A navigation model dynamic updating method and system based on a portable MRI device
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- RENMIN HOSPITAL OF WUHAN UNIVERSITY (HUBEI GENERAL HOSPITAL)
- Filing Date
- 2026-06-03
- Publication Date
- 2026-07-14
AI Technical Summary
Existing medical navigation methods are prone to distortion when dealing with brain displacement, as the navigation model built based on preoperative MRI images is easily distorted, resulting in severe navigation drift. Furthermore, existing methods lack real-time performance and dynamic navigation accuracy in low-field environments.
A portable MRI device was used in conjunction with rigid and non-rigid registration methods to acquire the spatial position of optical markers in real time, establish real-time pose information, and dynamically update the 3D navigation model. This included rigid registration and confidence-weighted non-rigid registration to ensure that the model was consistent with the current tissue state.
It effectively avoids navigation drift, improves the safety and accuracy of medical procedures, and ensures the stability and real-time performance of navigation under low-field MRI conditions.
Smart Images
Figure CN122376263A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method and system for dynamically updating a navigation model based on a portable MRI device, belonging to the field of medical imaging and navigation technology. Background Technology
[0002] In medical procedures, it is necessary to establish an effective navigation model and an effective navigation method for devices entering the human body. Existing navigation methods are mainly based on preoperative high-field MRI images to establish navigation models. However, the position of the same area is not constant during actual operation. For example, brain shift can occur during neurosurgical operations, causing the navigation model based on preoperative MRI images to become distorted, resulting in severe drift and poor navigation effect.
[0003] While preoperative navigation models based on operational images such as low-field MRI / ultrasound can be corrected, they are mostly updated intermittently or offline, lacking real-time capability. Furthermore, in existing methods, optical tracking and image registration are often used independently, lacking fusion, and cannot guarantee dynamic navigation accuracy in low-field environments. Summary of the Invention
[0004] The technical problem that the invention aims to solve
[0005] This invention addresses the problems existing in the above-mentioned medical operation navigation methods by proposing a navigation model dynamic update method and system based on a portable MRI device. The method can continuously correct and dynamically update the navigation model based on MRI images of the operation process.
[0006] Technical solution
[0007] To achieve the above objectives, the technical solution provided by the present invention is as follows:
[0008] The method for dynamically updating the navigation model based on a portable MRI device includes the following steps:
[0009] Step 1: Before the operation, a high-field magnetic resonance imaging scan of the target area is performed using a portable MRI device to obtain high-resolution MRI image data covering the target area and process the images to establish a three-dimensional navigation model of the target area.
[0010] Step 2: Fix the rigid reference frame with multiple optical markers to the target area so that the rigid reference frame maintains a fixed spatial relationship with the target area during operation.
[0011] Step 3: During the operation, the spatial position of the optical markers on the rigid reference frame is acquired in real time. The real-time pose information of the target area is obtained based on the rigid geometric relationship between the markers. A three-dimensional coordinate system for synchronous motion during the operation is established based on the real-time pose information of the target area.
[0012] Step 4: During the operation phase, a portable MRI device is used to scan the target area to obtain MRI images that reflect the current tissue morphology, and then noise reduction and enhancement processing is performed.
[0013] Step 5: Based on the real-time pose information of the target area obtained in Step 3, perform rigid registration on the 3D navigation model established in Step 1. The method is as follows: construct a rotation matrix and a translation vector and apply them to the 3D navigation model established in Step 1 to achieve overall alignment between the 3D navigation model and the current surgical position.
[0014]
[0015] in Represents a point in a 3D navigation model. Let R represent the point after the transformation, where R is a 3x3 rotation matrix and T is a 3x1 translation vector.
[0016] Step 6, performing confidence-weighted non-rigid registration based on MRI images acquired during the operation phase, includes the following steps:
[0017] Step 6-1: Perform noise suppression and intensity normalization on the MRI images during the operation phase.
[0018] Step 6-2: Calculate spatial confidence information w(x) based on the signal-to-noise ratio and image gradient intensity in the voxel x-neighborhood of the MRI image during the operation phase. Assign higher confidence to regions with high signal-to-noise ratio and clear structural boundaries, and assign lower confidence to regions with large noise or artifacts.
