A three-dimensional multi-nuclear spectral scanning positioning method and system based on region of interest optimization

Abstract

The present application relates to a kind of three-dimensional multi-nuclear spectrum scanning positioning method and system based on region of interest optimization, method includes: obtaining high-resolution structural image data, one or more regions of interest are divided, and the corresponding mask information is obtained;Optimization is carried out based on the type of region of interest, and the optimal positioning grid information of magnetic resonance spectroscopy imaging is obtained in high-resolution anatomical image space by multi-scale search;T1 weighted structural image is obtained as the reference space of magnetic resonance spectroscopy imaging scan, based on the mask information, the magnetic resonance spectroscopy imaging positioning grid information is mapped to reference space;Based on the magnetic resonance spectroscopy imaging positioning grid information registered to reference space, magnetic resonance spectroscopy imaging scan is carried out, and the closed loop of scan-analysis-optimization-scan is used to realize multi-nuclear spectrum scanning automatic optimization positioning.The present application realizes the accurate alignment of region of interest and magnetic resonance spectroscopy voxel, real-time three-dimensional magnetic resonance spectrum optimal positioning and simplifies post-processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical imaging technology, in particular to a three-dimensional multi-nuclear spectrum scanning positioning method and system based on region of interest optimization. BACKGROUND

[0002] The traditional Magnetic resonance spectroscopic imaging (MRSI) positioning scheme acquires the whole brain anatomy by scanning T1-Weighted Imaging (T1WI), and based on this image, the angle and position of the MRSI acquisition grid are adjusted by visual observation to make the brain region of interest as much as possible to be located in certain voxels of the MRSI acquisition grid. This method highly depends on the experience and clinical knowledge of the operator, lacks a unified standard, and is prone to inaccurate positioning due to cognitive bias and the inherent low resolution of MRSI data. In the case of complex shape of the region of interest, both reliability and efficiency are challenged. Due to the large size of MRSI voxels, non-optimal positioning may result in the proportion of the region of interest in all voxels not reaching the maximum, which affects the quantitative accuracy of metabolites in the region of interest under the effect of partial volume effect.

[0003] In addition, existing open source software such as jMRUI, inspector and Tarquin, etc. have not effectively solved the fusion problem with T1WI when processing 3D ³¹P-MRSI data, and there is a lack of unified standard for the definition of spatial coordinates of MRSI voxels, so the voxel selection process in the post-processing stage is relatively complex. In addition, there is still a lack of effective scheme to identify the accuracy of selected voxels.

[0004] Chinese patent application publication No. CN119723021A provides a magnetic resonance spectrum automatic positioning method, device, equipment and medium, which realizes automatic positioning of magnetic resonance spectrum imaging by constructing a lesion recognition model and combining multi-directional scanning, solves the problem of inaccurate lesion positioning in the prior art, and improves the accuracy and efficiency of positioning. The application only automatically positions single-voxel magnetic resonance spectrum by identifying the lesion center, and cannot be applied to the automatic positioning of complex three-dimensional magnetic resonance spectrum involving multiple regions of interest. In addition, the lesion recognition model based on patient history images is used for automatic positioning of spectrum, which still has certain limitations in model generalization ability and application scope. SUMMARY

[0005] The purpose of the present application is to overcome the defects of the prior art and provide a three-dimensional multi-nuclear spectrum scanning positioning method and system based on region of interest optimization to solve or partially solve the problem of difficult real-time positioning grid information optimization in multi-nuclear spectrum scanning positioning.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] One aspect of the present invention provides a three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization, comprising the following steps:

[0008] Utilizing a multi-channel skull 1 H-coil acquires high-resolution structural image data, divides one or more regions of interest, and obtains the corresponding mask information;

[0009] Based on the type of region of interest, the magnetic resonance spectroscopy imaging localization grid is optimized, and the magnetic resonance spectroscopy imaging localization grid information of high-resolution anatomical image space is obtained through multi-scale search.

[0010] use 1 H / 31 The T1-weighted structural image obtained by the P-dual-tuned cranial coil is used as the reference space for magnetic resonance spectroscopy scanning. Based on the mask information, the magnetic resonance spectroscopy positioning grid information is registered to the reference space.

[0011] Magnetic resonance spectroscopy imaging is performed based on the positioning grid information mapped to the reference space. Automatic positioning of three-dimensional multi-nucleus spectroscopy scanning is achieved through a closed loop of scanning-analysis-optimization-scanning.

[0012] As a preferred technical solution, the process of obtaining magnetic resonance spectroscopy imaging localization grid information based on the type of region of interest includes the following steps:

[0013] For a single region of interest, optimization is performed with the goal of maximizing the coverage of the region of interest, under the constraints of preset translation and rotation angle ranges.

