Three-dimensional reconstruction method and related device for cardiac chambers based on echocardiogram

Abstract

The present application discloses a three-dimensional reconstruction method and related device for a cardiac chamber based on echocardiogram, which relates to the technical fields of artificial intelligence and three-dimensional reconstruction. Three-dimensional point cloud data is generated based on the echocardiogram of the cardiac chamber to be reconstructed, the three-dimensional point cloud data is normalized, uniform sampling is performed on a preset three-dimensional region to obtain a plurality of query points, and the three-dimensional coordinates of all query points are normalized. Using the normalized point cloud data and the normalized three-dimensional coordinates of each query point as inputs, the trained occupancy function prediction model is used to obtain the occupancy result of each query point. Based on the normalized three-dimensional coordinates and occupancy results of all query points, three-dimensional reconstruction of the cardiac chamber to be reconstructed is performed, which can simply and efficiently complete the three-dimensional reconstruction of the cardiac chamber and help doctors accurately, intuitively, and quantitatively analyze the cardiac function of patients through echocardiogram.

Description

Technical Field

[0001] The present application relates to the technical fields of artificial intelligence and three-dimensional reconstruction, and particularly to a method and related device for three-dimensional reconstruction of cardiac chambers based on echocardiogram. Background Art

[0002] Echocardiogram is one of the most commonly used imaging methods in cardiac examinations. It uses ultrasonic scanning to observe the cardiac structure and blood flow through the human skin and muscles, and to clarify the functional conditions and integrity of the cardiac chambers and large blood vessels. Obtaining a three-dimensional model of the cardiac chambers on an echocardiogram can help doctors more accurately quantify and analyze the cardiac function of patients, such as measuring ventricular volume, ejection fraction, etc. However, echocardiogram, as an ultrasonic image, has a lot of noise and sparse sampling, and it is relatively difficult to directly obtain a three-dimensional model of the cardiac chambers from an echocardiogram. Marking the anatomical structure position points on the echocardiogram and converting the two-dimensional coordinates of the anatomical structure position points into a three-dimensional coordinate system to form a three-dimensional point cloud, and reconstructing the three-dimensional model of the cardiac chambers based on the three-dimensional point cloud is an effective method.

[0003] Currently, the method for three-dimensional reconstruction of cardiac chambers based on echocardiogram is mainly a knowledge-based reconstruction algorithm (KBR). This knowledge-based reconstruction algorithm uses a piece-wise smooth subdivision surface (PSSS) to fine-tune a template mesh based on the three-dimensional point cloud, and uses this template mesh to fit the three-dimensional point cloud to reconstruct the three-dimensional model of the cardiac chambers. However, this knowledge-based reconstruction algorithm requires marking the structural features of each point, each edge, and each face of the template mesh, which not only requires professional knowledge but also is time-consuming and laborious.

[0004] Based on this, there is an urgent need for a technology that can simply and efficiently reconstruct the three-dimensional model of the cardiac chambers. Summary of the Invention

[0005] The purpose of the present application is to provide a method and related device for three-dimensional reconstruction of cardiac chambers based on echocardiogram, which can simply and efficiently reconstruct the three-dimensional model of the cardiac chambers.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] In a first aspect, the present application provides a method for three-dimensional reconstruction of cardiac chambers based on echocardiogram. The method for three-dimensional reconstruction of cardiac chambers based on echocardiogram includes:

[0008] Generate three-dimensional point cloud data based on the echocardiogram of the heart chamber to be reconstructed; the three-dimensional point cloud data includes a plurality of three-dimensional points and the three-dimensional coordinates and categories of each three-dimensional point; the three-dimensional points are the anatomical structure position points of the heart chamber to be reconstructed, and the category is the anatomical structure category of the anatomical structure position point;

[0009] Perform normalization processing on the three-dimensional point cloud data to obtain normalized point cloud data; the three-dimensional coordinates of the center point of the normalized point cloud data are the coordinate origin, and the three-dimensional coordinates of each three-dimensional point of the normalized point cloud data are within a preset range;

[0010] Perform uniform sampling on a preset three-dimensional region to obtain a plurality of query points, and perform normalization processing on the three-dimensional coordinates of all the query points to obtain the normalized three-dimensional coordinates of each query point; the normalized three-dimensional coordinates of the center point of all the query points are the coordinate origin, and the normalized three-dimensional coordinates are within a preset range;

[0011] Using the normalized point cloud data and the normalized three-dimensional coordinates of each query point as inputs, use the trained occupancy function prediction model to obtain the occupancy result of each query point; the occupancy result includes that the query point is located inside the heart chamber to be reconstructed and that the query point is located outside the heart chamber to be reconstructed;

[0012] Perform three-dimensional reconstruction on the heart chamber to be reconstructed based on the normalized three-dimensional coordinates and occupancy results of all the query points to obtain an initial three-dimensional model of the heart chamber to be reconstructed;

[0013] Perform anti-normalization processing on the initial three-dimensional model to obtain a three-dimensional model of the heart chamber to be reconstructed.

[0014] Optionally, the trained occupancy function prediction model includes an encoder, a fusion module, and a decoder connected in sequence;

[0015] The encoder is used to process the normalized point cloud data to obtain point cloud features;

[0016] The fusion module is used to process the point cloud features and the normalized three-dimensional coordinates of each query point to obtain fusion features;

[0017] The decoder is used to process the fusion features to obtain the occupancy result of each query point.

[0018] Optionally, the encoder includes a plurality of first multi-layer perceptron pooling modules and a multi-layer perceptron connected in sequence; the first multi-layer perceptron pooling module includes a multi-layer perceptron and a pooling layer connected in sequence;

[0019] The fusion module includes a second multi-layer perceptron pooling module and a multi-layer perceptron connected in sequence. The second multi-layer perceptron pooling module includes a multi-layer perceptron and a pooling layer connected in parallel; alternatively, the fusion module includes a feature projection layer, a segmentation network, a feature query layer, and a pooling layer connected in sequence; wherein, the feature projection layer is used to project the point cloud features onto the xoy plane, xoz plane, and yoz plane respectively to obtain three plane vectors; the segmentation network is used to process the three plane vectors to obtain three plane features; the feature query layer is used to determine the query point feature of each query point based on the normalized three-dimensional coordinates of each query point and the three plane features.

