Focus positioning method based on cardiac ultrasonic image

Abstract

The invention provides a focus positioning method based on a cardiac ultrasound image, which belongs to the technical field of cardiac ultrasound images and comprises the following steps: firstly, preprocessing an original image to remove noise and artifacts; then, through multi-dimensional matrix comparison and multi-scale decomposition, two types of abnormal feature matrixes are extracted; after the two abnormal feature matrixes are fused, a heart abnormal region can be positioned more accurately. And then determining the boundary of the abnormal region by adopting a density clustering algorithm, thereby completing the accurate positioning of the heart focus. And finally, marking a positioning result on the processed cardiac ultrasound image to form a visual focus positioning image. According to the method, the technologies of image preprocessing, feature extraction, anomaly detection, region segmentation and the like are combined, the focus region of the heart disease can be accurately positioned, and the technical problem that an existing neural network image recognition method is prone to ignoring details due to detail differences existing in heart ultrasonic images of different patients is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cardiac ultrasonic imaging, and in particular, relates to a lesion positioning method based on cardiac ultrasonic imaging. Background Art

[0002] Cardiac ultrasound examination is one of the important means of clinical diagnosis of heart disease. By collecting and analyzing ultrasound images of various parts of the heart, doctors can comprehensively evaluate the structure and function of the heart, providing key evidence for the diagnosis of heart disease. In recent years, with the continuous advancement of ultrasound imaging technology, the resolution and clarity of cardiac ultrasound images have been continuously improved, providing more abundant information for clinical diagnosis. However, due to the complex anatomical structure and functional characteristics of the heart itself, as well as factors such as large individual differences, diagnosis based solely on human eye observation of cardiac ultrasound images often has certain subjectivity and errors. In particular, for some heart disease lesions with strong occultness, such as myocardial infarction and myocardial hypertrophy, the performance on ultrasound images may not be obvious enough and can be easily overlooked. In the prior art, the lesion localization of cardiac ultrasound images often uses the method of neural network image recognition. Due to the differences in details of cardiac ultrasound images of different patients, the existing neural network image recognition method is prone to ignore details, which affects the reliability of lesion localization. Summary of the invention

[0003] In view of this, the present invention provides a lesion localization method based on cardiac ultrasound images, which can solve the technical problem that the existing neural network image recognition method easily ignores details due to the differences in details of cardiac ultrasound images of different patients.

[0004] The present invention is achieved in that:

[0005] The present invention provides a method for locating a lesion based on cardiac ultrasound imaging, comprising the following steps:

[0006] S10, collecting an original cardiac ultrasound image, and preprocessing the original cardiac ultrasound image to obtain a preprocessed cardiac ultrasound image;

[0007] S20, performing a multi-dimensional matrix comparison calculation on the pre-processed cardiac ultrasound image and a standard cardiac ultrasound image to obtain a first abnormal feature matrix;

[0008] S30, performing multi-scale decomposition calculation on the preprocessed cardiac ultrasonic image to obtain a stable component and a variable component of the cardiac ultrasonic image;

[0009] S40, establishing a correlation coefficient matrix based on the stable component and the variable component of the cardiac ultrasonic image to obtain a second abnormal feature matrix;

[0010] S50, performing a fusion operation on the first abnormal feature matrix and the second abnormal feature matrix to obtain a fused abnormal feature matrix;

[0011] S60, extracting an abnormal point coordinate set according to the fused abnormal feature matrix, wherein the abnormal point coordinate set includes a plurality of abnormal points;

[0012] S70, performing density cluster analysis on the outliers in the outlier coordinate set to obtain a plurality of cluster center points;

[0013] S80, establishing the lesion area boundary based on the multiple cluster center points to obtain the lesion localization result;

[0014] S90: Mark the lesion localization result on the preprocessed cardiac ultrasound image to generate a lesion localization image.

[0015] The calculation process involved in the present invention is described in detail as follows:

[0016] 1. The specific representation of the first abnormal feature matrix is ​​as follows:

[0017]

[0018] Where, d ij represents the difference between the preprocessed cardiac ultrasound image and the standard cardiac ultrasound image at position (i, j); m and n represent the number of rows and columns of the image, respectively.

