Search

CN-121982329-A - Ultrasonic characteristic point extraction method, and 3D imaging method and device based on double-layer catheter

CN121982329ACN 121982329 ACN121982329 ACN 121982329ACN-121982329-A

Abstract

The invention relates to the technical field of ultrasonic image processing, which comprises the steps of calculating three different feature algorithms on an obtained ultrasonic training image to obtain a clear ultrasonic feature point image, processing the ultrasonic training image based on a preset first feature algorithm to obtain a first ultrasonic image, processing the first ultrasonic image based on a preset second feature algorithm to obtain a second ultrasonic image, processing the second ultrasonic image based on a preset third feature algorithm to obtain an enhanced ultrasonic image, processing the ultrasonic training image through the obtained enhanced ultrasonic image and an initial feature extraction model to obtain an initial feature image determination loss function, and improving the quality of 2D feature points through the three feature algorithms so as to solve the problem that the feature point extraction of the 2D ultrasonic image is not accurate enough.

Inventors

  • WANG ZHIHUA

Assignees

  • 杭州心影医疗科技有限公司

Dates

Publication Date
20260505
Application Date
20251210

Claims (11)

  1. 1. An ultrasonic image feature point extraction method, which is characterized by comprising the following steps: acquiring a plurality of ultrasonic training images, and training an initial feature extraction model by using the plurality of ultrasonic training images to obtain a trained feature extraction model, wherein the feature extraction model is used for extracting feature points from the ultrasonic images; Wherein, in the training process of the initial feature extraction model: processing the ultrasonic training image based on a preset first characteristic algorithm to obtain a first ultrasonic image, processing the first ultrasonic image based on a preset second characteristic algorithm to obtain a second ultrasonic image, and processing the second ultrasonic image based on a preset third characteristic algorithm to obtain an enhanced ultrasonic image; Respectively processing the ultrasonic training image and the corresponding enhanced ultrasonic image based on a preset initial feature extraction model to obtain an initial feature image and an enhanced feature image; And determining a loss function according to the initial characteristic image and the enhanced characteristic image, and restricting the training process of the initial characteristic extraction model by using the loss function.
  2. 2. The method for extracting feature points of an ultrasound image according to claim 1, wherein the processing the ultrasound training image based on the preset first feature algorithm to obtain a first ultrasound image includes: aiming at the ultrasonic training image, dividing a central area of each first target pixel point in the ultrasonic training image to obtain a first target neighborhood which corresponds to the first target pixel point serving as the center; For each first target pixel point, determining the brightness difference between each pixel point in the first target neighborhood and the first target pixel point; determining the edge intensity difference between each pixel point in the first target adjacent area and the first target pixel point, and calculating according to the brightness difference and the edge intensity difference through a weight function to obtain a self-adaptive weight corresponding to the first target pixel point; for each first target pixel point, determining the occurrence probability of all brightness values in the first target neighborhood to obtain a first target neighborhood brightness probability value; calculating through an integration function according to the first target neighborhood brightness probability value to obtain texture entropy corresponding to the first target pixel point; and obtaining a first ultrasonic image according to the self-adaptive weights and the texture entropy corresponding to all the first target pixel points.
  3. 3. The method for extracting feature points of an ultrasound image according to claim 2, wherein the processing the first ultrasound image based on the preset second feature algorithm to obtain a second ultrasound image includes: Performing center region division on each second target pixel point in the first ultrasonic image to obtain a second target neighborhood which corresponds to the second target pixel point serving as the center; For each second target pixel point, determining an average gray value of a second target neighborhood corresponding to the second target pixel point, processing the gray value of each pixel point in the second target neighborhood and the average gray value of the second target neighborhood to obtain a second target neighborhood gray dispersion; for each second target pixel point, determining the distance between the pixel point farthest from the second target pixel point in the second target adjacent area corresponding to the second target pixel point and the second target pixel point to obtain the farthest target distance; For each second target pixel point, determining an included angle between each pixel point in a second target adjacent area corresponding to the second target pixel point and the second target pixel point to obtain target pixel included angle data; and obtaining a second ultrasonic image according to the second target neighborhood contrast difference value, the spatial distance attenuation weight and the edge direction fitting weight corresponding to all the second target pixel points.
  4. 4. The method for extracting feature points of an ultrasound image according to claim 3, wherein the processing the second ultrasound image based on a preset third feature algorithm to obtain an enhanced ultrasound image comprises: performing a separation operation on the second ultrasonic image to obtain a separation information ultrasonic image; Determining a corresponding adjusting coefficient according to the separation information ultrasonic image, and processing the second ultrasonic image through the adjusting coefficient to obtain an effective characteristic image; and processing the separation information ultrasonic image through a preset counteracting coefficient to obtain an invalid characteristic image, and obtaining an enhanced ultrasonic image according to the valid characteristic image and the invalid characteristic image.
