligence of the ultrasonic image is provided.

[0065] Figure 3 A connection relationship diagram of the kidney stone extracorporeal lithotripsy detection system based on the artificial intelligence of the ultrasonic image is provided. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the present application clearer, the following further describes the present application in conjunction with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application, i.e., the described specific embodiments are only a part of the embodiments of the present application, but not all the specific embodiments. The components of the specific embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations, and the present application can also have other embodiments.

[0067] Therefore, the following detailed description of the specific embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected specific embodiments of the present application. Based on the specific embodiments of the present application, all other specific embodiments obtained by those skilled in the art without making creative efforts fall within the scope of the present application.

[0068] To further understand the invention's content, features, and effects, the following specific embodiments are provided, along with accompanying drawings. Figure 1 -Appendix Figure 3 Detailed explanation is as follows:

[0069] Example 1:

[0070] An artificial intelligence-based extracorporeal shock wave lithotripsy (ESWL) method for detecting kidney stones using ultrasound images includes the following steps:

[0071] S1. Obtain kidney ultrasound images of patients with kidney stones, classify the kidney ultrasound images using image annotation tools, and construct a dataset;

[0072] S2. The dataset obtained in step S1 is processed by the data processing unit, which performs data cleaning, image interpolation, and standardization to obtain the processed dataset, which is then divided into a training set and a validation set.

[0073] Furthermore, the specific implementation method of image interpolation in step S2 includes the following steps:

[0074] S2.1. For the image data in the dataset obtained in step S1, firstly... Linear interpolation in the direction:

[0075]

[0076] Where f is the interpolation function, and R1 is the coordinate of the first pixel in the image. For Q 11 of Axis coordinates For Q 21 of Axis coordinates The coordinates of the first pixel Q axis coordinates 11 For R1 in The first interpolation point in the direction, Q 21 For R1 in The second interpolation point in the direction, The coordinates of the first pixel Axis coordinates

[0077]

[0078] Among them, Q 12 For R2 in The first interpolation point in the direction, Q 22 For R2 in The second interpolation point in the direction, R2 is the coordinate of the second pixel in the image. The coordinates of the second point axis coordinates,

[0079] S2.2. Then linear interpolation is performed in the direction of

[0080]

[0081] where f(P) is a function of linear interpolation of R1 and R2 in the direction of

[0082] S2.3. The image interpolation result is obtained by combining step S2.1 and step S2.2 as follows:

[0083]

[0084] Further, the image interpolation method is used to preprocess the ultrasound image. Image interpolation can increase the resolution of the image by increasing the number of pixels in the image, thereby more clearly showing the detailed structure of the stone.

[0085] S3. The artificial intelligence algorithm unit adopts the HRNe model combined with mean filtering and Sobel operator processing technology to construct a kidney stone recognition model of the ultrasound image, trains the constructed kidney stone recognition model of the ultrasound image using the training set of step S2, obtains a trained kidney stone recognition model of the ultrasound image, and outputs the kidney stone recognition result of the ultrasound image;

[0086] Further, the specific implementation method of step S3 includes the following steps:

[0087] S3.1. Constructing the HRNe model, including down-sampling through two convolutional layers with a convolution kernel size of 3x3 and a step of 2, then repeatedly stacking through the Layer1 module, adding the stacking result, and finally obtaining the fusion output of the down-sampling 4 times branch through the ReLU function; the fusion output of the down-sampling 4 times branch includes combining the output of the down-sampling 8 times branch again through 2 times up-sampling, the output of the down-sampling 16 times branch again through 4 times up-sampling, and the outputs of the three branches again through ReLU activation to form the final fusion output of the model;

[0088] The expression of the HRNe model is finally obtained as follows:

[0089] F out = H HRNet (F in )

[0090] where F out represents the feature map output by the HRNe model, H HRNet represents the processing process of the HRNet algorithm, and F in represents the input convolutional image.​​

[0091] Further, HRNet (High-Resolution Network) is used as the core algorithm for feature extraction of B-ultrasound images. HRNet can maintain high-resolution features throughout the network process through parallel multi-resolution paths and repeated multi-scale feature fusion, thereby effectively capturing the subtle structures in ultrasound images.

