CN-121981936-A - Ankle landmark point identification method, electronic device and computer program product

CN121981936ACN 121981936 ACN121981936 ACN 121981936ACN-121981936-A

Abstract

The embodiment of the application is suitable for the technical field of computer-assisted medical treatment and image processing, and provides an ankle joint landmark point identification method, electronic equipment and a computer program product, wherein the method comprises the steps of determining a first position point of an ankle joint landmark point to be identified in three-dimensional image data; the method comprises the steps of determining a first position point positioned in a bone, determining a second position point corresponding to the first position point on a bone surface, determining a curvature value of the second position point on the bone surface, and optimizing the second position point based on the curvature value to obtain a target position point of the ankle joint landmark point on the bone surface. By adopting the method, the position of the projection point on the bone surface can be optimized based on the curvature of the projection point, so that a more accurate ankle joint position point is obtained, and the recognition accuracy of the ankle joint landmark point is improved.

Inventors

CHAI WEI
Meng Liaili
LI ANG
LIU XIANGDONG

Assignees

骨圣元化机器人(深圳)有限公司
中国人民解放军总医院第四医学中心

Dates

Publication Date: 20260505
Application Date: 20251127

Claims (10)

1. An ankle landmark point identification method, comprising: Determining a first position point of an ankle joint landmark point to be identified in three-dimensional image data; determining a second position point corresponding to the first position point on the bone surface aiming at the first position point positioned in the bone; Determining a curvature value at the second location point on the bone surface; and optimizing the second position point based on the curvature value to obtain a target position point of the ankle joint landmark point on the bone surface.
2. The method of claim 1, wherein optimizing the second location point based on the curvature value results in a target location point of the ankle landmark point on the bone surface, comprising: determining three-dimensional coordinates of the second position point on the bone surface, wherein the three-dimensional coordinates are coordinate values under a coordinate system corresponding to the three-dimensional image data; invoking a trained deep learning model, and importing a curvature value at the second position point and a three-dimensional coordinate of the second position point into the deep learning model to obtain a target coordinate, wherein the target coordinate is used for representing a target position point of the ankle joint landmark point on the bone surface; the deep learning model is a curvature-position probability model obtained by training a Gaussian model based on curvature distribution.
3. The method of claim 2, wherein the importing the curvature value at the second location point and the three-dimensional coordinates of the second location point into the deep learning model to obtain target coordinates comprises: Importing the curvature value at the second position point and the three-dimensional coordinates of the second position point into the deep learning model to obtain the coordinates of each position point and the corresponding probability value output by the deep learning model; Determining a plurality of candidate location points located within the second location point neighborhood; And determining a target position point with the maximum probability value from a plurality of candidate position points.
4. The method of claim 2, wherein prior to invoking the trained deep learning model, the method further comprises: When training the deep learning model, sampling in the neighborhood of each labeling point in sample data to obtain a plurality of candidate labeling points, and calculating curvature distribution based on the plurality of candidate labeling points; fitting the Gaussian model according to curvature distribution of the plurality of marked points; constructing a loss function based on model parameter values in the fitted Gaussian model, wherein the loss function is used for verifying the convergence of the model in the training process of the deep learning model.
5. The method of any one of claims 1 to 4, wherein the determining a curvature value at the second location point on the bone face comprises: Determining a plurality of computational scales; A curvature value of the second location point on the bone surface is calculated based on a plurality of the calculation scales.
6. The method of claim 5, wherein calculating a curvature value of the second location point on the bone surface based on the plurality of calculation scales comprises: Respectively calculating the curvature value of the second position point under each calculation scale; And calculating an average value of curvature values of the second position point at a plurality of calculation scales as the curvature value of the second position point on the bone surface.
7. The method of any one of claims 1 to 4 or 6, wherein after optimizing the second location point based on the curvature value to obtain a target location point of the ankle landmark point on the bone surface, the method further comprises: Determining anatomical constraints for a target location point on the bone surface characterized by the target coordinates; validating the target location point based on the anatomical constraint; And identifying the target position point as a corresponding ankle joint landmark point under the condition that the target position point passes verification.
8. The method of claim 7, wherein the anatomical constraint comprises an ankle mirrored constraint, the verifying the target location point based on the anatomical constraint comprising: Determining the ankle where the target position point is located, and determining a corresponding position point on the ankle on the other side corresponding to the target position point; Verifying whether the target position point and the corresponding position point meet the ankle joint mirror image constraint condition; And respectively identifying the target position point and the corresponding position point as ankle joint landmark points on corresponding lateral ankles under the condition that the target position point and the corresponding position point meet the ankle joint mirror image constraint condition.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the computer program, when executed by the processor, causes the electronic device to implement the method of any one of claims 1 to 8.
10. A computer program product comprising a computer program which, when run, causes the method of any one of claims 1 to 8 to be performed.

