Magnetic resonance imaging parameter virtual simulation method based on pca
By using the PCA algorithm to reduce the dimensionality of the magnetic resonance imaging database, the problems of high computational cost and high storage requirements in existing technologies are solved, enabling efficient simulation imaging in resource-constrained environments.
Patent Information
- Application Number
- CN202111264379.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-28
- Publication Date
- 2025-12-30
- Estimated Expiration
- 2041-10-28
AI Technical Summary
Existing virtual simulation methods for magnetic resonance imaging suffer from high computational and storage requirements, making it difficult to achieve efficient simulation, especially in environments with limited computing platforms and storage resources.
The PCA algorithm is used to reduce the dimensionality of the magnetic resonance multi-parameter combined pre-imaging database, storing only low-dimensional compressed vectors and principal component images. The three-dimensional volume data is then reconstructed through interpolation, reducing computational load and storage requirements.
It enables efficient virtual simulation of magnetic resonance imaging parameters in computationally and storage-constrained environments, reducing computational and storage requirements, and is suitable for web platforms or mobile devices.
Smart Images
Figure CN113963131B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of magnetic resonance imaging, and more particularly to a virtual simulation method for magnetic resonance imaging parameters based on PCA. Background Technology
[0002] Magnetic resonance imaging (MRI) systems, as large-scale medical imaging devices with high technological content and added value, are characterized by complex principles, high cost, and limited time constraints in clinical hospital applications. Therefore, there is a need for virtual simulation of MRI parameters in university courses and hospital clinical training. Specifically, for a virtual case or phantom, the user can adjust various imaging-related parameters, and the simulation system, based on these parameter settings, provides simulated imaging results that closely resemble those of a real MRI scan under the same parameters. In this process, the user is not concerned with the physical principles of the actual imaging process, but rather with understanding how different parameter settings affect the imaging results. In past practice, virtual simulation of MRI has mainly included the following two implementation methods.
[0003] One approach is real-time computational simulation based on a digital phantom and an idealized physical model. This method simplifies the physical principles of the magnetic resonance imaging (MRI) process and, based on the user-provided imaging parameters and the numerical distribution of MRI-related physical quantities in the digital phantom, uses the finite element method to calculate and deduce the imaging process at a specific time step and spatial resolution. After the deduction is complete, the simulated raw MRI signal is obtained. Subsequently, the raw signal is processed using the same image reconstruction algorithm as the real imaging process to obtain the simulated imaging result. This simulation method only requires storing the digital phantom and the calculation program, thus occupying relatively little storage space. However, it involves a very large number of computational steps, including many trigonometric and exponential operations, and requires the use of double-precision floating-point data types for accuracy. To meet the user's real-time requirements, high-performance professional computing accelerator cards (general GPUs cannot meet this method due to insufficient double-precision computing capabilities) or computing servers with dozens of cores are typically used for computation, resulting in high hardware costs.
[0004] Another approach is an interpolation simulation method based on a multi-parameter combined pre-imaging database. This method first takes a finite number of discrete values for each imaging parameter based on its adjustable range, then generates a full combination of all discrete values. For each parameter combination in the full combination, actual imaging or the computational simulation described in the previous method is performed to obtain and save the imaging results, constructing a pre-imaging database. In actual use, the neighboring parameter combinations in the full combination are determined based on the imaging parameters provided by the user. The pre-imaging results corresponding to the neighboring parameter combinations are retrieved from the database and linearly interpolated to obtain the final simulation result. This method does not require real-time computational simulation of the imaging process and does not require an expensive computing platform, but the capacity of the pre-imaging database is extremely large. First, the number of full combinations is very large. Second, MRI imaging results are 3D high-precision volumetric data, and the size of a single image can reach tens of megabytes. For example, with 6 adjustable parameters to meet basic requirements, and each parameter having 5 discrete values in the database, the total capacity of the entire database will reach over 700GB, which puts a significant burden on data publishing and transmission. On the other hand, although this method reduces the computational cost compared to real-time simulation, interpolating the three-dimensional volume data of imaging results corresponding to many adjacent parameter combinations still requires considerable computation. Summary of the Invention
[0005] This invention proposes using the PCA algorithm to reduce the dimensionality of a multi-parameter combined pre-imaging database for magnetic resonance imaging. After dimensionality reduction, only low-dimensional compressed vectors and a small number of principal component images need to be stored, resulting in low storage requirements. Furthermore, it proposes that during the simulation process, only the compressed vectors need to be interpolated, followed by image restoration, resulting in low computational requirements. This invention is suitable for virtual simulation of magnetic resonance imaging parameters in scenarios with limited storage and computation, such as web platforms or mobile device platforms.