[0020] The decoder includes a number of multi-layer perceptrons connected in sequence.

[0021] Optionally, generating three-dimensional point cloud data based on the echocardiogram of the heart chamber to be reconstructed specifically includes:

[0022] Based on the echocardiogram of the heart chamber to be reconstructed, determine the two-dimensional coordinates of each anatomical structure position point;

[0023] For each anatomical structure position point, use the three-dimensional pose information of the ultrasound probe when generating the echocardiogram to perform coordinate transformation on the two-dimensional coordinates of the anatomical structure position point to obtain the three-dimensional coordinates of the anatomical structure position point; record the anatomical structure position point as a three-dimensional point, record the three-dimensional coordinates of the anatomical structure position point as the three-dimensional coordinates of the three-dimensional point, record the anatomical structure category of the anatomical structure position point as the category of the three-dimensional point, and form three-dimensional point cloud data from all the three-dimensional points and the three-dimensional coordinates and categories of each three-dimensional point.

[0024] Optionally, before using the normalized point cloud data and the normalized three-dimensional coordinates of each query point as inputs and using the trained occupancy function prediction model to obtain the occupancy result of each query point, the three-dimensional reconstruction method of the heart chamber based on echocardiogram further includes:

[0025] Obtain a data set; the data set includes multiple sample normalized point cloud data and the corresponding sample initial three-dimensional model for each sample normalized point cloud data;

[0026] For each sample normalized point cloud data, determine a sampling area based on the sample normalized point cloud data, perform random sampling on the sampling area to obtain multiple sample query points; for each sample query point, determine the actual occupancy result of the sample query point based on the three-dimensional coordinates of the sample query point and the sample initial three-dimensional model corresponding to the sample normalized point cloud data.

[0027] Using the sample-normalized point cloud data and the three-dimensional coordinates of each sample query point as inputs, and using the actual occupancy result of each sample query point as a label, train an initial occupancy function prediction model to obtain a trained occupancy function prediction model.

[0028] Optionally, perform three-dimensional reconstruction on the heart chamber to be reconstructed based on the normalized three-dimensional coordinates and occupancy results of all the query points to obtain an initial three-dimensional model of the heart chamber to be reconstructed, specifically including:

[0029] Using the normalized three-dimensional coordinates and occupancy results of all the query points as inputs, use the marching cube algorithm to perform three-dimensional reconstruction on the heart chamber to be reconstructed to obtain an initial three-dimensional model of the heart chamber to be reconstructed.

[0030] In a second aspect, the present application provides a three-dimensional reconstruction device for a heart chamber based on echocardiogram. The three-dimensional reconstruction device for a heart chamber based on echocardiogram includes:

[0031] A point cloud generation module for generating three-dimensional point cloud data based on the echocardiogram of the heart chamber to be reconstructed; the three-dimensional point cloud data includes a plurality of three-dimensional points and the three-dimensional coordinates and categories of each three-dimensional point; the three-dimensional points are anatomical structure position points of the heart chamber to be reconstructed, and the category is the anatomical structure category of the anatomical structure position point;

[0032] A normalization module for performing normalization processing on the three-dimensional point cloud data to obtain normalized point cloud data; the three-dimensional coordinates of the center point of the normalized point cloud data are the coordinate origin, and the three-dimensional coordinates of each three-dimensional point of the normalized point cloud data are within a preset range;

[0033] A query point determination module for uniformly sampling a preset three-dimensional region to obtain a plurality of query points, and performing normalization processing on the three-dimensional coordinates of all the query points to obtain the normalized three-dimensional coordinates of each query point; the normalized three-dimensional coordinates of the center point of all the query points are the coordinate origin, and the normalized three-dimensional coordinates are within a preset range;

[0034] A model prediction module for using the normalized point cloud data and the normalized three-dimensional coordinates of each query point as inputs, and using the trained occupancy function prediction model to obtain the occupancy result of each query point; the occupancy result includes that the query point is located inside the heart chamber to be reconstructed and that the query point is located outside the heart chamber to be reconstructed;

[0035] A three-dimensional reconstruction module for performing three-dimensional reconstruction on the heart chamber to be reconstructed based on the normalized three-dimensional coordinates and occupancy results of all the query points to obtain an initial three-dimensional model of the heart chamber to be reconstructed;

[0036] A model restoration module, configured to perform denormalization processing on the initial three-dimensional model to obtain a three-dimensional model of the heart chamber to be reconstructed.

[0037] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the echocardiogram-based three-dimensional reconstruction method of the heart chamber described in any one of the above.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the echocardiogram-based three-dimensional reconstruction method of the heart chamber described in any one of the above.

[0039] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the echocardiogram-based three-dimensional reconstruction method of the heart chamber described in any one of the above.

[0040] According to the specific embodiments provided by the present application, the following technical effects are disclosed in the present application:

[0041] The present application provides an echocardiogram-based three-dimensional reconstruction method of the heart chamber and related devices. Three-dimensional point cloud data is generated based on the echocardiogram of the heart chamber to be reconstructed, the three-dimensional point cloud data is normalized to obtain normalized point cloud data, a preset three-dimensional region is uniformly sampled to obtain a plurality of query points, and the three-dimensional coordinates of all query points are normalized to obtain the normalized three-dimensional coordinates of each query point. Using the normalized point cloud data and the normalized three-dimensional coordinates of each query point as inputs, the trained occupancy function prediction model is used to obtain the occupancy result of each query point. Based on the normalized three-dimensional coordinates and occupancy results of all query points, three-dimensional reconstruction of the heart chamber to be reconstructed is performed to obtain an initial three-dimensional model of the heart chamber to be reconstructed, and the initial three-dimensional model is denormalized to obtain a three-dimensional model of the heart chamber to be reconstructed. There is no need to introduce a template mesh, nor to label the template mesh, saving the time-consuming and laborious labeling work that requires professional knowledge, and being able to simply and efficiently complete the three-dimensional reconstruction of the heart chamber, helping doctors accurately, intuitively, and quantitatively analyze the heart function of patients through echocardiograms. Description of the Drawings

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0043] Figure 1 This is an application environment diagram of a three-dimensional reconstruction method for cardiac chambers based on echocardiogram provided in Embodiment 1 of the present application.

