A liver cirrhosis CT image liver edge sawtooth feature extraction method

Abstract

The present application relates to the technical field of edge feature extraction, and particularly relates to a liver cirrhosis CT image liver edge jagged feature extraction method. The method acquires abdominal portal phase CT images, segments liver, portal vein and inferior vena cava regions, extracts a cross section where a portal vein main stem bifurcation point is located and related vein center coordinates, constructs a plane topological relationship, and then divides the liver into a compensatory reconstruction analysis area, a fibrosis stress analysis area and a reference control analysis area on the cross section, and acquires a jagged feature set of the outer edge contour of each analysis area; measures a caudate lobe-right lobe ratio and combines macroscopic pathological indexes to acquire a reliability parameter and a synergistic contribution value of each index, and linearly weights to generate a reliability weight of each analysis area; based on the reliability weight, the jagged feature set is corrected and fused to output a whole liver edge jagged quantification representation. The present application can avoid dilution of local lesion features, strengthen the clinical confidence of edge features, and improve feature extraction accuracy.

Description

Technical Field

[0001] This invention relates to the field of edge feature extraction technology, specifically to a method for extracting serrated features of the liver edge in CT images of cirrhosis. Background Technology

[0002] Liver cirrhosis is a common chronic, progressive, and diffuse liver disease caused by various etiologies. During the progression of cirrhosis, the proliferation of fibrous tissue and the formation of pseudolobules within the liver lead to nodular remodeling on the liver surface. On medical imaging (such as CT images), this manifests primarily as an uneven, serrated, or wavy appearance of the liver margins. Therefore, accurately extracting and quantifying the serrated features of the liver margins from CT images is of crucial clinical value for the auxiliary diagnosis of cirrhosis, disease grading assessment, and prediction of the risk of decompensation.

[0003] Current methods for extracting serrated features from liver edges typically begin by obtaining the overall edge contour of the entire liver through image segmentation algorithms. This is then combined with methods such as Canny edge detection, local binary mode, or conventional curvature calculation to perform unified mathematical calculations and evaluations of the undulation features of the entire liver's outer contour. However, liver morphological remodeling caused by cirrhosis exhibits strong spatial heterogeneity. Specifically, during pathological evolution, the right lobe of the liver, due to long-term influence from enterogenic venous blood flow, experiences significant fibrous tissue contraction and traction, easily forming prominent serrated defects and depressions on its surface. Meanwhile, regions such as the caudate lobe often show compensatory hypertrophy with relatively smooth edges. Existing techniques, based on a unified contour analysis of the entire liver's edge, tend to average the high-intensity serrated features with relatively smooth regional features, resulting in severe dilution and weakening of typical local pathological features. Consequently, it becomes difficult to objectively reflect the true degree of lesions in different liver lobes. Secondly, existing serrated feature extraction methods mostly rely on pure two-dimensional or three-dimensional geometric edge measures. During CT imaging, there are respiratory motion artifacts, partial volume effects, and image reconstruction noise. These non-pathological factors cause contour burrs and false edge fluctuations, which are easily misidentified as pathological serrated structures by existing algorithms. Due to the lack of multi-dimensional medical pathological information for cross-validation and constraint, the serrated features extracted by existing methods are prone to false positives and have low feature confidence, making them difficult to use as reliable auxiliary decision-making basis in clinical practice. Summary of the Invention

[0004] To address the technical problem that existing methods cannot accurately extract the serrated edges of the liver, the present invention aims to provide a method for extracting the serrated edges of the liver in CT images of cirrhosis. The specific technical solution adopted is as follows: This invention provides a method for extracting the serrated edge features of the liver in CT images of liver cirrhosis. The method includes the following steps: CT images of the abdominal portal vein phase were acquired and segmented to obtain the liver region, portal vein region, and inferior vena cava region. Based on the portal vein region and inferior vena cava region, the cross-section where the main bifurcation point of the portal vein is located, as well as the coordinates of the center of the portal vein and the center of the inferior vena cava on the cross-section, were extracted to construct the planar topological relationship. Based on planar topological relationships, the liver region is divided into a compensatory reconstruction analysis area, a fibrosis stress analysis area, and a reference control analysis area on the cross section; based on the corresponding outer edge contour of each analysis area on the cross section, the serrated feature set of each analysis area is obtained. The caudate lobe-to-right lobe ratio was measured based on planar topological relationships. Combined with macroscopic pathological indicators, the reliability parameters and synergistic contribution values ​​of the caudate lobe-to-right lobe ratio and each macroscopic pathological indicator were obtained. The reliability weights of each analysis region were generated by linear weighting. Based on the credibility weight, the serrated feature set is modified and fused to obtain the whole liver multidimensional feature dictionary and the whole liver comprehensive deviation index, so as to output the quantitative representation of the whole liver edge serration.

[0005] Furthermore, the method for obtaining the cross-section is as follows: The portal vein region is processed into a skeleton to obtain the portal vein centerline, and the bifurcation nodes where the main trunk and the left and right first-order branches on the portal vein centerline intersect are identified. The axial slice containing the bifurcation node is used as the cross section where the main portal vein bifurcation point is located.

[0006] Furthermore, the method for obtaining the planar topological relationship is as follows: Extract the inferior vena cava centerline in the inferior vena cava region, and then obtain the intersection points of the portal vein centerline and the inferior vena cava centerline with the cross section in turn. These points are used as the coordinates of the portal vein center and the inferior vena cava center on the cross section in turn. Construct a central line connecting the coordinates of the portal vein center and the inferior vena cava center, and extract the relative positional relationship between the coordinates of the portal vein center and the inferior vena cava center by combining the sagittal and coronal directions of the CT image; A two-dimensional anatomical reference system is constructed on the cross-section using the central connecting line and the relative positional relationship, which serves as the planar topological relationship.

[0007] Furthermore, the method for obtaining the compensatory reconstruction analysis region, the fibrosis stress analysis region, and the reference control analysis region is as follows: Based on planar topological relationships, the boundaries of the Couinaud functional segments of the liver are determined on the cross section. The region within the cross section belonging to the caudate lobe of Couinaud 1 segment is designated as the compensatory reconstruction analysis region; The area within the cross-section belonging to the right lobe of the liver, from Couinaud segments 5 to 8, was designated as the fibrosis stress analysis area. The region within the cross-section belonging to the left lobe of the liver, segments 2 to 4 of Couinaud, was designated as a reference control analysis area.