[0019] Step 6-3: Extract the anatomical features that can be stably matched in the 3D navigation model and the MRI images during the operation phase;
[0020] Step 6-4: Calculate the deformation relationship f(x) between the voxel x corresponding to the anatomical features extracted in Step 6-3 and the MRI image during the operation phase, using it as the objective function:
[0021]
[0022] In the formula, w(x) represents the spatial confidence information generated in step 6-2, intra(x) represents the image intensity value at voxel x in the MRI image, pre(.) represents the image intensity value in the preoperative navigation model, and u(x) is a three-dimensional vector. Let represent the distance the voxel x moves in three directions. The optimal three-dimensional vector u*(x) is found through an iterative search method to minimize the objective function f(x).
[0023] Step 6-5: Merge the u*(x) values of all voxels obtained in Step 6-4 into a three-dimensional deformation field, and update the position of each voxel point in the three-dimensional navigation model according to x′=x+u*(x) to keep the local structural position of each voxel consistent with the tissue state during the operation.
[0024] Step 6-6: Compare the intensity values of the MRI images during the operation process with the updated images in Step 6-5 at the same voxel position x, calculate the difference as the spatial residual, and map the spatial residual values to a color distribution map corresponding to the spatial position of the 3D navigation model to intuitively reflect the registration reliability of different spatial regions. Regions with smaller residuals indicate higher registration consistency, while regions with larger residuals indicate lower registration reliability.
[0025] Step 7: The system dynamically corrects the 3D navigation model based on the rigid registration results in Step 5 and the 3D deformation field generated by the non-rigid registration in Step 6.
[0026] Step 8: Calculate the spatial residual between the updated 3D navigation model and the MRI images of the operation process, generate the corresponding residual distribution, and when the residual exceeds the preset threshold, indicate that the reliability of the current registration result has decreased, and suggest re-acquiring the MRI images of the operation process or re-performing the registration.
[0027] Furthermore, the iterative search process for the optimal three-dimensional vector u*(x) in step 6-4 is as follows:
[0028] Step 6-4-1, initialize u0(x) = 0.
[0029] Step 6-4-2: Calculate the difference between the MRI image of the operation process corresponding to the voxel x to be updated and the MRI image of the deformed 3D navigation model. The smaller the difference, the closer the two images are.
[0030] Step 6-4-3: Randomly generate the deformation update amount Δu(x), and update the deformation field u(x) by combining it with the voxel confidence w(x) obtained in step 6-4-2. The method is: u′(x)=u(x)+w(x)*Δu(x).
[0031] Step 6-4-4: Repeat steps 6-4-2 to 6-4-3 until the update magnitude of u′(x) is less than the preset threshold, the image difference no longer decreases, or the maximum number of iterations is reached, at which point the search process ends. At this point, u′(x) is the optimal three-dimensional vector u*(x).
[0032] The system for implementing the above-described method for dynamically updating the surgical navigation model based on a portable MRI device comprises a portable MRI device, a data acquisition module, an image processing module, an optical tracking and positioning module, a registration module, a dynamic update module, a residual calculation module, and a model export module, wherein:
[0033] Portable MRI devices acquire MRI images of the target region during the operational phase;
[0034] The data acquisition module is used to acquire MRI images of the 3D navigation model, MRI images during the operation phase, and optical positioning data.
[0035] The image processing module processes the acquired 3D navigation model and MRI images of the operation process.
[0036] The optical tracking and positioning module is used to acquire the spatial position of optical markers on the rigid reference frame in real time to achieve target area positioning.
[0037] The registration module is used to implement the rigid registration and non-rigid registration steps in the method.
[0038] The dynamic update module updates the 3D navigation model dynamically based on the registration results.
[0039] The residual calculation module is used to calculate the spatial residual between the 3D navigation model and the MRI images of the operation process and to provide a prompt when the residual exceeds a threshold.
[0040] The model export module exports the updated navigation model through a standardized export interface.
[0041] Beneficial effects
[0042] The navigation dynamic update method of this invention combines rigid registration and non-rigid registration methods to dynamically update the navigation model, which can effectively avoid navigation drift and improve the safety and accuracy of actual operation.
[0043] The navigation model dynamic update system of this invention can be effectively used in conjunction with portable MRI devices, operates stably under low-field MRI conditions, is compatible with existing navigation systems, and has practical application potential. Attached Figure Description
[0044] Figure 1 This is a flowchart illustrating the steps of the navigation dynamic update method of the present invention;
[0045] Figure 2 This is a diagram showing the composition of the navigation model dynamic update system of the present invention. Detailed Implementation
[0046] To further understand the content of this invention, a detailed description of the invention will be provided in conjunction with the accompanying drawings and specific embodiments.
