Ultrasonic elastography optimization method and application thereof in diffuse lesion ultrasonic elastography classification and grading

Abstract

The invention relates to the technical field of ultrasonic imaging, in particular to an ultrasonic elastography optimization method and application of the ultrasonic elastography optimization method in diffuse lesion ultrasonic elastography classification and grading, and the optimization method comprises the following steps: obtaining multiple groups of shear wave elastography data to construct a three-dimensional space data set, and then, local curve fitting and abnormal peak detection are adopted to eliminate operation errors, spherical space weighted mean filtering is adopted to suppress motion artifacts, and finally, an optimized two-dimensional elastic distribution diagram is generated. And the image classification and grading system associates the optimized data with pathological grading, and constructs a classification and grading model by utilizing machine learning. According to the method, the problem of measurement deviation caused by factors such as operator manipulation difference, patient body position change and tissue motion artifacts in an existing ultrasonic elastography technology is effectively solved, and the accuracy and reliability of elastography data are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ultrasonic imaging, and more particularly to an optimization method for ultrasonic elastography and its application in ultrasonic elastography classification and grading of diffuse lesions. Background Art

[0002] Pathological examination of diffuse lesions usually uses methods such as puncture biopsy, surgical biopsy, and internal diameter biopsy, all of which require obtaining partial tissue for biopsy. As a pathological examination, biopsy is the most accurate method, but it still has certain problems and risks, such as sampling error, the risk of tissue damage interference, and the inhomogeneity of diffuse lesions, which makes it difficult for local biopsy to accurately reflect the overall lesion situation. At the same time, patients and their families have low acceptance of this method.

[0003] Ultrasound technology has always played an important role in clinical work, among which ultrasound elastography is based on imaging the elastic differences of biological tissues. Normal and diseased tissues have different elastic moduli, such as normal liver and cirrhotic liver. By applying external pressure, soft tissues deform more and hard tissues deform less. The equipment transmits ultrasonic pulses and receives reflected signals. After signal processing, the tissue elasticity information is converted into visual images to assist diagnosis. As a new ultrasound diagnostic technology, ultrasound elastography can study tumors and diffuse diseases that cannot be detected by traditional ultrasound. It can broaden ultrasound images, make up for the shortcomings of conventional ultrasound, and more vividly display and locate lesions. Ultrasound elastography mainly includes strain elastography (SE) and shear wave elastography (SWE). Among them, liver shear wave elastography has now become a routine examination item in the ultrasound medicine department of many hospitals.

[0004] Although ultrasound elastography technology holds significant value for disease diagnosis, clinical accuracy and reliability still face challenges. These challenges often stem from both operator and patient factors. For example, uneven probe application and inconsistent manipulation can lead to excessive pressure in a specific area, resulting in overcompression of the tissue. Pressure in adjacent areas may be normal, leading to deviations in the measured tissue elasticity. Furthermore, the angle and direction of the probe held by the operator can alter the incident angle of the ultrasound beam, causing additional intensity and phase shifts in the reflected ultrasound signal, leading to reduced measurement accuracy. Furthermore, changes in patient position, voluntary muscle movement, and respiratory movement during ultrasound elastography can all lead to additional, unintended tissue deformation during probing. This additional deformation, superimposed on the deformation caused by normal pressure, complicates the signal received by the elastography device and makes it impossible to accurately distinguish between normal deformation and deformation caused by muscle movement, resulting in deviations in the measured elasticity.