  5. 5. The method of claim 4, wherein the using the loss function to perform constraint training on an initial feature extraction model comprises: Based on the convergence condition in training, feeding back the training process of the initial feature extraction model; Processing through a loss function according to the initial characteristic image and the enhanced characteristic image to obtain image characteristic parameters; when the image characteristic parameters do not accord with the corresponding convergence conditions, the corresponding algorithm parameters are adjusted to continuously train the characteristic extraction model; And when the image characteristic parameters accord with the corresponding convergence conditions, obtaining a trained characteristic extraction model.
  6. 6. A method of bilayer catheter-based 3D imaging, the method comprising: Acquiring an ultrasonic image to be processed, and recording a time node corresponding to the image to be processed to obtain an image time data set; Inputting an ultrasonic image to be processed into the trained feature extraction model according to any one of claims 1-5 to obtain a processed feature ultrasonic image; Obtaining a projection image through a convolution algorithm according to the processed characteristic ultrasonic image; performing projection establishment on the projection images to obtain associated feature information between the projection images corresponding to each processed feature image; obtaining 3D ultrasonic images of the corresponding images from all the associated characteristic information through a 3D reconstruction algorithm; And calculating through a dynamic matching algorithm according to the image time data set and the 3D ultrasonic image of the corresponding image to obtain a 3D dynamic model.
  7. 7. The dual catheter-based 3D imaging method of claim 6, wherein the catheter for acquiring the image to be processed comprises an inner catheter and an outer catheter, the inner catheter having a cardiac ultrasound image acquisition device disposed thereon and the outer catheter having a cardiac beat follow-up rotation device disposed thereon, the method further comprising, prior to acquiring the image to be processed: determining an electronic automatic controller on the heart ultrasonic image acquisition device; obtaining different three-dimensional rotation angles of the heart ultrasonic image acquisition device according to the electronic automatic controller, wherein the electronic automatic controller is used for inputting a control instruction for adjusting the angle; According to the heart beating follow-up rotating device on the outer catheter, the heart ultrasonic image acquisition system periodically and automatically rotates along with the heart cycle of the heart beating; Transmitting and receiving ultrasonic beams through a heart ultrasonic image acquisition device to obtain an ultrasonic beam data set, and obtaining an ultrasonic image to be processed according to the ultrasonic beam data set.
  8. 8. An ultrasound image feature point extraction apparatus, characterized in that the apparatus comprises: the model training module is used for acquiring a plurality of ultrasonic training images, training the initial feature extraction model by using the plurality of ultrasonic training images to obtain a trained feature extraction model, wherein the feature extraction model is used for extracting feature points from the ultrasonic images; Wherein, in the training process of the initial feature extraction model: processing the ultrasonic training image based on a preset first characteristic algorithm to obtain a first ultrasonic image, processing the first ultrasonic image based on a preset second characteristic algorithm to obtain a second ultrasonic image, and processing the second ultrasonic image based on a preset third characteristic algorithm to obtain an enhanced ultrasonic image; Respectively processing the ultrasonic training image and the corresponding enhanced ultrasonic image based on a preset initial feature extraction model to obtain an initial feature image and an enhanced feature image; And determining a loss function according to the initial characteristic image and the enhanced characteristic image, and restricting the training process of the initial characteristic extraction model by using the loss function.
  9. 9. A dual layer catheter-based 3D imaging device, the device comprising: The data acquisition module is used for acquiring an ultrasonic image to be processed, and recording a time node corresponding to the image to be processed to obtain an image time data set; The data processing module is used for inputting an ultrasonic image to be processed into the trained feature extraction model according to any one of claims 1-5 to obtain a processed feature ultrasonic image; The 3D image reconstruction module is used for obtaining projection images through a convolution algorithm according to the processed characteristic ultrasonic images, carrying out projection establishment on the projection images to obtain associated characteristic information among the projection images corresponding to each processed characteristic image, and obtaining 3D ultrasonic images of the corresponding images through a 3D reconstruction algorithm on all the associated characteristic information; the dynamic model acquisition module is used for calculating through a dynamic matching algorithm according to the image time dataset and the 3D ultrasonic image of the corresponding image to obtain a 3D dynamic model.
  10. 10. An apparatus comprising a memory and a processor, wherein the apparatus comprises a memory storing executable program code, and a processor coupled to the memory, the processor invoking the executable program code stored in the memory to perform the ultrasound image feature point extraction method of any of claims 1-5 or the bi-layer catheter based 3D imaging method of claims 6-7.
  11. 11. A computer storage medium storing computer instructions for performing the ultrasound image feature point extraction method of any one of claims 1-5 or the bilayer catheter-based 3D imaging method of claims 6-7 when invoked by a processor.