Description

Ankle landmark point identification method, electronic device and computer program product Technical Field The embodiment of the application belongs to the technical field of computer-aided medical treatment and image processing, and particularly relates to an ankle joint landmark point identification method, electronic equipment and a computer program product. Background The automatic identification of the human body joint landmark points has important application in the fields of medical image analysis, orthopedic operation planning, auxiliary diagnosis and the like. For example, in some procedures, it is desirable to identify the patient's ankle joint. The widely used method for identifying the center point of the femoral head in the prior art can be applied to the identification of the ankle joint landmark point. However, since the femoral head center is actually inside the bone and the ankle boundary point is on the bone surface, applying the method of identifying the femoral head center point to identify the ankle boundary point may result in output results that are not on the bone surface. The traditional approach is to project the output points directly onto the bone surface. For example, using the closest plane projection, the point in the identified bone is projected onto the bone plane so as to be closest to the bone plane, and the point obtained on the bone plane after the projection is used as the corresponding femoral head landmark point. However, the position point error obtained by directly adopting the nearest plane projection mode is also large. Disclosure of Invention In view of the foregoing, embodiments of the present application provide an ankle landmark point recognition method, an electronic device, and a computer program product, which are used for reducing a calculation error in a projection process of an ankle landmark point, and improving recognition accuracy of the ankle landmark point. A first aspect of an embodiment of the present application provides an ankle landmark point identification method, including: Determining a first position point of an ankle joint landmark point to be identified in three-dimensional image data; determining a second position point corresponding to the first position point on the bone surface aiming at the first position point positioned in the bone; Determining a curvature value at the second location point on the bone surface; and optimizing the second position point based on the curvature value to obtain a target position point of the ankle joint landmark point on the bone surface. Optionally, the optimizing the second location point based on the curvature value obtains a target location point of the ankle landmark point on the bone surface, including: determining three-dimensional coordinates of the second position point on the bone surface, wherein the three-dimensional coordinates are coordinate values under a coordinate system corresponding to the three-dimensional image data; invoking a trained deep learning model, and importing a curvature value at the second position point and a three-dimensional coordinate of the second position point into the deep learning model to obtain a target coordinate, wherein the target coordinate is used for representing a target position point of the ankle joint landmark point on the bone surface; the deep learning model is a curvature-position probability model obtained by training a Gaussian model based on curvature distribution. Optionally, the importing the curvature value at the second location point and the three-dimensional coordinate of the second location point into the deep learning model to obtain the target coordinate includes: Importing the curvature value at the second position point and the three-dimensional coordinates of the second position point into the deep learning model to obtain the coordinates of each position point and the corresponding probability value output by the deep learning model; Determining a plurality of candidate location points located within the second location point neighborhood; And determining a target position point with the maximum probability value from a plurality of candidate position points. Optionally, before invoking the trained deep learning model, the method further comprises: When training the deep learning model, sampling in the neighborhood of each labeling point in sample data to obtain a plurality of candidate labeling points, and calculating curvature distribution based on the plurality of candidate labeling points; fitting the Gaussian model according to curvature distribution of the plurality of marked points; constructing a loss function based on model parameter values in the fitted Gaussian model, wherein the loss function is used for verifying the convergence of the model in the training process of the deep learning model. Optionally, the determining a curvature value at the second location point on the bone surface includes: Determining a plurality