[0005] Therefore, how to solve the problem of unsatisfactory accuracy and reliability of results caused by operators and patients in ultrasonic elastography is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0006] In view of this, the present invention provides an optimization method for ultrasound elastography and its application in the classification and grading of diffuse lesions using ultrasound elastography, which effectively solves the measurement deviation problem caused by factors such as operator technique differences, patient position changes, and tissue motion artifacts in existing ultrasound elastography technology, and significantly improves the accuracy and reliability of elastography data.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a method for optimizing ultrasound elastography, comprising the following steps:

[0009] S1. Get Z k The two-dimensional plane image data points of the shear wave elastography of the target object [X n , Y m ] and its elasticity value I1, stacking the two-dimensional plane image data points of group Zk to form a first three-dimensional space image data point set {(X n , Y m , Z k , I1)}, where X n 、Y m , Z kThe horizontal coordinate, vertical coordinate, and depth coordinate of the midpoint of the three-dimensional space image data point set respectively; n represents the column index, m represents the row index, and k represents the depth index;

[0010] S2, taking [Y m , Z k ] The corresponding data point in the horizontal coordinate direction is X n Construct a data point set {(X n , I1)}, and perform curve fitting to obtain m*k fitting curves; remove the abnormal peaks on each fitting curve to obtain a noise reduction curve, and obtain the X on each noise reduction curve n The elastic value I2 of the corresponding data point is obtained, and the second three-dimensional space image data point set {(X n , Y m , Z k , I2)};

[0011] S3, taking [X n , Y m , Z k ] data point as the center, take the mean elasticity value I3 of all data points in the spherical space around it as [X n , Y m , Z k ] data point new elastic value, forming the third three-dimensional space image data point set {(X n , Y m , Z k , I3)};

[0012] S4, the third three-dimensional space image data point set [X n , Y m ] corresponds to the mean elastic value of all data points in the depth coordinate direction I4 as the two-dimensional plane image data point [X n , Y m ] elastic value, and obtain the final two-dimensional space image data point set {(X n , Y m , I4)}, converted to the corresponding optimized ultrasound elastography.

[0013] Furthermore, in S1, the elastic value is Young's modulus or shear wave velocity; and the values ​​of n, m, and k are 1 to 5000.

[0014] Furthermore, in S2, the curve fitting process includes:

[0015] For the data point set {(X n , I1)} for each data point X n, calculate the weight ω within its neighborhood i , based on the weight ω i , for X n Perform local quadratic polynomial fitting on the data points within the neighborhood to calculate the data point X n The predicted elastic value I1' is traversed through all data points in the horizontal axis direction to generate a complete fitting curve;

[0016] in,

[0017] σ=0.2d,

[0018] f is the smoothing parameter, the value range of f is 0.1≤f≤0.3, N is the data point set {(X n ,I1)} in X n The total number of Indicates rounding down, X i For X n Other data points in the neighborhood.

[0019] Furthermore, in S2, the step of removing abnormal peaks on each fitting curve includes:

[0020] Traverse the elastic value gradient changes of adjacent data points on the fitting curve. When the absolute value of the gradient of a consecutive data point exceeds p times the average value of the gradient of the b data points before and after it, it is marked as a candidate abnormal peak.

[0021] The reference interval is extended to both sides of the candidate abnormal peak to the area where the gradient returns to a stable state. The statistical median of the elasticity value in the reference interval is calculated. If the difference between the peak value of the candidate peak and the median exceeds L% of the median, it is confirmed as a background intensity abnormal peak.

[0022] The elasticity values ​​in the abnormal peak area are replaced by the cubic spline interpolation results of the elasticity values ​​in the reference interval to correct the curve.

[0023] Furthermore, in S2, a=3-10, b=5-20, p≥3; L=20-40.

[0024] Furthermore, in S3, the spherical space is a spherical space with a radius r of 5 to 100 data points; or, the spherical space is an ellipsoidal spherical space with a major axis radius of 1.1r to 1.5r data points and a minor axis radius of 0.5r to 0.8r data points.

[0025] Furthermore, the calculation process of the elasticity value I3 in S3 includes:

[0026] Calculating the median M and standard deviation σ of the elasticity values ​​of all data points in the spherical space, and extracting the interference-reducing data points in the spherical space that meet the condition |I2-M|≤2σ~3σ;

[0027] Assign weight λ to the interference reduction data point i =exp(-d0 / (pr)), where d0 is the Euclidean distance between the noise reduction data point and the center point of the spherical space, and p=0.4-0.6;

[0028] Calculate the weighted average value of each interference reduction data point I3=Σ(λ i ×I2) / Σλ i .

