A Multimodal Image Localization and Calibration Analysis Method and System Based on Magnetic Induction Array

Abstract

This invention discloses a multimodal image localization and calibration analysis method and system based on a magnetic induction array, belonging to the field of data analysis technology. It is equipped with a magnetic induction array composed of magnetic sensors; acquires preoperative images, constructs a three-dimensional model of the surgical area based on the images, plans the stimulation target area and navigation path, and generates a preoperative planning point set; selects a reference point, completes rigid registration with the actual intraoperative reference point, and establishes an initial coordinate mapping relationship; based on contact with the surgical target area boundary, the probe's magnetic excitation coil excites a magnetic field, and the array sensor acquires triaxial magnetic field data according to a preset sampling period, generating an intraoperative measured point set; inputs the measured point set acquired by the sensor into an LM solver, and outputs the probe coordinates in real time; performs coordinate correction based on a deformation compensation mechanism, and the system performs real-time calibration based on the corrected coordinates. This application combines an affine transformation matrix with a least-squares optimized deformation compensation mechanism, which can effectively correct intraoperative tissue non-rigid deformation deviations.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, specifically a method and system for multimodal image localization and calibration analysis based on a magnetic induction array. Background Technology

[0002] In the field of digital surgery, neurosurgery, orthopedics, and minimally invasive laparoscopic surgery have increasingly stringent requirements for precise intraoperative anatomical positioning. This necessitates avoiding dangerous structures and adapting to dynamic tissue changes with millimeter-level precision. However, existing positioning and calibration technologies face multiple bottlenecks: traditional optical navigation systems rely on rigid markers, which cannot compensate for positioning errors caused by non-rigid deformation during surgery, and are susceptible to line-of-sight obstruction and sensitive to preoperative and intraoperative positioning deviations; traditional electromagnetic positioning systems use single sensors, resulting in poor resistance to environmental magnetic interference, low dynamic refresh rates, and a lack of multimodal image fusion capabilities, leading to lagging positioning data and insufficient operational intuitiveness; existing systems primarily use fixed-shape core components, making planar arrays unsuitable for the irregular curved surfaces of open surgery, and large arrays difficult to adapt to the closed spaces of laparoscopic surgery, limiting intraoperative adaptability; even when some systems attempt deformation compensation, they can only handle rigid deformations, and complex nonlinear models suffer from poor real-time performance and a lack of multi-physical quantity fusion verification. In summary, there is an urgent clinical need for a technical solution that has low positioning error, strong anti-interference, no line-of-sight limitations, flexible array morphology adjustment, deep fusion of multimodal images, and efficient automated calibration. Against this backdrop, developing a multimodal image positioning and calibration analysis method based on magnetic induction arrays has become a key direction to overcome existing deficiencies and meet the needs of fine positioning and calibration. Summary of the Invention

[0003] The purpose of this invention is to provide a multimodal image localization and calibration analysis method and system based on magnetic induction array to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a multimodal image localization and calibration analysis method based on a magnetic induction array, specifically including the following steps:

[0005] Equipped with a magnetic induction array composed of magnetic sensors, the magnetic induction array is preprocessed before startup;

[0006] Preoperative images of the surgical site are collected, a three-dimensional model of the surgical area is constructed based on the images, the stimulation target area and navigation path are planned, and a preoperative planning point set {P} is generated.

[0007] Select several reference points from the preoperative planning point set {P} within the surgical area, and use a magnetically excited handheld probe to contact the reference points in sequence to trigger the array sensor response, thereby completing the rigid registration between the preoperative planning point set {P} and the actual intraoperative reference points and establishing the initial coordinate mapping relationship.

[0008] Based on the boundary of the surgical target area, the probe's magnetic excitation coil generates a magnetic field, and the array sensor collects triaxial magnetic field data according to a preset sampling period to generate an intraoperative measured point set {P'}. The measured point set {P'} collected by the sensor is input into the LM solver, and the 5-DOF pose is iteratively solved based on the magnetic dipole model, and the probe coordinates are output in real time.

[0009] Coordinate correction is performed based on a deformation compensation mechanism, and the system performs real-time calibration based on the corrected coordinates.

[0010] Furthermore, the preprocessing includes:

[0011] The magnetic sensitivity of the sensor units of the magnetic induction array is calibrated to ensure that the triaxial measurement accuracy meets the error convergence threshold requirements;

[0012] Inspect the handheld probe with the magnetic excitation coil and calibrate the magnetic signal using a magnetic calibration head to ensure the accuracy of the input parameters of the magnetic dipole model.

[0013] Furthermore, preoperative images of the patient's surgical site are acquired, a three-dimensional model of the surgical area is constructed based on the images, the stimulation target area and navigation path are planned, and a preoperative planning point set {P} is generated.

[0014] Step S3-1: Select the appropriate imaging modality to ensure that the scanning range covers the surgical operation area and the magnetic induction array bonding area. When acquiring images, fix the body position with a non-magnetic positioning patch on the body surface to avoid the initial offset between the preoperative model and the actual anatomical structure during the operation due to differences in body position.

[0015] Step S3-2: Noise processing is performed on the acquired images using a Kalman filter to eliminate intraoperative magnetic interference and reduce the impact of environmental noise on sensor measurements; multimodal influences are fused into a unified DICOM standard coordinate system through registration using preset marker points, and the regions of interest (ROIs) of relevant structures are extracted simultaneously.

[0016] Step S3-3: Using geometric modeling software, construct a three-dimensional model based on the image after noise processing, and mark the target area and the fitting area matching the magnetic induction array type in the three-dimensional model; wherein, the fitting area matching the magnetic induction array type includes the planar array size or the radius of curvature of the arc array;

[0017] Step S3-4: Delineate the stimulation target area based on the three-dimensional model and record the target area boundary;

[0018] Step S3-5: Plan the navigation path, ensuring that the path is within the sensing range of the magnetic induction array, and obtain the starting point, intermediate nodes, and ending point of the navigation path; wherein, the intermediate nodes are several nodes in the planned navigation path;

[0019] Steps S3-6: The preset marker points, the outline points of the target area boundary, and the navigation path points are used as the preoperative planning point set {P}. All points use the DICOM standard coordinate system, with the fixed point of the array contact area as the origin, the X-axis along the horizontal direction of the array, the Y-axis along the vertical direction of the array, and the Z-axis perpendicular to the array plane, consistent with the intraoperative magnetic induction array positioning coordinate system.

