A flexible brain signal probe array positioning and registration method

Abstract

The application provides a flexible brain signal detection array positioning and registration method, realizes automatic and high-precision positioning of a multi-modal flexible brain signal detector, and quickly completes registration with an MRI / CT structural image of a subject individual, and solves technical problems of difficult deformation adaptation, low positioning efficiency, poor universality and low registration automation degree of the flexible array.

Description

Technical Field

[0001] This application relates to the field of brain signal detection technology, specifically to a method for locating and registering a brain signal detection array suitable for multimodal flexible devices. Background Technology

[0002] With the rapid advancement of microelectronics and micromachining technologies, extracranial brain signal detectors such as EEG electrodes, functional near-infrared spectroscopy detectors, optically pumped magnetometers (OPMs), and ultrasonic transducers are continuously miniaturizing, making it possible to integrate a large number of miniature detectors onto flexible substrates to form flexible detector arrays. Flexible arrays can adaptively conform to the curvature of different subjects' heads, acquiring brain signals with higher signal-to-noise ratios, thus becoming a trend in research and application. However, the deformation of the flexible array varies in each subject's head, resulting in differences in the spatial position of the detectors, requiring individual positioning. Current technologies have the following shortcomings: Rigid array positioning schemes are not applicable: Traditional MEG helmets and other devices use rigid structures with fixed relative positions between detectors, requiring only one overall registration. However, after the flexible array deforms, the relative positions of each detector change, rendering the rigid positioning scheme completely ineffective.

[0003] Manual marking is inefficient: manually marking the position of each detector for each subject is time-consuming and labor-intensive, and the marking accuracy is greatly affected by the operator's experience, making it difficult to meet the needs of high-throughput experiments.

[0004] Lack of a universal cross-modal localization framework: Existing solutions are mostly designed for a single modality (such as EEG only or fNIRS only), and lack a universal localization and registration method that can uniformly handle scalar and vector detectors and is applicable to multiple brain signal modalities.

[0005] The registration of the detector array with individual structural images is not highly automated: Currently, the process from spatial positioning of the detector to registration with the subject's MRI / CT structural images often requires multiple manual steps, which is cumbersome and leads to the accumulation of errors.

[0006] Therefore, there is an urgent need in this field for a positioning and registration method that is highly automated, has high positioning accuracy, is compatible with multimodal flexible detectors, and can directly complete anatomical spatial registration. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for positioning and registering a flexible brain signal detector array. This method enables automated and high-precision positioning of a multimodal flexible brain signal detector and rapidly completes registration with the MRI / CT structural images of the subject. It solves the technical problems of difficult deformation adaptation of flexible arrays, low positioning efficiency, poor versatility, and low degree of registration automation.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for locating and registering a flexible brain signal detection array. The method includes the following steps: Step S1: According to the type of detector in the flexible detection array, a corresponding number of marker points are pasted on the detectors to establish a geometric mapping relationship between the marker points and the detectors; Step S2: The set of marker points on the subject's head and the point cloud on the subject's head surface are acquired simultaneously through optical scanning; Step S3: A marker point template containing the correspondence between the marker point coordinates and the detectors is constructed on a standard head model; Step S4: The set of marker points and the marker point template acquired in the new scan are normalized, and a non-rigid point set registration algorithm is used to establish the point correspondence relationship, automatically completing the marker point marking and using the geometric mapping relationship between the marker points and the detectors established in Step S1 to calculate the pose of each detector in the new scan; Step S5: The point cloud on the subject's head surface is automatically rigidly registered with the scalp point cloud reconstructed from the subject's MRI / CT structural images to obtain a spatial transformation matrix, and the spatial transformation matrix is ​​applied to the pose of each detector to realize the location and registration of the detection array in the subject's anatomical space.

[0009] Furthermore, in step S1, the types of detectors in the flexible detection array include scalar detectors, single-axis vector detectors, dual-axis vector detectors, and tri-axis vector detectors; The scalar detector has one marker point attached; the single-axis vector detector has two marker points attached or one marker point with normal information attached; the dual-axis vector detector and the triaxial dual-axis vector detector have three non-collinear marker points attached.

