A Low-Dose PET Image Restoration Method, System, Device, and Medium

Abstract

The present invention provides a method, system, device and medium for restoring low-dose PET images. The method includes: S1. Performing block processing on training images including low-dose PET images, MR images and standard-dose PET images to obtain first patches, and performing first preprocessing on the first patches to obtain second patches; S2. According to the second patches, using sparse coding and dictionary update to obtain a first joint dictionary; S3. Restoring the low-dose PET image into a restored image of a standard-dose PET according to the first joint dictionary. The present invention solves the defects in the prior art such as poor restoration effect of low-dose PET images, complex image restoration process, and low image restoration accuracy, can solve the problems of serious noise and detail loss in low-dose PET images, and improve the image restoration efficiency and image restoration accuracy at the same time.

Description

Technical Field

[0001] The present invention relates to the field of PET image processing, and in particular to a low-dose PET image restoration method, system, device and medium. Background Art

[0002] PET / MR is a large-scale functional metabolic and molecular imaging diagnostic device that combines positron emission tomography (PET) with magnetic resonance imaging (MRI). It has the inspection functions of both PET and MR, and has the advantages of high sensitivity, good accuracy, and low radiation. It is effective in the diagnosis of many diseases, especially tumors and cardiovascular and cerebrovascular diseases. The PET imaging dose range is 4.43-7.35mSv, with an average of 5.89mSv. The standard dose of PET imaging agent still has a certain amount of radiation to the human body, which will increase the possibility of various diseases under the cumulative effect and affect human health. Therefore, reasonably reducing the PET dose to below the PET imaging dose range can reduce the impact of radiation on the human body. The restoration from low-dose PET images to standard-dose PET images is of great significance to the application of PET / MR technology and medical diagnosis.

[0003] Low-dose PET images reduce the dose of developer during imaging, resulting in a lot of noise after imaging, and the image details are lost, which seriously affects the imaging diagnosis effect of PET / MR equipment on patient lesions. In addition, due to the high resolution of PET images, the image restoration process is computationally intensive, complex, and takes a long time, which seriously affects the diagnostic efficiency.

[0004] The existing technology for restoring low-dose PET images generally has problems such as poor restoration effect, complicated restoration process, and low restoration accuracy, which leads to poor restored image quality, long restoration time, and missing restored image content. Therefore, a restoration method for low-dose PET images is needed that can solve the problems of severe noise and loss of details, and at the same time improve the efficiency of image restoration and improve the accuracy of restored images. Summary of the invention

[0005] Based on the problems existing in the prior art, the present invention provides a low-dose PET image restoration method, and the specific scheme is as follows:

[0006] A low-dose PET image restoration method comprises the following steps: S1, performing block processing on training images including low-dose PET images, MR images and standard-dose PET images to obtain a first patch, and performing a first preprocessing on the first patch to obtain a second patch; S2, obtaining a first joint dictionary based on the second patch by using sparse coding and dictionary updating; S3, restoring the low-dose PET image into a restored image of the standard-dose PET based on the first joint dictionary.

[0007] Furthermore, S2 also includes the following steps: S21, using the second patch as a sample to obtain an initialization dictionary including a low-dose PET dictionary, an MR dictionary and a standard-dose PET dictionary; S22, constructing an initialization joint dictionary based on the initialization dictionary, and constructing a target matrix based on the second patch; S23, obtaining sparse coding based on the target matrix, iteratively updating the sparse coding and the initialization joint dictionary until an iteration stop condition is met, and obtaining a first joint dictionary including a first low-dose PET dictionary, a first MR dictionary and a first standard-dose PET dictionary.

[0008] Furthermore, in S23, each iteration includes first fixing the dictionary to update the sparse code, and then fixing the sparse code to update the dictionary.

[0009] Furthermore, in S23, each iteration randomly selects some samples for sparse coding and dictionary updating.

[0010] Furthermore, in the S3, the following steps are also included: S31, performing block processing on the low-dose PET image and the MR image to obtain a third patch, and performing a second preprocessing on the third patch to obtain a fourth patch; S32, merging the first low-dose PET dictionary and the first MR dictionary obtained in the S2 into a second joint dictionary, and obtaining a second sparse code according to the fourth patch and the second joint dictionary; S33, obtaining a predicted patch according to the first standard dose PET dictionary and the second sparse code obtained in the S2, restoring the predicted patch into a two-dimensional dot matrix, and obtaining a restored image of the standard dose PET.

[0011] In particular, the first patch is a one-dimensional vector formed by randomly selecting image blocks from multiple frames of images and extending them, including the first patch of the low-dose PET image, the first patch of the MR image and the first patch of the standard-dose PET image; the first patch is in the same position in the multiple frames of images.

