A B-ultrasound image analysis method for pneumothorax diagnosis

Abstract

The present invention relates to a B-ultrasound image analysis method for pneumothorax diagnosis, comprising the following steps: obtaining an ultrasound examination video of a lung area and preprocessing it; using a one-dimensional block matching algorithm to process the ultrasound examination video to obtain a video feature matrix; inputting the video feature matrix into a trained machine learning model, the output of the machine learning model being a diagnosis result; and outputting the video feature matrix and the diagnosis result. Pneumothorax can be diagnosed based on lung sliding signs in B-ultrasound images, but it takes a certain amount of time for doctors to judge by naked eyes and requires certain skills and experience for doctors. Compared with the prior art, the present invention uses a one-dimensional block matching algorithm to process ultrasound examination videos, and the obtained video feature matrix can intuitively and quantitatively reflect the movement of lung sliding signs in ultrasound videos, thereby improving the efficiency and accuracy of pneumothorax diagnosis through computer technology.

Description

Technical Field

[0001] The invention relates to the technical field of pneumothorax diagnosis, and in particular to a B-ultrasound image analysis method for pneumothorax diagnosis. Background Art

[0002] Pneumothorax refers to the state of gas accumulation caused by the entry of gas into the pleural cavity. It is mostly caused by lung disease or external force, which causes the rupture of lung tissue and visceral pleura, or the rupture of tiny emphysematous bubbles near the surface of the lung, and the air in the lung and bronchi escapes into the pleural cavity. Pneumothorax caused by chest wall or lung trauma is called traumatic pneumothorax; pneumothorax caused by spontaneous rupture of lung tissue due to disease is called "spontaneous pneumothorax", and artificial injection of air into the pleural cavity for treatment or diagnosis is called "artificial pneumothorax". Pneumothorax can be divided into closed pneumothorax, open pneumothorax and tension pneumothorax. Spontaneous pneumothorax is more common in young and middle-aged men or those with chronic bronchitis, emphysema, and tuberculosis. This disease is one of the pulmonary emergencies. In severe cases, it can be life-threatening and can be cured with timely treatment.

[0003] In recent years, ultrasound has been used to diagnose pneumothorax, with good results. Ultrasound images of the lungs can be divided into three types based on texture changes: lung sliding (normal), lung point sign, and disappearance of lung sliding (pneumothorax), which are the main basis for diagnosing whether a person has pneumothorax. However, the disadvantage of lung ultrasound is that its use requires certain technical skills from medical personnel, and experience will affect the results of the diagnosis. The increase in the number of chest radiographs and the technical requirements for diagnosing pneumothorax have increased the workload of medical personnel and reduced the efficiency of diagnosis. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a B-ultrasound image analysis method for pneumothorax diagnosis.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A B-ultrasound image analysis method for pneumothorax diagnosis comprises the following steps:

[0007] S1, obtaining an ultrasound examination video of the lung area and preprocessing it;

[0008] S2, using a one-dimensional block matching algorithm to process the ultrasound inspection video to obtain a video feature matrix;

[0009] S3, inputting the video feature matrix into a trained machine learning model, wherein the output of the machine learning model is a diagnosis result;

[0010] S4. Output the video feature matrix and diagnosis results.

[0011] Preferably, in step S1, a B-mode ultrasound device is used to obtain an ultrasound examination video of the lung area.

[0012] Preferably, in step S1, the preprocessing of the ultrasound examination video is specifically as follows: in each frame of the ultrasound examination video, grayscale processing is performed on each frame based on brightness.

[0013] Preferably, step S2 includes the following steps:

[0014] S21. Set n×m macroblocks in the first frame of the ultrasound examination video. The size of each macroblock is k×k pixels. Record the center points of each macroblock, set a 1×k weight coefficient matrix A, and the value of the elements in the weight coefficient matrix is 1 / 256;

[0015] S22. Obtain the pixel values of each pixel point in each macroblock to obtain n×m macroblock pixel matrices. Multiply the weight coefficient matrix A by each macroblock pixel matrix to obtain the central feature matrix of each macroblock. The shape of the central feature matrix is 1×k. Let i = 2;

