Registration algorithm for preoperative CTMR scanning image and intraoperative patient coordinate
By designing a registration algorithm combining adaptive filtering, grayscale mapping, scale-invariant feature transformation, singular value decomposition and Levenberg-Marquardt algorithm, the problems of low registration accuracy and poor robustness of CTMR images in the prior art are solved, and a higher accuracy and robustness registration effect is achieved.
Patent Information
- Application Number
- CN202510016279.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-06
- Publication Date
- 2025-05-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art has problems such as low accuracy, poor robustness and insufficient adaptability to complex environments in the registration of preoperative CTMR scan images and intraoperative patient coordinates.
A registration algorithm including adaptive filtering algorithm, a grayscale mapping method based on region features, a scale-invariant feature transformation algorithm, a singular value decomposition rigid transformation algorithm and a Levenberg-Marquardt optimization algorithm are designed. This algorithm improves registration accuracy and robustness by removing noise, realizing grayscale normalization, extracting feature points, performing initial registration and iterative optimization, and combining with a dynamic weight adjustment mechanism.
Higher precision image registration is achieved, the robustness of the algorithm and adaptability to complex environments are enhanced, and the sub-millimeter-level accuracy and stability are ensured in complex electromagnetic environments.
Smart Images
Figure CN119941811A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing technology, and more specifically, to a registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates. Background Art
[0002] With the rapid development of medical imaging technology, CT (computed tomography) and MR (magnetic resonance imaging) have become indispensable means in clinical diagnosis. CT has advantages in displaying bone structure and calcification due to its high-density resolution and rapid imaging capability; while MR is better in displaying soft tissue structures such as muscles, ligaments, and nerves due to its high soft tissue resolution and no radiation. These two imaging technologies have their own characteristics, but in clinical applications, they often need to be combined to obtain more comprehensive diagnostic information.
[0003] However, the fusion and registration of CT and MR images face a series of technical challenges. First, the image features of the two modalities are quite different, such as contrast, texture and structural information, which makes direct image registration complicated. Second, the patient's position changes, breathing, heartbeat and other physiological movements during surgery cause the actual anatomical structure to be incompletely corresponding to the preoperative image, which increases the difficulty of registration. In addition, the existing registration algorithms still have deficiencies in accuracy, robustness and automation. Especially when processing multimodal images, how to effectively combine the advantages of the two modalities to improve the accuracy and efficiency of registration is a hot topic of current research.
[0004] Therefore, the existing technology has problems such as low registration accuracy, poor robustness and insufficient adaptability to complex environments. Summary of the invention
[0005] In order to overcome the problems of low registration accuracy, poor robustness and insufficient adaptability to complex environments in the prior art, the present invention designs a registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates that can effectively solve the above technical problems.
[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows:
[0007] A registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates includes the following steps:
[0008] The CT / MR scan images of the patient before the operation are obtained respectively, and the patient's body position coordinate information is collected in real time through the positioning sensor during the operation;
[0009] Adopting an adaptive filtering algorithm to remove the scan image noise from the acquired CT / MR scan image, and realizing grayscale normalization through a grayscale mapping method based on regional features, so that the grayscale values of scan images of different modalities are comparable;
[0010] Extracting feature points from the preprocessed CT / MR scan image using a scale-invariant feature transformation algorithm, wherein the scale-invariant feature transformation algorithm extracts key bone landmarks as spatial positioning features for the body position coordinate information of the patient during surgery by introducing local texture direction constraints;
[0011] Based on the extracted feature points, an initial registration is performed using a rigid transformation algorithm based on singular value decomposition to calculate an initial transformation matrix, to preliminarily align the CT / MR scan images, and to preliminarily map the body position coordinate information of the patient during the operation to the image space;
[0012] The Levenberg-Marquardt algorithm is adopted as the optimization strategy, combined with the body position coordinate information updated in real time during the operation, the distance between the feature points and the bone landmark points is used as the error metric, and the initial transformation matrix is iteratively optimized. A dynamic weight adjustment mechanism is introduced in the iterative process, and different weights are assigned according to the credibility and distance of the feature points to improve the registration accuracy.
[0013] Preferably, the adaptive filtering algorithm is a denoising method based on wavelet transform.
[0014] Preferably, the grayscale mapping method based on regional features includes histogram equalization and contrast-limited adaptive histogram equalization.
