Congenital heart disease four-dimensional echocardiogram dynamic segmentation system

Abstract

The invention relates to the field of medical image processing, in particular to a congenital heart disease four-dimensional echocardiogram dynamic segmentation system, which acquires four-dimensional echocardiogram data of a fetal heart through an image acquisition module, corrects fetal movement artifacts by an optical flow field correction module by using an optical flow field time sequence analysis technology, improves the image quality, and improves the segmentation accuracy of the fetal movement artifacts. The heart structure time sequence alignment module is used for aligning a heart structure and extracting a ventricular septal defect area through a multi-scale similarity transformation strategy, the defect analysis module is used for calculating the edge line diameter and the shunt direction of a defect part and generating defect parameter information, and the result output module is used for generating a congenital heart disease screening report. And a foundation is laid for subsequent analysis, and the defect part is kept at a stable position in a time sequence by accurately aligning the heart structure, so that accurate analysis is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a four-dimensional echocardiogram dynamic segmentation system for congenital heart disease, which can be applied to the automated screening and analysis of fetal congenital heart disease in prenatal diagnosis. Background Art

[0002] Congenital heart disease (CHD) is the most common congenital malformation in newborns, with an incidence of approximately 8‰. Ventricular septal defect is one of the most common types of CHD. Prenatal ultrasound diagnosis is an important means of screening for CHD, and four-dimensional echocardiography allows doctors to observe the spatial structure and temporal changes of the fetal heart, providing a crucial basis for CHD diagnosis.

[0003] However, existing fetal heart ultrasound image analysis methods face numerous challenges: First, image artifacts caused by fetal movement severely impact diagnostic quality; second, the dynamic changes in cardiac structure during contraction and relaxation make image alignment difficult; and third, edge detection and measurement of defects have long relied on manual labor by physicians, which is highly subjective and inefficient. This is especially true during the first and second trimesters, when the fetal heart is smaller and more active, further complicating accurate detection.

[0004] While some automated segmentation algorithms have been applied to echocardiographic analysis, most are designed for adult hearts and are difficult to directly apply to fetal heart analysis. Furthermore, most existing methods fail to adequately address artifacts caused by fetal movement and lack effective alignment strategies for temporal changes in cardiac structure, resulting in unstable diagnostic results. Furthermore, existing technologies have limited accuracy in automated measurement of defect locations, particularly for complex ventricular septal defects, making it difficult to provide accurate diameter and shunt direction information.

[0005] Therefore, there is an urgent need to develop an intelligent system that can effectively handle fetal movement artifacts, accurately align cardiac structures, and accurately segment and measure defect sites, so as to improve the accuracy and efficiency of prenatal screening for congenital heart disease. Summary of the Invention

[0006] The purpose of the present invention is to provide a four-dimensional echocardiographic dynamic segmentation system for congenital heart disease. Through innovative image processing and artificial intelligence technologies, it solves technical problems such as fetal movement artifact correction, cardiac structure timing alignment, and precise measurement of defect sites, thereby achieving early and accurate diagnosis of congenital heart disease.

[0007] The present invention proposes a four-dimensional echocardiographic dynamic segmentation system for congenital heart disease, comprising:

[0008] an image acquisition module for acquiring four-dimensional echocardiographic data of the fetal heart;

[0009] an optical flow correction module, connected to the image acquisition module, configured to receive the four-dimensional echocardiographic data, correct fetal movement artifacts based on an optical flow time series analysis technique, and generate corrected four-dimensional echocardiographic data;

[0010] a cardiac structure time series alignment module, connected to the optical flow field correction module, configured to receive the corrected four-dimensional echocardiographic data, align the cardiac structure using a multi-scale similarity transformation strategy, extract the ventricular septal defect region, and generate aligned cardiac structure data;

[0011] a defect analysis module connected to the cardiac structure timing alignment module, configured to receive the aligned cardiac structure data, calculate the defect edge line diameter and shunt direction, and generate defect parameter information;

[0012] The result output module is connected to the defect analysis module and is used to receive the defect parameter information and generate a congenital heart disease screening report.

[0013] Preferably, the cardiac structure timing alignment module includes:

[0014] a multi-scale pyramid construction unit, configured to construct a multi-level pyramid structure for the received four-dimensional echocardiographic data, wherein each level contains a heart image of a different resolution;

[0015] a feature extraction unit, connected to the multi-scale pyramid construction unit, for extracting cardiac structural features in different scale spaces;

[0016] A feature fusion unit, connected to the feature extraction unit, for weighted integration of features at different levels to generate a multi-scale feature description;

[0017] a model matching unit, connected to the feature fusion unit, for retrieving the most similar reference model from a pre-established cardiac structure model library based on the multi-scale feature description;

[0018] A non-rigid transformation unit is connected to the model matching unit and is used to perform non-rigid transformation based on the reference model to align the heart structure.

[0019] Preferably, the non-rigid transformation unit comprises:

[0020] a global transform processor for performing an overall coarse alignment of the cardiac structures;

[0021] A local transformation processor, connected to the global transformation processor, configured to perform local fine transformation based on global alignment;

[0022] An iterative optimization processor is connected to the local transformation processor and is used to adjust the transformation parameters through multiple iterations until a preset alignment quality threshold is reached.

[0023] Preferably, the optical flow field correction module includes:

[0024] an optical flow field estimation unit, configured to estimate the motion field of the fetal heart by analyzing image changes between adjacent ultrasound frames;

[0025] a motion separation unit, connected to the optical flow field estimation unit, for separating fetal movement from the heart's own movement;

[0026] The artifact elimination unit is connected to the motion separation unit and is used to eliminate image artifacts caused by fetal movement based on the separated motion information.

[0027] Preferably, the defect analysis module includes:

[0028] an edge detection unit, configured to identify an edge of a ventricular septal defect in the aligned cardiac structure data;

[0029] a size measurement unit, connected to the edge detection unit, and configured to calculate the diameter of the ventricular septal defect;

[0030] a shunt direction analysis unit connected to the edge detection unit, configured to determine the shunt direction of the defect based on the morphological characteristics and blood flow information of the ventricular septal defect;

[0031] The severity assessment unit is connected to the size measurement unit and the shunt direction analysis unit, and is used to comprehensively analyze defect parameters and assess the severity of congenital heart disease.

[0032] Preferably, the model matching unit includes:

[0033] A feature vector generator, used to convert the current ultrasound sequence into a feature vector;

[0034] A similarity calculator, connected to the feature vector generator, for calculating the similarity between the current feature vector and the reference model stored in the model library;

[0035] The best matching selector is connected to the similarity calculator and is used to select the reference model with the highest similarity as the matching result.