[0020] Furthermore, several reference points are selected from the preoperative planned point set {P} within the surgical area. A magnetically excited handheld probe is used to sequentially contact these reference points, triggering the array sensor response. This completes the rigid registration between the preoperative planned point set {P} and the actual intraoperative reference points, establishing an initial coordinate mapping relationship. Specifically:

[0021] Step S4-1: Select n planning points from the preoperative planning point set {P} within the surgical area as reference points, denoted as p. pre，i =[x pre，i ,y pre，i ,z pre，i ] T ;where x pre，i ,y pre，i ,z pre，i Let B represent the x, y, and y coordinates of the i-th reference point, respectively; i = 1, 2, ..., n, where n is the number of reference points; a handheld probe with a magnetic excitation coil is placed against the reference point, and the probe's magnetic excitation coil excites a magnetic field of a preset intensity. The triaxial magnetic field data B at that reference point is collected by an array sensor. int (X, Y, Z), where X, Y, and Z represent the magnetic field data of the horizontal, vertical, and longitudinal axes, respectively; the collected magnetic field data are input into the LM solver to iteratively solve for the three-dimensional coordinates of each actual reference point, denoted as p. int，i =[x int，i ,y int，i ,z int，i ] T ;where x int，i ,y int，i ,z int，i These represent the horizontal, vertical, and y-axis coordinates of the i-th actual reference point, respectively.

[0022] Step S4-2: Solve the 3×3 rotation matrix R and the 3×1 translation vector t to make the reference point p pre，i After rotation and translation transformations, it is compared with the actual reference point p. int，i The spatial position error is minimized;

[0023] Step S4-3: Calculate the registration error e of all reference points. i e i =||p int，i -(R*p pre，i+t)||2;When the average registration error of all reference points is less than the error convergence threshold, the initial coordinate mapping relationship is valid; otherwise, the process is repeated from step S4-1.

[0024] Preferably, the solution method for the 3×3 rotation matrix R and the 3×1 translation vector t is as follows:

[0025] Step s1: Calculate the centroids of the n reference points and the actual reference point respectively, using the formula: μ pre =(Σ n p pre，i ) / n; where μ pre This represents the centroid of n reference points; similarly, the centroid μ of the actual reference point is obtained. int ;

[0026] Step s2: Decentrifuge the reference point to eliminate the interference of translation on the rotation solution. The centrifuged reference point q is then obtained. pre，i =p pre，i -μ pre The actual reference point q after centroid removal int，i =p int，i -μ int ;

[0027] Step s3: Construct the covariance matrix H and solve for the rotation matrix R;

[0028] Wherein, the covariance matrix H=Σ n q int，i *q T pre，i Singular value decomposition of the covariance matrix H yields H = U·Λ·V T Where Λ represents a 3rd order diagonal matrix; U and V represent 3rd order orthogonal matrices; when det(U·V) T If ) = 1, then the rotation matrix R = U·V T When det(U·V) T If ) = -1, then adjust the sign of the last column of U before calculating R; ensure the orthogonality of the rotation matrix;

[0029] Step s4: Substitute the rotation matrix R into the centroid relation to obtain the translation vector t = μ int -R·μ pre ;

[0030] Furthermore, based on the boundary of the surgical target area, the probe's magnetic excitation coil generates a magnetic field, and the array sensor collects triaxial magnetic field data according to a preset sampling period, generating an intraoperative measured point set {P'}. The measured point set {P'} collected by the sensor is input into the LM solver, and the 5-DOF pose is iteratively solved based on the magnetic dipole model, outputting the probe coordinates in real time. Specifically:

[0031] Based on the magnetic field generated by the probe's magnetic excitation coil when contacting the surgical target area boundary, the array sensor collects triaxial magnetic field data from N sensors according to a preset sampling period, generating an intraoperative measured point set {P'}, where each element in {P'} corresponds to a measured magnetic field vector B from one sensor. meas,k k = 1, 2, ..., N; k represents the number of sensors. The specific steps include:

[0032] The pose vector of the probe's built-in spherical permanent magnet is F=[x,y,z,θ,Φ]. T Where (x,y,z) represents the three-dimensional coordinates of the permanent magnet in the DICOM standard coordinate system; θ represents the pitch angle of the permanent magnet's magnetic moment; Φ represents the yaw angle of the permanent magnet's magnetic moment; where the pitch angle is the angle between the magnetic moment and the vertical axis of the coordinate system, ranging from [0, Π]; the yaw angle is the angle between the projection of the magnetic moment onto the XY plane and the X-axis, ranging from [0, 2Π].

[0033] Wherein, the three-dimensional coordinates of the fixed position of the k-th sensor in the sensor array are r. s,k =[x s,k ,y s,k ,z s,k ] T ;

[0034] The triaxial measured magnetic field of the k-th sensor is B. meas,k =[B x,meas,k B y,meas,k B z,meas,k ] T ;

[0035] The predicted magnetic field at the k-th sensor, calculated based on the magnetic dipole model, is denoted as:

[0036] B model,k (F)=[B x,model,k (F), B y,model,k (F), B z,model,k (F)] T ;

[0037] The optimal pose F is solved iteratively using the LM solver. * The mathematical expression is: F * =argmin F S(F); where S(F) is the cost function; the optimal pose F is obtained by solving... * As probe coordinate output;

[0038] S(F)=Σ k=1 N Σ j=x,y,z (B j,meas,k -B j,model,k (F)) 2; where j represents the identifier of the three-dimensional coordinate.