[0006] To achieve the above objectives, this invention provides a virtual simulation method for magnetic resonance imaging parameters based on PCA, comprising the following steps:
[0007] S1: Determine the required types of magnetic resonance imaging parameters and calculate the full combination of all parameter types;
[0008] S2: For each parameter combination in the full combination, perform scanning imaging or finite element calculation simulation imaging on a real magnetic resonance scanner according to the parameter settings to obtain a database of three-dimensional volume data of the imaging.
[0009] S3: Use the PCA algorithm to reduce the dimensionality of the database to obtain a one-dimensional compressed vector;
[0010] S4: Based on the imaging parameter settings, find the nearest parameter combination, query its corresponding compression vector, perform linear interpolation, and refold it into three-dimensional volume data to obtain virtual simulation imaging.
[0011] Preferably, in step S1, a finite number of discrete values are determined within the adjustable range of each imaging parameter, depending on the type of magnetic resonance imaging parameter required.
[0012] Preferably, in step S2, for each parameter combination in the full combination, after obtaining the imaging result, it is saved to construct a pre-imaging database. This database stores the magnetic resonance imaging parameter combinations and the corresponding imaging three-dimensional volume data.
[0013] As a preferred approach, when using the PCA algorithm to reduce the dimensionality of the database, each 3D volume data is first expanded into a one-dimensional compressed vector. Then, all the original 3D volume data in the database constitute the first matrix. The row average vector of the first matrix is calculated, and finally, the normalized second matrix is obtained.
[0014] As a preferred approach, the covariance matrix is obtained by substituting the second matrix into the formula, and the eigenvalues and eigenvectors of the covariance matrix are obtained. The top few eigenvectors with the largest corresponding eigenvalues are arranged to form a transformation matrix, and the resulting matrix and principal component matrix after dimensionality reduction are calculated.
[0015] Preferably, in step S4, based on the imaging parameter settings given by the user, the two discrete values that are closest to each other in the upward and downward directions of several parameters are found. Then, the nearest neighbor parameter combination that is closest to the user's parameter settings and the corresponding compression vector in the whole combination are determined. Linear interpolation is performed, and then reconstruction is performed according to the relevant formula. The data is then folded back into three-dimensional volume data to obtain the virtual simulation imaging result.
[0016] The beneficial effects of this invention are as follows: Compared with the prior art, the PCA-based virtual simulation method for magnetic resonance imaging parameters provided by this invention obtains simulation results by interpolating images of neighboring parameters in the pre-imaging database. It does not require real-time calculation simulation of the imaging process, nor does it require an expensive computing platform. The PCA algorithm is used to reduce the dimensionality of the pre-imaging data, transforming a large amount of high-capacity three-dimensional volume data into low-dimensional compressed vectors and a small number of principal component images. During the simulation imaging process, the compressed vectors are interpolated before data reconstruction, rather than interpolating the three-dimensional volume data, which further reduces the amount of computation. Attached Figure Description
[0017] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation
[0018] To more clearly illustrate the present invention, it is further described below with reference to the accompanying drawings and embodiments. Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application's specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that the term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.
[0019] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.