[0036] As an option, it also includes:

[0037] The spatiotemporal fusion network module is arranged between the cardiac structure timing alignment module and the defect analysis module, and is used to integrate the features of the aligned cardiac structure data in the time dimension and the space dimension to enhance the feature expression of the defect area.

[0038] Preferably, the spatiotemporal fusion network module includes:

[0039] The time series channel is used to process the time series characteristics of dynamic changes of the heart;

[0040] Spatial channel, used to process the spatial geometric characteristics of cardiac structures;

[0041] The fusion layer is connected to the temporal channel and the spatial channel, and is used to integrate temporal and spatial features to generate an enhanced defect area expression.

[0042] Preferably, the result output module includes:

[0043] a defect visualization unit, configured to convert the defect parameter information into an intuitive graphical representation;

[0044] a report generating unit connected to the defect visualization unit, for generating a congenital heart disease screening report including diagnosis results and recommendations based on the defect parameter information and the graphical representation;

[0045] The data storage unit is connected to the report generating unit and is used to store the congenital heart disease screening report and related analysis data.

[0046] As an option, it also includes:

[0047] The adaptive model updating module is connected to the cardiac structure timing alignment module and is used to update the reference model in the cardiac structure model library based on new analysis cases to improve the system's adaptability to different fetal physiological characteristics.

[0048] The beneficial effects of the present invention include:

[0049] 1. Through optical flow field timing correction technology, image artifacts caused by fetal movement are effectively eliminated, image quality is improved, and the foundation for subsequent analysis is laid;

[0050] 2. A multi-scale similarity transformation strategy is used to achieve precise alignment of cardiac structures, ensuring that the defect remains in a stable position in the time series, facilitating accurate analysis.

[0051] 3. Combined with spatiotemporal fusion network technology, it enhances the feature expression of defective areas and improves segmentation accuracy;

[0052] 4. Automatically calculates the diameter of the defect margin and the direction of shunt flow, increasing detection efficiency by 15 times compared to manual annotation and supporting congenital heart disease screening as early as 16 weeks of gestation;

[0053] 5. Continuously optimize system performance through an adaptive model library to improve adaptability to different fetal physiological characteristics.

[0054] In summary, the present invention significantly improves the accuracy, efficiency and early screening capability of prenatal diagnosis of congenital heart disease, and has important clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is the overall structural block diagram of the congenital heart disease four-dimensional echocardiography dynamic segmentation system of the present invention;

[0056] Figure 2 This is a structural block diagram of the cardiac structure timing alignment module of the present invention;

[0057] Figure 3 This is a structural block diagram of the optical flow field correction module of the present invention;

[0058] Figure 4 This is a structural block diagram of the defect analysis module of the present invention;

[0059] Figure 5 This is a structural block diagram of the spatiotemporal fusion network module of the present invention;

[0060] Figure 6 Schematic diagram of the multi-scale pyramid structure of the present invention;

[0061] Figure 7 Schematic diagram of the non-rigid transformation optimization process of the present invention;

[0062] Figure 8 This is an example diagram of the ventricular septal defect measurement results of the present invention. DETAILED DESCRIPTION

[0063] Please refer to the attached Figure 1-8 , the specific implementation of the present invention is further described in detail below with reference to the accompanying drawings.

[0064] See also Figure 1 The four-dimensional echocardiographic dynamic segmentation system for congenital heart disease provided by the present invention includes an image acquisition module 1, an optical flow field correction module 2, a cardiac structure timing alignment module 3, a defect analysis module 4, a result output module 5, an adaptive model updating module 6 and a spatiotemporal fusion network module 7.

[0065] Image acquisition module 1 is used to acquire four-dimensional echocardiographic data of the fetal heart. In one embodiment of the present invention, the four-dimensional echocardiographic data acquired by this module typically has an acquisition frequency of 25 frames per second and a spatial resolution of 224×224×96 voxels, which can clearly capture the three-dimensional structure of the fetal heart and its four-dimensional information that changes over time.

[0066] Optical flow correction module 2, connected to image acquisition module 1, receives the 4D echocardiographic data and corrects fetal movement artifacts using optical flow time-series analysis techniques to generate corrected 4D echocardiographic data. This module addresses the problem of fetal movement interfering with ultrasound image quality, laying the foundation for subsequent accurate analysis.

[0067] The cardiac structure temporal alignment module 3 is connected to the optical flow correction module 2. It receives the corrected 4D echocardiographic data and uses a multiscale similarity transformation strategy to align the cardiac structures, extract the ventricular septal defect region, and generate aligned cardiac structural data. This module is the core innovation of the system and solves the structural alignment difficulties caused by dynamic changes in the heart.

[0068] The defect analysis module 4 is connected to the cardiac structure timing alignment module 3 and is used to receive the aligned cardiac structure data, calculate the defect edge diameter and shunt direction, and generate defect parameter information. This module realizes the automated and accurate measurement of ventricular septal defects.

[0069] The result output module 5 is connected to the defect analysis module 4 and is used to receive defect parameter information and generate a congenital heart disease screening report. This module converts the analysis results into diagnostic information that can be understood by clinicians.

[0070] Preferably, the system also includes a spatiotemporal fusion network module 7, which is arranged between the cardiac structure timing alignment module 3 and the defect analysis module 4, and is used to integrate the features of the aligned cardiac structure data in the time dimension and the space dimension to enhance the feature expression of the defect area.

[0071] In addition, the system also includes an adaptive model updating module 6, which is connected to the cardiac structure timing alignment module 3 and is used to update the reference model in the cardiac structure model library based on new analysis cases to improve the system's adaptability to different fetal physiological characteristics.

[0072] The specific implementation of each module is described in detail below.

[0073] See also Figure 3 The optical flow field correction module 2 includes an optical flow field estimation unit 21, a motion separation unit 22 and an artifact elimination unit 23.

[0074] The optical flow field estimation unit 21 is used to estimate the motion field of the fetal heart by analyzing the image changes between adjacent ultrasound frames. Specifically, the unit calculates the motion vector of each pixel in the ultrasound image sequence based on the optical flow principle.

[0075] In the implementation process, the optical flow field estimation adopts the improved Lucas-Kanade algorithm, and the calculation formula is:

[0076] ,

[0077] in: is the image spatial gradient, which is a three-dimensional vector that represents the rate of change of the image brightness in the x, y, and z directions; is the velocity vector field to be determined, which is a three-dimensional vector , represents the displacement rate of the pixel in the x, y, and z directions; is the time derivative of the image, which represents the rate of change of image brightness over time.

[0078] In fetal heart ultrasound image processing, the optical flow equations described above are based on the assumption of constant brightness. This assumption assumes that the brightness of the same tissue remains constant over a short period of time, with only the position changing. This assumption applies to most cardiac tissues in ultrasound sequences, but areas with large brightness variations, such as those in the ventricular septal defect region, require special consideration.