[0012] In particular, when selecting a position, the third patch covers the entire frame of image in the order of multiple frames of image.

[0013] In particular, in S1, the first preprocessing includes mapping the first patch of the low-dose PET image and the first patch of the MR image to the imaging space of the standard-dose PET image through a preset matrix to obtain the second patch of the low-dose PET image and the second patch of the MR image.

[0014] In particular, in S31, the second preprocessing includes mapping the third patch of the low-dose PET image and the third patch of the MR image to the imaging space of the standard-dose PET image through a preset matrix to obtain the fourth patch of the low-dose PET image and the fourth patch of the MR image.

[0015] In particular, S21 includes adopting a K-means clustering algorithm, taking the second patch as a sample, obtaining K cluster centers as an initialization dictionary, and normalizing the initialization dictionary.

[0016] Furthermore, in S22, the expressions of the initialized joint dictionary and the target matrix are respectively:

[0017]

[0018] Among them, D represents the initialized joint dictionary, Y represents the target matrix, Dl represents the low-dose PET dictionary, Dr represents the MR dictionary, Ds represents the standard-dose PET dictionary, Yl represents the second patch of the low-dose PET image, Yr represents the second patch of the MR image, and Ys represents the second patch of the standard-dose PET image.

[0019] Furthermore, in S23, the expression of the sparse coding includes:

[0020]

[0021] Where X represents sparse coding, Λ is a diagonal matrix whose diagonal elements are Λ q =d q -y i , here d q is the qth atom in dictionary D, y i is the i-th element of Y, and λ represents the sparse constraint coefficient.

[0022] Furthermore, in S23, the dictionary update expression includes:

[0023]

[0024] Among them, y i is the i-th element of Y, is the absolute value of each element of sample x, ψ i For is a diagonal matrix with diagonal elements, y i is the i-th element of Y, and μ represents the sparse constraint coefficient.

[0025] In particular, the expression of the dictionary update is solved by the gradient descent method, and the expression of the gradient descent method is:

[0026]

[0027] Among them, d q is the qth atom in dictionary D, k is the number of iterations, a q for The qth column element of b q for The qth column element of a qq for The element at row q and column q.

[0028] A low-dose PET image restoration system comprises: a sample acquisition unit, used for performing block processing on training images including low-dose PET images, MR images and standard-dose PET images to obtain first patches, and performing first preprocessing on the first patches to obtain second patches; a joint dictionary acquisition unit, used for obtaining a first joint dictionary through sparse coding and dictionary updating according to the second patches; and an image restoration unit, used for restoring the low-dose PET images into restored images of standard-dose PET according to the first joint dictionary.

[0029] Furthermore, the joint dictionary acquisition unit also includes: an initialization unit: used to acquire an initialization dictionary including a low-dose PET dictionary, an MR dictionary and a standard-dose PET dictionary using the second patch as a sample; a construction unit: used to construct an initialization joint dictionary based on the initialization dictionary, and to construct a target matrix based on the second patch; an iteration unit: used to acquire sparse coding based on the target matrix, iteratively update the sparse coding and the initialized joint dictionary until an iteration stop condition is met, and acquire the first joint dictionary including a first low-dose PET dictionary, a first MR dictionary and a first standard-dose PET dictionary.

[0030] In particular, the iteration unit further includes: an iteration update unit: used to first update the sparse coding with a fixed dictionary and then update the dictionary with a fixed sparse coding in each iteration; a sample selection unit: used to randomly select some samples for sparse coding and dictionary update in each iteration.

[0031] A computer device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the low-dose PET image restoration method as described above.

[0032] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the low-dose PET image restoration method as described above.

[0033] The present invention has the following beneficial effects:

[0034] In view of the serious noise and detail loss of low-dose PET images, the present invention proposes a low-dose PET image restoration method, system, device and medium, which solves the common defects of the prior art, such as poor image restoration effect, complex image restoration process, and low image restoration accuracy. It can effectively solve the serious noise and detail loss of low-dose PET images, and at the same time improve the efficiency of image restoration and the accuracy of restored images. Applying the low-dose PET restoration method to a specific system, computer equipment and computer storage medium, and concretizing the method, is of great significance to the development of the medical field.

