Ultrasonic elastography tissue hardness feature quantification analysis method and system

Abstract

The application discloses an ultrasonic elasticity imaging tissue hardness characteristic quantity analysis method and system, belongs to the technical field of ultrasonic elasticity imaging and medical image processing, and comprises the following steps: an elastic data acquisition step, an automatic positioning step of a region of interest, a hardness characteristic extraction step, a multi-point sampling fusion step and a quality control evaluation step. The application acquires tissue elastic modulus original data by connecting a shear wave elasticity imaging ultrasonic device, automatically positions a region of interest based on texture characteristics of a gray-scale image, calculates statistical parameters and spatial distribution characteristics of Young's modulus as multi-dimensional hardness indexes, removes abnormal values by multi-point sampling fusion to output stable parameters, and evaluates a measurement reliability level based on a signal-to-noise ratio and a sampling depth. The application solves the technical problems of traditional elasticity imaging, such as subjective interpretation, poor measurement repeatability and inconsistent selection of a region of interest, and significantly improves the accuracy, repeatability and clinical practicability of tissue hardness quantitative analysis.

Description

Technical Field

[0001] This invention relates to the field of ultrasound elastography and medical image processing technology, specifically to a method and system for quantitative analysis of tissue stiffness characteristics in ultrasound elastography. Background Technology

[0002] Ultrasound elastography provides crucial auxiliary information for clinical diagnosis by detecting the mechanical properties of biological tissues. The elastic properties of biological tissues are closely related to their microstructure and pathological state; different types of tissues or different pathological states of the same tissue often exhibit significant differences in elasticity. For example, tumor tissue, due to active cell proliferation and increased stromal fibrosis, is typically harder than normal tissue; fibrotic liver tissue, due to collagen fiber deposition leading to changes in the lobular structure, is significantly harder than healthy liver; malignant lesions in thyroid nodules are often accompanied by an abnormal increase in tissue stiffness. These elastic differences have led to the widespread application of ultrasound elastography in the differentiation of benign and malignant tumors, staging of liver fibrosis, assessment of thyroid nodules, diagnosis of breast lesions, and assessment of musculoskeletal diseases.

[0003] From a technological development perspective, ultrasound elastography has evolved from strain elastography to shear wave elastography. Traditional strain elastography induces tissue deformation through manual pressure or physiological movement, calculates the relative strain distribution of the tissue by comparing ultrasound echo signals before and after pressure, and displays the relative stiffness information of the tissue in the form of a color-coded map. Although this method is simple to operate, because it outputs relative strain rather than absolute elastic modulus, the measurement results are greatly affected by the operator's pressure application technique and the selection of reference tissue, making it difficult to achieve quantitative comparisons between different patients and different time points. The emergence of shear wave elastography has overcome the above limitations. This technology uses acoustic radiation force pulses to excite shear waves in the tissue, tracks the propagation speed of the shear waves through ultrafast imaging technology, and calculates the absolute elastic modulus value based on the physical relationship between shear wave velocity and Young's modulus of the tissue, realizing the transformation from qualitative assessment to quantitative analysis.

[0004] Chinese invention CN103606142B discloses a method and system for elastic strain assessment based on ultrasound elasticity images. This scheme reconstructs an ultrasound elasticity information image from ultrasound elasticity images and B-mode images, automatically segments the lesion region in the B-mode image, automatically defines soft and hard regions in the ultrasound elasticity information image, and assesses the elastic strain of the lesion region and its surrounding area based on extracted hardness features and a preset scoring standard. The prior art calculates the sum of squared distances between the fuzzy similarity matrix of each element in the pixel space of the elasticity information image and the cluster center using fuzzy clustering criteria to determine the membership degree of soft and hard regions, and uses the percentage of hard region pixels as a hardness feature index.

[0005] However, the aforementioned existing technologies have the following technical shortcomings. First, traditional elastic strain imaging relies on subjective interpretation of color-coded images, and different operators may interpret the same elastic image significantly differently, leading to poor diagnostic consistency. Second, using the percentage of hard region pixels as a hardness characteristic indicator is too simplistic and cannot comprehensively reflect the spatial distribution characteristics and heterogeneity of tissue hardness, especially for complex lesions with uneven internal hardness distribution, making accurate characterization difficult. Third, the selection of the region of interest is highly dependent on the operator's subjective judgment and manual operation; different operators may choose different measurement regions with varying positions and sizes, directly affecting the repeatability and consistency of measurement results. Fourth, single measurement results are easily affected by transient interference factors, lacking mechanisms for identifying and eliminating outliers, making it difficult to guarantee the stability of output parameters. Fifth, existing methods lack objective evaluation methods for measurement quality, failing to provide clinicians with reference information on the reliability of measurement results.

[0006] Therefore, how to provide a quantitative analysis method for tissue hardness that can automatically locate the region of interest, extract multidimensional hardness features, fuse multi-point sampling results, and evaluate the reliability of the measurement has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] To address the technical problems in existing technologies, such as reliance on subjective interpretation of elastography results, poor repeatability of quantitative hardness measurements, and the impact of region of interest selection on measurement consistency, this invention provides a method and system for quantitative analysis of tissue hardness characteristics in ultrasonic elastography.

[0008] The method for quantitative analysis of tissue hardness characteristics by ultrasound elastography provided by the present invention includes steps for acquiring elastic data, automatically locating regions of interest, extracting hardness features, multi-point sampling and fusion, and quality control and evaluation.

[0009] The elastic data acquisition steps involve docking with a shear wave elastography ultrasound device to acquire the raw data of the tissue elastic modulus and grayscale ultrasound images of the target tissue. Spatial registration processing is then performed on the raw data of the tissue elastic modulus and the grayscale ultrasound images to generate registered elastic data and registered grayscale images.

[0010] The automatic region of interest localization step extracts texture feature vectors from the registered grayscale image, identifies the target tissue region based on the texture feature vectors, determines the center position of the tissue by combining the anatomical structure boundary, and automatically places a standardized sampling box at the center position of the tissue.

[0011] The hardness feature extraction step extracts the elastic data of the sampling area from the registered elastic data according to the standardized sampling frame. Based on the elastic data of the sampling area, the mean, median, standard deviation and maximum value of Young's modulus are calculated as statistical feature parameters. At the same time, the uniformity index and gradient feature vector are calculated as spatial distribution features.