[0010] Further, in step S1, establishing the geometric mapping relationship between the marker points and the detector includes: The position of the scalar detector is the superposition of the coordinates of the marker point and a known preset offset vector. The position of the single-axis vector detector is the superposition of the midpoint of two marker points and a known preset offset vector, and the sensitive axis is the unit vector of the line connecting the two marker points; or it is the superposition of the coordinates of a marker point with normal information and the product of the known scalar offset along the normal and the normal of the marker point. The positions of the dual-axis vector detector and the triaxial vector detector are obtained by superimposing a known preset offset vector on the centroids of three marker points. The three-axis attitude is obtained by constructing a rotation matrix from the three marker points through Gram-Schmidt orthogonalization. Let the first Each detector corresponds to Marker point ( ), the first The locations of the detectors are The geometric mapping relationship between the marker points and the detector is as follows:

[0011] in, Mark the coordinates of the point. This corresponds to the local number within the detector. For the detector position, , Here is the attitude rotation matrix. It is a special three-dimensional orthogonal group, representing the set of three-dimensional attitude rotation matrices. The known mapping function is determined by the geometric mapping relationship of the detector.

[0012] Furthermore, in step S2, the optical scanning employs a structured light scanner or other stereo vision system.

[0013] Furthermore, in step S3, the marker template only needs to be constructed once, and can be reused by all subsequent subjects; The marker template includes a mapping relationship between the global index of the marker, the corresponding detector number, and the local number within the detector.

[0014] Furthermore, in step S4, before normalizing the newly acquired set of marker points and the marker point template, the following steps are also included: The newly acquired set of marker points is aligned with the marker point template in terms of overall orientation. The overall orientation alignment method is either a fixed scanning orientation or automatic identification of facial orientation using the point cloud of the subject's head surface. A rotation matrix is ​​calculated to rotate the facial orientation to the standard orientation of the marker point template and applied to the set of marker points.

[0015] Furthermore, the point set of the marker template and the point set of the marker points obtained by the new scan are normalized respectively. The normalization process includes centroid translation and scale normalization. The scale normalization uses the sixth root of the determinant of the covariance matrix as a scaling factor to eliminate differences in the subject's head size.

[0016] Further, in step S4, the step of establishing point correspondence using a non-rigid point set registration algorithm, automatically marking the marker points, and calculating the pose of each detector using the geometric mapping relationship between the marker points and the detectors established in step S1 includes: The non-rigid point set registration algorithm is the CPD algorithm. It regards the normalized marker template point set as the center set of the Gaussian mixture model, and regards the marker point set obtained by the new scan after normalization as the observation data generated by the Gaussian mixture model. By maximizing the likelihood function estimation, the marker template point set is deformed into the new scan marker point set through a non-rigid transformation, and the transformed template point set is obtained. From the transformed template point set, find the nearest neighbor of each point in the new scan marker point set, and use the template marker information of the nearest neighbor as the label information of the new scan marker point; After each point in the new scan marker set obtains the tag information, for each new scan detector, the position corresponding to the new scan detector is extracted from the new scan marker set. The mapping function between the markers and the detector established in step S1 is used. Calculate the pose of each detector: ; in, new The coordinates of the new scan marker point, This corresponds to the local number within the detector. For the first The location of a new scanning detector , For the first The attitude rotation matrix of a new scanning detector. It is a special three-dimensional orthogonal group, representing the set of three-dimensional attitude rotation matrices. The known mapping function is determined by the geometric mapping relationship of the detector.

[0017] Further, in step S5, the point cloud on the subject's head surface is automatically and rigidly registered with the scalp point cloud reconstructed from the subject's MRI / CT structural images to obtain a spatial transformation matrix. This spatial transformation matrix is ​​then applied to the pose of each detector to achieve the positioning and registration of the detector array in the subject's anatomical space, including: Voxel downsampling was performed on the point cloud of the subject's head surface and the point cloud of the scalp, respectively; Fast point feature histograms were calculated for the point cloud of the subject's head surface and scalp after voxel downsampling, and used as local geometric feature descriptors. Coarse registration is performed based on local geometric feature descriptor matching and random sampling consensus algorithm to obtain the initial pose transformation matrix. The final spatial transformation matrix is ​​obtained by optimizing the initial pose transformation matrix using a fine registration algorithm. The final spatial transformation matrix is ​​applied to the poses of each detector in the newly scanned image to obtain the detector poses in the anatomical image coordinate system. Furthermore, the coarse registration uses the RANSAC algorithm, and the fine registration uses the Iterative Nearest Neighbor (ICP) algorithm.

[0018] Furthermore, the flexible brain signal detection array includes electroencephalography (EEG), functional near-infrared spectroscopy, optically pumped magnetometer, and ultrasound single-modal or multi-modal integrated flexible array.