[0016] S23. Obtain the i-th frame image, obtain the center points of each macroblock in the (i - 1)-th frame image, set search points in the i-th frame image according to the center points of each macroblock in the (i - 1)-th frame image, calculate the feature matrix of the macroblock at the search points, compare the feature matrices of each search point of the macroblock with the central feature matrix of the macroblock in the (i - 1)-th frame image, determine the center points of each macroblock in the i-th frame image, and calculate the central feature matrix of each macroblock;

[0017] S24. For each macroblock, record the distance between its center point in the (i - 1)-th frame image and the center point in the i-th frame image as the modulus of the (i - 1)-th horizontal motion vector of the macroblock;

[0018] S25. Let i + 1, and repeat step S23 until the moduli of the 1st, 2nd, …, K - 1 horizontal motion vectors of each macroblock are obtained, where K is the total number of frames of the ultrasound examination video;

[0019] S26. For each macroblock, accumulate the moduli of the 1st, 2nd, …, K - 1 horizontal motion vectors of the macroblock to obtain the motion feature of each macroblock, and establish an n×m video feature matrix. The elements in the video feature matrix are the motion features of each macroblock.

[0020] Preferably, in step S21, the n×m macroblocks are evenly distributed in the first frame image, and the distribution shape is n rows and m columns, and n×m×k×k < N×M, where N×M is the size of each frame image in the ultrasound examination video.

[0021] Preferably, in step S23, determining the center point of a macroblock in the i-th frame image includes the following steps:

[0022] A1. Project the center point of the macroblock in the i-1th frame image to the i-th frame image, record it as the center search point, and record the r pixels to the left and right of the center search point as the left search point and the right search point, where r is an integer greater than or equal to 1;

[0023] A2. Take the center search point, left search point and right search point as the center point of the macroblock, obtain the pixel value of each pixel in the macroblock, and obtain the macroblock pixel matrix. Multiply the weight coefficient matrix A with the macroblock pixel matrix to obtain the feature matrix of the macroblock at the center search point, left search point and right search point, denoted as T L , T E and T R ;

[0024] A3. Let the central feature matrix of the macroblock in the i-1th frame image be T, and calculate T and T respectively. L , T E and T R The absolute error sum is the sum of the absolute values ​​of the differences between the corresponding elements in the two matrices;

[0025] A4, judging whether the search termination condition is met, if yes, outputting the coordinates of the search point with the minimum absolute error as the center point of the macroblock in the i-th frame image, and calculating the central feature matrix of the macroblock, otherwise, executing step A5;

[0026] A5. Take the search point with the smallest absolute error as the central search point, reduce r according to the preset step size, obtain the left search point and the right search point, and execute step A2.

[0027] Preferably, in step A4, the search termination condition is: the sum of absolute errors of the central search point is less than the sum of absolute errors of the left search point and the right search point or r<1.

[0028] Preferably, in step A5, r is set to half of its original value to obtain a new r value, and r pixels to the left and right of the center search point are recorded as the left search point and the right search point.

[0029] Preferably, in step S24, the modulus of the horizontal motion vector is a natural number.

[0030] Preferably, the calculation part required for the one-dimensional block matching algorithm is converted into a digital gate circuit through FPGA, and the calculation is performed in hardware, which is implemented by a hardware acceleration unit. The hardware acceleration unit includes a data interface unit, a data register and a parallel operation unit. The data interface unit is used to realize data transmission between the hardware acceleration unit and an external device. The data register is used to temporarily store data in the calculation process of the one-dimensional block matching algorithm. The parallel operation unit includes multiple parallel operation units for realizing the calculation of the one-dimensional block matching algorithm.

[0031] Preferably, the machine learning model is a k-nearest neighbor classifier.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] (1) In B-ultrasound images, pneumothorax can be diagnosed based on the lung sliding sign, but it takes a certain amount of time for doctors to judge by naked eyes and requires certain skills and experience. Therefore, this application uses a one-dimensional block matching algorithm to process the ultrasound examination video. The obtained video feature matrix can intuitively and quantitatively reflect the movement of the lung sliding sign in the ultrasound video, thereby improving the efficiency and accuracy of pneumothorax diagnosis through computer technology.