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 is a flow chart of the low-dose PET image restoration method of the present invention;

[0038] Figure 2 is a specific flow chart of S2 of the low-dose PET image restoration method of the present invention;

[0039] Figure 3 is a block diagram of a low-dose PET image restoration system of the present invention;

[0040] Figure 4 is a specific block diagram of the low-dose PET image restoration system of the present invention;

[0041] Figure 5It is a schematic diagram of applying the low-dose PET image restoration method of the present invention to a computer device. DETAILED DESCRIPTION

[0042] Example 1

[0043] In view of the serious noise and detail loss of low-dose PET images, this embodiment proposes a low-dose PET image restoration method. The specific method steps are as shown in the attached manual. Figure 1 The specific plan is as follows:

[0044] S1, image preprocessing: performing block processing on training images including low-dose PET images, MR images and standard-dose PET images to obtain a first patch, and performing a first preprocessing on the first patch to obtain a second patch;

[0045] S2, obtaining a first joint dictionary: according to the second patch, obtaining a first joint dictionary by using sparse coding and dictionary updating;

[0046] S3. Obtaining a restored image: restoring the low-dose PET image to a restored image of the standard-dose PET according to the first joint dictionary.

[0047] Specifically, S1 is to obtain a first patch from a training sample, and to perform a first preprocessing on the first patch to obtain a second patch. The training images targeted by this embodiment include low-dose PET images, MR images, and standard-dose PET images, wherein the low-dose PET images are images to be restored. PET images are a set of continuous human tomographic images, and multiple single-frame images constitute multi-frame images. In this embodiment, the first patch is first obtained from the training image, and the first patch is a one-dimensional vector randomly selected from image blocks in multiple frames and extended. Since the first patch is randomly selected from multiple frames, there will be repetitions, so it is necessary to remove the repeated parts and then participate in the training as samples. In particular, the first patch of the low-dose PET image, the first patch of the MR image, and the first patch of the standard-dose PET image are obtained at the same position on the multiple frames. Since the imaging space of the low-dose PET image and the MR image is different from the imaging space of the standard-dose PET image, it cannot be directly used for subsequent dictionary learning, and the first patch needs to be preprocessed first. The first preprocessing includes obtaining a mapping matrix M to map the first patch of the low-dose PET image and the first patch of the MR image to the imaging space of the standard-dose PET image, and obtain the mapped first patch, that is, the second patch. The mapping matrix can be used in point-to-point, edge-to-edge, etc. to make the mapping both accurate and general. The specific mapping matrix is ​​expressed as:

[0048] f(Y)=M*Y

[0049] Where Y is the first patch vector matrix and M is the mapping matrix.

[0050] Specifically, S2 mainly obtains the first joint dictionary. S2 contains two main modules: sparse coding and dictionary update. The dictionary is acquired by alternating sparse coding and dictionary update. S2 specifically includes the following steps: S21. Acquire the initialization dictionary: use the second patch as a sample to acquire the initialization dictionary including the low-dose PET dictionary, MR dictionary and standard-dose PET dictionary; S22. Construct the initialization dictionary and the target matrix: construct the initialization joint dictionary according to the initialization dictionary, and construct the target matrix according to the second patch; S23. Acquire the first joint dictionary: acquire the sparse coding according to the target matrix, and iteratively update the sparse coding and the initialized joint dictionary according to the K-SVD idea until the iteration stop condition is met. Each iteration includes first fixing the dictionary to update the sparse coding, and then fixing the sparse coding to update the dictionary. The first joint dictionary acquired includes the first low-dose PET dictionary, the first MR dictionary and the first standard-dose PET dictionary. The specific steps of S2 are as shown in the attached manual. Figure 2 Preferably, this embodiment adopts Local Coordinate Coding for dictionary learning, and the specific expression is:

[0051]

[0052] Among them, X represents sparse coding, D represents the feature matrix (dictionary), x i represents the sparse coefficient of the i-th sample, d represents the feature dimension of the feature dictionary, q represents the q-th column of the dictionary, and the main purpose of dictionary learning is to obtain D.

[0053] S21 uses the second patch as a sample to obtain an initialization dictionary including a low-dose PET dictionary, an MR dictionary, and a standard-dose PET dictionary. Specifically, through a K-means clustering algorithm, the second patch is used as a sample to obtain K cluster centers as the initialization dictionary, and the initialization dictionary includes a low-dose PET dictionary, an MR dictionary, and a standard-dose PET dictionary. In addition, the initialization dictionary needs to be normalized.

[0054] S22 constructs an initialization joint dictionary according to the initialization dictionary, and constructs a target matrix according to the second patch. The expressions for initializing the joint dictionary D and the target matrix Y are:

[0055]

[0056] Among them, Dl represents the low-dose PET dictionary, Dr represents the MR dictionary, Ds represents the standard-dose PET dictionary, Yl represents the second patch of the low-dose PET image, Yr represents the second patch of the MR image, and Ys represents the second patch of the standard-dose PET image.

[0057] S23 iteratively updates the sparse code and dictionary according to the K-SVD idea until the iteration stop condition is met. Each iteration includes first fixing the dictionary to update the sparse code, and then fixing the sparse code to update the dictionary.