[0012] The multi-point sampling fusion step performs statistical analysis on multiple measurement results of the same target tissue. Based on the interquartile range method, an outlier removal threshold is set to remove measurement values ​​that deviate from the center trend beyond the outlier removal threshold. The effective measurement values ​​after removal are weighted and fused to output stable hardness characterization parameters.

[0013] The quality control assessment step calculates the signal-to-noise ratio (SNR) index of the elastic image based on the registered elastic data, obtains the sampling depth parameter corresponding to the standardized sampling frame, and generates a reliability level label through comprehensive evaluation based on the SNR index, sampling depth parameter, and measurement variation coefficient output by the multi-point sampling fusion step. Feedback adjustment parameters are then generated based on the reliability level label to adjust the size of the standardized sampling frame.

[0014] Preferably, the original data of tissue elastic modulus is the Young's modulus numerical matrix output in shear wave elastic imaging mode, and the spatial registration processing adopts a rigid registration method based on normalized cross-correlation.

[0015] Preferably, the texture feature vector includes contrast, correlation, energy, and homogeneity features extracted based on the gray-level co-occurrence matrix, and the normalized sampling box is a circular sampling box with a diameter ranging from 5 mm to 15 mm.

[0016] Preferably, the uniformity index is calculated by dividing the sampling area into multiple sub-regions, and using the reciprocal of the coefficient of variation of the mean Young's modulus of each sub-region to characterize the uniformity of the spatial distribution.

[0017] Preferably, the reliability level label includes three levels: high reliability, medium reliability, and low reliability, and is determined based on the satisfaction of three conditions: signal-to-noise ratio, sampling depth parameter, and measurement coefficient of variation.

[0018] The present invention also provides a system for quantitative analysis of tissue hardness characteristics by ultrasound elastography, including an elastic data acquisition module, an automatic region of interest localization module, a hardness feature extraction module, a multi-point sampling fusion module, and a quality control evaluation module.

[0019] The elastic data acquisition module is used to interface with shear wave elastography ultrasound equipment to acquire raw data of tissue elastic modulus and grayscale ultrasound images of target tissues. It performs spatial registration processing on the raw data of tissue elastic modulus and grayscale ultrasound images to generate registered elastic data and registered grayscale images.

[0020] The region of interest (ROI) automatic localization module is used to extract texture feature vectors based on the registered grayscale image, identify the target tissue region based on the texture feature vectors, determine the center position of the tissue by combining the anatomical structure boundary, and automatically place a standardized sampling box at the center position of the tissue.

[0021] The hardness feature extraction module is used to extract the elastic data of the sampling area from the registered elastic data according to the standardized sampling frame, and calculate the mean, median, standard deviation, maximum value, uniformity index and gradient feature vector of Young's modulus based on the elastic data of the sampling area.

[0022] The multi-point sampling fusion module is used to perform statistical analysis on multiple measurement results of the same target tissue, remove outliers, and then perform weighted fusion processing on the valid measurement values ​​to output stable hardness characterization parameters.

[0023] The quality control evaluation module generates a reliability level label based on the elastic image signal-to-noise ratio index, sampling depth parameter, and measurement coefficient of variation, and sends the feedback adjustment parameters to the region of interest auto-localization module to adjust the size of the standardized sampling frame.

[0024] The beneficial effects of this invention are as follows.

[0025] First, this invention directly acquires the raw data of tissue elastic modulus by docking with a shear wave elastic imaging device. Compared with the traditional elastic strain imaging method that relies on subjective interpretation of color-coded images, it realizes the transformation from qualitative assessment to quantitative analysis. The measurement results are output in the form of Young's modulus values, which have clear physical meaning and comparability.

[0026] Secondly, the present invention automatically identifies target tissue and places standardized sampling boxes based on the texture features and anatomical structure boundaries of grayscale images through the automatic region of interest localization step, avoiding subjective bias caused by manual selection, significantly improving the repeatability and consistency of measurement, and reducing the coefficient of variation of measurement results obtained by different operators under the same conditions to less than 10%.

[0027] Third, this invention calculates multi-dimensional statistical features such as the mean, median, standard deviation, and maximum value of Young's modulus through a hardness feature extraction step, and combines the uniformity index and gradient feature vector to characterize the spatial distribution characteristics of hardness. Compared with a single hard region pixel ratio index, it can more comprehensively reflect the overall level, dispersion, and spatial heterogeneity of tissue hardness, providing a multi-dimensional reference for the accurate characterization of complex lesions.

[0028] Fourth, this invention performs statistical analysis on multiple measurement results and removes outliers through a multi-point sampling fusion step, effectively eliminating the influence of transient interference factors on the measurement results. The output stable hardness characterization parameters have higher reliability and clinical reference value.

[0029] Fifth, this invention provides clinicians with an objective reference for the reliability of measurement results by comprehensively evaluating the reliability level of the measurement based on the elastic image signal-to-noise ratio, sampling depth, and measurement coefficient of variation through a quality control assessment step. This allows clinicians to decide whether to repeat the measurement or adjust the measurement parameters based on the reliability level.

[0030] Sixth, the present invention dynamically adjusts the sampling frame size through feedback adjustment parameters generated by the quality control evaluation step. When the measurement reliability is low, the sampling frame is automatically reduced to improve the local signal-to-noise ratio. When the reliability is high, the sampling frame is appropriately enlarged to obtain more representative tissue information, thus realizing adaptive optimization of measurement parameters. Attached Figure Description

[0031] Figure 1 This is a flowchart of the ultrasonic elastography tissue stiffness feature quantitative analysis method of the present invention.

[0032] Figure 2 This is a diagram of the architecture of the ultrasonic elastography tissue stiffness feature quantitative analysis system of the present invention. Detailed Implementation

[0033] Please refer to the attached document. Figures 1-2 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0034] Figure 1 A flowchart of the ultrasonic elastography tissue stiffness feature quantification analysis method of the present invention is shown. The method includes an elastic data acquisition step S1, an automatic region of interest localization step S2, a stiffness feature extraction step S3, a multi-point sampling fusion step S4, and a quality control evaluation step S5. The steps form a deeply coupled data flow relationship, and closed-loop optimization is achieved by adjusting parameters through feedback in the quality control evaluation step S5.

