Correction method and device, imaging method and system, electronic equipment and storage medium

Abstract

The invention discloses a correction method and device, an imaging method and system, electronic equipment and a storage medium. The correction method comprises the steps that an uncorrected PET image is input into a trained correction network model, a simulated MRI image is obtained, and the correction network model is obtained based on a reversible generative network and through training of a PET training image and a corresponding MRI label image; and correcting the uncorrected PET detection data corresponding to the uncorrected PET image by using the simulated MRI image to obtain corrected PET detection data. According to the scheme, the simulated MRI image can be directly obtained from the uncorrected PET image, so that the method is applied to correction of PET detection data, an extra MRI or CT scanning process does not need to be carried out after a patient is subjected to PET detection, the discomfort of the patient and the radiation dosage to the patient are reduced, especially for children, detection is safer, and the detection efficiency is improved. And the use cost is greatly reduced.

Description

Technical Field

[0001] The present application relates to the field of nuclear detection, and specifically to a correction method and device, an imaging method and system, an electronic device and a storage medium. Background Art

[0002] Positron Emission Computed Tomography (PET) is a relatively advanced clinical examination imaging technology in the field of nuclear medicine. PET is a non-invasive imaging method. Its principle is to provide direct imaging by using radioactive labeled molecules, such as fluorine-18-fluorodeoxyglucose (FDG) as a biomarker to evaluate glucose metabolism. The radioactive substances injected into the subject release positrons during the decay process. When the positrons encounter electrons, they will annihilate, thus generating a pair of photons with opposite directions and energies of 511 KeV. PET observes the physiological and biochemical changes of the subject by detecting a large number of gamma rays emitted by radioactive labeled molecules.

[0003] In PET imaging, when gamma photons generated by positron annihilation pass through the tissue of the object being measured, the different densities and properties of different tissues will cause a certain degree of energy attenuation of the gamma photons, thus affecting the quality of PET imaging. For example, this attenuation will cause inaccurate PET imaging and thus affect the results of medical tests. Therefore, in the process of PET image reconstruction, the PET raw data needs to be corrected according to the tissue structure of the object being measured, such as bones and skin.

[0004] Traditional PET attenuation correction solutions include PET / CT correction technology and PET / MRI correction technology. Both use dual-modality imaging methods to perform PET correction. PET / CT combines uncorrected PET detection data with X-ray computed tomography (CT) imaging, and uses the human tissue information obtained from CT scans to assist in PET correction; PET / MRI combines uncorrected PET detection data with magnetic resonance imaging (MRI) images, and uses the human tissue information obtained from MRI scans to assist in PET correction.

[0005] However, both of the above methods have their disadvantages. For example, in the PET / CT correction method, CT images are used as an auxiliary for PET correction. Although CT images can well distinguish between bone structures and air, skin and other tissues, on the one hand, they will cause a certain amount of radiation to patients during clinical use, especially for pediatric patients, who should avoid radiation examinations as much as possible. On the other hand, additional CT examinations lead to higher costs. For another example, in the PET / MRI correction method, MRI images are used as an auxiliary for PET correction. Although it will not cause radiation impact on patients, patients need to undergo both PET and magnetic resonance imaging scans, which not only greatly increases costs, but may also cause more discomfort to some patients.

[0006] In addition, for the PET attenuation problem, it is currently proposed to use machine learning or deep learning methods to correct it in the field of medical imaging, but the correction effect of these correction methods based on machine learning or deep learning needs to be improved. Moreover, some correction methods based on machine learning or deep learning still require additional examinations in addition to PET scans to obtain MRI images or CT images to assist in correction.

[0007] In view of this, there is an urgent need to provide a correction method and apparatus, an imaging method and system, a device and a storage medium so that PET correction does not require additional MRI or CT scanning.

[0008] The content of the background technology description is only for facilitating understanding of the relevant technology in this field and is not regarded as an admission of the prior art. Summary of the invention

[0009] Therefore, the present application intends to provide correction methods and devices, imaging methods and systems, electronic devices and storage media. These schemes can directly obtain simulated MRI images from uncorrected PET images, thereby being applied to the correction of PET detection data. After the PET test, the patient does not need to undergo additional MRI or CT scanning, which reduces the patient's discomfort and radiation dose to the patient. The detection method is safer for child patients, and since no additional scanning examinations are required, the cost of use is greatly reduced.

[0010] In a first aspect, a correction method is provided, comprising: inputting an uncorrected PET image into a trained correction network model to obtain a simulated MRI image, wherein the correction network model is based on a reversible generative network and is trained using PET training images and corresponding MRI label images; and using the simulated MRI image to correct uncorrected PET detection data corresponding to the uncorrected PET image to obtain corrected PET detection data.

[0011] In some embodiments, the correction network model is based on a reversible generative network and is trained using PET training images and corresponding MRI label images, including: acquiring multiple PET training images and corresponding multiple MRI label images; taking the multiple PET training images as input, the multiple MRI label images and the multiple PET training images as labels, and the simulated MRI images and the PET generated images as mutually inverse outputs, iteratively training the reversible generative network until a preset training completion condition is reached, and storing at least part of the reversible generative network models obtained by the iterative training of batches and / or generations; and determining the optimal model in the stored reversible generative network models as the correction network model.