[0058] For the sparse coding part, the sparse coding expression for the dictionary D and the target matrix Y is:

[0059]

[0060] Where X represents sparse coding, Λ is a diagonal matrix whose diagonal elements are Λ q =d q -y i , here d q is the qth atom of D, y i is the ith element of Y, and λ represents the sparse constraint coefficient. For the above formula, this embodiment selects to use the MP (Matching Pursuits) algorithm, the OMP (Orthogonal Matching Pursuit) algorithm or the LASSO (Least absolute shrinkage and selection operator) algorithm for solving.

[0061] For the dictionary update part, the dictionary update expression for the dictionary D and the target matrix Y is:

[0062]

[0063] Among them, y i is the i-th element of Y, is the absolute value of each element of sample x, ψ i For is a diagonal matrix with diagonal elements, y i is the i-th element of Y, and μ represents the sparse constraint coefficient. This embodiment uses the gradient descent method to solve the above formula, and the specific expression is as follows:

[0064]

[0065] Among them, k is the number of iterations, a q for The qth column of b q for The qth column element of a qq for The element at row q and column q.

[0066] This embodiment iteratively updates the sparse code and the dictionary based on the K-SVD concept until the iteration stop condition is met. Each iteration includes first fixing the dictionary D to update the sparse code X, and then fixing the sparse code X to update the dictionary D. The K-SVD concept is a classic dictionary training algorithm. According to the minimum error principle, the error term is decomposed by SVD, and the decomposition term with the minimum error is selected as the updated dictionary atom and the corresponding atomic coefficient. After continuous iteration, an optimized solution is obtained.

[0067] In particular, this embodiment adopts an online learning method, that is, N samples are randomly selected for training in each iteration. Different from the method of iterating all training samples at the same time in the prior art, this embodiment randomly selects some samples for iteration, which greatly improves the training speed while ensuring the training accuracy, thereby shortening the entire low-dose image restoration time. Randomly selecting samples for training ensures that the training accuracy will not differ due to chance, and selecting some samples for training avoids repeated training of samples, greatly shortens the iteration time, and improves the training efficiency.

[0068] Specifically, S3 restores the low-dose PET image to a restored image of the standard-dose PET according to the joint dictionary. The specific steps of S3 include: S31 performs block processing on the low-dose PET image and the MR image to obtain a third patch, and performs a second preprocessing on the third patch to obtain a fourth patch; S32 combines the first low-dose PET dictionary and the first MR dictionary obtained in S2 into a second joint dictionary, and obtains a second sparse code according to the fourth patch and the second joint dictionary; S33 obtains a predicted patch according to the first standard-dose PET dictionary and the second sparse code obtained in S2, and restores the predicted patch to a two-dimensional dot matrix to obtain a restored image of the standard-dose PET.

[0069] Among them, S31 performs block processing on the low-dose PET image and the MR image to obtain the third patch, and performs a second preprocessing on the third patch to obtain the fourth patch. The block method is the same as S1, randomly selecting image blocks from multiple frames of images and extending them into a one-dimensional vector to obtain the third patch. The third patch includes the third patch of the low-dose PET image and the third patch of the MR image. In particular, the selection position of the third patch needs to be in accordance with the arrangement order of the multiple frames of images so that it can cover the entire frame of the image. In particular, there can be overlapping parts between the third patches, which can reduce the block effect of the result. Similarly, since the imaging space of the low-dose PET image and the MR image is different from the imaging space of the standard-dose PET image, it cannot be directly used for subsequent dictionary learning, and the third patch needs to be preprocessed for the second time. The second preprocessing includes obtaining a mapping matrix M to map the third patch of the low-dose PET image and the third patch of the MR image to the imaging space of the standard-dose PET image, and obtain the mapped third patch, that is, the fourth patch. The mapping matrix can be used in a point-to-point, edge-to-edge, and other ways to make the mapping both accurate and general.

[0070] Among them, S32 combines the first low-dose PET dictionary and the first MR dictionary obtained in S2 into a second joint dictionary, and obtains the corresponding second sparse code according to the second joint dictionary D and the fourth patch. The specific expression is as follows:

[0071]

[0072] Dl represents the first low-dose PET dictionary, Dr represents the first MR dictionary, Yl represents the fourth patch of the low-dose PET image, and Yr represents the fourth patch of the MR image.

[0073] Finally, S33 obtains the prediction patch according to the first standard dose PET dictionary and the second sparse coding, restores the prediction patch into a two-dimensional dot matrix, and obtains the restored image of the standard dose PET. The corresponding prediction patch is predicted by the first standard dose PET dictionary obtained in S2 and the second sparse coding obtained, and the prediction patch is restored into a two-dimensional dot matrix to obtain the final restored image of the standard dose PET. The final restored image of the standard dose PET is the restored image of the low dose PET image.