[0034] (2) In the one-dimensional block matching algorithm, considering that the actual lung sliding movement in the column direction has little impact on the diagnosis of pneumothorax and adding the movement calculation in the column direction will increase the complexity of the calculation and reduce the operation rate, the calculation of the motion vector in the column direction is cancelled, and only the motion vector in the horizontal direction is used as the judgment basis. This simplifies the algorithm and reduces the possibility of algorithm error to a certain extent.

[0035] (3) Based on FPGA technology and the algorithmic characteristics of the one-dimensional block matching algorithm, a hardware acceleration strategy combining software and hardware can be implemented. The algorithm operation part is converted into hardware through customized instructions. With its high speed and functional specialization, it can meet the higher speed requirements of certain specific situations. Customized hardware designed for different needs can also broaden its scope of use to a certain extent.

[0036] (4) Although the possibility of error is extremely small, in order to avoid errors and provide a variety of judgment bases, the present application simultaneously outputs the diagnosis results and the corresponding video feature matrix, which can not only assist doctors in diagnosis but also verify each other, thereby reducing the burden on doctors and improving patient experience. At the same time, the joint diagnosis of the system and doctors makes the diagnosis and treatment process more accurate and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of the present invention;

[0038] Figure 2 Flowchart of the one-dimensional block matching algorithm. DETAILED DESCRIPTION

[0039] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0040] In the drawings, components with the same structure are indicated by the same numerical labels, and components with similar structures or functions are indicated by similar numerical labels. The size and thickness of each component shown in the drawings are arbitrarily shown, and the present invention does not limit the size and thickness of each component. In order to make the illustration clearer, some parts in the drawings are appropriately exaggerated.

[0041] Embodiment 1:

[0042] A B-ultrasound image analysis method for pneumothorax diagnosis, such as Figure 1 As shown, the following steps are included:

[0043] S1, obtaining an ultrasound examination video of the lung area and preprocessing it;

[0044] S2, using a one-dimensional block matching algorithm to process the ultrasound inspection video to obtain a video feature matrix;

[0045] S3. Input the video feature matrix into a trained machine learning model, and the output of the machine learning model is the diagnosis result. In this embodiment, the machine learning model is a k-nearest neighbor classifier.

[0046] S4. Output the video feature matrix and diagnosis results.

[0047] Specifically, in step S1: use the ultrasound probe of a B-type ultrasound instrument to perform ultrasound examination on the human lungs to obtain an ultrasound examination video of the lung area. The B-type ultrasound instrument is integrated with an integrated display. The collected ultrasound examination video is displayed on the integrated display for real-time observation by the doctor, and on the other hand, it is preprocessed and the diagnosis result is obtained through steps S2 to S4. The preprocessing of the ultrasound examination video is specifically: in each frame of the ultrasound examination video, each frame is grayed based on the brightness, and the formula is as follows:

[0048] Gray=W R ×R+W G ×G+W B ×B

[0049] Among them, W R , W G , W B is the preset weight value, Gray represents the gray value after grayscale processing, R, G, B represent the RGB channels of the pixel before grayscale processing, and W R +W G +W B =1, if we can set W R =W G =W B =1 / 3, we can also let W R =0.3,W G =0.59,WB = 0.11 and so on.

[0050] Specifically, as Figure 2 shown, step S2 includes the following steps:

[0051] S21. Set n×m macroblocks in the first frame of the ultrasound examination video. The size of each macroblock is k×k pixels. Record the center point of each macroblock. Set a 1×k weight coefficient matrix A, and the value of the elements in the weight coefficient matrix is 1 / 256; the n×m macroblocks are evenly distributed in the first frame image, and the distribution shape is n rows and m columns, where n×m×k×k < N×M, and N×M is the size of each frame image in the ultrasound examination video.

[0052] Considering that a search will be performed in subsequent step S23 and the search is carried out to the left and right from the center point of the macroblock, the distance between the macroblock and the left and right edges of the graph can be appropriately controlled to avoid out-of-bounds during the search process.

