Two-stage automatic registration method for brain magnetic measurement helmet and brain MRI data

Abstract

The invention relates to the technical field of data processing, in particular to a brain magnetic measurement helmet and brain MRI data two-stage automatic registration method, which comprises the following steps of: acquiring brain magnetic measurement helmet and brain MRI data, and storing the brain magnetic measurement helmet and brain MRI data; generating a point cloud based on a brain magnetic measurement helmet and a three-dimensional model of MRI data; performing preliminary registration on the generated point cloud by applying a random sampling consistency algorithm; performing further fine alignment on the point clouds subjected to deicing configuration by using an iterative nearest point algorithm; according to the method, an automatic key point extraction and registration algorithm is adopted, human intervention is greatly reduced, errors in the manual calibration process are avoided, and the consistency and high precision of the registration result are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly relates to a two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data. Background Art

[0002] Magnetoencephalography is a non-invasive brain functional imaging technique that infers brain activities based on weak magnetic field signals generated by nerve currents outside the head. With its high temporal resolution at the millisecond level and spatial resolution at the millimeter level, this technique is widely used in the field of neuroscience and can provide important support for brain function research and brain disease diagnosis.

[0003] In magnetoencephalography analysis, the precise positioning of MEG sensor positions is crucial. Especially under the condition of a multi-sensor array, any position deviation will have an adverse impact on the accuracy of subsequent brain activity source localization and signal tracing.

[0004] To improve the diagnostic and research accuracy of magnetoencephalography, it is usually necessary to perform high-precision registration on magnetoencephalography data and brain magnetic resonance imaging data to establish a unified spatial coordinate system. Due to the different coordinate systems and significant differences in resolution between MEG sensors and MRI systems when collecting data, there are significant technical difficulties in spatial alignment between the two types of data.

[0005] Traditionally, the registration process of magnetoencephalography and MRI relies on manually selecting feature points. Usually, operators need to mark reference points on brain structure images. This process is not only cumbersome and time-consuming but also leads to inconsistencies in registration results due to human operation errors, thereby affecting the matching accuracy and analysis reliability of the data.

[0006] In response to the above problems, some registration methods based on automated algorithms have gradually emerged in recent years. These methods usually rely on automatically extracting data feature points or performing rough registration on point clouds. To a certain extent, these methods reduce manual intervention and improve the registration efficiency. However, current automated methods often require a large amount of computing resources during the registration process, and are sensitive to the initial position for complex anatomical structures or non-standard data, and are prone to falling into local optima, making it difficult to balance registration accuracy and real-time performance simultaneously.

[0007] Therefore, the prior art has not fully solved the problem of automatic and high-precision registration between magnetoencephalography and MRI data. Achieving precise alignment between data sources under different spatial resolutions and different scanning conditions remains a technical bottleneck in this field; Therefore, there is an urgent need to develop an optimized solution that ensures accuracy while improving the registration efficiency. Summary of the Invention

[0008] In view of the above-mentioned disadvantages of the prior art, the present invention provides a two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data, which solves the technical problems proposed in the above background art.

[0009] To achieve the above objectives, the present invention is realized through the following technical solutions: A two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data, comprising the following steps: Obtain the magnetoencephalography measurement helmet and brain MRI data, and store the magnetoencephalography measurement helmet and brain MRI data; generate a point cloud based on the three-dimensional models of the magnetoencephalography measurement helmet and MRI data; apply the random sample consensus algorithm to perform a preliminary registration on the generated point cloud; use the iterative closest point algorithm to perform further fine alignment on the point cloud after de-icing configuration; output the final transformation matrix according to the alignment result.

[0010] Furthermore, the acquisition operation process of the magnetoencephalography measurement helmet and brain MRI data is as follows: Load model files: Load the magnetoencephalography measurement helmet and MRI model files from the specified path, and the file formats include STL; Check file validity: Confirm that each model file contains readable three-dimensional data; Verify data format: Load the STL file in the three-dimensional grid; Among them, in the stage of checking file validity, when an unreadable model file is identified, the model file is discarded, and a check message is generated synchronously. The content of the check message includes the success probability of reading the model file and the name of the model file for which the discard operation is performed.

