An enhanced denoising method for clinical ultrasound images

Abstract

The present invention relates to the field of image processing, and particularly relates to an enhanced denoising method for clinical ultrasound images, including: collecting knee joint grayscale images; presetting a search window and a sliding window, dividing the search window into upper and lower half-region grayscale images, and stratifying the upper and lower half-region grayscale images according to the sliding window; obtaining a feature extraction factor for each layer according to the pixel points of each layer; obtaining the filter weights of the upper and lower half-region grayscale images according to the feature extraction factors of each layer; obtaining the weight of each pixel point within the search window according to the differences between the pixel points in the sliding window within the search window; obtaining the weighted grayscale value of each pixel point according to the filter weights of the upper and lower half-region grayscale images, the weight of each pixel point within the search window, and the grayscale value; and obtaining the denoised grayscale image according to the weighted grayscale value of each pixel point. The present invention determines the pixel point weight values by image processing, weakening or even eliminating the influence of noise interference in ultrasound images.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to an enhanced denoising method for clinical ultrasound images. Background Art

[0002] In recent years, ultrasound technology has been widely used in preoperative localization of diseases. Compared with CT images, ultrasound imaging technology has the advantages of being non-invasive, non-radiative, simple, rapid, convenient, intuitive, and inexpensive. It has a fast imaging speed and has outstanding advantages in the observation of moving organs. Ultrasound images can clearly show the size, shape, internal echo, and blood flow distribution of the lesion area.

[0003] However, the ultrasound imaging process is rather special. Due to the random interference of scatterers, when ultrasonic waves pass through a slightly inhomogeneous medium, multiplicative speckle noise is generated in the image, resulting in a reduction in the resolution of the ultrasound image, a large amount of detailed information in the image being lost, and the image edges becoming blurred, which seriously interferes with the subsequent clinical diagnosis process of differentiating between diseased tissues and normal tissues of human soft tissues. Summary of the Invention

[0004] In order to solve the above problems, the present invention provides an enhanced denoising method for clinical ultrasound images, and the method includes:

[0005] Processing the collected knee ultrasound image to obtain a knee grayscale image;

[0006] Presetting a search window and a sliding window, dividing the search window into an upper half area and a lower half area, denoting the upper half area as the upper half area grayscale image, and denoting the lower half area as the lower half area grayscale image; stratifying the upper half area grayscale image and the lower half area grayscale image according to the width of the sliding window to obtain several layers; obtaining a feature extraction factor for each layer according to the pixel points of each layer; obtaining a filter weight for the lower half area grayscale image and a filter weight for the upper half area grayscale image according to the feature extraction factor of each layer;

[0007] Obtaining the weight of each pixel point in the search window according to the difference between the pixel points in the sliding window corresponding to the central pixel point in the search window and the pixel points in the sliding window corresponding to the remaining pixel points in the search window; obtaining the weighted grayscale value of each pixel point according to the filter weight of the lower half area grayscale image, the filter weight of the upper half area grayscale image, the weight of each pixel point in the search window, and the grayscale value of each pixel point in the search window;

[0008] Obtaining a denoised grayscale image according to the weighted grayscale value of each pixel point;

[0009] Automatically identifying the osteophyte region according to the denoised grayscale image.

[0010] Further, the specific steps for stratifying the upper half grayscale image and the lower half grayscale image according to the width of the sliding window to obtain several layers are as follows:

[0011] After obtaining the upper half grayscale image and the lower half grayscale image, divide the upper half grayscale image and the lower half grayscale image respectively according to the width of the sliding window. In the vertical direction, equally divide them with the width L of the sliding window to obtain several layers.