[0044] Figure 2 This is a schematic flowchart of a three-dimensional reconstruction method for cardiac chambers based on echocardiogram provided in Embodiment 1 of the present application.

[0045] Figure 3 This is a schematic structural diagram of a trained occupancy function prediction model provided in Embodiment 1 of the present application.

[0046] Figure 4 This is a first schematic structural diagram of a fusion module provided in Embodiment 1 of the present application.

[0047] Figure 5 This is a second schematic structural diagram of a fusion module provided in Embodiment 1 of the present application.

[0048] Figure 6 This is a training flowchart of a trained occupancy function prediction model provided in Embodiment 1 of the present application.

[0049] Figure 7 This is a comparison schematic diagram of a target three-dimensional model and a predicted three-dimensional model provided in Embodiment 1 of the present application.

[0050] Figure 8 This is a schematic diagram of functional modules of a three-dimensional reconstruction device for cardiac chambers based on echocardiogram provided in Embodiment 2 of the present application.

[0051] Figure 9 This is a schematic structural diagram of a computer device provided in Embodiment 3 of the present application. Detailed implementation manners

[0052] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0053] Embodiment 1

[0054] The three-dimensional reconstruction method for cardiac chambers based on echocardiogram provided in the embodiments of the present application can be applied, for example, as Figure 1In the application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be set separately, integrated on the server, or placed on the cloud or other servers. The terminal can send the echocardiogram to be processed to the server. After the server receives the echocardiogram to be processed, for the echocardiogram to be processed, the server generates three-dimensional point cloud data based on the echocardiogram of the heart chamber to be reconstructed, normalizes the three-dimensional point cloud data to obtain normalized point cloud data, uniformly samples a preset three-dimensional region to obtain multiple query points, and normalizes the three-dimensional coordinates of all query points to obtain the normalized three-dimensional coordinates of each query point. Using the normalized point cloud data and the normalized three-dimensional coordinates of each query point as inputs, the trained occupancy function prediction model is used to obtain the occupancy result of each query point. Based on the normalized three-dimensional coordinates and occupancy results of all query points, the heart chamber to be reconstructed is three-dimensionally reconstructed to obtain an initial three-dimensional model of the heart chamber to be reconstructed, and the initial three-dimensional model is anti-normalized to obtain the three-dimensional model of the heart chamber to be reconstructed. The server can feedback the obtained three-dimensional model for the echocardiogram to the terminal.

[0055] In addition, in some embodiments, the three-dimensional reconstruction method of the heart chamber based on echocardiogram can also be implemented separately by the server or the terminal. For example, the terminal can directly process the echocardiogram to be processed, or the server can obtain the echocardiogram to be processed from the data storage system and process the echocardiogram to be processed.

[0056] Among them, the terminal can be, but is not limited to, various desktop computers, laptop computers, smartphones, tablets, Internet of Things devices, and portable wearable devices. The server can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0057] Such as Figure 2 As shown, this embodiment provides a three-dimensional reconstruction method of the heart chamber based on echocardiogram. This method is executed by a computer device, and can be specifically executed alone by a computer device such as a terminal or a server, or jointly executed by the terminal and the server. In the embodiments of the present application, taking this method applied to Figure 1 the server in as an example for illustration, the three-dimensional reconstruction method of the heart chamber based on echocardiogram includes the following steps:

[0058] S1: Generate three-dimensional point cloud data based on the echocardiogram of the heart chamber to be reconstructed; the three-dimensional point cloud data includes multiple three-dimensional points and the three-dimensional coordinates and categories of each three-dimensional point; the three-dimensional points are anatomical structure position points of the heart chamber to be reconstructed, and the category is the anatomical structure category of the anatomical structure position points.

[0059] S2: Normalize the three-dimensional point cloud data to obtain normalized point cloud data; the three-dimensional coordinates of the center point of the normalized point cloud data are the origin of coordinates, and the three-dimensional coordinates of each three-dimensional point of the normalized point cloud data are within a preset range.

[0060] S3: Uniformly sample a preset three-dimensional region to obtain a plurality of query points, and normalize the three-dimensional coordinates of all the query points to obtain the normalized three-dimensional coordinates of each query point; the normalized three-dimensional coordinates of the center point of all the query points are the origin of coordinates, and the normalized three-dimensional coordinates are within a preset range.

[0061] S4: Use the normalized point cloud data and the normalized three-dimensional coordinates of each query point as inputs, and use the trained occupancy function prediction model to obtain the occupancy result of each query point; the occupancy result includes that the query point is inside the heart chamber to be reconstructed and that the query point is outside the heart chamber to be reconstructed.

[0062] S5: Based on the normalized three-dimensional coordinates and occupancy results of all the query points, perform three-dimensional reconstruction on the heart chamber to be reconstructed to obtain an initial three-dimensional model of the heart chamber to be reconstructed.

[0063] S6: Perform denormalization processing on the initial three-dimensional model to obtain a three-dimensional model of the heart chamber to be reconstructed.

[0064] By implementing the above steps S1 - S6, in this embodiment, deep learning technology is used to fit the implicit function of the three-dimensional structure of the heart chamber, and a trained occupancy function prediction model is obtained. Using this trained occupancy function prediction model, three-dimensional reconstruction of the heart chamber based on echocardiography is realized. Specifically, taking the three-dimensional sparse point cloud of the heart anatomical structure as the input, a neural network is used to fit the occupancy function (Occupancy function, that is, the implicit function) of the heart chamber to be reconstructed, and predict whether any query point in space is inside the heart chamber to be reconstructed, so as to realize the three-dimensional reconstruction of the heart chamber to be reconstructed. There is no need to introduce a template mesh, nor to label the template mesh, saving the time-consuming and laborious labeling work that requires professional knowledge. A simple and efficient method for three-dimensional reconstruction of the heart chamber based on the implicit function can be proposed for echocardiography, and a three-dimensional model of the heart chamber can be obtained. Doctors can further analyze through this three-dimensional model to help doctors accurately, intuitively, and quantitatively analyze the heart function of patients through echocardiography.