[0019] Compared with the prior art, this application has the following advantages: This invention provides a method for localization and registration of a flexible brain signal detection array. The method includes the following steps: S1: According to the type of detector in the flexible detection array, a corresponding number of marker points are pasted on the detectors to establish a geometric mapping relationship between the marker points and the detectors; the marker points pasted on the detectors can be flexibly placed to ensure that they are imaged by optical scanning; S2: The set of marker points on the subject's head and the point cloud on the subject's head surface are acquired simultaneously through optical scanning; S3: A marker point template containing the correspondence between marker point coordinates and detectors is constructed on a standard head model; and the detector number corresponding to each marker point coordinate is also determined in this step, the detector number preferably corresponding to the channel number of data acquisition; S4: The newly acquired set of marker points and the marker point template are normalized, and a non-rigid point set registration algorithm is used. A point correspondence is established, and the marking of the marker points is automatically completed. The geometric mapping relationship between the marker points and the detectors established in step S1 is used to calculate the pose of each detector in the new scan. The detector number corresponding to the marker point acquired in the new scan is also determined. Therefore, the detector pose in this step corresponds to the detector number determined in step S3, and a connection between brain signal data and acquisition location is established. Step S5: The point cloud of the subject's head surface and the scalp point cloud reconstructed from the subject's MRI / CT structural image are automatically rigidly registered to obtain a spatial transformation matrix. The spatial transformation matrix is ​​applied to the pose of each detector to realize the positioning and registration of the detector array in the subject's anatomical space. It has the technical effects of strong versatility, high positioning accuracy, and the ability to quickly complete the matching and registration with the individual MRI / CT structural image of the subject, with good adaptability. Attached Figure Description

[0020] The above and other features, advantages, and aspects of the embodiments of this application will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein: Figure 1 This is a flowchart of a flexible brain signal detection array localization and registration method in an embodiment of this application; Figure 2 This is a schematic diagram showing the structure of attaching one marker point to the scalar detector in this embodiment and establishing its geometric mapping relationship with the scalar detector; Figure 3 This is a schematic diagram showing the structure in which two marker points are pasted on a single-axis vector detector in this embodiment of the application, and the geometric mapping relationship between them and the single-axis vector detector is established. Figure 4 A schematic diagram showing the structure in which three non-collinear markers are pasted on the dual-axis / triaxial vector detector in the embodiment of this application, and the geometric mapping relationship between them and the triaxial vector detector is established; Figure 5This embodiment of the application shows a subject's head wearing a flexible array, and a structural schematic diagram of the facial point cloud of the subject's head, which is simultaneously acquired by optical scanning, including the set of marker points on the subject's head and the point cloud on the surface of the subject's head. Figure 6 The three-dimensional visualization of the detector array on the subject's MRI structural images in the embodiments of this application includes the position (sphere) and axial direction (arrow) of each detector. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the appendices in the embodiments of this disclosure will be described below. Figures 1-6 The technical solutions in the embodiments of this disclosure are clearly and completely described. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0022] The technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of a person skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0023] This application provides a method for positioning and registering a flexible brain signal detection array, applicable to single-modal or multi-modal integrated flexible brain signal detection arrays such as EEG, functional near-infrared spectroscopy, optically pumped magnetometer, and ultrasound. It is fully compatible with various measurement detectors, including scalar, single-axis vector, and three-axis vector types, and can adapt to differences in head shape among different subjects and flexible array deformation scenarios. It achieves fully automatic high-precision positioning of the multi-modal flexible detection array detector pose and accurate registration in individual anatomical space.

[0024] In this embodiment, a structured light scanner or a stereo vision system is used to complete the optical scanning of the head. The static positioning accuracy of the device is not less than 0.1mm. It can simultaneously collect the three-dimensional coordinate data of the marker points on the subject's head and the selected three-dimensional point cloud data of the head surface in a single transaction, namely the marker point set and the point cloud of the subject's head surface. Based on the standard head model, a universal marker point template is constructed in one go. The template can be reused for all test subjects without repeated calibration.

[0025] like Figure 1 As shown in the figure, the flexible brain signal detection array localization and registration method provided in this application embodiment specifically includes the following steps: Step S1: Based on the type of detector in the flexible detection array, paste the corresponding number of markers on the detectors to establish the geometric mapping relationship between the markers and the detectors; Differentiated marker placement schemes are adopted for detectors of different measurement types, specifically: See Figure 2 For scalar detectors that only need to determine spatial location and not axial information (such as some EEG electrodes), a single marker point can be attached to the scalar detector. Let the coordinates of the marker point be... There is a known preset offset vector between the sensitive point and the marker point of the scalar detector. (In the detector's body coordinate system), the position of the scalar detector is: ; When the marker is directly pasted above the sensitive surface of the scalar detector Only the normal component exists, at this time for Projection along the local normal.

[0026] See Figure 3 For single-axis vector detectors that require determining the direction of the sensitive axis (such as OPM single-axis mode), attach two marker points to the single-axis vector detector. The position and sensitive axis direction of the single-axis vector detector are calculated as follows: , ; in, The known preset offset vector is the distance from the midpoint of the marker point to the sensitive point of the detector. is the unit direction vector of the sensitive axis.