[0012] Further, the specific steps for obtaining the feature extraction factor of each layer are as follows:

[0013] Denote any layer in the lower half grayscale image as the target layer. The formula for the feature extraction factor of the target layer is:

[0014]

[0015] In the formula, represents the feature extraction factor of the target layer, represents the grayscale value of the i-th pixel point in the middle row of the target layer, represents the grayscale value of the z-th pixel point in the neighborhood of the sliding window centered on the i-th pixel point in the middle row of the target layer. L represents the width of the sliding window, represents the total number of pixel points in the sliding window, represents the number of pixel points in the target layer;

[0016] Similarly, obtain the feature extraction factor of each layer.

[0017] Further, the specific steps for obtaining the filter weights of the lower half grayscale image and the upper half grayscale image are as follows:

[0018] The formula for the filter weight of the lower half grayscale image is:

[0019]

[0020] In the formula, represents the difference between the feature extraction factors of the m-th layer and the adjacent lower layer in the lower half grayscale image, represents the average value of the differences between the feature extraction factors of all adjacent layers in the lower half grayscale image, represents the number of layers in the lower half grayscale image, represents the hyperbolic tangent function, represents the filter weight in the lower half grayscale image;

[0021] According to the filter weight in the lower half grayscale image, the filter weight in the upper half grayscale image is .

[0022] Further, the specific steps for obtaining the weight of each pixel point within the search window are as follows:

[0023] The formula for the weight of the v-th pixel point within the search window is:

[0024]

[0025] In the formula, represents the mean square error corresponding to the v-th pixel point in the lower half grayscale image, represents the number of pixel points within the search window, represents the mean square error corresponding to the j-th pixel point in the lower half grayscale image, represents the weight of the v-th pixel point in the lower half grayscale image, represents the exponential function with the natural constant as the base.

[0026] Further, the specific steps for obtaining the mean square error corresponding to the v-th pixel point in the lower half grayscale image are as follows:

[0027] Obtain the central pixel point of the search window in the lower half grayscale image, denoted as the target central pixel point, and obtain the sliding window of the target central pixel point, denoted as the target sliding window; obtain the sliding window of the v-th pixel point within the search window, denoted as the marked sliding window; take the result of summing the squares of the differences between the pixel points in the target sliding window and all pixel points in the marked sliding window as the mean square error corresponding to the v-th pixel point in the lower half grayscale image.

[0028] Further, the specific steps for obtaining the gray value of each pixel point after weighting are as follows:

[0029] The formula for the gray value of each pixel point after weighting is:

[0030]

[0031] In the formula, represents the gray value of the v-th pixel point in the lower half grayscale image, represents the gray value of the r-th pixel point in the upper half grayscale image, represents the weight of the v-th pixel point in the lower half grayscale image, represents the weight of the r-th pixel point in the upper half grayscale image, represents the number of pixel points within the search window, represents the gray value of the central pixel point of the search window after weighting, represents the filter weight in the lower half grayscale image, represents the filter weight in the upper half grayscale image.

[0032] Further, the steps of dividing the search window into an upper half grayscale image and a lower half grayscale image are as follows:

[0033] Divide the search window into an upper half and a lower half. Denote the upper half as the upper half grayscale image and the lower half as the lower half grayscale image.

[0034] Further, the steps of obtaining the denoised grayscale image according to the weighted grayscale value of each pixel point are as follows:

[0035] Take the weighted grayscale value of each pixel point as the true grayscale value of each pixel point, and denote the image composed of the true grayscale values of each pixel point in the knee joint grayscale image as the denoised grayscale image.