[0079] In the specific implementation, for each voxel point , the velocity vector is obtained by solving the following equations :

[0080] ,

[0081] in: 、 、 are the partial derivatives of the image in the three spatial directions of x, y, and z, respectively, indicating the rate of change of the brightness of the image in each direction; is the partial derivative of the image with respect to time, which represents the rate of change of image brightness over time; 、 、 They are the velocity components in the x, y, and z directions to be determined, and the unit is pixel / frame.

[0082] To improve computational stability, this paper uses a 5×5×5 Gaussian kernel for spatial filtering, with a standard deviation σ set to 1.5. In practice, the Gaussian kernel size and standard deviation are determined based on the size and motion characteristics of the fetal heart. For a typical second-trimester fetal heart (20-28 weeks), a kernel size of 5×5×5 balances the requirements for preserving local detail and suppressing noise, while σ=1.5 ensures adequate smoothing.

[0083] The motion separation unit 22 is connected to the optical flow estimation unit 21 and is used to separate fetal movement from the heart's intrinsic motion. This unit distinguishes different sources of motion based on their frequency characteristics. Fetal movement typically manifests as low-frequency global displacement, while intrinsic heart motion manifests as periodic localized deformation.

[0084] In the specific implementation, the frequency domain filtering method is used to design a bandpass filter to extract the cardiac cycle motion:

[0085] ,

[0086] in: is the signal frequency in Hertz (Hz); The low cutoff frequency is set to 0.5 Hz to filter out breathing and other low-frequency movements; is the starting frequency of the high-pass region, set to 1.0 Hz; is the high-pass cutoff frequency, set to 2.5Hz; is half of the signal sampling frequency, that is, the Nyquist frequency. For a sampling rate of 25 frames / second, Hz; It is the frequency response function with a range of [0,1], indicating the retention ratio of the corresponding frequency component.

[0087] These parameter settings are based on the actual situation that the fetal heart rate is usually 110-160 beats per minute (about 1.8-2.7Hz). In applications at different gestational weeks, these parameters can be dynamically adjusted according to the changes in the fetal heart rate to obtain the best separation effect. For example, in the early pregnancy (16-20 weeks), when the fetal heart rate is high, and Adjusted up to 1.2Hz and 2.8Hz.

[0088] The artifact removal unit 23 is connected to the motion separation unit 22 and is used to remove image artifacts caused by fetal movement based on the separated motion information. This unit uses inverse transformation to compensate for the fetal movement between frames to achieve image stabilization.

[0089] The core of the compensation algorithm is:

[0090] ,

[0091] in: is the corrected image; is the original image; is the estimated fetal movement velocity vector, in pixels / frame; is the time interval between frames, in seconds. For an acquisition rate of 25 frames per second, Second; is the voxel coordinate in the image; The time point is in seconds.

[0092] In actual fetal heart ultrasound image processing, the displacement calculated from the fetal movement velocity vector is typically a non-integer pixel, necessitating pixel resampling. This invention employs a bicubic interpolation algorithm for resampling to ensure image quality. While preserving image detail, the bicubic interpolation algorithm effectively suppresses artifacts, making it particularly suitable for medical images rich in detail, such as fetal heart ultrasound.

[0093] Artifact elimination is particularly important for ventricular septal defect detection. Clinical validation has shown that processing with the optical flow correction module of the present invention can reduce image artifacts caused by fetal movement by over 85%, particularly during the second and third trimesters (24-34 weeks), when fetal movement is frequent. For example, in a test of 30 fetuses in the second and third trimesters, the average signal-to-noise ratio in the ventricular septum region was 15.3dB before correction, which increased to 23.7dB after correction, providing a high-quality image foundation for subsequent defect detection.

[0094] See also Figure 2 The cardiac structure timing alignment module 3 is the core innovative part of this system, which includes a multi-scale pyramid construction unit 31, a feature extraction unit 32, a feature fusion unit 33, a model matching unit 34 and a non-rigid transformation unit 35.

[0095] The multi-scale pyramid construction unit 31 is used to construct a multi-level pyramid structure for the received four-dimensional echocardiographic data, each level containing a heart image of different resolutions. Figure 6 ,This unit constructs an image pyramid that usually contains 5 layers through recursive downsampling,,achieving multi-scale expression of cardiac structure from macro to micro.

[0096] In one embodiment of the present invention, the pyramid is constructed using the following recursive formula:

[0097] ,

[0098] in: Indicates the layer pyramid image, The value range of is [1,5]. is the original input image; is the downsampling function, which represents the operation of reducing the resolution of the input image; is the downsampling factor, usually set to 0.5, which means that the linear resolution of each layer is half of the previous layer.

[0099] In fetal heart ultrasound image processing, the original image The three-dimensional data is usually 224×224×96 voxels. After the five-layer pyramid is constructed, the size of each layer image is as follows: 、 、 、 and This multi-scale representation can simultaneously capture the overall morphology (low-resolution layer) and detailed features (high-resolution layer) of the heart, and is particularly suitable for detecting ventricular septal defects of different sizes.

[0100] The feature extraction unit 32 is connected to the multi-scale pyramid construction unit 31 and is used to extract cardiac structural features in different scale spaces. This unit extracts key features such as ventricular contours, interventricular septum, and myocardial boundaries at each pyramid level.

[0101] In a preferred embodiment of the present invention, feature extraction uses an improved SIFT (Scale Invariant Feature Transform) algorithm that is optimized for three-dimensional ultrasound image characteristics. , calculate its local feature descriptor:

[0102] ,

[0103] in: is a voxel point The feature descriptor at is a dimensional vector; represents the 3D gradient histogram centered at the point, The value range of is [1,n], Typically 128; each Indicates the cumulative strength of the gradient in a specific direction, and the unit is a dimensionless normalized value.

[0104] In fetal heart ultrasound image processing, the feature extraction process usually divides the space into For each sub-block, the gradient histogram in 8 directions is calculated to form a 128-dimensional feature vector. To adapt to the noise characteristics of ultrasound images, anisotropic diffusion filtering is used for preprocessing before gradient calculation. The parameters are set to diffusion coefficient K = 25 and the number of iterations is 10.

[0105] These parameter settings are determined based on the characteristics of fetal heart ultrasound images. A diffusion coefficient of K = 25 is suitable for preserving the edges of key structures such as the ventricular septum while effectively suppressing speckle noise. Ten iterations strike a balance between image smoothing and computational efficiency. In practice, for ultrasound images of varying quality, K = 20-30 can be fine-tuned to achieve optimal results.

[0106] The feature fusion unit 33 is connected to the feature extraction unit 32 and is used to perform weighted integration of features at different levels to generate a multi-scale feature description. This unit achieves effective fusion of multi-scale information and improves the robustness of feature expression.