[0092] Further, the process of upsampling or downsampling Xi from resolution i to resolution k is contained in the function a(Xi, k), and the expression is:

[0093] X i = a(X i , k)

[0094] The calculation formula of loss M is:

[0095]

[0096] Where n represents the number of samples, y i represents the actual value, y p represents the predicted value, and ∑ represents the summation of all samples.

[0097] S3.2. The feature map output by the HRNe model obtained in step S3.1 is smoothed using the mean filtering method. For each pixel point in the image, the pixel value of the filtered pixel point is calculated as follows:

[0098]

[0099] Where M represents the length of the mean filtering window, N represents the width of the mean filtering window, is the pixel value of the point in the feature map output by the HRNe model;

[0100] After the HRNet algorithm is processed, the extracted feature map is smoothed using the mean filtering method. Mean filtering is a simple image processing technique that removes noise by calculating the average value of the neighborhood around the pixel point. The window size of mean filtering can be adjusted according to the actual situation to remove noise while ensuring image clarity.

[0101] S3.3. The pixel value filtered in step S3.2 is processed using the Sobel operator to detect edges in the image, and the horizontal direction template is set as:

[0102]

[0103]

[0104] ​For a pixel point in the image The gradient amplitude in the horizontal direction is:

[0105]

[0106] The template in the vertical direction is set as:

[0107]

[0108] The gradient amplitude in the vertical direction is:

[0109]

[0110] The total gradient amplitude is:

[0111]

[0112] The main purpose of mean filtering is to remove noise and smooth the image, making the model identification more accurate, while the main purpose of Sobel operator is to detect edges in the image. Mean filtering achieves smoothing effect by calculating the mean value of pixels within the window, while Sobel operator detects edges by calculating the approximate value of image gradient. Both mean filtering and Sobel operator use convolution kernel, but their convolution kernel design and function are different. The convolution kernel of mean filtering is usually square, and the weight of each element is equal; while the convolution kernel of Sobel operator is specially designed to capture the gray level change in the image and highlight the edge features. The combination of the two methods helps to improve the accuracy of image recognition.

[0113] S4. The model optimization unit evaluates the trained ultrasonic image kidney stone recognition model obtained in step S3 using the validation set, calculates the accuracy, sensitivity and specificity, evaluates the accuracy and reliability of the trained ultrasonic image kidney stone recognition model in the kidney stone recognition task, further optimizes the trained ultrasonic image kidney stone recognition model, and obtains an optimized ultrasonic image kidney stone recognition model;

[0114] Further, the calculation formula of the accuracy acc in step S4 is:

[0115]

[0116] Where TP is the number of samples that actually have stones and are predicted by the model to have stones; TN is the number of samples that actually have no stones and are predicted by the model to have no stones; FP is the number of samples that actually have no stones but are predicted by the model to have stones; FN is the number of samples that actually have stones but are predicted by the model to have no stones;

[0117] The calculation formula of the sensitivity TPR is:

[0118]

[0119] The formula for calculating specificity TNR is:

[0120]

[0121] S5. The results interaction unit calculates the extracorporeal shock wave lithotripsy energy for kidney stones based on the stone characteristics output by the kidney stone identification model of the ultrasound image optimized in step S4, and plans an extracorporeal shock wave lithotripsy scheme for kidney stones.

[0122] Furthermore, the specific implementation method of step S5 includes the following steps:

[0123] S5.1. Collect the stone features output by the kidney stone identification model from the ultrasound images optimized in step S4, and combine them with the patient's specific situation, including stone size, location, and shape.