[0027] Alternatively, when the vector detector is a single-axis vector detector, a marker point with normal information, such as a planar marker point, can be used. The normal of this marker point can be directly obtained through optical scanning. In this case, the single-axis vector detector only needs one marker point with normal information. Let the position of this marker point be... The normal of the marker point is ,but: , ; in, It is a known scalar offset along the normal direction.

[0028] like Figure 4 As shown, for dual-axis / triaxial vector detectors that require complete three-axis attitude information, such as the OPM three-axis mode, three non-collinear markers are pasted on the three-axis vector detector. These three points form the detector's body coordinate system, ensuring the accuracy of the three-axis attitude calculation. Specifically, the detector's position is first calculated. The formula for superimposing a known preset offset vector on the centroids of three marker points is as follows: ; The above-described detector location definition is a preferred example. Other linear combinations of offset vectors can also represent the detector position.

[0029] Then construct three orthogonal axis directions and define two edge vectors. , The details are as follows: ; The three orthogonal unit axis vectors are obtained through the Gram-Schmidt orthogonalization process, and the specific calculation formula is as follows: , , , ; in, Along the direction of the base of the triangle, Let be the normal vector of the triangular plane. Orthogonal to the previous two, forming a right-handed coordinate system, where R is the three-axis attitude rotation matrix. It represents a three-dimensional special orthogonal group (attitude rotation matrix space).

[0030] The above definition of the three-axis vector is a preferred example. , , Linear combinations of these can also construct other three-axis attitude rotation matrices, which can also be used to represent the detector's detection axis.

[0031] The formula for the general detector pose estimation algorithm is given by the following: Each detector corresponds to Marker point ( ), the first The locations of the detectors are The geometric mapping relationship between the marker points and the detector is as follows: ; in, Mark the coordinates of the point. This corresponds to the local number within the detector. For the detector position, Here is the attitude rotation matrix. The known mapping function is determined by the geometric mapping relationship of the detector.

[0032] This application employs a geometric relationship-based detector pose inversion design, allowing the marker points to be flexibly attached and arranged on the detector. For example, for detectors with cables, these marker points can be attached and arranged across the cables to ensure they can be identified during optical scanning; for detectors with uneven surfaces, planar fixtures can be constructed and fixed to the detector surface to place the marker points.

[0033] Step S2: Simultaneously acquire the set of marker points on the subject's head and the point cloud on the subject's head surface through optical scanning; like Figure 5 The image shows the head of a subject wearing a flexible array with multiple marker points (dots). An optical scanner was used to simultaneously acquire the set of marker points and the facial point cloud (displayed on a computer screen) of the subject's head surface. An optical 3D scanner or stereo vision system was used to scan the subject's head wearing the flexible detector array, simultaneously acquiring the set of marker points. Facial point clouds and the point cloud of the subject's head surface ,in The total number of marked points. , representing the coordinates of j marked points. The number of points in the facial point cloud is used. This embodiment uses the facial point cloud as an example. Of course, depending on the actual situation, the surface point clouds of other parts of the subject's head can also be selected and automatically rigidly registered with the scalp point cloud reconstructed from the subject's MRI / CT structural images.

[0034] When using planar marker points, the scan can also simultaneously acquire the normal vectors of each marker point. .

[0035] Step S3: Construct a reusable standard head model template. On the standard head model, construct a marker point template containing the correspondence between marker point coordinates and detectors. Through manual annotation, establish a one-to-one correspondence between each marker point and the detector to generate a standardized marker point template.

[0036] Specifically, a flexible detection array is worn on a standard head model, and a set of marker points is acquired through optical scanning. ; The operator manually marks the correspondence between each marker point and the detector to establish a mapping table: ; Marker template is defined as a tuple ; in, Represented as the global index of the marker point. For the corresponding detector number, The local number of this marker point within the detector marker point group ( The marker template only needs to be constructed once and can be reused for all subsequent subjects. The detector number usually corresponds to the data acquisition channel number, therefore the marker template already establishes the connection between brain signals and acquisition locations.

[0037] Step S4: Normalize the set of marker points obtained in the new scan and the marker point template, establish the point correspondence relationship using a non-rigid point set registration algorithm, automatically complete the marker point marking, and use the geometric mapping relationship between the marker points and the detector established in step S1 to calculate the pose of each detector in the new scan. Specifically, before normalizing the newly acquired set of marker points and the marker point template, the method further includes: The newly acquired set of marker points is aligned with the marker point template in terms of overall orientation. The overall orientation alignment method is either a fixed scanning orientation or automatic identification of facial orientation using the point cloud of the subject's head surface. A rotation matrix is ​​calculated to rotate the facial orientation to the standard orientation of the marker point template and applied to the set of marker points.

