A left ventricular compliance evaluation method and system based on medical images

Abstract

The present application relates to the technical field of medical image analysis, in particular to a left ventricular compliance evaluation method and system based on medical images, comprising the following steps: obtaining three-dimensional coordinates of the left ventricular apex to the mitral annulus plane through CT images, calling a thin plate spline interpolation algorithm to establish a left ventricular long axis model, synchronously obtaining an end-diastolic diameter, inputting the left ventricular long axis model and the end-diastolic diameter into an Euclidean distance calculation function to generate a left ventricular long diameter ratio. In the present application, the left ventricular long axis model is constructed by integrating CT three-dimensional coordinates and ultrasound inner diameter, the long diameter ratio is calculated in combination with the Euclidean distance, the point cloud is registered and the myocardial wall area is segmented, the local deformation is quantified by using the Gaussian curvature gradient, the correction factor is input into the radial basis function network to generate, and the parameter coupling mechanism is established by fusing the aortic root angle, the dynamic model is established by optimizing the pressure-volume equation weight, and the quantitative evaluation system of spatial heterogeneity and time continuity is realized.

Description

Technical Field

[0001] This invention relates to the field of medical image analysis technology, and in particular to a method and system for assessing left ventricular compliance based on medical images. Background Technology

[0002] The field of medical image analysis technology encompasses a technical system that acquires and quantitatively analyzes information on human anatomical structures and physiological functions through medical imaging technology. Its core content involves key technical aspects such as data acquisition from image acquisition equipment, optimization of image reconstruction algorithms, extraction of anatomical structural features, and calculation of pathophysiological parameters. It focuses on addressing the subjective limitations of traditional qualitative interpretation of medical images, achieving precise quantitative assessment from morphology to function. In the cardiovascular field, it is mainly applied to clinical diagnostic support scenarios such as cardiac chamber volume measurement, myocardial motion tracking, and hemodynamic parameter calculation.

[0003] The left ventricular compliance assessment method refers to a technical solution based on the quantitative analysis of left ventricular diastolic functional characteristics using preoperative medical imaging data. This solution integrates left ventricular end-diastolic diameter and relative wall thickness parameters measured by echocardiography, aortic valve calcification integral and annular diameter data obtained by CT scan, and basic clinical characteristics such as patient age and gender to establish a standardized parameter assignment system. A mathematical model of the left ventricular diastolic pressure-volume relationship is constructed using multivariate regression analysis, ultimately forming a graded and quantifiable left ventricular stiffness assessment index. Its implementation relies on the systematic coupling analysis of medical imaging measurement data and clinical parameters.

[0004] Current technologies rely on single-dimensional analysis of ultrasound inner diameter and wall thickness parameters, lacking three-dimensional modeling of the heart's long-axis morphology using CT three-dimensional spatial coordinates, leading to planarization errors in ventricular geometric remodeling assessment. Traditional methods use single physiological parameters to construct mathematical models, failing to achieve systematic fusion of multi-source image features, and the parameter assignment system is limited by the inherent spatial resolution bias of single-modality imaging. Existing assessment systems use overall ventricular wall thickness indicators, ignoring the curvature gradient differences between the anterior septum and lateral walls, resulting in insufficient sensitivity in assessing local stiffness. The influence of aortic root anatomy on ventricular filling pressure is not included in the computational model, causing systematic bias in diastolic function assessment. Existing regression analysis methods use fixed weighting coefficients, failing to dynamically respond to the cardiac deformation process during diastole, reducing the ability to identify early functional abnormalities. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for assessing left ventricular compliance based on medical imaging.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for assessing left ventricular compliance based on medical imaging, comprising the following steps:

[0007] S1: Obtain the three-dimensional coordinates of the plane from the left ventricular apex to the mitral valve annulus through CT images, call the thin plate spline interpolation algorithm to establish the left ventricular long axis model, simultaneously obtain the end-diastolic diameter of ultrasound, and input the left ventricular long axis model and the end-diastolic diameter of ultrasound into the Euclidean distance calculation function to generate the left ventricular long axis ratio;

[0008] S2: Based on the left ventricular length-to-diameter ratio, an adaptive morphological adjustment process is used to register the CT left ventricular cavity point cloud data. The anterior septum and lateral wall regions are segmented by a region growing algorithm, and the curvature gradient difference value is extracted using a Gaussian curvature calculation model.

[0009] S3: The curvature gradient difference value is calculated as a ratio to the ultrasound interventricular septum thickness and sidewall thickness, and the ratio is input into the radial basis function network to generate a regional stiffness correction factor, while simultaneously obtaining the aortic root tilt angle of CT angiography.

[0010] S4: Input the regional stiffness correction factor, the aortic root tilt angle, and the ultrasound E / e' ratio into the multi-parameter coupled model, and iteratively calculate the weight coefficients of the left ventricular filling pressure equation through the gradient descent optimization algorithm, and output the left ventricular compliance assessment value.

[0011] As a further aspect of the present invention, the left ventricular length-to-diastolic ratio is specifically the geometric ratio of the long axis distance to the end-diastolic diameter; the curvature gradient difference value includes the anterior septal curvature distribution characteristics and the lateral wall curvature distribution characteristics; the regional stiffness correction factor specifically refers to the product of the interventricular septum-lateral wall thickness ratio and the curvature gradient; the aortic root tilt angle includes the coronal projection angle and the sagittal projection angle; and the left ventricular compliance assessment value includes the filling pressure weighting coefficient and the multimodal parameter coupling value.

[0012] As a further embodiment of the present invention, the thin plate spline interpolation algorithm satisfies the condition of continuity of the first derivative at the node;

[0013] The Euclidean distance calculation function is defined as follows:

[0014] ;

[0015] in, The left ventricular long axis model is represented by the first The three-dimensional coordinates of each vertex This represents the standardized spatial coordinates corresponding to the end-diastolic diameter in ultrasound. This represents the total number of vertices in the model.

[0016] In the process of calculating the curvature gradient difference value, the spatial coordinate units of CT point cloud and the ultrasonic thickness measurement value are uniformly converted into millimeter units.

[0017] Before the ratio calculation, a normalization layer is set to eliminate the dimensional differences between the curvature gradient difference value and the interventricular septum thickness and sidewall thickness.

[0018] As a further aspect of the present invention, the step of obtaining the left ventricular length-diameter ratio specifically includes:

[0019] S101: Acquire CT image data, and based on the left ventricular apex and mitral valve annulus plane in the CT image, detect the spatial positioning points of the left ventricular apex and the anterior and posterior mitral valves, organize the three-dimensional coordinates of multiple positioning points, summarize the coordinate data into a spatial sequence, and generate a spatial coordinate sequence.

[0020] S102: Call the thin plate spline interpolation algorithm, perform long axis curve fitting based on multiple three-dimensional coordinate data of the spatial coordinate sequence, establish the three-dimensional structure of the left ventricular long axis, collect the left ventricular internal diameter at the end of diastole by ultrasound, and collect the three-dimensional structure parameters and internal diameter values ​​into a parameter group to generate a three-dimensional structure parameter group.

[0021] S103: Based on the three-dimensional structural parameter group, call the Euclidean distance calculation function to combine the left ventricular long axis model with the ultrasound end-diastolic diameter value to complete the calculation of the Euclidean distance between each parameter and obtain the left ventricular long axis ratio.

[0022] As a further aspect of the present invention, the step of obtaining the curvature gradient difference value specifically includes:

[0023] S201: Based on the left ventricular length-to-diameter ratio, an adaptive morphological adjustment process is used to register the left ventricular cavity point cloud data acquired by CT. According to the three-dimensional spatial distribution of the point cloud, the deformation parameters are matched with the left ventricular length-to-diameter ratio value, the spatial offset between multiple point clouds is adjusted, and a registration spatial coordinate set is generated.