[0036] The beneficial effects of the technical solution of the present invention are as follows: Making an accurate identification of the situation in the osteophyte area is an important method in medical image processing. However, due to image quality problems, there is noise information in the image, and the visualization degree of the bone spur area is not very high. Therefore, the image needs to be denoised first. In order to better retain the information of the feature area, the present invention sets a search window, divides the search window into upper and lower halves along the center point, stratifies the search window along the y-axis direction, constructs a model to extract the feature information of each layer, fits the feature values output by each layer, and uses the fitting residual as the total weight of the upper and lower halves within the search window. Then, by calculating the similarity between the sliding window in the lower half area and the sliding window of the central pixel point, the weights of each half area are allocated instead of local mean filtering weighted smoothing. The present invention makes the filtering weights concentrate on the bone spur area, so that the effective information of the bone spur area can be retained as much as possible during the filtering operation, improving the image quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0038] Figure 1 It is a flowchart of the steps of an enhanced denoising method for clinical ultrasound images of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following specifically describes, in conjunction with the accompanying drawings and preferred embodiments, a method for enhancing and denoising clinical ultrasound images according to the present invention, including its specific implementation manner, structure, features, and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0041] The following specifically describes, in conjunction with the accompanying drawings, the specific solution of a method for enhancing and denoising clinical ultrasound images provided by the present invention.

[0042] Please refer to Figure 1 , which shows a flowchart of the steps of a method for enhancing and denoising clinical ultrasound images provided by an embodiment of the present invention. The method includes the following steps:

[0043] Step S001: Obtain a knee joint ultrasound image and preprocess the knee joint ultrasound image to obtain a knee joint grayscale image.

[0044] It should be noted that knee joint bone hyperplasia is not an ordinary joint inflammation, but a long-term degenerative lesion of cartilage. The patella is composed of the lower end of the femur, the upper end of the tibia, and the patella, surrounded by a joint capsule, with cruciate ligaments and menisci inside. It is the most complex and the joint surface with the largest load-bearing joint in the human body. Therefore, the identification of knee joint bone hyperplasia is of utmost importance and has certain therapeutic significance.

[0045] Specifically, collect a knee joint ultrasound image and perform grayscale preprocessing on the knee joint ultrasound image to obtain a preprocessed knee joint grayscale image.

[0046] Thus, a knee joint grayscale image is obtained.

[0047] Step S002: Dynamically adjust the size of the search window according to the edge characteristics of the knee joint, use the width of the sliding window to layer the upper half grayscale image and the lower half grayscale image, obtain the feature extraction factor of each layer according to the pixel points of each layer, obtain the filter weight according to the feature extraction factor of each layer, and use the filter weight to obtain the weighted grayscale value of each pixel point.

[0048] It should be noted that knee joint bone spurs are small protruding bones formed under the knee joint cartilage, usually appearing on the joint surface between the femoral condyle and the tibial plateau. The joint edge under normal cartilage is relatively flat without protrusions, and the gap between the knee joints is relatively sufficient. However, for the knee joint with bone spurs, the gap between the joints becomes smaller, and abnormal protrusions appear at the bone edge. It is necessary to analyze the detailed information between the knee joints in the blurred image.

[0049] Since the gray value of the bone is relatively high, it shows a relatively bright feature in the image. The same is true for bone spurs. Since bone spurs grow on the surface of the cartilage, the gray value in the joint space is slightly weaker, and the bone density is also slightly weaker.

[0050] Since two windows are required when using non-local mean filtering, a search window width M and a sliding window width L are set here. In this embodiment, the search window width M = 49 and the sliding window width L = 7 are used as examples for description. This embodiment does not make specific limitations, and M and L can be determined according to specific implementation situations. Both the search window and the sliding window slide from left to right and from top to bottom with a step size of 1. Also, because bone spurs only exist on the upper surface or the lower surface of the joint space, the search window is stratified vertically along the y-axis direction in the rectangular coordinate system to obtain multiple horizontal layers. As the different levels increase in the y-axis direction, the distribution of pixel points in the sliding window at different levels is analyzed and statistically counted.

[0051] Specifically, the area corresponding to the search window each time it slides on the knee joint gray image is recorded as the gray image of the search window. First, due to the characteristic of the bone spur being in the middle position, the gray image of the search window is equally divided into two upper and lower area gray images, denoted as the upper half area gray image and the lower half area gray image. Now, the lower half area gray image is analyzed, and the analysis process of the upper half area gray image is the same as that of the lower half area gray image. After obtaining the lower half area gray image, it is divided again with the sliding window width. It is evenly divided vertically with a width of L to obtain multiple horizontal layers. Similarly, the upper half area gray image is also divided into layers.