[0053] S22. Obtain the pixel values of each pixel point in each macroblock to obtain n×m macroblock pixel matrices. Multiply the weight coefficient matrix A by each macroblock pixel matrix to obtain the central feature matrix of each macroblock. The shape of the central feature matrix is 1×k. Let i = 2;

[0054] For example, if the center point coordinates of a macroblock are (50, 50), its macroblock pixel matrix is the grayscale values of all pixel points within the range of 30×30 around the coordinates (50, 50). A×macroblock pixel matrix gives the central feature matrix of this macroblock.

[0055] S23. Obtain the i-th frame image, obtain the center points of each macroblock in the (i - 1)-th frame image, set search points in the i-th frame image according to the center points of each macroblock in the (i - 1)-th frame image, calculate the feature matrix of the macroblock at the search points, compare the feature matrices of each search point of the macroblock with the central feature matrix of the macroblock in the (i - 1)-th frame image, determine the center points of each macroblock in the i-th frame image, and calculate the central feature matrix of each macroblock;

[0056] Specifically, determining the center point of a macroblock in the i-th frame image includes the following steps:

[0057] A1. Project the center point of the macroblock in the (i - 1)-th frame image onto the i-th frame image, denoted as the center search point. Denote the pixels at r pixels to the left and right of the center search point as the left search point and the right search point respectively, where r is an integer greater than or equal to 1; in this embodiment, r = 20;

[0058] A2. Take the center search point, left search point and right search point as the center point of the macroblock, obtain the pixel value of each pixel in the macroblock, and obtain the macroblock pixel matrix. Multiply the weight coefficient matrix A with the macroblock pixel matrix to obtain the feature matrix of the macroblock at the center search point, left search point and right search point, denoted as T L , T E and T R ;

[0059] A3. Let the central feature matrix of the macroblock in the i-1th frame image be T, and calculate T and T respectively. L , T E and T R The absolute error sum is the sum of the absolute values ​​of the differences between the corresponding elements in the two matrices;

[0060] like t d is the dth element in the matrix T, t Ld is the matrix T L The dth element in .

[0061] A4. Determine whether the search termination condition is met. If so, output the coordinates of the search point with the smallest absolute error as the center point of the macroblock in the i-th frame image, and calculate the central feature matrix of the macroblock. Otherwise, execute step A5. In this embodiment, the search termination condition is: the absolute error of the center search point is less than the absolute error of the left search point and the right search point, or r<1.

[0062] A5. Take the search point with the smallest absolute error as the central search point, reduce r according to the preset step size, obtain the left search point and the right search point, and execute step A2. In this embodiment, r is set to half of the original value to obtain a new r value, and the left and right r pixels of the central search point are recorded as the left search point and the right search point.

[0063] If r is not an integer after being half of its original value, but a decimal such as 7.5 or 2.5, it can be rounded to 7, 2 or 8, 3.

[0064] In the one-dimensional block matching algorithm, considering that the actual lung sliding movement in the column direction has little impact on the diagnosis of pneumothorax and adding the movement calculation in the column direction will increase the complexity of the calculation and reduce the operation rate, the calculation of the motion vector in the column direction of the motion vector is cancelled, and only the motion vector in the horizontal direction is used as the basis for judgment. This simplifies the algorithm and reduces the possibility of algorithm error to a certain extent.

[0065] S24. For each macroblock, record the distance between its center point in the i-1th frame image and the center point of the i-th frame image as the modulus of the i-1th horizontal motion vector of the macroblock; the modulus of the horizontal motion vector is a natural number, such as 0, 1, 3, etc.

[0066] S25, let i+1, repeat step S23 until the modulus of the 1st, 2nd, ..., K-1th horizontal motion vectors of each macroblock is obtained, where K is the total number of frames of the ultrasound examination video;

[0067] S26. For each macroblock, accumulate the moduli of the 1st, 2nd, ..., K-1st horizontal motion vectors of the macroblock to obtain the motion features of each macroblock, and establish an n×m video feature matrix, where the elements in the video feature matrix are the motion features of each macroblock.

[0068] Specifically, in step S3, a large number of training samples, i.e., video feature matrices, are obtained, and their corresponding diagnostic results are marked. The absolute error sum is used as the distance between two samples to train the k nearest neighbor classifier. After obtaining a good classification effect, the parameters are recorded to complete the training of the k nearest neighbor classifier. The video feature matrix obtained in step S2 is sent to the k nearest neighbor classifier to obtain the diagnostic result.