[0012] In some embodiments, the reversible generation network includes a plurality of reversible modules connected in sequence, each reversible module being characterized by a reversible bijective function.

[0013] In some embodiments, each reversible module includes a reversible 1x1 convolution for shunt, multiple coupling layers located in a first shunt path, and multiple coupling layers located in a second shunt path, wherein the coupling layers of different shunt paths are interconnected through a transformation function, and the first and second shunt paths are spliced ​​with each other after passing through the coupling layer.

[0014] In some embodiments, the coupling layer is an affine coupling layer and is represented by the following formula:

[0015] n 1：d =m 1：d

[0016] n d+1：D =m d+1：D ⊙exp(s(m 1：d ))+t(m 1：d )

[0017] Where D is a given dimension, d is a natural number and d<D, m is the input, n is the output, s and t represent the scale transformation function and the translation transformation function respectively, and ⊙ represents the Hadamard product.

[0018] In some embodiments, the reversible generation network includes a first channel replication layer located forwardly upstream of the plurality of sequentially connected reversible modules and a second channel replication layer located reversely upstream of the plurality of sequentially connected reversible modules.

[0019] In some embodiments, the reversible generation network includes a first normalization layer located forward upstream of the plurality of sequentially connected reversible modules and a second normalization layer located reverse upstream of the plurality of sequentially connected reversible modules.

[0020] In some embodiments, each iteration includes: forward inputting the PET training image into the reversible generative network and outputting a corresponding simulated MRI image; reverse inputting the simulated MRI image into the reversible generative network and outputting a corresponding PET generated image; determining a forward loss based on the simulated MRI image and the MRI label image; determining a reverse loss based on the PET generated image and the PET training image; and updating the parameters of the reversible generative network based on the total loss determined by the forward loss and the reverse loss.

[0021] In some embodiments, determining the optimal model in the stored reversible generative network model as the correction network model includes: selecting multiple preferred models in the stored reversible generative network model based on a first judgment criterion; selecting the optimal model from the multiple preferred models based on a second judgment criterion, and the first judgment criterion is different from the second judgment criterion.

[0022] In some embodiments, the selecting of multiple preferred models from the stored reversible generative network models based on the first judgment criterion includes: determining a first similarity index between the simulated MRI image and the MRI label image corresponding to the stored reversible generative network model; and selecting the multiple preferred models based on the first similarity index.

[0023] In some embodiments, the selecting the optimal model from the multiple preferred models based on the second judgment criterion includes: performing a first correction on the uncorrected PET detection data corresponding to the uncorrected PET image using the simulated MRI images corresponding to the multiple preferred models to obtain a first corrected PET image; performing a second correction on the uncorrected PET detection data corresponding to the uncorrected PET image using the MRI label images corresponding to the multiple preferred models to obtain a second corrected PET image; determining a second similarity index between the first corrected PET image and the second corrected PET image corresponding to the multiple preferred models; and selecting the optimal model based on the second similarity index.

[0024] In some embodiments, the similarity metric includes one or more of peak signal-to-noise ratio, structural similarity, root mean square error, mean square error, or mean absolute error.

[0025] In some embodiments, the acquiring of multiple PET training images and corresponding multiple MRI label images includes: acquiring multiple original PET images; acquiring corresponding multiple original MRI images; registering the multiple original PET images with the multiple original MRI images; and based on the registration, adjusting the sizes of the multiple original MRI images to be consistent with those of the multiple original PET images, to obtain multiple MRI label images and multiple PET training images.

[0026] In some embodiments, the acquiring of multiple original PET images includes: performing a first detection on multiple objects to obtain first detection data; and reconstructing the first detection data to obtain the multiple original PET images; and the acquiring of the corresponding multiple original MRI images includes: performing a second detection on the multiple objects to obtain second detection data; and reconstructing the multiple second detection data to obtain the multiple original MRI images.

[0027] In some embodiments, the preset training completion condition is that a predetermined number of generations / batches / iterations or a total loss reaches stable convergence, or a combination of convergence and a predetermined number of generations / batches / iterations.

[0028] In some embodiments, the uncorrected PET detection data corresponding to the uncorrected PET image is corrected using the simulated MRI image to obtain corrected PET detection data, including: converting the simulated MRI image into an attenuation factor; and correcting the uncorrected PET detection data corresponding to the uncorrected PET image using the attenuation factor to obtain corrected PET detection data.

[0029] In some embodiments, the uncorrected PET detection data comprises the uncorrected PET sinogram.

[0030] According to a second aspect of the present application, an imaging method is provided, comprising: acquiring uncorrected PET detection data obtained by detecting a target object; performing a first image reconstruction on the uncorrected PET detection data to obtain a reconstructed uncorrected PET image; correcting the uncorrected PET detection data according to the correction method described in an embodiment of the present application to obtain corrected PET detection data; performing a second image reconstruction on the corrected PET detection data to obtain a corrected PET image as an imaging image.

