Intelligent quantification analysis method and system for myocardial texture based on cmr images

Abstract

The application relates to the technical field of medical image analysis, and discloses a myocardial texture intelligent quantitative analysis method and system based on a CMR image. The method comprises the following steps: acquiring T1 Mapping original images of three layers of a short-axis direction base part, a middle part of a ventricle and an apical part of a heart; performing mask processing on the images to extract a myocardial region and acquire T1 physical values of the myocardial region of each layer; identifying a reference point based on geometric concave feature points, calculating a rotation angle, aligning the three layer images, and segmenting a plurality of divided regions according to a preset standard; independently constructing a gray level co-occurrence matrix for each divided region, and extracting quantitative indexes such as texture entropy and texture energy; calculating a plurality of heterogeneity indexes between layers, between ventricular walls and through walls based on the quantitative indexes of the regions, and performing multidimensional analysis to obtain quantitative evaluation results of different dimensions. The application can quantify the heterogeneity and spatial distribution mode of myocardial tissue from a microscopic level without injecting exogenous contrast agents.

Description

Technical Field

[0001] This application relates to the field of medical image analysis technology, and in particular to a method and system for intelligent quantitative analysis of myocardial texture based on CMR images. Background Technology

[0002] In the clinical imaging diagnosis of cardiovascular diseases, cardiovascular magnetic resonance (CMR) is widely used for qualitative and quantitative analysis of myocardial tissue. Traditional methods for assessing myocardial lesions, such as late gadolinium enhancement (LGE), rely on the injection of exogenous gadolinium-containing contrast agents. This method is not only time-consuming and costly, but also has absolute or relative contraindications for patients with severe renal insufficiency, pregnant women, and some pediatric patients, making it difficult to popularize in special populations. To overcome this limitation, contrast agent-free T1 mapping imaging technology is gradually gaining attention.

[0003] However, conventional T1 mapping image analysis still has significant limitations in clinical applications: On the one hand, most existing methods only focus on the simple average of myocardial T1 values. This macroscopic mean can easily mask the heterogeneity of local microstructures, making it impossible to identify diffuse myocardial fibrosis in the very early stages of diseases such as dilated cardiomyopathy, when small lesions have not yet significantly increased the overall mean. On the other hand, existing analytical methods lack in-depth exploration of the anatomical spatial topology of the heart, failing to systematically combine the long-axis spatial distribution of the myocardium from the base to the apex, the asymmetric characteristics of the interventricular septum and lateral walls, and the transmural heterogeneity from the endocardium to the epicardium. Therefore, it is impossible to accurately differentiate complex pathological patterns such as ischemic cardiomyopathy, asymmetric hypertrophic cardiomyopathy, or diffuse damage in the absence of contrast agent focal imaging.

[0004] Therefore, there is an urgent need for an intelligent quantitative analysis method that can deeply mine the microscopic texture features and spatial distribution of myocardium without the need for contrast agents. Accordingly, this invention provides an intelligent quantitative analysis method and system for myocardial texture based on CMR images. Summary of the Invention

[0005] This application provides a method and system for intelligent quantitative analysis of myocardial texture based on CMR images, which can be used to achieve early and accurate identification and classification of myocardial lesions such as diffuse myocardial fibrosis and ischemic injury.

[0006] In a first aspect, this application provides a method for intelligent quantitative analysis of myocardial texture based on CMR images, the method comprising:

[0007] Step S1: Obtain T1 Mapping raw images of the heart in three layers along the short axis, including the basal layer, the mid-ventricular layer, and the apical layer.

[0008] Step S2: Perform masking on the original images of the three layers respectively, extract the closed-loop myocardial region, and obtain the T1 physical value of the myocardial region of each layer.

[0009] Step S3: Identify the reference point based on the myocardial region at the base level, calculate the rotation angle based on the reference point and the centroid of the internal region, and rotate the myocardial regions at the base level, the middle ventricular level, and the apical level based on the rotation angle to obtain aligned images of the myocardial regions at the three levels. Divide the myocardial regions at the three levels into multiple subdivision regions according to a preset standard.

[0010] Step S4: For each segmented region, construct the gray-level co-occurrence matrix independently, and extract quantitative indicators representing organizational characteristics based on the gray-level co-occurrence matrix;

[0011] Step S5: Calculate multiple heterogeneity indicators based on the quantitative indicators of each region, and perform multi-dimensional calculations based on the quantitative indicators and multiple heterogeneity indicators of each region to obtain quantitative evaluation results in different dimensions.

[0012] Secondly, this application provides an intelligent quantitative analysis system for myocardial texture based on CMR images, the system comprising:

[0013] The image acquisition module is used to acquire raw T1 mapping images of the heart in three layers along the short axis: the basal layer, the mid-ventricular layer, and the apical layer.

[0014] The myocardial extraction module is used to perform masking processing on the original images of the three layers respectively, extract the closed-loop myocardial region, and obtain the T1 physical value of the myocardial region of each layer.

[0015] The alignment and segmentation module is used to identify the reference point based on the myocardial region at the base level, calculate the rotation angle based on the reference point and the centroid of the internal region, and rotate the myocardial region at the base level, the middle ventricular level and the apical level based on the rotation angle to obtain the aligned myocardial region images at the three levels. The myocardial region at the three levels is segmented according to the preset standard to obtain multiple division regions.

[0016] The indicator extraction module is used to independently construct a gray-level co-occurrence matrix for each segmented region, and extract quantitative indicators representing organizational characteristics based on the gray-level co-occurrence matrix.

[0017] The multidimensional evaluation module is used to calculate multiple heterogeneity indicators based on the quantitative indicators of each region, and to perform multidimensional calculations based on the quantitative indicators and multiple heterogeneity indicators of each region to obtain quantitative evaluation results in different dimensions.