[0065] In S1, the heart chamber to be reconstructed can be the left atrium, right atrium, left ventricle, and right ventricle.

[0066] In S1, three-dimensional point cloud data is generated based on the echocardiogram of the cardiac chamber to be reconstructed, specifically including: based on the echocardiogram of the cardiac chamber to be reconstructed, determining the two-dimensional coordinates of each anatomical structure position point; for each anatomical structure position point, using the three-dimensional pose information of the ultrasound probe when generating the echocardiogram to perform coordinate transformation on the two-dimensional coordinates of the anatomical structure position point to obtain the three-dimensional coordinates of the anatomical structure position point. Denote the anatomical structure position point as a three-dimensional point, denote the three-dimensional coordinates of the anatomical structure position point as the three-dimensional coordinates of the three-dimensional point, denote the anatomical structure category of the anatomical structure position point as the category of the three-dimensional point, and form three-dimensional point cloud data from all three-dimensional points and the three-dimensional coordinates and categories of each three-dimensional point.

[0067] In this embodiment, first, an echocardiogram of the cardiac chamber to be reconstructed is collected by an ultrasound device, and during the process of collecting the echocardiogram, a sensor is used to collect the movement of the ultrasound probe of the ultrasound device to obtain the three-dimensional pose information of the ultrasound probe. Then, key anatomical structure position points are marked on the 2D echocardiogram to obtain the two-dimensional coordinates of each anatomical structure position point. Finally, the two-dimensional coordinates of each anatomical structure position point are transformed into a three-dimensional coordinate system through the recorded three-dimensional pose information to obtain the three-dimensional coordinates of each anatomical structure position point, and at the same time, the anatomical structure category information of each anatomical structure position point is recorded to obtain three-dimensional point cloud data.

[0068] Among them, considering that the echocardiogram of one angle may not be able to fully display all anatomical structure position points, in this embodiment, when collecting the echocardiogram, multiple views at different angles can be collected to facilitate determining the two-dimensional coordinates of all anatomical structure position points.

[0069] Among them, the anatomical structure position points can be determined according to user needs. As an example, for the right ventricle, the anatomical structure position points include seven types: tricuspid annulus, right ventricular septum, right ventricular endocardium, base, apex, pulmonary valve annulus, and right ventricular septal margin.

[0070] Among them, the anatomical structure category of each anatomical structure position point is recorded in the form of a one-hot code.

[0071] In S2, normalization processing is performed on the three-dimensional point cloud data, specifically including: translating and scaling the three-dimensional point cloud data, and recording the translation parameters and scaling parameters. The preset range can be [-1, 1]. Of course, other preset ranges with different values can also be selected according to needs, and this embodiment does not make any limitations in this regard.

[0072] In S3, the preset three-dimensional region can be a voxel grid of (32, 32, 32). Each voxel grid corresponds to a query point. Of course, voxel grids with other resolutions can also be selected according to needs, and other types of three-dimensional regions can be selected according to needs. By uniformly sampling the three-dimensional region, multiple query points can be obtained. This embodiment does not make any limitation on this.

[0073] In S3, normalize the three-dimensional coordinates of all query points, which specifically includes: translate and scale the three-dimensional coordinates of all query points to obtain the normalized three-dimensional coordinates of each query point. The preset range where the normalized three-dimensional coordinates are located is the same as the preset range where the three-dimensional coordinates of each three-dimensional point in the normalized point cloud data are located.

[0074] In S4, the trained occupancy function prediction model is used to extract point cloud feature information and predict the OCC value of query points in space. OCC is a binary variable. If OCC is equal to 1, it means that the query point is inside the heart chamber to be reconstructed. If OCC is equal to 0, it means that the query point is not inside the heart chamber to be reconstructed. As Figure 3 shown, the trained occupancy function prediction model includes an encoder, a fusion module, and a decoder connected in sequence. The encoder is used to process the normalized point cloud data to extract point cloud features. The fusion module is used to process the point cloud features and the normalized three-dimensional coordinates of each query point to obtain fusion features. The decoder is used to process the fusion features to predict the OCC value of each query point and obtain the occupancy result of each query point.

[0075] Among them, the input of the encoder is the normalized point cloud data (np, 3 + c), which is composed of the three-dimensional coordinates (np, 3) of all three-dimensional points in the normalized point cloud data and the one-hot code of the categories of all three-dimensional points (np, c). Here, np is the number of three-dimensional points in the normalized point cloud data, 3 is the number of dimensions of the three-dimensional coordinates of the three-dimensional points, and c is the number of categories of the three-dimensional points. The output of the encoder is the point cloud features (np, d), and d is the feature dimension of the three-dimensional points.

[0076] The encoder includes a number of first multi-layer perceptron pooling modules connected in sequence and a multi-layer perceptron (MLP). The first multi-layer perceptron pooling module includes a multi-layer perceptron and a pooling layer connected in sequence. The working process of the first multi-layer perceptron pooling module is as follows: The first multi-layer perceptron pooling module is divided into two parts. One part is the MLP in the channel dimension, which is used to extract the local features (np, d) of all three-dimensional points in the normalized point cloud data. The other part is the pooling layer in the point dimension, which is used to extract the d-dimensional global features (d) of the local features (np, d) of all three-dimensional points output by the MLP. The global features are copied np times and spliced onto the local features of each three-dimensional point respectively to obtain features with a size of (np, 2*d). The last multi-layer perceptron processes the features with a size of (np, 2*d) to obtain point cloud features (np, d).

[0077] The input of the fusion module is the point cloud features (np, d) extracted by the encoder and the normalized three-dimensional coordinates (nq, 3) of all query points, where nq is the number of query points. The output of the fusion module is the fused features (nq, d).