[0015] Preferably, the SIFT algorithm further comprises:
[0016] constructing a Gaussian difference scale space for the preprocessed CT / MR scan image, and detecting extreme value points in the Gaussian difference scale space;
[0017] Performing curve fitting on the detected extreme value points to obtain the positions and scales of the feature points;
[0018] Calculating the local gradient direction of the CT / MR scan image;
[0019] Selecting the feature points according to the local gradient direction and the local texture intensity, and assigning one or more main directions to each of the feature points;
[0020] Constructing the feature descriptor of the feature point includes selecting a region around each feature point and dividing it into a plurality of sub-regions, calculating the gradient direction histogram of each sub-region, and finally forming the feature descriptor of the feature point.
[0021] Preferably, the rigid transformation algorithm based on singular value decomposition further comprises:
[0022] Calculating the correspondence between the feature points;
[0023] Constructing a covariance matrix corresponding to the feature points;
[0024] Decompose the covariance matrix through SVD and calculate the rotation matrix and translation vector;
[0025] The calculated rotation matrix and translation vector are used to transform the preoperative CT / MR scan image to achieve initial alignment with the patient's body position coordinate information during the operation.
[0026] Preferably, the Levenberg-Marquardt algorithm further comprises:
[0027] Calculate the objective function value and Jacobian matrix under the current transformation matrix;
[0028] Calculate the Levenberg-Marquardt damping parameter according to the objective function value and the Jacobian matrix;
[0029] updating the transformation matrix according to the damping parameter to reduce the objective function value;
[0030] Repeat the above steps until the predetermined convergence condition or number of iterations is reached.
[0031] Preferably, the dynamic weight adjustment mechanism further comprises:
[0032] Evaluating the credibility of each of the feature points;
[0033] Dynamically adjust the weight according to the credibility of each feature point and the distance between them and the corresponding point, wherein the feature point with high credibility is assigned a higher weight, and the feature point with close distance is assigned a higher weight than the feature point with far distance;
[0034] The adjusted weights are applied in an iterative process to optimize the transformation matrix.
[0035] Compared with the prior art, the beneficial effects of the present invention are as follows: image noise is removed by an adaptive filtering algorithm and wavelet transform, grayscale normalization is achieved by combining a grayscale mapping method based on regional features, the comparability of grayscale values of scanned images of different modalities is improved, feature points are extracted by using a scale-invariant feature transform (SIFT) algorithm, and local texture direction constraints are introduced to enhance the recognition and stability of feature points, singular value decomposition (SVD) is used for initial registration, and the transformation matrix is iteratively optimized by a Levenberg-Marquardt algorithm, and more accurate registration is achieved by combining body position coordinate information updated in real time during surgery, a dynamic weight adjustment mechanism assigns different weights according to the credibility and distance of feature points, and the robustness of registration is improved, especially in complex electromagnetic environments, the submillimeter accuracy and stability of the positioning sensor ensure the reliability of the algorithm, in summary, the technical solution not only improves the registration accuracy, but also enhances the robustness of the algorithm and its adaptability to complex environments, providing strong technical support for precision medicine. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived based on the provided drawings without paying any creative work.
[0037] Figure 1 A step-by-step diagram of an algorithm for registering preoperative CTMR scan images with intraoperative patient coordinates. DETAILED DESCRIPTION
[0038] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;
[0039] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;
[0040] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0041] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0042] Example
[0043] A registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates, such as Figure 1 As shown, the following steps are included:
[0044] The CT / MR scan images of the patient before the operation are obtained respectively, and the patient's body position coordinate information is collected in real time through the positioning sensor during the operation;
[0045] Adopting an adaptive filtering algorithm to remove the scan image noise from the acquired CT / MR scan image, and realizing grayscale normalization through a grayscale mapping method based on regional features, so that the grayscale values of scan images of different modalities are comparable;
[0046] Extracting feature points from the preprocessed CT / MR scan image using a scale-invariant feature transformation algorithm, wherein the scale-invariant feature transformation algorithm extracts key bone landmarks as spatial positioning features for the body position coordinate information of the patient during surgery by introducing local texture direction constraints;
[0047] Based on the extracted feature points, an initial registration is performed using a rigid transformation algorithm based on singular value decomposition to calculate an initial transformation matrix, to preliminarily align the CT / MR scan images, and to preliminarily map the body position coordinate information of the patient during the operation to the image space;
[0048] The Levenberg-Marquardt algorithm is adopted as the optimization strategy, combined with the body position coordinate information updated in real time during the operation, the distance between the feature points and the bone landmark points is used as the error metric, and the initial transformation matrix is iteratively optimized. A dynamic weight adjustment mechanism is introduced in the iterative process, and different weights are assigned according to the credibility and distance of the feature points to improve the registration accuracy.