[0107] Feature fusion uses an adaptive weighting scheme:

[0108] ,

[0109] in: is the fused feature, which is a dimension and Same vector; is the feature extracted from the i-th layer, The value range is [0,n], where n is the number of pyramid levels minus 1, usually 4; is the corresponding weight factor, which indicates the importance of the i-th layer feature, satisfying Represents a vector weighted sum operation.

[0110] In fetal heart ultrasound image processing, weights Dynamically determined by the feature quality evaluation function:

[0111] ,

[0112] in: is the quality score of the i-th layer feature, which is a non-negative real number; Represents the sum of all layer feature quality scores. Feature Quality Score It is calculated based on comprehensive indicators such as gradient amplitude, contrast and entropy.

[0113] In ventricular septal defect detection, a typical weight distribution is: (Original layer), (first floor), =0.20(second layer), (Third floor), This allocation method emphasizes the edge details of the original high-resolution layer while taking into account the overall structural information of the low-resolution layer. It is particularly suitable for detecting ventricular septal defects of different sizes.

[0114] The model matching unit 34 is connected to the feature fusion unit 33 and is used to retrieve the most similar reference model from the pre-established cardiac structure model library based on the multi-scale feature description. This unit realizes the intelligent matching of the current ultrasound sequence with the standard cases in the model library.

[0115] The model matching unit 34 includes a feature vector generator 341 , a similarity calculator 342 and an optimal matching selector 343 .

[0116] Feature vector generator 341 is used to convert the current ultrasound sequence into a feature vector. In fetal heart ultrasound image processing, a feature vector typically contains various information, including cardiac anatomical structural features (such as the location and thickness of the ventricular septum), texture features, and dynamic features, forming a high-dimensional vector that comprehensively represents the current case.

[0117] The similarity calculator 342 is connected to the feature vector generator 341 and is used to calculate the similarity between the current feature vector and the reference model stored in the model library. In the present invention, the similarity calculation adopts the cosine distance:

[0118] ,

[0119] in: is the similarity value, the value range is [-1,1], the larger the value, the more similar it is; is the feature vector of the current ultrasound sequence; is the feature vector of the mth model in the model library; Represents the dot product of two vectors; and They represent the Euclidean norm of the two vectors (i.e., the length of the vector).

[0120] In fetal heart ultrasound images, when the similarity value When , it is usually considered that the current sequence has a high similarity with the reference model and is suitable for the initial parameter setting of subsequent non-rigid transformations.

[0121] The best match selector 343 is connected to the similarity calculator 342 and is used to select the reference model with the highest similarity as the matching result. In an actual system, in order to improve the reliability of matching, clinical factors such as gestational age and fetal position are usually considered for auxiliary screening.

[0122] In one embodiment of the present invention, the model library contains 1,000 sets of standard images, categorized and indexed by gestational age (16-40 weeks), maternal characteristics (such as BMI), fetal position (cephalic, breech, transverse), and cardiac anatomy (normal, ventricular septal defect, other congenital heart disease). This structured organization significantly improves search efficiency, typically enabling optimal matching within 0.5 seconds.

[0123] The non-rigid transformation unit 35 is connected to the model matching unit 34 and is used to perform non-rigid transformation based on the reference model to align the cardiac structure. This is the core step of the entire alignment process and enables accurate modeling of the complex deformation of the cardiac structure.

[0124] The non-rigid transformation unit 35 includes a global transformation processor 351 , a local transformation processor 352 and an iterative optimization processor 353 .

[0125] The global transformation processor 351 is used to perform the overall coarse alignment of the cardiac structure. This processor implements rigid transformations, including translation, rotation, and scaling. The basic transformation matrix is:

[0126] ,

[0127] in: is the global transformation matrix, the size is 3×3 (two-dimensional case) or 4×4 (three-dimensional case); is the scaling factor, which indicates the ratio of overall size adjustment, usually in the range of 0.8-1.2; is the rotation angle, in radians, which represents the amount of rotation around the center point; is the translation vector, in pixels, representing the amount of translation in the x and y directions. For 3D data, it is expanded to a 4×4 transformation matrix, adding a transformation parameter in the z direction.

[0128] In fetal heart ultrasound image processing, global transformation parameters are usually initialized based on a matched reference model. For example, for a 24-week gestational fetus in the head position, typical transformation parameters might be: =1.05 (slightly enlarged), = 0.1 radian (about 5.7 degrees rotation), =5 pixels, = 3 pixels. These parameters vary with gestational age and fetal position.

[0129] The local transformation processor 352 is connected to the global transformation processor 351 and is used to perform local fine transformation based on global alignment. The processor adopts a free-form deformation (FFD) model and implements local deformation based on B-spline function.

[0130] The specific formula is:

[0131] ,

[0132] in: is the local transformation function, which represents the point The new position after the transformation; 、 、 is the cubic B-spline basis function, 、 、 The value range of is [0,3]; are the coordinates of the control points in the control point grid, where , , , is the control point grid spacing; is the normalized coordinate, calculated as , , .

[0133] The cubic B-spline basis function is defined as:

[0134] ,

[0135] ,

[0136] ,

[0137] ,

[0138] in: is the normalized coordinate, and its value range is [0,1]; are four basis functions used for interpolation calculation. In fetal heart ultrasound image processing, the control point grid spacing is usually set to 8-16 voxels. For example, for 224 Image, you can set the control point grid spacing to , generating approximately 14×14×12 control points. In high gradient areas such as the interventricular septum, the control points can be dynamically encrypted to 8×8×4 spacing to obtain more precise deformation control.

[0139] The iterative optimization processor 353 is connected to the local transformation processor 352 and is used to adjust the transformation parameters through multiple iterations until a preset alignment quality threshold is reached. This processor implements the automatic optimization process of the transformation parameters.

[0140] The optimization objective function is:

[0141] ,

[0142] in: is the total objective function, which is a non-negative real number. The smaller the value, the higher the alignment quality. is the image similarity measure, usually using mutual information or normalized cross-correlation; It is the smoothness constraint of the deformation field, used to prevent excessive deformation; and is a trade-off factor used to balance the importance of similarity and smoothness, usually set , The value range is 0.01-0.1.

[0143] In fetal heart ultrasound image processing, the calculation formula of mutual information (MI) is:

[0144] ,

[0145] in: is the mutual information value, in bits, and the larger the value, the higher the matching degree between the two images; and are the two images to be aligned; and Images and Entropy, for and The joint entropy of .

[0146] The calculation formula of the deformation field smoothness constraint is:

[0147] ,

[0148] in: is the smoothness constraint term, the smaller the value, the smoother the deformation field; is the image domain; is the Laplace operator, representing the second-order derivative; is the transformation function; represents the Euclidean norm.