[0031] According to the third aspect of the present application, a correction device is provided, comprising: a correction network model, configured to input an uncorrected PET image into a trained correction network model to obtain a simulated MRI image, wherein the correction network model is based on a reversible generative network and is trained using PET training images and corresponding MRI label images; and a correction unit, configured to use the simulated MRI image to correct uncorrected PET detection data corresponding to the uncorrected PET image to obtain corrected PET detection data.

[0032] In some embodiments, the correction unit is further configured to convert the simulated MRI image into an attenuation factor; and use the attenuation factor to correct the uncorrected PET detection data to obtain corrected PET detection data.

[0033] According to a fourth aspect of the present application, an imaging system is provided, comprising: an acquisition unit, configured to acquire uncorrected PET detection data obtained by detecting a target object; a first reconstruction unit, configured to perform a first image reconstruction on the uncorrected PET detection data to obtain a reconstructed uncorrected PET image; the correction device described in an embodiment of the present application is configured to input the uncorrected PET image into a trained correction network model to obtain a simulated MRI image; the correction network model is based on a reversible generative network and is trained using PET training images and corresponding MRI label images; the uncorrected PET detection data is corrected using the simulated MRI image to obtain corrected PET detection data; and a second reconstruction unit, configured to perform a second image reconstruction on the corrected PET detection data to obtain a corrected PET image as an imaging image.

[0034] In some embodiments, the acquisition unit is a PET detection device, which is configured to perform radiation detection on a target object to obtain the detection data; the PET detection device includes a plurality of scintillation crystal arrays, a plurality of photoelectric conversion units coupled to the plurality of scintillation crystal arrays, and a readout circuit module electrically connected to the plurality of photoelectric conversion units.

[0035] According to a fifth aspect of the present application, an electronic device is provided, comprising: a processor and a memory storing a computer program, wherein the processor is configured to execute any one of the methods described in the embodiments of the present application when running the computer program.

[0036] According to a sixth aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and the computer program is configured to execute any one of the methods in the embodiments of the present application when executed.

[0037] The solution proposed in the embodiment of the present application can directly obtain a simulated MRI image from an uncorrected PET image, and use the obtained simulated MRI image to correct the PET detection data, that is, use the simulated MRI image to correct the uncorrected PET detection data corresponding to the uncorrected PET image to obtain the corrected PET detection data. This does not require an additional MRI or CT scanning process, reduces the patient's discomfort and radiation dose to the patient, makes the detection method safer for child patients, and greatly reduces the cost of use because no additional scanning examination is required.

[0038] The optional features and other effects of the embodiments of the present application are partially described below, and partially can be understood by reading this document. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The embodiments of the present application are described in detail with reference to the accompanying drawings. The elements shown are not limited by the proportions shown in the accompanying drawings. The same or similar reference numerals in the accompanying drawings represent the same or similar elements, wherein:

[0040] Figure 1A A flow chart of a correction method according to an embodiment of the present application is shown.

[0041] Figure 1B A flow chart of a correction method according to an embodiment of the present application is shown.

[0042] Figure 2 A flowchart of a training method for a correction network model according to an embodiment of the present application is shown.

[0043] Figure 3 A flowchart of a training method for a correction network model according to an embodiment of the present application is shown.

[0044] Figure 4 A flow chart of an imaging method according to an embodiment of the present application is shown.

[0045] Figure 5 A schematic diagram of a correction device module according to an embodiment of the present application is shown.

[0046] Figure 6 A module schematic diagram of an imaging system according to an embodiment of the present application is shown.

[0047] Figure 7 A schematic diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific implementation methods and drawings. Here, the exemplary implementation methods of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0049] As used herein, the term "including" and its variations mean open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "based at least in part on". The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0050] The correction and training methods and related imaging schemes provided in the present application relate to radiation tomography technology, especially positron radiation tomography technology, and can be applied to various fields.

[0051] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0052] In an embodiment of the present application, the inventors intentionally designed a correction method based on a reversible generative network, using a deep learning method to directly obtain a simulated MRI image from an uncorrected PET image, thereby applying the simulated MRI image to the correction and image reconstruction of PET detection data.

[0053] In some embodiments, Figure 1A As shown, a calibration method 100 is provided, which may include:

[0054] S110: inputting the uncorrected PET image into a trained correction network model to obtain a simulated MRI image; the correction network model is based on a reversible generative network and is trained using PET training images and corresponding MRI label images;

[0055] S120: Using the simulated MRI image, correct the uncorrected PET detection data corresponding to the uncorrected PET image to obtain corrected PET detection data.

[0056] In some embodiments, Figure 1B As shown, step S120 includes:

[0057] S1210: converting the simulated MRI image into an attenuation factor;

[0058] S1220: Correcting the uncorrected PET detection data corresponding to the uncorrected PET image using the attenuation factor to obtain corrected PET detection data.

[0059] In some embodiments, the specific operations of steps S1210 and S1220 included in step S120 can refer to the prior art and will not be described in detail here.

[0060] In some embodiments, the uncorrected PET detection data is obtained based on detecting a target object, and the uncorrected PET detection data includes an uncorrected PET sinogram.

