Ultrasonic image analysis method

Abstract

The present invention discloses an ultrasonic image analysis method, which relates to the technical field of medical image analysis. The present invention includes the following steps: starting an ultrasonic device, collecting ultrasonic images of the patient's examination site through an ultrasonic probe; pre-processing the ultrasonic image and performing edge detection; collecting historical ultrasonic images and converting the format; resampling all ultrasonic images; selecting appropriate window widths and window heights for ultrasonic images of different sites, completing the conversion of ultrasonic images, and constructing an ultrasonic diagnostic data set; inputting the ultrasonic diagnostic data set into a convolutional neural network for ultrasonic diagnostic model training; performing edge detection on the ultrasonic image to be detected and then inputting it into the trained ultrasonic diagnostic model. The present invention trains the ultrasonic diagnostic model by constructing an ultrasonic diagnostic data set, pre-processes the image collected by the ultrasonic probe and performs edge detection, and inputs the edge detection results into the trained ultrasonic diagnostic model, thereby improving the accuracy and efficiency of ultrasonic image diagnosis.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image analysis, and in particular relates to an ultrasonic image analysis method. Background Art

[0002] Ultrasound data is the ultrasound images and diagnostic data generated by doctors during patient diagnosis and treatment. Currently, ultrasound data is typically stored in ultrasound equipment, workstations, or hospital PACS (Picture Archiving and Communication Systems). However, ultrasound equipment, workstations, and PACS systems only store ultrasound data and are unable to identify or analyze the ultrasound images. This often requires manual identification and analysis by ultrasound physicians, which is a heavy workload and can easily lead to missed diagnoses. Summary of the Invention

[0003] The purpose of the present invention is to provide an ultrasonic image analysis method, which solves the problems of large deviation and low efficiency of existing manual diagnosis of ultrasonic images by constructing an ultrasonic diagnostic data set to train an ultrasonic diagnostic model, preprocessing the images collected by the ultrasonic probe and performing edge detection, and inputting the edge detection results into the trained ultrasonic diagnostic model.

[0004] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0005] The present invention is an ultrasonic image analysis method, comprising the following steps:

[0006] Step S1: Start the ultrasound equipment and collect ultrasound images of the patient's examination area through the ultrasound probe;

[0007] Step S2: preprocessing the ultrasound image;

[0008] Step S3: performing edge detection on the preprocessed ultrasound image;

[0009] Step S4: collecting historical ultrasound images and performing format conversion;

[0010] Step S5: resampling all ultrasound images;

[0011] Step S6: Select appropriate window width and window height for the ultrasound images of different parts, complete the conversion of the ultrasound images, and construct an ultrasound diagnostic data set;

[0012] Step S7: inputting the ultrasound diagnosis data set into the convolutional neural network to train the ultrasound diagnosis model;

[0013] Step S8: The ultrasound image to be detected is subjected to edge detection and then input into the trained ultrasound diagnosis model.

[0014] As a preferred technical solution, in step S2, the specific process of pre-processing is as follows:

[0015] Step S21, color space conversion: converting the original image from RGB color space to HSV color space or Lab color space;

[0016] Step S22, image segmentation: using the color information and brightness information, using an image segmentation algorithm to separate the shadow area from the non-shadow area in the processed image;

[0017] Step S23, shadow correction: retrieve the brightness value of the area or adjust the color value to offset the effect;

[0018] Step S24, fusion processing: re-fusing the processed shadow area with the original non-shadow area to obtain an image with the shadow removed;

[0019] Step S25, image contrast enhancement: the image is separated into three BGR channel images, and the three channels are subjected to histogram equalization processing using local histogram equalization processing, and the channel images are merged into the target image.

[0020] As a preferred technical solution, in step S21, the original ultrasound image is set to f(x,y)=c(x,y)×l(x,y)+(1-f(x,y)); wherein c(x,y) is the color value at the pixel (x,y), and l(x,y) is the transmittance of the ultrasound film at the pixel (x,y). The specific process of color space conversion is as follows:

[0021] Step S211: converting the original ultrasound image into a grayscale image;

[0022] Step S212: calculating the local contrast of the grayscale image;

[0023] Step S213: Calculate the gain factor according to the local contrast. The specific calculation formula is as follows:

[0024]

[0025] Where M is the local contrast, Q is the predefined threshold, Gain factor;

[0026] Step S214: performing a gain operation on the original image using the gain factor, multiplying the grayscale value of each pixel by the gain factor at the corresponding position;

[0027] Step S215: Obtain the dark primary color pixel distribution function I of the dark primary color ultrasound image by adjusting the parameters. dark (x,y):

[0028]

[0029] Where c dark (x,y) is the dark original color value at pixel (x,y).