[0047] like Figure 1 As shown, the method for dynamically updating the navigation model based on a portable MRI device, and an embodiment of its application in neurosurgical procedures, includes the following specific steps:
[0048] Step 1: Before the operation, a high-field magnetic resonance imaging (MRI) scan is performed on the target area to acquire high-resolution MRI image data covering the target surgical area. The images are then processed to establish a three-dimensional navigation model of the target area. In this embodiment, image processing includes target tissue region extraction, key anatomical structure identification, and segmentation of related tissue structures.
[0049] Step 2: Before operation, fix the rigid reference frame with multiple optical markers to the target area. During operation, the rigid reference frame maintains a fixed spatial relationship with the target area.
[0050] Step 3: During the operation, the spatial position of the optical markers on the rigid reference frame is acquired in real time. The real-time pose information of the target area is obtained based on the rigid geometric relationship between the markers. A three-dimensional coordinate system that moves synchronously with the target area during the operation is established based on the real-time pose information.
[0051] Step 4: During the operation, a portable magnetic resonance imaging device is used to scan the target area to obtain MRI images reflecting the current tissue morphology, and then denoising and enhancement processing is performed.
[0052] Step 5: Based on the real-time pose information of the target area obtained in Step 3, perform rigid registration on the 3D navigation model established in Step 1.
[0053] Rigid registration process: Based on the real-time head pose information acquired by the optical tracking system, a rotation matrix and translation vector are constructed, and a rigid transformation is applied to the 3D navigation model to achieve overall alignment between the 3D navigation model and the current target area. The method is as follows:
[0054] ,
[0055] in Represents a point in a 3D navigation model. Let R represent the point after the transformation, where R is a 3x3 rotation matrix and T is a 3x1 translation vector.
[0056] If there is a point with coordinates in the 3D navigation model before operation... The optical tracking module measures the rotation matrix R= Translation vector T = Substituting into the above formula, we get That is, the point that was originally at coordinates [100, 40, 60] in the 3D navigation model becomes [87.2, 63.5, 63] in the current target area.
[0057] Step 6: Perform confidence-weighted non-rigid registration based on the MRI images of the operation process. Generate spatial confidence information according to the quality of the MRI images of the operation process. Assign higher deformation weights to anatomical regions with higher confidence and limit the deformation amplitude to regions with more noise or obvious artifacts.
[0058] After the rigid registration process, the target area is aligned as a whole, but there may still be problems such as tissue displacement, tissue surface collapse, and local tissue deformation. Using the current brain images obtained by portable MRI, the three-dimensional navigation model before operation is spatially registered to allow local deformation, and the deformation calculation is weighted according to the reliability of images in different regions, so as to correct brain tissue displacement errors while ensuring stability.
[0059] The process of non-rigid registration is as follows:
[0060] 1) Preprocessing of the MRI images during the operation: In this embodiment, the preprocessing includes noise suppression and intensity normalization. Noise suppression reduces random noise in the image through median filtering and Gaussian filtering while preserving tissue boundary information. Intensity normalization refers to standardizing the image grayscale values. Through linear normalization or histogram matching methods, the grayscale distribution of the MRI images during the operation is made consistent with that of the MRI images before the operation, thereby reducing intensity differences caused by different scanning conditions.
[0061] 2) Based on the signal-to-noise ratio (SNR) and image gradient intensity within the x-neighborhood of voxels in the MRI image during the operation process, spatial confidence information w(x) is calculated. Regions with high SNR and clear structural boundaries are assigned higher confidence, while regions with high noise or artifacts are assigned lower confidence. Confidence reflects the reliability of image data in different regions of the MRI image.
[0062] 3) Extract stable anatomical features from the pre-operation model and MRI images during the operation, such as ventricular boundaries, brain surface contours, and midline structures.
[0063] 4) Calculate the deformation relationship f(x) between the voxel x corresponding to the anatomical features extracted in step 3) and the MRI images during the operation, using it as the objective function:
[0064]
[0065] In the formula, w(x) represents the spatial confidence information generated in step 2), intra(x) represents the image intensity value at voxel x in the MRI image, pre(.) represents the image intensity value in the three-dimensional navigation model before the operation, and u(x) is a three-dimensional vector. , represents the distance voxel x moves in three directions, i.e., how far voxel x in the pre-operation 3D navigation model should move in three directions in the MRI image during the operation to align with the current tissue state. f(x) represents the degree of mismatch between the pre-operation model and the MRI image during the operation at position x. The goal is to find the optimal 3D vector u*(x) that minimizes the objective function f(x). The optimal 3D vector u*(x) is found through an iterative search method.