[0014] For multiple regions of interest, the importance of the disease is ranked and weighted based on each region of interest, with the goal of prioritizing the coverage of high-weight regions of interest;

[0015] For tumor boundaries, optimization is performed based on the probability distribution map of tumor regions, with the goal of maximizing the coverage of important tumor areas and minimizing the inclusion of healthy tissue.

[0016] As a preferred technical solution, the multi-scale search includes the following steps:

[0017] A coarse search is performed with a relatively large rotation step size and translation step size;

[0018] A fine search is performed with relatively small rotation and translation step sizes.

[0019] As a preferred technical scheme, the process of registering the magnetic resonance spectroscopy positioning grid information to the reference space comprises the following steps:

[0020] Based on the mask information, the high-resolution structural image is registered to the reference space by linear / nonlinear registration to obtain a mapping matrix, and the mapping matrix is used to register the magnetic resonance spectroscopy positioning grid information to the reference space; or

[0021] The preset standard space structure template is registered to the high-resolution structural image space by nonlinear registration to obtain a first mapping matrix, the high-resolution structural image is registered to the reference space to obtain a second mapping matrix, and the mask information of the standard space and the magnetic resonance spectroscopy positioning grid information are registered to the reference space by using the first mapping matrix and the second mapping matrix.

[0022] As a preferred technical scheme, in the process of optimizing the magnetic resonance spectroscopy positioning grid, the optimization target comprises:

[0023] The region of interest is covered as completely as possible by the least number of voxels in the magnetic resonance spectroscopy positioning grid;

[0024] The direction of the voxel in the magnetic resonance spectroscopy positioning grid is as consistent as possible with the region of interest.

[0025] As a preferred technical scheme, after the magnetic resonance spectroscopy scan based on the magnetic resonance spectroscopy positioning grid information mapped to the reference space, it further comprises:

[0026] Obtaining the optimized magnetic resonance spectroscopy scan positioning parameters and the spatial coordinate index of the voxels of interest, wherein the magnetic resonance spectroscopy scan positioning parameters comprise a translation amount, a rotation angle, and a voxel spatial coordinate of the region of interest.

[0027] As a preferred technical scheme, the mask information is obtained by registering the high-resolution structural image data with a preset standard space region of interest template, or the corresponding mask information is obtained by segmenting the region of interest of the high-resolution structural image data using a pre-trained 3D U-Net deep learning model.

[0028] As a preferred technical scheme, the high-resolution structural image data is any one of a T1-weighted image (T1WI), a T2-weighted image (T2W), a T2 fluid-attenuated inversion recovery sequence (T2FLAIR), a diffusion-weighted imaging (DWI), and a quantitative susceptibility mapping (QSM), and the magnetic resonance spectroscopy positioning grid information comprises a positioning grid translation amount, a rotation angle, and a voxel spatial coordinate information of a region of interest.

[0029] Another aspect of the present application provides a three-dimensional multi-nuclear spectrum scanning positioning system based on region of interest optimization, comprising a scanning subsystem and an analysis subsystem,

[0030] The scanning subsystem comprises:

[0031] A multi-nuclear magnetic resonance imaging scanner and coils, the coils comprising multi-channel head-neck 1 H coils and 1 H / 31 P dual-tuned coils, the multi-nuclear magnetic resonance imaging scanner being configured to acquire magnetic resonance data, the magnetic resonance data comprising high-resolution 1 H structural images obtained using multi-channel head-neck 1 H coils, and T1-weighted structural images as a reference space obtained using 1 H / 31 P dual-tuned coils, the multi-nuclear magnetic resonance imaging scanner being further configured to receive and configure the analysis subsystem optimized 31 P-magnetic resonance spectrum imaging scanning positioning parameters;

[0032] A console connected to the multi-nuclear magnetic resonance imaging scanner, configured to display the spatial position of the region of interest, grid coverage, and perform scanning and positioning operations;

[0033] The analysis subsystem comprises:

[0034] A server configured to implement the aforementioned three-dimensional multi-nuclear spectrum scanning positioning method based on region of interest optimization.

[0035] Compared with the prior art, the present application has at least one of the following beneficial effects:

[0036] (1) Achieving precise alignment of the region of interest and the magnetic resonance spectrum imaging voxel: To solve the problem of quantitative accuracy affected by experience and partial volume effect in traditional manual positioning, the present application adopts real-time anatomical analysis of the scanning end based on high-resolution structural image automatic segmentation of the region of interest, obtains optimal positioning grid information of the magnetic resonance spectrum imaging through multi-scale search, and realizes precise alignment of the region of interest and the MRSI voxel, significantly improving positioning accuracy and further improving metabolite quantitative accuracy and reducing analysis errors caused by positioning deviation.