[0018] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0019] The technical solution provided in this application directly acquires non-invasive T1 mapping raw images of three layers: the base of the short axis of the heart, the mid-ventricle, and the apex. This avoids the dependence of traditional imaging techniques on exogenous gadolinium-containing contrast agents, solving the problem of inaccurate screening for cardiomyopathy in special populations such as patients with severe renal insufficiency, pregnant women, and infants. It not only reduces examination costs and shortens imaging time but also ensures the safety of the detection. Based on the curvature extremum method, it identifies geometric concave feature points and automatically completes cross-layer anatomical position rotation alignment and standard segment division, ensuring spatial registration across individuals and scanning devices under the same anatomical benchmark. This allows for the in-depth extraction of texture entropy and texture energy, reflecting the microscopic disorder and uniformity of tissues. This overcomes the shortcomings of traditional methods that rely solely on T1 mean values, which easily mask subtle local lesions, revealing early signs that are not visible to the naked eye. The system transforms diffuse myocardial fibrosis into calculable digital indicators. By deeply integrating the first heterogeneity indicator characterizing the long axis deviation of the heart, the second heterogeneity indicator characterizing the asymmetry between the interventricular septum and lateral walls, and the third and fourth heterogeneity indicators characterizing the transmural differences between the intima, middle and adventitia, it can keenly capture the subtle distortions and spatial covariance of microstructures when there are no macroscopic changes in cardiac morphology and the mean T1 value is still in the normal pseudomorphic period. Under the premise of no contrast agent assistance, it can automatically generate specific markers such as diffuse myocardial fibrosis early warning, topological overlap identification, global homogeneity, and transmural ischemic injury, providing a reliable intelligent assessment system for the early detection, accurate identification, and personalized clinical intervention of various complex myocardial disease patterns. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of an embodiment of an intelligent quantitative analysis method for myocardial texture based on CMR images in this application.

[0022] Figure 2 This is a T1 mapping image of the base layer;

[0023] Figure 3T1 mapping image of the mid-ventricular plane;

[0024] Figure 4 T1 mapping image at the apex of the heart;

[0025] Figure 5 This is a schematic diagram of one embodiment of an intelligent quantitative analysis system for myocardial texture based on CMR images in this application. Detailed Implementation

[0026] This application provides a method and system for intelligent quantitative analysis of myocardial texture based on CMR images. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent quantitative analysis method for myocardial texture based on CMR images in this application includes:

[0028] Step S1: Obtain T1 Mapping raw images of the heart in three layers along the short axis, including the basal layer, the mid-ventricular layer, and the apical layer.

[0029] Specifically, clinical diagnosis of cardiomyopathy (such as LGE technique) heavily relies on the injection of gadolinium-containing contrast agents, which is time-consuming and costly. Furthermore, contrast agents are unsuitable for patients with severe renal insufficiency, pregnant women, and some pediatric patients. To overcome the dependence on exogenous contrast agents, non-invasive T1 mapping technology can be used to address the challenge of qualitative and quantitative myocardial histology analysis in specific populations. To solve these problems, raw T1 mapping images of the subject's heart along the short axis are obtained. These raw images contain two-dimensional parametric maps covering three independent layers of the left ventricular anatomy: the basal layer, the mid-ventricular layer, and the apical layer. The basal layer is located at the level of the mitral valve leaflets, where the left ventricular chamber appears as a large ring, surrounded by the right ventricular free wall attached anteriorly and posteriorly to the left ventricular septum, forming the anatomical depressions of the anterior and posterior interventricular grooves. Figure 2 The image shown is a T1 mapping image at the basal level. The mid-ventricular level is located at the level of the papillary muscles. The left ventricular annulus is slightly smaller than at the basal level, but the anatomical structures of the right ventricular free wall and the anterior and posterior interventricular grooves are still visible. Figure 3 The image shown is a T1-mapped image of the mid-ventricular plane. In standard three-layer T1-mapped acquisition, the right ventricular free wall attachment is usually still visible at the apical level, but the geometric indentation of the interventricular groove may be shallower than at the basal and mid-ventricular levels. Therefore, its detection reliability is not as good as at the basal and mid-ventricular levels. Figure 4 The image shown is a T1 mapping image at the apex of the heart. The value of each pixel in the original image represents the absolute physical quantity of the longitudinal relaxation time of the microscopic myocardial tissue at that spatial location, in milliseconds.

[0030] Step S2: Perform masking on the original images of the three layers respectively, extract the closed-loop myocardial region, and obtain the T1 physical value of the myocardial region of each layer.

[0031] Specifically, to extract the myocardial region, masking processing is performed on the original images of the three layers mentioned above. Masking relies on the coordinates of the closed contour of the left ventricular endocardium and epicardium. These coordinates can be pre-obtained using automated image segmentation algorithms or manual delineation. Automated image segmentation algorithms refer to fully automatic segmentation of the left ventricular myocardium based on a deep learning semantic segmentation network, outputting a sequence of endocardial and epicardial coordinate points. Manual delineation involves experienced radiologists using specialized medical image processing software to manually delineate the endocardial and epicardial boundaries, generating a closed contour. The contour coordinates are obtained through these methods, based on the endocardial... The contour coordinates of the endocardium and epicardium are used to construct a zero-matrix of the same size as the original image, where 0 represents black (background). The annular region between the endocardium and epicardium contours is defined as the myocardial region. All pixels within this annular region are assigned a value of 1, generating a binary mask containing only 0s and 1s. Multiplying the original T1 physical values ​​pixel-by-pixel with the binary mask sets the original T1 values ​​outside the myocardial region to 0, while retaining the original T1 values ​​within the annular myocardial region, thus obtaining the myocardial region. A basic threshold filter is then applied to the pixels within this myocardial region to remove extreme free pixels belonging to surrounding adipose tissue or background gas, yielding the T1 physical values ​​for each layer of the myocardium. In the mapping image, the T1 value of normal myocardium is usually between 900-1100 ms at a 1.5T field strength and between 1050-1300 ms at a 3.0T field strength. The T1 value of adipose tissue is usually below 300 ms, while the signal value of background gas is close to zero. Based on the above physical characteristics, a basic effective range is set, and pixels that do not obviously belong to myocardial tissue are removed to obtain the initial physical T1 values ​​of myocardial regions at each level.