[0149] Iterative optimization uses gradient descent with a learning rate of 0.01 and a maximum number of iterations of 100. The optimization is terminated early if the objective function changes by less than 0.001 between two consecutive iterations. In practice, convergence is typically achieved within 15–30 iterations.

[0150] After processing by the Cardiac Structure Time Series Alignment Module 3, cardiac structures across different time frames are precisely aligned. In particular, the ventricular septal defect region maintains a stable position throughout the entire sequence, laying the foundation for subsequent precise analysis. Clinical validation has shown that this module improves cardiac structure alignment accuracy from the original pixel level (approximately 1-2 mm) to the sub-pixel level (approximately 0.3-0.5 mm), significantly improving the accuracy of subsequent defect analysis.

[0151] For example, in a test of 40 cases of ventricular septal defects at different gestational ages (18-32 weeks), the overlap coefficient (Dice coefficient) of the ventricular septum area before and after alignment increased from an average of 0.68 to 0.92. The improvement was particularly significant for cases with obvious fetal movement.

[0152] See also Figure 4 The defect analysis module 4 includes an edge detection unit 41 , a size measurement unit 42 , a diversion direction analysis unit 43 and a severity assessment unit 44 .

[0153] The edge detection unit 41 is used to identify the edge of the ventricular septal defect in the aligned cardiac structure data. This unit accurately segments the ventricular septal defect area and generates a defect edge contour.

[0154] In a preferred embodiment of the present invention, edge detection uses an improved level set method, and its evolution equation is:

[0155] ,

[0156] in: is the level set function, representing the implicit surface, It is located at the edge; is the evolution time, which indicates the number of iteration steps; is the partial derivative of the level set function with respect to time, which represents the surface evolution rate; is the Dirac function, which is used to restrict the evolution to proceed only near the edge; is the edge stopping function based on image gradient; is the curvature term, used to keep the surface smooth; is the gradient of the level set function; is the gradient of the edge stopping function; and To control the parameters, usually set .

[0157] Dirac function Defined as:

[0158] ,

[0159] Where: ϵ is the bandwidth parameter, which controls the range of action and is usually set to 1.5 pixels.

[0160] Edge stop function Defined as:

[0161] ,

[0162] in: is a Gaussian kernel with a standard deviation of σ, used to smooth the image; * represents the convolution operation; I is the input image; ∇ is the gradient operator; represents the gradient amplitude of the smoothed image; g is the edge stop function, which approaches 0 at the edge (large gradient) and approaches 1 in the smooth area (small gradient).

[0163] In fetal heart ultrasound image processing, σ is typically set to 1.0, based on the characteristics of ultrasound images and the typical size of ventricular septal defects. The initial contours of the level set method are often based on candidate regions from the previous cardiac structure alignment step. This "warm start" strategy significantly improves segmentation efficiency and accuracy.

[0164] The size measurement unit 42 is connected to the edge detection unit 41 and is used to calculate the diameter of the ventricular septal defect. The unit calculates the maximum diameter and effective diameter of the defect based on the defect edge contour.

[0165] In actual applications, since ventricular septal defects usually have irregular shapes, the system calculates the diameters in multiple directions and takes the maximum value as the maximum diameter:

[0166] ,

[0167] in: is the maximum diameter in millimeters; is the defect contour point set; and are any two points on the contour; Represents the Euclidean distance between two points; Indicates the maximum value operation. For three-dimensional data, and is the three-dimensional coordinate .

[0168] To more fully describe the defect size, the system also calculates the equivalent circular diameter as the effective diameter:

[0169] ,

[0170] in: is the effective diameter in millimeters; is the area of ​​the defect area, in square millimeters; is the ratio of pi to 3.14159. For three-dimensional data, the equivalent spherical diameter of the defect is calculated: ,in is the defect volume in cubic millimeters.

[0171] In clinical practice, the size of the ventricular septal defect is an important indicator for assessing its severity. Typically, Considered a minor defect, For medium defects, These thresholds are determined based on a large amount of clinical data statistics and expert consensus and are applicable to fetal heart assessment at different gestational ages.

[0172] Shunt direction analysis unit 43 is connected to edge detection unit 41 and is used to determine the shunt direction of the VSD based on the morphological characteristics of the defect and blood flow information. This unit identifies the direction of blood flow from the left ventricle to the right ventricle (left-to-right shunt) or from the right ventricle to the left ventricle (right-to-left shunt).

[0173] In one embodiment of the present invention, the flow direction analysis is combined with morphological features and Doppler data. Based on the morphological analysis, the system calculates the pressure gradient estimate on both sides of the defect area:

[0174] ,

[0175] in: is the estimated pressure gradient in millimeters of mercury (mmHg); is the proportionality coefficient, and its empirical value is set to 0.05, which is based on the calibration of the fluid dynamics model and clinical data; is the effective diameter in millimeters; and are the estimated radii of the left and right ventricles, respectively, in millimeters.

[0176] The sign of the pressure gradient indicates the direction of flow diversion: when When , it is judged as left to right shunt (normal physiological state); when When When , it is judged as bidirectional diversion.

[0177] In actual application, the system is further verified by combining Doppler color flow data. In the standard four-chamber view, left-to-right shunts are usually displayed as red or blue jets from the left ventricle to the right ventricle in color Doppler mode (depending on the Doppler angle setting).

[0178] The severity assessment unit 44 is connected to the size measurement unit 42 and the shunt direction analysis unit 43 and is used to comprehensively analyze the defect parameters and assess the severity of the CHD. This unit classifies the severity of the CHD based on factors such as the defect diameter, shunt direction, and location.

[0179] In clinical application, the system adopts the following grading standards:

[0180] Mild: Defect diameter <3mm, left to right shunt, located in the membranous part

[0181] Moderate: Defect diameter 3-6 mm, left-to-right shunt, or small defect located in the muscle

[0182] Severe: Defect diameter > 6 mm, or any right-to-left shunt, or defect located at the outlet

[0183] This grading system, based on authoritative international cardiology guidelines and extensive clinical data, is applicable to the assessment of congenital heart disease in fetuses of different gestational ages. In practice, the system also considers factors such as gestational age and makes adjustments. For example, for early gestation fetuses (18-24 weeks), the defect diameter threshold is adjusted proportionally based on heart size.

[0184] The system also considers the impact of defect location on prognosis. Ventricular septal defects can be categorized by anatomical location as membranous, muscular, and outlet defects. Membranous defects generally have a better prognosis, while muscular defects vary depending on their specific location and size. Outlet defects often coexist with other cardiac abnormalities and have a poorer prognosis. The system automatically determines the defect type by analyzing the defect's position relative to the aortic and tricuspid valves.