[0008] Furthermore, the method for obtaining the sawtooth feature set is as follows: For any analysis region, the discrete points on the outer edge contour of the analysis region are smoothed to generate a reference curve. The Euclidean distance between each discrete point and the corresponding point on the reference curve is calculated and used as the reference distance. The average of all reference distances is used as the contour deviation of the analysis region. Arrange the reference distances according to the specified directions of the discrete point positions to obtain a reference distance sequence; obtain the sum of the number of local maxima and local minima in the reference distance sequence as the reference quantity; and take the ratio of the reference quantity to the total arc length of the outer edge contour of the analysis area as the contour peak-valley density of the analysis area. Spline fitting is performed on the outer edge contour of the analysis area, and the standard deviation of the curvature at each sampling point on the fitted curve is calculated as the contour curvature fluctuation of the analysis area. Set a preset number of different preset grid side lengths, count the number of grids containing the outer edge contour points of the analysis area, and fit the linear slope between the logarithm of the reciprocal of the preset grid side length and the logarithm of the number of grids using the least squares method, and use it as the contour complexity of the analysis area. The set consisting of the contour deviation magnitude, contour peak and valley density, contour curvature fluctuation, and contour complexity of the analysis area is taken as the serrated feature set of the analysis area.

[0009] Furthermore, the method for obtaining the caudate leaf-right leaf ratio is as follows: Draw a first reference line on the cross section that passes through the right side wall of the bifurcation point of the portal vein and is parallel to the sagittal plane of the human body, and a second reference line that passes through the leftmost edge of the caudate lobe and is parallel to the sagittal plane of the human body. Calculate the vertical distance between the first reference line and the second reference line, and use it as the caudate leaf span value; Calculate the vertical distance between the first reference line and the outermost edge of the right lobe of the liver, and use it as the right lobe span value; Divide the caudate leaf span value by the right leaf span value to obtain the caudate leaf-right leaf ratio.

[0010] Furthermore, the method for obtaining the credibility parameter and the collaborative contribution value is as follows: The caudate lobe-to-right lobe ratio and various macroscopic pathological indicators were normalized to obtain standardized parameters for each indicator; among them, macroscopic pathological indicators included spleen diameter parameters, portal vein sign parameters, and hemodynamic parameters. Subtracting the preset pathological threshold from the standardized parameters and multiplying it by the preset adjustment coefficient, the negative number of the result is used as the exponent of the natural constant for exponentiation. The result of exponentiation is then increased by 1 and the reciprocal is taken to obtain the reliability parameters of each indicator. The credibility parameters of each indicator are multiplied in pairs to obtain the corresponding collaborative contribution value.

[0011] Furthermore, the method for obtaining the credibility weight is as follows: For the fibrotic stress analysis area, the corresponding reference synergistic contribution values ​​are obtained through the confidence parameters of the caudate lobe-to-right lobe ratio, the portal vein sign parameters, and the hemodynamic parameters. The confidence parameters of the caudate lobe-to-right lobe ratio, portal vein sign parameters, and hemodynamic parameters, along with the maximum reference synergistic contribution value, are weighted and summed according to the preset weight allocation ratio to obtain the confidence weight of the fibrotic stress analysis area. For the compensatory reconstruction analysis region, the credibility parameters of the caudate lobe-right lobe ratio and the spleen diameter parameter, as well as the collaborative contribution values ​​corresponding to the caudate lobe-right lobe ratio and the spleen diameter parameter, are weighted and summed according to the preset weight allocation ratio to obtain the credibility weight of the compensatory reconstruction analysis region. For the reference control analysis area, the credibility weight of the reference control analysis area is obtained by weighting and summing the maximum values ​​of the credibility parameters of all indicators and all collaborative contribution values ​​according to the preset weight allocation ratio.

[0012] Furthermore, the method for obtaining the whole liver multidimensional feature dictionary is as follows: For any analysis region, the confidence weight of the analysis region is multiplied by the contour deviation magnitude, contour peak and valley density and contour curvature fluctuation of the analysis region respectively, and the original value of the contour complexity of the analysis region is retained to generate the corrected jagged feature set of the analysis region. The modified serrated feature sets of each analysis region are structurally spliced ​​and combined to form a complete multidimensional feature dictionary of the whole liver that retains the feature representation of each anatomical region.

[0013] Furthermore, the method for obtaining the whole liver comprehensive deviation index is as follows: Extract the corrected contour deviation magnitude of each analysis region from the corrected serrated feature set of each analysis region; The weighted summation of the deviation magnitudes of the corrected contours in each analysis region yields the overall liver deviation index.