[0038] To ensure the convergence and accuracy of non-rigid point set registration, the newly acquired set of marker points is first... With marker template Overall orientation alignment can be achieved in the following ways: Method A (Fixed Scan Orientation): Each scan begins from a uniform standard orientation (e.g., the subject faces directly forward), ensuring consistency of the scanning coordinate system across different subjects. Method B (Facial Orientation Recognition): Utilizes point clouds of the subject's head surface. Automatically identify facial orientation (e.g., based on nose tip detection or PCA principal axis analysis) and calculate the rotation matrix to rotate the facial orientation to a standard orientation. And it applies to the set of marked points, as shown in the following formula:

[0039] in, new Indicates the coordinates of the new scan marker point. This indicates the coordinates of the new scan marker point after alignment. This indicates all newly scanned marker points.

[0040] The point set of the marker template and the point set of the marker points obtained by the new scan are normalized respectively. The normalization process includes centroid translation and scale normalization. The scale normalization uses the sixth root of the determinant of the covariance matrix as a scaling factor to eliminate the differences in the subject's head size.

[0041] Specifically, to eliminate the differences in overall scale and position caused by different head sizes, the point set of the marker template and the point set of the newly acquired markers are normalized to obtain the normalized marker template point set. The set of markers obtained from the new scan after normalization. The specific normalization process is as follows: Step 1: Translate the centroid and calculate the centroid. : ; Translate each point to a point with the origin as its centroid, and obtain the coordinates of the centroid-free marked points. : ; Step 2: Scale normalization, calculate the covariance matrix of the centroid-free point set. :

[0042] Calculate the determinant of the covariance matrix:

[0043] in, for The three eigenvalues ​​(i.e., the variance of the point set along the three principal axes), and the determinant Equal to the product of the variances of the three principal axes, it reflects the "variance volume" of the point set in three-dimensional space, and its dimensions are... Calculate the 6th root of the determinant as the scalar scaling factor: ; Normalized coordinates for: ; This leads to the normalized set of marker template points. The set of marker points obtained from the new scan after normalization. .

[0044] The reason for using the 6th root of the determinant of the covariance matrix as the scaling factor is as follows: the determinant of the covariance matrix It is the product of the variances of the three principal axes, reflecting the variance "volume" of the point set in three-dimensional space, and its dimensions are... Take its 6th root: ; This scaling factor converts the three-dimensional variance volume into the standard deviation order with the dimension of length (i.e., the cube root of the volume, corresponding to the side length). While preserving the absolute scale information of the point set, it effectively compresses the overall scale difference caused by different subject head sizes, enabling the two point sets to be registered in a similar numerical range in subsequent non-rigid registration, achieving a good balance between scale normalization and deformation information preservation.

[0045] By employing the CPD non-rigid point set registration algorithm to establish point correspondence, the marking of marker points is automatically completed, and the pose of each detector is calculated using the geometric mapping relationship between the marker points and the detectors established in step S1. Specifically, this includes: The normalized marker template point set Considered as the center set of the Gaussian mixture model, the set of marker points obtained after normalization of the new scan is... The observed data is considered to be generated by the Gaussian mixture model. By maximizing the likelihood function estimation, the template point set of the marker points is deformed into a new set of scanned marker points through a non-rigid transformation, resulting in the transformed template point set. Specifically, Template points are used as GMMs In this application scenario, both point sets consist of the same number of marked points. When establishing the point correspondence using the CPD algorithm, there are no outliers; therefore, the outlier weights are taken. For new scan markers probability density function for: ; The probability density of each component is an isotropic Gaussian: ; in, For the first The displacement vector of each template point The variance is Gaussian (shared by all components).

[0046] Displacement vector Kernel function parameterization is used to ensure space coherence. ; in, Let be the weight vector to be estimated. Here is the Gaussian radial basis kernel function: ; in, The smoothness of deformation is controlled and represented in matrix form; a kernel matrix is ​​defined. Its elements Weight matrix Then the displacement matrix for: ; Transformed template point set matrix for: ; The transformed first Template points: ; Transformed template point set ; in, This is a template point set matrix (one point per row).

[0047] Maximizing the regularized log-likelihood function is equivalent to minimizing the negative log-likelihood plus a regularization term: ; in, For the transformed first One template point, For regularization parameters, the regularization terms are... Constraining the smoothness of deformation.

[0048] The Expectation-Maximization (EM) algorithm is used to solve the problem iteratively. E-step (expectation step): Calculate the posterior probability matrix. Its elements are: ; Indicates new scan marker points By the The posterior probabilities generated by the GMM components.