[0049] The adaptive filtering algorithm is a denoising method based on wavelet transform.
[0050] The grayscale mapping method based on regional features includes histogram equalization and contrast-limited adaptive histogram equalization.
[0051] The SIFT algorithm further comprises:
[0052] constructing a Gaussian difference scale space for the preprocessed CT / MR scan image, and detecting extreme value points in the Gaussian difference scale space;
[0053] Performing curve fitting on the detected extreme value points to obtain the positions and scales of the feature points;
[0054] Calculating the local gradient direction of the CT / MR scan image;
[0055] Selecting the feature points according to the local gradient direction and the local texture intensity, and assigning one or more main directions to each of the feature points;
[0056] Constructing the feature descriptor of the feature point includes selecting a region around each feature point and dividing it into a plurality of sub-regions, calculating the gradient direction histogram of each sub-region, and finally forming the feature descriptor of the feature point.
[0057] The rigid transformation algorithm based on singular value decomposition further comprises:
[0058] Calculating the correspondence between the feature points;
[0059] Constructing a covariance matrix corresponding to the feature points;
[0060] Decompose the covariance matrix through SVD and calculate the rotation matrix and translation vector;
[0061] The calculated rotation matrix and translation vector are used to transform the preoperative CT / MR scan image to achieve initial alignment with the patient's body position coordinate information during the operation.
[0062] The Levenberg-Marquardt algorithm further comprises:
[0063] Calculate the objective function value and Jacobian matrix under the current transformation matrix;
[0064] Calculate the Levenberg-Marquardt damping parameter according to the objective function value and the Jacobian matrix;
[0065] updating the transformation matrix according to the damping parameter to reduce the objective function value;
[0066] Repeat the above steps until the predetermined convergence condition or number of iterations is reached.
[0067] The dynamic weight adjustment mechanism further includes:
[0068] Evaluating the credibility of each of the feature points;
[0069] Dynamically adjust the weight according to the credibility of each feature point and the distance between them and the corresponding point, wherein the feature point with high credibility is assigned a higher weight, and the feature point with close distance is assigned a higher weight than the feature point with far distance;
[0070] The adjusted weights are applied in an iterative process to optimize the transformation matrix.
[0071] In the specific implementation, professional equipment that can obtain high-quality preoperative CT and MR scan images of patients is prepared. At the same time, high-precision positioning sensors that can collect real-time patient intraoperative body position coordinate information are installed on the operating table in the operating room to ensure that the sensors can operate stably and accurately capture changes in body position.
[0072] Before the operation, the patient should undergo CT and MR scans as per the specifications to obtain the corresponding images. Pay attention to covering the surgical-related parts to ensure that the images are complete and clear. During the operation, the patient should be in place, the positioning sensor should be turned on, and the patient's body position coordinate information should be continuously collected, ready to provide data support for subsequent alignment at any time.
[0073] An adaptive filtering algorithm based on wavelet transform is used to remove noise from the acquired CT / MR scan images. The filtering parameters are dynamically adjusted by analyzing the image characteristics to retain the key details of the image. The grayscale of the image is normalized using a method including histogram equalization and contrast-limited adaptive histogram equalization, so that the grayscale values of scanned images of different modalities can be comparable to each other, which is convenient for subsequent operations.
[0074] The scale-invariant feature transform (SIFT) algorithm is used to process the preprocessed images. First, a Gaussian difference scale space is constructed, in which extreme points are carefully detected. Curve fitting is used to accurately determine the position and scale of the detected extreme points. The local gradient direction of the image is calculated, and feature points are selected based on local texture direction constraints and local texture intensity. The main direction is assigned, and the area around the feature points is divided into sub-regions. The gradient direction histogram of each sub-region is calculated, and a feature descriptor that can fully characterize the image features is constructed. According to the intraoperative patient's body position coordinate information, the key bone landmarks are extracted using appropriate methods and used as important spatial positioning features.
[0075] For the initial registration, based on the extracted feature points, we first find the correspondence between them, then construct the covariance matrix, calculate the rotation matrix and translation vector through singular value decomposition, and finally use these calculated parameters to transform the preoperative CT / MR scan images, preliminarily align the images, and preliminarily map the intraoperative patient's body position coordinate information to the image space.