[0014] The present invention has the following beneficial effects: This invention first segments abdominal CT images during the portal venous phase to obtain the liver, portal vein, and inferior vena cava regions, facilitating precise localization of the most typical anatomical structures in the pathological manifestations of liver cirrhosis. To reduce the complexity of three-dimensional surface analysis to two-dimensional slices—the most stable anatomical structures and the most commonly used in clinical assessment—and avoid algorithmic deadlock caused by tracking three-dimensional surface undulations, the invention further extracts cross-sections of the portal vein bifurcation points, along with the coordinates of the portal vein and inferior vena cava centers on these cross-sections, based on the portal vein and inferior vena cava regions. This constructs a planar topological relationship, providing an absolute spatial positioning benchmark unaffected by patient position for subsequent precise partitioning and feature extraction. Furthermore, to avoid the pitfalls of whole-liver contour analysis leading to localization issues... The typical high-intensity serrated features of the liver are weakened by averaging relatively smooth regional features. Therefore, based on planar topological relationships, the liver region is divided into a compensatory remodeling analysis area, a fibrosis stress analysis area, and a reference control analysis area on the cross-section, enabling targeted and independent assessment of different spatially heterogeneous pathological manifestations. To objectively quantify the irregular morphological changes caused by nodular remodeling on the liver surface, the serrated feature set of each analysis area is obtained based on the corresponding outer edge contour on the cross-section. This accurately characterizes the roughness of the macroscopic deformation and microscopic undulations of the edges of different analysis areas, providing rich underlying quantitative evidence for assessing the progression of cirrhosis. This approach aims to achieve morphological underlying measurement and objective pathogenesis in humans. This approach demonstrates in-depth interaction and provides cross-dimensional evidence for comprehensive decision support. It measures the caudate lobe-to-right lobe ratio based on planar topological relationships, combining this with macroscopic pathological indicators to obtain the confidence parameters and synergistic contribution values ​​of the caudate lobe-to-right lobe ratio and each macroscopic pathological indicator. This accurately reflects the certainty of the pathological state indicated by individual or combined indicators. Utilizing multi-dimensional macroscopic pathological data strengthens the clinical confidence of corresponding region edge features as a substantial representation of cirrhosis. By introducing a system-level observation dimension, it significantly reduces the comprehensive assessment error caused by the lack of system constraints when relying solely on unidirectional low-level geometry and contour operations for state characterization. Furthermore, it generates confidence weights for each analysis region through linear weighting, reasonably explaining... The importance contribution of edge serration degree in each region to the prediction of medical history; in order to provide a comprehensive assessment scale that takes into account both spatial heterogeneity and the severity of macroscopic nodulation of the whole liver, the serration feature set is modified and fused based on the credibility weight to obtain a multidimensional feature dictionary of the whole liver and a comprehensive deviation index of the whole liver. This effectively harmonizes the key weights of feature quantities of different functional blocks in the criterion system, so that the final output morphological quantification matrix is ​​more in line with the development trend of clinical pathology, thereby accurately outputting the quantitative characterization of edge serration of the whole liver. This provides an objective and reliable imaging-assisted decision-making basis for the accurate grading of clinical cirrhosis, monitoring of disease progression, and prediction of decompensation risk, greatly improving the accuracy of feature extraction and clinical practical value. Attached Figure Description

[0015] Figure 1This is a schematic flowchart illustrating a method for extracting serrated features of the liver edge from CT images of cirrhosis, as provided in one embodiment of the present invention. Detailed Implementation

[0016] The following describes in detail, with reference to the accompanying drawings, a specific scheme for extracting the serrated edge features of liver in CT images of liver cirrhosis provided by the present invention.

[0017] Example 1: This invention proposes a method for extracting the serrated edge features of the liver in CT images of liver cirrhosis. Please refer to [link / reference]. Figure 1 The diagram illustrates a schematic flowchart of a method for extracting serrated features of the liver edge in CT images of liver cirrhosis, according to an embodiment of the present invention. The method includes the following steps: Step S1: Obtain CT images of the abdominal portal vein phase and segment them to obtain the liver region, portal vein region, and inferior vena cava region; based on the portal vein region and inferior vena cava region, extract the cross-section where the main bifurcation point of the portal vein is located, as well as the coordinates of the center of the portal vein and the center of the inferior vena cava on the cross-section, and construct the planar topological relationship.

[0018] Specifically, this embodiment uses a patient with cirrhosis as an example for analysis; all subsequent mentions of cirrhosis patients refer to this patient. To objectively and accurately assess the liver morphological remodeling in cirrhotic patients, the patient's abdominal enhanced CT scan sequence was first acquired. Considering the uniform enhancement of liver parenchyma, clear visualization of the liver capsule boundary, and the highest contrast of the main portal vein and its branches during the portal venous phase, which facilitates accurate identification and localization of anatomical structures, the abdominal CT images from the portal venous phase were used as the analysis images to improve the stability and reliability of low-level feature extraction. Furthermore, to overcome the limitations of single morphological fluctuation assessment and ensure the smooth calculation of subsequent combined weights, the system simultaneously receives the patient's macroscopic pathological indicators at similar time points through the data interface of the hospital information system or electronic medical record system while acquiring the CT images. These indicators include spleen diameter parameters, portal vein sign parameters, and hemodynamic parameters, and are temporarily stored in system memory for later retrieval. To accurately separate the target structures required for topology analysis, a pre-trained 3D medical image segmentation network (such as a 3D nnU-Net model or a similar 3D convolutional neural network) is used to segment the abdominal CT image during the portal vein phase to obtain 3D voxel masks for the liver, portal vein, and inferior vena cava regions. Further, residual segmentation noise is removed through 3D maximum connected component preservation and morphological smoothing to ensure the continuity and integrity of the acquired regions. The image segmentation network, 3D maximum connected component preservation, and morphological smoothing are all well-known techniques and will not be elaborated further.

[0019] Considering that directly tracking the undulations on the three-dimensional whole liver surface can easily lead to algorithm deadlock due to the complexity of the surface geometry and topology, and cannot highlight the typicality of local lesions, in order to reduce the complexity of three-dimensional spatial analysis to two-dimensional slices with the most significant clinical pathological features and the most stable anatomical structure, we extract the cross-section where the main bifurcation point of the portal vein is located, as well as the coordinates of the center of the portal vein and the center of the inferior vena cava on the cross-section, based on the portal vein region and the inferior vena cava region, to construct a planar topological relationship, providing an absolute spatial positioning benchmark, which is conducive to accurately analyzing the lesion status of the whole liver.

[0020] The method for obtaining the cross-section is as follows: First, the portal vein region is processed into a skeleton using a 3D thinning algorithm or topological skeleton extraction technique to obtain the portal vein centerline with a monomer width, eliminating interference caused by uneven vessel diameter. Then, a bifurcation point detection algorithm is used to identify the bifurcation node where the main trunk and the left and right first-order branches (i.e., the left and right branches of the portal vein) intersect on the portal vein centerline. This bifurcation node is a core anatomical landmark of the first hepatic hilum region. The axial slice containing this bifurcation node is then used as the cross-section where the main trunk of the portal vein bifurcation point is located, allowing subsequent analysis to accurately focus on the horizontal level where liver atrophy and compensatory remodeling are most typical. The 3D thinning algorithm, topological skeleton extraction technique, and bifurcation point detection algorithm are all well-known techniques and will not be described in detail here.