[0037] (2) Simplifying the post-processing procedure: To solve the technical problem of complex voxel selection, lack of unified coordinate standard, and low efficiency in the post-processing stage, the present application directly outputs standardized region of interest voxel spatial coordinates through scanning end optimization and sends them to the scanning subsystem, writes them into the Digital Imaging and Communications in Medicine (DICOM) Tag, skips the traditional manual selection step, realizes seamless connection of scanning and post-processing, greatly simplifies the post-processing procedure, and improves overall operation efficiency.

[0038] (3) Wide range of applications: This invention adopts a multi-target optimization and scan-analysis linkage scheme, supports the localization optimization of multiple ROIs and custom lesions, and is compatible with ¹H-MRSI and other multidimensional magnetic resonance spectroscopy (such as... 13 C, 2 Imaging (such as H) expands the application scenarios of the technology and promotes the widespread application of magnetic resonance spectroscopy in clinical practice and scientific research. Attached Figure Description

[0039] Figure 1 This is a flowchart of the three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization in the embodiments;

[0040] Figure 2 This is a schematic diagram of a three-dimensional multi-kernel spectral scanning and positioning system based on region of interest optimization in the embodiment.

[0041] Figure 3 This is a schematic diagram of the electronic device in the embodiment. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0043] Example 1

[0044] To address the problems existing in the aforementioned prior art, this embodiment provides a three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization. This method dynamically adjusts the localization process through real-time image processing and optimization algorithms. 31 The localization parameters of P-MRSI scanning enable the scanning area to accurately cover the target anatomical structure, such as specific brain regions of interest, brain tumors, and areas of metabolic abnormalities, thereby improving the quantitative accuracy of phosphorus metabolism imaging.

[0045] See Figure 1 The method includes the following steps:

[0046] Step S1: Region of Interest (ROI) segmentation.

[0047] Preferably, automatic or semi-automatic segmentation algorithms are used to segment the Region of Interest (ROI). These algorithms may include traditional image processing techniques (such as thresholding, region growing, edge detection, morphological operations, etc.), atlas / template-based registration and segmentation, and machine learning and deep learning-based segmentation models (such as U-Net and its variants). The output will be an accurate 3D ROI mask for the individual subject, identifying the target anatomical structure or lesion region.

[0048] In one embodiment, the ROI is segmented in real time based on the individual's anatomical space.

[0049] Specifically, first, high-resolution structural image data (e.g., T1WI, T2WI, T2FLAIR, DWI, or QSM) is obtained by scanning the scanning subsystem. Then, in the analysis subsystem, the high-resolution structural image data (e.g., T1WI, T2WI, T2FLAIR, DWI, QSM, etc.) is input, and the ROI is segmented in real time using traditional image processing techniques or deep learning models such as pre-trained 3D U-Net, etc. The binary mask of the ROI is output. For a single ROI (e.g., unilateral substantia nigra, hippocampus, etc.), multiple ROIs (e.g., bilateral substantia nigra, basal ganglia, etc.) output binary masks, and for tumor regions, probability masks can be output. Binary mask and / or probability mask information is formed. This step uses a graphics processing unit (GPU) for acceleration, and the time required for a single segmentation is less than 15 seconds to meet the real-time processing requirements of clinical scanning.

[0050] In another embodiment, the standard space ROI template is registered to the individual's space.

[0051] Specifically, standard space templates such as ICBM152 template, AAL90 template, Hammersmith template, SUIT cerebellum template, etc. are built into the analysis system. The labels of the brain regions of interest in the built-in templates are selected as the binary masks of the ROIs, and the mask information is formed.

[0052] Preferably, after receiving the high-resolution structural image data obtained by the scanning subsystem, the necessary format conversion, resampling, registration, and other preprocessing steps are also included.

[0053] Step S2, magnetic resonance spectroscopy imaging (MRSI) positioning grid optimization.

[0054] According to 31 P-MRSI sequence parameters, field of view (FOV), "matrix" virtual 3D MRSI positioning grid, and ROI spatial mask and corresponding high-resolution anatomical image data. For different ROIs, the following optimization strategies are used:

[0055] (1) Single ROI (e.g., unilateral substantia nigra).

[0056] The optimization goal is to maximize the coverage of the target region, and the constraint condition is to prevent excessive tilt by setting the range of translation and rotation angle;

[0057] (2) Multiple ROIs (e.g., basal ganglia).