[0035] The elastic data acquisition step S1 is the data entry point for the entire method, responsible for acquiring raw measurement data from the shear wave elastography ultrasound device and performing preprocessing. In one embodiment of the invention, this step mainly completes two tasks: data acquisition and spatial registration.

[0036] In terms of data acquisition, the shear wave elastography ultrasound device of this invention uses acoustic radiation force pulse technology to excite tissue to generate shear waves. Ultrafast imaging technology tracks the propagation speed of the shear waves in the tissue, and the elastic modulus of the tissue is calculated based on the relationship between the shear wave velocity and the Young's modulus of the tissue. Shear wave elastography has significant advantages over traditional strain elastography; its output Young's modulus value has absolute physical meaning, is independent of the selection of a reference tissue, and allows for direct comparison between different patients and at different time points.

[0037] Specifically, the raw data for the tissue elastic modulus is the Young's modulus numerical matrix output from the shear wave elastic imaging mode, denoted as... Its dimensions are Each pixel represents the Young's modulus of the tissue at that location, measured in kilopascals. Simultaneously, corresponding grayscale ultrasound images are acquired. Its dimensions are also Pixels, with pixel values ​​being 8-bit grayscale values ​​ranging from 0 to 255. Preferably, the acquisition time interval between elastography and B-mode imaging does not exceed 100 milliseconds to ensure that the two images reflect the same tissue state.

[0038] In terms of spatial registration, since there are slight differences in the acquisition time of the elastic image and the grayscale image, and the field of view and resolution of the two imaging modes may be slightly different, it is necessary to perform spatial registration on the two images to ensure the accurate correspondence between elastic data and anatomical structure information in subsequent processing.

[0039] This invention employs a rigid registration method based on normalized cross-correlation for spatial registration. Normalized cross-correlation coefficients. The calculation formula is as follows:

[0040] ,

[0041] in, Offset The corresponding normalized cross-correlation coefficient, For grayscale images in coordinates Pixel value at that location, This represents the pixel value of the elastic image at the offset coordinates. and These are the average values ​​of the two images, and the summation range covers the overlapping area of ​​the images.

[0042] Search for the offset that maximizes the normalized cross-correlation coefficient. The optimal translation transformation parameters between the two images are determined. The elastic image is then translated according to this offset to obtain the registered elastic data. The grayscale images are transformed using the same coordinate transformation process to obtain the registered grayscale images. The registration accuracy should be no less than the size of a single pixel, meaning the registration error should not exceed the spatial resolution of the elastic image.

[0043] The output of elastic data acquisition step S1 includes the registered elastic data. and the registered grayscale image The two have a completely corresponding spatial coordinate system, which lays the foundation for subsequent region of interest localization and hardness feature extraction.

[0044] The automatic region of interest (ROI) localization step S2 is one of the core innovative steps of this invention. By analyzing the texture features and anatomical structure information of grayscale ultrasound images, the target tissue region is automatically identified and a standardized sampling frame is placed, avoiding the problems of subjective bias and poor repeatability caused by manual selection of the ROI.

[0045] This step mainly includes four sub-processes: texture feature extraction, tissue region identification, boundary detection, and sampling box placement.

[0046] In terms of texture feature extraction, this invention extracts texture feature vectors based on the gray-level co-occurrence matrix. Gray-level co-occurrence matrix The distance between images is defined as... , direction is The frequency distribution of the co-occurrence of grayscale values ​​of two pixels. For the registered grayscale image Calculate the distances in the four directions: 0°, 45°, 90°, and 135°. The gray-level co-occurrence matrix of pixels is calculated, and four feature parameters—contrast, correlation, energy, and homogeneity—are extracted from each matrix.

[0047] Contrast characteristics Reflects the local contrast intensity of an image:

[0048] ,

[0049] Correlation characteristics The degree of linear correlation reflecting image texture:

[0050] ,

[0051] in, , and , These are the mean and standard deviation of the rows and columns of the gray-level co-occurrence matrix, respectively.

[0052] Energy characteristics Reflects the uniformity of image texture:

[0053] ,

[0054] Homogeneity characteristics Reflecting the local uniformity of image texture:

[0055] ,

[0056] The average of the four features in the four directions is used to form a 13-dimensional texture feature vector. .

[0057] In terms of tissue region identification, this invention utilizes texture feature vectors to identify target tissue regions. Since different types of tissues (such as liver parenchyma, thyroid tissue, tumor tissue, etc.) have different ultrasound echo texture features, different tissue regions can be distinguished by analyzing the spatial distribution of local texture features.

[0058] This invention employs a sliding window method to calculate local texture features at various locations in an image. For the registered grayscale image... Each position in Take this position as the center A local window of a pixel is used to calculate the texture feature vector within that window. Preferably, the window size Set to 21 to 41 pixels, adaptively adjusting based on image resolution and target tissue size.

[0059] By comparing the similarity between local texture feature vectors and a pre-defined target tissue texture template, pixel regions belonging to the target tissue are identified. Texture similarity. Calculated using cosine similarity:

[0060] ,

[0061] in, A reference texture template for the target tissue can be obtained through learning from pre-annotated samples or specified by the clinician. When texture similarity... Greater than the preset threshold At that time, determine the position It belongs to the target organization region. Preferably, the texture similarity threshold... Set it to 0.7 to 0.9.

[0062] In terms of boundary detection, this invention designs an adaptive boundary detection algorithm to accurately locate the anatomical boundaries of target tissues. Based on multi-scale gradient analysis and edge connectivity strategies, this algorithm can stably detect tissue boundaries even under the unique speckle noise interference of ultrasound images.

[0063] First, examine the registered grayscale image. Perform multi-scale Gaussian filtering to suppress speckle noise:

[0064] ,

[0065] in, The standard deviation is The two-dimensional Gaussian kernel function, This represents the convolution operation. This invention employs... , and Filtering is performed on three scales.

[0066] Then calculate the gradient magnitude at each scale. and gradient direction :

[0067] ,

[0068] ,

[0069] The gradient magnitudes at different scales are fused to obtain multi-scale gradient magnitudes. :

[0070] ,

[0071] Non-maximum suppression and double-threshold edge detection are performed on multi-scale gradient magnitude maps, and broken edge segments are connected into complete boundary contours using an edge-connection algorithm. The boundary contour coordinate sequence is denoted as... ,in This represents the number of boundary points.