[0039] Furthermore, the steps for calculating the predicted magnetic field at the k-th sensor based on the magnetic dipole model include:

[0040] Based on the unified DICOM standard coordinate system, and according to the pose vector of the probe's built-in spherical permanent magnet as F=[x,y,z,θ,Φ] T The three-dimensional coordinate vector Fpos=[x,y,z] of the permanent magnet in the DICOM standard coordinate system is obtained. T ;

[0041] The magnetic moment vector of the magnet is obtained as g(θ, Φ) = g[cosθcosΦ,cosθsinΦ,sinΦ]. T Where g represents the magnitude of the magnetic moment of the magnet;

[0042] Calculate the position vector r from the permanent magnet to the sensor k =r s,k -Fpos;

[0043] The predicted magnetic field at the k-th sensor in space is:

[0044] B model,k (F)=μ0g*[3(g(θ,Φ)·r k ^)r k ^-g(θ,Φ)] / 4Πr k 3 ;

[0045] Where μ0 represents the free permeability; r k Represents the distance from the permanent magnet to the k-th sensor; r is the position vector r of the permanent magnet to the sensor. k The modulus length; r k ^ represents the position vector r from the permanent magnet to the k-th sensor. k The unit vector.

[0046] Since the preoperative planning point set {P} (including target area and path nodes) is generated based on preoperative static images (CT / MRI), the coordinates are established on the assumption that the tissue is without deformation and are bound to the magnetic induction array positioning coordinate system (with the array contact area as the origin); during the operation, the tissue will undergo non-rigid deformation (different from the rigid change of overall displacement), which manifests as local stretching, shearing, and slight torsion.

[0047] If preoperative coordinate navigation is used directly, the navigation path will be misaligned with the actual tissue position. Traditional rigid registration (rotation and translation only) cannot cover non-rigid deformations and requires targeted modeling and compensation. Therefore, this application proposes a deformation compensation mechanism for coordinate correction.

[0048] Optionally, coordinate correction is performed based on a deformation compensation mechanism, and the system performs real-time calibration based on the corrected coordinates, specifically as follows:

[0049] When the deviation between the navigation path points in the preoperative planning point set {P} and the corresponding probe coordinates exceeds a preset deviation threshold, coordinate correction is performed through a deformation compensation mechanism. Initial matching of the navigation path points in the preoperative planning point set {P} with the probe coordinates is then performed to obtain data pairs, denoted as (p...). feat,l ,p' feat,l ); where l represents the identifier of the data pair, l is a positive integer, l∈[1,m], and m represents the total number of navigation path points; p feat,l p' represents the l-th navigation path point; feat,l Indicate the coordinates of the l-th probe; assign navigation path point p feat,l =[x feat,l ,y feat,l ,z feat,l ] T Mapped to predict intraoperative coordinates p'' feat,l Specifically:

[0050] ;where x feat,l ,y feat,l ,z feat,l x'' represents the three-dimensional coordinates of the l-th navigation path point; feat,l ,y'' feat,l ,z'' feat,l These represent the three-dimensional coordinates of the l-th predicted intraoperative coordinate; a 11 ,…,a 33 Used to describe the linear deformation of an organization; t x , t y , and t z Used for overall translation of the organization; the last row is an auxiliary row for homogeneous coordinates;

[0051] The affine transformation matrix [T] is obtained by minimizing the sum of squares of the predicted intraoperative coordinates and the probe coordinates using least squares optimization.

[0052] The points in the preoperative planning point set {P} are corrected using the affine transformation matrix [T].

[0053] It should be noted that the preoperative target area feature point coordinates are generated based on static images of tissue without deformation. During the operation, the target area may shift or change shape due to traction or resection of surrounding tissues. By mapping these preoperative feature points to the corrected intraoperative coordinates through [T], it can be ensured that the probe can accurately locate the actual boundary of the target area during the operation, thus avoiding misalignment and echoing the technical goal of improving the accuracy of target area positioning in the document.

[0054] Since dangerous structures will shift synchronously with tissue deformation, if the preoperative coordinates are still used to avoid risks, accidental injury may occur due to coordinate misalignment. By correcting the preoperative dangerous structure coordinates to the actual intraoperative coordinates through [T], it can be ensured that the corrected navigation path maintains a safe distance from the dangerous structure.

[0055] A multimodal image localization and calibration analysis system based on a magnetic induction array includes an initial registration module, a 3D modeling module, a point set acquisition and mapping module, an LM solver module, a deformation compensation module, and a visualization module.

[0056] The initial registration module is used to equip a magnetic induction array composed of magnetic sensors and to preprocess the magnetic induction array before startup.

[0057] The 3D modeling module is used to acquire preoperative images of the patient's surgical site and construct a 3D model of the surgical area based on the images.

[0058] The point set acquisition and mapping module is used to plan the stimulation target area and navigation path based on the three-dimensional model, generate a preoperative planning point set {P}, select several reference points in the preoperative planning point set {P} within the surgical area, use a magnetically excited handheld probe to contact the reference points in sequence, trigger the array sensor response, complete the rigid registration between the preoperative planning point set {P} and the actual intraoperative reference points, and establish an initial coordinate mapping relationship.

[0059] The LM solver module is used to input the measured point set {P'} collected by the sensor into the LM solver, iteratively solve the 5-DOF pose based on the magnetic dipole model, and output the probe coordinates in real time.

[0060] The deformation compensation module is used to perform coordinate correction based on the deformation compensation mechanism;

[0061] The visualization module is used to display the generated 3D model and the current probe coordinates in real time.