[0024] S202: Based on the registered spatial coordinate set, the region growing algorithm is applied to identify the front gap region and the side wall region, the starting point of the region growing is selected, and the multi-point partition is determined according to the density of the spatial coordinates and the topological structure characteristics. The point cloud is grouped to form a spatial partition, and the partition point cloud set is obtained.

[0025] S203: Based on the aforementioned set of partitioned points, using the Gaussian curvature calculation model, select the three-dimensional spatial neighborhood of multiple points within the partition, analyze the trend of surface normal variation between point sets, and simultaneously perform curvature gradient calculation through the numerical distribution of the angle between adjacent normals between points to obtain the curvature gradient difference value.

[0026] As a further aspect of the present invention, in the Gaussian curvature calculation model, the curvature gradient is associated with the spatial distribution characteristics of the myocardial fiber orientation.

[0027] As a further aspect of the present invention, the step of obtaining S3 specifically comprises:

[0028] S301: Obtain the original data of the curvature gradient difference value, detect the thickness measurement values ​​of the interventricular septum and lateral wall in the ultrasound image, use point-by-point division to calculate the ratio of the curvature gradient difference value to the interventricular septum thickness and lateral wall thickness respectively, establish the interventricular septum thickness ratio parameter and the lateral wall thickness ratio parameter, and fuse the two parameters through a weighted average algorithm to generate the interventricular wall thickness ratio parameter.

[0029] S302: Call the wall thickness ratio parameter as the network input layer node, construct the hidden layer Gaussian kernel function of the radial basis function network, calculate the Euclidean distance between the input parameter and the center point of the hidden layer, activate the hidden layer node through the exponential function, calculate the output layer weight coefficient using the least squares method, perform linear weighted summation operation, and obtain the final output of the radial basis function network as the regional stiffness correction factor.

[0030] S303: Calls DICOM format image data from CT angiography, extracts the contour of the aortic root through a threshold segmentation algorithm, establishes a three-dimensional rectangular coordinate system, calculates the angle between the central axis of the aortic root and the spatial vector of the horizontal plane, and generates the tilt angle of the aortic root.

[0031] As a further aspect of the present invention, the step of obtaining the left ventricular compliance assessment value specifically includes:

[0032] S401: Obtain the regional stiffness correction factor and the aortic root tilt angle, input the ultrasound E / e' ratio into the multi-parameter coupling model, establish the parameter correlation matrix through matrix determinant operation, eliminate collinearity interference between parameters using the least squares method, and generate the initial value of the filling pressure weight.

[0033] S402: Call the initial value of the filling pressure weight, construct the left ventricular filling pressure equation, calculate the partial derivative of the weight coefficient through the gradient descent algorithm, update the weight parameters along the negative gradient direction, terminate the iteration when the sum of squared residuals is less than the convergence threshold, and obtain the optimized value of the weight coefficient.

[0034] S403: Substitute the optimized weight coefficient values ​​into the left ventricular filling pressure equation, calculate the convolution integral of myocardial stress tensor and ventricular volume change rate, extract the principal strain components through eigenvalue decomposition, and output the left ventricular compliance assessment value.

[0035] As a further aspect of the present invention, the myocardial stress tensor is defined as the stress distribution matrix of myocardial tissue in three-dimensional space;

[0036] The rate of change of ventricular volume is the rate of change of left ventricular cavity volume per unit time.

[0037] A medical imaging-based left ventricular compliance assessment system, wherein the medical imaging-based left ventricular compliance assessment system is used to implement the aforementioned medical imaging-based left ventricular compliance assessment method, the system comprising:

[0038] The long axis modeling module is used to obtain the three-dimensional coordinates of the left ventricular apex and mitral valve annulus plane through CT images, call the thin plate spline interpolation algorithm to construct the left ventricular long axis geometric model, simultaneously acquire end-diastolic ultrasound data, input the vertex coordinates of the long axis geometric model and the ultrasound data into the Euclidean distance calculation function to generate the left ventricular long axis ratio, and transfer the left ventricular long axis ratio to the curvature registration module;

[0039] The curvature registration module is used to perform adaptive morphological adjustment registration of the CT left ventricular cavity point cloud data based on the left ventricular length-to-diameter ratio. It segments the anterior septum and lateral wall anatomical regions using a region growing algorithm, extracts the curvature distribution features of the anterior septum and lateral wall using a Gaussian curvature calculation model, calculates the curvature gradient difference value between the two regions, and transmits the curvature gradient difference value to the stiffness correction module.

[0040] The stiffness correction module is used to calculate the ratio of the curvature gradient difference value to the ultrasound interventricular septum thickness and sidewall thickness, input the calculation result into the radial basis function network to generate a regional stiffness correction factor, and simultaneously acquire the aortic root tilt angle measurement value in CT angiography, and transmit the regional stiffness correction factor and aortic root tilt angle to the compliance coupling module.

[0041] The compliance coupling module is used to compensate for the weight of the ultrasound E / e' ratio by the regional stiffness correction factor, construct a multi-parameter coupling equation by combining the aortic root tilt angle, call the gradient descent optimization algorithm to iteratively solve the weight coefficients of early diastolic pressure and atrial systolic pressure in the left ventricular filling pressure equation, and output the quantitative evaluation value of left ventricular compliance.

[0042] The input parameters of the multi-parameter coupling equation correspond one-to-one with the regional stiffness correction factor, aortic root tilt angle, and ultrasound E / e' ratio.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0044] In this invention, by integrating CT three-dimensional coordinates and ultrasound ventricular diameter data, a left ventricular long axis model is constructed using a thin-plate spline interpolation algorithm. The length-to-diameter ratio is calculated using the Euclidean distance function, establishing a quantitative correlation between spatial geometric parameters and functional parameters. Based on an adaptive morphological adjustment process, CT point cloud data is dynamically registered, and a region growing algorithm is used to segment the myocardial wall region. Gaussian curvature gradient differences are used to quantify local deformation characteristics, achieving spatial analysis of myocardial motion heterogeneity. The curvature gradient difference and the ultrasound wall thickness ratio are input into a radial basis function network to generate regional correction factors. Simultaneously, the aortic root tilt angle is fused to construct a multimodal parameter coupling mechanism, enhancing the correlation analysis between anatomical structure and hemodynamics. The weight coefficients of the filling pressure equation are iteratively optimized using a gradient descent algorithm, establishing a dynamic pressure-volume relationship model. This addresses the problem of insufficient characterization of time-varying characteristics by static regression models, forming a quantitative evaluation system that combines spatial heterogeneity and temporal continuity. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the workflow of the present invention;

[0046] Figure 2 This is a flowchart of the steps for obtaining the left ventricular length-diameter ratio according to the present invention;

[0047] Figure 3 This is a flowchart of the steps for obtaining the curvature gradient difference value in this invention;

[0048] Figure 4 This is a flowchart of the acquisition steps in S3 of the present invention;

[0049] Figure 5 This is a flowchart illustrating the steps for obtaining the left ventricular compliance assessment value according to the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0051] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0052] Example 1

[0053] Please see Figure 1 This invention provides a technical solution: a method for assessing left ventricular compliance based on medical imaging, comprising the following steps:

[0054] S1: Obtain the three-dimensional coordinates of the plane from the left ventricular apex to the mitral valve annulus through CT images, call the thin plate spline interpolation algorithm to establish the left ventricular long axis model, simultaneously obtain the end-diastolic diameter of ultrasound, and input the left ventricular long axis model and the end-diastolic diameter of ultrasound into the Euclidean distance calculation function to generate the left ventricular long axis ratio;

[0055] S2: Based on the left ventricular length-to-diameter ratio, an adaptive morphological adjustment process is used to register the CT left ventricular cavity point cloud data. The anterior septum and lateral wall regions are segmented by a region growing algorithm, and the curvature gradient difference value is extracted using a Gaussian curvature calculation model.

[0056] S3: Calculate the ratio of the curvature gradient difference value to the ultrasound interventricular septum thickness and sidewall thickness, input it into the radial basis function network to generate a regional stiffness correction factor, and simultaneously obtain the aortic root tilt angle of CT angiography.