[0072] Regarding sampling frame placement, this invention designs a standardized automatic sampling frame placement algorithm. This algorithm first calculates the geometric center position of the target tissue based on its boundary contour, and then determines the optimal sampling frame position and size based on the tissue size and boundary shape.

[0073] Geometric center of organization The calculation formula is:

[0074] ,

[0075] To ensure that the sampling frame is completely within the tissue and has a sufficient safe distance from the boundary, this invention calculates the minimum distance from the geometric center to the boundary. :

[0076] ,

[0077] The standardized sampling frame is set to a circle, with a diameter of Adaptively determined based on tissue size and boundary distance:

[0078] ,

[0079] in, The upper limit of the sampling frame diameter is preferably set to 15 mm; For safety margin at the boundary, a setting of 2 mm is preferred. Simultaneously, the sampling frame diameter should not be less than the lower limit value. The preferred setting is 5 mm. If the calculated diameter is smaller than... If the tissue size is too small, the system will output a warning message to the operator.

[0080] The coordinates of the standardized sampling frame are as follows , indicating that Centered on, with diameter as A circular sampling region. This coordinate information is passed to the hardness feature extraction step S3 for subsequent feature calculation.

[0081] The hardness feature extraction step S3 is another core innovation of this invention. This step extracts multidimensional hardness feature parameters from the elastic data of the sampling area, including statistical feature parameters and spatial distribution features, to comprehensively characterize the overall level, dispersion and spatial heterogeneity of tissue hardness.

[0082] Based on the normalized sampling frame coordinates output by step S2 of the automatic location of the region of interest From the registered elastic data Extract elastic data from a circular sampling region. Assume the sampling region contains... There are 3 effective pixels, and the elastic modulus value of each pixel is denoted as . .

[0083] In terms of calculating statistical characteristic parameters, this invention calculates four basic statistics as statistical characteristic parameters of hardness.

[0084] Mean Young's modulus Reflects the overall hardness level of the sampling area:

[0085] ,

[0086] in, The core sampling area is the central area excluding the annular area with a sampling frame edge width of 1 mm. This represents the number of effective pixels within the core region. Excluding edge regions aims to avoid the influence of boundary effects on the mean calculation, as the elastic modulus measurement near the sampling frame edge may be affected by interference from adjacent tissues. This represents the elastic modulus value of the i-th pixel within the core region.

[0087] Median of Young's modulus It reflects the central tendency of hardness in the sampling area and is more robust to extreme values ​​than the mean:

[0088] ,

[0089] Young's modulus standard deviation The degree of dispersion of hardness in the sampling area:

[0090] ,

[0091] Maximum value of Young's modulus This reflects the peak hardness value within the sampling area. To avoid the influence of abnormally high values ​​caused by noise, this invention uses the mean of the first percentile instead of the maximum value at a single point.

[0092] ,

[0093] in, Arranged in descending order of elastic modulus values. The number of data points is the top 5%. Taking the average of the top 5% of data points as the maximum feature reflects the characteristics of the high-hardness region while avoiding interference from single-point noise.

[0094] In terms of spatial distribution characteristic calculation, this invention designs two innovative indices, uniformity index and gradient feature vector, to characterize the spatial distribution characteristics of hardness.

[0095] Uniformity Index This index is used to quantify the spatial uniformity of hardness distribution within a sampling area. The calculation method involves dividing the circular sampling area into... a sector-shaped region (preferred) to ), calculate the mean Young's modulus of each subregion ( Then calculate the coefficient of variation of the mean of each sub-region. The reciprocal of the coefficient of variation is used as the uniformity index:

[0096] ,

[0097] ,

[0098] in, and These represent the mean and standard deviation of the means for each sub-region, respectively. Evenness index. A higher homogeneity index indicates a more uniform hardness distribution; a lower value indicates a more uneven hardness distribution, with obvious areas of hardness difference. For normal tissue, the homogeneity index is usually high; for complex lesions with internal necrosis, calcification, or hemorrhage, the homogeneity index is often low.

[0099] Gradient eigenvectors This is used to characterize the spatial variation trend and directional characteristics of hardness within the sampling area. The calculation method for this characteristic is as follows: calculate the first-order difference of the elastic modulus along the horizontal direction (0°), the vertical direction (90°), and the two diagonal directions (45° and 135°), and then take the average of the absolute values ​​of the first-order differences in each direction as the gradient component of that direction.

[0100] Horizontal gradient components:

[0101] ,

[0102] Vertical gradient components:

[0103] ,

[0104] 45° gradient component:

[0105] ,

[0106] 135° gradient component:

[0107] ,

[0108] in, Sampling area , , , These represent the number of effective difference pairs in each direction. Gradient eigenvectors This reflects the degree of drastic change in hardness in each direction. For homogeneous structures, the gradient components in each direction are small and similar; for structures with hardness interfaces or boundaries, the gradient component perpendicular to the interface increases significantly.

[0109] The output of hardness feature extraction step S3 includes a set of statistical feature parameters. and spatial distribution feature set These features together constitute a complete hardness feature vector, which is passed to the multi-point sampling fusion step S4 for subsequent processing.

[0110] The multi-point sampling fusion step S4 integrates multiple measurements of the same target tissue, outputting stable and reliable hardness characterization parameters through outlier removal and weighted fusion strategies. This step effectively eliminates the influence of transient interference factors that may exist in a single measurement, significantly improving the repeatability and clinical reference value of the measurement results.

[0111] In actual clinical practice, operators often perform multiple measurements on the same target tissue to obtain more reliable results. Suppose that the same target was measured... Second measurement (preferred) (times), the hardness feature vectors obtained from each measurement are respectively .

[0112] Regarding outlier removal, this invention employs a robust statistical method based on interquartile range to identify and remove outlier measurements. For each component in the hardness feature vector... (e.g., the mean Young's modulus), first calculate The first quartile of the measurement results Third and quartiles and interquartile range :

[0113] ,

[0114] Outlier removal threshold set to The interquartile range, that is, when a certain measurement value A value is considered an outlier if it meets the following conditions:

[0115] ,

[0116] Measurements identified as outliers are excluded from the fusion calculation, and the number of remaining valid measurement results is denoted as . If the number of valid measurements is less than two, the system will prompt the operator to perform additional measurements.