[0019] 2. The specific process of performing multi-dimensional matrix comparison calculation on the pre-processed cardiac ultrasound image and the standard cardiac ultrasound image is as follows:

[0020]

[0021] In the formula, I ij is the gray value of the preprocessed cardiac ultrasound image at position (i, j); I std,ij is the gray value of the standard cardiac ultrasound image at position (i, j); and Respectively represent the gradient values ​​of the two images at position (i, j); ΔI ij and ΔI std,ij Respectively represent the Laplace operator values ​​of the two images at position (i, j); α1, α2, α3 are weight coefficients, and satisfy α1+α2+α3=1, and the value range is [0,1]; ε1 is the error term.

[0022] 3. The specific process of performing multi-scale decomposition calculation on the pre-processed cardiac ultrasound image is as follows:

[0023]

[0024] I variable =II stable ;

[0025] In the formula, I stable I is the cardiac ultrasound image stabilization component; variable is the variation component of the cardiac ultrasound image; I is the preprocessed cardiac ultrasound image; The scale parameter is σ k Gaussian kernel function; K is the number of decomposition scales, ranging from [3,5]; β k is the weight coefficient of the kth scale and satisfies * indicates convolution operation; ε2 is the error term.

[0026] 4. The specific expression of the correlation coefficient matrix is ​​as follows:

[0027]

[0028] In the formula, r ij Represents the correlation coefficient between the stable component and the variable component at position (i, j).

[0029] 5. The specific process of establishing the correlation coefficient matrix based on the stable component and the variable component of the cardiac ultrasonic image is as follows:

[0030]

[0031] In the formula, I stabke,ij and I variable,ij Respectively represent the values ​​of the stable component and the variable component at position (i, j); μ stable and μ variable They represent the means of the stable component and the variable component respectively; γ is the weight coefficient, and its value range is [0,1]; ε3 is the error term.

[0032] 6. The specific representation of the second abnormal feature matrix is ​​as follows:

[0033] M2=R⊙(|I stable -I variable |);

[0034] Where ⊙ represents the Hadamard product of the matrices.

[0035] 7. The specific process of fusing the abnormal feature matrix is ​​as follows:

[0036] M fusion =λ1M1+λ2M2+ε4;

[0037] Where λ1, λ2 are fusion weight coefficients, and satisfy λ1+λ2=1, and the value range is [0,1]; ε4 is the error term.

[0038] 8. The specific process of extracting the abnormal point coordinate set based on the fused abnormal feature matrix is ​​as follows:

[0039] P={(i,j)|m ij >μ fusion +δσ fusion};

[0040] Where P is the coordinate set of the outlier points; m ij is the value at position (i, j) in the fused abnormal feature matrix; μ fusion is the mean of the fused abnormal feature matrix; σ fusion is the standard deviation of the fused abnormal feature matrix; δ is the threshold coefficient, and its value range is [2,3].

[0041] 9. The specific process of performing density cluster analysis on the outliers in the outlier coordinate set is as follows:

[0042]

[0043] In the formula, ρ i is the local density of the ith point; d ij is the Euclidean distance between point i and point j; η is the attenuation parameter of density calculation, ranging from [0.1, 0.5]; δ i is the distance from the ith point to the nearest point with a greater density than it.

[0044] The construction principles and meanings of these equations are as follows:

[0045] 1. The first abnormal feature matrix adopts a multi-scale comparison method to comprehensively capture the abnormal features of the image by considering the gray value, gradient value and Laplace operator value of the image;

[0046] 2. Multi-scale decomposition uses Gaussian kernel function, which can effectively separate the stable and variable components of the image, facilitating subsequent analysis;

[0047] 3. The correlation coefficient matrix takes into account the correlation and difference between stable components and variable components, and can better identify abnormal areas;

[0048] 4. The fusion of abnormal feature matrices comprehensively considers the information of multiple feature matrices through weight fusion;

[0049] 5. The outlier point extraction adopts the adaptive threshold method, which can automatically adjust the threshold according to the image features;

[0050] 6. Density cluster analysis uses dual indicators of local density and distance to effectively identify cluster centers.