[0011] Furthermore, the three-dimensional models of the magnetoencephalography measurement helmet and MRI data applied in the point cloud generation stage are derived from the loaded magnetoencephalography measurement helmet and MRI model files opened through three-dimensional grid loading; Steps for generating a point cloud based on the three-dimensional models of the magnetoencephalography measurement helmet and MRI data: Convert the three-dimensional grid to a point cloud: Generate a point cloud from the three-dimensional surface of the helmet and MRI models through a uniform sampling algorithm, so that each three-dimensional grid area is sampled into several points; Point cloud quantity configuration: According to the parameters in the configuration file, specify the quantity of the point cloud generated by each three-dimensional model; Point cloud preprocessing: Perform denoising processing on the generated point cloud data, remove isolated noise points, and use voxel downsampling technology to reduce the quantity of the point cloud data to two-thirds of the original point cloud data.

[0012] Furthermore, the denoising processing logic of the point cloud data is as follows: Let the point cloud data set be , where , which are the three-dimensional coordinates of the i-th point. For each point in the point cloud, calculate the average distance to its k nearest neighbor points ; , is the j-th nearest neighbor point of point ; is the Euclidean distance; Calculate the mean and standard deviation of the average distances of all points and standard deviation :

[0013] Among them, a threshold T is set by the client, and by default, take , where m is a constant. Points with an average distance greater than the threshold T are determined as outliers and removed.

[0014] Furthermore, the Random Sample Consensus (RANSAC) algorithm obtains the maximum consistent point pairs by randomly selecting sample points from the point cloud and performing model fitting.

[0015] Furthermore, the initial point cloud registration step includes: Feature calculation: Calculate the local feature descriptors of the magnetoencephalography (MEG) measurement helmet point cloud and the magnetic resonance imaging (MRI) point cloud; Coarse alignment: Use the RANSAC algorithm to perform coarse alignment based on the extracted feature descriptors. Determine the initial transformation matrix by calculating the matching point pairs between the point clouds.

[0016] Distance threshold: During the application stage of the RANSAC algorithm, synchronously set a distance threshold to determine which point pairs can be regarded as matching points and control the accuracy of the initial registration; Among them, the feature calculation results are calculated for the features of each point cloud through the Fast Point Feature Histogram (FPFH) algorithm. The distance threshold is user-defined by the client and is initially set to 4.0 mm by default.

[0017] Furthermore, the fine alignment operation after the point cloud completes the initial registration is as follows: Fine alignment: Use the initial registration result obtained from the RANSAC algorithm as the initial transformation matrix of the Iterative Closest Point (ICP) algorithm to perform fine alignment; Distance threshold: During the ICP algorithm process, set a smaller distance threshold; Maximum number of iterations: Set the maximum number of iterations until the algorithm converges; Among them, the ICP algorithm adjusts the position of the point cloud iteratively. During the distance threshold setting stage, it is default set to 0.5 times the voxel size, and the maximum number of iterations is set to 2000 times.

[0018] Furthermore, after the point cloud undergoes two-stage registration using the Random Sample Consensus (RANSAC) algorithm and the Iterative Closest Point (ICP) algorithm, the resulting final transformation matrix contains the rotation and translation information between the magnetoencephalography (MEG) measurement helmet and the MRI point cloud.

[0019] Furthermore, the upper-left 3x3 part of the final transformation matrix represents the rotation relationship, and the translation vector is located in the last column of the matrix, representing the translation relationship, that is, the translation information. At the output stage of the final transformation matrix, the final transformation matrix is saved to an XML file. Among them, at the output stage of the final transformation matrix, a 3D visualization tool is used to check the registration effect and verify whether the MEG measurement helmet and the MRI point cloud are accurately aligned. When the inspection result is positive, the output of the final transformation matrix is further executed. Otherwise, the obtained MEG measurement helmet and brain MRI data are reset and executed.