[0061] In some embodiments, the target object can be a patient (such as a human or an animal) or a prosthesis.

[0062] In some embodiments, the correction network model mentioned in step S110 is based on a reversible generative network and is trained using PET training images and corresponding MRI label images. In some specific embodiments, Figure 2 As shown, it may include steps S1110 to S1140:

[0063] S1110: Acquire multiple PET training images and corresponding multiple MRI label images;

[0064] S1120: Iteratively train the reversible generative network until a preset training completion condition is met;

[0065] S1130: Storing at least part of the reversible generative network model obtained by batch and / or epoch iterative training;

[0066] S1140: Determine an optimal model among the stored reversible generative network models as the correction network model.

[0067] In some specific embodiments, the step S1110 may include the following steps S1111 to S1114:

[0068] S1111: Acquire multiple original PET images.

[0069] In some embodiments, the step S1111 may include:

[0070] S11111: Performing a first detection on a plurality of objects to obtain first detection data;

[0071] S11112: Reconstruct the first detection data to obtain the multiple original PET images, that is, uncorrected PET images.

[0072] S1112: Acquire multiple corresponding original MRI images.

[0073] In some embodiments, step S1112 may include:

[0074] S11121: Perform a second detection on the multiple objects to obtain second detection data;

[0075] S11122: Reconstruct the multiple second detection data to obtain the multiple original MRI images.

[0076] S1113: Align the multiple original PET images with the multiple original MRI images.

[0077] S1114: Based on the registration, adjust the sizes of the multiple original MRI images to be consistent with the multiple original PET images, to obtain multiple MRI label images and multiple PET training images.

[0078] In step S110, a plurality of PET training images and a plurality of corresponding MRI label images are acquired. In other words, the acquired training data is data in the image domain.

[0079] As a specific example, the above data is, for example, the patient's brain data, but data from other parts of the patient (human or animal), or data from a prosthesis, or data obtained by simulation can be imagined.

[0080] The following description will be made using patient's brain data as a specific example.

[0081] For example, in sub-step S11111 of step S1111, PET detection is performed on multiple patients, such as 36 patients, to obtain first detection data, which is the uncorrected PET detection data mentioned in step S120, for example, including PET detection sinograms. Similarly, in sub-step S11121 of step S1112, MRI detection can be performed on these patients to obtain second detection data. In some embodiments, the first detection (PET detection) and the second detection (MRI detection) can be implemented in the same device, for example, in a PET / MRI device.

[0082] Further, for example, in sub-step S11112 of step S1111 and sub-step S11122 of step S1112, the first detection data (i.e., uncorrected PET detection data) such as the PET detection sinusoid and the second detection data (i.e., MRI detection data) can be reconstructed into uncorrected PET images and scanned MRI images using a reconstruction algorithm, respectively, and the scanned MRI image can be applied to step S120 to correct the first detection data. The above-mentioned reconstruction algorithm can use a known reconstruction algorithm, which is not described here, and they can be the same or different.

[0083] In step S1113, the multiple original PET images can be registered with the multiple original MRI images. In other words, the uncorrected PET images are registered with the MRI images one by one to ensure that each uncorrected PET image has an MRI image at its corresponding position. In some embodiments, the registration can be an absolute registration or a relative registration. In some specific embodiments, the registration can adopt one or more of the cross-correlation method, the inertial similarity detection matching method, the mutual information method, the image registration method based on the transform domain, and the image registration method based on the feature. The registration method can utilize various known methods, which will not be repeated here.

[0084] Further, in step S1114, the size of the multiple original MRI images is adjusted to be consistent with the size of the multiple original PET images (uncorrected PET images), so as to obtain multiple MRI label images and multiple PET training images. In some exemplary embodiments, the size of the PET image can be fixed, and the MRI image can be reduced to the same size as the uncorrected PET image. For example, a 512×512 MRI image can be reduced to a size of 250×250, which is the same as the uncorrected PET image. As an optional means, it is also possible to check whether there are any redundant parts in the image other than brain data that interfere with the training process, and if so, they can be appropriately cropped.

[0085] like Figure 2 As shown, in the iterative training of the above step S1120, the multiple PET training images are used as input, the multiple MRI label images and the multiple PET training images are used as labels, and the simulated MRI image and the PET generated image are used as the inverse output. Specifically, the forward process of the reversible network generates a synthetic image, and the reverse process restores the original image, thereby combining the forward loss with the reverse loss to jointly optimize the network model training result, as further described below.

[0086] In some embodiments, in combination Figure 3 Describe the iterative training process.

[0087] like Figure 3 As shown, each iteration of the above step S1120 may include the following steps until the preset training completion condition S1126 is reached:

[0088] S1121: forward inputting the PET training image into the reversible generative network, and outputting a corresponding simulated MRI image;

[0089] S1122: Reversely input the simulated MRI image into the reversible generative network, and output a corresponding PET generated image;

[0090] S1123: Determine a forward loss based on the simulated MRI image and the MRI label image;

[0091] S1124: Determine an inverse loss based on the PET generated image and the PET training image;

[0092] S1125: Based on the total loss determined by the forward loss and the reverse loss, update the parameters of the reversible generative network.