[0049] Define a diagonal matrix ,in It is a column vector of all 1s. It is the sum of posterior probabilities.

[0050] M-step (maximization step): Fix the posterior probability and update the parameters. and : ; Simplifying, we get: ; in, This is the matrix of the new scan marker set.

[0051] Variance update: ; Iteration termination condition: When the change in the objective function is less than a threshold. (like It may stop when the maximum number of iterations is reached.

[0052] From the transformed template point set, find the nearest neighbor of each point in the new scan marker point set, and use the template marker information of the nearest neighbor as the label information of the new scan marker point; Specifically, after CPD registration is completed, for each marker point in the new scan point set... Find the transformed template point set Nearest neighbor in: ; This establishes a mapping between the new scan marker points and the template marker points. The marking information of template marker points New scan markers automatically inherit the detector's attribution label from the corresponding template markers. ; Detector pose reconstruction: After obtaining the tag information for each point in the new scan marker set, for each new scan detector, the position corresponding to the new scan detector is extracted from the new scan marker set. The mapping function between the markers and the detector established in step S1 is used. Calculate the pose of each detector: ; in, new The coordinates of the new scan marker point. This corresponds to the local number within the detector. For the first The location of a new scanning detector For the first The attitude rotation matrix of a new scanning detector. A known mapping function is designed to determine the geometric mapping relationship for the detector. Original coordinates are used here instead of normalized coordinates to ensure that the detector pose is in the scanner's physical coordinate system. The detector numbers in the marker template are mapped one-to-one with the new detector pose to establish a connection between brain signals and acquisition locations.

[0053] Step S5 is the registration step between the subject's head surface point cloud and the subject's structural image. This step involves registering the facial point cloud of the subject's head surface point cloud acquired by optical scanning. Scalp point cloud reconstructed from structural images (MRI or CT) of the subjects Registration is performed to obtain a spatial transformation matrix, thereby transforming the detector pose from the scanner coordinate system to the subject's anatomical image coordinate system, achieving the positioning and registration of the detector array in the subject's anatomical space, including: Voxel downsampling was performed on the point cloud of the subject's head surface and the point cloud of the scalp, specifically on the facial point cloud of the subject's head surface. ( (points) and the aforementioned scalp point cloud ( Voxel downsampling is performed on each point to reduce computational complexity and enhance registration robustness.

[0054] For a given voxel size Divide the three-dimensional space into sections with sides of length . For a voxel grid, for each non-empty voxel, calculate the centroid of all points within it as the representative point after downsampling:

[0055] in Let the point set within this voxel be denoted as and the downsampled point cloud as . and .

[0056] Fast Point Feature Histograms (FPFHs) were calculated for the point clouds of the subject's head surface and scalp after voxel downsampling, serving as local geometric feature descriptors. Specifically: Normal vector estimation: For points in the subject's head surface point cloud and scalp point cloud after voxel downsampling In its radius Neighborhood Within the region, the normal vector is estimated by fitting the local plane using PCA, and the neighborhood covariance matrix is ​​calculated. The calculation formula is as follows: ; in, Represents points within the neighborhood. Let the neighborhood centroid and normal vector be... Pick The eigenvector corresponding to the smallest eigenvalue.

[0057] SPFH (Simplified Point Feature Histogram) calculation: For points and every point in its neighborhood Establish the Darboux coordinate system:

[0058]

[0059]

[0060] Calculate the three angular features:

[0061]

[0062]

[0063] Will Divided into A range (usually) ), and the frequency of each interval is counted to form an SPFH histogram. .

[0064] FPFH calculation: FPFH integrates neighborhood information based on SPFH: ; RANSAC-based coarse registration: Utilizing the FPFH local geometric feature descriptor, coarse registration is performed using the Random Sample Consensus (RANSAC) algorithm to find a rigid transformation that roughly aligns the two point clouds. .

[0065] Feature correspondence establishment: For Each point in ,exist Search for the point closest to the FPFH feature as the candidate corresponding point:

[0066] for One point in it.

[0067] RANSAC iteration: Randomly sample 3 pairs of corresponding points from the candidate corresponding set. .

[0068] Calculate candidate rigid transformations using 3 sets of corresponding point pairs , Representing the Euclidean transformation group, specifically, it is solved using the SVD method: Let the source point be the center of mass. Target point centroid Calculate the cross covariance matrix :

[0069] right Perform SVD decomposition ,but:

[0070]

[0071] Calculate the number of interior points: Acting on All points, after statistical transformation, and... The nearest neighbor distance is less than the threshold The logarithm of the point.

[0072] Repeat the above operations for a total of Next, the transformation with the most interior points is selected as the initial pose transformation matrix for the coarse registration result. .