[0029] Furthermore, the direction of the major axis radius is parallel to the longitudinal coordinate direction.

[0030] In a second aspect, the present invention provides an application of the above-mentioned optimization method in the classification and grading of diffuse lesion ultrasound elastography images, comprising:

[0031] Optimize the data set according to the optimization method described above;

[0032] The optimized two-dimensional space image data point set {(X n , Y m , I4)} is associated with the diffuse lesion pathology grade of the target object to construct a training dataset;

[0033] Based on the training data set, a pre-built classification and grading model is trained;

[0034] Based on the trained classification and grading model, the two-dimensional spatial image data point set of the new target object is analyzed, and the corresponding prediction results of the diffuse lesion pathological grade are output.

[0035] Furthermore, the process of constructing the training data set includes:

[0036] Get the optimized two-dimensional space image data point set {(X n , Y m , I4)} the average value M1 of the elasticity values ​​of all points; obtain the optimized two-dimensional space image data point set {(X n , Y m , I4)} the average elasticity value of column X1 and X n The ratio M2 of the average values ​​of the elasticity values ​​on the column is obtained as data (M1, M2);

[0037] The data (M1, M2) are associated with the diffuse lesion pathology grade of the target object to construct a training dataset.

[0038] In a third aspect, the present invention provides a diffuse lesion ultrasound elastography image classification and grading system, comprising:

[0039] A preprocessing module, used to optimize the data set according to the optimization method described above;

[0040] The dataset construction module is used to transform the optimized two-dimensional space image data point set {(X n , Y m , I4)} is associated with the diffuse lesion pathology grade of the target object to construct a training dataset;

[0041] A training module, configured to train a pre-built classification and grading model based on the training data set;

[0042] The prediction module is used to analyze the two-dimensional spatial image data point set of the new target object based on the trained classification and grading model, and output the corresponding prediction result of the diffuse lesion pathological grade.

[0043] It can be seen from the above technical solutions that compared with the prior art, the present invention has the following beneficial effects:

[0044] (1) The present invention can reduce operator dependence and improve measurement consistency. Through three-dimensional spatial data stacking and multi-dimensional noise reduction processing, it eliminates or reduces the interference of manual operation factors such as uneven probe pressure and grip angle deviation on elasticity values. The use of local curve fitting and spherical spatial mean filtering can suppress abnormal peaks in local elasticity values ​​caused by fluctuations in probe pressure, making the data more consistent with the actual tissue mechanical properties.

[0045] (2) The present invention can suppress motion artifacts of the target object and enhance data stability. Through abnormal peak detection and adaptive smoothing correction, it eliminates the interference of unexpected deformation caused by respiratory movement, muscle contraction, etc. on the elastic signal. Based on the weighted mean calculation of the median and standard deviation, it further filters outlier data points caused by changes in body position and preserves the true distribution characteristics of the tissue elastic gradient.

[0046] (3) This invention optimizes the conversion of 3D to 2D data to generate elastic imaging images with a high signal-to-noise ratio, more clearly reflecting the overall hardness distribution of the lesion area. Combined with a machine learning algorithm, the optimized elastic data is correlated with pathological grade to construct a high-precision classification model, achieving non-invasive, quantitative grading of diffuse lesions and reducing reliance on needle biopsies.

[0047] (4) This invention does not require modification of existing ultrasound hardware and is compatible with existing equipment. It has strong clinical applicability and can improve ultrasound image classification performance simply through image optimization, making it easy to promote in clinical practice. It can adapt to a variety of diffuse lesions and output discrete grades or continuous scores to meet the personalized needs of different departments.

[0048] (5) Through multi-level data optimization and intelligent analysis, the present invention has achieved a leap from "experience-based" to "objective quantification" in ultrasound elastography, which has significant clinical value and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0050] Figure 1 A flow chart of the method for optimizing ultrasonic elastography provided by the present invention;

[0051] Figure 2 A flow chart for removing abnormal peaks on each fitting curve provided by the present invention;

[0052] Figure 3 This is a structural block diagram of the diffuse lesion ultrasound elastic imaging image classification and grading system provided by the present invention. DETAILED DESCRIPTION

[0053] 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 creative efforts are within the scope of protection of the present invention.