[0062] Compared with existing technologies, the beneficial effects of this invention are as follows: This application's multimodal image localization and calibration analysis method based on a magnetic induction array addresses the core defects of existing technologies, such as insufficient positioning accuracy, weak anti-interference, poor adaptability, and low real-time performance. Through key technological innovations, it achieves a comprehensive breakthrough: Utilizing a magnetic induction array composed of multiple sensors and an LM solver iteratively solving the 5-DOF pose based on a magnetic dipole model, combined with an affine transformation matrix and a least-squares optimized deformation compensation mechanism, it can effectively correct intraoperative tissue non-rigid deformation deviations, stabilizing the positioning error within a convergence threshold of ≤0.3mm, far superior to traditional optical navigation and electromagnetic positioning, meeting accuracy requirements; simultaneously, the preprocessing stage calibrates the sensor magnetic sensitivity and probe magnetic signals, and, combined with Kalman filter noise reduction and multi-sensor redundancy design, can... It resists interference from metal instruments and environmental electromagnetic fields, ensuring stable and reliable data. The array sensor collects data at a preset sampling period, and the LM solver dynamically balances convergence speed and stability, adapting to rapid intraoperative probe movement without data lag. Preoperative multimodal images are fused into a unified DICOM standard coordinate system through preset markers, solving the adaptation problem of traditional arrays with fixed shapes. After deep fusion of multimodal images and intraoperative positioning data, the 3D model can clearly mark the target area and dangerous structures. The deformation compensation mechanism can also correct the intraoperative coordinates of dangerous structures in real time, ensuring that the navigation path maintains a safe distance from dangerous structures, significantly reducing the risk of accidental injury. Furthermore, rigid registration and error verification are fully automated, greatly shortening the average registration time, reducing manual intervention and time consumption, and ultimately achieving high-precision, strong anti-interference, high adaptability, and high safety intraoperative positioning and calibration. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating a multimodal image localization and calibration analysis method based on a magnetic induction array according to the present invention.

[0064] Figure 2 This is a structural diagram of the magnetic excitation handle in an embodiment of the present invention;

[0065] Figure 3 This is a schematic diagram of the magnetic induction array layout in an embodiment of the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Example: Figures 1-3As shown, the present invention provides a technical solution, a multimodal image localization and calibration analysis method based on a magnetic induction array, which specifically includes the following steps:

[0068] Equipped with a magnetic induction array composed of magnetic sensors, the magnetic induction array is preprocessed before startup;

[0069] Furthermore, the preprocessing includes:

[0070] The magnetic sensitivity of the sensor units of the magnetic induction array is calibrated to ensure that the triaxial measurement accuracy meets the error convergence threshold requirement; in this embodiment, the error convergence threshold is 0.3 mm.

[0071] like Figure 3 The diagram shown is a schematic of the magnetic induction array layout.

[0072] Where AF is the row identifier of the sensor, and 1-4 are the column identifiers of the sensor, which are used together to determine the sensor position;

[0073] Inspect the handheld probe with the magnetic excitation coil and calibrate the magnetic signal using a magnetic calibration head to ensure the accuracy of the input parameters of the magnetic dipole model.

[0074] Preoperative images of the surgical site are collected, a three-dimensional model of the surgical area is constructed based on the images, the stimulation target area and navigation path are planned, and a preoperative planning point set {P} is generated.

[0075] Furthermore, preoperative images of the patient's surgical site are acquired, a three-dimensional model of the surgical area is constructed based on the images, the stimulation target area and navigation path are planned, and a preoperative planning point set {P} is generated.

[0076] Step S3-1: Select the appropriate imaging modality to ensure that the scanning range covers the surgical operation area and the magnetic induction array bonding area. When acquiring images, fix the body position with a non-magnetic positioning patch on the body surface to avoid the initial offset between the preoperative model and the actual anatomical structure during the operation due to differences in body position.

[0077] Step S3-2: Noise processing is performed on the acquired images using a Kalman filter to eliminate intraoperative magnetic interference and reduce the impact of environmental noise on sensor measurements; multimodal influences are fused into a unified DICOM standard coordinate system through registration using preset marker points, and the regions of interest (ROIs) of relevant structures are extracted simultaneously.

[0078] Step S3-3: Using geometric modeling software, construct a three-dimensional model based on the image after noise processing, and mark the target area and the fitting area matching the magnetic induction array type in the three-dimensional model; wherein, the fitting area matching the magnetic induction array type includes the planar array size or the radius of curvature of the arc array;

[0079] Step S3-4: Delineate the stimulation target area based on the three-dimensional model and record the target area boundary;

[0080] Step S3-5: Plan the navigation path, ensuring that the path is within the sensing range of the magnetic induction array, and obtain the starting point, intermediate nodes, and ending point of the navigation path; wherein, the intermediate nodes are several nodes in the planned navigation path;

[0081] Steps S3-6: The preset marker points, the outline points of the target area boundary, and the navigation path points are used as the preoperative planning point set {P}. All points use the DICOM standard coordinate system, with the fixed point of the array contact area as the origin, the X-axis along the horizontal direction of the array, the Y-axis along the vertical direction of the array, and the Z-axis perpendicular to the array plane, consistent with the intraoperative magnetic induction array positioning coordinate system.

[0082] Select several reference points from the preoperative planning point set {P} within the surgical area, and use a magnetically excited handheld probe to contact the reference points in sequence to trigger the array sensor response, thereby completing the rigid registration between the preoperative planning point set {P} and the actual intraoperative reference points and establishing the initial coordinate mapping relationship.

[0083] like Figure 2 The diagram shown is a structural diagram of the magnetic excitation handle;

[0084] Furthermore, several reference points are selected from the preoperative planned point set {P} within the surgical area. A magnetically excited handheld probe is used to sequentially contact these reference points, triggering the array sensor response. This completes the rigid registration between the preoperative planned point set {P} and the actual intraoperative reference points, establishing an initial coordinate mapping relationship. Specifically:

[0085] Step S4-1: Select n planning points from the preoperative planning point set {P} within the surgical area as reference points, denoted as p. pre，i =[x pre，i ,y pre，i ,z pre，i ] T ;where x pre，i ,y pre，i ,z pre，i Let B represent the x, y, and y coordinates of the i-th reference point, respectively; i = 1, 2, ..., n, where n is the number of reference points; a handheld probe with a magnetic excitation coil is placed against the reference point, and the probe's magnetic excitation coil excites a magnetic field of a preset intensity. The triaxial magnetic field data B at that reference point is collected by an array sensor. int (X, Y, Z), where X, Y, and Z represent the magnetic field data of the horizontal, vertical, and longitudinal axes, respectively; the collected magnetic field data are input into the LM solver to iteratively solve for the three-dimensional coordinates of each actual reference point, denoted as p. int，i =[x int，i ,y int，i ,zint，i ] T ;where x int，i ,y int，i ,z int，i These represent the horizontal, vertical, and y-axis coordinates of the i-th actual reference point, respectively.