[0030] As a preferred technical solution, in step S3, the process of performing edge detection on the preprocessed ultrasound image is as follows:

[0031] Step S31: Based on the correlation between the noise pixel points and the adjacent elements, a partition mean filtering algorithm is used to smooth and reduce the noise points of the ultrasound image;

[0032] Step S32: performing non-maximum suppression processing on the image after noise smoothing processing;

[0033] Step S33: Pixel points are connected one by one to form a final edge detection map of the noise target image.

[0034] As a preferred technical solution, in step S31, the function formula for the smoothing noise reduction process is as follows:

[0035]

[0036] Where σ is the spatial scale coefficient of the Gaussian filter;

[0037] After smoothing the target ultrasound image, the partial derivative functions in the x and y directions are calculated respectively. The specific formulas are as follows:

[0038]

[0039] The original ultrasound image f(x,y) is convolved to obtain the gradient direction θ(i,j) and gradient value of each pixel in the target image. The calculation formulas are:

[0040]

[0041] As a preferred technical solution, in step S32, the specific process of non-maximum suppression processing is as follows:

[0042] Step S321: Calculate the gradient value S(i, j) of each pixel point in the gradient direction θ(i, j) on the image to determine whether it is a point on the edge;

[0043] Step S322: Taking each pixel as the center, compare the gradient values of the pixels adjacent to it in four adjacent directions. If the gradient value S(i, j) of the pixel is the largest, it is an edge point; otherwise, set the gradient value of the pixel to zero.

[0044] Step S323: Set the high threshold to T hand the low threshold is T l , if the gradient value of the pixel point S(i,j)≥T h , then the pixel is an edge point of the image. If the gradient value of the pixel S(i,j)<T h , then the pixel is not an image edge point;

[0045] Step S324: If the gradient value T of the pixel point h <S(i,j)<T l , then compare the gradient values of the pixel point with those of the adjacent pixels in four directions;

[0046] Step S325: Connect the pixels that meet the conditions to form a final edge detection map of the noisy target image.

[0047] As a preferred technical solution, in step S6, the conversion process of the ultrasound image is as follows:

[0048] Step S61: Perform average pooling on the input ultrasound image along the x and y directions, and encode each direction; the specific formula is as follows:

[0049]

[0050] Where x c (h,i) represents the eigenvalue of the i-th position on the c channel and h window height, x c (j,w) represents the eigenvalue of the jth position on the c channel and w window height, Represents the information of the feature map along the x-axis direction, Represents the information of the feature map along the y-axis;

[0051] Step S62: perform dimensionality conversion again to obtain a feature map with encoding information;

[0052] Step S63: Concat the information in the two directions to splice the feature map;

[0053] Step S64: Normalization is performed to generate an intermediate feature vector, and a feature map with the same number of channels is obtained. The specific formula is as follows:

[0054] F=δ(F1([z h ,z w ]));

[0055] In the formula, for z h ,z w Perform Concat operation, F1 represents convolution operation, and δ represents nonlinear activation function.

[0056] As a preferred technical solution, in step S7, the convolutional layer of the convolutional neural network performs feature extraction on the input feature map. The specific formula is as follows:

[0057]

[0058] Where i is the image sequence number, j is the image feature number, ζ is the output function, and m j is the image with image feature number j corresponding to the feature map, is the convolution result of the i-th image in the l-1th convolution layer, * is the convolution operator symbol, is the convolution kernel with image feature map number j in the i-th image when the current convolution layer number is l, is the offset;

[0059] The pooling layer of the convolutional neural network performs an average pooling operation, and the specific formula is as follows:

[0060]

[0061] Where, e i For the region The activation value of each neuron in Output result for the pooling layer;

[0062] The fully connected layer of the convolutional neural network converts the abstract features learned by the convolutional layer into classifiable outputs and activates each layer using the Softmax function. The specific formula is as follows:

[0063]

[0064] The present invention has the following beneficial effects:

[0065] The present invention trains an ultrasound diagnostic model by constructing an ultrasound diagnostic data set, performs edge detection on images collected by an ultrasound probe after preprocessing, and inputs the edge detection results into the trained ultrasound diagnostic model to improve the accuracy and efficiency of ultrasound image diagnosis.