[0032] Step S3: Identify the reference point based on the myocardial region at the base level, calculate the rotation angle based on the reference point and the centroid of the internal region, and rotate the myocardial regions at the base level, the middle ventricular level, and the apical level based on the rotation angle to obtain aligned images of the myocardial regions at the three levels. Divide the myocardial regions at the three levels into multiple subdivision regions according to a preset standard.

[0033] Furthermore, the reference point is identified based on the myocardial region at the basal level, specifically including the following steps:

[0034] The curvature of each pixel is calculated along the outer contour of the myocardium. Local minimum points where the curvature is concave inward are obtained as geometric concavity feature points. Anterior interventricular groove feature points are identified from the geometric concavity feature points and defined as reference points.

[0035] Furthermore, the rotation angle is calculated based on the reference point and the centroid of the internal region, specifically including the following steps:

[0036] The centroid of the inner region surrounding the myocardial region is calculated as the centroid of the inner region. The line connecting the reference point and the centroid of the inner region is used as the reference axis. Based on the preset standard axis, the angle difference between the reference axis and the standard axis is obtained. The angle difference is used as the rotation angle. Based on the rotation angle, the myocardial region images of the three layers are rotated and aligned in the direction of the standard axis.

[0037] Specifically, in order to ensure that the same anatomical location is obtained with the same segmentation number among different patients, so that all subsequent texture features across populations and time locations are more meaningful, the myocardial region needs to be segmented according to the standard to achieve anatomical location registration across individuals and scanning devices. To achieve the above requirements, for the basal and mid-ventricular levels, the geometric concave feature points where the left ventricular epicardial edge intersects with the right ventricular free wall are automatically searched on the myocardial outer contour. First, an ordered pixel sequence of the myocardial outer contour is extracted. For each pixel on the contour, the local curvature is calculated using the curvature extremum method. In normal cardiac anatomy, two significant concave points are usually found at the basal and mid-ventricular levels. Therefore, the two points with the largest inward concavity are selected as geometric concave feature points. Typically, one geometric concave feature point is located on the right anterior side, corresponding to the anterior interventricular groove feature point, and the other is located on the right posterior side, corresponding to the posterior interventricular groove feature point. In order to identify the anterior interventricular groove feature point from the geometric concave feature points, a judgment rule based on anatomical prior knowledge is introduced. The concave point that is closest to the azimuth angle of the right ventricular centroid and is located within its preset azimuth interval is identified as the anterior interventricular groove feature point.

[0038] Further, identifying the anterior interventricular groove feature points from the geometric concave feature points specifically includes the following steps: determining the region where the right ventricular free wall is located based on the outer contour of the myocardium, calculating the geometric centroid of the right ventricular region as the right ventricular centroid, establishing a polar coordinate system with the geometric centroid of the left ventricular chamber as the pole, calculating the azimuth angle of the right ventricular centroid in the polar coordinate system as the azimuth angle of the right ventricular centroid; calculating the azimuth angle of each geometric concave feature point in the polar coordinate system, identifying the geometric concave feature point whose azimuth angle is closest to the azimuth angle of the right ventricular centroid and whose difference is within a preset angle range as the anterior interventricular groove feature point, and identifying another geometric concave feature point that is geometrically roughly opposite to the anterior interventricular groove feature point as the posterior interventricular groove feature point.

[0039] After identifying the anterior interventricular groove feature points from the geometric concave feature points, the anterior interventricular groove feature points are defined as reference points. The rotation angle is calculated based on the reference points and the centroid of the internal region. The calculation of the rotation angle based on the reference points and the centroid of the internal region specifically includes the following steps:

[0040] The centroid of the inner region surrounding the myocardial region is calculated as the centroid of the inner region. The line connecting the reference point and the centroid of the inner region is used as the reference axis. Based on the preset standard axis, the angle difference between the reference axis and the standard axis is obtained. The angle difference is used as the rotation angle. Based on the rotation angle, the myocardial region images of the three layers are rotated and aligned in the direction of the standard axis.

[0041] To establish a unified coordinate system based on anatomical landmarks across three layers of myocardial region images, ensuring that regardless of the subject's tilt in the scanner, the same anatomical stage corresponds to the same orientation in image space, and guaranteeing multi-anatomical consistency for subsequent texture feature extraction and cross-individual comparisons, since anterior interventricular groove feature points can be identified at both the basal and mid-ventricular levels, the centroid of the inner region surrounding the myocardial region is calculated based on the basal level. The centroid of the inner region is defined as the average of all pixel coordinates within the area enclosed by the endocardial contour. The line connecting the reference point and the centroid of the inner region is then drawn. As the reference axis, the preset standard axis is usually the 12 o'clock direction, which is the vertical upward direction of the image coordinate system. For the base layer, the angle between the reference axis and the standard axis is calculated and used as the rotation angle of that layer. Since the human heart remains stationary during a single magnetic resonance scan, the overall orientation of the left ventricle should not change from the base to the middle of the ventricle. Therefore, the three layers can be rotated and aligned according to the calculated rotation angle. This method uses the calculated angle difference (the angle between the reference axis and the standard axis) as the rotation angle and rotates and aligns the myocardial region images of the three layers towards the direction of the standard axis based on the rotation angle.