[0054] like Figure 1 As shown, an embodiment of the present invention discloses an optimization method for ultrasonic elastography, comprising the following steps:

[0055] S1. Get Z k The two-dimensional plane image data points of the shear wave elastography of the target object [X n , Y m ] and its elasticity value I1, stacking the Zk groups of two-dimensional plane image data points to form the first three-dimensional space image data point set {(X n , Y m , Z k , I1)}, where X n 、Y m , Z k The horizontal coordinate, vertical coordinate, and depth coordinate of the midpoint of the three-dimensional space image data point set respectively; n represents the column index, m represents the row index, and k represents the depth index;

[0056] S2, taking the first three-dimensional space image data point set [Y m , Z k ] The corresponding horizontal coordinate data point X n Construct a data point set {(X n , I1)}, and perform curve fitting to obtain m*k fitting curves; remove the abnormal peaks on each fitting curve to obtain a noise reduction curve, and obtain the X on each noise reduction curve n The elastic value I2 of the corresponding data point is obtained, and the second three-dimensional space image data point set {(X n , Y m , Z k , I2)};

[0057] S3, taking the second three-dimensional space image data point set [X n , Y m , Z k ] data point as the center, take the mean elasticity value I3 of all data points in the spherical space around it as [X n , Y m , Z k ] data point new elastic value, forming the third three-dimensional space image data point set {(X n , Y m , Z k , I3)};

[0058] S4, the third three-dimensional space image data point set [X n , Y m ] corresponds to the mean elastic value of all data points in the depth coordinate direction I4 as the two-dimensional plane image data point [X n , Y m ] elastic value, and obtain the final two-dimensional space image data point set {(X n , Y m , I4)}, the conversion is to form the corresponding optimized ultrasound elasticity imaging. The logic of the conversion is that the elasticity value of each point corresponds to a different color.

[0059] The above steps of the present invention are further described below.

[0060] S1. Shear wave elastography is a two-dimensional graphic that displays the elastic distribution of tissue within a certain plane. Each pixel in the image corresponds to an elastic value (such as Young's modulus or shear wave velocity), which is used to quantify the hardness of the tissue.

[0061] The method for obtaining shear wave elastography of the target object (which may be the right lobe of the liver) in group Zk is as follows: prepare to hold the ultrasound probe and place it in the test area of ​​the target object (for example, a liver area of ​​5 cm × 5 cm × 3 cm) for shear wave elastography, and continuously acquire the two-dimensional plane image data points of the shear wave elastography of group Zk [X n , Y m ] and its elasticity value I1, Zk=30~50, and the collection interval is greater than or equal to 0.5 seconds to cover the inspiration-expiration cycle.

[0062] Stack Zk groups of two-dimensional plane image data points to form a first three-dimensional space image data point set {(X n , Y m , Z k , I1)}, where X n 、Y m , Z k The horizontal coordinate, vertical coordinate, and depth coordinate of the midpoint of the three-dimensional space image data point set respectively; n represents the column index, m represents the row index, and k represents the depth index, 256≤n≤2048, 256≤m≤2048; the elastic value is the Young's modulus or shear wave velocity value.

[0063] S2, taking the first three-dimensional space image data point set [Y m , Z k ] The corresponding data point in the horizontal coordinate direction is X n Construct a data point set {(X n , I1)}, and perform curve fitting to obtain m*k fitting curves; the specific curve fitting process includes:

[0064] For the data point set {(X n , I1)} for each data point X n , calculate the weight ω within its neighborhood i , based on the weight ω i , for X n The data points in the neighborhood are fitted with a local quadratic polynomial (i.e., weighted least squares method) to calculate the data point X n The predicted elastic value I1' is traversed through all data points in the horizontal axis direction to generate a complete fitting curve;

[0065] in,

[0066] σ=0.2d, used to adjust the smoothing strength of the Gaussian kernel,

[0067] f is the smoothing parameter, the value range of f is 0.1≤f≤0.3, N is the data point set {(X n ,I1)} in Xn The total number of Indicates rounding down, X i For X n Other data points in the neighborhood, that is, for each data point X n , X i The choice is X n All the neighbors that satisfy |Xi-X n ∣≤d data points.