[0020] Please see Figure 1 This invention discloses a virtual simulation method for magnetic resonance imaging parameters based on PCA, comprising the following steps: S1: Determine the required types of magnetic resonance imaging parameters and calculate all combinations of all parameter types; S2: For each parameter combination in the full combination, perform scanning imaging on a real magnetic resonance scanner according to the parameter settings to obtain a database of three-dimensional volume data with imaging; S3: Perform dimensionality reduction processing on the database using the PCA algorithm to obtain a one-dimensional compressed vector; S4: Find the nearest parameter combination according to the imaging parameter settings, query its corresponding compressed vector, perform linear interpolation, and refold it into three-dimensional volume data to obtain virtual simulation imaging. In the specific implementation process, magnetic resonance imaging requires multiple parameter information to be accurately established. Therefore, it is necessary to collect the types of magnetic resonance imaging parameters. Only after collecting these parameter information can the magnetic resonance image be established. In order to achieve the transfer and reconstruction of these parameter information and avoid excessive data volume during the transfer process, it is necessary to perform dimensionality reduction processing on the data, transforming the original three-dimensional volume data into a one-dimensional compressed vector. This greatly reduces the data volume while ensuring that no data is lost. Finally, the one-dimensional compressed vector is folded back into three-dimensional volume data through linear interpolation, thereby achieving the effect of virtual simulation imaging.
[0021] To achieve the above objectives, in step S1, a finite number of discrete values are determined within the adjustable range of each imaging parameter, based on the required types of magnetic resonance imaging parameters. In step S2, each parameter in the full combination is combined, and finite element simulation can be performed based on a numerical phantom to obtain and save the imaging results, constructing a pre-imaging database. This database protects the combinations of magnetic resonance imaging parameters and the corresponding three-dimensional imaging data. In the specific implementation process, the required types of magnetic resonance imaging parameters are determined, and their number is set to n; a finite number of discrete values are determined within the adjustable range of each imaging parameter; the number of discrete values for the n parameters are set to M1, M2, ..., Mn; then, the full combination is obtained based on the origins of these parameters, and the number of full combinations is... Let it be denoted as m.
[0022] When using the PCA algorithm for dimensionality reduction of a database, each 3D volume data is first expanded into a one-dimensional compressed vector. The original 3D volume data in the database then constitutes the first matrix. The row average vector of the first matrix is calculated, resulting in a normalized second matrix. The covariance matrix is then calculated using the formula after obtaining the second matrix. The eigenvalues and eigenvectors of the variance matrix are obtained. The top few eigenvectors with the largest corresponding eigenvalues are arranged to form a transformation matrix. The resulting matrix and principal component matrix after dimensionality reduction are then calculated. In the specific implementation, each parameter combination in the full combination is scanned and imaged on a real magnetic resonance scanner according to these parameter settings, or a finite element simulation is performed based on a digital phantom to construct a pre-imaging database. The database contains m magnetic resonance imaging parameter combinations and corresponding 3D volume data. Then, the PCA algorithm is used for dimensionality reduction. First, each 3D volume data is expanded into a one-dimensional compressed vector of length p. The original 3D volume data in the database then constitutes a p-row, m-column matrix X'. The row average vector μ of X' is calculated, and μ is subtracted from each column of X' to obtain the normalized matrix X. The formula is used... Calculate the covariance matrix C, where the symbol T denotes matrix transpose; find the eigenvalues and eigenvectors of matrix C, and arrange the k largest eigenvectors in rows to form the transformation matrix V. Calculate the dimension-reduced matrix Y = VX; if the calculation is correct, the size of Y should be k rows and m columns. Then calculate the principal component matrix P = VX. T If the calculation is correct, the size of P should be p rows and k columns; then save the principal component matrix P, the row average vector μ and the result matrix Y, and use each column of Y as a compression vector y, which corresponds to an imaging parameter combination.