[0086] Step S4-2: Solve the 3×3 rotation matrix R and the 3×1 translation vector t to make the reference point p pre，i After rotation and translation transformations, it is compared with the actual reference point p. int，i The spatial position error is minimized;

[0087] Step S4-3: Calculate the registration error e of all reference points. i e i =||p int，i -(R*p pre，i +t)||2;When the average registration error of all reference points is less than the error convergence threshold, the initial coordinate mapping relationship is valid; otherwise, the process is repeated from step S4-1.

[0088] Preferably, the solution method for the 3×3 rotation matrix R and the 3×1 translation vector t is as follows:

[0089] Step s1: Calculate the centroids of the n reference points and the actual reference point respectively, using the formula: μ pre =(Σ n p pre，i ) / n; where μ pre This represents the centroid of n reference points; similarly, the centroid μ of the actual reference point is obtained. int ;

[0090] Step s2: Decentrifuge the reference point to eliminate the interference of translation on the rotation solution. The centrifuged reference point q is then obtained. pre，i =p pre，i -μ pre The actual reference point q after centroid removal int，i =p int，i -μ int ;

[0091] Step s3: Construct the covariance matrix H and solve for the rotation matrix R;

[0092] Wherein, the covariance matrix H=Σ n q int，i *q T pre，i Singular value decomposition of the covariance matrix H yields H = U·Λ·V T Where Λ represents a 3rd order diagonal matrix; U and V represent 3rd order orthogonal matrices; when det(U·V) T If ) = 1, then the rotation matrix R = U·V TWhen det(U·V) T If ) = -1, then adjust the sign of the last column of U before calculating R; ensure the orthogonality of the rotation matrix;

[0093] Step s4: Substitute the rotation matrix R into the centroid relation to obtain the translation vector t = μ int -R·μ pre ;

[0094] Based on the contact with the surgical target area boundary, the probe's magnetic excitation coil generates a magnetic field, and the array sensor collects triaxial magnetic field data according to a preset sampling period, generating an intraoperative measured point set {P'}; the measured point set {P'} collected by the sensor is input into the LM solver, and the 5-DOF pose is iteratively solved based on the magnetic dipole model, and the probe coordinates are output in real time; when the navigation probe contacts the tumor boundary, it excites the array sensor response. In this embodiment, the sampling period is 2ms;

[0095] Furthermore, based on the boundary of the surgical target area, the probe's magnetic excitation coil generates a magnetic field, and the array sensor collects triaxial magnetic field data according to a preset sampling period, generating an intraoperative measured point set {P'}. The measured point set {P'} collected by the sensor is input into the LM solver, and the 5-DOF pose is iteratively solved based on the magnetic dipole model, outputting the probe coordinates in real time. Specifically:

[0096] Based on the magnetic field generated by the probe's magnetic excitation coil when contacting the surgical target area boundary, the array sensor collects triaxial magnetic field data from N sensors according to a preset sampling period, generating an intraoperative measured point set {P'}, where each element in {P'} corresponds to a measured magnetic field vector B from one sensor. meas,k k = 1, 2, ..., N; k represents the number of sensors. The specific steps include:

[0097] The pose vector of the probe's built-in spherical permanent magnet is F=[x,y,z,θ,Φ]. T Where (x,y,z) represents the three-dimensional coordinates of the permanent magnet in the DICOM standard coordinate system; θ represents the pitch angle of the permanent magnet's magnetic moment; Φ represents the yaw angle of the permanent magnet's magnetic moment; where the pitch angle is the angle between the magnetic moment and the vertical axis of the coordinate system, ranging from [0, Π]; the yaw angle is the angle between the projection of the magnetic moment onto the XY plane and the X-axis, ranging from [0, 2Π].

[0098] Wherein, the three-dimensional coordinates of the fixed position of the k-th sensor in the sensor array are r. s,k =[x s,k ,y s,k ,z s,k ] T ;

[0099] The triaxial measured magnetic field of the k-th sensor is B. meas,k =[B x,meas,k By,meas,k B z,meas,k ] T ;

[0100] The predicted magnetic field at the k-th sensor, calculated based on the magnetic dipole model, is denoted as:

[0101] B model,k (F)=[B x,model,k (F), B y,model,k (F), B z,model,k (F)] T ;

[0102] The optimal pose F is solved iteratively using the LM solver. * The mathematical expression is: F * =argmin F S(F); where S(F) is the cost function; the optimal pose F is obtained by solving... * As probe coordinate output;

[0103] S(F)=Σ k=1 N Σ j=x,y,z (B j,meas,k -B j,model,k (F)) 2 ; where j represents the identifier of the three-dimensional coordinate.

[0104] Furthermore, the steps for calculating the predicted magnetic field at the k-th sensor based on the magnetic dipole model include:

[0105] Based on the unified DICOM standard coordinate system, and according to the pose vector of the probe's built-in spherical permanent magnet as F=[x,y,z,θ,Φ] T The three-dimensional coordinate vector Fpos=[x,y,z] of the permanent magnet in the DICOM standard coordinate system is obtained. T ;

[0106] The magnetic moment vector of the magnet is obtained as g(θ, Φ) = g[cosθcosΦ,cosθsinΦ,sinΦ]. T Where g represents the magnitude of the magnetic moment of the magnet;

[0107] Calculate the position vector r from the permanent magnet to the sensor k =r s,k -Fpos;

[0108] The predicted magnetic field at the k-th sensor in space is:

[0109] B model,k (F)=μ0g*[3(g(θ,Φ)·r k ^)r k ^-g(θ,Φ)] / 4Πrk 3 ;

[0110] Where μ0 represents the free permeability; r k Represents the distance from the permanent magnet to the k-th sensor; r is the position vector r of the permanent magnet to the sensor. k The modulus length; r k ^ represents the position vector r from the permanent magnet to the k-th sensor. k The unit vector.

