METHODS FOR CONSTRUCTING THREE-DIMENSIONAL ANATOMICAL MODELS OF THE HEART AND SYSTEMS FOR THREE-DIMENSIONAL CARDIAC MAPPING

Claims

1. A method for constructing a three-dimensional anatomical model of the heart based on point cloud data, including: obtaining an atrial template model, wherein the atrial template model is a three-dimensional mesh model of a predetermined boundary of the atrium and is provided with a plurality of predetermined control feature points, and the atrial template model is deformed based on the displacement of the control feature points; obtaining point cloud data of a simulated atrium inside the heart, wherein the point cloud data of the simulated atrium contains information on the positional coordinates of individual characteristic points inside the simulated atrium; mapping individual feature points in the simulated atrium to the atrial template model using an affine transformation network, thereby obtaining point cloud data mapped to the atrial template model; inputting the point cloud data mapped to the atrial template model into a prediction network to predict offsets of control feature points in the atrial template model; and obtaining a three-dimensional anatomical model of the simulated atrium by deforming the atrium template model based on the displacement of control characteristic points.

2. The method according to claim 1, wherein the offsets of the control characteristic points have a linear transformation relationship with the offsets of the grid vertices in the atrial template model.

3. The method according to claim 1, wherein obtaining data of a point cloud of a simulated atrium inside the heart includes: obtaining information about the positional coordinates of individual characteristic points inside the simulated atrium using an electrode, thereby obtaining point cloud data of the simulated atrium.

4. The method of claim 1, wherein matching individual feature points in the modeled atrium with the atrial template model using an affine transform network, thereby obtaining point cloud data matched with the atrial template model, comprises: normalization of the point cloud data of the simulated atrium; inputting the normalized point cloud data into a fully connected neural network to predict an affine transformation matrix between the simulated atrium and the template atrium model; and matching individual feature points in the simulated atrium with the atrial template model based on the predicted affine transformation matrix, thereby obtaining point cloud data matched with the atrial template model.

5. The method according to claim 4, wherein the fully connected neural network is trained using a backpropagation algorithm using a first loss function defined by the expression: L1=||V init -(T1V T groundtruth ) T || 2 2, where L1 denotes the value of the first loss function, V init denotes the matrix corresponding to the information about the positional coordinates of the mesh vertices in the atrial template model, T1 denotes the affine transformation matrix, V groundtruth denotes the matrix corresponding to the true values ​​of the coordinates of the mesh vertices in the three-dimensional network corresponding to the modeled atrium, and || || 2 2 denotes the square of the Euclidean norm.

6. The method according to paragraph 1, wherein: The prediction network contains multiple point set abstraction modules; Inputting the point cloud data mapped to the atrial template model into the prediction network to predict the offsets of the control feature points in the atrial template model involves: inputting the point cloud data mapped to the atrial template model into the first point set abstraction module, iteratively feeding the output of each previous point set abstraction module into the subsequent point set abstraction module sequentially, and obtaining the predicted offsets of the control feature points in the atrial template model as output from the last point set abstraction module; where each point set abstraction module performs operations that include: randomly sampling k points from the input data and treating each sampled point as a centroid point; for each centroid point, select m points within a spherical region of given radius centered at each specified centroid point; grouping the selected m sample points corresponding to each centroid point into a set of points; extracting feature information for each set of points and performing a max-pooling operation to obtain global feature information corresponding to each selected point; and output of the selected k points and global feature information corresponding to each selected point.

7. The method of claim 6, wherein the prediction network is trained using a backpropagation algorithm using a second loss function defined by the expression: L2=||V p -V groundtruth || 2 2, where L2 denotes the value of the second loss function, V p denotes the matrix corresponding to the predicted values ​​of the coordinates of the grid vertices in the three-dimensional grid model of the simulated atrium, V groundtruth denotes the matrix corresponding to the true values ​​of the coordinates of the mesh vertices in the three-dimensional anatomical model corresponding to the modeled atrium, and || || 2 2 denotes the square of the Euclidean norm.

8. A three-dimensional cardiac mapping system, including: a data acquisition device configured to acquire point cloud data of a simulated atrium within the heart; a processor configured to perform the method for constructing a three-dimensional anatomical model of the heart according to any one of claims 1-7, to output a three-dimensional anatomical model of the modeled atrium; and a display device configured to display a three-dimensional anatomical model of the simulated atrium.

9. The three-dimensional cardiac mapping system of claim 8, wherein the data collection device includes at least one of the following elements: an electrode catheter and an electrode pad.

10. A machine-readable storage medium on which a program is stored which, when executed by a processor, is capable of implementing the method according to any of paragraphs 1-7.