[0124] S5.2. Apply SVM to construct a classification model to comprehensively calculate the different features of the stones obtained in step S5.1. The formula for dividing the hyperplane in the stone sample space is as follows:

[0125] w T +b=y i

[0126] Where w = (w1, w2, ..., w d ) represents the parametric characteristics of the stone, b represents the distance to the hyperplane, and y represents the distance to the hyperplane. i Select the appropriate lithotripsy energy level for the patient;

[0127] S5.3. Based on experience and the model parameters obtained in step S5.2, the physician selects the parameters of the extracorporeal shock wave lithotripsy device, including shock wave intensity, frequency, and focal point, and plans an extracorporeal shock wave lithotripsy plan for kidney stones.

[0128] Example 2:

[0129] An artificial intelligence-based extracorporeal shock wave lithotripsy (ESWL) system for detecting kidney stones using ultrasound images is implemented based on the artificial intelligence-based ESWL method for detecting kidney stones using ultrasound images described in Example 1. The system is characterized by comprising a data processing unit, an artificial intelligence algorithm unit, a model optimization unit, and a result interaction unit. The data processing unit is connected to the artificial intelligence algorithm unit, the artificial intelligence algorithm unit is connected to the model optimization unit, and the model optimization unit is connected to the result interaction unit.

[0130] It has to be noted that the terms "first", "second", and the like in connection with an entity or action refer to this entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without further constraints, exclude the presence of additional elements of the process, method, article, or apparatus.

[0131] While the application has been described with reference to specific implementations thereof, it should be understood that various modifications and substitutions can be made by those skilled in the art without departing from the scope of the present application. In particular, any one of the features of the present application disclosed above can be utilized independently of any other and the scope of the application should not be limited by the specific embodiments disclosed herein, but should be given the widest coverage possible in its true scope.

Claims

1. A method for detecting kidney stones via extracorporeal shock wave lithotripsy based on ultrasound images using artificial intelligence, characterized in that, Includes the following steps: S1. Obtain kidney ultrasound images of patients with kidney stones, classify the kidney ultrasound images using image annotation tools, and construct a dataset; S2. The data processing unit processes the dataset obtained in step S1 by performing data cleaning, image interpolation, and standardization to obtain the processed dataset, which is then divided into a training set and a validation set. S3. The artificial intelligence algorithm unit uses the HRNe model combined with mean filtering and Sobel operator processing techniques to construct a kidney stone recognition model for ultrasound images. The constructed kidney stone recognition model for ultrasound images is trained using the training set from step S2 to obtain a trained kidney stone recognition model for ultrasound images, and the kidney stone recognition result of ultrasound images is output. S4. The model optimization unit uses the validation set to evaluate the trained ultrasound image kidney stone recognition model obtained in step S3, calculates the accuracy, sensitivity and specificity, evaluates the accuracy and reliability of the trained ultrasound image kidney stone recognition model in the kidney stone recognition task, and further optimizes the trained ultrasound image kidney stone recognition model to obtain the optimized ultrasound image kidney stone recognition model. S5. The result interaction unit calculates the extracorporeal lithotripsy energy for kidney stones based on the stone characteristics output by the kidney stone identification model of the ultrasound image optimized in step S4, and plans an extracorporeal lithotripsy scheme for kidney stones. The specific implementation method of step S5 includes the following steps: S5.

1. Collect the stone features output by the kidney stone identification model from the ultrasound images optimized in step S4, and combine them with the patient's specific situation, including stone size, location, and shape. S5.

2. Apply SVM to construct a classification model to comprehensively calculate the different features of the stones obtained in step S5.

1. The formula for dividing the hyperplane in the stone sample space is as follows: ； in, =(w1, w2, ..., w d ) represents the parameter characteristics of the stone. The distance is the hyperplane. Select the appropriate lithotripsy energy level for the patient; S5.

3. Based on experience and the model parameters obtained in step S5.2, the physician selects the parameters of the extracorporeal shock wave lithotripsy device, including shock wave intensity, frequency, and focal point, and plans an extracorporeal shock wave lithotripsy plan for kidney stones.