[0185] After processing by the Defect Analysis Module 4, the system automatically provides key diagnostic parameters for ventricular septal defects, including diameter, shunt direction, and severity assessment, providing strong support for clinical decision-making. In a clinical validation study of 50 ventricular septal defect cases, the system's grading results achieved a Kappa coefficient of 0.87 consistent with those of three experienced ultrasound physicians, demonstrating excellent clinical practicality.

[0186] The spatiotemporal fusion network module 7 of the present invention is arranged between the cardiac structure timing alignment module 3 and the defect analysis module 4, and is used to integrate the features of the aligned cardiac structure data in the time dimension and the space dimension to enhance the feature expression of the defect area.

[0187] like Figure 5 As shown, the spatiotemporal fusion network module 7 includes a temporal channel 71 , a spatial channel 72 and a fusion layer 73 .

[0188] The timing channel 71 is used to process the time series characteristics of cardiac dynamic changes. This channel uses a long short-term memory network (LSTM) structure to capture the temporal dependencies of cardiac motion.

[0189] In fetal heart ultrasound image processing, the number of hidden units in an LSTM unit is typically set to 128, a value determined based on the input data dimension and the balance between computational resources. The sequence length is set to the number of cardiac cycle frames, typically 20-30 frames (corresponding to one cardiac cycle at a 25 frames / second acquisition rate).

[0190] The LSTM network effectively captures the dynamic changes in the ventricular septum during cardiac contraction and relaxation, which is crucial for distinguishing true ventricular septal defects from artifacts that resemble defects (such as normal anatomical structures at certain angles). For example, a true ventricular septal defect exists throughout the entire cardiac cycle, while certain artifacts may only appear during specific phases of the cardiac cycle.

[0191] The spatial channel 72 is used to process the spatial geometric features of the cardiac structure. This channel uses a multi-layer codec structure and an improved U-Net architecture to extract spatial features.

[0192] The encoding part of U-Net consists of 4 consecutive downsampling blocks, each of which contains:

[0193] Two 3×3×3 three-dimensional convolutional layers, each followed by batch normalization and ReLU activation function

[0194] A 2×2×2 max pooling layer to reduce the spatial resolution

[0195] The specific parameters of the U-Net encoding part are set to:

[0196] Input layer: 224 × 224 × 96 × 1 (size × number of channels);

[0197] First downsampling block: 112×112×48×64;

[0198] Second downsampling block: 56×56×24×128;

[0199] The third downsampling block: 28×28×12×256;

[0200] Fourth downsampling block: 14×14×6×512;

[0201] The decoding part corresponds to 4 upsampling blocks, each of which contains:

[0202] A 2×2×2 transposed convolution upsampling layer to increase the spatial resolution;

[0203] Feature connections (skip connections) with the corresponding layers of the encoding part are used to fuse low-level and high-level features

[0204] Two 3×3×3 three-dimensional convolutional layers, each followed by batch normalization and ReLU activation function;

[0205] In addition, a residual connection is added after each encoding block to improve the efficiency of feature transfer and alleviate the gradient vanishing problem. The implementation formula of the residual connection is:

[0206] ,

[0207] in: is the output feature of the block; The output of the regular processing path (convolution, activation, etc.); is the input feature of the block.

[0208] In fetal heart ultrasound image processing, the U-Net architecture uses a small 3×3×3 convolution kernel in each convolutional layer. This configuration reduces the number of parameters while maintaining the receptive field, making it suitable for processing ultrasound images with complex textures. Furthermore, the skip connection design ensures the transfer of high-resolution features, which is particularly important for detecting small defects.

[0209] The fusion layer 73 is connected to the temporal channel 71 and the spatial channel 72 to integrate temporal and spatial features to generate an enhanced representation of the defective region. This layer effectively integrates temporal and spatial information, improving the recognition of defective region features.

[0210] The fusion layer uses the attention mechanism to achieve adaptive fusion of spatiotemporal features:

[0211] ,

[0212] in: It is the fused feature, and its dimension is the same as the input feature; is the timing feature, the output from the timing channel; is the spatial feature, the output from the spatial channel; and is the learning weight, which is a trainable parameter matrix; is the bias parameter, which is also a trainable parameter vector; It is a sigmoid function that maps the input to the interval [0,1]; Represents an element-wise multiplication operation.

[0213] This attention mechanism allows the network to adaptively adjust the weights of temporal and spatial features based on the characteristics of different regions. For example, in the ventricular septal defect region, temporal features are generally given a higher weight due to the significant changes in its characteristics during the cardiac cycle; whereas in more stable regions such as the myocardium, spatial features are given a higher weight.

[0214] During training, the spatiotemporal fusion network module 7 uses data augmentation techniques, including random rotation (±15°), scaling (0.9-1.1 times), and elastic deformation, to improve model generalization. Optimization uses the Adam algorithm, with an initial learning rate of 0.0001, which decays to 0.5 every 50 epochs. The batch size is 4, and training is terminated early after 200 epochs or if the validation loss does not decrease for 20 consecutive epochs.

[0215] The learning rate setting was determined based on extensive experimental testing. An initial value of 0.0001 ensures convergence stability while providing sufficient learning speed. Decreasing the learning rate to 0.5 of its original value every 50 epochs is a common learning rate scheduling strategy that can help the model converge more precisely to a local optimum in the later stages of training. The batch size was set to 4 to account for network complexity and GPU memory limitations, ensuring efficient operation on most medical-grade GPU workstations.

[0216] After processing by the spatiotemporal fusion network module 7, the feature expression of the defect area is significantly enhanced, especially for small defects and those with blurred boundaries, with a significant improvement in detection capabilities. Clinical validation has shown that this module improves the Dice coefficient for defect segmentation from 0.77 to 0.89, with particularly significant results for detecting small defects in the fetal heart during the 16-22th week of gestation.

[0217] For example, in a test of 35 cases of ventricular septal defects in early pregnancy (16-22 weeks), the system's detection rate for small defects (<3 mm in diameter) was 76.5% without the spatiotemporal fusion network. With the addition of the spatiotemporal fusion network, the detection rate increased to 92.3%. This has important clinical value for early prenatal diagnosis, providing a longer window for clinical intervention and family decision-making.

[0218] The result output module 5 includes a defect visualization unit 51 , a report generation unit 52 and a data storage unit 53 .

[0219] The defect visualization unit 51 is used to convert the defect parameter information into an intuitive graphical representation. This unit generates a visual image including defect area segmentation, diameter marking and shunt direction marking, which is convenient for doctors to understand intuitively.