[0111] Coordinate correction is performed based on a deformation compensation mechanism, and the system performs real-time calibration based on the corrected coordinates.

[0112] Since the preoperative planning point set {P} (including target area and path nodes) is generated based on preoperative static images (CT / MRI), the coordinates are established on the assumption that the tissue is without deformation and are bound to the magnetic induction array positioning coordinate system (with the array contact area as the origin); during the operation, the tissue will undergo non-rigid deformation (different from the rigid change of overall displacement), which manifests as local stretching, shearing, and slight torsion.

[0113] If preoperative coordinate navigation is used directly, the navigation path will be misaligned with the actual tissue position. Traditional rigid registration (rotation and translation only) cannot cover non-rigid deformations and requires targeted modeling and compensation. Therefore, this application proposes a deformation compensation mechanism for coordinate correction.

[0114] Optionally, coordinate correction is performed based on a deformation compensation mechanism, and the system performs real-time calibration based on the corrected coordinates, specifically as follows:

[0115] When the deviation between the navigation path points in the preoperative planning point set {P} and the corresponding probe coordinates exceeds a preset deviation threshold, coordinate correction is performed through a deformation compensation mechanism. Initial matching of the navigation path points in the preoperative planning point set {P} with the probe coordinates is then performed to obtain data pairs, denoted as (p...). feat,l ,p' feat,l ); where l represents the identifier of the data pair, l is a positive integer, l∈[1,m], and m represents the total number of navigation path points; p feat,l p' represents the l-th navigation path point; feat,l Indicate the coordinates of the l-th probe; assign navigation path point p feat,l =[x feat,l ,y feat,l ,z feat,l ] T Mapped to predict intraoperative coordinates p'' feat,l Specifically:

[0116] ;where x feat,l ,y feat,l ,z feat,lx'' represents the three-dimensional coordinates of the l-th navigation path point; feat,l ,y'' feat,l ,z'' feat,l These represent the three-dimensional coordinates of the l-th predicted intraoperative coordinate; a 11 ,…,a 33 Used to describe the linear deformation of an organization; t x , t y , and t z Used for overall translation of the organization; the last row is an auxiliary row for homogeneous coordinates;

[0117] The affine transformation matrix [T] is obtained by minimizing the sum of squares of the predicted intraoperative coordinates and the probe coordinates using least squares optimization.

[0118] It should be noted that the submatrix element a in the affine transformation matrix [T] 11 ,…,a 33 Describing the linear deformation of an organization, a 11 a is the scaling factor in the X-axis direction. 22 a is the scaling factor in the Y-axis direction. 33 The scaling factor is the Z-axis scaling factor; a 12 Let a be the shear coefficient in the XY plane. 13 Let a be the shear coefficient in the XZ plane. 21 The shear coefficient in the YX plane is related to a. 12 Combined with complete modeling of shear deformation in the XY plane; a 23 Let a be the shear coefficient in the YZ plane. 31 Let a be the shear coefficient of the ZX plane, and a 13 Combined with complete modeling of shear deformation in the ZX plane; a 32 The shear coefficient in the ZY plane is related to a. 23 The shear deformation in the YZ plane is modeled in conjunction with the complete model; the rotation of the tissue is not determined by a single parameter, but is achieved through a combination of multiple parameters in the submatrix. Due to the limitation of the angle, the shear coefficients need to work together to describe the shear deformation.

[0119] The mapping implementation logic is matrix multiplication. Expanding it allows us to observe how each parameter participates in coordinate transformation. The solution is obtained by minimizing the sum of squares, which is a current technology and will not be elaborated here. However, it should be noted that the object of minimizing the sum of squares is the predicted intraoperative coordinates and probe coordinates, not the three-dimensional coordinates of the navigation path points.

[0120] The points in the preoperative planning point set {P} are corrected using the affine transformation matrix [T].

[0121] Based on the calculated affine transformation matrix [T], correction calculations are performed using matrix multiplication. It should be noted that when performing matrix calculations, the last row of the coordinate vectors of the points in the preoperative planning point set {P} is padded with 1s as an auxiliary row for joint calculation.

[0122] It should be noted that the preoperative target area feature point coordinates are generated based on static images of tissue without deformation. During the operation, the target area may shift or change shape due to traction or resection of surrounding tissues. By mapping these preoperative feature points to the corrected intraoperative coordinates through [T], it can be ensured that the probe can accurately locate the actual boundary of the target area during the operation, thus avoiding misalignment and echoing the technical goal of improving the accuracy of target area positioning in the document.

[0123] Since dangerous structures will shift synchronously with tissue deformation, if the preoperative coordinates are still used to avoid risks, accidental injury may occur due to coordinate misalignment. By correcting the preoperative dangerous structure coordinates to the actual intraoperative coordinates through [T], it can be ensured that the corrected navigation path maintains a safe distance from the dangerous structure.

[0124] In this embodiment, the preset deviation threshold is 0.2mm; the threshold for error convergence is set to 0.3mm.

[0125] A multimodal image localization and calibration analysis system based on a magnetic induction array includes an initial registration module, a 3D modeling module, a point set acquisition and mapping module, an LM solver module, a deformation compensation module, and a visualization module.

[0126] The initial registration module is used to equip a magnetic induction array composed of magnetic sensors and preprocesses the magnetic induction array before startup.

[0127] The 3D modeling module is used to acquire preoperative images of the patient's surgical site and construct a 3D model of the surgical area based on the images;

[0128] The point set acquisition and mapping module is used to plan the stimulation target area and navigation path based on the three-dimensional model, generate a preoperative planning point set {P}, select several reference points in the preoperative planning point set {P} in the surgical area, use a magnetically excited handheld probe to contact the reference points in sequence, trigger the array sensor response, complete the rigid registration between the preoperative planning point set {P} and the actual reference points during the operation, and establish the initial coordinate mapping relationship.

[0129] The LM solver module is used to input the measured point set {P'} collected by the sensor into the LM solver, iterate the 5-DOF pose based on the magnetic dipole model, and output the probe coordinates in real time.