[0093] In some specific embodiments, the reversible generation network includes a plurality of sequentially connected reversible modules, also referred to as reversible blocks, each of which is composed of a reversible bijective function Characterization, the reversible structure of the reversible generation network can be expressed as where f k is a bijective function. In some specific examples, k=8, but other numbers of reversible modules (bijective functions) are conceivable.

[0094] In some embodiments, each reversible module may include a reversible 1x1 convolution for shunting, a plurality of coupling layers located in a first shunting path, and a plurality of coupling layers located in a second shunting path, wherein the coupling layers of different shunting paths are interconnected by a transformation function, and the first and second shunting paths are spliced ​​with each other after passing through the coupling layers. In some specific embodiments, the coupling layers in the first and second shunting paths can be interconnected to another shunting path through a transformation function r, a scale transformation function s, a translation transformation function t, etc.

[0095] In a specific example, the coupling layer may be an affine coupling layer (AffineCouplingLayer), and is represented by the following formula:

[0096] n 1：d =m 1：d

[0097] n d+1：D =m d+1：D ⊙exp(s(m 1：d ))+t(m 1：d )

[0098] Where D is a given dimension, d is a natural number and d<D, m is the input, n is the output, s and t represent the scale transformation function and the translation transformation function respectively, and ⊙ represents the Hadamard product.

[0099] Therefore, in the network structure designed in the embodiment of the present application, the reversibility of the network is achieved by introducing an affine coupling layer, and the reversible 1x1 convolution is used as a learnable permutation function between the affine coupling layers, thereby achieving the ability of mutually inverse generation of the reversible network structure.

[0100] In some embodiments, the affine coupling layer and the reversible 1x1 convolution can refer to Kingma DP, Dhariwal P. Glow: Generative Flow with Invertible 1x1 Convolutions:, 10.48550 / arXiv.1807.03039[P].2018., the entire text of which is incorporated herein by reference.

[0101] In an alternative example, the coupling layer may be an additive coupling layer (Addictive Coupling Layer), which falls within the scope of the present application.

[0102] In some embodiments, the reversible generation network includes a first channel replication layer located upstream of the plurality of sequentially connected reversible modules in the forward direction and a second channel replication layer located upstream of the plurality of sequentially connected reversible modules in the reverse direction. During training, the single-channel PET data and MRI data can be replicated in channels to achieve the effect of variable enhancement to improve the training effect. Continuing to refer to the above specific example, in the forward direction, a first channel replication layer can be set upstream of the reversible module (f1) for replicating uncorrected PET images; correspondingly, in the reverse direction, a second channel replication layer can be set upstream of the reversible module (f8) for replicating MRI images, such as simulated MRI images. In some embodiments, although the reversible module is reversible, the channel replication layer is only set upstream of the reversible module, so that, for example, in the forward direction, the PET training data can be processed by the first channel replication layer and the plurality of reversible modules without passing through the second channel replication layer, and vice versa. Although not shown, it is conceivable that a channel merging layer with the opposite function to the channel replication layer is set downstream of the reversible module, but it is also conceivable to replicate multiple labels for loss calculation, which all fall within the scope of the present application.

[0103] In some embodiments, the reversible generation network may further include a first normalization layer located upstream of the plurality of sequentially connected reversible modules and a second normalization layer located upstream of the plurality of sequentially connected reversible modules. As an explanation, since the original values ​​of the PET images and the MRI images are large, it may be necessary to normalize the values ​​of the PET images and the MRI images to a range of 0-1 during training to increase the matching degree of the data, so as to avoid loss explosion, failure to converge, etc. In some preferred embodiments, when the PET training images are batch PET training images, the normalization layer may be a batch normalization (BN) layer.

[0104] Continue to refer Figure 3In the network training phase, the PET training images can be input into the network for iterative learning, and the MRI label images (and PET training images) themselves can be used as labels. As previously mentioned, in some embodiments, the images are batched images, so that each iteration can perform training for a batch of images, and after all images (all batches) have completed the iteration, it can be called an epoch. For example, assuming there are 19,600 PET training images, such as uncorrected PET images, if there are 100 PET training images in each batch, then each iteration or each batch will execute these 100 PET training images in parallel; and when all 19,600 PET training images are executed once, that is, 196 batches are executed, it can be said that an epoch or an "era" of PET training images has been executed. In the embodiment of the present application, several generations of iterations can be performed as needed, such as 300 epochs. Although it is described that PET training images are performed in batches, it is conceivable that the PET training images are not batched, that is, one PET training image is performed per batch, which also falls within the scope of the present application and does not conflict with the execution of a generation (epoch) or an "era" of PET training images described in this article.

[0105] Continue to refer Figure 3 In each iteration of the above step S1120, before reaching the preset training completion condition S1126, the PET training image can be forward input into the reversible generative network in step S1121 to output the corresponding simulated MRI image, and the simulated MRI image can be reversely input into the reversible generative network in step S1122 to output the corresponding PET generated image.