[0058] The method of dynamic weighting is used to solve the priority conflict problem of multiple ROIs to ensure the coverage of high weight ROIs when it is difficult to cover perfectly. The multiple ROIs are sorted and given weights based on their contribution to the disease, for example, for Parkinson's patients, the substantia nigra is given the highest weight as the main lesion brain area, the putamen is given a medium weight as the secondary metabolic change area, and the globus pallidus is given a lower weight as the background reference area. When optimizing, the weighted centroid alignment is used to realize the overall translation positioning, and then the rotation angle is further optimized, so as to realize the optimal configuration of the positioning grid under the cooperation of multiple ROIs.

[0059] (3) tumor boundary.

[0060] A differentiated processing mechanism is used to maximize the coverage of the tumor important area by the Voxel Of Interest (VOI) and minimize the inclusion of healthy tissue as the optimization goal. Based on the probability distribution map of the tumor area, a coverage reward term is set to preferentially cover the high-probability area of the tumor, and a healthy tissue penalty term is set to constrain the inclusion tendency of the area. In the edema or low-probability infiltration area, a penalty attenuation strategy is adopted to reduce the constraint strength of the area, and the positioning box is allowed to moderately cover the potential suspicious area. When optimizing, the tumor centroid is used as the initial VOI, and different VOI parameters are iteratively evaluated under the constraint condition to maximize the objective function.

[0061] In the optimization process, GPU parallel computing is used to improve the operation speed, and a multi-scale search mode combining coarse search (rotation step 5°, translation step 5mm) with fine search (step reduced to 1° and 1mm) is used to speed up the optimization efficiency. Through the optimization strategy, the high-resolution anatomical image space 31 The MRSI optimal positioning grid translation [Δx, Δy, Δz], rotation angle [θ x , θ y , θ z ] of P-MRSI positioning, and ROI voxel space coordinates are obtained as the high-resolution magnetic resonance spectroscopy positioning grid information in the T1WI space.

[0062] In one embodiment, step S2 includes the following sub-steps:

[0063] 1. Input the mask information of the ROI generated in step S1, and the T1W positioning reference image (consistent with the P-MRSI space) acquired by the scanning subsystem through the dual-tuned coil to obtain the preset MRSI scanning constraints (such as voxel size, sequence FOV, matrix size, etc.). 31 P-MRSI space).

[0064] 2. The process is to run an optimization algorithm. The goal of the algorithm is to automatically adjust the positioning parameters (translation and rotation angles) of the 3D MRSI scan according to the spatial distribution of the ROI, so that the ROI region is covered as completely as possible by the least number of voxels in the MRSI positioning grid.

[0065] The specific optimization method may be based on the corresponding objective function for different ROI cases (such as maximizing ROI coverage, minimizing non-ROI volume, etc.).

[0066] 3. Output the optimized 3D 31 P-MRSI scan positioning parameters as the positioning grid information of the region of interest-based magnetic resonance spectroscopy, including: translation [Δx, Δy, Δz], rotation angles [θ x , θ y , θ z ], ROI voxel spatial coordinates, etc.

[0067] 4. Parameter transmission. The output parameters include: the optimized positioning parameters are sent back to the quality control computer of the scanning subsystem through the data transmission protocol, and are applied to the 3D 31 P-MRSI scan sequence; the spatial coordinates of the voxels of interest are used for spectral post-processing analysis.

[0068] Step S2 is used to pre-calculate the best positioning scheme based on high-resolution T1WI before scanning, to achieve optimal region of interest-based magnetic resonance spectroscopy precise positioning.

[0069] Step S3, the positioning grid is transformed to the 3D 31 P-MRSI space.

[0070] It should be noted that while steps S1 and S2 are being performed, the scanning of the T1WI positioning reference image in step S3 can be performed.

[0071] Using 1 H / 31 P dual-tuned coil scanning to obtain T1WI (which can be performed synchronously with steps S1 and S2), as the 31 P-MRSI space reference, the scanning grid and related positioning parameters are mapped to the T1WI ( 31 P-MRSI space reference) space, and finally the 31 P-MRSI optimal positioning grid translation [Δx, Δy, Δz], rotation angles [θ 31 , θ x , θ y ] and ROI voxel spatial coordinates, etc. information of T1WI ( z P-MRSI reference space) space are obtained and transmitted back to the scanning subsystem, and T1WI ( 31The ROI mask of the ROI in the standard space is fused with the T1WI for the technician to check the spatial position.

[0072] In one embodiment, based on the binary mask obtained by real-time segmentation of the ROI of the individual anatomy, the high-resolution structural images (T1WI, T2WI, T2FLAIR, DWI, QSM, etc.) are linearly or nonlinearly registered to the T1WI positioning reference image (T1WI) 31 P-MRSI reference space), a mapping matrix is obtained, and the ROI mask and the MRSI positioning grid are registered to the T1WI space using the mapping matrix.