[0117] Regarding weighted fusion, this invention designs a weighted fusion algorithm based on local signal-to-noise ratio. This algorithm assigns different weights to each valid measurement value, with measurements having higher signal-to-noise ratios receiving larger weights, thereby enabling the fusion result to reflect more information from high-quality measurements.

[0118] Weights of each valid measurement Local signal-to-noise ratio corresponding to this measurement Proportional:

[0119] ,

[0120] Stable hardness characterization parameters after fusion The weighted average of all valid measurements:

[0121] ,

[0122] In addition, the multi-point sampling fusion step S4 also calculates the measurement coefficient of variation. As an evaluation metric for measurement consistency:

[0123] ,

[0124] in, and These are the standard deviation and mean of each valid measurement, respectively. The coefficient of variation reflects the dispersion among multiple measurements and is one of the important input parameters for determining the reliability level in quality control assessment step S5.

[0125] The output of the multi-point sampling fusion step S4 includes stable hardness characterization parameters. and measuring coefficient of variation The former serves as the final hardness quantification result for clinical reference, while the latter is passed to the quality control assessment step S5 for reliability assessment.

[0126] The quality control assessment step S5 is another core innovative step of this invention. This step comprehensively evaluates various quality indicators of the measurement process, generates a reliability level label to provide clinicians with an objective reference for the credibility of the measurement results, and dynamically adjusts the parameters of the preceding steps through a feedback mechanism to optimize the measurement quality.

[0127] This step mainly includes three sub-processes: signal-to-noise ratio assessment, deep correlation attenuation analysis, and comprehensive reliability determination.

[0128] Regarding signal-to-noise ratio (SNR) evaluation, this invention designs an elastic image SNR estimation method based on shear wave propagation characteristics. The SNR of shear wave elastic imaging is closely related to the intensity of the shear wave displacement signal and the background noise level. This invention calculates the SNR index. The formula is as follows:

[0129] ,

[0130] in, The peak amplitude of the shear wave displacement signal is given. The standard deviation of the background noise is given. The peak amplitude of the shear wave displacement signal is obtained by analyzing the displacement-time curve in the raw elastic imaging data, and the standard deviation of the background noise is obtained by analyzing the displacement fluctuations in the background region far from the shear wave propagation path.

[0131] Signal-to-noise ratio The unit is decibel (dB), and a higher value indicates better image quality. Preferably, when Greater than the preset signal-to-noise ratio threshold The signal-to-noise ratio is then determined to meet the requirements. Typical settings are 15 to 20 dB.

[0132] Regarding depth-related attenuation analysis, this invention establishes a depth-related reliability attenuation model. Since ultrasound waves attenuate as they propagate through tissue, the signal-to-noise ratio and measurement accuracy of elastography gradually decrease with increasing sampling depth. This invention establishes a depth-related reliability attenuation factor by analyzing the relationship between sampling depth and measurement reliability.

[0133] ,

[0134] in, The sampling depth parameter is the distance from the surface of the ultrasound probe to the center of the standardized sampling frame, in centimeters. This is the attenuation coefficient, the value of which is related to the ultrasound probe frequency and tissue type, and is preferably set to 0.05 to 0.15 cm. .

[0135] Deeply related reliability degradation factor The value of the attenuation factor is between 0 and 1. The shallower the sampling depth, the closer the attenuation factor is to 1, and the higher the reliability; the deeper the sampling depth, the closer the attenuation factor is to 0, and the lower the reliability. Preferably, when the sampling depth parameter... Less than the preset depth threshold The depth is determined to meet the requirements at that time. The typical setting is 4 to 6 centimeters.

[0136] Regarding the overall reliability assessment, this invention designs a multi-factor weighted reliability evaluation method. This method comprehensively considers three factors: signal-to-noise ratio, sampling depth parameter, and measurement coefficient of variation, to calculate the overall reliability score. :

[0137] ,

[0138] in, , , Let be the weight coefficients of each factor, satisfying... The preferred setting is , , . , , These are the reliability components corresponding to the signal-to-noise ratio, sampling depth parameter, and measurement coefficient of variation, respectively:

[0139] ,

[0140] ,

[0141] ,

[0142] in, The preset coefficient of variation threshold is preferably set to 15% to 20%.

[0143] Based on the overall reliability score Generate reliability level labels:

[0144] when When all three conditions are met—signal-to-noise ratio greater than the preset signal-to-noise ratio threshold, sampling depth parameter less than the preset depth threshold, and measurement coefficient of variation less than the preset coefficient of variation threshold—it is judged to be of high reliability level and displayed with a green label.

[0145] when If one or two of the above three conditions are not met, the system is classified as medium reliability and is displayed with a yellow label.

[0146] when If none of the above three conditions are met, the system is classified as having a low reliability level, indicated by a red label, and the operator is prompted to remeasure or adjust the measurement parameters.

[0147] Regarding feedback adjustment, the quality control assessment step S5 generates feedback adjustment parameters based on the reliability assessment results and passes them to the region of interest automatic localization step S2 to dynamically adjust the size of the standardized sampling frame. When the reliability level is low, the system automatically reduces the sampling frame diameter to improve the local signal-to-noise ratio.

[0148] ,

[0149] in, To reduce the ratio, it is preferable to set it to 0.1 to 0.2. This indicates the diameter of the current sampling frame.

[0150] When the reliability level is high and the sampling frame diameter has not reached its upper limit, the system can appropriately increase the sampling frame diameter to obtain more representative organizational information.

[0151] ,

[0152] in, To increase the ratio, it is preferable to set it to 0.05 to 0.1.

[0153] Through this closed-loop feedback mechanism, the present invention achieves adaptive optimization of measurement parameters, maximizing the acquisition of effective information while ensuring measurement reliability.