[0020] Adopting the technical solution provided by the present invention, compared with the known public technologies, it has the following beneficial effects: 1. The present invention provides a two-stage automatic registration method for a magnetoencephalography (MEG) measurement helmet and brain MRI data. This method effectively reduces human intervention during execution and improves the registration accuracy: Traditional registration methods for magnetoencephalogram (MEG) and MRI data usually rely on manual selection of feature points, which are easily interfered by factors such as the operator's experience, hand tremors, and skin deformation, resulting in registration errors. In contrast, the present invention uses automated key point extraction and registration algorithms, greatly reducing human intervention, avoiding errors in the manual calibration process, and ensuring the consistency and high accuracy of the registration results. 2. Combining a two-stage process of coarse alignment and fine registration, the present invention uses the Random Sample Consensus (RANSAC) algorithm for preliminary coarse alignment and then uses the Iterative Closest Point (ICP) algorithm for fine alignment. This two-stage registration strategy not only improves the robustness of the registration but also can handle data under different resolutions and different scanning conditions. Through the coarse alignment of the RANSAC algorithm, the error sources can be quickly reduced, and the ICP algorithm further refines the registration, avoiding the local optimum problem of a single method in the case of complex anatomical structures or more noise.

[0021] 3. Optimizing the balance between computational efficiency and accuracy: Existing automated registration methods are usually sensitive to the initial position and require a large amount of computational resources, making it difficult to balance registration accuracy and real-time performance. In contrast, the present invention uses a two-stage registration strategy with step-by-step optimization, greatly improving the computational efficiency while ensuring the registration accuracy, making this method more practical in real-time applications. Especially in the case of multi-sensor arrays, it can ensure the accurate alignment of MEG and MRI data and is applicable to various scenarios such as brain function research and brain disease diagnosis.

[0022] 4. Adapt to different resolutions and scanning conditions: Due to the resolution differences between MEG and MRI data, the registration method of the present invention can process data with different resolutions, ensuring stable registration results under various scanning conditions. This advantage enables the method to have wide applicability in different devices and scanning environments.

[0023] 5. Improve the stability of the registration result: By adopting a fine alignment step with a small distance threshold and a high number of iterations, the present invention can ensure the high stability of the final registration result, avoiding the accuracy problems caused by insufficient iteration times or too large distance thresholds in traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0025] Figure 1 It is a schematic flowchart of a two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data; Figure 2 It is a point cloud map before registration in the present invention; Figure 3 It is a point cloud map after registration in the present invention; Figure 4 It is a schematic diagram of manual registration; Figure 5 It is a schematic diagram of the first effect example of automatic registration; Figure 6 It is a schematic diagram of the second effect example of automatic registration. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0027] The following further describes the present invention with reference to the embodiments.

[0028] Embodiment: A two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data in this embodiment is as Figure 1As shown, it includes the following steps: Obtain magnetoencephalography measurement helmet and brain MRI data, and store the magnetoencephalography measurement helmet and brain MRI data; The acquisition operation process of the magnetoencephalography measurement helmet and brain MRI data is as follows: Load model files: Load magnetoencephalography measurement helmet and MRI model files from the specified path, and the file format includes STL; Check file validity: Confirm that each model file contains readable three-dimensional data; Verify data format: Load the STL file in the three-dimensional grid; Among them, in the stage of checking file validity, when an unreadable model file is identified, discard the model file, and synchronously generate a check message. The content of the check message includes the success probability of reading the model file and the name of the model file for which the discard operation is performed; Generate a point cloud based on the three-dimensional models of the magnetoencephalography measurement helmet and MRI data; The three-dimensional models of the magnetoencephalography measurement helmet and MRI data applied in the point cloud generation stage are derived from the loaded magnetoencephalography measurement helmet and MRI model files opened through three-dimensional grid loading; Steps to generate a point cloud based on the three-dimensional models of the magnetoencephalography measurement helmet and MRI data: Convert the three-dimensional grid to a point cloud: Generate a point cloud from the three-dimensional surface of the helmet and MRI models through a uniform sampling algorithm, so that each three-dimensional grid area is sampled into several points; Point cloud quantity configuration: Specify the number of point clouds generated for each three-dimensional model according to the parameters in the configuration file; Point cloud preprocessing: Denoise the generated point cloud data, remove isolated noise points, and use voxel downsampling technology to reduce the quantity of point cloud data to two-thirds of the original point cloud data; The denoising processing logic of the point cloud data is as follows: Let the point cloud data set be , where , is the three-dimensional coordinate of the i-th point. For each point in the point cloud, calculate its average distance to k nearest neighbor points ; , is the j-th nearest neighbor point of point ; is the Euclidean distance; Calculate the mean value and standard deviation of the average distances of all points:

[0029] Among them, a threshold T is set by the user side, and the default value is taken where \(m\) is a constant, points with an average distance greater than the threshold \(T\) are determined as outliers and removed; Through the above formula, denoising processing logic is provided for the point cloud data; Apply the Random Sample Consensus (RANSAC) algorithm to perform preliminary registration on the generated point cloud; The Random Sample Consensus (RANSAC) algorithm randomly selects sample points from the point cloud and performs model fitting to obtain the maximum consistent point pairs; The steps of preliminary point cloud registration include: Feature calculation: Calculate the local feature descriptors of the magnetoencephalography (MEG) measurement helmet point cloud and the magnetic resonance imaging (MRI) point cloud; Coarse alignment: Use the Random Sample Consensus (RANSAC) algorithm to perform coarse alignment based on the extracted feature descriptors. By calculating the matching point pairs between the point clouds, the preliminary transformation matrix is determined.

[0030] Distance threshold: During the application stage of the Random Sample Consensus (RANSAC) algorithm, a distance threshold is synchronously set to determine which point pairs can be regarded as matching points and control the accuracy of preliminary registration; Among them, the feature calculation results are calculated for the features of each point cloud through the Fast Point Feature Histogram (FPFH) algorithm. The distance threshold is user-defined, and the initial default setting is 4.0 mm; Use the Iterative Closest Point (ICP) algorithm to perform further fine alignment on the point cloud after de-icing configuration; The fine alignment operation of the point cloud after preliminary registration is as follows: Fine alignment: Based on the preliminary registration result obtained by the Random Sample Consensus (RANSAC) algorithm as the initial transformation matrix of the ICP algorithm, perform fine alignment; Distance threshold: During the ICP algorithm process, set a smaller distance threshold; Maximum number of iterations: Set the maximum number of iterations until the algorithm converges; Among them, the ICP algorithm adjusts the position of the point cloud iteratively. During the distance threshold setting stage, the default setting is 0.5 times the voxel size, and the maximum number of iterations is set to 2000 times; After two-stage registration of the point cloud by the Random Sample Consensus (RANSAC) algorithm and the Iterative Closest Point (ICP) algorithm, the obtained final transformation matrix contains the rotation and translation information between the magnetoencephalography (MEG) measurement helmet and the magnetic resonance imaging (MRI) point cloud; Output the final transformation matrix according to the alignment result; The upper left 3x3 part of the final transformation matrix represents the rotation relationship, and the translation vector is located in the last column of the matrix, representing the translation relationship, that is, the translation information. During the final transformation matrix output stage, the final transformation matrix is saved to an XML file; Among them, in the output stage of the final transformation matrix, a three-dimensional visualization tool is used to check the registration effect and verify whether the magnetoencephalography measurement helmet and the MRI point cloud are accurately aligned. When the inspection result is yes, the output of the final transformation matrix is further executed. Otherwise, the obtained magnetoencephalography measurement helmet and brain MRI data are reset and executed.

[0031] In this embodiment, through the method in the above embodiment, the automatic key point extraction technology is used to realize the step-by-step optimization process combining rough alignment and fine registration, so as to efficiently and accurately align the magnetoencephalogram and MRI data under different resolutions and scanning conditions. Compared with the traditional manual method, this method reduces the influence of human intervention on the registration result, significantly improves the registration accuracy and stability, and has wide applicability and promotion value in real-time application scenarios.