[0073] In another embodiment, the ROI binary mask of the brain region of interest in the standard space is obtained. Two-step registration is performed, first, a first mapping matrix is obtained by using a nonlinear registration method to register the standard space structural template to the individual high-resolution structural image space, and then the high-resolution structural image is registered to the T1WI positioning reference image (T1WI) 31 P-MRSI reference space), a second mapping matrix is obtained, and the standard space ROI mask and the MRSI positioning grid are registered to the T1WI positioning reference image (T1WI) 31 P-MRSI reference space).

[0074] Step S4: Transmit the positioning parameters to the scanning subsystem.

[0075] The transformed MRSI positioning grid information, including the grid translation [Δx, Δy, Δz], the rotation angles [θ x , θ y , θ z ], and the ROI voxel space coordinates, is transmitted back to the scanning subsystem and applied to 31 the positioning of P-MRSI.

[0076] The positioning parameters are automatically loaded in the MRI scanning subsystem, the scanning protocol is updated, the technician can visually confirm the ROI spatial position and grid coverage through the console, and the 31 P-MRSI scan is performed. After the scan, the spatial coordinates of the voxels of interest are written into the DICOM Tag, and finally the closed loop of scan-analysis-optimization-scan is realized to achieve accurate positioning of the ROI, solving the problems of difficult ROI positioning in the post-processing stage, poor alignment of the voxels of interest and the region of interest, and large quantitative errors.

[0077] Step S5: Positioning of voxels of interest in the post-processing end.

[0078] Specifically, this step includes the following sub-steps:

[0079] Step S501: Data preparation.

[0080] High resolution T1WI (HR-T1WI), obtained from multi-channel 1 H-Head coil, voxel ~1 mm x 1 mm x 1 mm.

[0081] T1W scout image (T1W scout) 31 P-MRSI reference space), obtained using single-channel dual-tuned coil and finally used for 31 localization of P-MRSI.

[0082] 3D 31 P-MRSI data, spatial resolution ~20 mm x 20 mm x 20 mm, DICOM format file.

[0083] Region of interest label, which can be either a binary mask NIFTI file generated from manual / automatic segmentation of individual space on high resolution T1 image or a NIFTI file of pre-defined region of interest in standard space atlas such as MNI-152.

[0084] Step S502, registration of region of interest label to T1W scout image (T1W scout) 31 P-MRSI reference space).

[0085] In one embodiment, the region of interest label is in standard MNI space.

[0086] MNI152 space label is mapped to individual HR-T1WI space using non-linear registration algorithm, then affine registration of HR-T1WI to T1W scout image (T1W scout) 31 P-MRSI reference space) is performed using FLIRT algorithm, concatenated registration transformation matrices are applied, and the region of interest label mask in 31 P-MRSI reference space) is output.

[0087] In another embodiment, the region of interest label is in individual space.

[0088] Affine registration of HR-T1WI to T1W scout image (T1W scout) 31 P-MRSI reference space) is performed, the affine matrix is applied to the registration of region of interest label to T1W scout image (T1W scout) 31 P-MRSI reference space), and the region of interest label mask in 31 P-MRSI reference space) is output.

[0089] Step S503, fusion of MRSI to T1W scout image (T1W scout) 31 P-MRSI reference space).

[0090] Firstly, the voxel grid origin coordinates, rotation matrix, and voxel size information (for example, 20mm x 20mm x 20mm) are extracted from the DICOM header file to construct the physical coordinate system of the MRSI voxel grid, and a positioning grid is generated.

[0091] Then, the MRSI voxel grid is projected onto the T1W positioning reference image 31 P-MRSI reference space, and a unique label value (such as voxel index n = 1, 2, 3, …) is assigned to each MRSI voxel, which satisfies the formula n = (z-1) x 100 + (y-1) x 10 + x, and a label synthesis image is generated.

[0092] Step S504, MRSI voxel positioning of the region of interest label.

[0093] Extract 31 All vertex coordinates of the P-MRSI reference space label synthesis image that overlap with the region of interest label, if the region of interest label spans multiple voxels, output all voxel index sets and all voxel masks.

[0094] Embodiment 2

[0095] Based on embodiment 1, the three-dimensional multi-nuclear spectrum scanning positioning system based on region of interest optimization is provided, which is used to realize the multi-nuclear spectrum scanning positioning method as described in embodiment 1, and aims to use high-resolution structural images to guide the optimization of 3D 31 P-MRSI scanning positioning in real time, so as to improve the imaging efficiency and accuracy for specific regions of interest.