[0021] The method for obtaining the planar topological relationships is as follows: Considering that the spatial connection between the portal vein and the inferior vena cava and the angle between them and the sagittal plane are important natural anatomical landmarks for dividing the left and right halves of the liver and the caudate lobe, in order to establish a standard reference system unaffected by the patient's individual body position, the inferior vena cava centerline is extracted in the inferior vena cava region using a centerline extraction algorithm. Then, the intersections of the portal vein centerline and the inferior vena cava centerline with the cross-section are obtained sequentially and used as the coordinates of the portal vein center and the inferior vena cava center on the cross-section, thereby accurately mapping the three-dimensional venous conduit to anchor coordinate points on a two-dimensional plane; in order to determine the geometric orientation benchmark of the cross-section, a connection is constructed... By connecting the centers of the portal vein and inferior vena cava, and combining the sagittal (Y-axis, defined in DICOM image metadata) and coronal (X-axis, defined in DICOM image metadata) coordinates of the CT image, the relative positional relationship between the portal vein and inferior vena cava centers (i.e., the spatial distance between them and the angle between the connecting line and the sagittal or coronal plane of the body) is extracted. To eliminate slight deviation errors during image acquisition, a two-dimensional anatomical reference system is constructed on the cross-section using the connecting line and the relative positional relationship. This serves as a planar topological relationship, laying the geometric coordinate foundation for subsequent accurate measurements and multi-region division within the cross-section. The centerline extraction algorithm is well-known and will not be elaborated further.

[0022] Step S2: Based on the planar topological relationship, the liver region is divided into a compensatory reconstruction analysis area, a fibrosis stress analysis area, and a reference control analysis area on the cross section; based on the corresponding outer edge contour of each analysis area on the cross section, the serrated feature set of each analysis area is obtained.

[0023] Specifically, considering the significant spatial heterogeneity of pathological remodeling and fibrotic contraction in different hepatic lobe regions during the progression of cirrhosis, and to avoid the problem that whole-liver contour analysis can easily lead to the weakening of local high-intensity serrated features by averaging with smoothed region features, the liver region is divided into a compensatory remodeling analysis area, a fibrotic stress analysis area, and a reference control analysis area on the cross-section based on planar topological relationships. This facilitates targeted and independent assessment of areas with different pathological manifestations, improving the accuracy of the analysis. Furthermore, considering that irregular protrusions and depressions at the edges are the most direct imaging manifestation of nodular remodeling on the liver surface, in order to objectively quantify this edge irregularity, a set of serrated features for each analysis area is obtained based on the corresponding outer edge contour on the cross-section. This accurately reflects the roughness and undulation of the edges of different analysis areas, providing a quantitative basis for the objective assessment of the progression of cirrhosis.

[0024] Preferably, in one feasible embodiment of this invention, the method for obtaining the compensatory reconstruction analysis area, the fibrosis stress analysis area, and the reference control analysis area is as follows: Based on the planar topological relationship (i.e., the two-dimensional anatomical reference system formed by the central line connecting the portal vein and the inferior vena cava), the boundaries of the Couinaud functional segments of the liver are determined on the cross-section using a spatial coordinate mapping algorithm. For example, using the central line as the orientation reference, a reference segmentation line is drawn that passes through the central coordinate of the portal vein and is perpendicular to the central line. The mask area near the ventral side of the reference segmentation line and located within a specified angle range to the left of the central line is assigned to the caudate lobe of Couinaud segment 1; the area to the right and posterior of the central line is divided into Couinaud segments 5 to 8; and the remaining anterior and left anterior areas are assigned to Couinaud segments. Segments 2 through 4 are used to define the specific range of each anatomical segment. It is known that the caudate lobe often undergoes compensatory hypertrophy and has relatively smooth edges during the progression of cirrhosis. However, the right lobe, due to long-term influence from enterogenic venous blood flow, experiences significant fibrous tissue contraction and traction, resulting in an uneven surface prone to prominent serrations. The morphological changes in the left lobe are relatively central. Therefore, the area within the cross-section belonging to the caudate lobe of segment 1 of Couinaud is designated as the compensatory remodeling analysis area; the area within the cross-section belonging to the right lobe of segments 5 through 8 of Couinaud is designated as the fibrotic stress analysis area; and the area within the cross-section belonging to the left lobe of segments 2 through 4 of Couinaud is designated as the reference control analysis area. The spatial coordinate mapping algorithm and Couinaud functional segmentation are well-known and will not be elaborated further.

[0025] Preferably, in one feasible embodiment, the method for obtaining the serrated feature set is as follows: For any analysis region, in order to eliminate the macroscopic geometric curvature trend of the edge to extract pure microscopic serrated undulations, a reference curve is generated by smoothing discrete points on the outer edge contour of the analysis region using a Gaussian kernel filtering algorithm or a moving average smoothing algorithm, thus removing the local protrusions caused by nodules; in order to quantify the absolute magnitude of the actual edge deviating from the ideal smooth state, the Euclidean distance between each discrete point and the corresponding point on the reference curve is calculated and used as a reference distance to accurately reflect the concavity and convexity depth of the local edge; in order to obtain the overall undulation scale of the analysis region, the average of all reference distances is used as the contour deviation magnitude of the analysis region to accurately reflect the macroscopic edge deformation degree of the analysis region; wherein, the Gaussian kernel filtering algorithm, the moving average smoothing algorithm, and the method for obtaining Euclidean distance are all well-known and will not be described in detail. Considering that the jagged features are manifested not only in the depth of amplitude but also in the frequency density of undulation inflection points, in order to quantify the spatial frequency of such edge fluctuations, reference distances are arranged according to a specified direction of discrete point positions to obtain a reference distance sequence. In this embodiment, the specified direction is set to clockwise to ensure the continuity and unidirectional traversal of the sequence. Implementers can set the specified direction according to actual conditions, and it is not limited here. In order to capture each feature node that bulges outward or indents inward, the sum of the number of local maxima and local minima in the reference distance sequence is obtained as the reference quantity. Then, the ratio of the reference quantity to the total arc length of the outer edge contour of the analysis area is used as the contour peak-valley density of the analysis area, which accurately reflects the density of edge bulges and depressions within a unit length of the analysis area. Meanwhile, considering that regenerating nodules on the liver surface can cause drastic abrupt changes in the tangential direction of the edge, in order to reflect the local sharpness of the edge geometry, a cubic spline interpolation algorithm is used to fit the outer edge contour of the analysis area, generating a continuously differentiable fitting curve. The standard deviation of the curvature at each sampling point on the fitting curve is calculated as the contour curvature fluctuation of the analysis area, accurately reflecting the intensity of the edge morphology fluctuation of the analysis area. The cubic spline interpolation algorithm is well known and will not be described in detail here. Furthermore, considering the geometric self-similarity of cirrhotic marginal nodules at different observation scales, a preset number of different preset grid side lengths are set to quantify the spatial occupancy characteristics of this irregular and complex structure. For example, at least three preset grids are set, with the grid side lengths increasing sequentially to 2 pixels, 4 pixels, and 8 pixels to ensure sufficient scale gradients for calculating the fractal dimension. The implementer can set the preset number and grid side lengths according to the image resolution accuracy; no limitation is imposed here. To statistically analyze the spatial volume occupied by the edge at different observation scales, the area containing the analysis region is statistically analyzed separately. The number of grid points on the outer edge contour (i.e., the number of non-empty boxes in the standard box dimension algorithm) is used to transform observation results at different scales into objective quantitative indicators that are stable and executable by computers and to avoid the risk that the limit formula is not computable. Therefore, the least squares method is used to fit the linear slope between the logarithm of the reciprocal of the preset grid side length (as the x value) and the logarithm of the number of grids (as the y value) (i.e., the slope value of the corresponding regression line), which is used as the contour complexity of the analysis area and accurately reflects the fractal roughness of the overall spatial structure of the analysis area. The standard box dimension algorithm and the least squares method are well known and will not be described in detail here. In order to construct a comprehensive and multidimensional regional feature representation matrix, the set consisting of the contour deviation magnitude, contour peak and valley density, contour curvature fluctuation and contour complexity of the analysis area is used as the jagged feature set of the analysis area, providing rich low-level quantitative input features for subsequent comprehensive liver assessment.