[0032] In summary, the method in the above embodiments effectively reduces human intervention during execution and improves the registration accuracy: Traditional magnetoencephalogram (MEG) and magnetic resonance imaging (MRI) data registration methods usually rely on manual selection of feature points, which are easily interfered by factors such as the experience of operators, hand tremors, and skin deformation, resulting in registration errors. In contrast, the present invention adopts an automated key point extraction and registration algorithm, which greatly reduces human intervention, avoids errors in the manual calibration process, and ensures the consistency and high accuracy of the registration results; Combining the two-stage process of rough alignment and fine registration, the present invention uses the Random Sample Consensus (RANSAC) algorithm for preliminary rough alignment, and then uses the Iterative Closest Point (ICP) algorithm for fine alignment. This two-stage registration strategy not only improves the robustness of registration, but also can process data under different resolutions and different scanning conditions. Through the rough alignment of the RANSAC algorithm, the error sources can be quickly reduced, while the ICP algorithm further refines the registration, avoiding the local optimum problem of a single method in the case of complex anatomical structures or more noise; Optimize the balance between computational efficiency and accuracy: Existing automated registration methods are usually sensitive to the initial position and require a large amount of computational resources, making it difficult to balance registration accuracy and real-time performance. In contrast, the present invention greatly improves the computational efficiency through the two-stage registration strategy of step-by-step optimization, while ensuring the registration accuracy, making the method have better practicality in real-time applications. Especially in the case of multi-sensor arrays, it can ensure the accurate alignment of MEG and MRI data, and is applicable to various scenarios such as brain function research and brain disease diagnosis; Adapt to different resolutions and scanning conditions: Due to the resolution differences between MEG and MRI data, the registration method of the present invention can process data with different resolutions, ensuring stable registration results under various scanning conditions. This advantage makes the method widely applicable in different devices and different scanning environments; Improve the stability of the registration results: By adopting a fine alignment step with a small distance threshold and a high number of iterations, the present invention can ensure the high stability of the final registration results, avoiding the accuracy problems caused by insufficient iteration times or too large distance thresholds in traditional methods; To verify the effectiveness and superiority of the technical solutions in the above embodiments, we conducted a series of comparative experiments. The purpose of this experiment is to evaluate the performance of the method of the present invention in terms of registration accuracy, registration time, and robustness to different initial conditions.

[0033] Experimental setup: The data used in this invention was sourced from 30 healthy adult volunteers (aged 20 - 40 years old, 15 males and 15 females, all of whom had signed informed consent forms). Each volunteer underwent the following data collection: T1-weighted high-resolution anatomical MRI scans were obtained using a 3.0T MRI scanner, with an echo time of 31 milliseconds, a flip angle of 46°, and a repetition time of 45 milliseconds. The shape of each magnetic resonance volume was 512x512x192, with a resolution of 0.49x0.49x0.8 millimeters. Additionally, there were corresponding helmet optical scan results for registering MEG and MRI. Among them, at least 3 clearly distinguishable anatomical landmark points were selected as references for evaluating the registration accuracy.

[0034] Hardware settings: The experiment used an Intel Core i7-10700K CPU @ 3.80 GH processor and 16 GB of RAM.

[0035] Comparison experiment: Manual registration: Three experienced operators manually selected corresponding feature points for registration. The average time required for each operator to complete one registration was recorded, and the FLE after registration was calculated.

[0036] ICP standard algorithm: The source point cloud (MEG helmet) and the target point cloud (scalp surface segmented from MRI) were directly input into the standard point-to-point ICP algorithm. The initial transformation matrix was set to the identity matrix (i.e., assuming no initial offset). The specific point cloud preprocessing (such as fine denoising and multi-step downsampling strategies) or coarse registration steps in this invention were not adopted. Among them, the parameters of ICP were kept consistent with those in the ICP stage of the method of this invention to ensure fairness.

[0037] The method of this invention: It was strictly implemented according to the two-stage automatic registration process described in this patent specification, including point cloud generation and preprocessing (uniform sampling, denoising, voxel downsampling), coarse registration based on RANSAC and FPFH features, and fine registration based on ICP.

[0038] Among them, the parameters were: Number of helmet point clouds (after sampling): 6000 points Number of MRI point clouds (after sampling): 8000 points RANSAC distance threshold: Initially 4.0 mm ICP distance threshold: 0.5 times the voxel size, maximum number of iterations: 2000 times.

[0039] The following evaluation metrics were used: After registration is completed, calculate the average of the three-dimensional Euclidean distances between the anatomical landmark points on the source data (MEG helmet) after the registration transformation and the corresponding anatomical landmark points on the target data (MRI). This metric directly reflects the accuracy of the registration on key anatomical structures. The calculation formula is: ; where Nf is the number of anatomical landmark points (Nf ≥ 3 in this experiment), landmarkMEG,i is the coordinate of the i-th landmark point on the MEG helmet, landmarkMRI,i is the coordinate of the corresponding i-th landmark point on the MRI, and R and T are the rotation matrix and translation vector obtained from the registration, respectively.