[0051] Compared with the prior art, the beneficial effects of the lesion localization method based on cardiac ultrasound imaging provided by the present invention are:

[0052] 1. Make full use of multi-scale image analysis to extract the features of stable components and variable components, which can more comprehensively characterize the abnormal features of cardiac ultrasound images and improve positioning accuracy. Existing methods based on segmentation or feature extraction often only focus on local areas or fixed features, and it is difficult to capture the full picture of complex lesions. The present invention achieves multi-scale decomposition through Gaussian filtering to obtain the stable component and variable component of the image. The correlation analysis can well reflect the abnormal relationship between these two components, thereby more accurately identifying the heart disease focus area.

[0053] 2. Adopting an adaptive anomaly detection strategy, the anomaly judgment threshold can be automatically adjusted according to the image characteristics, thereby improving the robustness of the positioning result. Existing methods mostly use empirical thresholds or trained models, which are difficult to cope with the image differences of different patients. The present invention automatically determines the appropriate anomaly judgment standard by calculating the statistical characteristics of the fusion feature matrix, which can adapt to different image data, thereby improving the overall positioning reliability.

[0054] 3. The regional segmentation strategy combined with density clustering can more accurately describe the boundary shape of the lesion, which is helpful for clinical diagnosis. The existing segmentation-based methods often find it difficult to obtain a lesion area that is close to the actual situation. The present invention uses the idea of ​​density clustering to determine the cluster center according to the distribution characteristics of abnormal points, and then constructs a more reasonable lesion boundary, solving the technical problem that the existing neural network image recognition method easily ignores details due to differences in details in cardiac ultrasound images of different patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0057] like Figure 1 FIG. 1 is a flow chart of a method for locating a lesion based on cardiac ultrasound imaging provided by the present invention. The method comprises the following steps:

[0058] S10, collecting original cardiac ultrasound images, and preprocessing the original cardiac ultrasound images to obtain preprocessed cardiac ultrasound images;

[0059] S20, performing multi-dimensional matrix comparison calculation on the preprocessed cardiac ultrasound image and the standard cardiac ultrasound image to obtain a first abnormal feature matrix;

[0060] S30, performing multi-scale decomposition calculation on the preprocessed cardiac ultrasound image to obtain a stable component and a variable component of the cardiac ultrasound image;

[0061] S40, establishing a correlation coefficient matrix based on the stable component and the variable component of the cardiac ultrasonic image to obtain a second abnormal feature matrix;

[0062] S50, performing a fusion operation on the first abnormal feature matrix and the second abnormal feature matrix to obtain a fused abnormal feature matrix;

[0063] S60, extracting an abnormal point coordinate set according to the fused abnormal feature matrix, where the abnormal point coordinate set includes multiple abnormal points;

[0064] S70, performing density cluster analysis on the outliers in the outlier coordinate set to obtain multiple cluster center points;

[0065] S80, establishing the boundary of the lesion area based on the multiple cluster center points to obtain the lesion positioning result;

[0066] S90: Mark the lesion localization result on the preprocessed cardiac ultrasound image to generate a lesion localization image.

[0067] The specific implementation methods of the above steps are described in detail below:

[0068] First, the implementation method of step S10. This step includes preprocessing the original cardiac ultrasound image, and the specific process is as follows:

[0069] Step 101, collecting an original cardiac ultrasound image, and graying the original cardiac ultrasound image:

[0070]

[0071] In the formula, I gray is the grayscale image, R, G, B are the red, green and blue three-channel values ​​of the original color image. Grayscale can convert the original color image into a single-channel grayscale image, laying the foundation for subsequent image processing.

[0072] Step 102, performing noise removal processing on the grayscaled cardiac ultrasound original image to obtain a first processed image:

[0073] I denoised =Median(I gray ,k×k);

[0074] In the formula, I denoisedis the image after noise removal, Median(·, k×k) represents the median filtering operation of a k×k window on the input image, usually k=3 or 5. Median filtering can effectively remove impulse noise and speckle noise in the image and improve the signal-to-noise ratio of the image.