[0068] The final fitting curve is f(X n , I1'), where I1' is the X after curve fitting n The elasticity value corresponding to the data point.

[0069] After curve fitting, it is necessary to detect and remove abnormal peaks on each fitting curve. The noise reduction curve after removing abnormal peaks is f(X n , I2), such as Figure 2 As shown, the specific steps include:

[0070] 1) Traverse the elastic value gradient changes of adjacent data points on the fitting curve. When the absolute value of the gradient of a consecutive data point exceeds p times the average value of the gradient of b data points before and after it, it is marked as a candidate abnormal peak, where a = 3-10, b = 5-20, and p ≥ 3. This step is used to detect abnormal peaks with gradient mutation.

[0071] 2) Centered around the candidate outlier peak, extend to the region where the gradient returns to a stable state, forming a reference interval. Calculate the statistical median of the elasticity values ​​within this reference interval. If the difference between the candidate peak's apex and the median exceeds L% of the median, the peak is identified as a background intensity outlier, where L = 20-40. This step is used to verify background intensity separation. The background intensity at this location is derived from the elasticity background enhancement caused by pressure errors caused by manual probe grip during shear wave elastography of the target object.

[0072] 3) The elasticity values ​​in the abnormal peak area are replaced by the cubic spline interpolation results of the elasticity values ​​in the reference interval to correct the curve.

[0073] S3, in the second three-dimensional space image data point set [X n , Y m , Z k ] data point as the center, take the mean elasticity value I3 of all data points in the spherical space around it as [X n , Y m , Z k ] data point new elastic value, forming the third three-dimensional space image data point set {(X n , Y m , Z k, I3)}.

[0074] The spherical space is a spherical space with a radius r of 5 to 100 data points, preferably 5 to 15 data points. Alternatively, the spherical space is an ellipsoidal space with a major axis radius of 1.1r to 1.5r data points and a minor axis radius of 0.5r to 0.8r data points. The major axis radius is parallel to the ordinate, matching the high-resolution characteristics of the probe cross-section. The ellipsoidal space is used to reduce interslice artifacts. More preferably, the major axis radius is 1.2r, resulting in higher resolution along the probe cross-section; the minor axis radius is 0.8r, resulting in lower resolution along the probe depth.

[0075] The calculation process of the elasticity value I3 includes:

[0076] 1) Calculate the median M and standard deviation σ of the elasticity values ​​of all data points in the spherical space, and extract the interference-reducing data points in the spherical space that meet the condition |I2-M|≤2σ~3σ;

[0077] The median M sorts the elasticity values ​​I2 of all data points in the spherical space and takes the value in the middle. If the number of data points is even, the average of the two middle values ​​is taken. The standard deviation σ is calculated based on all retained I2 values ​​in the ellipsoid neighborhood, using the formula:

[0078]

[0079] Wherein, μ is the mean of the elasticity values ​​I2 of the data points in the spherical space, N is the total number of data points, and i is the sequence number after sorting the elasticity values ​​I2 of all data points in the spherical space.

[0080] A median-based threshold (2σ–3σ) was used to further exclude outliers caused by human-induced probe pressure changes while retaining the true tissue gradient boundary.

[0081] 2) Assign weight λ to the interference reduction data points i =exp(-d0 / (pr)), where d0 is the Euclidean distance between the interference-reduced data point and the center point of the spherical space, and p=0.4~0.6.

[0082] The center point refers to the geometric center of all data points in the spherical space [X*, Y*, X*], and the Euclidean distance d0 is the distance between any data point [X n , Y m , Z k ] to the center point, the calculation formula is:

[0083]

[0084] 3) Calculate the weighted average value of each interference reduction data point I3=Σ(λ i ×I2) / Σλ i .