[0066] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0068] Figure 1This is a flow chart of an ultrasonic image analysis method of the present invention;

[0069] Figure 2 This is the ultrasound image preprocessing flow chart;

[0070] Figure 3 Flowchart for edge detection of preprocessed ultrasound images;

[0071] Figure 4 Schematic diagram of the convolutional neural network structure. DETAILED DESCRIPTION

[0072] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0073] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0074] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-4 It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0075] Ultrasound images are acquired using an ultrasound probe that transmits ultrasound waves, which propagate through tissue. When ultrasound waves encounter tissues of varying densities, they produce echoes. The probe receives these echoes and converts them into electrical signals, which are then processed to form an ultrasound image.

[0076] See also Figure 1 As shown, the present invention is an ultrasonic image analysis method, comprising the following steps:

[0077] Step S1: Start the ultrasound equipment and collect ultrasound images of the patient's examination area through the ultrasound probe;

[0078] Step S2: preprocessing the ultrasound image;

[0079] Step S3: performing edge detection on the preprocessed ultrasound image;

[0080] Step S4: collecting historical ultrasound images and performing format conversion;

[0081] Step S5: resampling all ultrasound images;

[0082] Step S6: Select appropriate window width and window height for the ultrasound images of different parts, complete the conversion of the ultrasound images, and construct an ultrasound diagnostic data set;

[0083] Step S7: inputting the ultrasound diagnosis data set into the convolutional neural network to train the ultrasound diagnosis model;

[0084] Step S8: The ultrasound image to be detected is subjected to edge detection and then input into the trained ultrasound diagnosis model.

[0085] In step S1, during the process of ultrasonic acquisition, the specific operation process is as follows:

[0086] 1. Preparation stage: determine the examination site and purpose, select a suitable ultrasound probe, and apply coupling agent to the patient's examination site;

[0087] 2. Positioning and scanning: Place the ultrasound probe on the inspection area and adjust the probe's angle and position to allow the ultrasound to penetrate and reflect back. Start the ultrasound device and begin the scanning process. During the scanning process, the probe will emit ultrasound and receive the reflected signal.

[0088] 3. Image Acquisition: Ultrasonic waves propagate and reflect within the human body, forming echoes. Ultrasonic equipment receives these echoes and converts them into electrical signals. After amplification and filtering, these signals are converted into digital signals. These digital signals are further processed by a computer to generate two-dimensional or three-dimensional ultrasound images. This embodiment primarily analyzes and processes two-dimensional ultrasound images.

[0089] See also Figure 2 As shown, in step S2, the specific process of preprocessing is as follows:

[0090] Step S21, color space conversion: converting the original image from RGB color space to HSV color space or Lab color space;

[0091] Step S22, image segmentation: using color information and brightness information, using an image segmentation algorithm to separate shadow areas from non-shadow areas in the processed image;

[0092] Step S23, shadow correction: retrieve the brightness value of the area or adjust the color value to offset the effect;

[0093] Step S24, fusion processing: re-fusing the processed shadow area with the original non-shadow area to obtain an image with the shadow removed;

[0094] Step S25, image contrast enhancement: the image is separated into three BGR channel images, and the three channels are subjected to histogram equalization processing using local histogram equalization processing, and the channel images are merged into the target image.

[0095] In step S21, the original ultrasound image is set to f(x,y)=c(x,y)×l(x,y)+(1-f(x,y)); where c(x,y) is the color value at pixel (x,y), and l(x,y) is the transmittance of the ultrasound image at pixel (x,y). The specific process of color space conversion is as follows:

[0096] Step S211: converting the original ultrasound image into a grayscale image;

[0097] Step S212: calculating the local contrast of the grayscale image;

[0098] Step S213: Calculate the gain factor according to the local contrast. The specific calculation formula is as follows:

[0099]

[0100] Where M is the local contrast, Q is the predefined threshold, Gain factor;

[0101] Step S214: performing a gain operation on the original image using the gain factor, multiplying the grayscale value of each pixel by the gain factor at the corresponding position;

[0102] Step S215: Obtain the dark primary color pixel distribution function I of the dark primary color ultrasound image by adjusting the parameters. dark (x,y):

[0103]

[0104] Where c dark (x,y) is the dark primary color value at pixel (x,y)

[0105] See also Figure 3 As shown in Figure 1, in step S3, edge detection is one of the most important operations in image processing, and is usually used to identify areas in an image where grayscale or color changes dramatically. These changes are usually caused by the structure and texture of the object, external lighting, and the reflection of light on the surface of the object. Specifically, the accuracy, stability, and processing speed of image processing are improved by eliminating noise, enhancing contrast, and standardizing image size. The purpose of preprocessing is to provide higher-quality input data for subsequent algorithms, thereby improving the effect and efficiency of overall image processing. The specific process of edge detection on preprocessed ultrasound images is as follows:

[0106] Step S31: Based on the correlation between the noise pixel points and the adjacent elements, a partition mean filtering algorithm is used to smooth and reduce the noise points of the ultrasound image;

[0107] Step S32: performing non-maximum suppression processing on the image after noise smoothing processing;

[0108] Step S33: Pixel points are connected one by one to form a final edge detection map of the noise target image.