[0052] Further, it should be noted that since the specific area of the bone spur is unknown, only that the bone spur area is the distribution of pixel points with a certain shape, and the gray levels of the pixel points in the bone spur area should be similar to those of the pixel points in the bone area. Due to the presence of noise, the gray levels of some pixel points are not so obvious; however, the gray levels of the pixel points outside the bone spur area are similar and the gray values are relatively low. The average of all pixel points within the sliding window centered on the pixel points in the bone spur area will be higher, while the average gray level of all pixel points within the sliding window centered on the pixel points outside the bone spur area is lower. The gray value of the central pixel point of the sliding window in the bone spur area is higher, and the distribution density of the high-brightness points within the sliding window is also higher.

[0053] Specifically, the layers divided in the lower half gray-scale image are analyzed. Any layer in the lower half gray-scale image is denoted as the target layer. The feature extraction factor of the target layer in the lower half gray-scale image is obtained based on all the pixel points in the middle row of the target layer. Then, the feature extraction factor of the target layer in the lower half gray-scale image can be expressed by the following formula:

[0054]

[0055] In the formula, represents the feature extraction factor of the target layer, represents the gray value of the i-th pixel point in the middle row of the target layer, represents the gray value of the z-th pixel point within the window neighborhood of the sliding window centered on the i-th pixel point in the middle row of the target layer, that is, the gray value of the z-th pixel point within the window except the central pixel point. L represents the width of the sliding window, represents the total number of pixel points within the sliding window, represents the length of the target layer, that is, the number of pixel points in the middle row of the target layer.

[0056] Similarly, the feature extraction factor of each layer is obtained.

[0057] Among them, represents the distribution density of the high-brightness pixel points within the sliding window centered on the i-th pixel point in the middle row of the target layer. Since there is noise in the sliding window and it is impossible to accurately capture the pixel points with distinct brightness, the distribution density cannot be represented by the number of high-brightness pixel points. The average gray level within the window can be used to indirectly represent the distribution density of the high-brightness pixel points. The higher the average gray level of the pixel points within the sliding window, the more high-brightness pixel points there are in the window, that is, the distribution density of the high-brightness pixel points is large. represents the Euclidean norm of the gray value of the central pixel point and the distribution density of the high-brightness pixel points within the sliding window centered on this pixel point; there are M sliding windows in each layer.

[0058] Obtain the feature extraction factors of each layer in the lower-half gray-scale image according to the feature extraction factors of the target layer. Sequentially obtain the feature extraction factors of each layer in the lower-half gray-scale image from top to bottom to obtain a sequence A, which are sequentially denoted as where n represents the total number of layers divided in the lower-half gray-scale image, represents the feature extraction factor of the i-th layer in the lower-half gray-scale image. Similarly, the feature extraction factors of all layers in the upper-half gray-scale image can be obtained.

[0059] It should be noted that then analyze the fitting degree of the fitting line to obtain the difference between the actual fitting degree and the fitting degree in the ideal case. Use this difference to allocate the filter weights to retain as much feature information of the bone spur area as possible during the filtering process. In the actual scenario, the bone spur presents a "pyramid shape" from top to bottom, that is, the feature extraction of each layer from top to bottom increases layer by layer. Therefore, analyze through the difference between the feature extraction factors of adjacent layers. If there is a bone spur, it presents a "pyramid shape", that is, the difference between the feature extraction factors of adjacent layers is not significant.