[0096] Referring to Figure 2 , the system includes a scanning subsystem and an analysis subsystem connected to each other.

[0097] The scanning subsystem includes a multi-channel head and neck 1 H coil, 1 H / 31 P dual-tuned coil, which supports multi-nuclear (at least containing 1 H and 31 P) imaging MRI scanner for magnetic resonance signal generation, data acquisition and image data transmission to the analysis system. The subsystem communicates with the analysis subsystem bidirectionally through data transmission protocols (such as DICOM standard network transmission, private TCP / IP protocol, contribution storage, etc.), and transmits the acquired high-resolution multi-channel 1 H structural image and T1W structural image collected by the dual-tuned coil to the analysis subsystem, and receives the calculated optimized 31P-MRSI scan positioning parameters. The positioning parameters (MRSI optimal positioning grid translation [Δx, Δy, Δz], rotation angle [θx, θy, θz], ROI voxel space coordinates, etc.) are automatically updated within the scanning system. The technician performs a visual inspection, then performs the MRSI scan, and records the ROI corresponding spatial coordinates in the DICOM data for export.

[0098] The analysis subsystem includes a processing server and a visualization terminal, which are used to accept MRI outgoing data, perform ROI segmentation, positioning optimization algorithm, and other data analysis, and return the positioning parameters to the scanning subsystem.

[0099] The server includes hardware modules and software modules. The hardware module is a separate computing server, which is used to receive image data from the scanning subsystem, perform real-time image processing, analysis and optimization calculation, and feed back the results to the scanning system. The hardware requirements include: high-performance multi-core CPU for processing complex image segmentation and optimization algorithms; large-capacity high-speed RAM for fast loading and processing of three-dimensional MRI image data; GPU acceleration card for acceleration of segmentation model inference of deep learning; high-speed network interface card to ensure low delay of data transmission between the MRI scanning system; high-speed storage for fast reading and writing of image data and intermediate processing results; special software environment for installing software related to various medical image reading and format conversion, image preprocessing, ROI segmentation and optimization algorithms.

[0100] The software module is used to implement the multi-nuclear spectrum scanning positioning method of embodiment 1, and specifically includes the following steps:

[0101] (1) Image preprocessing, receiving images from the scanning system, and possibly performing necessary format conversion, resampling, registration and other preprocessing.

[0102] (2) ROI segmentation, by inputting high-resolution structural images, running automatic or semi-automatic segmentation algorithms, which may include traditional image processing techniques (such as thresholding, region growing, edge detection, morphological operations, etc.), atlas / template-based registration and segmentation, machine learning and deep learning-based segmentation models (such as U-Net and its variants). The output is an accurate subject spatial 3D ROI mask, which identifies the target anatomical structure or lesion area.

[0103] (3) Positioning optimization.

[0104] a. Input the ROI 3D mask generated in step (2), the T1W positioning reference image (spatially consistent with the final P-MRSI scan) acquired by the scanning subsystem through the dual-tuned coil, and the preset MRSI scan constraints (voxel size, sequence FOV, matrix size, etc.). 31

[0105] ​b. The processing procedure is to run an optimization algorithm. The goal of the algorithm is to automatically adjust the positioning parameters of the 3D MRSI scan according to the spatial distribution of the ROI, so as to meet the following conditions: the ROI region is covered as completely as possible by the least number of voxels in the MRSI positioning grid; the direction of the voxel is as close as possible to the ROI to improve the effective SNR and spatial resolution; the preset scan constraint condition is met; the specific optimization method may involve optimization based on the objective function for different ROI conditions (such as maximizing ROI coverage and minimizing non-ROI volume).

[0106] c. Output the optimized 3D 31 P-MRSI scan positioning parameters, including: positioning grid translation [Δx, Δy, Δz], rotation angle [θx, θy, θz], and ROI voxel spatial coordinates, etc.

[0107] d. Parameter transmission. The output parameters include: the optimized positioning parameters are sent back to the quality control computer of the scanning subsystem through a data transmission protocol, and are applied to the 3D 31 P-MRSI scan sequence; the spatial coordinate index of the voxel of interest is used for spectral post-processing analysis.

[0108] The present application will establish a full-automatic MRSI positioning optimization system based on the region of interest from the scanning end, to solve the problems of 3D MRSI scanning and post-processing voxel positioning, and to improve the accuracy of MRS metabolite quantification.