[0023] In step S4, based on the imaging parameter settings provided by the user, the two most closely spaced discrete values of several parameters are found, one upward and one downward. Then, the nearest neighbor parameter combination and its corresponding compression vector are determined within the entire combination. Linear interpolation is performed, and then reconstruction is carried out according to relevant formulas, folding the data back into three-dimensional volume data to obtain the virtual simulation imaging result. In specific implementation, based on the imaging parameter settings provided by the user, the two most closely spaced discrete values of n parameters are found, one upward and one downward. Then, the nearest neighbor parameter combination and its corresponding compression vector within the entire combination are determined. n Find the nearest parameter combinations. Locate the compressed vector y corresponding to all nearest parameter combinations and interpolate them linearly to obtain y'. Reconstruct the data using the following formula: x' = Py' + μ, where x' is a one-dimensional compressed vector of length p. Fold this vector back into three-dimensional volume data to obtain the virtual simulation imaging result.
[0024] The advantages of this invention are:
[0025] 1) Virtual simulation of magnetic resonance imaging parameters is achieved through pre-imaging interpolation;
[0026] 2) The PCA algorithm is used for dimensionality reduction and compression of the pre-imaging database;
[0027] 3) The interpolation step is applied to the compressed vector of the 3D volume data rather than the 3D volume data itself.
[0028] The above-disclosed embodiments are merely specific examples of the present invention, but the present invention is not limited thereto. Any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.
Claims
1. A method for PCA-based virtual simulation of magnetic resonance imaging parameters, characterized in that, The method comprises the following steps: S1: determining the required magnetic resonance imaging parameter types, and calculating all parameter types of the full combination; S2: for each parameter combination in the full combination, performing scanning imaging or finite element calculation simulation imaging on the real magnetic resonance scanner according to the parameter setting, to obtain a database of three-dimensional body data of imaging; S3: performing dimension reduction processing on the database by using a PCA algorithm to obtain a one-dimensional compressed vector; S4: finding a nearest parameter combination according to the imaging parameter setting, querying the corresponding compressed vector, performing linear interpolation, and refolding the three-dimensional body data to obtain virtual simulation imaging; In step S4, according to the imaging parameter setting given by the user, find two discrete values closest to the user's parameter setting, and then determine the nearest parameter combination in the full combination and the corresponding compressed vector, perform linear interpolation, and then reconstruct according to the related formula to refold the three-dimensional body data, that is, obtain the result of virtual simulation imaging; When the PCA algorithm is used to reduce the dimension of the database, first, each three-dimensional body data is unfolded into a one-dimensional compressed vector, then all three-dimensional body data in the database form a first matrix, the row average vector of the first matrix is calculated, and finally the normalized second matrix is obtained; After obtaining the second matrix, the covariance matrix is calculated by substituting the formula, the eigenvalue and eigenvector of the covariance matrix are obtained, the first several eigenvectors corresponding to the maximum eigenvalue are arranged to form a transformation matrix, and the result matrix after dimension reduction and the principal component matrix are calculated. The database contains m combinations of magnetic resonance imaging parameters and corresponding three-dimensional imaging data, and then the PCA algorithm is used for dimension reduction. First, each three-dimensional data is unfolded into a one-dimensional compressed vector, and the length is p. All three-dimensional data in the database originally forms a matrix X' of p rows and m columns; the row average vector μ of X' is calculated, and each column of X' is subtracted from μ to obtain the normalized matrix X; according to the formula The covariance matrix C is calculated, where the symbol T represents matrix transposition; the eigenvalues and eigenvectors of the matrix C are obtained, the first k eigenvectors corresponding to the largest eigenvalues are arranged in rows to form a transformation matrix V, and the reduced result matrix Y = VX is calculated; if the calculation is correct, the size of Y should be k rows and m columns, and the principal component matrix is calculated, and if the calculation is correct, the size of P should be p rows and k columns; then the principal component matrix P, the row average vector μ and the result matrix Y are saved, and each column of Y is used as a compressed vector y, which corresponds to an imaging parameter combination.
2. The method of claim 1, wherein, In step S1, according to the types of the required magnetic resonance imaging parameters, a limited number of discrete values are determined in the adjustable range of each imaging parameter.
3. The method of claim 1, wherein, In step S2, after obtaining the imaging result, save it for each parameter combination in the full combination, and construct a pre-imaging database, which saves the magnetic resonance imaging parameter combination and the corresponding three-dimensional body data of imaging.