[0075] Step 103, performing histogram equalization processing on the first processed image to obtain a second processed image:

[0076] I equalized =HistogramEqualization(I denoised );

[0077] In the formula, I equalized is the image after histogram equalization, and HistogramEqualization(·) indicates that the input image is subjected to histogram equalization. Histogram equalization can stretch the grayscale distribution of the image and enhance the contrast of the image, which is beneficial for subsequent feature extraction.

[0078] Step 104: perform edge enhancement processing on the second processed image to obtain a third processed image:

[0079]

[0080] In the formula, I enhanced is the image after edge enhancement, I equalized The gradient image obtained by Sobel operator calculation, α is the weight coefficient, and the value range is [0.1, 0.5]. Edge enhancement can highlight the edge information in the image, enhance the detailed features of the image, and provide more effective information for subsequent analysis.

[0081] Step 105: Output the third processed image as a preprocessed cardiac ultrasound image. At this point, the preprocessing process of the original cardiac ultrasound image is completed.

[0082] Next is step S20, which is to perform a multi-dimensional matrix comparison calculation on the pre-processed cardiac ultrasound image and the standard cardiac ultrasound image to obtain the first abnormal feature matrix M1. The specific process is as follows:

[0083]

[0084] Among them, d ij It represents the difference between the preprocessed cardiac ultrasound image and the standard cardiac ultrasound image at position (i, j), and the calculation formula is:

[0085]

[0086] In the formula, I ij and Istd,ij Respectively represent the grayscale values ​​of the preprocessed cardiac ultrasound image and the standard cardiac ultrasound image at position (i, j); and Respectively represent the gradient values ​​of the two images at position (i, j); ΔI ij and ΔI std,ij Respectively represent the Laplace operator values ​​of the two images at position (i, j); α1, α2, α3 are weight coefficients, satisfying α1+α2+α3=1, and ε1 is the error term. By calculating the difference between the gray value, gradient value and Laplace operator value, the difference between the preprocessed cardiac ultrasound image and the standard image can be fully characterized, thereby obtaining the first abnormal feature matrix M1.

[0087] Then comes step S30, i.e., performing multi-scale decomposition calculation on the pre-processed cardiac ultrasound image to obtain the stable component and the variable component of the cardiac ultrasound image. The specific process is as follows:

[0088]

[0089] i variable =II stable ;

[0090] In the formula, I stable is the stable component of cardiac ultrasound image, I variable is the variation component of the cardiac ultrasound image, I is the preprocessed cardiac ultrasound image, The scale parameter is σ k Gaussian kernel function, K is the number of decomposition scales, the value range is [3,5], β k is the weight coefficient of the kth scale, satisfying * represents convolution operation, and ε2 is the error term. Through multi-scale Gaussian filtering, relatively stable components (background information) and relatively changing components (foreground information) in the image can be extracted, providing a basis for subsequent anomaly detection and region segmentation.

[0091] Next is step S40, i.e., establishing a correlation coefficient matrix R based on the stable component and the variable component of the cardiac ultrasound image to obtain a second abnormal feature matrix M2. The specific process is as follows:

[0092]

[0093] Among them, r ij It represents the correlation coefficient between the stable component and the variable component at position (i, j), and the calculation formula is:

[0094]

[0095] In the formula, I stable,ij and Ivariable,ij Respectively represent the values ​​of the stable component and the variable component at position (i, j), μ stable and μ variable Represent the mean of the stable component and the variable component respectively, γ is the weight coefficient, the value range is [0,1], and ε3 is the error term. By calculating the correlation coefficient and the difference, the relationship between the stable component and the variable component can be quantified, thereby constructing the second abnormal feature matrix:

[0096] M2=R⊙(|I stable -I variable |);

[0097] Where ⊙ represents the Hadamard product of the matrices.