[0078] This embodiment presents two structures of the fusion module. The first is to extract features using global pooling + splicing, and the second is to extract features using planar convolution.

[0079] As Figure 4 shown, it is a schematic diagram of the first structure. The fusion module includes a second multi-layer perceptron pooling module and a multi-layer perceptron connected in sequence. The second multi-layer perceptron pooling module includes a multi-layer perceptron and a pooling layer connected in parallel. At this time, the working process of the fusion module is as follows: First, perform max pooling on the point cloud features (np, d) in the point dimension to obtain a d-dimensional global feature vector (d); then use the MLP in the channel dimension to extract features from the normalized three-dimensional coordinates (nq, 3) of all query points to obtain a feature vector (nq, d); finally, copy the d-dimensional global feature vector nq times, splice it onto the feature vector (nq, d), and use the MLP to reduce the dimension to d to obtain the fused features (nq, d).

[0080] As Figure 5As shown in the figure, it is a schematic diagram of the second structure. The fusion module includes a feature projection layer, a segmentation network, a feature query layer, and a pooling layer connected in sequence. The feature projection layer is used to project the point cloud features onto the xoy plane, xoz plane, and yoz plane respectively to obtain three plane vectors. The segmentation network is used to process the three plane vectors to obtain three plane features. The feature query layer is used to determine the query point features of each query point based on the normalized three-dimensional coordinates of each query point and the three plane features. At this time, the working process of the fusion module is as follows: First, according to the three-dimensional coordinates of each three-dimensional point in the three-dimensional point cloud data, project the point cloud features (np, d) onto the xoy, xoz, and yoz planes (the resolution h*w uses 32*32) to obtain three plane vectors 3*(h, w, d); then use a segmentation network (such as UNet) to process the three plane vectors to extract three plane features 3*(h, w, d); finally, extract the query point features (nq, 3, d) of all query points according to the projection positions of each query point on the xoy, xoz, and yoz planes, and use the pooling layer to pool the query point features of all query points to obtain the fused features (nq, d) of all query points on the three planes.

[0081] Among them, the feature projection process includes: Each three-dimensional point in the point cloud features has three-dimensional coordinates (x, y, z) and d-dimensional features. For each three-dimensional point, the process of projecting the three-dimensional point onto the xoy plane is as follows:

[0082] Initialize the features of the xoy plane as a three-dimensional matrix with a shape of (32*32*d) and all values being 0. Vertically map the three-dimensional point to the xoy plane according to the three-dimensional coordinates (x, y, z) of the three-dimensional point to obtain the projection point of the three-dimensional point on the xoy plane. The specific process of vertical mapping is: Set the z coordinate of the three-dimensional point to 0, and translate and scale the values of the x coordinate and y coordinate to [0, 31]. Since the xoy plane is a 32*32 two-dimensional grid and its coordinate points are integers in [0, 31], while the coordinates of the projection point of the three-dimensional point on the xoy plane may be decimals, so round or interpolate the coordinates of the projection point to obtain the rounded coordinates (x0, y0) of the projection point. Then the d-dimensional feature of the three-dimensional point is the feature at the position (x0, y0) on the xoy plane. When performing the above operations on all three-dimensional points in the point cloud features, multiple three-dimensional points may correspond to the same position on the xoy plane. At this time, take the average value of the features of these three-dimensional points to obtain the feature at this position, so as to obtain the features at each position on the xoy plane and form the plane vector of the xoy plane.

[0083] The projection processes of the xoz and yoz planes are the same as the above process and will not be elaborated here.

[0084] Among them, the feature query process includes: according to the projection process in the feature projection process, determining the projection positions of each query point on the xoy, xoz, and yoz planes, and selecting the average value of the features at the projection positions on the xoy, xoz, and yoz planes as the query point feature of the query point.

[0085] The input of the decoder is the fused feature (nq, d), and the output of the decoder is the OCC value (nq, 1) of all query points. The OCC value characterizes the occupancy result of the query point.

[0086] The decoder includes a number of multi-layer perceptrons connected in sequence.

[0087] In S5, based on the normalized three-dimensional coordinates and occupancy results of all query points, three-dimensional reconstruction of the heart chamber to be reconstructed is performed to obtain an initial three-dimensional model of the heart chamber to be reconstructed, specifically including: using the normalized three-dimensional coordinates and occupancy results of all query points as input, and using the marching cube algorithm to perform three-dimensional reconstruction on the heart chamber to be reconstructed to obtain an initial three-dimensional model of the heart chamber to be reconstructed.

[0088] In S6, anti-normalization processing is performed on the initial three-dimensional model, specifically referring to using the translation parameter and scaling parameter to process the initial three-dimensional model. For example, if the three-dimensional point cloud data is translated 1 unit to the right, the initial three-dimensional model is translated 1 unit to the left; if the three-dimensional point cloud data is scaled down by a factor of 10, the initial three-dimensional model is scaled up by a factor of 10 to restore the size and pose, and a three-dimensional model of the heart chamber to be reconstructed is obtained.

[0089] As Figure 6 shown, before using the trained occupancy function prediction model to obtain the occupancy result of each query point with the normalized point cloud data and the normalized three-dimensional coordinates of each query point as input, the three-dimensional reconstruction method of the heart chamber based on echocardiogram in this embodiment further includes:

[0090] (1) Obtain a dataset, where the dataset includes multiple sample normalized point cloud data and the sample initial three-dimensional model corresponding to each sample normalized point cloud data.

[0091] In this embodiment, first, the three-dimensional point cloud data and the real three-dimensional model (i.e., the target three-dimensional model) of the target heart chamber (i.e., the heart structure) are obtained. The acquisition process of the three-dimensional point cloud data is the same as the acquisition process of the three-dimensional point cloud data of the heart chamber to be reconstructed described above, and will not be elaborated here. The real three-dimensional model represents the real model of the target heart chamber represented by the three-dimensional point cloud data, and its shape and pose are consistent with the structure of the target heart chamber represented by the three-dimensional point cloud data.