[0060] Specifically, obtain the differences between the feature extraction factors of all adjacent layers in the lower-half gray-scale image, which is represented by sequence B and is respectively denoted as where, represents the difference between the feature extraction factors of the m-th layer and the (m + 1)-th layer in the lower-half gray-scale image, and n - 1 represents the lower-half gray-scale image. According to the characteristics of the bone spur, if there is a bone spur in the image, the variance in sequence B should be very small, that is, each value in sequence B is regarded as equal. Therefore, when sequence B conforms to the bone spur characteristics, all values in sequence A can be linearly fitted to obtain a straight line. Therefore, when there is a bone spur, the curve fitted by the feature extraction factors of each layer in sequence A is closer to a straight line, and the variance of all data in the corresponding sequence B is smaller, that is, when the variance approaches 0, the possibility of the existence of a bone spur in the search window is greater.

[0061] Then the formula for the filter weights in the lower-half gray-scale image can be expressed as:

[0062]

[0063] In the formula, represents the difference between the feature extraction factor of the m-th layer and the adjacent lower layer in the lower-half gray-scale image, represents the mean value of the differences between the feature extraction factors of all adjacent layers in the lower-half gray-scale image, represents the number of layers in the lower-half gray-scale image, represents the hyperbolic tangent function, represents the filter weights in the lower-half gray-scale image.

[0064] Among them, use to represent the degree of change of each layer of output values in the lower half region . Ideally, it is considered that in the region from the "tip of the spur" to the bone in the spur region, the change of each layer should be a linearly increasing fitting effect. The value will be close to 0. Use the th function to normalize it to between 0 and 1; The smaller the value, the more it indicates that each layer of output values shows a linearly increasing situation, and it is considered that the central pixel point of the sliding window is more likely to be a pixel point in the spur region. Therefore, at this time is also the effect of true fitting, that is, it shows that the filter weights from to 0 are a fitting effect that does not belong to the spur region. The value can be used as the weight of the pixel points in the non-spur region.

[0065] According to the weights of the filters in the lower half grayscale image, the weights of the filters in the upper half grayscale image are .

[0066] The above operations are described for the lower spurs with the horizontal line of the center point of the search window as the dividing line. As for the calculation of the upper spurs, it can be obtained in the same way. Compare the fitting results of the upper and lower half grayscale images, and take the side with the smallest value as the suspicious spur region.

[0067] It should be noted that the calculated above is the difference under ideal conditions, that is, the distribution of the true pixel points in the spur region. The overall weight of the region below the horizontal dividing line of the search window is ; while the region above the dividing line is considered to be a region without spurs, and the weight of this region is ; The region below the dividing line not only contains the spur region, but also the noise region outside the spurs is in the region below the search window dividing line. However, the distribution of the spur region is like a "pyramid" shape. The closer to the cartilage surface, the more pixel points are distributed in the spur region, the more sliding windows there are, and the higher the similarity between the sliding windows. 1 - The weight should be more allocated to the pixel points in the spur region, so as to ensure that the characteristic information of the spur region will not be overly smoothed as much as possible during the filtering process.

[0068] Specifically, obtain the central pixel point of the search window in the lower half-region grayscale image, denoted as the target central pixel point, and obtain the sliding window of the target central pixel point, denoted as the target sliding window; then obtain the sliding windows of other pixel points within the search window, denoted as the marked sliding windows. Take the sum of the squares of the differences between the pixel points in the target sliding window and the grayscale values of all pixel points in the marked sliding window as the grayscale mean square error between the target sliding window and the marked sliding window, which is used to represent the mean square error corresponding to the central pixel point within the marked sliding window. Since there is no difference between itself and itself, the mean square error of the target central pixel point is 0. After obtaining the mean square error corresponding to each pixel point, the weight of each pixel point in the lower half-region grayscale image can be represented by the proportion that the mean square error corresponding to each pixel point occupies in the total sum of the mean square errors corresponding to all pixel points in the lower half-region grayscale image.