[0042] It is important to note that this method calculates the rotation angle only based on the basal level. Since a reference point can also be found in the myocardial region at the mid-ventricular level, the same method can be used to calculate the rotation angle based on the mid-ventricular level. Because the human heart remains stationary during a single MRI scan, the overall orientation of the left ventricle should not change from the basal level to the mid-ventricular level. Therefore, the rotation angle calculated based on the mid-ventricular level and the rotation angle calculated based on the basal level will be highly similar. Thus, this method calculates the rotation angle only based on the basal level image and applies the calculated rotation angle to the mid-ventricular and apical levels. However, the interventricular groove at the apical level is shallower than at the basal and mid-ventricular levels, making its detection reliability lower than at the basal and mid-ventricular levels. Therefore, the rotation angle calculated at the apical level is used.

[0043] Furthermore, the three layers of myocardial regions are divided according to preset standards to obtain multiple subdivision regions, including:

[0044] For the basal and mid-ventricular levels, the heart is divided into six local myocardial regions by dividing the region into 60-degree units, with the centroid of the internal region as the center and the standard axis as the starting point. For the apical level, the heart is divided into four local myocardial regions by dividing the region into 90-degree units, with the centroid of the internal region as the center and the standard axis as the starting point.

[0045] Specifically, differential segmentation was performed on the aligned myocardial region image based on the 16-segment topology definition recommended by the American Heart Association (AHA) standards. In the rotated and aligned image coordinate system, the standard axis corresponds to the 0-degree starting direction, and the angle increases clockwise. The specific segmentation steps are as follows: For the basal level and the mid-ventricular direction, with the centroid of the internal region as the center and the standard axis as the starting point, a sector was divided every 60 degrees clockwise, resulting in 6 local myocardial regions. These correspond to the AHA standards for the anterior basal wall, anterior basal septum, subbasal septum, inferior basal wall, inferior basal lateral wall, and anterior basal lateral wall, respectively, and are numbered one, two, three, four, five, and six. For example, the anterior basal wall region is the first segmented region. For the mid-ventricular level, again with the centroid of the internal region as the center, a sector was divided every 60 degrees, resulting in 6 local myocardial regions: mid-anterior wall, mid-anterior septum, mid-inferior septum, mid-inferior wall, mid-inferior lateral wall, and mid-anterior lateral wall. At the apical level, the area is divided into sectors every 90 degrees, centered on the centroid of the internal region. These sectors correspond to the anterior apical wall, apical septum, inferior apical wall, and lateral apical wall according to the AHA standard. Each segmented local myocardial region (i.e., each divided region) corresponds to a specific anatomical segment in the AHA standard.

[0046] Step S4: For each segmented region, construct the gray-level co-occurrence matrix independently, and extract quantitative indicators representing organizational characteristics based on the gray-level co-occurrence matrix;

[0047] Specifically, routine clinical assessments typically focus only on the average T1 value of the myocardium. However, the average value can mask the heterogeneity of local microstructures. For example, in the very early stages of dilated cardiomyopathy, the average T1 value of the entire myocardium may be completely normal, but scattered micro-collagen deposits may already be present in the myocardial interstitium. These micro-lesions do not significantly increase the overall average value, but they can disrupt the originally uniform myocardial texture. In order to extract mathematical features that can quantitatively describe the degree of disorder and uniformity of myocardial microstructure from T1 mapping images without using any exogenous contrast agents, and to transform early diffuse fibrosis or extracellular matrix remodeling that is not visible to the naked eye into calculable digital indicators, the continuous T1 physical values ​​of each segmented region are first converted into discrete gray levels to facilitate the statistical analysis of the contribution frequency of pixel pairs. Then, a gray-level co-occurrence matrix is ​​constructed independently for each segmented region.

[0048] Furthermore, for each segmented region, the gray-level co-occurrence matrix is ​​constructed independently, specifically including the following steps:

[0049] The T1 values ​​in each divided region are resampled into N discrete gray levels. The spatial relative pixel distance parameter d and the direction angle parameter θ are set. For each direction angle parameter θ, the probability distribution of the simultaneous occurrence of a pixel with gray level i and a pixel with gray level j at a distance d along the θ direction is statistically analyzed to form a gray co-occurrence matrix for the corresponding direction angle parameter.

[0050] Since the T1 value is a continuous value, in order to divide the continuous value space into several discrete values ​​to facilitate the statistical analysis of the co-occurrence frequency of pixel pairs, the T1 value of each divided region is resampled into N discrete gray levels. N is usually 8, 16, or 32. Because N is too small, it will mask subtle texture differences, and too large, it will make the matrix sparse and the statistics unstable. Therefore, this embodiment uses 16. For example, the range of T1 values ​​in each divided region is divided into 16 smaller ranges in ascending order. A corresponding gray level is set for each smaller range, and each T1 value is mapped to the corresponding smaller range to obtain the corresponding gray level. The spatial relative pixel distance parameter d and the orientation angle parameter θ are set, where d refers to the number of related pixels between two pixels when statistically analyzing pixel pairs. Because changes in microstructure such as myocardial interstitial fibrosis are usually first reflected in the fluctuation of T1 values ​​between adjacent pixels, a smaller distance can more sensitively capture subtle texture numbers. Therefore, in this embodiment... The value of d is 1, which means that only directly adjacent pixel pairs in the image are counted. θ defines the relative direction of the pixel pairs. In order to fully capture the texture features in all directions, this implementation constructs gray-level co-occurrence matrices in four typical directions. The values ​​of θ are 0, 45, 90 and 135, where 0 represents the horizontal direction (including adjacent pixel pairs from left to right), 45 represents the main diagonal direction (including adjacent pixel pairs from top left to bottom right), 90 represents the vertical direction (including adjacent pixel pairs from top to bottom), and 135 represents the secondary diagonal direction (including adjacent pixel pairs from top right to bottom left).