[0057] S4: Input the regional stiffness correction factor, aortic root tilt angle and ultrasound E / e' ratio into the multi-parameter coupled model, and iteratively calculate the weight coefficients of the left ventricular filling pressure equation through the gradient descent optimization algorithm, and output the left ventricular compliance assessment value.

[0058] The left ventricular length-to-diastolic ratio is the geometric ratio of the long axis distance to the end-diastolic diameter. The curvature gradient difference value includes the curvature distribution characteristics of the anterior septum and the curvature distribution characteristics of the lateral wall. The regional stiffness correction factor is the product of the ratio of interventricular septum to lateral wall thickness and the curvature gradient. The aortic root tilt angle includes the coronal projection angle and the sagittal projection angle. The left ventricular compliance assessment value includes the filling pressure weighting coefficient and the multimodal parameter coupling value.

[0059] The thin-plate spline interpolation algorithm satisfies the condition of continuity of the first derivative at the nodes;

[0060] The Euclidean distance calculation function is defined as follows:

[0061] ;

[0062] in, The left ventricular long axis model is represented by the first The three-dimensional coordinates of each vertex This represents the standardized spatial coordinates corresponding to the end-diastolic diameter in ultrasound. This represents the total number of vertices in the model.

[0063] In the process of calculating the curvature gradient difference value, the spatial coordinate units of CT point cloud and the ultrasonic thickness measurement value are uniformly converted to the millimeter dimension;

[0064] Before the ratio calculation, a normalization layer is set to eliminate the dimensional differences between the curvature gradient difference value and the interventricular septum thickness and sidewall thickness.

[0065] Please see Figure 2 The specific steps for obtaining the left ventricular length-diameter ratio are as follows:

[0066] S101: Acquire CT image data, detect spatial positioning points of the left ventricular apex and mitral valve annulus plane based on the left ventricular apex and mitral valve anterior and posterior leaflets in the CT image, organize the three-dimensional coordinates of multiple positioning points, summarize the coordinate data into a spatial sequence, and generate a spatial coordinate sequence.

[0067] To perform this step, firstly, the raw image data of the patient acquired during end-diastolic cardiac CT scans using a 64-slice or higher computed tomography (CT) device is retrieved from the Picture Archiving and Communication System (PACS). This data is in DICOM format, with a slice thickness set to 0.625 mm and a reconstruction matrix of 512x512 pixels to ensure that the image quality meets the requirements for subsequent 3D reconstruction and precise measurement. The image data is then imported into a dedicated medical image post-processing workstation.

[0068] On the post-processing workstation, the multiplanar reconstruction (MPR) function was used to simultaneously observe axial, sagittal, and coronal images of the heart to identify and locate the apex of the left ventricle. The apex of the left ventricle is determined by its appearance as a gradually narrowing apical structure at the distal end of the cardiac chamber in continuous tomographic images. The point at the most distal end of the endocardium at the apex was selected, and its three-dimensional coordinates were recorded. The coordinates of this apex point were determined to be (X:132.5, Y:98.2, Z:215.0), all in millimeters (mm). Subsequently, the mitral valve annulus plane was identified. This plane is the plane between the left atrium and the left ventricular ... In the MPR view, the anatomical boundaries of the mitral valve are clearly identified, showing the positions where the roots of the anterior and posterior leaflets attach to the valve annulus. The spatial locations of the midpoints of the attachment points of the anterior and posterior leaflets are measured, and their three-dimensional coordinates are recorded as (X: 135.8, Y: 120.5, Z: 160.3) mm for the anterior leaflet and (X: 125.2, Y: 123.8, Z: 160.1) mm for the posterior leaflet. To more accurately describe the morphology of the mitral valve annulus, at least four additional anatomical landmarks are selected along the annulus circumference, and their coordinates are also recorded.

[0069] The three-dimensional coordinate data of all detected localization points, including the left ventricular apex and various localization points on the mitral valve annulus, were systematically organized. These coordinate data were then connected sequentially to specific points on the mitral valve annulus according to the logical order of the anatomical structure, starting from the apex, thus forming an ordered spatial coordinate sequence. This sequence is... The ellipsis represents the coordinates of other lobe ring positioning points, thus generating a spatial coordinate sequence for subsequent model construction.

[0070] Table 1. Three-dimensional coordinate data of anatomical localization points of the left ventricle

[0071]

[0072] As shown in Table 1, this table details the three-dimensional coordinates of key anatomical locations of the left ventricle obtained in a single CT image analysis. These data form the basis for subsequent calculations of the left ventricle's long axis length and morphological analysis.

[0073] S102: Call the thin plate spline interpolation algorithm, perform long axis curve fitting based on multiple three-dimensional coordinate data of the spatial coordinate sequence, establish the three-dimensional structure of the left ventricular long axis, collect the left ventricular internal diameter at the end of diastole by ultrasound, and collect the three-dimensional structure parameters and internal diameter values ​​into a parameter group to generate a three-dimensional structure parameter group.

[0074] Based on the spatial coordinate sequence containing the three-dimensional coordinates of multiple positioning points of the left ventricular apex and mitral annulus generated in step S101, a precise reconstruction of the three-dimensional structure of the left ventricular long axis is performed. This process employs thin-plate spline interpolation technology, using each three-dimensional coordinate point in the spatial coordinate sequence as a control point for interpolation. By solving a mathematical problem that minimizes the bending energy of the interpolation function, a smooth three-dimensional curve passing through all control points is generated. This curve is the three-dimensional geometric representation of the left ventricular long axis, accurately reflecting the actual spatial path from the apex to the center of the mitral annulus. The distance from the apex to the center of the mitral annulus on this three-dimensional curve is calculated (this center point is obtained by averaging the coordinates of multiple positioning points on the annulus). For example, by calculating the average of the coordinates of the midpoint of the anterior leaflet attachment point, the midpoint of the posterior leaflet attachment point, auxiliary point 1, and auxiliary point 2 in Table 1, the coordinates of the mitral valve annulus center point are obtained as the arc length between ((135.8+125.2+130.5+128.9) / 4, (120.5+123.8+115.2+128.5) / 4, (160.3+160.1+161.0+159.8) / 4, i.e. (130.10, 122.00, 160.30) mm), and the left ventricular long axis length (L_LAX_model) is obtained. In this case, the calculated L_LAX_model is 98.5 mm.

[0075] Meanwhile, the left ventricular diameter at end-diastole (LVIDd) was acquired from the patient's two-dimensional echocardiogram. In a standard parasternal long-axis view of the left ventricle, based on the peak value of the R wave on the electrocardiogram (representing the end-diastolic phase), the LVIDd value was obtained using electronic calipers of M-mode or two-dimensional ultrasound, perpendicular to the interventricular septum and the inner edge of the posterior wall of the left ventricle. In this case, the LVIDd was measured to be 47.0 mm. This measurement followed the standard method recommended by the American Society of Echocardiography.

[0076] Finally, the three-dimensional structural parameters, namely the calculated left ventricular long axis length L_LAX_model (98.5 mm) and the left ventricular end-diastolic diameter LVIDd measured by ultrasound (47.0 mm), are aggregated to form a parameter set containing these two key physiological parameters. This parameter set is specifically represented as {L_LAX_model:98.5 mm, LVIDd:47.0 mm}, thereby generating a three-dimensional structural parameter set for subsequent calculations.

[0077] S103: Based on the three-dimensional structural parameter set, call the Euclidean distance calculation function to combine the left ventricular long axis model with the ultrasound end-diastolic diameter value to complete the calculation of the Euclidean distance between each parameter and obtain the left ventricular long axis ratio.