[0026] Step S3: Measure the caudate lobe-right lobe ratio based on planar topological relationships, and combine it with macroscopic pathological indicators to obtain the confidence parameters and synergistic contribution values ​​of the caudate lobe-right lobe ratio and each macroscopic pathological indicator. Generate the confidence weight of each analysis region through linear weighting.

[0027] Specifically, considering that the most typical anatomical changes in the compensated stage of cirrhosis are compensatory hypertrophy of the caudate lobe and atrophy and shrinkage of the right lobe, the relative size changes of these two are the key gold standard for assessing the process of liver remodeling. To obtain quantitative evidence of this anatomical change, the caudate lobe-to-right lobe ratio is measured based on planar topological relationships to reflect the degree of liver volume remodeling and the macroscopic anatomical manifestations of cirrhosis. In addition, considering that relying solely on the edge geometry features on two-dimensional sections and the single caudate lobe-to-right lobe ratio is easily affected by image reconstruction errors, artifacts, and local tissue defects, in order to comprehensively verify the true pathological attributes of the edge features from multiple dimensions, various macroscopic pathological indicators pre-received by the system through the data interface are invoked. These include spleen diameter parameters reflecting the degree of portal hypertension (such as the maximum long diameter of the spleen), portal vein signs reflecting collateral circulation or dilation, and hemodynamic parameters reflecting pressure gradients. This provides multi-faceted pathological evidence from anatomical, blood flow, and complication manifestations. To unify various indicators with different dimensions and physical meanings into mathematical probabilities that measure the "possibility of disease," and thus obtain the confidence parameters and synergistic contribution values ​​of the caudate lobe-to-right lobe ratio and various macroscopic pathological indicators, accurately reflecting the degree of certainty of the pathological state indicated by each indicator and when they occur together; furthermore, to generate confidence weights for each analysis region through linear weighting, accurately reflecting the possibility that the serrated features extracted from each analysis region originate from actual cirrhotic fibrosis and regenerative nodules. It should be noted that the caudate lobe-to-right lobe ratio is considered as a single indicator by default.

[0028] Preferably, in one feasible embodiment, the method for obtaining the caudate lobe-right lobe ratio is as follows: A first reference line (serving as the natural anatomical boundary between the caudate lobe and the right lobe) is drawn on the transverse plane, passing through the right side wall of the bifurcation point of the portal vein and parallel to the sagittal plane of the human body, and a second reference line (serving as the abduction pole of caudate lobe hyperplasia) is drawn through the leftmost edge of the caudate lobe and parallel to the sagittal plane of the human body. To quantify the transverse development width of the caudate lobe and the right lobe of the liver on the coronal plane, the vertical distance between the first and second reference lines is calculated as the caudate lobe span value. The vertical distance between the first reference line and the outermost edge of the right lobe of the liver is calculated as the right lobe span value. To eliminate absolute size deviations caused by individual patient body shape differences and obtain standardized anatomical proportion parameters, the caudate lobe span value is divided by the right lobe span value to obtain the caudate lobe-right lobe ratio.