[0085] S4, the third three-dimensional space image data point set [X n , Y m ] corresponds to the mean elastic value of all data points in the depth coordinate direction I4 as the two-dimensional plane image data point [X n , Y m ] elastic value, and obtain the final two-dimensional space image data point set {(X n , Y m , I4)}, converted to the corresponding optimized ultrasound elastography.

[0086] In other embodiments, the present invention further provides an application of the above-mentioned optimization method in the classification and grading of diffuse lesion ultrasound elastography images, including:

[0087] Optimize the data set according to the optimization method described above;

[0088] The optimized two-dimensional space image data point set {(X n , Y m , I4)} is associated with the diffuse lesion pathology grade of the target object to construct a training dataset;

[0089] Based on the training data set, a pre-built classification and grading model is trained;

[0090] Based on the trained classification and grading model, the two-dimensional spatial image data point set of the new target object is analyzed, and the corresponding prediction results of the diffuse lesion pathological grade are output.

[0091] The process of constructing the training data set includes:

[0092] Get the optimized two-dimensional space image data point set {(X n , Y m , I4)} the average value M1 of the elasticity values ​​of all points; obtain the optimized two-dimensional space image data point set {(X n , Y m , I4)} the average elasticity value of column X1 and X n The ratio M2 of the average values ​​of the elasticity values ​​on the column is obtained as data (M1, M2);

[0093] The data (M1, M2) are associated with the diffuse lesion pathology grade of the target object to construct a training dataset.

[0094] In another embodiment, Figure 3As shown, the present invention also provides a diffuse lesion ultrasound elastic imaging image classification and grading system, including a data set construction module, a training module and a prediction module:

[0095] The preprocessing module is used to optimize the data set according to the above optimization method;

[0096] The dataset construction module is used to transform the optimized two-dimensional space image data point set {(X n , Y m , I4)} is associated with the diffuse lesion pathological grade of the target object to construct a training data set; the construction process of the training data set is shown in the previous article.

[0097] The training module is used to train the pre-built classification and grading model based on the training data set.

[0098] The prediction module is used to analyze the two-dimensional spatial image data point set of the new target object based on the trained classification and grading model, and output the corresponding prediction result of the diffuse lesion pathological grade.

[0099] The optimization method of the present invention is further described below with reference to a specific embodiment.

[0100] S1. 3D data construction:

[0101] Use the shear wave elastography function of clinical ultrasound equipment to obtain Z k The two-dimensional plane image data points of the shear wave elastography of the target object [X n , Y m ] and its elastic value I1, where k = 30, m = n = 256, the elastic value is Young's modulus (kPa), and the resolution of each two-dimensional plane image is 256 × 256. The data points of the Zk groups of two-dimensional plane images are stacked along the depth direction to form a three-dimensional space image data point set {(X n , Y m , Z k , I1)};

[0102] S2. Curve fitting and abnormal peak removal:

[0103] 1) Local weighted curve fitting:

[0104] For each [Y m , Z k ] corresponding to the horizontal coordinate direction data point (X n , I1) Perform quadratic polynomial fitting to obtain m*k fitting curves: Calculate the neighborhood width data points (smoothing parameter f = 0.2) and then define the weight function And based on the weight ω iPerform local fitting on the data points in the neighborhood to generate a smooth curve.

[0105] 2) Abnormal peak detection and correction:

[0106] Gradient mutation detection: When the absolute value of the gradient of five consecutive data points exceeds three times the mean value of the gradient of the previous and next ten points (parameters a = 5, b = 10, p = 3, used to detect gradient mutation), it is marked as a candidate abnormal peak.

[0107] Background intensity verification: Calculate the reference interval median M, extending from the candidate peak to the gradient plateau region. If the peak apex value exceeds 30% of M (parameter L = 30%, used to exclude operational interference peaks), it is determined to be operational pressure interference.