[0098] Next is step S50, which is to fuse the first abnormal feature matrix M1 with the second abnormal feature matrix M2 to obtain a fused abnormal feature matrix M fusion The specific process is as follows:

[0099] M fusion =λ1M1+λ2M2+ε4;

[0100] In the formula, λ1,λ2 are fusion weight coefficients, satisfying λ1+λ2=1, with a value range of [0,1], and ε4 is the error term. By weighted fusion of two abnormal feature matrices, the abnormal information of the image at different feature levels can be comprehensively considered to obtain a more comprehensive abnormal feature representation.

[0101] Next is step S60, that is, according to the fusion abnormal feature matrix M fusion Extract the outlier point coordinate set p. The specific process is as follows:

[0102] p={(i,j)|m ij >μ fusion +δσ fusion};

[0103] In the formula, m ij is the value at position (i, j) in the fused abnormal feature matrix, μ fusion and σ fusion are the mean and standard deviation of the fused abnormal feature matrix, and δ is the threshold coefficient, with a value range of [2,3]. By setting an adaptive abnormality judgment threshold, a significant abnormal point coordinate set can be extracted from the fused abnormal feature matrix, providing a basis for subsequent lesion area identification.

[0104] Next is step S70, that is, performing density clustering analysis on the outlier point coordinate set P to obtain multiple cluster center points. The specific process is as follows:

[0105]

[0106] In the formula, ρ i is the local density of the ith point, d ij is the Euclidean distance between point i and point j, η is the attenuation parameter of density calculation, the value range is [0.1, 0.5], δ i is the distance from the ith point to the nearest point with a higher density. By calculating the local density of the outlier point and the shortest distance to the point with a higher density, the cluster center can be effectively identified, providing a basis for the subsequent determination of the boundary of the lesion area.

[0107] Finally, step S80 and step S90 are to establish the boundary of the lesion area based on the cluster center point and mark the result on the preprocessed cardiac ultrasound image. This part does not involve mathematical formulas, but mainly conducts subsequent boundary construction and visualization output based on the results obtained in the previous steps, which will not be repeated.

[0108] Specifically, the principle of the present invention is:

[0109] 1. Multi-scale image analysis. Cardiac ultrasound images often contain information of different scales, such as myocardial texture, cardiac cavity boundaries, and overall morphology. The present invention uses multi-scale Gaussian filtering technology to decompose the original image into a stable background component and a changing foreground component. The former contains the overall structural information of the heart, while the latter reflects local abnormal changes. By analyzing the relationship between these two components, the abnormal characteristics of cardiac ultrasound images can be more comprehensively portrayed.

[0110] 2. Multi-dimensional feature fusion. The present invention not only considers the difference in image grayscale values, but also introduces features such as gradient and Laplace operator to construct the first abnormal feature matrix. At the same time, a correlation coefficient matrix is ​​established based on the stable component and the variable component to obtain the second abnormal feature matrix. By weighted fusion of these two feature matrices, a variety of abnormal information can be comprehensively utilized to improve the accuracy of positioning.

[0111] 3. Adaptive anomaly detection. In order to cope with the differences in cardiac ultrasound images of different patients, the present invention adopts an adaptive anomaly determination strategy. Specifically, the statistical characteristics of the fused abnormal feature matrix, such as the mean and standard deviation, are first calculated, and then the abnormality determination threshold is dynamically determined based on the confidence interval coefficient set empirically. This can not only effectively identify significant abnormal areas, but also adapt to the characteristics of different image data, thereby improving overall robustness.

[0112] 4. Density-based region segmentation. After determining the abnormal point coordinate set, the present invention uses a density-based cluster analysis method to identify multiple cluster center points. These cluster center points represent the possible lesion area centers. Then, based on these cluster centers and combined with the distribution characteristics of the abnormal points, a more accurate lesion area boundary is constructed. This density-based segmentation method can better describe the irregular shape of the lesion and provide more valuable information for clinical diagnosis.

[0113] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: A 55-year-old male patient, Mr. Wang, was admitted to the cardiology department of a tertiary hospital. Mr. Wang was diagnosed with chest pain and was initially diagnosed with acute myocardial infarction. The doctor decided to perform a cardiac ultrasound examination on Mr. Wang to further clarify his condition.