[0106] Further, in step S1123, a forward loss L1 may be determined based on the simulated MRI image and the MRI label image; and in step S1124, a reverse loss L2 may be determined based on the PET generated image and the PET training image. Thus, in step S1125, a total loss L1 may be determined based on the forward loss and the reverse loss. total , update the parameters of the reversible generative network.

[0107] As mentioned above, in the forward process, the MRI labeled images will be used as labels, also known as ground truth images, while in the reverse process, the PET training images themselves will be used as labels.

[0108] The total loss L determined based on the forward loss and reverse loss total It can be the forward loss L l The weighted sum of the inverse loss L2 is as follows:

[0109] L total =λL1+L2=λ||f(X)-Y||2+||f -1 (Y)-X||2

[0110] Where Y is the MRI label image, f(X) is the output image obtained from the source image X in the network, ||·|| is the 2-norm, L1 represents the loss between the PET generated image and the MRI label image, that is, the forward loss, L2 represents the loss between the input image and the PET training image, that is, the reverse loss, and the hyperparameter λ is the weight used to balance the forward loss and the reverse loss.

[0111] Thus, during training, by calculating the loss and, for example, by using a gradient descent algorithm, the parameters of the reversible generative network can be iteratively updated until the preset training completion condition S1126 is reached.

[0112] In an embodiment of the present application, the preset training completion condition 1126 may have a variety of implementation forms. For example, the preset training completion condition 1126 may be a predetermined number of epochs, for example, the preset training completion condition 1126 is to reach 300 epochs. Alternatively, the preset training completion condition 1126 may be that the total loss reaches stable convergence. Alternatively, the preset training completion condition 1126 may be a combination of convergence and a predetermined number of epochs / batches / iterations, for example, the preset training completion condition 1126 may be that the total loss reaches stable convergence for a given number of epochs, batches, or iterations, for example, the total loss reaches stable convergence for 1 generation or 64 batches or 64 iterations, etc., all of which fall within the scope of the present application.

[0113] Continue to refer Figure 2 In step S1140, the reversible generative network model may be stored after at least a portion of batches / iterations and / or at least a portion of epochs of training, such as the parameters of the reversible generative network updated at this time.

[0114] In some embodiments, a network model can be saved each time all PET training images are trained, that is, the network model after each epoch training is stored. For example, when training 300 epochs, 300 network models after each epoch training are stored.

[0115] In some embodiments, only part of the network model may be stored, for example, the network model after 100 generations of training before completing the iteration may be stored.

[0116] In some embodiments, a model that combines generations and batches (or iterations) can be stored. For example, for a PET training image with 196 batches per epoch, the network models trained from all batches (iterations) of the last 10 generations can be stored, i.e., 10×196 network models.

[0117] Different embodiments of storing the network model can be combined with different embodiments of presetting the training completion conditions to obtain various different embodiments, which fall within the scope of the present application.

[0118] In some embodiments, the process may also include selecting an optimal model from the stored network models.

[0119] In some embodiments, step S1140 may include:

[0120] S1141: Selecting a plurality of preferred models from the stored reversible generative network models based on a first judgment criterion;

[0121] S1142: Select the optimal model from the multiple preferred models based on a second judgment criterion.

[0122] In the embodiment of the present application, the first judgment criterion is different from the second judgment criterion.

[0123] In some embodiments, the judgment criterion may be an index of similarity of the images.

[0124] In a further embodiment, the first judgment criterion may be a judgment of a similarity index of the MRI images. In some examples, step S1141 may include:

[0125] S11411: Determine a first similarity index between the simulated MRI image and the MRI label image corresponding to the stored reversible generative network model;

[0126] S11412: Based on the first similarity index, select the multiple preferred models.

[0127] In a further embodiment, the first judgment criterion may be a judgment of a similarity index of a PET image after correction processing of the MRI image. In some examples, step S1142 may include:

[0128] S11421: performing a first correction on the uncorrected PET detection data corresponding to the uncorrected PET image using the simulated MRI images corresponding to the multiple preferred models to obtain a first corrected PET image;

[0129] S11422: performing a second correction on the uncorrected PET detection data corresponding to the uncorrected PET image using the MRI label images corresponding to the multiple preferred models to obtain a second corrected PET image;

[0130] S11423: Determine a second similarity index between the first corrected PET image and the second corrected PET image corresponding to the plurality of preferred models;

[0131] S11424: Based on the second similarity index, select the optimal model.

[0132] In some embodiments, regarding performing a second correction on the uncorrected PET detection data corresponding to the uncorrected PET image using the MRI label images corresponding to the multiple preferred models to obtain a second corrected PET image, specific details can be referred to the prior art and will not be elaborated herein.

[0133] In an embodiment of the present application, the similarity index includes one or more of peak signal-to-noise ratio (PSNR), structural similarity (SSIM), root mean square error (RMSE), mean square error (MSE) or mean absolute error (MAE).