[0073] Fine registration based on iterative nearest point ICP: the initial pose transformation matrix of the coarse registration result is used. The data is applied to the original (non-downsampled) head surface point cloud and then compared with the original scalp point cloud. Fine registration is performed using the point-to-point ICP (Iterative ClosestPoint) algorithm, as detailed below: Let the first The transformation matrix for the next iteration is (Initial value) The iteration process is as follows: Correspondence establishment: For the transformed head surface point cloud Each point in ,exist Search for nearest neighbor in the search bar: ; for Hit the point.

[0074] Transformation estimation: Minimize point-to-point distance: ; Similarly, the SVD method is used to solve for the optimal rigid transformation, and the centroid of the source point set is set. target point set centroid Cross-covariance matrix:

[0075] SVD decomposition The incremental transformation is then: ; ; The cumulative transformation matrix is ​​updated as follows: ; Convergence criterion: When the change in the root mean square distance between two consecutive iterations is less than a threshold. The iteration may stop when the maximum number of iterations is reached, and the spatial transformation matrix updated in the last iteration is the final spatial transformation matrix, denoted as . .

[0076] The final spatial transformation matrix The poses of each detector in the newly scanned image are applied to obtain the detector poses in the anatomical image coordinate system, specifically: The representation of the detector array pose in anatomical space, and the final spatial transformation matrix obtained by registration. By applying the detector position and axis in the scanner coordinate system, the detector pose in the anatomical image coordinate system is obtained.

[0077] Position transformation: For the first Each detector location :

[0078] Right now:

[0079] Axial transformation: For direction vectors (no translation required), only the rotation matrix is ​​applied:

[0080] Complete attitude transformation:

[0081] At this point, the position and orientation of all detectors in the subject's individual anatomical image coordinate system have been determined and can be directly used for subsequent brain signal source localization and calibration.

[0082] In summary, the flexible brain signal detector array localization and registration method provided by this invention achieves automated and high-precision localization of multimodal flexible brain signal detectors and rapidly completes registration with the MRI / CT structural images of the subject, solving the technical problems of difficult adaptation to flexible array deformation, low localization efficiency, poor versatility, and low degree of registration automation. The specific technical effects achieved are as follows: High versatility: This invention proposes a universal layout scheme (1-3 marker points corresponding to scalar / single-axis / dual-axis / triaxial vector detectors), thus it is not dependent on the specific shape of the detector and is applicable to multimodal and multi-shaped detectors. It can be applied to multimodal flexible detection arrays such as EEG, near-infrared, magnetoencephalography, and ultrasound, and can uniformly process scalar and vector detectors, providing a universal localization and registration method for brain signal research and application of different modalities.

[0083] High positioning accuracy: Using an optical scanner with an accuracy of no less than 0.1mm, the positioning error of the detector can be controlled within 1mm, which can meet the requirements of brain science research and clinical diagnosis and treatment for the positioning accuracy of the detector.

[0084] High degree of automation: The detector array is positioned in the subject's anatomical space in one step, reducing manual operation steps, lowering human error, and improving experimental efficiency and data accuracy. In actual tests, the 71 triaxial detectors distributed across the entire head can complete positioning and transfer to the subject's anatomical space within 35 seconds, with an average repeatability error of only 0.48 mm.

[0085] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for localization and registration of a flexible brain signal detection array, characterized in that, Includes the following steps: Step S1: Based on the type of detector in the flexible detection array, paste the corresponding number of markers on the detectors to establish the geometric mapping relationship between the markers and the detectors; Step S2: Simultaneously acquire the set of marker points on the subject's head and the point cloud on the subject's head surface through optical scanning; Step S3: Construct a marker template on the standard head model that includes the correspondence between the marker coordinates and the detector; Step S4: Normalize the set of marker points obtained in the new scan and the marker point template, establish the point correspondence relationship using a non-rigid point set registration algorithm, automatically complete the marker point marking, and use the geometric mapping relationship between the marker points and the detector established in step S1 to calculate the pose of each detector in the new scan. Step S5: Perform automated rigid registration between the point cloud on the subject's head surface and the scalp point cloud reconstructed from the subject's MRI / CT structural images to obtain a spatial transformation matrix. Apply the spatial transformation matrix to the pose of each detector to achieve the positioning and registration of the detector array in the subject's anatomical space.

2. The method according to claim 1, characterized in that, In step S1, the types of detectors in the flexible detection array include scalar detectors, single-axis vector detectors, dual-axis vector detectors, and tri-axis vector detectors; The scalar detector has one marker point affixed; the single-axis vector detector has two marker points affixed or one marker point with normal information affixed; the dual-axis vector detector and the triaxial vector detector have three non-collinear marker points affixed.