[0051] To enhance statistical stability and ensure the symmetry of the gray-level co-occurrence matrix, pixel pairs in both the forward and reverse directions are counted simultaneously for each direction. Specifically: in the 0-degree direction, both horizontal right and horizontal left are counted; in the 90-degree direction, both vertical down and vertical up are counted; in the 45-degree direction, both lower right and upper left are counted; and in the 135-degree direction, both lower left and upper right are counted. For each directional angle parameter, the joint probability distribution of the simultaneous occurrence of a pixel with gray level i and a pixel with gray level j at a distance d along that direction is calculated. First, calculate the frequency matrix. Its element value is a gray level pair The total number of occurrences in that direction is then normalized to obtain... The calculation formula is as follows: Based on this method, an N×N gray-level co-occurrence matrix P0, P1, ..., P2 is formed in four directions. 45 P 90 P 135 .

[0052] The extraction of quantitative indicators representing tissue characteristics based on the gray-level co-occurrence matrix specifically includes the following steps:

[0053] Quantitative metrics include texture entropy and texture energy. Texture entropy is the negative of the sum of the products of the joint probabilities of all elements in the gray-level co-occurrence matrix and their own logarithms. Texture energy is the sum of the squares of the values ​​of each joint probability element.

[0054] After calculating the texture entropy and texture energy for each gray-level co-occurrence matrix, the texture entropy of each directional angle parameter is added together and the average value is taken as the final texture entropy, and the texture energy of each directional angle parameter is added together and the average value is taken as the final texture energy.

[0055] Specifically, quantitative indicators include texture entropy and texture energy. Texture entropy measures the randomness of elements in the gray-level co-occurrence matrix, i.e., the degree of texture disorder. Texture entropy is calculated as the negative of the sum of the products of the joint probabilities of all elements in the gray-level co-occurrence matrix and their own logarithms. The calculation formula is as follows: Texture energy is used to measure the smoothness and uniformity of gray-level distribution. Texture energy is calculated by summing the squares of the joint probability elements. The calculation formula is as follows:

[0056] To eliminate orientation dependence and obtain rotation-invariant texture features, the average values ​​of texture entropy and texture energy in the four directions are taken as the final texture entropy and final texture energy of the partitioned region.

[0057] Performing the above operation on all 16 partitioned regions yields 16 texture entropies and 16 texture energies.

[0058] Step S5: Calculate multiple heterogeneity indicators based on the quantitative indicators of each region, and perform multi-dimensional calculations based on the quantitative indicators and multiple heterogeneity indicators of each region to obtain quantitative evaluation results in different dimensions.

[0059] Specifically, the quantitative indicators for each segmented region have been calculated through previous steps. These quantitative indicators include texture entropy and texture energy. In order to reorganize the isolated and fragmented local texture features (quantitative indicators) according to the anatomical spatial logic of the heart, the spatial distribution pattern of myocardial texture is transformed into quantifiable digital indicators by calculating the difference gradient between different regions. In order to capture early microstructural abnormalities and issue early warning signals when there are no macroscopic changes in the heart morphology and the average T1 value of the myocardium is still in the normal reference range, the spatial variation relationship between all segmented regions is analyzed.

[0060] First, multiple heterogeneity indicators are calculated based on the quantitative indicators for each region. The calculation of multiple heterogeneity indicators specifically includes the following steps:

[0061] For each segmented region, the quantitative index is grouped according to the layer, the average texture energy of each layer is calculated, and the difference between the average texture energy of the basal layer and the apical layer is calculated as the first heterogeneity index.

[0062] For the basal level and the mid-ventricular level, the average texture entropy of the second and third division regions of each level is calculated as the interventricular septum texture entropy, the average texture entropy of the fifth and sixth division regions of each level is calculated as the lateral wall texture entropy, and the absolute difference between the interventricular septum texture entropy and the lateral wall texture entropy is calculated as the second heterogeneity index.

[0063] Each region is divided into three layers according to the distance of each pixel from the endocardium. The average T1 value is calculated for each layer. The T1 skewness of each layer is also calculated based on the T1 value. The absolute difference between the average T1 values ​​of the inner and middle layers is calculated as the third heterogeneity index. The absolute difference between the average T1 value of the outer layer and the average T1 value of the standard outer layer is calculated as the fourth heterogeneity index.

[0064] Some cardiomyopathies, such as early-stage myocardial amyloidosis, may only affect the apex or basal region of the heart, resulting in a significant difference in texture energy along the long axis. In contrast, the texture energy at the basal and apical regions of a normal heart should be roughly equivalent. To measure the overall difference in texture smoothness between the ventricular basal and apical regions and to capture gradient anomalies along the long axis of the heart, each region was grouped into three levels: six regions at the basal level, six regions at the mid-ventricular level, and four regions at the apical level. The average texture energy of each level was calculated, and the difference between the basal and apical levels was used as the first heterogeneity index. The first heterogeneity index refers to the energy difference between the basal and apical levels. The larger the absolute value of the first heterogeneity index, the more significant the difference in texture smoothness between the basal and apical regions, providing input for subsequent multidimensional assessments.