[0114] The inspection process is as follows:

[0115] Collect original cardiac ultrasound images

[0116] The doctor performed a standard transthoracic ultrasound examination on Mr. Wang and collected a series of original cardiac ultrasound images including conventional sections such as the two-chamber, four-chamber, and short-axis sections. The resolution of these images was 512×512 pixels and the color depth was 8-bit grayscale.

[0117] In order to enhance the quality and features of the images, the doctors preprocessed the collected original cardiac ultrasound images, including the following steps:

[0118] Grayscale processing: Convert the original color cardiac ultrasound image into a single-channel grayscale image to facilitate subsequent image analysis. The grayscale value calculation formula is:

[0119]

[0120] Among them, I gray is a grayscale image, and R, G, and B are the three channel values ​​of the original color image.

[0121] Noise removal: Perform a 3×3 window median filter on the grayscale image to remove impulse noise and speckle noise in the image. The median filter formula is:

[0122] I denoised =Median(I gray ,3×3);

[0123] Histogram equalization: Perform histogram equalization on the denoised image to stretch the grayscale distribution of the image and enhance the contrast. The histogram equalization formula is:

[0124] I equalized=HistogramEqualization(I denoised );

[0125] Edge enhancement: Perform Sobel operator edge enhancement on the image after histogram equalization to highlight the details in the image. The edge enhancement formula is:

[0126]

[0127] After the above preprocessing steps, the original cardiac ultrasound image has been greatly improved, the noise has been effectively suppressed, and the contrast and edge information have been enhanced, laying a good foundation for subsequent feature extraction and analysis. Preprocessed cardiac ultrasound image.

[0128] Extract abnormal features of cardiac ultrasound images: In order to locate possible myocardial infarction lesions, doctors then performed multi-dimensional feature extraction and abnormal analysis on the pre-processed cardiac ultrasound images, including:

[0129] Comparison with standard images: The doctor first obtained a set of standard normal cardiac ultrasound images as a reference. Then, Mr. Wang's pre-processed images were compared with the standard images one by one in pixel-wise difference calculation to obtain the first abnormal feature matrix M1. The calculation formula is:

[0130]

[0131] Among them, d ij Represents the difference between the preprocessed image and the standard image at position (i, j); I ij ,I std,ij are the gray values ​​of the preprocessed image and the standard image at that position respectively; is the gradient value of the two images at this position; ΔI ij ,ΔI std,ij is the Laplacian operator value of the two images at this position. By comprehensively considering the grayscale, gradient and Laplacian features, the difference between the images can be more comprehensively characterized.

[0132] Multi-scale decomposition: The doctor also performed multi-scale decomposition on the pre-processed cardiac ultrasound image to extract the stable background component I stable and the changing outlook component I variable The specific formula is:

[0133]

[0134] I variable =II stable ;

[0135] in, The scale parameter is σk The Gaussian kernel function can separate the stable overall structural information and the changing local features in the image through Gaussian filtering at five different scales.

[0136] Correlation analysis: Based on the stable and variable components obtained above, the doctor further calculated the correlation coefficient matrix R between them to reflect the characteristics of the abnormal area in the image:

[0137]

[0138] Among them, μ stable ,μ variable are the means of the stable component and the variable component respectively. By considering the correlation coefficient and the difference, the abnormal area in the image can be better identified.

[0139] Combining the above three features, the doctor constructed two abnormal feature matrices M1 and M2, and weightedly fused them to obtain the final fused abnormal feature matrix M fusion :

[0140] M fusion =0.6M1+0.4M2;

[0141] Extracting the lesion area: Based on the fusion abnormal feature matrix M fusion ,The doctor then took the following steps to extract the area where the lesion might be:

[0142] Abnormal point extraction: First, the doctor calculated M fusion The mean μ fusion and standard deviation σ fusion , and then set the abnormality judgment threshold to μ fusion +2.5σ fusion . fusion The coordinates corresponding to the elements greater than the threshold are extracted to form the outlier point coordinate set P.