[0069] Then the weight of the v-th pixel point in the lower half-region grayscale image can be expressed as:

[0070]

[0071] In the formula, represents the mean square error corresponding to the v-th pixel point in the lower half-region grayscale image, represents the number of pixel points within the search window, represents the mean square error corresponding to the j-th pixel point in the lower half-region grayscale image, represents the weight of the v-th pixel point in the lower half-region grayscale image, represents the exponential function with the natural constant as the base.

[0072] Among them, represents the total sum of the mean square errors corresponding to all pixel points in the lower half-region grayscale image, represents the proportion that the mean square error corresponding to a single pixel point in the lower half-region grayscale image occupies in the total sum of the mean square errors corresponding to all pixel points in the lower half-region grayscale image, which is used to represent the weight of a single pixel point in the lower half-region grayscale image.

[0073] Under normal circumstances, the weight of the search window is divided into two equal parts, with each of the upper and lower regions accounting for 0.5. After the filter weights determined above, the weight of the lower region is (For the upper bone spurs in the upper region, the calculation method is the same as that of the lower region), and then according to the value of the overall weight of the lower region is redistributed.

[0074] Then the grayscale value corresponding to each pixel point after weighting is:

[0075]

[0076] In the formula, represents the gray value of the v-th pixel in the lower half gray image, represents the gray value of the r-th pixel in the upper half gray image, represents the weight of the v-th pixel in the lower half gray image, represents the weight of the r-th pixel in the upper half gray image, represents the number of pixels within the search window, represents the gray value after weighting the central pixel of the search window, represents the filter weight in the lower half gray image, represents the filter weight in the upper half gray image.

[0077] Similarly, the gray value after weighting each pixel in the knee joint gray image is obtained, and the gray value after weighting each pixel is used as the true gray value of each pixel.

[0078] Among them, represents the weighted weight of each pixel that may belong to the bone spur area in the lower half area; represents the weighted weight of each pixel in the upper half area; the weighted average of the pixels in the upper and lower areas can obtain the true gray value of the central pixel. This makes the weight of the pixels that are considered most likely to belong to the bone spur area larger, and during the non-local mean filtering process, it can better ensure that these pixels will not be over-smoothed and can better retain the effective information.

[0079] Thus far, the true gray value of each pixel in the knee joint gray image is obtained.

[0080] Step S003: Obtain the denoised gray image according to the gray value after weighting each pixel.

[0081] By analyzing the above-mentioned search window, the true gray value of the central pixel of the search window is obtained, and then by analyzing the search window corresponding to each pixel in the knee joint gray image, the true gray value corresponding to each pixel is obtained. Based on the true gray value of each pixel, the denoised gray image is obtained.

[0082] Thus far, the removal of the noise in the knee joint ultrasound image is completed.

[0083] Thus far, this embodiment is completed.

[0084] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An enhanced denoising method for clinical ultrasound images, characterized in that, The method includes the following steps: Process the collected knee joint ultrasound images to obtain knee joint grayscale images; Preset a search window and a sliding window in the knee joint grayscale image. Divide the search window into an upper half area and a lower half area. Denote the upper half area as the upper half area grayscale image and the lower half area as the lower half area grayscale image; Stratify the upper half area grayscale image and the lower half area grayscale image according to the width of the sliding window to obtain several layers; Obtain the feature extraction factor for each layer based on the pixel points of each layer; Obtain the filter weights of the lower half area grayscale image and the filter weights of the upper half area grayscale image according to the change of the feature extraction factor for each layer; Obtain the weight of each pixel point in the search window according to the difference between the pixel points in the sliding window corresponding to the central pixel point in the search window and the pixel points in the sliding window corresponding to the remaining pixel points in the search window; Obtain the weighted grayscale value of each pixel point according to the filter weights of the lower half area grayscale image, the filter weights of the upper half area grayscale image, the weight of each pixel point in the search window, and the grayscale value of each pixel point in the search window; Obtain the denoised grayscale image according to the weighted grayscale value of each pixel point; The specific steps for obtaining the feature extraction factor for each layer are as follows: Denote any layer in the lower half area grayscale image as the target layer. The formula for the feature extraction factor of the target layer is: In the formula, represents the feature extraction factor of the target layer, represents the gray value of the i-th pixel of the middle row of the target layer, represents the gray value of the z-th pixel within the sliding window neighborhood centered on the i-th pixel of the middle row of the target layer, where L represents the width of the sliding window, represents the total number of pixels within the sliding window, represents the number of pixels of the target layer; Similarly, obtain the feature extraction factor for each layer; The step of stratifying the upper half area grayscale image and the lower half area grayscale image according to the width of the sliding window to obtain several layers includes the following specific steps: After obtaining the upper half area grayscale image and the lower half area grayscale image, divide the upper half area grayscale image and the lower half area grayscale image respectively according to the width of the sliding window. Make an equal division in the vertical direction with the width L of the sliding window to obtain several layers.