[0092] For convenient model training, all three-dimensional point cloud data and their corresponding real three-dimensional models are normalized to the same size and position through translation and scaling. Meanwhile, the translation parameters and scaling parameters used for translation and scaling are recorded, which are subsequently used to restore the size and pose of the initially predicted three-dimensional model. Thus, multiple samples of normalized point cloud data and the sample initial three-dimensional models corresponding to each sample of normalized point cloud data can be obtained to form a dataset.

[0093] (2) For each sample of normalized point cloud data, determine the sampling region based on the sample normalized point cloud data, and randomly sample the sampling region to obtain multiple sample query points; for each sample query point, determine the actual occupancy result of the sample query point based on the three-dimensional coordinates of the sample query point and the sample initial three-dimensional model corresponding to the sample normalized point cloud data.

[0094] Training data sampling includes sampling the dataset for model training. Specifically, first densely sample nq sample query points in space, where the space is the sampling region, and the sampling region can be a region that completely contains the sample normalized point cloud data, serving as the model input data (nq, 3); then obtain the occupancy function values OCC(nq) of these nq sample query points.

[0095] (3) Use the sample normalized point cloud data and the three-dimensional coordinates of each sample query point as inputs, and the actual occupancy result of each sample query point as the label to train the initial occupancy function prediction model, obtaining a trained occupancy function prediction model.

[0096] In this embodiment, first build a point cloud implicit function prediction model with an encoder-decoder structure, that is, build an initial occupancy function prediction model, which has exactly the same structure as the trained initial occupancy function prediction model, and then train the initial occupancy function prediction model. Specifically, use the above sample normalized point cloud data, the three-dimensional coordinates of the sample query points, and the OCC values to train the initial occupancy function prediction model.

[0097] After obtaining the trained initial occupancy function prediction model, model inference and mesh extraction can be performed. Specifically, during the inference process, first uniformly sample a sufficient number of query points in space, normalize the query points, and input them together with the normalized point cloud data obtained after normalizing the three-dimensional point cloud data into the trained occupancy function prediction model to obtain the OCC value of each query point; then, through the marching cube algorithm, extract the predicted mesh model, that is, obtain the initial three-dimensional model; finally, according to the recorded translation parameters and scaling parameters, restore the mesh model to its true size and pose to obtain the three-dimensional model.

[0098] The 3D reconstruction method of this embodiment first collects 3D point cloud data and the target 3D model of the real heart structure as a data set, then constructs an initial occupancy function prediction model, and trains the initial occupancy function prediction model so that it can predict whether the input query point is inside the heart structure. Finally, the prediction results of the trained occupancy function prediction model are used, and a mesh is extracted according to the prediction results to complete the 3D model reconstruction. This embodiment provides an automated method framework for reconstructing the heart structure on echocardiograms. Different from existing methods, a latent function prediction model based on an encoder-decoder structure is used to reconstruct the 3D model of the heart structure, which can avoid complex manual feature design and still ensure sufficient accuracy and robustness on sparse and highly noisy point clouds.

[0099] Taking the 3D reconstruction of the right ventricle as an example, it specifically includes the following contents:

[0100] Step 1, obtain the 3D point cloud data and the target 3D model of the heart structure.

[0101] (1) The process of obtaining the 3D point cloud data is as follows: First, use an ultrasound device to collect echocardiograms, and collect 7 - 15 views at different angles for each case, and use a sensor to obtain the 3D pose information of the ultrasound probe. Then, mark the two-dimensional coordinates (np, 2) of the anatomical structure position points on the 2D echocardiogram. The anatomical structure position points specifically include: tricuspid annulus, right ventricular septum, right ventricular endocardium, base, apex, pulmonary valve annulus, right ventricular septal margin, etc., a total of 7 types. Finally, the two-dimensional coordinates are transformed into 3D space through the recorded 3D pose information, and the coordinate points of different views of the same heart at the same stage are merged to obtain the 3D point cloud (np, 3). In addition, the anatomical structure category of each 3D point is recorded in the form of a one-hot code (np, 7).

[0102] (2) The target 3D model can be in 3D representation forms such as mesh and voxel. As an example, in this embodiment, a triangular mesh is used, which consists of vertices (V: (nv, 3)), edges (E: (ne, 2)) and faces (F: (nf, 3)). The V matrix, with a shape of nv rows and 3 columns, represents the 3D coordinates of nv vertices, and each row is the 3D coordinates of a vertex. The E matrix, with a shape of ne rows and 2 columns, represents ne edges. The connection of two vertices is an edge, and the two values in each row are the subscripts of the two vertices corresponding to this edge. The subscript refers to the row number of the vertex in the V matrix. The F matrix, with a shape of nf rows and 3 columns, represents nf faces. A face is a triangle formed by connecting three vertices, and the three values in each row are the subscripts of the three vertices of the triangle in the V matrix.

[0103] For the convenience of subsequent training, all 3D point cloud data and the target 3D model are normalized to [-1, 1] through translation and scaling, and the translation parameters and scaling parameters are recorded.

[0104] Step 2: Construct training data (sample query points - OCC) pairs through sampling. First, randomly sample 2048 points within the coordinate range of [-1.1, 1.1] as sample query points. Then, according to the target 3D model, obtain the OCC values of these sample query points. The OCC of the sample query points inside the target 3D model is 1, and the OCC of the sample query points outside the target 3D model is 0.

[0105] Step 3: Train the initial occupancy function prediction model with the following specific configuration: Use adam as the optimizer, with an initial learning rate of 0.001, a batch size of 32, and use binary cross entropy as the loss function.

[0106] Step 4: Use the model trained above to predict and extract the mesh on the test set. As an example, first construct a voxel grid of (32, 32, 32), input its normalized 3D coordinates into the trained occupancy function prediction model to obtain the OCC values of the voxels in this grid. Then use the marching cube algorithm to extract the corresponding mesh. Finally, according to the recorded translation parameters and scaling parameters, restore the mesh to the real size and pose. The final predicted 3D model is as Figure 7 shown. The black mesh is the predicted 3D model, the white mesh is the target 3D model, and the white points are 3D point cloud data. Obviously, the predicted 3D model obtained by the 3D reconstruction method in this embodiment has high accuracy.