[0069] In addition, in other implementations, various machine learning models such as SVM classifiers and k-cluster classifiers may be used to obtain diagnostic results.

[0070] In step S4, the video feature matrix and the diagnosis results are output to assist doctors in making diagnoses, thereby reducing the experience and technical requirements for doctors, alleviating the diagnostic burden of clinicians, and improving detection efficiency and accuracy.

[0071] The calculation part required by the one-dimensional block matching algorithm is converted into a digital gate circuit through FPGA, and the calculation is performed in hardware, which is implemented by a hardware acceleration unit. The hardware acceleration unit includes a data interface unit, a data register and a parallel operation unit. The data interface unit is used to realize data transmission between the hardware acceleration unit and an external device. The data register is used to temporarily store data in the calculation process of the one-dimensional block matching algorithm. The parallel operation unit includes multiple parallel operation units for realizing the calculation of the one-dimensional block matching algorithm.

[0072] In the specific implementation process, the ultrasound probe of the B-type ultrasound instrument is used to examine the lung area, obtain the ultrasound examination video, and then grayscale it. The calculation part of the one-dimensional block matching algorithm is converted into a digital gate circuit through FPGA, and the calculation is performed directly in hardware, that is, the hardware acceleration unit is used for calculation. Compared with running in software on a computer, the hardware acceleration unit does not involve programming language conversion, instruction transmission and reception, etc. The required computing resources are unloaded from the CPU and designed into independent external hardware. Faster computing speed can be achieved through redundancy removal. Data input and output can be achieved through the interface, and a wider range of applications can be achieved. The computing speed is fast and can meet clinical use requirements. FPGA is already a relatively mature technology. For example, Xilinx has complete software to design simulations, and there are also some encapsulated libraries to provide some common functions. Relevant practitioners can implement them according to their understanding, and no further description will be given here. Through the one-dimensional block matching algorithm hardware acceleration unit, the video feature matrix is ​​obtained, and then sent to the trained k nearest neighbor classifier to obtain the diagnosis result, which assists doctors in diagnosing pneumothorax.

[0073] The video feature matrix can intuitively and quantitatively reflect the movement of lung sliding signs in ultrasound videos, which is convenient for doctors to observe and diagnose, and improves the efficiency and accuracy of pneumothorax diagnosis. The machine learning model used for classification can obtain diagnostic results based on the video feature matrix for doctors' reference. Based on the algorithm characteristics of FPGA-based technology and one-dimensional block matching algorithm, a hardware acceleration strategy combining software and hardware can be implemented to convert the algorithm operation part into hardware through customized instructions. With its high speed and functional specialization, it can meet the higher speed requirements of certain specific situations. Customized hardware designed for different needs can also broaden its scope of use to a certain extent. Finally, the diagnosis results and the corresponding video feature matrix are output simultaneously, which can not only assist doctors in diagnosis but also verify each other, thereby reducing the burden on doctors and improving patient experience. At the same time, the joint diagnosis of the system and doctors makes the diagnosis and treatment process more accurate and efficient.