[0220] In one embodiment of the present invention, visualization uses pseudo-color coding to display defect areas, with different colors representing different analysis results: red indicates the defect area, green indicates the normal ventricular septum, and blue arrows indicate the direction of the shunt. The system also provides an interactive display interface with multiple viewpoints (sagittal, coronal, and transverse), allowing physicians to rotate and zoom to view results from different angles.

[0221] Specifically, the system generates the following standard visualization views:

[0222] Four-chamber view: shows the location and size of the four chambers of the heart and the ventricular septal defect;

[0223] Left ventricular outflow tract section: shows the relationship between the defect and the aorta, especially suitable for observing the outlet defect;

[0224] Short axis section: shows the cross-sectional morphology of the defect and is suitable for observing muscle defects;

[0225] Three-dimensional reconstruction view: displays the three-dimensional shape and relative position of the defect;

[0226] Time series dynamic display: showing the dynamic changes of the defect during the cardiac cycle;

[0227] The report generation unit 52 is connected to the defect visualization unit 51 and is used to generate a congenital heart disease screening report containing diagnosis results and recommendations based on the defect parameter information and graphical representation. This unit converts the quantitative analysis results into a standardized diagnosis report.

[0228] The standard report includes the following:

[0229] Basic information: examination date, gestational age, basic fetal conditions (such as gestational age, fetal position, estimated weight, etc.);

[0230] Defect parameters: location (membranous, muscular, or outlet), maximum diameter, effective diameter, and shunt direction;

[0231] Severity assessment: mild, moderate or severe, with the basis for the assessment;

[0232] Diagnostic recommendations: recommended follow-up intervals, precautions, prenatal consultation recommendations, etc.;

[0233] Visualization images: defect display images of key sections, usually including static images and dynamic sequence links;

[0234] The report format is structured to facilitate quick access to key information while retaining detailed quantitative data for professional analysis. The system automatically adjusts the format based on the reporting standards of different hospitals to suit the needs of different clinical environments.

[0235] The data storage unit 53 is connected to the report generation unit 52 and is used to store the congenital heart disease screening report and related analysis data. This unit realizes the long-term storage and retrospective analysis functions of the data.

[0236] The system supports multiple data storage methods, including local databases and cloud storage, and provides access control and encryption protection. Patient data is stored in a mix of DICOM and custom formats, with raw ultrasound data and analysis results stored separately to optimize storage space and access efficiency.

[0237] The system also provides historical data comparison and analysis, tracking changes in fetal examination results at different gestational ages to aid in assessing disease progression. For example, for a confirmed ventricular septal defect, the system can automatically compare changes in defect size over time, providing objective evidence for prognosis assessment.

[0238] The adaptive model updating module 6 is connected to the cardiac structure timing alignment module 3 and is used to update the reference model in the cardiac structure model library based on new analysis cases to improve the system's adaptability to different fetal physiological characteristics.

[0239] This module implements the system's self-learning and continuous optimization capabilities, and mainly includes three functions:

[0240] First, the case evaluation function: Each newly analyzed case is evaluated for quality, including image quality, segmentation accuracy, and diagnostic consistency. Only high-quality cases are used for model updates. Quality scoring criteria include signal-to-noise ratio (>20dB), contrast (>0.4), segmentation consistency (Dice>0.85), etc.

[0241] These scoring thresholds are determined based on extensive clinical data analysis. For example, a signal-to-noise ratio (SNR) >20 dB is based on ultrasound equipment technical specifications and clinical image quality standards; images below this value often struggle to clearly display ventricular septal details. A contrast ratio >0.4 is based on the normal brightness difference between the ventricular septum and surrounding tissue; images below this value often have blurred ventricular septal boundaries. A Dice score >0.85 is based on professional standards in the field of medical image segmentation and indicates high consistency between the segmentation results and expert annotations.

[0242] Secondly, the incremental model update function integrates high-quality new cases into the existing model library and updates the feature representation of similar cases. The update adopts the sliding window weighted average method:

[0243] ,

[0244] in: The updated model is and Feature vectors of the same dimension; is the original model, which is the feature vector stored in the model library; is the new case feature, the feature vector extracted from the current case; To update the weight, the value range is [0,1], usually set to 0.1-0.3, and dynamically adjusted according to the quality score of the new case. The higher the quality score, Larger values ​​indicate that new cases have a greater impact on the model.

[0245] Update weights The dynamic adjustment formula is:

[0246] ,

[0247] in: is the base update weight, usually set to 0.3; Rate the quality of the current case; The maximum possible value for the quality score, usually 1.0. This dynamic adjustment mechanism ensures that high-quality cases have a more significant impact on the model, while low-quality cases have a relatively small impact.

[0248] Finally, the model library optimization function regularly evaluates the usage frequency and performance of each model in the model library, removing infrequently used or poorly performing models to maintain the efficiency of the model library. The system maintains a usage counter and performance score. If a model has not been used for 30 consecutive days or its performance score falls below the threshold of 0.6, it will be archived or removed.

[0249] The performance score is calculated as follows:

[0250] ,

[0251] in: is the performance score, the value range is [0,1]; is the accuracy rate (Accuracy); is the precision; The weights of each indicator (0.5, 0.3, and 0.2) are determined based on the different tolerances for false positives and false negatives in clinical diagnosis. In congenital heart disease screening, false negatives (missed diagnoses) are generally more serious than false positives (misdiagnoses), so the weight of recall is relatively low.

[0252] Through the adaptive model update module 6, the system can continuously learn from new cases, adapt to different population characteristics and equipment conditions, and continuously improve diagnostic performance. Clinical applications have shown that after three months of adaptive learning, the system's diagnostic accuracy for the Chinese population has increased from an initial 89.5% to 94.2%, significantly enhancing the system's value in practical applications.

[0253] This continuous learning capability is particularly important for system adaptability across different regions and populations. For example, fetal heart morphology varies slightly between different ethnic groups. Through adaptive learning, the system can gradually adjust model parameters to suit the characteristics of the local population. Similarly, the system can gradually adapt to the image characteristics of different brands and models of ultrasound equipment through learning, ensuring consistent diagnostic performance across different devices.

[0254] The overall workflow of the four-dimensional echocardiographic dynamic segmentation system for congenital heart disease provided by the present invention is as follows:

[0255] 1. Image acquisition module 1 acquires four-dimensional echocardiographic data of the fetal heart;

[0256] 2. The optical flow correction module 2 receives the four-dimensional echocardiographic data, corrects the fetal movement artifact based on the optical flow time series analysis technology, and generates corrected four-dimensional echocardiographic data;

[0257] 3. The cardiac structure time series alignment module 3 receives the corrected 4D echocardiographic data, aligns the cardiac structure using a multi-scale similarity transformation strategy, extracts the ventricular septal defect region, and generates aligned cardiac structure data;

[0258] 4. The spatiotemporal fusion network module 7 integrates the temporal and spatial features of the aligned cardiac structure data to enhance the feature expression of the defect area;

[0259] 5. The defect analysis module 4 receives the enhanced cardiac structure data, calculates the defect edge line diameter and shunt direction, and generates defect parameter information;

[0260] 6. The result output module 5 receives the defect parameter information and generates a congenital heart disease screening report;

[0261] 7. The adaptive model update module 6 updates the reference model in the cardiac structure model library based on new analysis cases to improve the system's adaptability to different fetal physiological characteristics.