[0078] The three-dimensional structural parameter set generated based on step S102 includes the left ventricular long axis length L_LAX_model calculated from the left ventricular long axis model as 98.5 mm, and the left ventricular end-diastolic diameter LVIDd measured by ultrasound as 47.0 mm. The length of the "left ventricular long axis model" here is directly derived from the apical coordinates. and the calculated coordinates of the mitral valve annulus center The spatial distance between them is used for precision, using the coordinates of the left ventricular apex in Table 1. The millimeters, and the coordinates of the mitral annulus center point calculated from the four points on the mitral annulus in Table 1. Millimeters.

[0079] The precise length L_LAX of the left ventricular major axis is calculated using the three-dimensional Euclidean distance formula:

[0080] ;

[0081] Substitute specific values:

[0082] millimeters;

[0083] millimeters;

[0084] millimeters;

[0085] Calculated in millimeters millimeters, this calculation yielded The value (59.70 mm) will replace the initial L_LAX_model value obtained in S102 as a more accurate left ventricular long axis length.

[0086] This precisely calculated left ventricular major axis length The data (59.70 mm) is combined with the acquired end-diastolic ultrasound diameter (LVIDd) (47.0 mm) to prepare for ratio calculation, completing the calculation of the left ventricular length-diastole ratio. This calculation involves taking the length of the left ventricular major axis... Divide by the left ventricular diameter at end-diastole (LVIDd) on ultrasound: Left ventricular length-to-diastole ratio = Substitute the value into / LVIDd: Left ventricular length-diameter ratio = mm / The result calculated in millimeters is approximately 1.27, which is a unitless, pure number that objectively reflects the overall geometric shape of the left ventricle, thus obtaining the left ventricular length-diameter ratio.

[0087] Please see Figure 3 The specific steps for obtaining the curvature gradient difference value are as follows:

[0088] S201: Based on the left ventricular length-to-diameter ratio, an adaptive morphological adjustment process is used to register the left ventricular cavity point cloud data acquired by CT. According to the three-dimensional spatial distribution of the point cloud, the deformation parameters are matched with the left ventricular length-to-diameter ratio values, the spatial offset between multiple point clouds is adjusted, and a registration spatial coordinate set is generated.

[0089] Based on the left ventricular length-diastole ratio obtained in step S103, which is 1.27, and using the three-dimensional point cloud data of the left ventricular cavity wall segmented and extracted from the same CT examination (this point cloud data contains approximately 60,000 XYZ coordinate points, accurately depicting the morphology of the left ventricular cavity at end-diastole), fine registration of the point cloud data was performed. This registration process was achieved through an adaptive morphological adjustment process, rather than simply applying a specific model name. This process first references a left ventricular digital template point cloud with a standard physiological morphology (its inherent length-diastole ratio, for example, 1.60). The left ventricular length-diastole ratio of 1.27 calculated for the current case was compared with the length-diastole ratio of 1.60 of this template. Since the length-diastole ratio of the current case is small (1.27 < 1.60), it indicates that its left ventricle is relatively more spherical, so morphological adjustment of the standard template is required.

[0090] During adjustment, keep the volume of the template point cloud basically constant, and apply a scaling factor along its major axis, which is approximately [value missing]. At the same time, a compensating scaling factor (approximately) is applied in the minor axis (radial) direction. To approximate the volume, the overall length-to-diameter ratio of the adjusted template is made close to 1.27. These non-rigid deformation parameters (i.e., major axis scaling factor 0.79375 and minor axis scaling factor 1.122) are set to match the geometric features of the deformed template with the left ventricular length-to-diameter ratio of the target case of 1.27.

[0091] Subsequently, the template point cloud, after the aforementioned morphological adjustments, is spatially aligned with the actual left ventricular cavity point cloud data extracted from the patient's CT data. This alignment employs a variant of the Iterative Closest Point (ICP) algorithm, which iteratively calculates and applies a rigid transformation matrix containing 3D translation and 3D rotation to minimize the average distance between corresponding point pairs between the two point clouds. Initially, the centroids of the two point clouds may have a spatial offset of approximately 15 mm. After, for example, 20 iterations, an optimal translation vector is calculated (…). millimeters millimeters (millimeters) and rotation matrix (describes the rotation angle about each axis, such as about the X-axis) , around the Y-axis , around the Z-axis This transformation matrix is ​​applied to the patient's original CT point cloud data to correct its position and orientation in three-dimensional space. This effectively adjusts the spatial offset between the template point cloud and the actual CT point cloud data, which are "multi-point clouds". Finally, the coordinates of each point in the patient's left ventricular CT point cloud are updated to a new coordinate system aligned with the morphologically adjusted template, thereby generating a registration spatial coordinate set.

[0092] S202: Based on the registered spatial coordinate set, the region growing algorithm is applied to identify the front gap region and the side wall region. The starting point of the region growing is selected. Based on the density of the spatial coordinates and the topological structure characteristics, the multiple points belong to the partition. The point cloud is grouped to form a spatial partition and the partition point cloud set is obtained.

[0093] Based on the registration spatial coordinate set generated in step S201, this dataset is a three-dimensional point cloud of the left ventricular cavity after precise alignment and morphological matching, containing approximately 60,000 points with corrected (X', Y', Z') coordinates. On this basis, the anterior septum and lateral wall regions within the ventricular cavity are precisely identified and segmented. This process is achieved through region growing. First, the starting point for region growing, i.e., the seed point, is selected. For the identification of the anterior septum region, the operator, referring to the standard anatomical orientation, selects a point in the registration point cloud corresponding to the midpoint of the surface of the interventricular septum facing the left ventricular cavity as the seed point. The coordinates of this point are... In millimeters (in the registered coordinate system), for the lateral wall region, a seed point is selected at the midpoint of the inner surface of the opposite left ventricular free wall, with coordinates as follows: Millimeters.

[0094] Starting from the selected seed point, based on the density of spatial coordinates and local topological characteristics, it is determined whether its neighboring points should belong to the same partition. Regarding the density of spatial coordinates, a neighborhood search radius of 3.5 mm is set. For a candidate neighboring point, if its nearest distance to an existing point in the current growth region is less than a distance threshold, it is considered a potential belonging point. This distance threshold is set with reference to the average point density of the point cloud. It is obtained by locally sampling the point cloud, calculating the average distance between adjacent points, and then multiplying it by a coefficient (such as 0.9), which is set to 1.8 mm. When the distance between any point in the point cloud and its nearest neighbor is less than 0.5 mm, the point is considered a dense point; when the distance is between 0.5 mm and 1.8 mm, it is considered a normal density; when it is greater than 1.8 mm, it is considered a sparse point. Growth prioritizes normal and dense points. To determine the characteristics of the local topology, the influence of adding candidate points on the consistency of the local surface normal vector is calculated. The normal vectors of the candidate points and their neighborhoods are estimated and compared with the average normal vector of the current growth region. If the angle between the two is less than an angle threshold, the topology is considered compatible. This angle threshold is set to 25 degrees. This value is based on the observation of the smooth continuity of the normal cardiac chamber wall to ensure that the growth process does not mistakenly cross the boundary or sharp edge of the anatomical structure.

[0095] In practice, for each seed point, unassigned points within its 3.5 mm neighborhood are checked. If a neighboring point simultaneously satisfies the following conditions: its distance from the nearest point in the current region is less than 1.8 mm, and its addition will not cause a change in the local normal vector exceeding 25 degrees, then the neighboring point is marked as the same partition (anterior septum or lateral wall) as the seed point. This newly added point is then used as a new growth source to continue iteratively expanding outwards until no neighboring point that meets the conditions can be added in the current partition. In this way, the original left ventricular point cloud is effectively divided into a point cloud subset representing the anterior septum region (containing approximately 10,500 points) and a point cloud subset representing the lateral wall region (containing approximately 11,200 points). These two subsets together constitute the spatial partition, thereby obtaining the partitioned point cloud set.

[0096] S203: Based on the partitioned point cloud, using the Gaussian curvature calculation model, select the three-dimensional spatial neighborhood of multiple points within the partition, analyze the trend of surface normal variation between point sets, and simultaneously perform curvature gradient calculation through the numerical distribution of the angle between adjacent normals between points to obtain the curvature gradient difference value.