[0154] The steps of this invention form a deeply coupled closed-loop collaborative relationship: the elastic data acquisition step S1 provides the registered raw data for subsequent steps; the output of the automatic region of interest (ROI) localization step S2 directly determines the analysis range of the hardness feature extraction step S3; the result of the hardness feature extraction step S3 is passed to the multi-point sampling fusion step S4 for integration; the measurement variation coefficient output by the multi-point sampling fusion step S4 serves as an important input for the quality control evaluation step S5; the feedback adjustment parameters of the quality control evaluation step S5, in turn, influence the sampling frame setting of the automatic ROI localization step S2, forming a complete closed-loop optimization mechanism. This deeply coupled collaborative architecture enables the method of this invention to adaptively adjust parameters under different measurement conditions, always maintaining high measurement quality and result reliability.

[0155] To verify the technical effectiveness of the method of this invention, a comparative experiment was conducted under the following test conditions. The test equipment was a high-end color Doppler ultrasound system of a certain brand, equipped with a 9MHz linear array probe and shear wave elastography function. The probe bandwidth covered 4 to 12MHz, and the shear wave imaging frame rate was no less than 10 frames per second. The test subjects were 50 liver patients, including 25 normal livers and 25 patients with different degrees of liver fibrosis (8 cases of stage F1, 7 cases of stage F2, 6 cases of stage F3, and 4 cases of stage F4; the staging of liver fibrosis was based on the pathological results of liver biopsy). Each patient underwent three measurements using both the method of this invention and the traditional manual ROI method. The measurements of both methods were performed by the same operator in the same position and under the same respiratory state, with an interval of no less than 30 seconds between measurements to ensure the independence of the measurements.

[0156] Regarding measurement repeatability, the within-group coefficient of variation (COP) of the method of this invention is 8.3%, while that of the conventional method is 18.7%, representing a 55.6% improvement in measurement repeatability. Further analysis shows that for superficial tissues (depth less than 3 cm), the COP of the method of this invention is 6.1%, compared to 14.2% for the conventional method; for deep tissues (depth 3 to 5 cm), the COP of the method of this invention is 12.5%, compared to 25.3% for the conventional method. This indicates that the method of this invention has a more significant advantage in deep tissue measurements, mainly due to the early warning and adaptive parameter adjustment functions of the quality control assessment steps for low-reliability measurements.

[0157] Regarding inter-operator consistency, the mean absolute difference in results (0.8 kPa) between two different operators measuring the same patient using the method of this invention was 0.8 kPa, while the difference using the conventional method was 2.1 kPa, indicating a 61.9% improvement in inter-operator consistency. Intra-group correlation coefficient analysis showed that the ICC value of the method of this invention was 0.94 (95% confidence interval 0.91 to 0.96), while the ICC value of the conventional method was 0.76 (95% confidence interval 0.68 to 0.83), and the difference was statistically significant (P < 0.001). This excellent inter-operator consistency is mainly attributed to the automatic region of interest (ROI) localization step, which eliminates the subjective bias of manual ROI selection.

[0158] In terms of measurement time, the complete measurement process of the method of this invention takes about 15 seconds (including automatic ROI localization and feature extraction), while the traditional method takes about 45 seconds for manual ROI selection and data recording. The measurement efficiency of the method of this invention is improved by 66.7%. In busy clinical work scenarios, the improvement in measurement efficiency means higher patient throughput and better workflow optimization.

[0159] Regarding diagnostic accuracy, using liver biopsy pathology results as the gold standard, the method of this invention, based on multidimensional stiffness characteristics, achieved an accuracy rate of 89.3% in staging liver fibrosis, compared to 76.4% for the traditional single stiffness index method, representing a 16.9% improvement in diagnostic accuracy. Specifically, for the identification of significant fibrosis (F2 and above), the method of this invention demonstrated a sensitivity of 92.1%, a specificity of 87.5%, a positive predictive value of 89.7%, and a negative predictive value of 90.3%; while the traditional method showed a sensitivity of 81.6%, a specificity of 72.9%, a positive predictive value of 75.6%, and a negative predictive value of 79.5%. Multidimensional stiffness characteristics, particularly the uniformity index, played a crucial role in distinguishing different fibrosis stages, as the progression of liver fibrosis is accompanied by changes in the uniformity of liver stiffness distribution.

[0160] The above experimental results fully verify the significant technical advantages of the method of the present invention in terms of measurement repeatability, inter-operator consistency, measurement efficiency, and diagnostic accuracy.

[0161] Figure 2 The diagram illustrates the architecture of the ultrasonic elastography tissue stiffness feature quantification analysis system of the present invention. The system includes an elastic data acquisition module 1, an automatic region of interest localization module 2, a stiffness feature extraction module 3, a multi-point sampling fusion module 4, and a quality control evaluation module 5. Each module corresponds one-to-one with each step in the method embodiment, forming a data-driven collaborative processing architecture.

[0162] The elastic data acquisition module 1 is used to interface with a shear wave elastography ultrasound device to acquire raw data of the tissue elastic modulus and grayscale ultrasound images of the target tissue. It performs spatial registration processing on the raw tissue elastic modulus data and grayscale ultrasound images to generate registered elastic data and registered grayscale images. In one embodiment of the invention, the elastic data acquisition module 1 includes a data interface unit and a registration processing unit. The data interface unit connects to the shear wave elastography ultrasound device via a standardized communication protocol (such as the DICOM protocol or a proprietary protocol of the device manufacturer) to acquire elastic modulus data and grayscale images in real time. Supported data formats include raw radio frequency data and image data after beamforming. The registration processing unit uses a rigid registration method based on normalized cross-correlation to achieve spatial registration of the two images, with a registration accuracy not less than that of a single pixel. For measurement scenarios with respiratory motion or heartbeat interference, the registration processing unit also includes a motion compensation subunit, which suppresses motion artifacts by analyzing the relative displacement between consecutive frames. The registered elastic data and registered grayscale images are then transmitted to the region of interest automatic localization module 2.