[0130] The deformation compensation module is used for coordinate correction based on the deformation compensation mechanism;

[0131] The visualization module is used to display the generated 3D model and the current probe coordinates in real time.

[0132] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A multimodal image localization and calibration analysis method based on a magnetic induction array, characterized in that: Specifically, the steps include the following: Equipped with a magnetic induction array composed of magnetic sensors, the magnetic induction array is preprocessed before startup; Preoperative images of the surgical site are collected, a three-dimensional model of the surgical area is constructed based on the images, the stimulation target area and navigation path are planned, and a preoperative planning point set {P} is generated. Select several reference points from the preoperative planning point set {P} within the surgical area, and use a magnetically excited handheld probe to contact the reference points in sequence to trigger the array sensor response, thereby completing the rigid registration between the preoperative planning point set {P} and the actual intraoperative reference points and establishing the initial coordinate mapping relationship. Based on the boundary of the surgical target area, the probe's magnetic excitation coil generates a magnetic field, and the array sensor collects triaxial magnetic field data according to a preset sampling period to generate an intraoperative measured point set {P'}. The measured point set {P'} collected by the sensor is input into the LM solver, and the 5-DOF pose is iteratively solved based on the magnetic dipole model, and the probe coordinates are output in real time. Coordinate correction is performed based on a deformation compensation mechanism, and the system performs real-time calibration based on the corrected coordinates. Specifically: When the deviation between the navigation path points in the preoperative planning point set {P} and the corresponding probe coordinates exceeds a preset deviation threshold, coordinate correction is performed through a deformation compensation mechanism. Initial matching of the navigation path points in the preoperative planning point set {P} with the probe coordinates is then performed to obtain data pairs, denoted as (p...). feat,l ,p' feat,l ); where l represents the identifier of the data pair, l is a positive integer, l∈[1,m], and m represents the total number of navigation path points; p feat,l p' represents the l-th navigation path point; feat,l Indicate the coordinates of the l-th probe; assign navigation path point p feat,l =[x feat,l ,y feat,l ,z feat,l ] T Mapped to predict intraoperative coordinates p'' feat,l Specifically: ;where x feat,l ,y feat,l ,z feat,l x'' represents the three-dimensional coordinates of the l-th navigation path point; feat,l ,y'' feat,l ,z'' feat,l These represent the three-dimensional coordinates of the l-th predicted intraoperative coordinate; a 11 ,…,a 33 Used to describe the linear deformation of an organization; t x , t y , and t z Used for the overall translation of an organization; The affine transformation matrix [T] is obtained by minimizing the sum of squares of the predicted intraoperative coordinates and the probe coordinates using least squares optimization. The points in the preoperative planning point set {P} are corrected using the affine transformation matrix [T].

2. The multimodal image localization and calibration analysis method based on a magnetic induction array according to claim 1, characterized in that: The preprocessing includes: The magnetic sensitivity of the sensor units of the magnetic induction array is calibrated to ensure that the triaxial measurement accuracy meets the error convergence threshold requirements; Inspect the handheld probe with the magnetic excitation coil and calibrate the magnetic signal using a magnetic calibration head to ensure the accuracy of the input parameters of the magnetic dipole model.

3. The multimodal image localization and calibration analysis method based on a magnetic induction array according to claim 2, characterized in that: Preoperative images of the patient's surgical site are acquired. A three-dimensional model of the surgical area is constructed based on the images. The stimulation target area and navigation path are planned, and a preoperative planning point set {P} is generated, specifically: Step S3-1: Select the appropriate imaging modality and fix the body position with a non-magnetic positioning patch on the body surface to avoid the initial deviation between the preoperative model and the actual anatomical structure during the operation due to differences in body position. Step S3-2: Noise processing is performed on the acquired images using a Kalman filter to eliminate magnetic interference. Multimodal effects are fused into a unified DICOM standard coordinate system by registration with preset marker points. At the same time, the regions of interest (ROIs) of relevant structures are extracted. Step S3-3: Using geometric modeling software, construct a three-dimensional model based on the image after noise processing, and mark the target area and the fitting area matching the magnetic induction array type in the three-dimensional model; wherein, the fitting area matching the magnetic induction array type includes the planar array size or the radius of curvature of the arc array; Step S3-4: Delineate the stimulation target area based on the three-dimensional model and record the target area boundary; Step S3-5: Plan the navigation path, ensuring that the path is within the sensing range of the magnetic induction array, and obtain the starting point, intermediate nodes, and ending point of the navigation path; wherein, the intermediate nodes are several nodes in the planned navigation path; Step S3-6: Use the preset marker points, the outline points of the target area boundary, and the navigation path points as the preoperative planning point set {P}.

4. The multimodal image localization and calibration analysis method based on a magnetic induction array according to claim 3, characterized in that: Several reference points are selected from the preoperative planned point set {P} within the surgical area. A magnetically excited handheld probe is used to sequentially contact these reference points, triggering the array sensor response. This completes the rigid registration between the preoperative planned point set {P} and the actual intraoperative reference points, establishing an initial coordinate mapping relationship. Specifically: Step S4-1: Select n planning points from the preoperative planning point set {P} within the surgical area as reference points, denoted as p. pre，i =[x pre，i ,y pre，i ,z pre，i ] T ;where x pre，i ,y pre，i ,z pre，i Let B represent the x, y, and y coordinates of the i-th reference point, respectively; i = 1, 2, ..., n, where n is the number of reference points; a handheld probe with a magnetic excitation coil is placed against the reference point, and the probe's magnetic excitation coil excites a magnetic field of a preset intensity. The triaxial magnetic field data B at that reference point is collected by an array sensor. int (X, Y, Z), where X, Y, and Z represent the magnetic field data of the horizontal, vertical, and longitudinal axes, respectively; the collected magnetic field data are input into the LM solver to iteratively solve for the three-dimensional coordinates of each actual reference point, denoted as p. int，i =[x int，i ,y int，i ,z int，i ] T ;where x int，i ,y int，i ,z int，i These represent the horizontal, vertical, and y-axis coordinates of the i-th actual reference point, respectively. Step S4-2: Solve the 3×3 rotation matrix R and the 3×1 translation vector t to make the reference point p pre，i After rotation and translation transformations, it is compared with the actual reference point p. int，i The spatial position error is minimized; Step S4-3: Calculate the registration error e of all reference points. i , e i =||p int，i -(R*p pre，i +t)||2;When the average registration error of all reference points is less than the error convergence threshold, the initial coordinate mapping relationship is valid; otherwise, the process is repeated from step S4-1.