[0029] Preferably, in one feasible manner of this embodiment, the method for obtaining the credibility parameter and collaborative contribution value is as follows: In order to eliminate the huge differences in the units and value ranges of each indicator and to avoid the weight distortion caused by indicators with excessively large values ​​in subsequent calculations, the caudate lobe-right lobe ratio and each macroscopic pathological indicator are normalized respectively to obtain standardized parameters of each indicator that converge to the interval of 0 to 1, and to unify the input dimensions of subsequent calculations; In this embodiment, the above indicators are normalized by the maximum-minimum normalization method. In this maximum-minimum normalization method, the dataset corresponding to each indicator is the sample set of the corresponding indicator in the clinical historical cirrhosis case database. At the same time, in order to avoid the denominator in the maximum-minimum normalization process being 0, the result of adding the denominator in the maximum-minimum normalization process to a preset minimum positive number (such as 0.001, which is not limited here) is used as the denominator. It should be noted that macroscopic pathological indicators include spleen diameter parameters, portal vein sign parameters, and hemodynamic parameters. Among them, portal vein sign parameters are converted into real-valued scalar values ​​based on medical examination findings and recorded accordingly. For example, a value of 0 is assigned when there are no abnormal signs, a value of 1 is assigned when mild varicose veins are diagnosed, and a value of 2 is assigned when severe varicose veins or portal vein thrombosis are diagnosed. The maximum and minimum normalization methods are well-known and will not be elaborated further. To map linearly changing standardized parameters to a nonlinear probability S-curve that aligns with clinical diagnostic intuition—that is, when an indicator's reliability rapidly increases and converges after crossing a certain high-risk threshold—the reliability parameters are calculated by subtracting a preset pathological threshold from the standardized parameters, multiplying the result by a preset adjustment coefficient, and then using the negative of the result as the exponent of the natural constant. The result of this exponentiation is then raised to the power of 1 and the reciprocal is taken to obtain the reliability parameters for each indicator. Therefore, the formula for calculating the reliability parameters is: In the formula, K is the credibility parameter of the indicator; h is the preset adjustment coefficient; h is the standardized parameter; Here, is a preset pathological threshold; e is a natural constant. This embodiment sets a preset pathological threshold. A value of 0.65 is used to ensure precise shifting of the activation center, resulting in a sensitive and reliable response from data at the clinically suspected borderline of cirrhosis. Implementers can set preset pathological thresholds based on epidemiological statistical reference intervals or clinical guidelines for different regions; no specific limitations are imposed here. A preset adjustment coefficient is also set. The value range is between 10 and 20 (e.g., 15) to ensure that the mapping function has an appropriate steepness of discrimination near the threshold. The implementer can set a preset adjustment coefficient according to the tolerance for false positives or false negatives, which is not limited here. Considering that when multiple complication indicators occur simultaneously in clinical practice (such as an abnormal caudate-to-right lobe ratio accompanied by portal hypertension), their significance in indicating the actual progression of cirrhosis far exceeds that of a single indicator. In order to reflect the pathological synergistic enhancement effect between multiple parameters, the confidence parameters of each indicator are multiplied in pairs to obtain the corresponding synergistic contribution value, thereby quantifying the confidence gain when indicators occur synergistically.

[0030] Preferably, in one feasible embodiment, the method for obtaining the confidence weight is as follows: For the fibrotic stress analysis area, considering that this area (i.e., the right lobe of the liver) is the area directly impacted by the right branch of the portal vein and where enterogenic toxins are deposited, its degree of fibrosis is most closely related to hemodynamics and anatomical macroscopic reconstruction. In order to extract the joint characterization with the strongest pathological correlation, the corresponding reference synergistic contribution value (i.e., the combined value obtained by multiplying these three parameters pairwise) is obtained through the confidence parameters of the caudate lobe-right lobe ratio, the portal vein sign parameters, and the hemodynamic parameters. Then, the confidence parameters of the caudate lobe-right lobe ratio, the portal vein sign parameters, and the hemodynamic parameters, along with the largest reference synergistic contribution value, are weighted and summed according to a preset weight allocation ratio to obtain the confidence weight of the fibrotic stress analysis area. This achieves a highly targeted confidence assessment of the severe fibrotic traction that is prone to occur in this area. In this embodiment, the preset weight allocation ratio corresponding to the four weighted items is set as follows: This ensures that anatomical reconstruction features dominate, while other features and synergistic effects are supplementary, with a total weight of 1. Implementers can set a preset weight allocation ratio based on the results of large-sample clinical regression analysis, without any restrictions here. For the compensatory remodeling analysis region, considering that this region (i.e., the caudate lobe) mainly manifests as compensatory hypertrophy, and its compensatory manifestations usually intensify synchronously when splenomegaly is caused by portal hypertension, in order to accurately characterize this compensatory mechanism, the confidence parameters of the caudate lobe-to-right lobe ratio, the confidence parameters of the spleen diameter parameter, and the synergistic contribution values ​​corresponding to the caudate lobe-to-right lobe ratio and the spleen diameter parameter are weighted and summed according to a preset weight allocation ratio to obtain the confidence weight of the compensatory remodeling analysis region, thereby giving the abnormal fluctuations in the compensatory remodeling analysis region a higher weight for determining the pathological authenticity; in this embodiment, the preset weight allocation ratio for the three weighted items here is set as follows: The anatomical characteristics of the caudate lobe-to-right lobe ratio are dominant, while spleen diameter parameters and synergistic effects are auxiliary, and the sum of the weights is 1. The implementer can set the preset weight allocation ratio according to the results of large-sample clinical regression analysis, and no limit is imposed here. For the reference control analysis area, considering that the morphological changes in this area (left lobe) during the sclerosis process are often intermediate or mixed, lacking a single absolute dominant expression, in order to provide a benchmark confidence level that comprehensively reflects systemic pathological pressure, the confidence parameters of all indicators and the maximum value of all synergistic contribution values ​​are weighted and summed according to a preset weight allocation ratio to obtain the confidence weight of the reference control analysis area, providing a unified and unbiased systematic confidence measure for the morphological assessment of the whole liver; in this embodiment, the preset weight allocation ratios for N items (N is the total number of confidence parameters of all indicators plus 1, where 1 represents the maximum value of all synergistic contribution values) are all... This ensures that the feature weights of each dimension of the reference area are balanced and that their sum is 1.

[0031] Step S4: Based on the confidence weight, the serrated feature set is corrected and fused to obtain the whole liver multidimensional feature dictionary and the whole liver comprehensive deviation index, so as to output the quantitative representation of the whole liver edge serration.

[0032] Specifically, considering that purely geometric measurement methods cannot independently distinguish the underlying physical causes of contour mutations, applying the original low-order calculation parameters without constraints can easily lead to biased and misleading assessments. To enhance the persuasiveness and systematic rigor of imaging representations in multiple dimensions, feedback weighting is applied based on the global-level physiological evolution trend. Then, based on the credibility weight, the serrated feature set is corrected and fused to obtain a multidimensional feature dictionary of the whole liver and a comprehensive deviation index of the whole liver. This accurately reflects the multidimensional pathological surface distortion of the whole liver and the overall severity of macroscopic nodulation, and outputs a quantitative representation of the serrated edges of the whole liver. This provides an objective and reliable imaging-assisted decision-making basis for the accurate grading of clinical cirrhosis, monitoring of disease progression, and prediction of decompensation risk.

