A method for local sensor array design for magnetoencephalography source localization

Abstract

The application provides a local sensor array design method for brain magnetic source positioning. The method comprises the following steps: obtaining a scalp surface and a cortex surface of a subject based on a head nuclear magnetic resonance image, obtaining a grid of the scalp surface by using a three-dimensional reconstruction technology, and further generating a candidate sensor array; selecting a dipole source of a brain functional area of interest based on a Brodmann partition system, and calculating array sensitivity corresponding to the selected dipole source by using a Frobenius norm; determining an initial position of the local sensor array based on the array sensitivity, and determining the volume of the sensor and the distance between adjacent sensors based on a size radius and a spacing radius. The local sensor array design method realized by the application is simple and efficient, has strong practicability and universality, and can obtain more accurate source estimation under the condition that the number of sensors is limited.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of biomedical engineering, and particularly relates to a local sensor array design method for brain magnetic source positioning. BACKGROUND

[0002] Magnetoencephalography (MEG) is a non-invasive functional imaging technique with important clinical applications. Currently, MEG is widely used with superconducting quantum interference devices (SQUIDs), which can measure the weak magnetic field generated by the current in the brain. However, the SQUID-MEG sensor array must be fixed in a sealed liquid helium to achieve good thermal insulation, which limits its application.

[0003] Unlike SQUID-MEG, MEG based on Spin-Exchange Relaxation-Free (SERF) atomic magnetometers can be placed closer to the scalp to obtain higher signal strength and enhanced spatial resolution. Moreover, SERF-MEG has flexible sensor configuration, which means that it can better match the head size of different participants (especially children) to ensure sensor coverage.

[0004] With limited number of sensors, SERF-MEG can be designed as a locally arranged sensor array (LASA) for specific brain regions. Compared with traditional SQUID-MEG arrays, LASA for specific brain regions can maintain the spatial resolution and sensitivity of the cortex of interest and achieve source positioning accuracy. Currently, researchers tend to place a limited number of sensors in the brain region of interest according to the brain anatomy to obtain brain magnetic field signals. In addition, methods have been proposed to optimize LASA based on different indicators, however, these methods are mostly in the simulation stage and do not consider the actual sensor size when optimizing, and cannot be applied to the construction of SERF-MEG systems. SUMMARY

[0005] The technical problem to be solved by the application is the design of a local sensor array for a limited number of SERF atomic magnetometers. Current methods do not consider sensor size and cannot be applied to actual SERF-MEG systems. To overcome the shortcomings of existing methods, the application provides a local sensor array design method for brain magnetic source positioning, which is simple and efficient, has strong universality and practicality, and helps to improve the accuracy of brain magnetic source positioning.

[0006] The local sensor array design method of the present application comprises the following steps:

[0007] Step 1: segmenting the brain of the individual head nuclear magnetic resonance image obtained by scanning, obtaining the head skin surface grid and the cortex surface grid by using three-dimensional reconstruction technology, and generating the candidate sensor array by using the head skin surface grid and the cortex surface grid;

[0008] Step 2: segmenting the cortex surface into 104 brain functional areas based on the Brodmann atlas, randomly selecting x dipole sources in the brain functional area of interest, and calculating the array sensitivity S corresponding to the selected dipole source by using the Frobenius norm;

[0009] Step 3: determining the initial position N1 of the sensor based on the array sensitivity S, restricting the volume of the sensor and the distance between adjacent sensors by using the size radius r and the spacing radius R, determining the position of each sensor in turn based on the user-set sensor number N, and generating the local sensor array.

[0010] Step 4: projecting the position information of the simulation-generated local sensor array to the real head skin surface of the subject, so as to apply it to the construction of the SERF-MEG system.

[0011] Further, the step 1 comprises:

[0012] Based on the nuclear magnetic resonance instrument, the head of the subject is scanned to obtain T1w-MRI data; the MRI data is segmented by using the FreeSurfer software to obtain the cortex surface C1 and the head skin surface S1; in order to ensure the practicability of the candidate sensor array, the head skin surface S1 is cropped, the cropping method is to delete all vertices below the plane determined by the lowest point of the cortex surface, and manually delete the vertices covering the ears and eyes of the subject; the topology connection of the cropped discrete point cloud is reconstructed by using the Delaunay triangulation algorithm to obtain a new head skin surface S2; the n vertices of the head skin surface S2 are used as the candidate sensor positions to further generate the candidate sensor array.