5. The multimodal image localization and calibration analysis method based on a magnetic induction array according to claim 4, characterized in that: The specific methods for solving the 3×3 rotation matrix R and the 3×1 translation vector t are as follows: Step s1: Calculate the centroids of the n reference points and the actual reference point respectively, using the following formula: μ pre =(Σ n p pre，i ) / n; where μ pre This represents the centroid of n reference points; similarly, the centroid μ of the actual reference point is obtained. int ; Step s2: Decentrifuge the reference point to eliminate the interference of translation on the rotation solution. The centrifuged reference point q is then obtained. pre，i =p pre，i -μ pre The actual reference point q after centroid removal int，i =p int，i -μ int ; Step s3: Construct the covariance matrix H and solve for the rotation matrix R; Wherein, the covariance matrix H=Σ n q int，i *q T pre，i Singular value decomposition of the covariance matrix H yields H = U·Λ·V T Where Λ represents a 3rd order diagonal matrix; U and V represent 3rd order orthogonal matrices; when det(U·V) T If ) = 1, then the rotation matrix R = U·V T When det(U·V) T If ) = -1, then adjust the sign of the last column of U and then calculate R; Step s4: Substitute the rotation matrix R into the centroid relation to obtain the translation vector t = μ int -R·μ pre .

6. The multimodal image localization and calibration analysis method based on a magnetic induction array according to claim 5, characterized in that: Based on the boundary of the surgical target area, the probe's magnetic excitation coil generates a magnetic field, and the array sensor collects triaxial magnetic field data according to a preset sampling period, generating an intraoperative measured point set {P'}. The measured point set {P'} collected by the sensor is input into the LM solver, and the 5-DOF pose is iteratively solved based on the magnetic dipole model, and the probe coordinates are output in real time, specifically: The pose vector of the probe's built-in spherical permanent magnet is F=[x,y,z,θ,Φ]. T Where (x, y, z) represent the three-dimensional coordinates of the permanent magnet in the DICOM standard coordinate system; θ represents the pitch angle of the permanent magnet's magnetic moment; and Φ represents the yaw angle of the permanent magnet's magnetic moment. Wherein, the three-dimensional coordinates of the fixed position of the k-th sensor in the sensor array are r. s,k =[x s,k ,y s,k ,z s,k ] T ; The triaxial measured magnetic field of the k-th sensor is B. meas,k =[B x,meas,k B y,meas,k B z,meas,k ] T ; The predicted magnetic field at the k-th sensor, calculated based on the magnetic dipole model, is denoted as: B model,k (F)=[B x,model,k (F),B y,model,k (F),B z,model,k (F)] T ; The optimal pose F is solved iteratively using the LM solver. * The mathematical expression is: F * =argmin F S(F); where S(F) is the cost function; the optimal pose F is obtained by solving... * As probe coordinate output; S(F)=Σ k=1 N Σ j=x,y,z (B j,meas,k -B j,model,k (F)) 2 ; where j represents the identifier of the three-dimensional coordinate.

7. The multimodal image localization and calibration analysis method based on a magnetic induction array according to claim 6, characterized in that: The steps for calculating the predicted magnetic field at the k-th sensor based on the magnetic dipole model include: Based on the unified DICOM standard coordinate system, and according to the pose vector of the probe's built-in spherical permanent magnet as F=[x,y,z,θ,Φ] T The three-dimensional coordinate vector Fpos=[x,y,z] of the permanent magnet in the DICOM standard coordinate system is obtained. T ; The magnetic moment vector of the magnet is obtained as g(θ, Φ) = g[cosθcosΦ,cosθsinΦ,sinΦ]. T Where g represents the magnitude of the magnetic moment of the magnet; Calculate the position vector r from the permanent magnet to the sensor k =r s,k -Fpos; The predicted magnetic field at the k-th sensor in space is: B model,k (F)=μ0g*[3(g(θ,Φ)·r k ^)r k ^-g(θ,Φ)] / 4Πr k 3 ; Where μ0 represents the free permeability; r k r represents the distance from the permanent magnet to the k-th sensor; k ^ represents the position vector r from the permanent magnet to the k-th sensor. k The unit vector.

8. A multimodal image localization and calibration analysis system based on a magnetic induction array, employing the multimodal image localization and calibration analysis method based on a magnetic induction array as described in any one of claims 1-7, characterized in that: It includes an initial registration module, a 3D modeling module, a point set acquisition and mapping module, an LM solver module, a deformation compensation module, and a visualization module; The initial registration module is used to equip a magnetic induction array composed of magnetic sensors and to preprocess the magnetic induction array before startup. The 3D modeling module is used to acquire preoperative images of the patient's surgical site and construct a 3D model of the surgical area based on the images. The point set acquisition and mapping module is used to plan the stimulation target area and navigation path based on the three-dimensional model, generate a preoperative planning point set {P}, select several reference points in the preoperative planning point set {P} within the surgical area, use a magnetically excited handheld probe to contact the reference points in sequence, trigger the array sensor response, complete the rigid registration between the preoperative planning point set {P} and the actual intraoperative reference points, and establish an initial coordinate mapping relationship. The LM solver module is used to input the measured point set {P'} collected by the sensor into the LM solver, iteratively solve the 5-DOF pose based on the magnetic dipole model, and output the probe coordinates in real time. The deformation compensation module is used to perform coordinate correction based on the deformation compensation mechanism; The visualization module is used to display the generated 3D model and the current probe coordinates in real time.

Images