[0134] Specifically, for example, the PET image of the verification data is input into each stored network model to obtain a simulated MRI image. Next, the simulated MRI image can be compared with the original MRI image (verification image) obtained by the device scan and the index is calculated to preliminarily find a better network model. Then, the simulated MRI image and the original MRI image (verification image) are used for PET correction to compare the similarity between the PET images obtained in this way.

[0135] In some embodiments, the peak signal-to-noise ratio (PSNR) may be determined according to the following formula:

[0136]

[0137] Among them, y, The two images used for comparison are respectively represented, depending on whether the comparison is an MRI image or a PET image.

[0138] In some embodiments, structural similarity (SSIM) can be determined according to the following formula:

[0139]

[0140] Among them, μ represents the mean, represents the standard deviation of y, express The standard deviation of Indicates y, The covariance of , and the rest are constants.

[0141] In some embodiments, the expressions of root mean square error (RMSE), mean square error (MSE) or mean absolute error (MAE) are conventional and will not be described in detail.

[0142] Based on the above content, it can be found that the correction network structure and the related correction method and imaging method of the embodiment of the present application have the following excellent effects:

[0143] (1) Compared with the known solution of the inventors that directly obtains the corrected PET image by inputting the uncorrected PET image into the trained deep learning or neural network, the solution of the embodiment of the present application makes full use of the tissue data in the simulated MRI image to correct the PET image, obtains a better correction effect, and avoids the overfitting phenomenon that may occur in the known solution of directly generating the corrected PET image. Moreover, for medical staff, there is no need to change their clinical skills or habits, and they can intuitively experience the MRI image data with important practical significance in medical applications.

[0144] (2) Compared with the known solution of applying the scanned CT images or MRI images to PET correction, the embodiment of the present application saves the additional CT or MRI scanning process during correction, reduces the patient's discomfort and radiation dose to the patient, especially for children, no additional CT examination is required, which reduces the radiation dose, is safer, and greatly reduces the cost of use. Moreover, it is important that although the MRI images generated by the solution of the embodiment of the present application do not seem to be "excellent", the images corrected by PET using these seemingly "not so excellent" simulated MRI images show excellent performance that is not inferior to that using real MRI scan images.

[0145] (3) The embodiment of the present application uses a reversible generative network in the correction model, and utilizes mutually inverse outputs and corresponding losses to train the network. Compared with known solutions, the training speed is extremely fast and problems such as overfitting are effectively avoided.

[0146] (4) The solution of the embodiment of the present application is not to directly use the trained correction model, but to store multiple reversible generative network models, and to select the best one from multiple stored models by simulating the image similarity of the MRI image itself and the image similarity index of the correction image. This can further reduce the number of training iterations / batches / rounds, obtain an excellent network model, and further avoid problems such as overfitting.

[0147] Accordingly, in Figure 1A and Figure 1B Based on the correction method embodiment shown, the embodiment of the present application also relates to an imaging method.

[0148] like Figure 4 As shown, an imaging method 400 is provided, which may include:

[0149] S410: Acquire uncorrected PET detection data obtained by detecting the target object;

[0150] S420: Performing a first image reconstruction on the uncorrected PET detection data to obtain a reconstructed uncorrected PET image;

[0151] S430: Exploitation Figure 1A and Figure 1B The correction method shown in the embodiment corrects the uncorrected PET detection data to obtain corrected PET detection data;

[0152] S440: Perform a second image reconstruction on the corrected PET detection data to obtain a corrected PET image as an imaging image.

[0153] In some embodiments, the first image reconstruction performed on the uncorrected PET detection data and the second image reconstruction performed on the corrected PET detection data may be the same or different image reconstructions. The specific reconstruction method may refer to the prior art and will not be described in detail herein.

[0154] In this embodiment, the uncorrected PET detection data may include an uncorrected PET sinogram. Accordingly, the first image reconstruction may be performed by reconstructing the uncorrected PET sinogram to obtain an uncorrected PET image. Accordingly, the uncorrected PET detection data may be corrected by using a simulated MRI image, thereby obtaining corrected PET detection data, and the corrected PET detection data may include a corrected PET sinogram. Accordingly, the second image reconstruction may be performed by reconstructing the corrected PET sinogram to obtain a corrected PET image.

[0155] Numerous embodiments of the correction method, the imaging method, etc. are described above. In these embodiments, the mapping between the uncorrected PET image and the simulated MRI image in the correction model is implemented in the image domain.

[0156] It will be understood that, unless otherwise specified, the features of the methods described in different embodiments can be combined in a non-contradictory manner into other embodiments to obtain new embodiments.

[0157] The methods and network models described in some embodiments of the present application can be implemented in, for example, Pytorch, and experiments can be performed on, for example, a computer equipped with an Intel Core i9-7900 CPU and a GeForce Titan XP. However, it is conceivable to implement the relevant methods and architectures in other machine learning platforms, and it is also conceivable to implement the required computing power in other structures, such as the electronic devices described in other parts of this article.

[0158] In the embodiments of the present application, Figure 5 As shown, a correction device 500 is also provided, which may include: a correction network model 510, configured to input an uncorrected PET image into a trained correction network model to obtain a simulated MRI image; the correction network model is based on a reversible generative network and is trained using PET training images and corresponding MRI label images; a correction unit 520, configured to use the simulated MRI image to correct the uncorrected PET detection data corresponding to the uncorrected PET image to obtain corrected PET detection data.