[0033] Preferably, in one feasible embodiment, the method for obtaining the whole liver multidimensional feature dictionary is as follows: Considering that the contour deviation amplitude, contour peak-valley density, and contour curvature fluctuation can be linearly correlated and reconfigured according to the confidence level of the corresponding region undergoing true irreversible fibrosis, and that contour complexity (i.e., fractal dimension) characterizes the geometric measure topology of the target structure within a finite space, its natural theoretical mathematical lower limit cannot be violated, and arbitrarily assigning scaling coefficients will cause it to lose its originally certain scientific definition of dimension; in order to reasonably utilize the region to objectively reshape the influence scale of the disease course with a high probability, while upholding the mathematical rigor of the standard spatial analysis model, for any analysis region, the confidence weight of the analysis region is respectively correlated with the contour deviation amplitude, contour peak-valley density, and... The contour curvature fluctuations are multiplied by scalars to adjust the contribution ratio of such geometric measures in the overall analysis matrix according to the reliability of regional diagnosis, thus widening and narrowing the fluctuations. At the same time, the contour complexity of the analysis area is selected to be the initial physical measurement value without coefficient contamination, thereby generating an improved matrix that conforms to cross-layer multimodal coordination control, namely the corrected serrated feature set. In order to avoid the dilution of locally significant pathological features (such as the high intensity serrations in the right lobe of the liver) by relatively smooth areas due to simple global summation or averaging, the corrected serrated feature sets of each analysis area are structured and spliced ​​together to form a complete liver multidimensional feature dictionary that retains the feature representation of each anatomical region. This makes the final output data structure take into account both spatial heterogeneity and feature multidimensionality, which is convenient for clinicians to perform targeted regional image reading.

[0034] Preferably, in one feasible embodiment, the method for obtaining the whole-liver comprehensive deviation index is as follows: Considering that although the multidimensional feature dictionary contains rich local microscopic information, for rapid clinical understanding of the overall macroscopic progression of cirrhosis, a single quantitative indicator that is intuitive and representative is needed. In order to provide a macroscopic comprehensive scale for assessing the degree of damage to the surface morphology of the whole liver, the modified contour deviation amplitude of each analysis area is extracted from the modified serrated feature set of each analysis area, because the deviation amplitude most intuitively represents the physical depth of the unevenness of the liver surface and is the most direct manifestation of macroscopic nodulation or local atrophy; in order to reasonably balance the differences in the degree of involvement of different analysis areas in the whole process of cirrhosis, the modified contour deviation amplitude of each analysis area is further... The contour deviation amplitudes are weighted and summed to obtain the whole liver comprehensive deviation index. Specifically, the corrected contour deviation amplitude of each analysis area is multiplied by the preset region fusion weight used to characterize the proportion of importance of each analysis area in the whole liver pathological assessment, and then summed. Considering that the right lobe of the liver (fibrotic stress area) is the most severely remodeled in sclerosis and best reflects the condition, it is given the highest weight. Therefore, in this embodiment, the preset region fusion weights set for the compensatory remodeling analysis area, the fibrotic stress analysis area, and the reference control analysis area are 0.2, 0.6, and 0.2, respectively, to ensure that the sum of the preset region fusion weights is 1, preventing the index from exploding uncontrollably. The implementer can adjust the preset region fusion weights according to the results of large-sample retrospective analysis, which is not limited here.

[0035] In summary, this embodiment acquires abdominal portal venous phase CT images, segments the liver, portal vein, and inferior vena cava regions, extracts the cross-sectional coordinates of the main portal vein bifurcation point and related venous center coordinates, constructs planar topological relationships, and then divides the liver into a compensatory reconstruction analysis area, a fibrosis stress analysis area, and a reference control analysis area on the cross-section, and obtains the serrated feature set of the outer edge contour of each analysis area; measures the caudate lobe-to-right lobe ratio and combines it with macroscopic pathological indicators to obtain the confidence parameters and synergistic contribution values ​​of each indicator, and linearly weights them to generate the confidence weights of each analysis area; based on the confidence weights, the serrated feature set is corrected and fused to output a quantitative representation of the serrations at the entire liver edge. This invention can avoid the dilution of local features, enhance the clinical confidence of edge features, and improve the accuracy of feature extraction.

Claims

1. A method for extracting serrated edges of the liver from CT images of liver cirrhosis, characterized in that, The method includes the following steps: CT images of the abdominal portal vein phase were acquired and segmented to obtain the liver region, portal vein region, and inferior vena cava region. Based on the portal vein region and inferior vena cava region, the cross-section where the main bifurcation point of the portal vein is located, as well as the coordinates of the center of the portal vein and the center of the inferior vena cava on the cross-section, were extracted to construct the planar topological relationship. Based on planar topological relationships, the liver region is divided into a compensatory reconstruction analysis area, a fibrosis stress analysis area, and a reference control analysis area on the cross section; based on the corresponding outer edge contour of each analysis area on the cross section, the serrated feature set of each analysis area is obtained. The caudate lobe-to-right lobe ratio was measured based on planar topological relationships. Combined with macroscopic pathological indicators, the reliability parameters and synergistic contribution values ​​of the caudate lobe-to-right lobe ratio and each macroscopic pathological indicator were obtained. The reliability weights of each analysis region were generated by linear weighting. Based on the credibility weight, the serrated feature set is modified and fused to obtain the whole liver multidimensional feature dictionary and the whole liver comprehensive deviation index, so as to output the quantitative representation of the whole liver edge serration.

2. The method for extracting serrated edges of the liver in CT images of liver cirrhosis as described in claim 1, characterized in that, The method for obtaining the cross-section is as follows: The portal vein region is processed into a skeleton to obtain the portal vein centerline, and the bifurcation nodes where the main trunk and the left and right first-order branches on the portal vein centerline intersect are identified. The axial slice containing the bifurcation node is used as the cross section where the main portal vein bifurcation point is located.