2. The enhanced denoising method for clinical ultrasound imaging according to claim 1, wherein The specific steps for obtaining the filter weights of the lower half area grayscale image and the filter weights of the upper half area grayscale image are as follows: The formula for the filter weights of the lower half area grayscale image is: In the formula, represents the difference between the feature extraction factors of the m-th layer and the adjacent lower layer in the lower half gray-scale image, represents the average value of the differences of the feature extraction factors between all adjacent layers in the lower half gray-scale image, represents the number of layers in the lower half gray-scale image, represents the hyperbolic tangent function, represents the filter weight in the lower half gray-scale image; The filter weights in the upper half grayscale image are obtained based on the filter weights in the lower half grayscale image as .

3. The enhanced denoising method for clinical ultrasound imaging according to claim 1, wherein, The specific steps for obtaining the weight of each pixel point in the search window are as follows: The formula for the weight of the v-th pixel point in the search window is: In the formula, represents the mean square error corresponding to the v-th pixel point in the lower half-region grayscale image, represents the number of pixel points within the search window, represents the mean square error corresponding to the j-th pixel point in the lower half-region grayscale image, represents the weight of the v-th pixel point in the lower half-region grayscale image, represents the exponential function with the natural constant as the base.

4. The enhanced denoising method for clinical ultrasound imaging according to claim 3, characterized in that The specific steps for obtaining the mean square error corresponding to the v-th pixel point in the lower half area grayscale image are as follows: Obtain the central pixel point of the search window in the lower half area grayscale image, denoted as the target central pixel point. Obtain the sliding window of the target central pixel point, denoted as the target sliding window; Obtain the sliding window of the v-th pixel point in the search window, denoted as the marked sliding window; Take the sum of the squares of the differences between the pixel points in the target sliding window and all the pixel points in the marked sliding window as the mean square error corresponding to the v-th pixel point in the lower half area grayscale image.

5. The enhanced denoising method for clinical ultrasound imaging according to claim 1, characterized in that, The specific steps for obtaining the weighted grayscale value of each pixel point are as follows: The formula for the weighted grayscale value of each pixel point is: Wherein, represents the gray value of the v-th pixel in the lower half gray image, represents the gray value of the r-th pixel in the upper half gray image, represents the weight of the v-th pixel in the lower half gray image, represents the weight of the r-th pixel in the upper half gray image, represents the number of pixels in the search window, represents the gray value of the center pixel of the search window after weighting, represents the filter weight in the lower half gray image, represents the filter weight in the upper half gray image.

6. The enhanced denoising method for clinical ultrasound imaging according to claim 1, wherein, The step of obtaining the denoised grayscale image according to the weighted grayscale value of each pixel point includes the following specific steps: Take the weighted grayscale value of each pixel point as the true grayscale value of each pixel point. Denote the image composed of the true grayscale values of each pixel point in the knee joint grayscale image as the denoised grayscale image.

Images