[0040] Registration time: For the automated registration method, it refers to the computing time consumed from loading the preprocessed point cloud data until the algorithm converges and outputs the final transformation matrix. For manual registration, it refers to the average operation time required for the operator to complete the entire registration process.

[0041] Success rate: For the automated registration algorithm, if the FLE of its registration result is less than a preset clinically acceptable threshold (set to 5.0 mm in this experiment), then this registration is considered successful. The success rate is defined as the percentage of the number of successful registrations to the total number of trials. This metric mainly evaluates the robustness of the algorithm and its stability on different data. For manual registration, since the operator will continuously adjust until satisfied, the success rate is usually defaulted to 100%, but its accuracy and time consumption are the main considerations.

[0042] Performance comparison of different registration methods (mean ± standard deviation):

[0043] The method of the present invention achieved the lowest FLE (1.3 ± 0.3 mm), indicating the highest registration accuracy. This result is not only better than the standard ICP algorithm (3.7 ± 1.8 mm), but even slightly better than the manual registration performed by experienced operators (1.5 ± 0.4 mm). Although the accuracy of manual registration is relatively high, its standard deviation also reflects a certain variability between and within human operations. Due to the lack of effective coarse registration guidance and robust handling of noise, the standard ICP algorithm is prone to falling into local optima in some cases where the initial pose deviation is large or the data quality is slightly poor, resulting in a relatively high average FLE and a large standard deviation. The method of the present invention can effectively guide the point cloud to an initial position close to the global optimum through the coarse registration step of RANSAC combined with FPFH features, and then fine-tuned by ICP, thus significantly improving the final registration accuracy and consistency. The accuracy of 1.3 mm is already better than the clinically acceptable error threshold of 2 mm often mentioned in the literature, which is crucial for subsequent high-precision brain functional source localization.

[0044] In terms of registration efficiency, the method of the present invention (75±15s) is much faster than manual registration (950±120s), and the time is shortened by more than an order of magnitude. Although the time consumption of the method of the present invention is slightly higher than that of the standard ICP algorithm (45±10s) due to the addition of point cloud preprocessing and coarse registration steps, this moderate increase in time is exchanged for a significant improvement in accuracy and robustness, and the cost-effectiveness is extremely high. For clinical and scientific research applications, an automated registration time of 1-2 minutes is completely acceptable and significantly improves work efficiency.

[0045] The proposed method achieved successful registration (FLE < 5.0 mm) in all 30 tests, with a success rate of 100%. In contrast, the success rate of the standard ICP algorithm was only 70%. In the other 30% (9 / 30) of the cases, its registration error exceeded the acceptable range, mainly due to its sensitivity to the initial point cloud pose. This fully demonstrates the superior robustness of the two-stage strategy of the proposed method (especially RANSAC coarse registration) when processing data from different individuals and different initial conditions. Although the operator will ensure that the final result of manual registration "looks correct", the process relies on subjective judgment and is time-consuming.

[0046] In summary, through comparative experimental analysis of real data from 30 volunteers, the "two-stage automatic registration method for magnetoencephalogram helmet and brain MRI data" proposed in the present invention shows significant comprehensive advantages in terms of registration accuracy, registration time and robustness. This method not only achieves high-precision automatic registration, effectively avoids interference from human factors, ensures the consistency and reliability of the results, but also greatly improves processing efficiency. These characteristics enable it to provide solid and reliable technical support for the precise fusion of magnetoencephalogram (MEG) and MRI data and subsequent brain function research and clinical diagnosis of neurological diseases, fully demonstrating the practicality and advancement of the present invention.

[0047] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will 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 two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data, characterized in that, It includes the following steps: Obtain magnetoencephalography measurement helmet and brain MRI data, and store the magnetoencephalography measurement helmet and brain MRI data; Generate a point cloud based on the three-dimensional models of the magnetoencephalography measurement helmet and MRI data; Apply the random sample consensus algorithm to perform preliminary registration on the generated point cloud; Use the iterative closest point algorithm to perform further fine alignment on the point cloud after completing the de-icing configuration; Output the final transformation matrix according to the alignment result.