[0108] Correction processing: replace the abnormal peak area (such as Xn=90-95) with the cubic spline interpolation result of the reference interval to ensure that the total energy difference of the curve before and after correction is less than 5%.

[0109] S3, three-dimensional spatial filtering:

[0110] 1) Ellipsoid neighborhood construction: take the current point [X n , Y m , Z k ] as the center, an ellipsoid neighborhood with a major axis radius of 1.2r and a minor axis radius of 0.8r is constructed to adapt to the probe resolution (reference radius r = 15 data points).

[0111] 2) Anti-interference weighted average: Calculate the median M and standard deviation σ of the elasticity value in the neighborhood, retain the data points that meet |I2-M|≤2.5σ, and assign weights to the retained points: w i =exp(-d / (0.5×15)), where d is the Euclidean distance between the interference reduction data point and the center point of the spherical space (k=0.5, used to control the spatial attenuation rate), and the weighted average value I3=Σ(w i ×I2) / Σw i .

[0112] S4. Optimized 2D image generation for ultrasound elastography:

[0113] Depth average: along Z k Direction for each [X n , Y m ] position I3 values ​​are taken as arithmetic average to generate a two-dimensional elastic data point set {(X n , Y m , I4)}, converted to the corresponding optimized ultrasound elastography.

[0114] The optimization method of ultrasound elastography is applied to the classification and grading system of diffuse lesion ultrasound elastography, including:

[0115] A preprocessing module, configured to optimize the ultrasound elastography dataset according to the above-mentioned optimization method;

[0116] Dataset building module, used to obtain the optimized ultrasound elasticity data point set {(X n , Y m , I4)}, the average M1 of the elasticity values ​​of all points, and the average of the elasticity values ​​on the X1 column and X n The ratio M2 of the average values ​​of the elasticity values ​​on the column is obtained to obtain data (M1, M2), and the data (M1, M2) are associated with the diffuse lesion pathology grade of the target object to construct a training data set.

[0117] The training module uses the support vector machine (SVM) algorithm and (M1, M2) as input features to train the liver fibrosis staging (F0-F4) model.

[0118] The prediction module is used to first remove abnormal peaks and spatially filter the new ultrasound elasticity image through the preprocessing module to eliminate local elasticity value abnormalities caused by uneven probe pressure. The optimized elasticity map is then input into the liver fibrosis staging (F0-F4) model to distinguish different liver fibrosis stages (F0-F4) and reduce dependence on puncture biopsy.

[0119] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0120] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing ultrasonic elastography, characterized in that: The following steps are involved: S1. Get Z k The two-dimensional plane image data points of the shear wave elastography of the target object [X n , Y m ] and its elasticity value I1, stacking the two-dimensional plane image data points of group Zk to form a first three-dimensional space image data point set {(X n , Y m , Z k , I1)}, where X n 、Y m , Z k The horizontal coordinate, vertical coordinate, and depth coordinate of the midpoint of the three-dimensional space image data point set respectively; n represents the column index, m represents the row index, and k represents the depth index; S2, taking [Y m , Z k ] The corresponding horizontal coordinate data point X n Construct a data point set {(X n , I1)}, and perform curve fitting to obtain m*k fitting curves; remove the abnormal peaks on each fitting curve to obtain a noise reduction curve, and obtain the X on each noise reduction curve n The elastic value I2 of the corresponding data point is obtained, and the second three-dimensional space image data point set {(X n , Y m , Z k , I2)}; S3, taking [X n , Y m , Z k ] data point as the center, take the mean elasticity value I3 of all data points in the spherical space around it as [X n , Y m , Z k ] data point new elastic value, forming the third three-dimensional space image data point set {(X n , Y m , Z k , I3)}; S4, the third three-dimensional space image data point set [X n , Y m ] corresponds to the mean elastic value of all data points in the depth coordinate direction I4 as the two-dimensional plane image data point [X n , Y m ] elastic value, and obtain the final two-dimensional space image data point set {(X n , Y m , I4)}, converted to the corresponding optimized ultrasound elastography.