[0013] Further, the step 2 comprises:

[0014] The cortex surface C1 of the step 1 is down-sampled into a triangular mesh with m vertices, and the cortex surface C1 is further generated into a source model, and the m vertices are used as candidate dipole sources; the guide field matrix L of the candidate sensor array is calculated by using the single-shell head model and the source model; the cortex surface C1 is segmented into 104 brain functional areas by using the Brodmann atlas, x dipole sources are randomly selected in the brain functional area of interest, and the guide field matrix l of the corresponding candidate sensor array is selected; the array sensitivity S corresponding to the selected dipole source is calculated by using the Frobenius norm S={||l1||,||l2||,…,||ln ||}。

[0015] Further, the step 3 comprises:

[0016] Step 3.1: setting the number N of local sensor array channels;

[0017] Step 3.2: arranging the selected sensor array sensitivity S corresponding to the selected dipole source pair obtained in step 2 in descending order;

[0018] Step 3.3: selecting the sensor with the largest sensitivity value as the initial position N1 of the local sensor array;

[0019] Step 3.4: setting the length and width of the sensor base, and then calculating the size radius r and the spacing radius R;

[0020] Step 3.5: in the sphere 1 with the initial position N1 as the center and the size radius r as the radius, calculating the candidate sensor positions contained in the sphere 1 and deleting them;

[0021] Step 3.6: in the sphere 2 with the initial position N1 as the center and the spacing radius R as the radius, calculating the candidate sensor positions contained in the sphere 2 and deleting them;

[0022] Step 3.7: selecting the sensor with the largest sensitivity value from the remaining candidate sensor positions as the second position N2 of the local sensor array;

[0023] Step 3.8: judging whether the local sensor array position point reaches the preset channel number, if not, repeating steps 3.5-3.7, if the preset channel number is reached, stopping the loop, and generating the local sensor array with the last N position points.

[0024] Further, the local sensor array obtained in step 3.8 is matched with the scalp surface S1 of the subject by using the ICP algorithm to obtain a head model with sensor position points; the head model is printed according to the original scale by using the 3D printing technology, and the position points of the local sensor array are marked with a marker pen.

[0025] Further, a blank flexible helmet is installed on the 3D printed head model to obtain the corresponding sensor position points and mark them with stickers; the marked points on the flexible helmet are used as the center points of the sensor slots, and the slots are installed to use the simulated local sensor array to construct a real SERF-MEG system.

[0026] The advantages of the present application compared with the prior art method are:

[0027] The application provides a practical local sensor array design method which can be used for constructing a SERF-MEG system, the method is simple and efficient, and has strong universality, the size and number of sensors can be changed at will according to actual conditions, and the local sensor array designed by using the method can obtain more accurate source estimation under the condition that the number of sensors is limited. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The flow chart of the local sensor array design method in the application is shown in the figure.

[0029] Figure 2 The schematic diagram of the local sensor array design method in the application is shown in the figure. Figure 2 (a) is a candidate sensor array. Figure 2 (b) is the initial position of the local sensor array. Figure 2 (c) is the determination of the sensor size. Figure 2 (d) is the determination of the sensor spacing. Figure 2 (e) is the schematic diagram of the sensor size. Figure 2 (f) is the schematic diagram of the local sensor array.

[0030] Figure 3 The schematic diagram of the local sensor array projected onto the real scalp surface of the subject in the application is shown in the figure. Figure 3 (a) is a head model with local sensor array position points. Figure 3 (b) is a 3D printed head model with marker points. Figure 3 (c) is a flexible helmet with marker points. Figure 3 (d) is a flexible helmet with sensor slot. DETAILED DESCRIPTION

[0031] The application will be described in detail below in combination with the drawings and embodiments, and it should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the application.

[0032] Before introducing the specific operation steps, first briefly introduce the relationship between the brain magnetic source positioning error and the array sensitivity:

[0033] Suppose that the brain activity source is equivalent to the current dipole distributed on the cortical surface, and the measured MEG data is described by a general linear model equation:

[0034] B = Lj + e (1)

[0035] Where, B ∈ R n*t represents the magnetic field amplitude at n sensor channels; j ∈ R m*t is the current dipole amplitude at the dipole position m; L ∈ R n*m is the lead field matrix; e is the additional random noise of each sensor.

[0036] The source position is estimated using a linearly constrained minimum variance beamformer:

[0037]

[0038] Here, ||L|| is the Frobenius norm of the steering matrix L, called the array sensitivity.

[0039] The total error in the beamformer reconstruction process is represented as the sum of the square of the reconstruction time point error:

[0040]

[0041] Equation (3) can be simplified as:

[0042]

[0043] Where v is the standard deviation of the noise at each sensor.