[0065] Hypertrophic cardiomyopathy often involves rotational attack on the interventricular septum, resulting in extremely high texture entropy while the lateral wall remains relatively normal. In contrast, primary dilated cardiomyopathy typically presents with diffuse damage, showing little difference between the two sides. To quantify the asymmetry in texture disorder between these two anatomically opposing regions—the interventricular septum and the left ventricular lateral wall—we extracted regions representing the interventricular septum from the basal and mid-ventricular levels: the second region (anterior septum) and the third region (inferior septum). Their mean texture entropy was calculated as the interventricular septal texture entropy. Similarly, regions representing the lateral wall from the basal and mid-ventricular levels: the fifth region (anterior lateral wall) and the sixth region (inferior lateral wall). Their mean texture entropy was calculated as the lateral wall texture entropy. Entropy is used to calculate the absolute difference between the interventricular septum texture entropy and the lateral wall texture entropy as a second heterogeneity index. The absolute difference between the interventricular septum texture entropy and the lateral wall texture entropy refers to the absolute value of the difference between the two texture entropies. The second heterogeneity index is used to quantify the degree of asymmetry in texture disorder between the two anatomically opposed regions of the interventricular septum and the left ventricular lateral wall. The larger the second heterogeneity index, the greater the degree of texture disorder in these two anatomically opposed regions, indicating an asymmetric damage pattern. This step, through the calculation of the second heterogeneity index, can quantify the imbalance of regional myocardial structural damage caused by pathogenic factors in the absence of focal contrast imaging.

[0066] Coronary microcirculatory disturbances are the earliest and most likely to cause subendocardial myocardial ischemia and edema, leading to elevated T1 values ​​in the subendocardium and relatively normal values ​​in the media, forming a significant transmural gradient. This phenomenon is usually absent in non-ischemic lesions. To capture the abrupt change in mean T1 values ​​from the endocardium to the media within the myocardial wall and identify subendocardial ischemia features, each region was divided into three layers based on the distance of each pixel from the endocardium: the inner layer, the middle layer, and the outer layer. The average T1 value was calculated layer by layer, and the T1 skewness of each layer was also calculated based on the T1 values. The formula for calculating T1 skewness is: , where x k It is the T1 value of the k-th pixel within each layer. σ is the mean of the T1 values ​​for each layer, σ is the standard deviation of the T1 values ​​for each layer, and n is the total number of pixels in each layer. The absolute difference between the average T1 values ​​of the inner and middle layers is calculated as the third heterogeneity index, which represents the through-wall gradient value. The third heterogeneity index is used to provide data quantification support for subsequent multi-dimensional evaluation.

[0067] In pure subendocardial ischemia, the subepicardial myocardium is usually unaffected, and its T1 value is no different from that of healthy individuals. If the subepicardial T1 value is significantly elevated, it indicates that the damage has penetrated the wall or is a non-ischemic full-thickness lesion, which does not meet the criteria for subendocardial ischemia. To test whether the T1 value of the subepicardial layer remains at the healthy baseline level, as an internal control, the absolute difference between the average T1 value of the outer layer and the average T1 value of the standard outer layer is calculated as the fourth heterogeneity index. The standard outer layer refers to the average T1 value of the subepicardial layer of the same anatomical region in the standard model of healthy individuals. By calculating the fourth heterogeneity index, an objective basis for excluding full-thickness lesions can be provided, which can be used in conjunction with the third heterogeneity index to determine the subendocardial ischemia pattern.

[0068] After calculating multiple heterogeneity indicators, multi-dimensional calculations can be performed based on the quantitative indicators and multiple heterogeneity indicators for each region to obtain quantitative evaluation results in different dimensions. The specific steps include the following:

[0069] Calculate the first distance between the current texture entropy and texture energy and the texture entropy and texture energy of healthy people. If the first distance is greater than the preset distance threshold, generate a diffuse myocardial fibrosis early warning marker.

[0070] When the second heterogeneity index is greater than the preset asymmetry judgment constant, a topological overlap identification marker is generated;

[0071] Calculate the texture energy variance for all partitioned regions. When the texture energy variance is less than the homogeneity judgment threshold and the first heterogeneity index is less than the preset index threshold, generate a global homogeneity judgment label.

[0072] For each segmented region, if there exists at least one segmented region whose corresponding third heterogeneity index is greater than a preset first judgment threshold and simultaneously satisfies the first and second conditions, a transmural ischemic injury pattern label is generated. The first condition is that the fourth heterogeneity index of the corresponding segmented region is less than the preset second judgment threshold, and the second condition is that the T1 skewness of the inner layer of the corresponding segmented region is greater than the preset first skewness threshold, and the absolute values ​​of the T1 skewness of the middle and outer layers of the corresponding segmented region are less than or equal to the preset second skewness threshold.

[0073] Specifically, in high-risk groups such as carriers of genes causing dilated cardiomyopathy and patients undergoing chemotherapy for cancer, the myocardial interstitium may begin to develop microedema or very early fibrosis before the left ventricular ejection fraction decreases. At this time, the variation in T1 value in the myocardial region is often smaller than the measurement error range of magnetic resonance imaging equipment and cannot be identified by the naked eye or traditional methods. However, the spatial correlation between multiple divided regions may have undergone small but statistically significant changes, which is what the first dimension aims to capture.

[0074] To detect subtle distortions in the covariance of texture features across multiple segmented regions during a pseudo-healthy period characterized by normal cardiac morphology and normal mean T1 values, and to issue early warnings, a first distance (Mahavira distance) is calculated between the current texture entropy and energy and those of healthy individuals. When this first distance exceeds a preset threshold, it indicates that microscopic texture covariance has become disordered, thus generating a diffuse myocardial fibrosis warning marker. This marker provides attending physicians with a basis for judgment, allowing them to conduct further examinations or initiate preventative interventions as needed.