[0074] The method proposed in this embodiment is compatible and can also be applied to other types of medical image reconstruction fields, such as CT images, etc., and can improve the image reconstruction effect by combining with deep learning related methods. In addition, the method proposed in this embodiment has a strong advantage in noise reduction and can also be applied to image denoising related fields.

[0075] This embodiment provides a low-dose PET image restoration method. Low-dose PET images are restored through dictionary learning and sparse matrices. After the dictionary is constructed, the dictionary is updated using a corresponding algorithm to obtain a joint dictionary that is more suitable for restoring standard-dose images, and the accuracy of restoration is improved by combining MR images. At the same time, online learning related methods are applied to sparse dictionary updates to speed up convergence, so that the time required for the entire image restoration process is greatly shortened, thereby improving the efficiency of image restoration.

[0076] Example 2

[0077] Based on Example 1, this example provides a low-dose PET image restoration system, which modularizes the low-dose PET image restoration method of Example 1. The specific scheme is as follows:

[0078] A low-dose PET image restoration system comprises: a sample acquisition unit, a joint dictionary acquisition unit and an image restoration unit, wherein the joint dictionary acquisition unit is connected to the sample acquisition unit and the image restoration unit respectively. Figure 3 shown.

[0079] Specifically, the sample acquisition unit is used to perform block processing on the training images including low-dose PET images, MR images and standard-dose PET images to obtain the first patch, and perform first preprocessing on the first patch to obtain the second patch. The user inputs the low-dose PET image as a sample through the sample acquisition unit, and the sample acquisition unit obtains the first patch according to the training image, and performs mapping processing on the first patch, and maps the first patch of the low-dose PET image and the first patch of the MR image to the imaging space of the standard-dose PET image through the mapping matrix M to obtain the mapped first patch, that is, the second patch. The second patch is then passed to the joint dictionary acquisition unit.

[0080] Specifically, the joint dictionary acquisition unit is used to acquire the first joint dictionary according to the sparse coding and dictionary update. The joint dictionary acquisition unit mainly includes an initialization unit, a construction unit and an iteration unit. The construction unit is respectively connected to the initialization unit and the iteration unit, as shown in the attached specification. Figure 4 shown.

[0081] Specifically, the initialization unit is used to obtain an initialization dictionary including a low-dose PET dictionary, an MR dictionary and a standard-dose PET dictionary using the second patch as a sample. The initialization unit receives the second patch transmitted from the sample acquisition unit, adopts the K-means clustering algorithm, uses the second patch as a sample, obtains K cluster centers as the initialization dictionary, and normalizes the initialization dictionary. The construction unit is used to construct an initialization joint dictionary according to the initialization dictionary, and construct a target matrix according to the second patch, wherein the initialization joint dictionary includes a low-dose PET dictionary, an MR dictionary and a standard-dose PET dictionary, and the target matrix includes the second patch of the low-dose PET image, the second patch of the MR image and the second patch of the standard-dose PET image; the iteration unit is used to obtain sparse coding according to the target matrix, iteratively update the sparse coding and the initialization joint dictionary according to the K-SVD idea until the iteration stop condition is met, and obtain the first joint dictionary including the first low-dose PET dictionary, the first MR dictionary and the first standard-dose PET dictionary. The iteration unit also includes an iteration update unit and a sample selection unit. The iteration update unit is used to first fix the dictionary to update the sparse coding, and then fix the sparse coding to update the dictionary in each iteration. The sample selection unit is used to randomly select some samples for sparse coding and dictionary update at each iteration. The iteration unit is the core processing unit, which uses the K-SVD idea to iteratively update the sparse coding and dictionary to ensure the accuracy of image restoration. In addition, the iteration unit is also equipped with a sample selection unit, which is used to randomly select some samples for sparse coding and dictionary update at each iteration, improving training efficiency while ensuring training accuracy.

[0082] Specifically, the image restoration unit is used to restore the low-dose PET image to a restored image of the standard-dose PET according to the first joint dictionary obtained by the joint dictionary acquisition unit. The image restoration unit is the final restoration module. First, the image restoration unit performs block processing on the low-dose PET image and the MR image to obtain the third patch, and then maps the third patch of the low-dose PET image and the third patch of the MR image to the imaging space of the standard-dose PET image through the mapping matrix M to obtain the mapped third patch, that is, the fourth patch. Secondly, the low-dose PET dictionary and the MR dictionary obtained in the joint dictionary acquisition unit are merged into a joint dictionary, and the second sparse code is obtained in combination with the fourth patch. Finally, the corresponding predicted patch is predicted by the standard-dose PET dictionary and the obtained second sparse code, and the predicted patch is restored to a two-dimensional dot matrix to obtain the final restored image of the standard-dose PET. The final restored image of the standard-dose PET is the restored image of the low-dose PET image.