2. The method for detecting kidney stones using extracorporeal shock wave lithotripsy based on ultrasound images according to claim 1, characterized in that, The specific implementation method of image interpolation in step S2 includes the following steps: S2.

1. For the image data in the dataset obtained in step S1, first in... Linear interpolation in the direction: f = f + f ; Where f is the interpolation function. The coordinates of the first pixel in the image. for of Axis coordinates for of Axis coordinates The coordinates of the first pixel Axis coordinates for exist The first interpolation point in the direction, for exist The second interpolation point in the direction, The coordinates of the first pixel Axis coordinates = ; f = f + f ; in, for exist The first interpolation point in the direction, for exist The second interpolation point in the direction, The coordinates of the second pixel in the image. The coordinates of the second point Axis coordinates = ; S2.

2. Then... Linear interpolation in the direction: f = f + f ; Among them, f for and exist A function that performs linear interpolation in a specific direction; S2.

3. Combining steps S2.1 and S2.2, the image interpolation result is as follows: f = f + f + f + f 。 3. The method for detecting kidney stones using extracorporeal shock wave lithotripsy based on ultrasound images according to claim 2, characterized in that, The specific implementation method of step S3 includes the following steps: S3.

1. Construct the HRNe model, which includes downsampling using two convolutional layers with a kernel size of 3x3 and a stride of 2, then repeatedly stacking them using the Layer1 module, summing the stacked results, and finally obtaining the fused output of the 4x downsampled branch through the ReLU function; the fused output of the 4x downsampled branch includes combining the output of the 8x downsampled branch and then upsampling it by 2x, the output of the 16x downsampled branch and then upsampling it by 4x, and the outputs of the three branches are activated again by ReLU to form the final fused output of the model; The final expression for the HRNe model is: ； in, This represents the feature map output by the HRNe model. This represents the processing steps of the HRNet algorithm. Represents the input convolutional image; S3.

2. The feature map output by the HRNe model obtained in step S3.1 is smoothed using the mean filtering method. For each pixel in the image, its filtered... pixel value g The calculation formula is as follows: ; in, This indicates the length of the mean filter window. This indicates the width of the mean filter window. ( ) is in the feature map output by the HRNe model. ) pixel value; S3.

3. Apply the Sobel operator to the filtered pixel values ​​from step S3.2 to perform edge processing in the image, setting... The template for the horizontal direction is: ； For a pixel (x, y) in an image, the gradient magnitude in the horizontal direction (𝒳,𝒴) is: (𝒳,𝒴)= + ; Set the template for the vertical direction as follows: ； The gradient magnitude in the vertical direction (𝒳,𝒴) is: (𝒳,𝒴)= + ; The total gradient magnitude is: G(𝒳,𝒴)= 。 4. The method for detecting kidney stones using extracorporeal shock wave lithotripsy based on ultrasound images according to claim 3, characterized in that, The formula for calculating accuracy acc in step S4 is: ； Wherein, TP is the number of samples that actually have stones and are predicted by the model to have stones; TN is the number of samples that actually do not have stones and are predicted by the model to have stones; FP is the number of samples that actually do not have stones but are predicted by the model to have stones; and FN is the number of samples that actually have stones but are predicted by the model to have stones. Sensitivity The calculation formula is: ； Specificity The calculation formula is: 。 5. An artificial intelligence-based extracorporeal shock wave lithotripsy (ESWL) system for detecting kidney stones using ultrasound images, implemented based on the artificial intelligence-based ESWL method for detecting kidney stones using ultrasound images as described in any one of claims 1-4, characterized in that, It includes a data processing unit, an artificial intelligence algorithm unit, a model optimization unit, and a result interaction unit. The data processing unit is connected to the artificial intelligence algorithm unit, the artificial intelligence algorithm unit is connected to the model optimization unit, and the model optimization unit is connected to the result interaction unit.

Images