[0107] The embodiment of the present application also provides an application scenario that applies the above-mentioned 3D reconstruction method for cardiac chambers based on echocardiography. Specifically, the 3D reconstruction method for cardiac chambers based on echocardiography provided in this embodiment can be applied in the 3D reconstruction and release scenario of cardiac chambers. The 3D reconstruction and release scenario of cardiac chambers includes an image acquisition link, a 3D reconstruction link, and a release link. The image acquisition link is used to acquire the echocardiogram of the cardiac chamber to be reconstructed. The 3D reconstruction link is used to obtain the 3D model of the cardiac chamber to be reconstructed based on the echocardiogram. The release link is used to release the 3D model. The 3D reconstruction method for cardiac chambers based on echocardiography provided in this embodiment belongs to the 3D reconstruction link.

[0108] Embodiment 2

[0109] Based on the same inventive concept, an embodiment of the present application further provides an echocardiogram-based three-dimensional reconstruction device for a cardiac chamber for implementing the above-mentioned echocardiogram-based three-dimensional reconstruction method of a cardiac chamber. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in the embodiment of the echocardiogram-based three-dimensional reconstruction device for a cardiac chamber provided below can refer to the limitations on the echocardiogram-based three-dimensional reconstruction method of a cardiac chamber in the above text, and will not be elaborated here.

[0110] As Figure 8 shown, this embodiment provides an echocardiogram-based three-dimensional reconstruction device for a cardiac chamber. The echocardiogram-based three-dimensional reconstruction device for a cardiac chamber includes:

[0111] A point cloud generation module M1, configured to generate three-dimensional point cloud data based on an echocardiogram of a cardiac chamber to be reconstructed; the three-dimensional point cloud data includes a plurality of three-dimensional points and the three-dimensional coordinates and categories of each three-dimensional point; the three-dimensional points are anatomical structure position points of the cardiac chamber to be reconstructed, and the category is the anatomical structure category of the anatomical structure position points.

[0112] A normalization module M2, configured to perform normalization processing on the three-dimensional point cloud data to obtain normalized point cloud data; the three-dimensional coordinates of the center point of the normalized point cloud data are the coordinate origin, and the three-dimensional coordinates of each three-dimensional point of the normalized point cloud data are within a preset range.

[0113] A query point determination module M3, configured to uniformly sample a preset three-dimensional region to obtain a plurality of query points, and perform normalization processing on the three-dimensional coordinates of all the query points to obtain the normalized three-dimensional coordinates of each query point; the normalized three-dimensional coordinates of the center points of all the query points are the coordinate origin, and the normalized three-dimensional coordinates are within a preset range.

[0114] A model prediction module M4, configured to use the normalized point cloud data and the normalized three-dimensional coordinates of each query point as inputs, and use a trained occupancy function prediction model to obtain the occupancy result of each query point; the occupancy result includes that the query point is located inside the cardiac chamber to be reconstructed, and the query point is located outside the cardiac chamber to be reconstructed.

[0115] A three-dimensional reconstruction module M5, configured to perform three-dimensional reconstruction on the cardiac chamber to be reconstructed based on the normalized three-dimensional coordinates and occupancy results of all the query points to obtain an initial three-dimensional model of the cardiac chamber to be reconstructed.

[0116] A model restoration module M6, configured to perform anti-normalization processing on the initial three-dimensional model to obtain a three-dimensional model of the cardiac chamber to be reconstructed.

[0117] Embodiment 3

[0118] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 9 . The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store echocardiograms. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a three-dimensional reconstruction method of cardiac chambers based on echocardiograms.

[0119] Those skilled in the art can understand that Figure 9 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0120] In an exemplary embodiment, a computer device is further provided, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement the three-dimensional reconstruction method of cardiac chambers based on echocardiograms described in Embodiment 1.

[0121] Embodiment 4

[0122] The embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the three-dimensional reconstruction method of cardiac chambers based on echocardiograms described in Embodiment 1.

[0123] Embodiment 5

[0124] The embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the three-dimensional reconstruction method of cardiac chambers based on echocardiograms described in Embodiment 1.

[0125] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0126] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0127] Specific examples are used in this article to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for three-dimensional reconstruction of cardiac chambers based on echocardiography, characterized in that: The method for three-dimensional reconstruction of cardiac chambers based on echocardiography comprises: Generate three-dimensional point cloud data based on the echocardiogram of the heart chamber to be reconstructed; the three-dimensional point cloud data includes a plurality of three-dimensional points and the three-dimensional coordinates and category of each three-dimensional point; the three-dimensional point is an anatomical structure position point of the heart chamber to be reconstructed, and the category is the anatomical structure category of the anatomical structure position point; Normalizing the three-dimensional point cloud data to obtain normalized point cloud data; the three-dimensional coordinates of the center point of the normalized point cloud data are the coordinate origin, and the three-dimensional coordinates of each three-dimensional point of the normalized point cloud data are within a preset range; Uniformly sampling a preset three-dimensional area to obtain a plurality of query points, and normalizing the three-dimensional coordinates of all the query points to obtain normalized three-dimensional coordinates of each query point; the normalized three-dimensional coordinates of the center point of all the query points are the coordinate origin, and the normalized three-dimensional coordinates are within a preset range; Taking the normalized point cloud data and the normalized three-dimensional coordinates of each query point as input, using the trained occupancy function prediction model to obtain an occupancy result of each query point; the occupancy result includes that the query point is located inside the heart chamber to be reconstructed, and that the query point is located outside the heart chamber to be reconstructed; Performing three-dimensional reconstruction of the heart chamber to be reconstructed based on the normalized three-dimensional coordinates and occupancy results of all the query points to obtain an initial three-dimensional model of the heart chamber to be reconstructed; The initial three-dimensional model is subjected to denormalization processing to obtain a three-dimensional model of the heart chamber to be reconstructed.