[0083] This embodiment proposes a low-dose PET image restoration system based on a low-dose PET image restoration method proposed in Example 1. The method of Example 1 can effectively solve the problems of severe noise and detail loss in low-dose PET images, while improving the efficiency of image restoration and the accuracy of restored images.

[0084] Example 3

[0085] Figure 5 A schematic diagram of the structure of a computer device provided in Example 3 of the present invention. Figure 5 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0086] like Figure 5 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 connecting different system components (including the system memory 28 and the processing unit 16). The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the device computer 12, including volatile and non-volatile media, removable and non-removable media. The system memory 28 may include computer system readable media in the form of volatile memory.

[0087] The computer device 12 may also communicate with one or more external devices 14 (e.g., a keyboard, a pointing device, a display, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices.

[0088] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing a low-dose PET image restoration method provided in Embodiment 1 of the present invention, the method comprising:

[0089] S1. Perform block processing on training images including low-dose PET images, MR images and standard-dose PET images to obtain a first patch, and perform a first preprocessing on the first patch to obtain a second patch; S2. According to the second patch, use sparse coding and dictionary update to obtain a first joint dictionary; S3. Restore the low-dose PET image into a restored image of the standard-dose PET according to the first joint dictionary.

[0090] Among them, S2 specifically includes: S21, using the second patch as a sample to obtain an initialization dictionary including a low-dose PET dictionary, an MR dictionary, and a standard-dose PET dictionary; S22, constructing an initialization joint dictionary based on the initialization dictionary, and constructing a target matrix based on the second patch; S23, obtaining sparse coding based on the target matrix, iteratively updating the sparse coding and the initialization joint dictionary until the iteration stop condition is met, and obtaining a first joint dictionary including a first low-dose PET dictionary, a first MR dictionary, and a first standard-dose PET dictionary. Among them, each iteration includes first fixing the dictionary to update the sparse coding, and then fixing the sparse coding to update the dictionary. Each iteration randomly selects some samples for sparse coding and dictionary update.

[0091] Among them, S3 specifically includes: S31, performing block processing on the low-dose PET image and the MR image to obtain a third patch, and performing a second preprocessing on the third patch to obtain a fourth patch; S32, merging the first low-dose PET dictionary and the first MR dictionary obtained in S2 into a second joint dictionary, and obtaining a second sparse code according to the fourth patch and the second joint dictionary; S33, obtaining a predicted patch according to the first standard dose PET dictionary and the second sparse code obtained in S2, restoring the predicted patch into a two-dimensional dot matrix, and obtaining a restored image of the standard dose PET.

[0092] This embodiment applies a low-dose PET image restoration method to a specific computer device, and stores the method in a memory. When an executor executes the memory, the method is run to perform low-dose PET image restoration. The method is quick and convenient to use and has a wide range of applications.

[0093] Of course, those skilled in the art can understand that the processor can also implement the technical solution of the PET image restoration method provided by any embodiment of the present invention.

[0094] Example 4

[0095] This embodiment 4 provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the low-dose PET image restoration method provided in any embodiment of the present invention are implemented. The method includes:

[0096] S1. Perform block processing on training images including low-dose PET images, MR images and standard-dose PET images to obtain a first patch, and perform a first preprocessing on the first patch to obtain a second patch; S2. According to the second patch, use sparse coding and dictionary update to obtain a first joint dictionary; S3. Restore the low-dose PET image into a restored image of the standard-dose PET according to the first joint dictionary.

[0097] Among them, S2 specifically includes: S21, using the second patch as a sample to obtain an initialization dictionary including a low-dose PET dictionary, an MR dictionary, and a standard-dose PET dictionary; S22, constructing an initialization joint dictionary based on the initialization dictionary, and constructing a target matrix based on the second patch; S23, obtaining sparse coding based on the target matrix, iteratively updating the sparse coding and the initialization joint dictionary until the iteration stop condition is met, and obtaining a first joint dictionary including a first low-dose PET dictionary, a first MR dictionary, and a first standard-dose PET dictionary. Among them, each iteration includes first fixing the dictionary to update the sparse coding, and then fixing the sparse coding to update the dictionary. Each iteration randomly selects some samples for sparse coding and dictionary update.

[0098] Among them, S3 specifically includes: S31, performing block processing on the low-dose PET image and the MR image to obtain a third patch, and performing a second preprocessing on the third patch to obtain a fourth patch; S32, merging the first low-dose PET dictionary and the first MR dictionary obtained in S2 into a second joint dictionary, and obtaining a second sparse code according to the fourth patch and the second joint dictionary; S33, obtaining a predicted patch according to the first standard dose PET dictionary and the second sparse code obtained in S2, restoring the predicted patch into a two-dimensional dot matrix, and obtaining a restored image of the standard dose PET.