2. The two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data according to claim 1, wherein The acquisition operation process of the magnetoencephalography measurement helmet and brain MRI data is as follows: Load model files: Load magnetoencephalography measurement helmet and MRI model files from the specified path, and the file format includes STL; Check file validity: Confirm that each model file contains readable three-dimensional data; Verify data format: Load the STL file in the three-dimensional grid; Among them, in the stage of checking file validity, when an unreadable model file is identified, the model file is discarded, and a check message is generated synchronously. The content of the check message includes the success probability of reading the model file and the name of the model file for which the discard operation is performed.

3. A two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data according to claim 1, characterized in that, The three-dimensional models of the magnetoencephalography measurement helmet and MRI data applied in the point cloud generation stage are derived from the magnetoencephalography measurement helmet and MRI model files loaded by opening through the three-dimensional grid; Steps for generating a point cloud based on the three-dimensional models of the magnetoencephalography measurement helmet and MRI data: Convert the three-dimensional grid to a point cloud: Generate a point cloud from the three-dimensional surface of the helmet and MRI models through a uniform sampling algorithm, so that each three-dimensional grid area is sampled into several points; Point cloud quantity configuration: Specify the number of point clouds generated for each three-dimensional model according to the parameters in the configuration file; Point cloud preprocessing: Perform denoising processing on the generated point cloud data, remove isolated noise points, and use voxel downsampling technology to reduce the quantity of point cloud data to two-thirds of the original point cloud data.

4. A two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data according to claim 3, characterized in that The denoising processing logic of the point cloud data is as follows: Let the point cloud data set be , where , is the three-dimensional coordinates of the i-th point. For each point in the point cloud, calculate its average distance to the k nearest neighbor points ; , is the j-th nearest neighbor point of ; is the Euclidean distance; Calculate the mean of the average distances of all points and the standard deviation : Among them, a threshold T is set by the client, and by default, , where m is a constant. Points with an average distance greater than the threshold T are determined as outliers and removed.

5. The two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data according to claim 1, characterized in that, The random sample consensus algorithm obtains the maximum consistent point pairs by randomly selecting sample points from the point cloud and performing model fitting.

6. The two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data according to claim 1, wherein The preliminary registration steps of the point cloud include: Feature calculation: Calculate the local feature descriptors of the magnetoencephalography measurement helmet point cloud and the MRI point cloud; Coarse alignment: Use the random sample consensus algorithm to perform coarse alignment based on the extracted feature descriptors, and determine the preliminary transformation matrix by calculating the matching point pairs between the point clouds. Distance threshold: In the application stage of the random sample consensus algorithm, a distance threshold is set synchronously to determine which point pairs can be regarded as matching points and control the accuracy of preliminary registration; Among them, the feature calculation result calculates the features of each point cloud through the FPFH algorithm, and the distance threshold is user-defined, with an initial default setting of 4.0 mm.

7. A two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data according to claim 1, characterized in that, The fine alignment operation of the point cloud after completing preliminary registration is as follows: Fine alignment: Based on the preliminary registration result obtained by the random sample consensus algorithm as the initial transformation matrix of the ICP algorithm, perform fine alignment; Distance threshold: Set a smaller distance threshold during the ICP algorithm process; Maximum number of iterations: Set the maximum number of iterations until the algorithm converges; Among them, the ICP algorithm adjusts the position of the point cloud iteratively. In the distance threshold setting stage, the default setting is 0.5 times the voxel size, and the maximum number of iterations is set to 2000 times.

8. A two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data according to claim 1, characterized in that, After the two-stage registration of the point cloud through the random sample consensus algorithm and the iterative closest point algorithm, the obtained final transformation matrix contains the rotation and translation information between the magnetoencephalography measurement helmet and the MRI point cloud.

9. A two-stage automatic registration method for a magnetoencephalography measurement helmet and brain MRI data according to claim 1, characterized in that The upper left 3x3 part of the final transformation matrix represents the rotation relationship, and the translation vector is located in the last column of the matrix, representing the translation relationship, that is, the translation information. In the final transformation matrix output stage, the final transformation matrix is saved to an XML file; Among them, in the output stage of the final transformation matrix, a three-dimensional visualization tool is used to check the registration effect and verify whether the magnetoencephalography measurement helmet and the MRI point cloud are accurately aligned. When the inspection result is yes, the output of the final transformation matrix is further executed. Otherwise, the obtained magnetoencephalography measurement helmet and brain MRI data are reset and executed.

Images