[0097] Based on the partitioned point cloud obtained in step S202, which includes the point cloud of the front gap region (10,500 points) and the point cloud of the sidewall region (11,200 points), local curvature analysis was performed on the point cloud data within these two partitions. For each point within the partition... First, its three-dimensional spatial neighborhood is selected, which is determined by identifying its nearest neighbor. Defined by neighboring points The value of is set based on the point cloud density and the desired curvature scale; here it is set to... That is, select the 25 nearest points around each point. This constitutes its local point set.

[0098] Then, for each point And its corresponding local point set, by fitting a quadratic surface to the local point set. (in the local coordinate system) (The following) is used to estimate its Gaussian curvature. The Gaussian curvature is calculated from the coefficients of the quadratic term obtained through fitting: The unit is mm. This method allows for the calculation of a Gaussian curvature value for each point in the anterior spacer region and the sidewall region.

[0099] Next, the curvature gradient is calculated. The curvature gradient reflects the rate and direction of change of curvature in space, for a point... Its curvature gradient It is a vector, its size Through calculation Gaussian curvature value and its neighboring points Gaussian curvature value It can be estimated by the ratio of the weighted difference between the two to the distance, or by... Within the neighborhood of the already computed The gradient of a function is obtained by fitting a locally linear function. The estimate is in mm. Calculate the arithmetic mean of the curvature gradient magnitudes at all points in the front gap region to obtain the average front gap curvature gradient. Calculations yielded mm Similarly, the average sidewall curvature gradient was calculated. Calculations yielded mm .

[0100] Finally, the absolute difference between the average curvature gradient values ​​of the two regions is calculated as the curvature gradient difference value: Curvature gradient difference value = Substitute the values: Curvature gradient difference = mm = mm This allows us to obtain the curvature gradient difference value.

[0101] In the Gaussian curvature calculation model, the curvature gradient is associated with the spatial distribution characteristics of myocardial fiber orientation.

[0102] Please see Figure 4 The specific steps to obtain S3 are as follows:

[0103] S301: Obtain the raw data of curvature gradient difference value, detect the thickness measurement values ​​of the interventricular septum and lateral wall in the ultrasound image, use point-by-point division to calculate the ratio of the curvature gradient difference value to the interventricular septum thickness and lateral wall thickness respectively, establish the interventricular septum thickness ratio parameter and the lateral wall thickness ratio parameter, and fuse the two parameters through a weighted average algorithm to generate the interventricular wall thickness ratio parameter.

[0104] Obtain the raw data of the curvature gradient difference values ​​calculated in step S203; its value is... Meanwhile, the end-diastolic thickness measurements of the interventricular septum and lateral wall (or left ventricular posterior wall, referring to the lateral wall) were retrieved from the two-dimensional echocardiogram report of the same patient. These thickness values ​​were obtained by using electronic calipers to accurately measure the distance from the inner edge to the outer edge of the myocardium at the peak of the R wave in the short-axis parasternal section at the level of the papillary muscles. The interventricular septal thickness IVSth was measured to be 14.0 mm and the lateral wall thickness PWth was measured to be 9.5 mm.

[0105] Next, the curvature gradient difference value is calculated as a ratio to the interventricular septum thickness and lateral wall thickness, respectively. First, the interventricular septum thickness ratio parameter (Ratio_IVS) is calculated: Ratio_IVS = Curvature gradient difference value / IVSth. Substituting the values: Ratio_IVS = Then calculate the sidewall thickness ratio parameter (Ratio_PW): Ratio_PW = Curvature gradient difference / PWth. Substitute the value: Ratio_PW = Retain four significant figures. The ratio parameter obtained from these two calculations is the interventricular septal thickness ratio parameter. The ratio parameter of sidewall thickness It was established.

[0106] Finally, by weighted averaging, these two ratio parameters are combined to generate a single wall thickness ratio parameter (WallThicknessRatioParam), calculated as follows: WallThicknessRatioParam = Weighting coefficient and The settings are based on previous clinical studies and expert consensus on the impact of lesions in different ventricular wall regions on overall left ventricular function, ensuring... Considering that the structural and functional characteristics of the interventricular septum are slightly more important than those of the free lateral wall in assessing ventricular stiffness, the following settings were adopted: ,but These weights were determined based on a retrospective analysis of data from 300 patients with different heart conditions. An optimization algorithm was used to find the weight combination that best distinguished between individuals with normal and abnormal diastolic function. Ultimately, a weight of 0.55 for the interventricular septum was determined to be optimal for model performance, with an area under the receiver operating characteristic (AUC) of 0.85. The values ​​were then used for calculation.

[0107] WallThicknessRatioParam= ;

[0108] WallThicknessRatioParam= ;

[0109] WallThicknessRatioParam= Retain four significant figures. This generates the wall thickness ratio parameter. For easier subsequent network input, this parameter is multiplied by... Scale the input to obtain a unitless value. .

[0110] S302: Call the wall thickness ratio parameter as the network input layer node, construct the hidden layer Gaussian kernel function of the radial basis function network, calculate the Euclidean distance between the input parameter and the center point of the hidden layer, activate the hidden layer node through the exponential function, calculate the output layer weight coefficient using the least squares method, perform linear weighted summation operation, and obtain the final output of the radial basis function network as the regional stiffness correction factor.

[0111] The scale-adjusted chamber wall thickness ratio parameter generated in step S301 is called, and its value is... This value serves as the single-node input to a radial basis function (RBF) neural network. The RBF network structure is predefined, with its hidden layers containing three neurons, each using a Gaussian kernel function as its activation function. Gaussian kernel function The mathematical expression is ,in It is the first The central point of each hidden layer neuron These are its width parameters, which (center point and width) are determined based on unsupervised learning (e.g., using K-means clustering to determine the center point) on a training dataset containing 250 cases with known wall thickness ratios and corresponding regional stiffness grades as assessed by clinical evaluation. The width is set according to the multiple of the standard deviation of the sample distribution of each cluster. The parameters were obtained through supervised fine-tuning, ensuring the representativeness and discriminative power of the network. The specific parameter values ​​are shown in the table below.

[0112] Table 2. Parameters of Hidden Layer Neurons in Radial Basis Function Network

[0113]

[0114] As shown in Table 2, this table lists the preset center point and width parameters of the three Gaussian kernel functions of the hidden layer of the RBF network in this embodiment.

[0115] Next, calculate the input parameters. With the center point of each hidden layer neuron The square of the Euclidean distance, i.e. :

[0116] For neurons : ;

[0117] For neurons : ;

[0118] For neurons : ;

[0119] Then, the activation output of each hidden layer neuron is calculated using an exponential function. :

[0120] ;

[0121] ;

[0122] ;

[0123] Output layer weight coefficients During the network training phase, this was determined by applying the least squares method to the aforementioned 250 training datasets. The objective was to minimize the mean square error between the network's predicted regional stiffness correction factor and the actual clinical assessment level, by solving a system of linear equations. (in It is the hidden layer output matrix of the training samples. This is the corresponding target stiffness correction factor vector. (The weight vector to be determined) is obtained. The output layer weight coefficients determined through this process are: , , These weights were cross-validated, and on an independent test set of 50 cases, the Kappa value of the prediction results was 0.78, which is consistent with the expert evaluation, indicating that the model has good generalization ability.

[0124] Finally, a linear weighted summation operation is performed to calculate the network's final output, namely the regional stiffness correction factor. :

[0125] ;

[0126] Substitute the values:

[0127] ;

[0128] Retain three significant figures This generates a regional stiffness correction factor.

[0129] S303: Calls DICOM format image data from CT angiography, extracts the contour of the aortic root through a threshold segmentation algorithm, establishes a three-dimensional rectangular coordinate system, calculates the angle between the central axis of the aortic root and the spatial vector of the horizontal plane, and generates the tilt angle of the aortic root.