[0099] The computer storage medium of the present embodiment can adopt any combination of one or more computer-readable media. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable storage media can be, for example, but not limited to: electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this document, computer-readable storage media can be any tangible medium containing or storing programs, which can be used by instruction execution systems, devices or devices or used in combination with them.

[0100] This embodiment applies a low-dose PET image restoration method to a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the low-dose PET image restoration method provided by the present invention are implemented, which is simple, fast, easy to store, and not easy to lose.

[0101] The existing technology for low-dose PET images generally has problems such as poor restoration effect, complex restoration process, and low restoration accuracy, which leads to poor restored image quality, long restoration time, and missing restored image content. Therefore, a restoration method for low-dose PET images is needed that can solve the problems of severe noise and loss of details, and at the same time improve the efficiency of image restoration and improve the accuracy of restored images.

[0102] In summary, the present invention proposes a low-dose PET image restoration method, system, device and medium, which solves the common defects of the prior art such as poor image restoration effect, complex image restoration process, and low image restoration accuracy, and can solve the problems of severe noise and detail loss in low-dose PET images, while improving the efficiency of image restoration and the accuracy of restored images. The standard-dose PET images can be restored from low-dose PET images through dictionary learning and sparse matrix, which overcomes the shortcoming that traditional denoising methods cannot retain details; online learning-related concepts are adopted to randomly obtain smaller training samples during the learning process, which accelerates the convergence speed while ensuring accuracy compared to traditional technologies. The low-dose PET restoration method is applied to specific systems, computer equipment and computer storage media, and the method is concretized, which is of great significance to the development of the medical impact field.

[0103] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0104] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

[0105] The above disclosure is only a few specific implementation scenarios of the present invention, but the present invention is not limited thereto, and any changes 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 reducing low-dose PET images, characterized in that, it comprises the following steps: S1. Perform block processing on training images including low-dose PET images, MR images, and standard-dose PET images to obtain first patches, and perform first preprocessing on the first patches to obtain second patches; S2. According to the second patches, use sparse coding and dictionary update to obtain a first combined dictionary; In the S2, it includes the following steps: S21. Use the second patches as samples to obtain an initial dictionary including a low-dose PET dictionary, an MR dictionary, and a standard-dose PET dictionary; S22. Construct an initial combined dictionary according to the initial dictionary, and construct a target matrix according to the second patches; S23. Obtain sparse coding according to the target matrix, and perform iterative update on the sparse coding and the initial combined dictionary until the iterative stop condition is met, and then obtain a first combined dictionary including a first low-dose PET dictionary, a first MR dictionary, and a first standard-dose PET dictionary; In the S23, each iteration includes first fixing the dictionary to update the sparse coding, and then fixing the sparse coding to update the dictionary. Each iteration randomly selects some samples for sparse coding and dictionary update; S3. Restore the low-dose PET image to a restored image of standard-dose PET according to the first combined dictionary; It includes the following steps: S31. Perform block processing on the low-dose PET image and the MR image to obtain third patches, and perform second preprocessing on the third patches to obtain fourth patches; S32. Merge the first low-dose PET dictionary and the first MR dictionary obtained in S2 into a second combined dictionary, and obtain second sparse coding according to the fourth patches and the second combined dictionary; S33. Obtain predicted patches according to the first standard-dose PET dictionary and the second sparse coding obtained in S2, and restore the predicted patches to a two-dimensional lattice to obtain a restored image of standard-dose PET; In the S31, the second preprocessing includes mapping the third patches of the low-dose PET image and the third patches of the MR image to the imaging space of the standard-dose PET image through a preset matrix to obtain the fourth patches of the low-dose PET image and the fourth patches of the MR image; The first patches are one-dimensional vectors randomly selected from multiple frames of images and extended, including the first patches of the low-dose PET image, the first patches of the MR image, and the first patches of the standard-dose PET image; The first patches are in the same position in multiple frames of images; when selecting positions for the third patches, the entire frame of image is covered in the order of multiple frames of images.

2. The method according to claim 1, characterized in that, in the S1, the first preprocessing includes mapping the first patches of the low-dose PET image and the first patches of the MR image to the imaging space of the standard-dose PET image through a preset matrix to obtain the second patches of the low-dose PET image and the second patches of the MR image.

3. The method according to claim 1, It is characterized in that in the S21, it includes using the K-means clustering algorithm, taking the second patch as a sample, obtaining K clustering centers as the initial dictionary, and performing normalization processing on the initial dictionary.