3. The method for extracting serrated edges of the liver in CT images of liver cirrhosis as described in claim 2, characterized in that, The method for obtaining the planar topological relationship is as follows: Extract the inferior vena cava centerline in the inferior vena cava region, and then obtain the intersection points of the portal vein centerline and the inferior vena cava centerline with the cross section in turn. These points are used as the coordinates of the portal vein center and the inferior vena cava center on the cross section in turn. Construct a central line connecting the coordinates of the portal vein center and the inferior vena cava center, and extract the relative positional relationship between the coordinates of the portal vein center and the inferior vena cava center by combining the sagittal and coronal directions of the CT image; A two-dimensional anatomical reference system is constructed on the cross-section using the central connecting line and the relative positional relationship, which serves as the planar topological relationship.

4. The method for extracting serrated edges of the liver in CT images of liver cirrhosis as described in claim 1, characterized in that, The methods for obtaining the compensatory reconstruction analysis region, the fibrosis stress analysis region, and the reference control analysis region are as follows: Based on planar topological relationships, the boundaries of the Couinaud functional segments of the liver are determined on the cross section. The region within the cross section belonging to the caudate lobe of Couinaud 1 segment is designated as the compensatory reconstruction analysis region; The area within the cross-section belonging to the right lobe of the liver, from Couinaud segments 5 to 8, was designated as the fibrosis stress analysis area. The region within the cross-section belonging to the left lobe of the liver, segments 2 to 4 of Couinaud, was designated as a reference control analysis area.

5. The method for extracting serrated edges of the liver in CT images of liver cirrhosis as described in claim 1, characterized in that, The method for obtaining the sawtooth feature set is as follows: For any analysis region, the discrete points on the outer edge contour of the analysis region are smoothed to generate a reference curve. The Euclidean distance between each discrete point and the corresponding point on the reference curve is calculated and used as the reference distance. The average of all reference distances is used as the contour deviation of the analysis region. Arrange the reference distances according to the specified directions of the discrete point positions to obtain a reference distance sequence; obtain the sum of the number of local maxima and local minima in the reference distance sequence as the reference quantity; and take the ratio of the reference quantity to the total arc length of the outer edge contour of the analysis area as the contour peak-valley density of the analysis area. Spline fitting is performed on the outer edge contour of the analysis area, and the standard deviation of the curvature at each sampling point on the fitted curve is calculated as the contour curvature fluctuation of the analysis area. Set a preset number of different preset grid side lengths, count the number of grids containing the outer edge contour points of the analysis area, and fit the linear slope between the logarithm of the reciprocal of the preset grid side length and the logarithm of the number of grids using the least squares method, and use it as the contour complexity of the analysis area. The set consisting of the contour deviation magnitude, contour peak and valley density, contour curvature fluctuation, and contour complexity of the analysis area is taken as the serrated feature set of the analysis area.

6. The method for extracting serrated edges of the liver in CT images of liver cirrhosis as described in claim 1, characterized in that, The method for obtaining the caudate leaf-right leaf ratio is as follows: Draw a first reference line on the cross section that passes through the right side wall of the bifurcation point of the portal vein and is parallel to the sagittal plane of the human body, and a second reference line that passes through the leftmost edge of the caudate lobe and is parallel to the sagittal plane of the human body. Calculate the vertical distance between the first reference line and the second reference line, and use it as the caudate leaf span value; Calculate the vertical distance between the first reference line and the outermost edge of the right lobe of the liver, and use it as the right lobe span value; Divide the caudate leaf span value by the right leaf span value to obtain the caudate leaf-right leaf ratio.

7. The method for extracting serrated edges of the liver in CT images of liver cirrhosis as described in claim 1, characterized in that, The method for obtaining the credibility parameter and collaborative contribution value is as follows: The caudate lobe-to-right lobe ratio and various macroscopic pathological indicators were normalized to obtain standardized parameters for each indicator; among them, macroscopic pathological indicators included spleen diameter parameters, portal vein sign parameters, and hemodynamic parameters. Subtracting the preset pathological threshold from the standardized parameters and multiplying it by the preset adjustment coefficient, the negative number of the result is used as the exponent of the natural constant for exponentiation. The result of exponentiation is then increased by 1 and the reciprocal is taken to obtain the reliability parameters of each indicator. The credibility parameters of each indicator are multiplied in pairs to obtain the corresponding collaborative contribution value.

8. The method for extracting serrated features of liver edges in CT images of liver cirrhosis as described in claim 7, characterized in that, The method for obtaining the credibility weight is as follows: For the fibrotic stress analysis area, the corresponding reference synergistic contribution values ​​are obtained through the confidence parameters of the caudate lobe-to-right lobe ratio, the portal vein sign parameters, and the hemodynamic parameters. The confidence parameters of the caudate lobe-to-right lobe ratio, portal vein sign parameters, and hemodynamic parameters, along with the maximum reference synergistic contribution value, are weighted and summed according to the preset weight allocation ratio to obtain the confidence weight of the fibrotic stress analysis area. For the compensatory reconstruction analysis region, the credibility parameters of the caudate lobe-right lobe ratio and the spleen diameter parameter, as well as the collaborative contribution values ​​corresponding to the caudate lobe-right lobe ratio and the spleen diameter parameter, are weighted and summed according to the preset weight allocation ratio to obtain the credibility weight of the compensatory reconstruction analysis region. For the reference control analysis area, the credibility weight of the reference control analysis area is obtained by weighting and summing the maximum values ​​of the credibility parameters of all indicators and all collaborative contribution values ​​according to the preset weight allocation ratio.

9. The method for extracting serrated features of liver edges in CT images of liver cirrhosis as described in claim 5, characterized in that, The method for obtaining the whole liver multidimensional feature dictionary is as follows: For any analysis region, the confidence weight of the analysis region is multiplied by the contour deviation magnitude, contour peak and valley density and contour curvature fluctuation of the analysis region respectively, and the original value of the contour complexity of the analysis region is retained to generate the corrected jagged feature set of the analysis region. The modified serrated feature sets of each analysis region are structurally spliced ​​and combined to form a complete multidimensional feature dictionary of the whole liver that retains the feature representation of each anatomical region.

10. The method for extracting serrated features of liver edges in CT images of liver cirrhosis as described in claim 9, characterized in that, The method for obtaining the whole liver comprehensive deviation index is as follows: Extract the corrected contour deviation magnitude of each analysis region from the corrected serrated feature set of each analysis region; The weighted summation of the deviation magnitudes of the corrected contours in each analysis region yields the overall liver deviation index.

Images