[0130] We reviewed the DICOM format images of the patient's thoracic aortic CT angiography (CTA) acquired during the same examination period. This data was obtained using rapid scanning technology after intravenous injection of iodine contrast agent, with a slice thickness of 0.75 mm, clearly displaying the three-dimensional anatomical structure of the aorta and its root. First, to accurately extract the three-dimensional contour of the aortic root, a threshold segmentation method based on Hounsfield units (HU) was used. The aortic lumen exhibits high density due to contrast agent filling, with HU values ​​typically between 200 HU and 550 HU. This HU range was set as the segmentation threshold. Thresholding is used to compare the HU value of each voxel in the CTA image with this threshold range. All voxels with HU values ​​within this range are initially labeled as belonging to the aortic vascular structure. For a voxel with a HU value of 380, it is classified as the aorta, while a voxel with a HU value of 80 is excluded. This initial segmentation result is further optimized by three-dimensional connected region analysis, retaining only the connected regions with the largest volume that are consistent with the anatomical location of the aorta, and supplemented by morphological closing operations (dilation followed by erosion) to fill the internal small holes and smooth the contour edges, thereby obtaining an accurate three-dimensional point cloud contour of the aortic root.

[0131] Based on the extracted aortic root contour, a local three-dimensional Cartesian coordinate system is established for standardized angle measurements. The origin of this coordinate system is... The center of the aortic valve annulus is set at its geometric center. This center is obtained by fitting a plane to the aortic valve annulus contour point set (usually three aortic valve junction points) and calculating the center of the junction ring between this plane and the initial segment of the aortic root. The Z-axis is defined as perpendicular to this aortic valve annulus plane and pointing in the direction of blood flow in the ascending aorta. The X-axis and Y-axis are perpendicular to each other in the annulus plane and are determined according to the standard anatomical orientation (e.g., the X-axis points to the left side of the patient and the Y-axis points to the front of the patient).

[0132] Subsequently, the midline of the aortic root was calculated. Extending upwards along the ascending aorta for 30 mm from the origin of the midline (center of the valve annulus), a cross-sectional profile of the ascending aorta perpendicular to the local Z-axis was obtained every 1.0 mm. The geometric center point of each cross-sectional profile was calculated. Connecting these consecutive center points formed a three-dimensional spatial curve representing the course of the aortic root, which is the midline of the aortic root. The first point on this midline closest to the valve annulus was selected. mm and the point 20 mm from the valve ring Millimeters (all coordinates are in a local coordinate system), thus defining the direction vector of the aortic root midline. .

[0133] Define the horizontal plane during patient scanning (usually corresponding to the plane of the CT scan table). In the standard CT coordinate system, the normal vector of the horizontal plane is usually in the Z-axis direction, i.e. (Assuming the Z-axis of the CT scanner is vertically upward), calculate the vector of the midline of the aortic root. Normal vector of this horizontal plane The angle between This included angle is calculated using the vector dot product formula:

[0134]

[0135] ;

[0136] ;

[0137] so but This angle The angle between the aortic root midline and the vertical direction (the direction of the horizontal plane normal) is the desired aortic root tilt angle, thus generating the aortic root tilt angle as follows: .

[0138] Please see Figure 5 The specific steps for obtaining the left ventricular compliance assessment value are as follows:

[0139] S401: Obtain the regional stiffness correction factor and the aortic root tilt angle, input the ultrasound E / e' ratio into the multi-parameter coupling model, establish the parameter correlation matrix through matrix determinant operation, use the least squares method to eliminate collinearity interference between parameters, and generate the initial value of the filling pressure weight.

[0140] Obtain the regional stiffness correction factor calculated in step S302, with a value of 0.608, and the aortic root tilt angle calculated in step S303, with a value of... Simultaneously, the E / e' ratio was obtained from the patient's echocardiogram report. This ratio was calculated by averaging the peak early diastolic velocity E (measured as E = 0.95 m / s) and the peak early diastolic velocity e' on the septal and lateral sides of the mitral valve annulus (measured as e' = 0.065 m / s, or 6.5 cm / s). First, the units were standardized, converting the E-wave velocity to centimeters per second. Then the E / e' ratio is .

[0141] The three core parameters are: regional stiffness correction factor (Factor=0.608, unitless), aortic root tilt angle (Angle= (unit: degrees), and the ultrasound E / e' ratio (E_e_prime=14.62, unitless), along with other clinical parameters that may have an impact on left ventricular filling pressure, such as patient age (Age=65 years) and left atrial volume index (LAVI=38 ml / m²). ), together serving as a set of input features.

[0142] Table 3 Input Feature Data of Multi-parameter Coupled Model

[0143]

[0144] As shown in Table 3, this table lists the key input features and their specific values ​​used for subsequent model analysis in this embodiment.

[0145] To explore these parameters (in the context of...) The relationship between the five characteristics mentioned above and left ventricular filling pressure (target variable Y, measured using gold standard methods such as right heart catheterization, in mmHg) is established, providing initial weights for the subsequent filling pressure equation. First, a system is constructed containing... Historical cases (e.g.) )of Data Matrix ( (features) and Target vector Calculate the correlation matrix among these features. (yes Matrix, its elements Features and The vector of Pearson correlation coefficients between features and the target variable and the correlation coefficients between features and the target variable. If significant collinearity exists among the features (e.g., the absolute value of the correlation coefficient between two features is greater than 0.8, or collinearity is considered to exist if the variance inflation factor (VIF) is greater than 5), then Ridge Regression is used to estimate the initial weights. Ridge Regression addresses the collinearity problem by adding an L2 regularization term to the loss function of ordinary least squares. Its weight vector... The calculation formula is: ,in It is the result of centralization and standardization Feature matrix, It is the centered target vector. It is the identity matrix. Ridge parameters (regularization strength). The value is selected on the training set through 10-fold cross-validation with the objective of minimizing the mean squared prediction error (MSE). For example, the optimal value is determined through cross-validation. , use The initial weight coefficient vector obtained from the value calculation (only for the first three core parameters) is as follows: , , These are the initial values ​​for the generated filling pressure weights.

[0146] S402: Call the initial value of the filling pressure weight, construct the left ventricular filling pressure equation, calculate the partial derivative of the weight coefficient through the gradient descent algorithm, update the weight parameters along the negative gradient direction, terminate the iteration when the sum of squared residuals is less than the convergence threshold, and obtain the optimized value of the weight coefficient.

[0147] Invoke the initial values ​​of the filling pressure weights for the core parameters generated in step S401. , , And initial weights for other auxiliary parameters (such as age, LAVI) and intercept terms (e.g., , , These are all obtained from ridge regression, and a predictive equation for left ventricular filling pressure is constructed. Its specific form is a linear combination: in It is a vector containing all the weight coefficients to be optimized, with its initial value being... .

[0148] Define a residual function for the nth element in the training dataset. One sample (containing the five input features mentioned above and the actual left ventricular filling pressure measured by the gold standard) ), its residual Defined as the difference between the predicted value and the actual value: The goal of optimization is to minimize all The sum of squared residuals (SSE) of each training sample, i.e., the loss function. .

[0149] The gradient descent algorithm is used to iteratively optimize the weight vector. In each iteration, the loss function is calculated. Relative to each weight coefficient partial derivatives For a linear model, this partial derivative is:

[0150] ;

[0151] in, , , And so on. The weight update rule is:

[0152] ;

[0153] in It is the number of iterations. It's the learning rate, set to a small positive number to ensure convergence. Using line search or empirically set to 0.0005, this value showed good convergence speed and stability in preliminary experiments. The iterative process continued, with each iteration recalculating the predicted filling pressure, residuals, and total residual sum of squares for all training samples using the updated weights, until the residual sum of squares was reached. If the convergence threshold is less than a preset convergence threshold, or if the preset maximum number of iterations (e.g., 500) is reached, the convergence threshold is set to... This threshold is based on historical model training experience. When the change in SSE is less than this value, further iterations have little effect on improving model performance. In this case, after 185 iterations, SSE reached [a certain value]. The iteration terminates when the convergence condition is met. The resulting weight vector is the optimized weight coefficient. For example, the final optimized core parameter weights are: , , (Other weights are also updated accordingly) to obtain the optimized weight coefficient values.