[0109] Traditional MRSI positioning belongs to an artificial experience-driven operation process, while the present application innovatively realizes the paradigm shift from post-selection to pre-optimization through scanning control by moving the anatomical analysis forward to the scanning end. Specifically, the spatial resolution difference between 3D 31 P-MRSI (20mm×20mm×20mm) and T1WI (1mm×1mm×1mm) is huge, and it is difficult to accurately position the region of interest and the 3D MRSI grid through manual positioning, especially when facing individual-specific regions of interest (such as tumors) or multiple regions of interest, immediate optimization of the scan positioning is urgently needed. In addition, based on a unified positioning standard, on the one hand, manual selection of voxels of interest is needed during data post-processing, which may not be accurate due to large positioning deviation, thereby causing quantification error; on the other hand, since the alignment of the voxel of interest and the region of interest may not be optimal, it will also affect the accuracy of quantitative analysis due to partial volume effect.

[0110] Therefore, the present application solves this series of problems from the scanning end, builds a scanning end adaptive positioning optimization system, uses the real-time obtained subject structure image information to perform interested region segmentation, automatically calculates the optimal scanning grid positioning through an optimization algorithm, ensures that the interested region occupies the maximum proportion in the related voxels, and simultaneously outputs the interested voxel space coordinates to facilitate MRS post-processing analysis.

[0111] The present application innovatively calculates the interested region positioning in real time in advance at the scanning end, combines the core design of a multi-target real-time optimization algorithm, and has breakthroughs in technical performance, operation efficiency and economic benefits, solves the problems of 31 A core bottleneck of P-MRSI technology clinical popularization, makes MRSI evolve from an experience-dependent technology into a precision tool, lays a foundation for clinical popularization of metabolic imaging, and realizes breakthroughs in precise positioning, efficient operation and wide compatibility. The following technical advantages are achieved:

[0112] (1) High accuracy. The anatomical and segmentation are moved to the scanning end, which fundamentally solves the problems of inaccurate positioning, suboptimal alignment and quantitative errors caused by manual positioning and post-selection of interested voxels. Through a multi-target real-time positioning optimization algorithm, a real-time linkage system of scanning end structure image data transmission-ROI segmentation-positioning image transmission-optimization parameter feedback-MRSI positioning is formed, the maximum interested coverage rate in MRSI voxels is realized, and the problems of precise positioning and quantification of multiple ROIs, complex anatomical structures and individualized ROIs are solved.

[0113] (2) Fast speed. The whole process of MRSI scanning positioning is realized automatically, the scanning process is optimized, and the scanning time is saved. In addition, the system outputs the standard space coordinates of the interested voxels, skips the step of manually selecting the interested voxels, and simplifies the MRS post-processing steps.

[0114] (3) Strong universality. Wide multi-dimensional magnetic resonance spectroscopy data applicability. The core positioning optimization method is not limited to 31 P-MRSI, and can be extended to 1H-MRSI and other multi-dimensional magnetic resonance spectroscopy data (such as 13 C, 2 H, etc.), promotes the clinical popularization process of magnetic resonance spectroscopy, and is suitable for optimal positioning of multiple ROI scenes.

[0115] (4) Real-time anatomical analysis at the scanning end. In the MRSI scanning preparation stage, the ROI is automatically segmented based on the real-time obtained T1W high-resolution structure image (or other structure image), and the scanning grid is optimized according to the ROI, which overturns the traditional “scan first and then register” process, and can ensure the optimal alignment of the interested voxels and the interested region in MRSI. The time sequence control logic of anatomical analysis moved to the scanning end needs to be protected.

[0116] Embodiment 3

[0117] On the basis of the foregoing embodiments, the present embodiment provides an electronic device, comprising: one or more processors and a memory, the memory having stored therein one or more programs, the one or more programs including instructions for performing the three-dimensional multi-nuclear spectrum scanning positioning method based on a region of interest optimization as described in Embodiment 1.

[0118] As Figure 3 described, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and of course can also include other hardware required by the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs to implement the method described above. Figure 1 Of course, in addition to the software implementation, the present application does not exclude other implementation manners, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or a logic device.

[0119] The memory can include non-permanent memory in a computer readable medium, random access memory (RAM) and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer readable medium.

[0120] The computer readable medium includes permanent and non-permanent, removable and non-removable media, which can be implemented by any method or technology to store information. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic disk storage or other magnetic storage device, or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition herein, computer readable medium does not include transitory computer readable media such as modulated data signals and carriers.