[0143] Density clustering analysis: For the abnormal point coordinate set P, the doctor uses a density-based clustering algorithm to analyze to determine the possible center of the lesion area. Specifically including:

[0144] Calculate the Euclidean distance between each point in P;

[0145] Set the density decay parameter η = 0.3;

[0146] Calculate the local density at each point

[0147] Calculate the distance from each point to the nearest point with a density greater than that of the point

[0148] According to i and δi Determine the cluster center point as the possible center of the lesion area.

[0149] Lesion area construction: Finally, the doctor constructs a preliminary lesion area boundary based on the determined cluster center point and the distribution characteristics of the abnormal point P. Then, the boundary is smoothed to obtain the final lesion localization result.

[0150] Through the above steps, the doctor successfully located the suspected myocardial infarction lesion in Mr. Wang's cardiac ultrasound image. The area is located in the middle and lower part of the left ventricular free wall, with an area of ​​about 2.5cm 2 , the edges are relatively irregular. The doctor marked the positioning results on the pre-processed cardiac ultrasound image to form an intuitive visualization effect. Based on the above positioning results, combined with clinical symptoms and other examinations, the doctor finally diagnosed Mr. Wang with acute anterior wall myocardial infarction. Subsequently, the doctor immediately arranged for Mr. Wang to undergo rescue percutaneous coronary intervention in order to restore blood supply to the infarct area as soon as possible and reduce myocardial damage.

[0151] During the follow-up, the doctor continued to pay close attention to Mr. Wang's condition. Through subsequent cardiac ultrasound examinations, it was found that after emergency treatment, Mr. Wang's myocardial infarction area gradually shrank and the ventricular wall motion function also recovered. Based on this, the doctor judged that Mr. Wang's prognosis was relatively optimistic and he was expected to recover good heart function in the future.

[0152] In general, the lesion localization method based on cardiac ultrasound images proposed in the present invention has played an important role in Mr. Wang's diagnosis and treatment. It can locate the lesion area of ​​myocardial infarction more accurately, providing an intuitive and reliable basis for clinical diagnosis. Compared with traditional single features or empirical judgments, this method integrates multiple image analysis technologies, which can more comprehensively characterize the abnormal features of cardiac ultrasound images and improve the accuracy and reliability of positioning. At the same time, it also has a certain degree of adaptability, and can dynamically adjust the threshold of abnormal detection according to the image characteristics of different patients, thereby enhancing the overall robustness.

[0153] It should be noted that the explanations of relevant parameters and variables are shown in Table 1 below:

[0154] Table 1 Variable and parameter explanation table

[0155]

[0156]

[0157] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for lesion localization based on cardiac ultrasound imaging, characterized in that: The following steps are involved: S10, collecting an original cardiac ultrasound image, and preprocessing the original cardiac ultrasound image to obtain a preprocessed cardiac ultrasound image; S20, performing a multi-dimensional matrix comparison calculation on the pre-processed cardiac ultrasound image and a standard cardiac ultrasound image to obtain a first abnormal feature matrix; S30, performing multi-scale decomposition calculation on the preprocessed cardiac ultrasonic image to obtain a stable component and a variable component of the cardiac ultrasonic image; S40, establishing a correlation coefficient matrix based on the stable component and the variable component of the cardiac ultrasonic image to obtain a second abnormal feature matrix; S50, performing a fusion operation on the first abnormal feature matrix and the second abnormal feature matrix to obtain a fused abnormal feature matrix; S60, extracting an abnormal point coordinate set according to the fused abnormal feature matrix, wherein the abnormal point coordinate set includes a plurality of abnormal points; S70, performing density cluster analysis on the outliers in the outlier coordinate set to obtain a plurality of cluster center points; S80, establishing the boundary of the lesion area based on the multiple cluster center points to obtain a lesion localization result; S90: Mark the lesion localization result on the preprocessed cardiac ultrasound image to generate a lesion localization image.