2. The method for optimizing ultrasonic elastography according to claim 1, wherein: In S1, the elastic value is Young's modulus or shear wave velocity; the values ​​of n, m, and k are 1 to 5000.

3. The method for optimizing ultrasonic elastography according to claim 1, wherein: In S2, the curve fitting process includes: For the data point set {(X n , I1)} for each data point X n , calculate the weight ω within its neighborhood i , based on the weight ω i , for X n Perform local quadratic polynomial fitting on the data points within the neighborhood to calculate the data point X n The predicted elastic value I1' is traversed through all data points in the horizontal axis direction to generate a complete fitting curve; in, σ=0.2d, f is the smoothing parameter, the value range of f is 0.1≤f≤0.3, N is the data point set {(X n ,I1)} in X n The total number of Indicates rounding down, X i For X n Other data points in the neighborhood.

4. The method for optimizing ultrasonic elastography according to claim 3, wherein: In S2, the steps of removing abnormal peaks on each fitting curve include: Traverse the elastic value gradient changes of adjacent data points on the fitting curve. When the absolute value of the gradient of a consecutive data point exceeds p times the average value of the gradient of the b data points before and after it, it is marked as a candidate abnormal peak. The reference interval is extended to both sides of the candidate abnormal peak to the area where the gradient returns to a stable state. The statistical median of the elasticity value in the reference interval is calculated. If the difference between the peak value of the candidate peak and the median exceeds L% of the median, it is confirmed as a background intensity abnormal peak. The elasticity values ​​in the abnormal peak area are replaced by the cubic spline interpolation results of the elasticity values ​​in the reference interval to correct the curve.

5. The method for optimizing ultrasonic elastography according to claim 4, characterized in that: In S2, a=3~10, b=5~20, p≥3; L=20~40.

6. The method for optimizing ultrasonic elastography according to claim 1, wherein: In S3, the spherical space is a spherical space with a radius r of 5 to 100 data points; or, the spherical space is an ellipsoidal spherical space with a major axis radius of 1.1r to 1.5r data points and a minor axis radius of 0.5r to 0.8r data points.

7. The method for optimizing ultrasonic elastography according to claim 6, characterized in that: The calculation process of the elasticity value I3 in S3 includes: Calculating the median M and standard deviation σ of the elasticity values ​​of all data points in the spherical space, and extracting the interference-reducing data points in the spherical space that meet the condition |I2-M|≤2σ~3σ; Assign weight λ to the interference reduction data point i =exp(-d0 / (pr)), where d0 is the Euclidean distance between the noise reduction data point and the center point of the spherical space, and p=0.4-0.6; Calculate the weighted average value of each interference reduction data point I3=Σ(λ i ×I2) / Σλ i .

8. The method for optimizing ultrasonic elastography according to claim 6, wherein: The direction of the major axis radius is parallel to the longitudinal coordinate direction.

9. An application of the optimization method according to any one of claims 1 to 8 in the classification and grading of diffuse lesion ultrasound elastography images, characterized in that: include: Optimizing the data set according to the optimization method according to any one of claims 1 to 8; The optimized two-dimensional space image data point set {(X n , Y m , I4)} is associated with the diffuse lesion pathology grade of the target object to construct a training dataset; Based on the training data set, a pre-built classification and grading model is trained; Based on the trained classification and grading model, the two-dimensional spatial image data point set of the new target object is analyzed, and the corresponding prediction results of the diffuse lesion pathological grade are output.

10. The use according to claim 9, characterized in that The process of constructing the training dataset includes: Get the optimized two-dimensional space image data point set {(X n , Y m , I4)} the average value M1 of the elasticity values ​​of all points; obtain the optimized two-dimensional space image data point set {(X n , Y m , I4)} the average elasticity value of column X1 and X n The ratio M2 of the average values ​​of the elasticity values ​​in the column is obtained to obtain the data (M1, M2); The data (M1, M2) are associated with the diffuse lesion pathology grade of the target object to construct a training dataset.

Images