[0076] Iterative optimization registration uses the Levenberg-Marquardt algorithm as the optimization strategy, combined with the real-time updated body coordinate information during surgery, and uses the distance between the feature points and the bone landmark points as the error metric to calculate the objective function value and Jacobian matrix under the current transformation matrix. The damping parameter is calculated based on the two, and then the transformation matrix is updated with the damping parameter to reduce the objective function value. The above steps are repeated until the predetermined convergence conditions are met, so that the preoperative image and the intraoperative patient coordinates are gradually and accurately matched.
[0077] Evaluate the credibility of each feature point, such as the stability and contrast of feature point extraction, and dynamically assign weights based on the credibility of feature points and the distance between them and corresponding points, so that feature points with high credibility and close distance have higher weights. In the iterative optimization process, use the adjusted weights to optimize the transformation matrix to improve the registration accuracy.
[0078] The same or similar reference numerals correspond to the same or similar components;
[0079] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;
[0080] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made on the basis of the above description. It is not necessary and impossible to list all the implementation methods here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.
Claims
1. A registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates, characterized in that: The steps include: The CT / MR scan images of the patient before the operation are obtained respectively, and the patient's body position coordinate information is collected in real time through the positioning sensor during the operation; Adopting an adaptive filtering algorithm to remove the scan image noise from the acquired CT / MR scan image, and realizing grayscale normalization through a grayscale mapping method based on regional features, so that the grayscale values of scan images of different modalities are comparable; Extracting feature points from the preprocessed CT / MR scan image using a scale-invariant feature transformation algorithm, wherein the scale-invariant feature transformation algorithm extracts key bone landmarks as spatial positioning features for the body position coordinate information of the patient during surgery by introducing local texture direction constraints; Based on the extracted feature points, an initial registration is performed using a rigid transformation algorithm based on singular value decomposition to calculate an initial transformation matrix, to preliminarily align the CT / MR scan images, and to preliminarily map the body position coordinate information of the patient during the operation to the image space; The Levenberg-Marquardt algorithm is adopted as the optimization strategy, combined with the body position coordinate information updated in real time during the operation, the distance between the feature points and the bone landmark points is used as the error metric, and the initial transformation matrix is iteratively optimized. A dynamic weight adjustment mechanism is introduced in the iterative process, and different weights are assigned according to the credibility and distance of the feature points to improve the registration accuracy.
2. The registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates according to claim 1, characterized in that: The adaptive filtering algorithm is a denoising method based on wavelet transform.
3. The registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates according to claim 1, characterized in that: The grayscale mapping method based on regional features includes histogram equalization and contrast-limited adaptive histogram equalization.
4. The registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates according to claim 1, characterized in that: The SIFT algorithm further comprises: constructing a Gaussian difference scale space for the preprocessed CT / MR scan image, and detecting extreme value points in the Gaussian difference scale space; Performing curve fitting on the detected extreme points to obtain the positions and scales of the feature points; Calculating the local gradient direction of the CT / MR scan image; Selecting the feature points according to the local gradient direction and the local texture intensity, and assigning one or more main directions to each of the feature points; Constructing the feature descriptor of the feature point includes selecting a region around each feature point and dividing it into a plurality of sub-regions, calculating the gradient direction histogram of each sub-region, and finally forming the feature descriptor of the feature point.
5. The registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates according to claim 1, characterized in that: The rigid transformation algorithm based on singular value decomposition further comprises: Calculating the correspondence between the feature points; Constructing a covariance matrix corresponding to the feature points; Decompose the covariance matrix through SVD and calculate the rotation matrix and translation vector; The calculated rotation matrix and translation vector are used to transform the preoperative CT / MR scan image to achieve initial alignment with the intraoperative patient's body position coordinate information.
6. The registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates according to claim 1, characterized in that: The Levenberg-Marquardt algorithm further comprises: Calculate the objective function value and Jacobian matrix under the current transformation matrix; Calculate the Levenberg-Marquardt damping parameter according to the objective function value and the Jacobian matrix; updating the transformation matrix according to the damping parameter to reduce the objective function value; Repeat the above steps until the predetermined convergence condition or number of iterations is reached.
7. The registration algorithm for preoperative CTMR scan images and intraoperative patient coordinates according to claim 1, characterized in that: The dynamic weight adjustment mechanism further includes: Evaluating the credibility of each of the feature points; Dynamically adjust the weight according to the credibility of each feature point and the distance between them and the corresponding point, wherein the feature point with high credibility is assigned a higher weight, and the feature point with close distance is assigned a higher weight than the feature point with far distance; The adjusted weights are applied in an iterative process to optimize the transformation matrix.