[0066] 5) Combine the u*(x) values of all voxels obtained in step 4) into a three-dimensional deformation field, and update the position of each voxel point in the three-dimensional navigation model according to x′=x+ u*(x) so that the local structural position of each voxel is consistent with the tissue state during the operation.
[0067] 6) Compare the intensity values of the MRI image during the operation with the updated image from step 5) at the same voxel position x, and calculate the difference as the spatial residual. The spatial residual represents the degree of matching between the updated 3D navigation model and the MRI image during the operation at the corresponding time point in local space. The spatial residual is mapped to a color distribution map corresponding to the spatial position of the 3D navigation model according to its value, so as to intuitively reflect the registration reliability of different spatial regions. The region with smaller residual indicates higher registration consistency, and the region with larger residual indicates lower registration reliability.
[0068] Step 7: The system dynamically corrects the three-dimensional navigation model before operation based on the rigid registration results in Step 5 and the three-dimensional deformation field generated by the non-rigid registration in Step 6.
[0069] Step 8: Calculate the spatial residual between the updated 3D navigation model and the MRI images during the operation and generate the corresponding residual distribution. When the residual exceeds the preset threshold, it indicates that the reliability of the current registration result has decreased and suggests re-acquiring intraoperative MRI images or re-performing registration.
[0070] The iterative search process for the optimal three-dimensional vector u*(x) in step 4) is as follows:
[0071] (1) Initialize u0(x) = 0;
[0072] (2) Calculate the difference between the MRI image during the operation and the preoperative MRI image after deformation at the voxel x to be updated. The smaller the difference, the closer the two images are.
[0073] (3) Randomly generate the deformation update amount Δu(x), and update the deformation field u(x) with the voxel confidence w(x) obtained in step 2). The method is: u′(x)=u(x)+w(x)*Δu(x). The effect of this update method is to increase the deformation update amplitude in the high confidence region and limit or freeze the deformation update amplitude in the low confidence region.
[0074] (4) Repeat steps (2) to (3) until the update magnitude of u′(x) is less than the preset threshold, the difference no longer decreases, or the maximum number of iterations is reached, and then the search process ends. At this time, the corresponding u′(x) is the optimal three-dimensional vector u*(x).
[0075] like Figure 2 As shown, the system for implementing the above-described method for dynamically updating the surgical navigation model based on a portable MRI device comprises a portable MRI device, a data acquisition module, an image processing module, an optical tracking and positioning module, a registration module, a dynamic update module, a residual calculation module, and a model export module, wherein:
[0076] Portable MRI devices acquire MRI images of the target area during the procedure.
[0077] The data acquisition module is used to acquire MRI images before the operation, MRI images during the operation, and optical positioning data.
[0078] The image processing module processes the acquired MRI images before and during the operation.
[0079] The optical tracking and positioning module is used to acquire the spatial position of optical markers on the rigid reference frame in real time to achieve target area positioning.
[0080] The registration module is used to implement the rigid registration and non-rigid registration steps in the method.
[0081] The dynamic update module dynamically updates the 3D navigation model based on the registration results.
[0082] The residual calculation module is used to calculate the spatial residual between the 3D navigation model and the MRI images during the operation, and to provide a prompt when the residual exceeds a threshold.
[0083] The model export module exports the updated navigation model. The system provides a standardized export interface that is compatible with existing navigation systems, such as JSON / XML and DICOM interfaces.
[0084] The present invention and its embodiments have been described above illustratively. This description is not restrictive, and the figures shown are only one embodiment of the present invention; the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.