[0262] In actual applications, the entire processing process usually takes 30-45 seconds to complete (based on standard medical workstation configuration: Intel Xeon E5-2690 CPU, NVIDIA RTX 3090 GPU, 64GB RAM), while traditional manual analysis of the same data usually takes 10-15 minutes, significantly improving efficiency.

[0263] The system of this invention has been clinically validated in several top-tier hospitals, collecting 350 fetal echocardiographic data, including 120 normal fetuses and 230 fetuses with different types of congenital heart disease (including 150 with ventricular septal defect). The validation results show that:

[0264] 1. Accuracy: The system achieved an accuracy of 94.2% for detecting ventricular septal defects, a sensitivity of 92.5%, and a specificity of 95.8%. The average diagnostic agreement (Kappa coefficient) with three experienced sonographers reached 0.89.

[0265] 2. Efficiency: The system's average diagnosis time is 45 seconds per case, 15 times faster than manual diagnosis (average 11 minutes per case);

[0266] 3. Early screening capability: The system can detect ventricular septal defects as early as 16 weeks of gestation, 4-8 weeks earlier than conventional manual screening (usually at 20-24 weeks), providing a time window for early intervention;

[0267] 4. Measurement accuracy: The system's average measurement error for defect diameter is 0.3mm, significantly better than manual measurement (average error 0.8mm).

[0268] At present, this system has been put into clinical trials in some hospitals. Feedback shows that the system has good practicality and stability, and can effectively assist doctors in congenital heart disease screening, especially in primary medical institutions and scenarios with high screening pressure.

[0269] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A four-dimensional echocardiographic dynamic segmentation system for congenital heart disease, characterized by: include: an image acquisition module for acquiring four-dimensional echocardiographic data of the fetal heart; an optical flow correction module, connected to the image acquisition module, configured to receive the four-dimensional echocardiographic data, correct fetal movement artifacts based on an optical flow time series analysis technique, and generate corrected four-dimensional echocardiographic data; a cardiac structure time series alignment module, connected to the optical flow field correction module, configured to receive the corrected four-dimensional echocardiographic data, align the cardiac structure using a multi-scale similarity transformation strategy, extract the ventricular septal defect region, and generate aligned cardiac structure data; a defect analysis module connected to the cardiac structure timing alignment module, configured to receive the aligned cardiac structure data, calculate the defect edge line diameter and shunt direction, and generate defect parameter information; The result output module is connected to the defect analysis module and is used to receive the defect parameter information and generate a congenital heart disease screening report.

2. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 1, characterized in that: The cardiac structure timing alignment module includes: a multi-scale pyramid construction unit, configured to construct a multi-level pyramid structure for the received four-dimensional echocardiographic data, wherein each level contains a heart image of a different resolution; a feature extraction unit, connected to the multi-scale pyramid construction unit, for extracting cardiac structural features in different scale spaces; A feature fusion unit, connected to the feature extraction unit, for weighted integration of features at different levels to generate a multi-scale feature description; a model matching unit, connected to the feature fusion unit, for retrieving the most similar reference model from a pre-established cardiac structure model library based on the multi-scale feature description; A non-rigid transformation unit is connected to the model matching unit and is used to perform non-rigid transformation based on the reference model to align the heart structure.

3. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 2, characterized in that: The non-rigid transformation unit includes: a global transform processor for performing an overall coarse alignment of the cardiac structures; A local transformation processor, connected to the global transformation processor, configured to perform local fine transformation based on global alignment; An iterative optimization processor is connected to the local transformation processor and is used to adjust the transformation parameters through multiple iterations until a preset alignment quality threshold is reached.

4. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 1, characterized in that: The optical flow field correction module includes: an optical flow field estimation unit, configured to estimate the motion field of the fetal heart by analyzing image changes between adjacent ultrasound frames; a motion separation unit, connected to the optical flow field estimation unit, for separating fetal movement from the heart's own movement; The artifact elimination unit is connected to the motion separation unit and is used to eliminate image artifacts caused by fetal movement based on the separated motion information.

5. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 1, characterized in that: The defect analysis module includes: an edge detection unit, configured to identify an edge of a ventricular septal defect in the aligned cardiac structure data; a size measurement unit, connected to the edge detection unit, and configured to calculate the diameter of the ventricular septal defect; a shunt direction analysis unit connected to the edge detection unit, configured to determine the shunt direction of the defect based on the morphological characteristics and blood flow information of the ventricular septal defect; The severity assessment unit is connected to the size measurement unit and the shunt direction analysis unit, and is used to comprehensively analyze defect parameters and assess the severity of congenital heart disease.

6. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 2, characterized in that: The model matching unit includes: A feature vector generator, used to convert the current ultrasound sequence into a feature vector; A similarity calculator, connected to the feature vector generator, for calculating the similarity between the current feature vector and the reference model stored in the model library; The best matching selector is connected to the similarity calculator and is used to select the reference model with the highest similarity as the matching result.

7. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 1, characterized in that: Also includes: The spatiotemporal fusion network module is arranged between the cardiac structure timing alignment module and the defect analysis module, and is used to integrate the features of the aligned cardiac structure data in the time dimension and the space dimension to enhance the feature expression of the defect area.

8. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 7, characterized in that: The spatiotemporal fusion network module includes: The time series channel is used to process the time series characteristics of dynamic changes of the heart; Spatial channel, used to process the spatial geometric characteristics of cardiac structures; The fusion layer is connected to the temporal channel and the spatial channel, and is used to integrate temporal and spatial features to generate an enhanced defect area expression.

9. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 1, characterized in that: The result output module includes: a defect visualization unit, configured to convert the defect parameter information into an intuitive graphical representation; a report generating unit connected to the defect visualization unit, for generating a congenital heart disease screening report including diagnosis results and recommendations based on the defect parameter information and the graphical representation; The data storage unit is connected to the report generating unit and is used to store the congenital heart disease screening report and related analysis data.

10. The congenital heart disease four-dimensional echocardiography dynamic segmentation system according to claim 1, characterized in that: Also includes: The adaptive model updating module is connected to the cardiac structure timing alignment module and is used to update the reference model in the cardiac structure model library based on new analysis cases to improve the system's adaptability to different fetal physiological characteristics.

Images