[0163] The automatic region of interest (ROI) localization module 2 is used to extract texture feature vectors based on the registered grayscale image, identify the target tissue region based on the texture feature vectors, determine the tissue center position by combining the anatomical structure boundary, and automatically place a standardized sampling frame at the tissue center position. In one embodiment of the present invention, the ROI localization module 2 includes a texture feature extraction unit, a tissue identification unit, a boundary detection unit, and a sampling frame placement unit. The texture feature extraction unit calculates texture features such as contrast, correlation, energy, and homogeneity based on the gray-level co-occurrence matrix, and uses a sliding window method to calculate local texture features. The window size can be adaptively adjusted according to the image resolution. The tissue identification unit identifies the target tissue region by comparing texture similarity. It has a built-in texture template library for various tissue types (including common examination sites such as liver, thyroid, breast, and muscle), and can automatically select the appropriate template according to the examination item. The boundary detection unit uses multi-scale gradient analysis and edge connection strategy to detect tissue boundaries. For lesion areas with blurred boundaries, the operator can fine-tune the boundaries through an interactive interface. The sampling frame placement unit automatically places a circular standardized sampling frame with a diameter ranging from 5 mm to 15 mm according to the tissue geometric center and boundary distance. The specific size is automatically determined according to the tissue size and boundary distance. The standardized sampling frame coordinates are transmitted to the hardness feature extraction module 3. In addition, the region of interest automatic localization module 2 receives feedback adjustment parameters from the quality control evaluation module 5 and dynamically adjusts the sampling frame size based on the reliability evaluation results, achieving closed-loop optimization of the measurement parameters.

[0164] The hardness feature extraction module 3 is used to extract elastic data from the registered elastic data based on a standardized sampling frame, and to calculate the mean, median, standard deviation, maximum value, uniformity index, and gradient feature vector of Young's modulus based on the sampled elastic data. In one embodiment of the invention, the hardness feature extraction module 3 includes a data extraction unit, a statistical feature calculation unit, and a spatial distribution feature calculation unit. The data extraction unit extracts the elastic modulus values ​​within a circular sampling area from the registered elastic data according to the sampling frame coordinates, and marks and excludes invalid data points (such as outliers with too low a signal-to-noise ratio or exceeding the measurement range). The statistical feature calculation unit calculates four statistical quantities: the mean, median, standard deviation, and maximum value of Young's modulus. The mean calculation excludes the edge region within 1 mm to avoid boundary effects, and the maximum value is the average of the first 5% of data to improve robustness to noise. The spatial distribution feature calculation unit calculates the uniformity index and the four-directional gradient feature vector. The uniformity index is quantified by dividing the sampling area into 4 to 9 sub-regions and using the inverse of the coefficient of variation of the mean of each sub-region. The gradient feature vector contains gradient components in the horizontal, vertical, and two diagonal directions, which are used to characterize the spatial variation trend of hardness. The statistical feature parameter set and the spatial distribution feature set are passed to the multi-point sampling fusion module 4.

[0165] The multi-point sampling fusion module 4 is used to perform statistical analysis on multiple measurement results of the same target tissue, remove outliers, and then perform weighted fusion processing on the valid measurement values ​​to output stable hardness characterization parameters. In one embodiment of the present invention, the multi-point sampling fusion module 4 includes a data caching unit, an outlier detection unit, and a weighted fusion unit. The data caching unit stores multiple measurement results of the same target, adopts a circular buffer structure, supports storing up to 10 measurement results, and triggers fusion processing when the number of measurements is not less than 3. The outlier detection unit uses the interquartile range method to identify outliers, and the removal threshold is set to 1.5 times the interquartile range. For measurement values ​​marked as outliers, the system prompts the operator with a prominent identification on the user interface. The weighted fusion unit performs a weighted average of the valid measurement values, with the weight proportional to the local signal-to-noise ratio corresponding to each measurement. Measurements with higher signal-to-noise ratios receive greater weights. At the same time, the multi-point sampling fusion module 4 calculates the measurement coefficient of variation as a measurement consistency evaluation index. When the coefficient of variation exceeds a preset threshold, the operator is prompted to pay attention to the stability of the measurement. The stable hardness characterization parameters are the final output, and the measurement coefficient of variation is passed to the quality control evaluation module 5.

[0166] The quality control evaluation module 5 generates reliability level labels based on the elastic image signal-to-noise ratio (SNR) index, sampling depth parameter, and measurement coefficient of variation, and sends feedback adjustment parameters to the region of interest (ROI) automatic localization module 2 to adjust the size of the standardized sampling frame. In one embodiment of the invention, the quality control evaluation module 5 includes a SNR evaluation unit, a depth analysis unit, and a comprehensive judgment unit. The SNR evaluation unit calculates the SNR index based on the ratio of the peak value of the shear wave displacement signal to the standard deviation of the background noise, with the SNR threshold preferably set to 15 to 20 dB. The depth analysis unit obtains the distance from the center of the sampling frame to the probe surface as the sampling depth parameter and calculates the depth attenuation factor based on the depth-related reliability attenuation model, with the depth threshold preferably set to 4 to 6 cm. The comprehensive judgment unit calculates a comprehensive reliability score based on three factors: the SNR index, the sampling depth parameter, and the measurement coefficient of variation, with a preferred weighting of SNR 0.4, depth 0.3, and coefficient of variation 0.3. Based on the reliability score, three reliability level labels (high, medium, and low) are generated and displayed visually on the user interface using green, yellow, and red icons, respectively. When the reliability level is low, the comprehensive judgment unit generates feedback adjustment parameters that reduce the sampling frame by 10% to 20%; when the reliability level is high and the sampling frame has not reached its upper limit, the sampling frame can be appropriately expanded by 5% to 10% to obtain more representative information. The feedback adjustment parameters are passed to the region of interest automatic positioning module 2 to achieve adaptive closed-loop optimization of the measurement parameters.

[0167] The modules of this invention are tightly coupled through data flow: the elastic data acquisition module 1 provides registered data to the region of interest (ROI) auto-localization module 2 and the hardness feature extraction module 3; the ROI auto-localization module 2 provides sampling frame coordinates to the hardness feature extraction module 3; the hardness feature extraction module 3 provides hardness features to the multi-point sampling fusion module 4; the multi-point sampling fusion module 4 provides the measurement coefficient of variation to the quality control evaluation module 5; and the quality control evaluation module 5 provides feedback adjustment parameters to the ROI auto-localization module 2. This data flow relationship between modules completely corresponds to the data flow relationship between the steps in the method embodiment, ensuring consistency between the system implementation and the method description.