Claims
1. A method for dynamically updating a navigation model based on a portable MRI device, characterized in that, Includes the following steps: Step S1: Before the operation, a high-field magnetic resonance imaging scan of the target area is performed using a portable MRI device to obtain high-resolution MRI image data covering the target area and process the images to establish a three-dimensional navigation model of the target area. Step S2: Fix the rigid reference frame with multiple optical markers to the target area so that the rigid reference frame maintains a fixed spatial relationship with the target area during operation; Step S3: During the operation, the spatial position of the optical markers on the rigid reference frame is collected in real time. The real-time pose information of the target area is obtained according to the rigid geometric relationship between the markers. A three-dimensional coordinate system that moves synchronously with the operation process is established based on the real-time pose information of the target area. Step S4: During the operation phase, a portable MRI device is used to scan the target area to obtain MRI images that reflect the current tissue morphology, and then denoising and enhancement processing is performed. Step S5: Based on the real-time pose information of the target area obtained in step S3, perform rigid registration on the 3D navigation model established in step S1. The method is as follows: construct a rotation matrix and a translation vector and apply them to the 3D navigation model established in step S1 to achieve overall alignment between the 3D navigation model and the current surgical position. ， in Represents a point in a 3D navigation model. Let R represent the point after the transformation, where R is a 3*3 rotation matrix and T is a 3*1 translation vector; Step S6, performing confidence-weighted non-rigid registration based on the MRI images acquired during the operation phase, includes the following steps: Step S6-1: Perform noise suppression and intensity normalization processing on the MRI images during the operation phase; Step S6-2: Calculate spatial confidence information w(x) based on the signal-to-noise ratio and image gradient intensity in the voxel x-neighborhood of the MRI image during the operation phase. Assign higher confidence to regions with high signal-to-noise ratio and clear structural boundaries, and assign lower confidence to regions with large noise or artifacts. Step S6-3: Extract the anatomical features that can be stably matched in the 3D navigation model and the MRI images during the operation phase; Step S6-4: Calculate the deformation relationship f(x) between the voxel x corresponding to the anatomical features extracted in step S6-3 and the MRI image during the operation phase, using it as the objective function: ， In the formula, w(x) represents the spatial confidence information generated in step S6-2, intra(x) represents the image intensity value at voxel x in the MRI image, pre(.) represents the image intensity value in the preoperative navigation model, and u(x) is a three-dimensional vector. Let represent the distance the voxel x moves in three directions. The optimal three-dimensional vector u*(x) is found by iterative search to minimize the objective function f(x). Step S6-5: Merge the u*(x) values of all voxels obtained in step S6-4 into a three-dimensional deformation field, and update the position of each voxel point in the three-dimensional navigation model according to x′=x+u*(x) so that the local structural position of each voxel is consistent with the tissue state during the operation. Step S6-6: Compare the intensity values of the MRI images during the operation process with the updated image values in step S6-5 at the same voxel position x, calculate the difference as the spatial residual, and map the spatial residual values to a color distribution map corresponding to the spatial position of the three-dimensional navigation model to intuitively reflect the registration reliability of different spatial regions. The region with smaller residuals indicates higher registration consistency, and the region with larger residuals indicates lower registration reliability. Step S7: Dynamically correct the three-dimensional navigation model based on the rigid registration result of step S5 and the three-dimensional deformation field generated by the non-rigid registration in step S6. Step S8: Calculate the spatial residual between the updated 3D navigation model and the MRI images of the operation process, generate the corresponding residual distribution, and when the residual exceeds the preset threshold, indicate that the reliability of the current registration result has decreased, and suggest re-acquiring the MRI images of the operation process or re-performing the registration.
2. The method for dynamically updating a navigation model based on a portable MRI device as described in claim 1, characterized in that, The iterative search process for the optimal three-dimensional vector u*(x) in step S6-4 is as follows: Step S6-4-1, initialize u0(x) = 0; Step S6-4-2: Calculate the difference between the MRI image of the operation process corresponding to the voxel x to be updated and the MRI image of the deformed 3D navigation model. The smaller the difference, the closer the two images are. Step S6-4-3: Randomly generate the deformation update amount Δu(x), and update the deformation field u(x) in combination with the voxel confidence w(x) obtained in step S6-4-2. The method is: u′(x)=u(x)+w(x)*Δu(x); Step S6-4-4: Repeat steps S6-4-2 to S6-4-3 until the update magnitude of u′(x) is less than the preset threshold, the image difference no longer decreases, or the maximum number of iterations is reached, and then the search process ends. At this time, the corresponding u′(x) is the optimal three-dimensional vector u*(x).
3. A system for implementing the navigation model dynamic update method based on a portable MRI device as described in any one of claims 1-2, characterized in that, It consists of a portable MRI device, a data acquisition module, an image processing module, an optical tracking and positioning module, a registration module, a dynamic update module, a residual calculation module, and a model export module. The portable MRI device acquires MRI images of the target area during the operation phase; the data acquisition module acquires MRI images of the 3D navigation model, MRI images during the operation phase, and optical positioning data; the image processing module processes the acquired 3D navigation model and MRI images during the operation phase; the optical tracking and positioning module acquires the spatial positions of optical markers on a rigid reference frame in real time to achieve target area positioning; the registration module implements the rigid and non-rigid registration steps in the method; the dynamic update module dynamically updates the 3D navigation model based on the registration results; and the residual calculation module calculates the spatial residual between the 3D navigation model and the MRI images during the operation phase and provides a prompt when the residual exceeds a threshold. The model export module exports the updated navigation model through a standardized export interface.