[0075] When the heart is enlarged, it is necessary to differentiate between other types of cardiomyopathy in the decompensated or end-stage stage and dilated cardiomyopathy. Due to significant differences in treatment strategies and prognoses, conventional imaging analysis is often insufficient to distinguish between the two, while texture entropy asymmetry is a strong differentiating factor. Therefore, when the second heterogeneity index exceeds the preset asymmetry judgment constant, a topological overlap differential marker is generated. This marker indicates to physicians that it may be another type of cardiomyopathy in the decompensated or end-stage stage, rather than primary dilated cardiomyopathy, providing crucial differential diagnostic evidence for clinicians.

[0076] To identify unique, uniformly diffuse injury patterns, most commonly seen in alcoholic cardiomyopathy or chemotherapy drug cardiotoxicity, where toxins directly attack all cardiomyocytes, leading to diffuse damage throughout the myocardium and lacking focal fibrosis, texture entropy typically exhibits uniformity. Therefore, we first calculate the texture energy variance for all segmented regions. When the texture energy variance is less than the homogeneity threshold and the first heterogeneity index is less than a preset threshold, a global homogeneity determination label is generated. By generating this global homogeneity determination label, uniform injury can be distinguished from diffuse fibrosis, guiding clinical adjustments to medication or other interventions.

[0077] To identify transmural gradient features of subendocardial ischemic injury and differentiate between ischemic and non-ischemic cardiomyopathy without contrast agents, a transmural ischemic injury pattern marker is generated when at least one segmented region has a corresponding third heterogeneity index greater than a preset first threshold, and both the first and second conditions are met. The third heterogeneity index being greater than the preset first threshold indicates a significant T1 difference in the inner and middle layers. The first condition is that the fourth heterogeneity index of the corresponding segmented region is less than the preset second threshold; meeting the first condition indicates that the subendocardium is unaffected. The second condition is that the T1 skewness of the inner layer of the corresponding segmented region is greater than the preset first skewness threshold, and the absolute values ​​of the T1 skewness of the middle and outer layers of the corresponding segmented region are less than or equal to the preset second skewness threshold; meeting the second condition indicates that high T1 pixels exhibit a focal right-skewed distribution and are limited to the inner layer. In this scenario, the myocardial T1 value exhibits a transmural attenuation gradient pattern along the radial direction of the ventricular wall, characterized by a significantly elevated mean T1 value in the inner layer, a positively skewed distribution of T1 in the inner layer, and outer layer T1 characteristics consistent with the healthy baseline. This gradient pattern is statistically associated with ischemic cardiomyopathy involving subendocardial ischemic remodeling along coronary artery branches. Therefore, generating transmural ischemic injury pattern markers when the above conditions are met can aid in the clinical diagnosis of ischemic cardiomyopathy without the use of contrast agents.

[0078] The above describes a method for intelligent quantitative analysis of myocardial texture based on CMR images in an embodiment of this application. The following describes a system for intelligent quantitative analysis of myocardial texture based on CMR images in an embodiment of this application. Please refer to [link to relevant documentation]. Figure 5 One embodiment of the intelligent quantitative analysis system for myocardial texture based on CMR images in this application includes:

[0079] The image acquisition module is used to acquire raw T1 mapping images of the heart in three layers along the short axis: the basal layer, the mid-ventricular layer, and the apical layer.

[0080] The myocardial extraction module is used to perform masking processing on the original images of the three layers respectively, extract the closed-loop myocardial region, and obtain the T1 physical value of the myocardial region of each layer.

[0081] The alignment and segmentation module is used to identify the reference point based on the myocardial region at the base level, calculate the rotation angle based on the reference point and the centroid of the internal region, and rotate the myocardial region at the base level, the middle ventricular level and the apical level based on the rotation angle to obtain the aligned myocardial region images at the three levels. The myocardial region at the three levels is segmented according to the preset standard to obtain multiple division regions.

[0082] The indicator extraction module is used to independently construct a gray-level co-occurrence matrix for each segmented region, and extract quantitative indicators representing organizational characteristics based on the gray-level co-occurrence matrix.

[0083] The multidimensional evaluation module is used to calculate multiple heterogeneity indicators based on the quantitative indicators of each region, and to perform multidimensional calculations based on the quantitative indicators and multiple heterogeneity indicators of each region to obtain quantitative evaluation results in different dimensions.

[0084] This application also protects a computer device, which includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the computer device can perform any of the aforementioned methods for intelligent quantitative analysis of myocardial texture based on CMR images.

[0085] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent quantitative analysis of myocardial texture based on CMR images, characterized in that, The method includes: Step S1: Obtain T1 Mapping raw images of the heart in three layers along the short axis, including the basal layer, the mid-ventricular layer, and the apical layer. Step S2: Perform masking on the original images of the three layers respectively, extract the closed-loop myocardial region, and obtain the T1 physical value of the myocardial region of each layer. Step S3: Identify the reference point based on the myocardial region at the base level, calculate the rotation angle based on the reference point and the centroid of the internal region, and rotate the myocardial regions at the base level, the middle ventricular level, and the apical level based on the rotation angle to obtain aligned images of the myocardial regions at the three levels. Divide the myocardial regions at the three levels into multiple subdivision regions according to a preset standard. Step S4: For each segmented region, construct the gray-level co-occurrence matrix independently, and extract quantitative indicators representing organizational characteristics based on the gray-level co-occurrence matrix; Step S5: Calculate multiple heterogeneity indicators based on the quantitative indicators of each region, and perform multi-dimensional calculations based on the quantitative indicators and multiple heterogeneity indicators of each region to obtain quantitative evaluation results in different dimensions.

2. The method according to claim 1, characterized in that, Reference points were identified based on the myocardial region at the basal level, including: The curvature of each pixel is calculated along the outer contour of the myocardium. Local minimum points where the curvature is concave inward are obtained as geometric concavity feature points. Anterior interventricular groove feature points are identified from the geometric concavity feature points and defined as reference points.