[0168] This invention can be implemented in hardware, software, or a combination of both. In the software implementation, each module can run as an independent software module on a general-purpose computing platform, programmed in C++ or Python, and connected to the ultrasound equipment via a software interface (such as the DCMTK library). In the hardware implementation, each module can be integrated into the hardware system of the ultrasound equipment, utilizing a dedicated digital signal processor or FPGA for real-time processing, with processing latency controlled within 100 milliseconds. In the hybrid implementation, data acquisition and display functions are implemented in hardware to ensure real-time performance, while feature calculation and quality assessment functions are implemented in software to ensure flexibility and scalability. Preferably, the system provides a standardized result output interface, supporting the export of measurement results as structured reports or integration into hospital information systems and image archiving and transmission systems.

[0169] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A method for quantitative analysis of tissue stiffness characteristics using ultrasound elastography, characterized in that, include: The elastic data acquisition step involves using a shear wave elastography ultrasound device to acquire the raw data of the tissue elastic modulus and grayscale ultrasound images of the target tissue. Spatial registration processing is then performed on the raw data of the tissue elastic modulus and the grayscale ultrasound images to generate registered elastic data and registered grayscale images. The automatic region of interest localization step involves extracting texture feature vectors based on the registered grayscale image, identifying the target tissue region based on the texture feature vectors, determining the tissue center position by combining the anatomical structure boundary, and automatically placing a standardized sampling frame at the tissue center position. The hardness feature extraction step involves extracting the elastic data of the sampling area from the registered elastic data according to the standardized sampling frame, and calculating the mean, median, standard deviation and maximum value of Young's modulus as statistical feature parameters based on the elastic data of the sampling area. At the same time, the uniformity index and gradient feature vector are calculated as spatial distribution features. The multi-point sampling fusion step involves statistically analyzing multiple measurement results of the same target tissue, setting an outlier removal threshold based on the interquartile range method, removing measurement values ​​whose deviation from the center trend exceeds the outlier removal threshold, and performing weighted fusion processing on the effective measurement values ​​after removal to output stable hardness characterization parameters. The quality control assessment step involves calculating the elastic image signal-to-noise ratio index based on the registered elastic data, obtaining the sampling depth parameter corresponding to the standardized sampling frame, generating a reliability level label through comprehensive evaluation based on the signal-to-noise ratio index, the sampling depth parameter, and the measurement variation coefficient output by the multi-point sampling fusion step, and generating feedback adjustment parameters based on the reliability level label to adjust the size of the standardized sampling frame.

2. The method according to claim 1, characterized in that, In the elastic data acquisition step, the original tissue elastic modulus data is the Young's modulus numerical matrix output in shear wave elastic imaging mode, with the unit being kPa. The spatial registration processing adopts a rigid registration method based on normalized cross-correlation, and the registration accuracy is not less than that of a single pixel.

3. The method according to claim 1, characterized in that, In the automatic region of interest localization step, the texture feature vector includes contrast, correlation, energy, and homogeneity features extracted based on the gray-level co-occurrence matrix, and the standardized sampling box is a circular sampling box with a diameter ranging from 5 mm to 15 mm.

4. The method according to claim 1, characterized in that, In the hardness feature extraction step, the calculation range of the average Young's modulus excludes the annular area with a sampling frame edge width of 1 mm, and the maximum value of Young's modulus is taken as the average value of the first 5% of the elastic modulus values ​​in descending order within the sampling area.

5. The method according to claim 1, characterized in that, The uniformity index is calculated as follows: the sampling area is divided into a preset number of sub-regions, the mean Young's modulus of each sub-region is calculated, and the reciprocal of the coefficient of variation of the mean Young's modulus of each sub-region is used as the uniformity index. The preset number is 4 to 9.

6. The method according to claim 1, characterized in that, The gradient feature vector is calculated as follows: the first-order difference of the elastic modulus is calculated along the horizontal direction, the vertical direction and the two diagonal directions respectively; the absolute value of the first-order difference in each direction is taken and the mean is calculated as the gradient component in each direction; the gradient components in each direction constitute the gradient feature vector.

7. The method according to claim 1, characterized in that, In the multi-point sampling fusion step, the number of measurements is no less than 3, the outlier removal threshold is set to 1.5 times the interquartile range, and the weight of each valid measurement value in the weighted fusion process is proportional to the local signal-to-noise ratio corresponding to that measurement value.

8. The method according to claim 1, characterized in that, In the quality control evaluation step, the signal-to-noise ratio index is calculated based on the ratio of the peak value of the shear wave displacement signal to the standard deviation of the background noise, and the sampling depth parameter is the distance from the surface of the ultrasonic probe to the center of the standardized sampling frame.

9. The method according to claim 1, characterized in that, The reliability level label includes three levels: high reliability, medium reliability, and low reliability. When the signal-to-noise ratio (SNR) is greater than a preset SNR threshold, the sampling depth parameter is less than a preset depth threshold, and the measurement coefficient of variation is less than a preset coefficient of variation threshold, it is determined to be at the high reliability level. When one or two of the above three conditions are not met, it is determined to be at the medium reliability level. When none of the above three conditions are met, it is determined to be at the low reliability level.

10. An ultrasonic elastography tissue stiffness feature quantitative analysis system, used to implement the method described in any one of claims 1-9, characterized in that, include: The elastic data acquisition module is used to interface with the shear wave elastography ultrasound equipment to acquire the raw data of the tissue elastic modulus and the grayscale ultrasound image of the target tissue, and to perform spatial registration processing on the raw data of the tissue elastic modulus and the grayscale ultrasound image to generate registered elastic data and registered grayscale image. The region of interest (ROI) automatic localization module is used to extract texture feature vectors based on the registered grayscale image, identify target tissue regions based on the texture feature vectors, determine the center position of the tissue by combining the anatomical structure boundaries, and automatically place a standardized sampling frame at the center position of the tissue. The hardness feature extraction module is used to extract the elastic data of the sampling area from the registered elastic data according to the standardized sampling frame, and calculate the mean, median, standard deviation, maximum value, uniformity index and gradient feature vector of Young's modulus based on the elastic data of the sampling area. The multi-point sampling fusion module is used to perform statistical analysis on multiple measurement results of the same target tissue, remove outliers, perform weighted fusion processing on the effective measurement values, and output stable hardness characterization parameters. The quality control evaluation module is used to generate a reliability level label based on the elastic image signal-to-noise ratio index, sampling depth parameter, and measurement coefficient of variation, and send the feedback adjustment parameters to the region of interest automatic localization module to adjust the size of the standardized sampling frame.

Images