[0074] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A B-ultrasound image analysis method for pneumothorax diagnosis, characterized in that: The following steps are involved: S1, obtaining an ultrasound examination video of the lung area and preprocessing it; S2, using a one-dimensional block matching algorithm to process the ultrasound inspection video to obtain a video feature matrix; S3, inputting the video feature matrix into a trained machine learning model, wherein the output of the machine learning model is a diagnosis result; S4, outputting the video feature matrix and the diagnosis result; Step S2 includes the following steps: S21, setting n×m macroblocks in the first frame of the ultrasound inspection video, the size of each macroblock is k×k pixels, recording the center point of each macroblock, setting a 1×k weight coefficient matrix A, and the value of the element in the weight coefficient matrix is ​​1 / 256; S22, obtaining the pixel value of each pixel point in each macroblock to obtain n×m macroblock pixel matrices, multiplying the weight coefficient matrix A by each macroblock pixel matrix to obtain the central feature matrix of each macroblock, wherein the shape of the central feature matrix is ​​1×k, and i=2; S23, obtaining the i-th frame image, obtaining the center point of each macroblock in the i-1-th frame image, setting a search point in the i-th frame image according to the center point of each macroblock in the i-1-th frame image, calculating the characteristic matrix of the macroblock at the search point, comparing the characteristic matrix at each search point of the macroblock with the central characteristic matrix of the macroblock in the i-1-th frame image, determining the center point of each macroblock in the i-th frame image, and calculating the central characteristic matrix of each macroblock; S24, for each macroblock, record the distance between the center point of the i-1th frame image and the center point of the i-th frame image as the modulus of the i-1th horizontal motion vector of the macroblock; S25, let i+1, repeat step S23 until the modulus of the 1st, 2nd, ..., K-1th horizontal motion vectors of each macroblock is obtained, where K is the total number of frames of the ultrasound examination video; S26, for each macroblock, accumulating the moduli of the 1st, 2nd, ..., K-1st horizontal motion vectors of the macroblock to obtain motion features of each macroblock, and establishing an n×m video feature matrix, where the elements in the video feature matrix are motion features of each macroblock; In step S23, determining the center point of a macroblock in the i-th frame image includes the following steps: A1. Project the center point of the macroblock in the i-1th frame image to the i-th frame image, record it as the center search point, and record the r pixels to the left and right of the center search point as the left search point and the right search point, where r is an integer greater than or equal to 1; A2. Take the center search point, left search point and right search point as the center point of the macroblock, obtain the pixel value of each pixel in the macroblock, and obtain the macroblock pixel matrix. Multiply the weight coefficient matrix A with the macroblock pixel matrix to obtain the feature matrix of the macroblock at the center search point, left search point and right search point, denoted as T L 、T E and T R ; A3. Let the central feature matrix of the macroblock in the i-1th frame image be T, and calculate T and T respectively. L 、T E and T R The absolute error sum is the sum of the absolute values ​​of the differences between the corresponding elements in the two matrices; A4, judging whether the search termination condition is met, if yes, outputting the coordinates of the search point with the minimum absolute error as the center point of the macroblock in the i-th frame image, calculating the central feature matrix of the macroblock, otherwise, executing step A5; A5. Take the search point with the smallest absolute error as the central search point, reduce r according to the preset step size, obtain the left search point and the right search point, and execute step A2.

2. The B-ultrasound image analysis method for pneumothorax diagnosis according to claim 1, characterized in that: In step S1, an ultrasound examination video of the lung area is acquired using a B-mode ultrasound machine.

3. The B-ultrasound image analysis method for pneumothorax diagnosis according to claim 1, characterized in that: In step S1, the ultrasound examination video is preprocessed specifically by: in each frame of the ultrasound examination video, grayscale processing is performed on each frame based on brightness.

4. The B-ultrasound image analysis method for pneumothorax diagnosis according to claim 1, characterized in that: In step S21, n×m macroblocks are evenly distributed in the first frame image, with a distribution shape of n rows and m columns, and n×m×k×k < N×M, where N×M is the size of each frame image in the ultrasound examination video.

5. The B-ultrasound image analysis method for pneumothorax diagnosis according to claim 1, characterized in that: In step A4, the search termination condition is: the sum of absolute errors of the central search point is less than the sum of absolute errors of the left search point and the right search point or r < 1.

6. The B-ultrasound image analysis method for pneumothorax diagnosis according to claim 1, characterized in that: In step A5, let r become half of its original value to obtain a new r value, and mark the pixels at r positions to the left and right of the central search point as the left search point and the right search point.

7. The B-ultrasound image analysis method for pneumothorax diagnosis according to claim 1, characterized in that: The computational part required by the one-dimensional block matching algorithm is converted into digital gate circuits through FPGA and calculated in a hardware manner, which is implemented by a hardware acceleration unit. The hardware acceleration unit includes a data interface unit, a data register, and a parallel operation unit. The data interface unit is used to implement data transmission between the hardware acceleration unit and external devices. The data register is used to temporarily store data during the calculation process of the one-dimensional block matching algorithm. The parallel operation unit includes multiple parallel operation units and is used to implement the calculation of the one-dimensional block matching algorithm.

8. The B-ultrasound image analysis method for pneumothorax diagnosis according to claim 1, characterized in that: The machine learning model is a k-nearest neighbor classifier.

Images