Description

Ultrasonic characteristic point extraction method, and 3D imaging method and device based on double-layer catheter Technical Field The invention relates to the technical field of ultrasonic image processing, in particular to an ultrasonic characteristic point extraction method, a 3D imaging method and a device based on a double-layer catheter. Background The heart is a core organ of the human circulatory system, and the evaluation of the structure and the function thereof is an important link of cardiovascular disease diagnosis, treatment planning and prognosis judgment. Traditional analysis of the heart relies on two-dimensional (2D) medical imaging techniques such as ultrasound, computed Tomography (CT), magnetic Resonance Imaging (MRI) and the like, which, although providing certain structural information, are not clear enough for the representation of the detailed features of the complex three-dimensional (3D) spatial configuration and dynamic motion of the heart. Along with the deep fusion of medical imaging and computer technology, 3D heart modeling technology appears in the market, and 3D heart modeling technology can construct a dynamic three-dimensional heart three-dimensional model by integrating multi-source image data, so that more visual and comprehensive diagnosis basis is provided for clinic, and the method has important application in specific scenes, for example, irreplaceable value is shown in scenes such as congenital heart disease screening, coronary heart disease interventional therapy planning, heart valve disease evaluation and the like. The core of the 3D dynamic heart modeling is to realize accurate projection from 2D image data to a 3D structure, and a 2D characteristic point extraction technology is an important link for acquiring the 3D dynamic heart modeling. The 2D feature points of the heart generally refer to key positions with clear anatomical meaning in the image, such as apex of the heart, apex of the annulus of the atrioventricular valve, coronary artery opening, turning points of ventricular wall, and the like, and these feature points are not only "anchor points" for heart structure identification, but also core basis for accurate registration of images, structure segmentation and morphological reconstruction when constructing a 3D dynamic model. When 2D feature points are not extracted accurately enough, 3D dynamic heart modeling is faced with the problems of fuzzy structural positioning, morphological distortion and the like, and the clinical application value of the model is directly affected. Practice shows that the feature point quality of the currently processed 2D ultrasonic image is low, so that the feature point extraction of the 2D ultrasonic image is not accurate enough, a new algorithm is required to be provided to solve the quality problem of the feature point of the 2D ultrasonic image, and the quality of the 2D feature point is improved through the new algorithm, so that the problem that the feature point extraction of the 2D ultrasonic image is not accurate enough is solved. Disclosure of Invention The invention aims to provide an ultrasonic image feature point extraction method which is used for improving the accuracy of ultrasonic image feature point extraction. In order to solve the technical problem, the first aspect of the present invention discloses an ultrasonic image feature point extraction method, which comprises the following steps: acquiring a plurality of ultrasonic training images, and training an initial feature extraction model by using the plurality of ultrasonic training images to obtain a trained feature extraction model, wherein the feature extraction model is used for extracting feature points from the ultrasonic images; Wherein, in the training process of the initial feature extraction model: processing the ultrasonic training image based on a preset first characteristic algorithm to obtain a first ultrasonic image, processing the first ultrasonic image based on a preset second characteristic algorithm to obtain a second ultrasonic image, and processing the second ultrasonic image based on a preset third characteristic algorithm to obtain an enhanced ultrasonic image; Respectively processing the ultrasonic training image and the corresponding enhanced ultrasonic image based on a preset initial feature extraction model to obtain an initial feature image and an enhanced feature image; And determining a loss function according to the initial characteristic image and the enhanced characteristic image, and restricting the training process of the initial characteristic extraction model by using the loss function. In a first aspect of the present invention, the processing the ultrasound training image based on the preset first feature algorithm to obtain a first ultrasound image includes: aiming at the ultrasonic training image, dividing a central area of each first target pixel point in the ultrasonic training image to obtain a first targ