4. The method according to claim 1, it is characterized in that in the S22, the expressions of the initial joint dictionary and the target matrix are respectively: where D represents the initial joint dictionary, Y represents the target matrix, Dl represents the low-dose PET dictionary, Dr represents the MR dictionary, Ds represents the standard-dose PET dictionary, Yl represents the second patch of the low-dose PET image, Yr represents the second patch of the MR image, and Ys represents the second patch of the standard-dose PET image.

5. The method according to claim 1, it is characterized in that in the S23, the expression of the sparse coding includes: Among them, D represents the initialized combined dictionary, Y represents the target matrix, X represents the sparse coding, Λ is a diagonal matrix, and its diagonal elements are Λ q = d q - y i , where d q is the q-th atom of the dictionary D, and y i is the i-th element of Y, and λ represents the sparse constraint coefficient.

6. The method according to claim 1, it is characterized in that in the S23, the expression of the dictionary update includes: Among them, D represents the initialized combined dictionary, Y represents the target matrix, and y i is the i-th element of Y, is the absolute value of each element of the sample x, and ψ i is a diagonal matrix with diagonal elements, and μ represents the sparse constraint coefficient.

7. The method according to claim 6, it is characterized in that solving the expression of the dictionary update by the gradient descent method, and the expression of the gradient descent method is: where d q is the q-th atom of dictionary D, k is the iteration number, a q is the q-th column element of , b q is the q-th column element of , and a qq is the element at the q-th row and q-th column of .

8. A low-dose PET image restoration system, it is characterized in that it includes: a sample acquisition unit, configured to perform block processing on training images including low-dose PET images, MR images, and standard-dose PET images to obtain first patches, and perform first preprocessing on the first patches to obtain second patches; the first patches are one-dimensional vectors randomly selected from multiple frames of images and extended, including the first patches of the low-dose PET images, the first patches of the MR images, and the first patches of the standard-dose PET images; the first patches are in the same position in multiple frames of images; when selecting the position of the third patch, the entire frame of image is covered in the order of multiple frames of images; a joint dictionary acquisition unit, configured to obtain a first joint dictionary according to the second patch through sparse coding and dictionary update; obtain an initial dictionary including a low-dose PET dictionary, an MR dictionary, and a standard-dose PET dictionary taking the second patch as a sample; construct an initial joint dictionary according to the initial dictionary, construct a target matrix according to the second patch; obtain sparse coding according to the target matrix, and perform iterative update on the sparse coding and the initial joint dictionary until the iterative stop condition is met, and then obtain a first joint dictionary including a first low-dose PET dictionary, a first MR dictionary, and a first standard-dose PET dictionary; each iteration includes first fixing the dictionary to update the sparse coding, and then fixing the sparse coding to update the dictionary, and randomly selecting some samples for sparse coding and dictionary update each time; an image restoration unit, configured to restore the low-dose PET image to a restored image of the standard-dose PET according to the first joint dictionary; The third patch is obtained by performing block processing on the low-dose PET image and the MR image, and the fourth patch is obtained by performing a second preprocessing on the third patch; the obtained first low-dose PET dictionary and the first MR dictionary are merged into a second combined dictionary, and a second sparse coding is obtained according to the fourth patch and the second combined dictionary; A predicted patch is obtained according to the obtained first standard-dose PET dictionary and the second sparse coding, and the predicted patch is restored to a two-dimensional dot matrix to obtain a restored image of the standard-dose PET; The second preprocessing includes mapping the third patch of the low-dose PET image and the third patch of the MR image to the imaging space of the standard-dose PET image through a preset matrix to obtain the fourth patch of the low-dose PET image and the fourth patch of the MR image.

9. The system according to claim 8, wherein, The combined dictionary obtaining unit further includes: An initialization unit: configured to obtain an initialization dictionary including a low-dose PET dictionary, an MR dictionary, and a standard-dose PET dictionary by using the second patch as a sample; A construction unit: configured to construct an initialization combined dictionary according to the initialization dictionary, and construct a target matrix according to the second patch; An iteration unit: configured to obtain a sparse coding according to the target matrix, and perform iterative update on the sparse coding and the initialization combined dictionary until the iterative stop condition is satisfied, and then obtain the first combined dictionary including the first low-dose PET dictionary, the first MR dictionary, and the first standard-dose PET dictionary.

10. The system according to claim 9, wherein, The iteration unit further includes: An iterative update unit: configured to, in each iteration, first fix the dictionary to update the sparse coding, and then fix the sparse coding to update the dictionary; A sample selection unit: configured to randomly select some samples for sparse coding and dictionary update in each iteration.

11. A computer device, wherein, The computer device includes: One or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the low-dose PET image restoration method according to any one of claims 1-7.

12. A computer-readable storage medium, on which a computer program is stored, wherein, When the program is executed by a processor, it implements the low-dose PET image restoration method according to any one of claims 1-7.

Images