[0121] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements shall be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization, characterized in that, Includes the following steps: Utilizing a multi-channel skull 1 H-coil acquires high-resolution structural image data, divides one or more regions of interest, and obtains the corresponding mask information; Based on the type of region of interest, the magnetic resonance spectroscopy imaging localization grid is optimized, and the magnetic resonance spectroscopy imaging localization grid information of high-resolution anatomical image space is obtained through multi-scale search. use 1 H / 31 The T1-weighted structural image obtained by the P-dual-tuned cranial coil is used as the reference space for magnetic resonance spectroscopy scanning. Based on the mask information, the magnetic resonance spectroscopy positioning grid information is registered to the reference space. Magnetic resonance spectroscopy imaging is performed based on the positioning grid information mapped to the reference space. Three-dimensional multi-nucleus spectral scanning positioning is achieved through a closed loop of scanning-analysis-optimization-scanning. The process of obtaining magnetic resonance spectroscopy imaging localization grid information based on the type of region of interest includes the following steps: For a single region of interest, optimization is performed with the goal of maximizing the coverage of the region of interest, under the constraints of preset translation and rotation angle ranges. For multiple regions of interest, the importance of the disease is ranked and weighted based on each region of interest, with the goal of prioritizing the coverage of high-weight regions of interest; For tumor boundaries, optimization is performed based on the probability distribution map of tumor regions, aiming to maximize the coverage of important tumor areas and minimize the inclusion of healthy tissue. The process of registering the magnetic resonance spectroscopy imaging positioning grid information to the reference space includes the following steps: Based on the mask information, the high-resolution structural image is linearly / nonlinearly registered to the reference space to obtain the mapping matrix, and the magnetic resonance spectroscopy imaging positioning grid information is registered to the reference space using the mapping matrix. or The preset standard spatial structure template is registered to the high-resolution structure image space using nonlinear registration to obtain the first mapping matrix. The high-resolution structure image is then registered to the reference space to obtain the second mapping matrix. The mask information of the standard space and the magnetic resonance spectroscopy imaging positioning grid information are then registered to the reference space using the first and second mapping matrices.

2. The three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization according to claim 1, characterized in that, The multi-scale search includes the following steps: A coarse search is performed with a relatively large rotation step size and translation step size; A fine search is performed with relatively small rotation and translation step sizes.

3. The three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization according to claim 1, characterized in that, The optimization objectives during the optimization of the magnetic resonance spectroscopy imaging localization grid include: The region of interest is covered as completely as possible by the minimum number of voxels in the magnetic resonance spectroscopy imaging localization grid; In magnetic resonance spectroscopy imaging, the orientation of voxels in the localization grid should be as close as possible to the region of interest.

4. The three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization according to claim 1, characterized in that, After performing magnetic resonance spectroscopy imaging scans based on the magnetic resonance spectroscopy imaging localization grid information mapped to the reference space, the process also includes: The optimized magnetic resonance spectroscopy imaging scanning positioning parameters and the spatial coordinate index of the voxel of interest are obtained. The magnetic resonance spectroscopy imaging scanning positioning parameters include translation, rotation angle and voxel spatial coordinates of the region of interest.

5. The three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization according to claim 1, characterized in that, The high-resolution structural image data is registered with a pre-set standard spatial region of interest template to obtain mask information, or a pre-trained 3D U-Net deep learning model is used to segment the region of interest in the high-resolution structural image data to obtain the corresponding mask information.

6. The three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization according to claim 1, characterized in that, The high-resolution structural image data is any one of T1WI, T2WI, T2 FLAIR, DWI, and QSM. The magnetic resonance spectroscopy imaging positioning grid information includes the positioning grid translation, rotation angle, and voxel spatial coordinate information of the region of interest.

7. A three-dimensional multi-kernel spectral scanning and positioning system based on region of interest optimization, characterized in that, Includes a scanning subsystem and an analysis subsystem. The scanning subsystem includes: Multinuclear imaging magnetic resonance scanner and coil, the coil including a multichannel head and neck 1 H coil and 1 H / 31 The P-type dual-tuned coil, the multi-core imaging magnetic resonance scanner is used to acquire magnetic resonance data, the magnetic resonance data including data from a multi-channel head and neck imaging system. 1 High resolution obtained from H coil 1 H-structure image, and the use of 1 H / 31 The T1-weighted structured image obtained from the P-dual-tuned coil serves as the reference space. The multi-nuclear imaging magnetic resonance scanner is also used to receive and configure the optimized analysis subsystem. 31 P-magnetic resonance spectroscopy imaging scanning localization parameters; The control console, connected to the multi-core imaging magnetic resonance scanner, is used to display the spatial location and grid coverage of the region of interest and to perform scanning and positioning operations. The analysis subsystem includes: A server is provided for implementing the three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization as described in any one of claims 1-6.

8. An electronic device, characterized in that, It includes one or more processors, a memory, and one or more programs stored in the memory, said one or more programs including instructions for executing the three-dimensional multi-kernel spectral scanning localization method based on region of interest optimization as described in any one of claims 1-6.

Images