3. The method according to claim 2, characterized in that, In step S1, establishing the geometric mapping relationship between the marker points and the detector includes: The position of the scalar detector is the superposition of the coordinates of the marker point and a known preset offset vector. The position of the single-axis vector detector is the superposition of the midpoint of two marker points and a known preset offset vector, and the sensitive axis is the unit vector of the line connecting the two marker points; or it is the superposition of the coordinates of a marker point with normal information and the product of the known scalar offset along the normal and the normal of the marker point. The positions of the dual-axis vector detector and the tri-axis vector detector are obtained by superimposing the centroids of three marker points with a known preset offset vector, and by constructing a rotation matrix from the three marker points through Gram-Schmidt orthogonalization. Let the first Each detector corresponds to Marker points, The first The locations of the detectors are The geometric mapping relationship between the marker points and the detector is as follows: in, Mark the coordinates of the point. This corresponds to the local number within the detector. For the detector position, , Here is the attitude rotation matrix. It is a special three-dimensional orthogonal group, representing the set of three-dimensional attitude rotation matrices. The known mapping function is determined by the geometric mapping relationship of the detector.

4. The method according to claim 1, characterized in that, In step S3, the marker template only needs to be constructed once, and can be reused by all subsequent subjects; The marker template includes a mapping relationship between the global index of the marker, the corresponding detector number, and the local number within the detector.

5. The method according to claim 1, characterized in that, In step S4, before normalizing the newly acquired set of marker points and the marker point template, the following steps are also included: The newly acquired set of marker points is aligned with the marker point template in terms of overall orientation. The overall orientation alignment method is either a fixed scanning orientation or automatic identification of facial orientation using the point cloud of the subject's head surface. A rotation matrix is ​​calculated to rotate the facial orientation to the standard orientation of the marker point template and applied to the set of marker points.

6. The method according to claim 5, characterized in that, The point set of the marker template and the point set of the marker points obtained by the new scan are respectively normalized. The normalization process includes centroid translation and scale normalization.

7. The method according to claim 3, characterized in that, In step S4, the step of establishing point correspondence using a non-rigid point set registration algorithm, automatically marking the marker points, and calculating the pose of each detector using the geometric mapping relationship between the marker points and the detectors established in step S1 includes: The non-rigid point set registration algorithm is the CPD algorithm. It regards the normalized marker template point set as the center set of the Gaussian mixture model, and regards the marker point set obtained by the new scan after normalization as the observation data generated by the Gaussian mixture model. By maximizing the likelihood function estimation, the marker template point set is deformed into the new scan marker point set through a non-rigid transformation, and the transformed template point set is obtained. From the transformed template point set, find the nearest neighbor of each point in the new scan marker point set, and use the template marker information of the nearest neighbor as the label information of the new scan marker point; After each point in the new scan marker set obtains the tag information, for each new scan detector, the position corresponding to the new scan detector is extracted from the new scan marker set. The mapping function between the markers and the detector established in step S1 is used. Calculate the pose of each detector: ； in, new The coordinates of the new scan marker point, This corresponds to the local number within the detector. For the first The location of a new scanning detector , For the first The attitude rotation matrix of a new scanning detector. It is a special three-dimensional orthogonal group, representing the set of three-dimensional attitude rotation matrices. The known mapping function is determined by the geometric mapping relationship of the detector.

8. The method according to claim 1, characterized in that... In step S5, the point cloud of the subject's head surface is automatically and rigidly registered with the scalp point cloud reconstructed from the subject's MRI / CT structural images to obtain a spatial transformation matrix. This spatial transformation matrix is ​​then applied to the pose of each detector to achieve the positioning and registration of the detector array in the subject's anatomical space. This includes: Voxel downsampling was performed on the point cloud of the subject's head surface and the point cloud of the scalp, respectively; Fast point feature histograms were calculated for the point cloud of the subject's head surface and scalp after voxel downsampling, and used as local geometric feature descriptors. Coarse registration is performed based on local geometric feature descriptor matching and random sampling consensus algorithm to obtain the initial pose transformation matrix. The final spatial transformation matrix is ​​obtained by optimizing the initial pose transformation matrix using a fine registration algorithm. The final spatial transformation matrix is ​​applied to the pose of each detector in the newly scanned image to obtain the detector pose in the anatomical image coordinate system.

9. The method according to claim 8, characterized in that, The coarse registration uses the RANSAC algorithm, and the fine registration uses the Iterative Nearest Neighbor (ICP) algorithm.

10. The method according to claim 1, characterized in that, The flexible brain signal detection array includes electroencephalography (EEG), functional near-infrared spectroscopy (FIR), optically pumped magnetometer, and ultrasound single-modal or multi-modal integrated flexible array.

Images