3. The method according to claim 2, characterized in that, The rotation angle is calculated based on the reference point and the centroid of the internal region, including: The centroid of the inner region surrounding the myocardial region is calculated as the centroid of the inner region. The line connecting the reference point and the centroid of the inner region is used as the reference axis. Based on the preset standard axis, the angle difference between the reference axis and the standard axis is obtained. The angle difference is used as the rotation angle. Based on the rotation angle, the myocardial region images of the three layers are rotated and aligned in the direction of the standard axis.

4. The method according to claim 1, characterized in that, The three layers of myocardial regions were divided according to preset criteria to obtain multiple subdivisions, including: For the basal and mid-ventricular levels, the area is divided into six local myocardial regions, with the centroid of the internal region as the center and the standard axis as the starting point, in units of 60 degrees. For the apical level, the area is divided into four local myocardial regions, with the centroid of the internal region as the center and the standard axis as the starting point, in units of 90 degrees.

5. The method according to claim 1, characterized in that, Independently construct the gray-level co-occurrence matrix, including: The T1 values ​​in each divided region are resampled into N discrete gray levels. The spatial relative pixel distance parameter d and the direction angle parameter θ are set. For each direction angle parameter θ, the probability distribution of the simultaneous occurrence of a pixel with gray level i and a pixel with gray level j at a distance d along the θ direction is statistically analyzed to form a gray co-occurrence matrix for the corresponding direction angle parameter.

6. The method according to claim 1, characterized in that, Quantitative indicators representing tissue characteristics are extracted based on the gray-level co-occurrence matrix, including: Quantitative metrics include texture entropy and texture energy. Texture entropy is the negative of the sum of the products of the joint probabilities of all elements in the gray-level co-occurrence matrix and their own logarithms; texture energy is the sum of the squares of the values ​​of each joint probability element. After calculating the texture entropy and texture energy for each gray-level co-occurrence matrix, the texture entropy of each directional angle parameter is added together and the average value is taken as the final texture entropy, and the texture energy of each directional angle parameter is added together and the average value is taken as the final texture energy.

7. The method according to claim 1, characterized in that, Multiple heterogeneity indicators were calculated based on the quantitative indicators for each partitioned region, including: For each segmented region, the quantitative index is grouped according to the layer, the average texture energy of each layer is calculated, and the difference between the average texture energy of the basal layer and the apical layer is calculated as the first heterogeneity index. For the basal level and the mid-ventricular level, the average texture entropy of the second and third division regions of each level is calculated as the interventricular septum texture entropy, the average texture entropy of the fifth and sixth division regions of each level is calculated as the lateral wall texture entropy, and the absolute difference between the interventricular septum texture entropy and the lateral wall texture entropy is calculated as the second heterogeneity index. Each region is divided into three layers according to the distance of each pixel from the endocardium. The average T1 value is calculated for each layer. The T1 skewness of each layer is also calculated based on the T1 value. The absolute difference between the average T1 values ​​of the inner and middle layers is used as the third heterogeneity index. The absolute difference between the average T1 value of the outer layer and the average T1 value of the standard outer layer is also used as the fourth heterogeneity index.

8. The method according to claim 1, characterized in that, Multi-dimensional calculations are performed to obtain quantitative evaluation results in different dimensions, including: Calculate the first distance between the current texture entropy and texture energy and the texture entropy and texture energy of healthy individuals; if the first distance is greater than a preset distance threshold, generate a diffuse myocardial fibrosis early warning marker. When the second heterogeneity index is greater than the preset asymmetry judgment constant, a topological overlap identification mark is generated. Calculate the texture energy variance for all partitioned regions. When the texture energy variance is less than the homogeneity judgment threshold and the first heterogeneity index is less than the preset index threshold, generate a global homogeneity judgment label. For each segmented region, if there exists at least one segmented region whose corresponding third heterogeneity index is greater than a preset first judgment threshold and simultaneously satisfies the first and second conditions, a transmural ischemic injury pattern label is generated. The first condition is that the fourth heterogeneity index of the corresponding segmented region is less than the preset second judgment threshold, and the second condition is that the T1 skewness of the inner layer of the corresponding segmented region is greater than the preset first skewness threshold, and the absolute values ​​of the T1 skewness of the middle and outer layers of the corresponding segmented region are less than or equal to the preset second skewness threshold.

9. A CMR image-based intelligent quantitative analysis system for myocardial texture, used to implement the CMR image-based intelligent quantitative analysis method for myocardial texture as described in any one of claims 1-8, characterized in that, The system includes: The image acquisition module is used to acquire raw T1 mapping images of the heart in three layers along the short axis: the basal layer, the mid-ventricular layer, and the apical layer. The myocardial extraction module is used to perform masking processing on the original images of the three layers respectively, extract the closed-loop myocardial region, and obtain the T1 physical value of the myocardial region of each layer. The alignment and segmentation module is used to identify the reference point based on the myocardial region at the base level, calculate the rotation angle based on the reference point and the centroid of the internal region, and rotate the myocardial region at the base level, the middle ventricular level and the apical level based on the rotation angle to obtain the aligned myocardial region images at the three levels. The myocardial region at the three levels is segmented according to the preset standard to obtain multiple division regions. The indicator extraction module is used to independently construct a gray-level co-occurrence matrix for each segmented region, and extract quantitative indicators representing organizational characteristics based on the gray-level co-occurrence matrix. The multidimensional evaluation module is used to calculate multiple heterogeneity indicators based on the quantitative indicators of each region, and to perform multidimensional calculations based on the quantitative indicators and multiple heterogeneity indicators of each region to obtain quantitative evaluation results in different dimensions.

Images