[0154] S403: Substitute the optimized weighting coefficients into the left ventricular filling pressure equation, calculate the convolution integral of myocardial stress tensor and ventricular volume change rate, extract the principal strain components through eigenvalue decomposition, and output the left ventricular compliance assessment value.

[0155] The core parameter weight coefficients obtained in step S402 are optimized ( , , ) and other parameter optimization weights (e.g., , , Substituting into the established left ventricular filling pressure equation, for the input characteristics of the current case (Factor=0.608, Angle= E_e_prime=14.62, Age=65 years old, LAVI=38 ml / m ), calculate the estimated value of its left ventricular end-diastolic filling pressure. :

[0156] ;

[0157] ;

[0158] mmHg;

[0159] Approximately 21.5 mmHg. As an important parameter for subsequent evaluation.

[0160] Next, the dynamic characteristics of ventricular diastole are assessed, which involves the myocardial stress tensor and the rate of change of ventricular volume. It describes how the myocardium withstands load during diastole. Symmetric tensors, their components (such as circumferential stress) Longitudinal stress radial stress (And shear stress) is calculated using a finite element model (FEM) combined with individualized left ventricular geometry (from CT 3D reconstruction) and the above-mentioned left ventricular intracavitary pressure. (As internal boundary conditions) and passive material properties of the myocardium (e.g., material constants of the Mooney-Rivlin model, C1 = 1.5 kPa, C2 = 0.5 kPa, obtained through literature references or fitting of isolated myocardial stretching experimental data), the calculated mean circumferential wall stress at end-diastole was 18.0 kPa, and the longitudinal wall stress was 14.5 kPa. Ventricular volume change rate Left ventricular volume throughout the entire cardiac cycle can be obtained using time-resolved 3D ultrasound or cardiac MRI. The curve is obtained by numerical differentiation during the rapid filling period. It can reach 450 ml / second; in late diastole (atrial systole). Approximately 150 ml / second.

[0161] Subsequently, an index characterizing overall ventricle-vascular coupling or energy transfer is calculated, which is expressed by myocardial stress (or some scalar representation thereof, such as mean wall stress). ) and ventricular volume change rate Throughout the diastolic period (from the opening of the mitral valve to the end of diastole, the time interval) Convolution integral To achieve: At discrete time points, this integral is approximately a summation. ,in It is the time step. Calculations yielded The value is 2500 kPa ml / s Second = 2500 kPa Milliliters. This integral value reflects the energy state during the filling process.

[0162] Meanwhile, the end-diastolic myocardial strain tensor calculated using FEM Extract the principal strain components from the strain tensor (derived from the stress tensor through the constitutive relation of the myocardium) and apply them to... Perform eigenvalue decomposition to solve the given equation. Three principal strain values ​​were obtained. These represent the maximum deformation of the myocardium in three mutually perpendicular principal directions. In this case, the end-diastolic principal strain (expressed as a unitless decimal representing the ratio of tension or compression) is: circumferential strain (Shortened by 18%), longitudinal strain (Reduced by 15%), radial strain (Thickened by 35%).

[0163] Ultimately, the Left Ventricular Compliance Assessment (LVCI) is a comprehensive indicator, calculated using a predefined empirical formula or machine learning model, combined with the left ventricular end-diastolic filling pressure obtained above. Convolution integral value and principal strain components (e.g., taking the overall volumetric strain) Or a certain combination of principal strains, such as Output a quantitative compliance score, for example, defining LVCI as: LVCI= in, For example, empirical coefficients. , Assuming the end-diastolic volume is 120 mL and the minimum early-diastolic volume is 50 mL, then mL. LVCI= First, To convert from mmHg to kPa, use the conversion rule: 1 mmHg 0.133322 kPa. kPa.

[0164] LVCI= ;

[0165] LVCI= ;

[0166] LVCI= mL / kPa. mL / kPa is the output left ventricular compliance assessment value. The lower the value, the worse the left ventricular compliance. For example, normal left ventricular compliance values ​​are usually in the range of 40-60 mL / kPa, while values ​​below 30 mL / kPa indicate a significant decrease in compliance.

[0167] The myocardial stress tensor is defined as the stress distribution matrix of myocardial tissue in three-dimensional space;

[0168] The rate of change of ventricular volume is the rate of change of left ventricular cavity volume per unit time.

[0169] A medical imaging-based left ventricular compliance assessment system is provided for performing the aforementioned medical imaging-based left ventricular compliance assessment method. The system includes:

[0170] The long axis modeling module is used to obtain the three-dimensional coordinates of the left ventricular apex and mitral valve annulus plane through CT images, call the thin plate spline interpolation algorithm to construct the left ventricular long axis geometric model, simultaneously acquire end-diastolic ultrasound data, input the vertex coordinates of the long axis geometric model and the ultrasound data into the Euclidean distance calculation function to generate the left ventricular long axis ratio, and transfer the left ventricular long axis ratio to the curvature registration module.

[0171] The curvature registration module is used to perform adaptive morphological adjustment registration of the CT left ventricular cavity point cloud data by using the left ventricular length-to-diameter ratio. It segments the anterior septum and lateral wall anatomical regions by using a region growing algorithm, extracts the curvature distribution features of the anterior septum and lateral wall using a Gaussian curvature calculation model, calculates the curvature gradient difference value between the two regions, and passes the curvature gradient difference value to the stiffness correction module.

[0172] The stiffness correction module is used to calculate the ratio of the curvature gradient difference value to the ultrasound interventricular septum thickness and sidewall thickness. The calculation result is input into the radial basis function network to generate a regional stiffness correction factor. Simultaneously, the measurement value of the aortic root tilt angle in CT angiography is obtained, and the regional stiffness correction factor and the aortic root tilt angle are transmitted to the compliance coupling module.

[0173] The compliance coupling module is used to compensate for the weight of the ultrasound E / e' ratio by the regional stiffness correction factor, construct a multi-parameter coupling equation by combining the aortic root tilt angle, call the gradient descent optimization algorithm to iteratively solve the weight coefficients of early diastolic pressure and atrial systolic pressure in the left ventricular filling pressure equation, and output the quantitative assessment value of left ventricular compliance.

[0174] The input parameters of the multi-parameter coupling equation correspond one-to-one with the regional stiffness correction factor, the aortic root tilt angle, and the ultrasound E / e' ratio.