[0044] As can be seen from equation (4), the higher the sensor array sensitivity, the smaller the overall error of the source reconstruction. Therefore, in the local sensor array design process, we consider the array sensitivity as an optimization index, and increase the physical size of the SERF magnetometer as a constraint condition for sequentially selecting sensor positions.

[0045] The application discloses a local sensor array design method for brain magnetic source positioning. The specific steps are as follows:

[0046] Step 1: Segment the brain region of the individual head nuclear magnetic resonance image obtained by scanning, obtain the head skin surface grid and the cortex surface grid by using three-dimensional reconstruction technology, and generate a candidate sensor array by using the head skin surface grid and the cortex surface grid, and the specific implementation is as follows:

[0047] Scan the head of the subject based on the nuclear magnetic resonance instrument to obtain T1w-MRI image data;

[0048] Segment the MRI data by using FreeSurfer software to obtain the cortex surface C1 and the head skin surface S1;

[0049] Cut the head skin surface S1, and the cutting method is to delete all vertices below the plane determined by the lowest point of the cortex surface, and manually delete the vertices covering the ears and eyes of the subject;

[0050] Reconstruct the topological connection of the discrete point cloud after cutting by using the Delaunay triangulation algorithm to obtain a new head skin surface S2;

[0051] The n vertices of the head skin surface S2 are regarded as candidate sensor positions, and a candidate sensor array is further generated.

[0052] Step 2: based on the Brodmann atlas, the cortical surface is segmented into 104 brain functional areas, x dipole sources are randomly selected in the brain functional area of interest, and the array sensitivity S corresponding to the selected dipole sources is calculated by using the Frobenius norm, and the specific implementation is as follows:

[0053] The cortical surface C1 in step 1 is down-sampled into a triangular mesh with m vertices, and a source model is further generated from the cortical surface C1, wherein the m vertices are regarded as candidate dipole sources;

[0054] The lead field matrix L of the candidate sensor array is calculated by using the single-shell head model and the source model, wherein L∈R n*m , n is the number of channels of the candidate sensor array, and m is the number of dipole sources of the source model;

[0055] The cortical surface C1 is segmented into 104 brain functional areas by using the Brodmann atlas;

[0056] x dipole sources and the lead field matrix l of the candidate sensor array corresponding to the x dipole sources are randomly selected in the brain functional area of interest, wherein l∈R n*x , n is the number of channels of the candidate sensor array, and x is the number of dipole sources in the brain functional area of interest;

[0057] The array sensitivity S corresponding to the selected dipole sources is calculated, S={||l1||,||l2||,…,||l n ||, wherein ||l n || represents the sensitivity of the nth sensor to the selected x dipole sources.

[0058] Step 3: based on the array sensitivity S, the initial position N1 of the sensor is determined, the volume of the sensor and the distance between adjacent sensors are restricted by using the size radius r and the spacing radius R, the position of each sensor is determined in turn based on the number of sensors N set by the user, and a local sensor array is generated. Figure 1 and Figure 2 are the flow chart and schematic diagram of the local sensor array design method of the application respectively, and the specific process of the algorithm is as follows:

[0059] Step 3.1: set the number of channels N of the local sensor array;

[0060] Step 3.2: arrange the selected dipole source corresponding to the selected sensor array sensitivity S obtained in step 2 in descending order;

[0061] Step 3.3: select the sensor with the largest sensitivity value as the initial position N1 of the local sensor array;

[0062] Step 3.4: Set the length and width of the sensor base, and then calculate the size radius r and the spacing radius R;

[0063] Step 3.5: In the sphere 1 with the initial position N1 as the center and the size radius r as the radius, calculate the candidate sensor positions contained in the sphere 1, and delete them;

[0064] Step 3.6: In the sphere 2 with the initial position N1 as the center and the spacing radius R as the radius, calculate the candidate sensor positions contained in the sphere 2, and delete them;

[0065] Step 3.7: Select the sensor with the maximum sensitivity value from the remaining candidate sensor positions as the second position N2 of the local sensor array;

[0066] Step 3.8: Determine whether the local sensor array position point reaches the preset channel number, if not, repeat steps 3.5-3.7, if the preset channel number is reached, stop the loop, and generate the local sensor array with the last N position points.

[0067] Step 4: Figure 3 The schematic diagram of the local sensor array projected onto the real scalp surface of the subject in the present application is as follows:

[0068] The ICP algorithm is used to match the local sensor array obtained in step 3.8 with the scalp surface S1 of the subject, to obtain a head model with sensor position points; the 3D printing technology is used to print the head model according to the original scale, and the local sensor array position points are marked with a marker pen; the blank flexible helmet is installed on the 3D printed head model, the corresponding sensor position points are obtained, and the marker points on the flexible helmet are marked with stickers; the marker points on the flexible helmet are used as the center points of the sensor slots, and the slots are installed, so that the simulated local sensor array can be used to construct a real SERF-MEG system.