[0109] In step S31, the function formula for smoothing noise reduction processing is as follows:

[0110]

[0111] Where σ is the spatial scale coefficient of the Gaussian filter;

[0112] After smoothing the target ultrasound image, the partial derivative functions in the x and y directions are calculated respectively. The specific formulas are as follows:

[0113]

[0114] The original ultrasound image f(x,y) is convolved to obtain the gradient direction θ(i,j) and gradient value of each pixel in the target image. The calculation formulas are:

[0115]

[0116] In step S32, the specific process of non-maximum suppression processing is as follows:

[0117] Step S321: Calculate the gradient value S(i, j) of each pixel point in the gradient direction θ(i, j) on the image to determine whether it is a point on the edge;

[0118] Step S322: Taking each pixel as the center, compare the gradient values of the pixels adjacent to it in four adjacent directions. If the gradient value S(i, j) of the pixel is the largest, it is an edge point; otherwise, set the gradient value of the pixel to zero.

[0119] Step S323: Set the high threshold to T h and the low threshold is T l , if the gradient value of the pixel point S(i,j)≥T h , then the pixel is an edge point of the image. If the gradient value of the pixel S(i,j)<T h , then the pixel is not an image edge point;

[0120] Step S324: If the gradient value T of the pixel point h <S(i,j)<T l , then compare the gradient values of the pixel point with those of the adjacent pixels in four directions;

[0121] Step S325: Connect the pixels that meet the conditions to form a final edge detection map of the noisy target image.

[0122] In step S6, the ultrasound image conversion process is as follows:

[0123] Step S61: Perform average pooling on the input ultrasound image along the x and y directions, and encode each direction; the specific formula is as follows:

[0124]

[0125] Where x c (h,i) represents the eigenvalue of the i-th position on the c channel and h window height, x c (j,w) represents the eigenvalue of the jth position on the c channel and w window height, Represents the information of the feature map along the x-axis direction, Represents the information of the feature map along the y-axis;

[0126] Step S62: perform dimensionality conversion again to obtain a feature map with encoding information;

[0127] Step S63: Concat the information in the two directions to splice the feature map;

[0128] Step S64: Normalization is performed to generate an intermediate feature vector, and a feature map with the same number of channels is obtained. The specific formula is as follows:

[0129] F=δ(F1([z h ,z w ]));

[0130] In the formula, for z h ,z w Perform Concat operation, F1 represents convolution operation, and δ represents nonlinear activation function.

[0131] See also Figure 4 As shown, in step S7, the convolutional layer of the convolutional neural network extracts features from the input feature map. The specific formula is as follows:

[0132]

[0133] Where i is the image sequence number, j is the image feature number, ζ is the output function, and m j is the image with image feature number j corresponding to the feature map, is the convolution result of the i-th image in the l-1th convolution layer, * is the convolution operator symbol, is the convolution kernel with image feature map number j in the i-th image when the current convolution layer number is l, is the offset;

[0134] The pooling layer of the convolutional neural network performs an average pooling operation. The specific formula is as follows:

[0135]

[0136] Where, e i For the region The activation value of each neuron in Output result for the pooling layer;

[0137] The fully connected layer of the convolutional neural network converts the abstract features learned by the convolutional layer into classifiable outputs and activates each layer using the Softmax function. The specific formula is as follows:

[0138]