2. The method according to claim 1, characterized in that The step S10 specifically includes: Step 101: collecting an original cardiac ultrasound image and performing grayscale processing on the original cardiac ultrasound image; Step 102, performing noise removal processing on the grayscaled original cardiac ultrasound image to obtain a first processed image; Step 103, performing histogram equalization processing on the first processed image to obtain a second processed image; Step 104: performing edge enhancement processing on the second processed image to obtain a third processed image; Step 105: Output the third processed image as a preprocessed cardiac ultrasound image.

3. The method according to claim 1, characterized in that The step S20 specifically includes: Step 201: Acquire a standard cardiac ultrasound image, and extract grayscale values ​​from the standard cardiac ultrasound image; Step 202: performing gradient calculation on the standard cardiac ultrasound image and the preprocessed cardiac ultrasound image to obtain a gradient value; Step 203, performing Laplace operator operation on the standard cardiac ultrasound image and the preprocessed cardiac ultrasound image to obtain a Laplace operator value; Step 204: Perform a weighted sum operation on the grayscale value, the gradient value, and the Laplace operator value to obtain a first abnormal feature matrix.

4. The method according to claim 1, characterized in that The step S30 specifically includes: Step 301, constructing Gaussian kernel functions of different scales; Step 302: performing a convolution operation on the preprocessed cardiac ultrasound image and the Gaussian kernel function to obtain a plurality of Gaussian blurred images; Step 303: performing weighted summation on the multiple Gaussian blurred images to obtain a stable component of the cardiac ultrasound image; Step 304: subtract the cardiac ultrasonic image stability component from the preprocessed cardiac ultrasonic image to obtain the cardiac ultrasonic image variation component.

5. The method according to claim 1, characterized in that The step S40 specifically includes: Step 401, calculating the mean value of the stable component of the cardiac ultrasonic image and the mean value of the variable component of the cardiac ultrasonic image; Step 402, calculating the correlation coefficient between the stable component of the cardiac ultrasonic image and the variable component of the cardiac ultrasonic image; Step 403, calculating the difference between the stable component of the cardiac ultrasonic image and the variable component of the cardiac ultrasonic image; Step 404: construct a correlation coefficient matrix according to the correlation coefficient and the difference; Step 405: Obtain a second abnormal feature matrix through a Hadamard product operation.

6. The method according to claim 1, characterized in that The step S50 specifically includes: Step 501: normalize the first abnormal feature matrix; Step 502: normalize the second abnormal feature matrix; Step 503: Setting the fusion weight coefficient; Step 504: Perform a weighted summation on the normalized first abnormal feature matrix and the normalized second abnormal feature matrix to obtain a fused abnormal feature matrix.

7. The method according to claim 1, characterized in that The step S60 specifically includes: Step 601, calculating the mean and standard deviation of the fused abnormal feature matrix; Step 602: Setting a threshold coefficient; Step 603: Calculate an abnormality determination threshold according to the mean, the standard deviation and the threshold coefficient; Step 604: compare each element in the fused abnormal feature matrix with the abnormality determination threshold; Step 605: extract the position coordinates corresponding to the elements greater than the abnormality determination threshold to form an abnormal point coordinate set.

8. The method according to claim 1, characterized in that The step S70 specifically includes: Step 701, calculating the Euclidean distance between each point in the abnormal point coordinate set; Step 702, setting the attenuation parameter of density calculation; Step 703, calculating the local density of each point in the abnormal point coordinate set; Step 704, calculating the distance from each point in the abnormal point coordinate set to the nearest point with a density greater than that point; Step 705: Determine multiple cluster center points according to the local density and the distance.

9. The method according to claim 1, characterized in that: The step S80 specifically includes: Step 801: construct an initial boundary with the multiple cluster center points as the center; Step 802: modifying the initial boundary according to the distribution of points in the abnormal point coordinate set; Step 803: setting boundary smoothing parameters; Step 804: Smooth the corrected boundary to obtain a lesion localization result.

10. The method according to claim 1, characterized in that The step S90 specifically includes: Step 901, converting the lesion localization result into a binary boundary image; Step 902: Overlay the binary boundary image with the preprocessed cardiac ultrasound image; Step 903, setting the boundary mark color; Step 904: coloring the superimposed image according to the boundary marker color to obtain a lesion localization image.

Images