[0175] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for assessing left ventricular compliance based on medical imaging, characterized in that, Includes the following steps: S1: Obtain the three-dimensional coordinates of the plane from the left ventricular apex to the mitral valve annulus through CT images, call the thin plate spline interpolation algorithm to establish the left ventricular long axis model, simultaneously obtain the end-diastolic diameter of ultrasound, and input the left ventricular long axis model and the end-diastolic diameter of ultrasound into the Euclidean distance calculation function to generate the left ventricular long axis ratio; The steps for obtaining the left ventricular length-diameter ratio are as follows: S101: Acquire CT image data, and based on the left ventricular apex and mitral valve annulus plane in the CT image, detect the spatial positioning points of the left ventricular apex and the anterior and posterior mitral valves, organize the three-dimensional coordinates of multiple positioning points, summarize the coordinate data into a spatial sequence, and generate a spatial coordinate sequence. S102: Call the thin plate spline interpolation algorithm, perform long axis curve fitting based on multiple three-dimensional coordinate data of the spatial coordinate sequence, establish the three-dimensional structure of the left ventricular long axis, collect the left ventricular internal diameter at end-diastole by ultrasound, and collect the parameters and internal diameter values ​​of the three-dimensional structure of the left ventricular long axis into a parameter group to generate a three-dimensional structure parameter group. S103: Based on the three-dimensional structural parameter group, call the Euclidean distance calculation function to combine the left ventricular long axis model with the ultrasound end-diastolic diameter value to complete the calculation of the Euclidean distance between each parameter and obtain the left ventricular long axis ratio, wherein the left ventricular long axis ratio is specifically the geometric ratio of the long axis distance to the end-diastolic diameter. S2: Based on the left ventricular length-diameter ratio, an adaptive morphological adjustment process is used to register the CT left ventricular cavity point cloud data. The anterior septum and lateral wall regions within the cardiac chamber are segmented using a region growing algorithm. The curvature gradient difference value is extracted using a Gaussian curvature calculation model. The specific steps for obtaining the curvature gradient difference value are as follows: S201: Based on the left ventricular length-diameter ratio, an adaptive morphological adjustment process is used to register the left ventricular cavity point cloud data acquired by CT, generating a registration spatial coordinate set. The adaptive morphological adjustment process is to compare the left ventricular length-diameter ratio calculated for the current case with the left ventricular digital template point cloud, with a reference to a left ventricular digital template point cloud with a standard physiological morphology. S202: Based on the registered spatial coordinate set, the region growing algorithm is applied to identify the anterior septum region and lateral wall region within the cardiac chambers. The starting point of the region growing is selected. Based on the density of the spatial coordinates and the topological characteristics, the multiple points are assigned to a partition. The point cloud is grouped to form a spatial partition, and the partition point cloud set is obtained. S203: Based on the partitioned point cloud, using the Gaussian curvature calculation model, select the three-dimensional spatial neighborhood of multiple points within the partition, analyze the trend of surface normal change between point sets, and simultaneously perform curvature gradient calculation through the numerical distribution of the angle between adjacent normals between points to obtain the curvature gradient difference value. S3: The curvature gradient difference value is calculated as a ratio to the ultrasound interventricular septum thickness and sidewall thickness, and the ratio is input into the radial basis function network to generate a regional stiffness correction factor, while simultaneously obtaining the aortic root tilt angle of CT angiography. The specific steps for obtaining S3 are as follows: S301: Obtain the original data of the curvature gradient difference value, detect the thickness measurement values ​​of the interventricular septum and lateral wall in the ultrasound image, use point-by-point division to calculate the ratio of the curvature gradient difference value to the interventricular septum thickness and lateral wall thickness respectively, establish the interventricular septum thickness ratio parameter and the lateral wall thickness ratio parameter, and fuse the two parameters through a weighted average algorithm to generate the interventricular wall thickness ratio parameter. S302: Call the wall thickness ratio parameter as the network input layer node, construct the hidden layer Gaussian kernel function of the radial basis function network, calculate the Euclidean distance between the input parameter and the center point of the hidden layer, activate the hidden layer node through the exponential function, calculate the output layer weight coefficient using the least squares method, perform linear weighted summation operation, and obtain the final output of the radial basis function network as the regional stiffness correction factor. S303: Call the DICOM format image data of CT angiography, extract the contour of the aortic root through the threshold segmentation algorithm, establish a three-dimensional rectangular coordinate system, calculate the angle between the spatial vector of the aortic root midline and the horizontal plane, and generate the aortic root tilt angle. S4: Input the regional stiffness correction factor, the aortic root tilt angle, and the ultrasound E / e' ratio into the multi-parameter coupled model, iteratively calculate the weight coefficients of the left ventricular filling pressure equation through the gradient descent optimization algorithm, and output the left ventricular compliance assessment value; The specific steps for obtaining the left ventricular compliance assessment value are as follows: S401: Establish a parameter correlation matrix based on the regional stiffness correction factor, the aortic root tilt angle, and the ultrasound E / e' ratio. Use the least squares method to eliminate collinearity interference between parameters and generate initial values ​​for filling pressure weights, where E is the peak velocity of early diastolic blood flow in the mitral valve, and e' is the peak velocity of early diastolic motion on the mitral valve annulus septum side and lateral wall side. S402: Call the initial value of the filling pressure weight, construct the left ventricular filling pressure equation, calculate the partial derivative of the weight coefficient through the gradient descent algorithm, update the weight parameters along the negative gradient direction, terminate the iteration when the sum of squared residuals is less than the convergence threshold, and obtain the optimized value of the weight coefficient. S403: Substitute the optimized weight coefficient values ​​into the left ventricular filling pressure equation, calculate the convolution integral of myocardial stress tensor and ventricular volume change rate, extract the principal strain components through eigenvalue decomposition, and output the left ventricular compliance assessment value.

2. The method for assessing left ventricular compliance based on medical imaging according to claim 1, characterized in that, The left ventricular length-to-diastolic ratio is specifically the geometric ratio of the long axis distance to the end-diastolic diameter. The curvature gradient difference value includes the anterior septal curvature distribution characteristics and the lateral wall curvature distribution characteristics. The regional stiffness correction factor specifically refers to the product of the interventricular septum to lateral wall thickness ratio and the curvature gradient. The aortic root tilt angle includes the coronal projection angle and the sagittal projection angle. The left ventricular compliance assessment value includes the filling pressure weighting coefficient and the multimodal parameter coupling value.

3. The method for assessing left ventricular compliance based on medical imaging according to claim 1, characterized in that, The thin-plate spline interpolation algorithm satisfies the condition of continuity of the first derivative at the nodes; The Euclidean distance calculation function is defined as follows: ； in, The left ventricular long axis model is represented by the first The three-dimensional coordinates of each vertex This represents the standardized spatial coordinates corresponding to the end-diastolic diameter in ultrasound. This represents the total number of vertices in the model. In the process of calculating the curvature gradient difference value, the spatial coordinate units of CT point cloud and the ultrasonic thickness measurement value are uniformly converted into millimeter units. Before the ratio calculation, a normalization layer is set to eliminate the dimensional differences between the curvature gradient difference value and the interventricular septum thickness and sidewall thickness.

4. The method for assessing left ventricular compliance based on medical imaging according to claim 1, characterized in that, In the Gaussian curvature calculation model, the curvature gradient is associated with the spatial distribution characteristics of myocardial fiber orientation.

5. The method for assessing left ventricular compliance based on medical imaging according to claim 1, characterized in that, The myocardial stress tensor is defined as the stress distribution matrix of myocardial tissue in three-dimensional space; The rate of change of ventricular volume is the rate of change of left ventricular cavity volume per unit time.

6. A left ventricular compliance assessment system based on medical imaging, characterized in that, The system is used to implement the left ventricular compliance assessment method based on medical imaging as described in any one of claims 1-5, and the system comprises: The long axis modeling module obtains the three-dimensional coordinates of the plane from the left ventricular apex to the mitral valve annulus through CT images, calls the thin plate spline interpolation algorithm to establish the left ventricular long axis model, and simultaneously obtains the end-diastolic diameter of ultrasound. The left ventricular long axis model and the end-diastolic diameter of ultrasound are input into the Euclidean distance calculation function to generate the left ventricular long axis ratio. The curvature registration module, based on the left ventricular length-to-diameter ratio, uses an adaptive morphological adjustment process to register the CT left ventricular cavity point cloud data, segments the anterior septum and lateral wall regions within the cardiac chambers using a region growing algorithm, and extracts the curvature gradient difference value using a Gaussian curvature calculation model. The stiffness correction module calculates the ratio of the curvature gradient difference value to the ultrasound interventricular septum thickness and sidewall thickness, inputs it into the radial basis function network to generate a regional stiffness correction factor, and simultaneously obtains the aortic root tilt angle of CT angiography. The compliance coupling module inputs the regional stiffness correction factor, the aortic root tilt angle, and the ultrasound E / e' ratio into a multi-parameter coupling model, iteratively calculates the weight coefficients of the left ventricular filling pressure equation through a gradient descent optimization algorithm, and outputs the left ventricular compliance assessment value.

Images