2. The method for three-dimensional reconstruction of cardiac chambers based on echocardiography according to claim 1, characterized in that: The trained occupancy function prediction model includes an encoder, a fusion module and a decoder connected in sequence; The encoder is used to process the normalized point cloud data to obtain point cloud features; The fusion module is used to process the point cloud features and the normalized three-dimensional coordinates of each query point to obtain fusion features; The decoder is used to process the fused features to obtain an occupancy result of each query point.

3. The method for three-dimensional reconstruction of cardiac chambers based on echocardiography according to claim 2, characterized in that: The encoder includes a plurality of first multi-layer perceptron pooling modules and a multi-layer perceptron connected in sequence; the first multi-layer perceptron pooling module includes a multi-layer perceptron and a pooling layer connected in sequence; The fusion module includes a second multi-layer perceptron pooling module and a multi-layer perceptron connected in sequence, and the second multi-layer perceptron pooling module includes a multi-layer perceptron and a pooling layer connected in parallel; or, the fusion module includes a feature projection layer, a segmentation network, a feature query layer and a pooling layer connected in sequence; wherein the feature projection layer is used to project the point cloud features onto the xoy plane, the xoz plane and the yoz plane respectively to obtain three plane vectors; the segmentation network is used to process the three plane vectors to obtain three plane features; the feature query layer is used to determine the query point feature of each query point based on the normalized three-dimensional coordinates of each query point and the three plane features; The decoder includes a plurality of multi-layer perceptrons connected in sequence.

4. The method for three-dimensional reconstruction of cardiac chambers based on echocardiography according to claim 1, characterized in that: Generate 3D point cloud data based on the echocardiogram of the heart chamber to be reconstructed, including: Determining the two-dimensional coordinates of each anatomical structure location point based on an echocardiogram of the heart chamber to be reconstructed; For each of the anatomical structure position points, the two-dimensional coordinates of the anatomical structure position point are transformed using the three-dimensional posture information of the ultrasound probe when the echocardiogram is generated to obtain the three-dimensional coordinates of the anatomical structure position point; the anatomical structure position point is recorded as a three-dimensional point, the three-dimensional coordinates of the anatomical structure position point are marked as the three-dimensional coordinates of the three-dimensional point, and the anatomical structure category of the anatomical structure position point is recorded as the category of the three-dimensional point, and all of the three-dimensional points and the three-dimensional coordinates and category of each of the three-dimensional points are combined into three-dimensional point cloud data.

5. The method for three-dimensional reconstruction of cardiac chambers based on echocardiography according to claim 1, characterized in that: Before using the normalized point cloud data and the normalized three-dimensional coordinates of each query point as input and obtaining the occupancy result of each query point using the trained occupancy function prediction model, the echocardiogram-based three-dimensional reconstruction method of the cardiac chamber further includes: Acquire a data set; the data set includes a plurality of sample normalized point cloud data and a sample initial three-dimensional model corresponding to each of the sample normalized point cloud data; For each of the sample normalized point cloud data, a sampling area is determined based on the sample normalized point cloud data, and the sampling area is randomly sampled to obtain a plurality of sample query points; for each of the sample query points, an actual occupancy result of the sample query point is determined based on the three-dimensional coordinates of the sample query point and the sample initial three-dimensional model corresponding to the sample normalized point cloud data; The sample normalized point cloud data and the three-dimensional coordinates of each sample query point are used as input, and the actual occupancy result of each sample query point is used as a label to train the initial occupancy function prediction model to obtain a trained occupancy function prediction model.

6. The method for three-dimensional reconstruction of cardiac chambers based on echocardiography according to claim 1, characterized in that: Performing three-dimensional reconstruction of the heart chamber to be reconstructed based on the normalized three-dimensional coordinates and occupancy results of all the query points to obtain an initial three-dimensional model of the heart chamber to be reconstructed, specifically including: The normalized three-dimensional coordinates and occupancy results of all the query points are used as input, and the marching cube algorithm is used to perform three-dimensional reconstruction on the heart chamber to be reconstructed, so as to obtain an initial three-dimensional model of the heart chamber to be reconstructed.

7. A device for three-dimensional reconstruction of cardiac chambers based on echocardiography, characterized in that: The cardiac chamber three-dimensional reconstruction device based on echocardiography comprises: A point cloud generation module, for generating three-dimensional point cloud data based on an echocardiogram of a heart chamber to be reconstructed; the three-dimensional point cloud data comprises a plurality of three-dimensional points and a three-dimensional coordinate and a category of each three-dimensional point; the three-dimensional point is an anatomical structure position point of the heart chamber to be reconstructed, and the category is an anatomical structure category of the anatomical structure position point; A normalization module, used for normalizing the three-dimensional point cloud data to obtain normalized point cloud data; the three-dimensional coordinates of the center point of the normalized point cloud data are the coordinate origin, and the three-dimensional coordinates of each three-dimensional point of the normalized point cloud data are within a preset range; A query point determination module is used to uniformly sample a preset three-dimensional area to obtain a plurality of query points, and normalize the three-dimensional coordinates of all the query points to obtain the normalized three-dimensional coordinates of each query point; the normalized three-dimensional coordinates of the center point of all the query points are the coordinate origin, and the normalized three-dimensional coordinates are within a preset range; a model prediction module, configured to use the normalized point cloud data and the normalized three-dimensional coordinates of each query point as input, and obtain an occupancy result of each query point using a trained occupancy function prediction model; the occupancy result includes that the query point is located inside the heart chamber to be reconstructed, and that the query point is located outside the heart chamber to be reconstructed; A three-dimensional reconstruction module, used for performing three-dimensional reconstruction of the heart chamber to be reconstructed based on the normalized three-dimensional coordinates and occupancy results of all the query points to obtain an initial three-dimensional model of the heart chamber to be reconstructed; The model recovery module is used to perform a denormalization process on the initial three-dimensional model to obtain a three-dimensional model of the heart chamber to be reconstructed.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for three-dimensional reconstruction of cardiac chambers based on echocardiography as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for three-dimensional reconstruction of cardiac chambers based on echocardiography described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for three-dimensional reconstruction of cardiac chambers based on echocardiography described in any one of claims 1 to 6 is implemented.

Images