[0139] It is worth noting that in the above system embodiment, the various units included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0140] In addition, those skilled in the art will appreciate that all or part of the steps in the above-mentioned embodiments can be accomplished by instructing related hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0141] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An ultrasonic image analysis method, characterized in that: The steps include: Step S1: Start the ultrasound equipment and collect ultrasound images of the patient's examination area through the ultrasound probe; Step S2: preprocessing the ultrasound image; Step S3: performing edge detection on the preprocessed ultrasound image; Step S4: collecting historical ultrasound images and performing format conversion; Step S5: resampling all ultrasound images; Step S6: Select appropriate window width and window height for the ultrasound images of different parts, complete the conversion of the ultrasound images, and construct an ultrasound diagnostic data set; Step S7: inputting the ultrasound diagnosis data set into the convolutional neural network to train the ultrasound diagnosis model; Step S8: The ultrasound image to be detected is subjected to edge detection and then input into the trained ultrasound diagnosis model; In step S3, the process of edge detection on the pre-processed ultrasound image is as follows: Step S31: Based on the correlation between the noise pixel points and the adjacent elements, a partition mean filtering algorithm is used to smooth and reduce the noise points of the ultrasound image; Step S32: performing non-maximum suppression processing on the image after noise smoothing processing; Step S33: connecting the pixels one by one to form the final edge detection map of the noise target image; In step S32, the specific process of non-maximum suppression processing is as follows: Step S321: Gradient direction of each pixel on the image Calculate the gradient value on , determine whether it is a point on the edge; Step S322: Taking each pixel as the center, compare the gradient values of the pixels adjacent to it in four adjacent directions. If the gradient value of the pixel is If the value is the largest, it is an edge point; otherwise, the gradient value of the pixel is set to zero; Step S323: Set the high threshold to and the low threshold is , if the gradient value of the pixel , then the pixel is the edge point of the image. If the gradient value of the pixel is , then the pixel is not an image edge point; Step S324: If the gradient value of the pixel , then compare the gradient values of the pixel point with those of the adjacent pixels in four directions; Step S325: connecting the pixels that meet the conditions to form a final edge detection map of the noise target image; In step S6, the conversion process of the ultrasound image is as follows: Step S61: The input ultrasound image is respectively and The direction is average pooled and each direction is encoded; the specific formula is as follows: ； Where, express Channel and The window is high, The eigenvalues at the positions, express Channel and The window is high, The eigenvalues at the positions, Indicates that the feature map is along Information on the axis direction, Indicates that the feature map is along Information on the axis direction; Step S62: perform dimensionality conversion again to obtain a feature map with encoding information; Step S63: Concat the information in the two directions to splice the feature map; Step S64: Normalization is performed to generate an intermediate feature vector, and a feature map with the same number of channels is obtained. The specific formula is as follows: ； Where, Perform Concat operation, represents the convolution operation, represents a nonlinear activation function; In step S7, the convolutional layer of the convolutional neural network performs feature extraction on the input feature map. The specific formula is as follows: ； Where, is the image sequence number, is the image feature number, is the output function, The image feature number corresponding to the feature map is images, For the The image in The convolution results in the number of convolution layers, is the convolution operator symbol, For the In the current convolution layer number of the image When the image feature map number is The convolution kernel, is the offset; The pooling layer of the convolutional neural network performs an average pooling operation, and the specific formula is as follows: ； Where, For the region The activation value of each neuron in Output result for the pooling layer; The fully connected layer of the convolutional neural network converts the abstract features learned by the convolutional layer into classifiable outputs and activates each layer using the Softmax function. The specific formula is as follows: 。 2. The ultrasonic image analysis method according to claim 1, characterized in that: In step S2, the specific process of pre-processing is as follows: Step S21, color space conversion: converting the original image from RGB color space to HSV color space or Lab color space; Step S22, image segmentation: using color information and brightness information, using an image segmentation algorithm to separate shadow areas from non-shadow areas in the processed image; Step S23, shadow correction: retrieve the brightness value of the area or adjust the color value to offset the effect; Step S24, fusion processing: re-fusing the processed shadow area with the original non-shadow area to obtain an image with the shadow removed; Step S25, image contrast enhancement: the image is separated into three BGR channel images, and the three channels are subjected to histogram equalization processing using local histogram equalization processing, and the channel images are merged into the target image.

3. The ultrasonic image analysis method according to claim 2, characterized in that: In step S21, the original ultrasound image is set to ;in, Pixels The color value at Pixels The specific color space conversion process is as follows: Step S211: converting the original ultrasound image into a grayscale image; Step S212: calculating the local contrast of the grayscale image; Step S213: Calculate the gain factor according to the local contrast. The specific calculation formula is as follows: ； Where M is the local contrast, Q is the predefined threshold, Gain factor; Step S214: performing a gain operation on the original image using the gain factor, multiplying the grayscale value of each pixel by the gain factor at the corresponding position; Step S215: Obtain the dark primary color pixel distribution function of the dark primary color ultrasound image by adjusting the parameters : ； Where, Pixels The dark primary value at .

4. The ultrasonic image analysis method according to claim 1, wherein: In step S31, the function formula for the smooth noise reduction process is as follows: ； Where, is the spatial scale coefficient of the Gaussian filter; When smoothing the target ultrasound image, calculate and The partial derivative function of the direction is as follows: ； ； The original ultrasound image Perform convolution operation to obtain the gradient direction of each pixel of the target image And the gradient value, the calculation formula is: ； 。

Images