[0069] The contents not described in detail in the specification of the present application belong to the prior art known to those skilled in the art.

[0070] The above embodiments are provided only for the purpose of describing the present application, and are not intended to limit the scope of the present application. The scope of the present application is defined by the appended claims. Various equivalent replacements and modifications made without departing from the spirit and principles of the present application should be covered within the scope of the present application.

Claims

1. A method for designing a local sensor array for locating magnetoencephalography (MEG) sources, characterized in that, Includes the following steps: Step 1: Segment the brain regions of the individual head MRI images obtained from the scan, use 3D reconstruction technology to obtain scalp surface mesh and cortical surface mesh, and use the scalp surface mesh and cortical surface mesh to generate a candidate sensor array; Step 2: Based on the Brodman map, the cortical surface was divided into 104 brain functional areas, and brain functional areas of interest were randomly selected. A number of dipole sources are used, and the array sensitivity corresponding to the selected dipole source is calculated using the Frobenius norm. ; Step 3: Based on array sensitivity Determine the initial position N1 of the sensor, constrain the volume of the sensor and the distance between adjacent sensors using the size radius r and the spacing radius R, determine the position of each sensor sequentially based on the number of sensors N set by the user, and generate a local sensor array; Step 4: Project the simulated local sensor array position information onto the subject's real scalp surface to apply it to the construction of the SERF-MEG system.

2. The local sensor array design method according to claim 1, characterized in that, Step 1 includes: T1w-MRI data was obtained by scanning the subject's head using an MRI scanner. The MRI data was segmented using FreeSurfer software to obtain the cortical surface C1 and the scalp surface S1. To ensure the practicality of the candidate sensor array, the scalp surface S1 was cropped by deleting all vertices below the plane determined by the lowest point of the cortical surface and manually deleting vertices covering the subject's ears and eyes. The topological connections of the cropped discrete point cloud were reconstructed using the Delaunay triangulation algorithm to obtain a new scalp surface S2. The n vertices of the scalp surface S2 were used as candidate sensor locations to further generate a candidate sensor array.

3. The local sensor array design method according to claim 1, characterized in that, Step 2 includes: The cortical surface C1 in step 1 is downsampled into a triangular mesh with m vertices. A source model is then generated from the cortical surface C1, with the m vertices used as candidate dipole sources. The guiding field matrix of the candidate sensor array is calculated using the single-shell head model and the source model. Using the Broadman atlas, the C1 cortical surface was segmented into 104 brain functional areas, and brain functional areas of interest were randomly selected. Each dipole source and its corresponding candidate sensor array's guiding field matrix The array sensitivity corresponding to the selected dipole source was calculated using the Frobenius norm. .

4. The local sensor array design method according to claim 1, characterized in that, Step 3 includes: Step 3.1: Set the number of channels N in the local sensor array; Step 3.2: Calculate the sensitivity of the candidate sensor array corresponding to the selected dipole source obtained in Step 2. Arranged in descending order; Step 3.3: Select the sensor with the highest sensitivity value as the initial position N1 of the local sensor array; Step 3.4: Set the length and width of the sensor base, and then calculate the dimensional radius r and the spacing radius R; Step 3.5: Within sphere 1, which has an initial position N1 as its center and a radius r as its size, calculate the positions of the candidate sensors contained in sphere 1 and delete them; Step 3.6: In sphere 2 with initial position N1 as the center and spacing radius R as the radius, calculate the positions of candidate sensors contained in sphere 2 and delete them; Step 3.7: Select the sensor with the highest sensitivity value from the remaining candidate sensor positions as the second position N2 of the local sensor array; Step 3.8: Determine whether the number of local sensor array positions has reached the preset number of channels. If not, repeat steps 3.5-3.

7. If the preset number of channels has been reached, stop the loop and generate a local sensor array from the final N positions.

5. The local sensor array design method according to claim 4, characterized in that: The ICP algorithm is used to match the local sensor array obtained in step 3.8 with the scalp surface S1 of the subject to obtain a head model with sensor location points; the head model is printed according to the original scale using 3D printing technology, and the location points of the local sensor array are marked with a marker.

6. The local sensor array design method according to claim 5, characterized in that: A blank flexible helmet is mounted on a 3D printed head model to obtain the corresponding sensor location points, which are then marked with stickers. The marked points on the flexible helmet are used as the center points of the sensor slots. By installing the slots, the simulated local sensor array can be used to build a real SERF-MEG system.

Images