[0159] In some embodiments, the correction unit is further configured to convert the simulated MRI image into an attenuation factor; and use the attenuation factor to correct the uncorrected PET detection data to obtain corrected PET detection data.

[0160] In some embodiments of the present application, Figure 6 As shown, an imaging system 600 is also provided, which may include: an acquisition unit 610, configured to acquire uncorrected PET detection data obtained by detecting a target object; a first reconstruction unit 620, configured to perform a first image reconstruction on the uncorrected PET detection data to obtain a reconstructed uncorrected PET image; a correction device 630, configured to input the uncorrected PET image into a trained correction network model to obtain a simulated MRI image; the correction network model is based on a reversible generative network and is trained using PET training images and corresponding MRI label images; the uncorrected PET detection data is corrected using the simulated MRI image to obtain corrected PET detection data; a second reconstruction unit 640, configured to perform image reconstruction on the corrected PET detection data to obtain a corrected PET image as an imaging image.

[0161] Preferably, the acquisition unit is a PET detection device, which is configured to perform radiation detection on the target object to obtain the detection data. Although not shown, the PET detection device can optionally include a plurality of scintillation crystal arrays, a plurality of photoelectric conversion units coupled to the plurality of scintillation crystal arrays, and a readout circuit module electrically connected to the plurality of photoelectric conversion units.

[0162] Those skilled in the art will understand that the devices and systems described in the embodiments of the present application can be combined with the method features described in the embodiments of the present application, and vice versa.

[0163] The embodiments of the present application also relate to electronic devices capable of implementing the methods of the embodiments of the present application.

[0164] In some embodiments, an electronic device is provided, which may include a processor and a memory storing a computer program, wherein the processor is configured to execute the method of any embodiment of the present application when running the computer program.

[0165] Figure 7 A schematic diagram of an exemplary electronic device that can implement the method of the embodiment of the present application is shown, which may include more or fewer electronic devices than shown in the figure in some embodiments. In some embodiments, it can be implemented using a single or multiple electronic devices. In some embodiments, it can be implemented using cloud or distributed electronic devices.

[0166] like Figure 7 As shown, the electronic device 2000 includes a processor 2001, which can perform various appropriate operations and processes according to the programs and / or data stored in the read-only memory (ROM) 2002 or the programs and / or data loaded from the storage part 2008 to the random access memory (RAM) 2003. Processor 2001 can be a single-core or multi-core processor, or it can include multiple processors. In some embodiments, processor 2001 may include a general-purpose main processor (such as a CPU) and one or more special coprocessors, such as a graphics processing unit (GPU), a neural network processor (NPU), a digital signal processor (DSP) or other general or application-specific integrated circuits. In RAM 2003, various programs and data required for the operation of the electronic device 2000 are also stored. Processor 2001, ROM 2002 and RAM 2003 are connected to each other via bus 2004. Input / output (I / O) interface 2005 is also connected to bus 2004.

[0167] The processor and the memory are used together to execute the program stored in the memory. When the program is executed by the computer, the steps or functions of the methods described in the above embodiments can be implemented.

[0168] The following components are connected to the I / O interface 2005: an input section 2006 including a keyboard, a mouse, etc.; an output section 2007 including a display and a speaker, etc.; a storage section 2008 including a hard disk, etc.; and a communication section 2009 including a network interface card such as a LAN card, a modem, etc. The communication section 2009 performs communication processing via a network such as the Internet. A drive 2010 is also connected to the I / O interface 2005 as needed. A removable medium 2011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 2010 as needed, so that a computer program read therefrom is installed into the storage section 2008 as needed.

[0169] Figure 7 Only schematically illustrated are schematic electronic devices, but the electronic devices according to the embodiments of the present application may include Figure 7 The electronic device shown may have more or fewer components or may have Figure 7 The devices of the illustrated embodiments may have the same, partially the same or different architectures.

[0170] In some embodiments, the electronic device may be combined with various components to obtain a method, device, and system having the advantages of the present invention.

[0171] Although not shown, in some embodiments, a computer-readable storage medium is also provided, storing a computer program, which is configured to execute any method of the present application embodiment when it is run. The computer program includes various program modules / units constituting the device according to the embodiment of the present application, and the computer program composed of various program modules / units can realize the functions corresponding to the various steps in the method described in the above embodiment when it is executed. The computer program can also be run on an electronic device as described in the embodiment of the present application.

[0172] The storage medium of the embodiment of the present application includes non-volatile and / or volatile articles that can be used to store information by any method or technology. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0173] Those skilled in the art should understand that the embodiments of this specification can be implemented in various forms such as methods, systems or computer program products. Therefore, those skilled in the art can imagine that the implementation of the functional modules / units or controllers and related method steps described in the above embodiments can be implemented in software, hardware or a combination of software / hardware.

[0174] Unless explicitly stated, the actions or steps of the methods, programs, and embodiments of the present application do not have to be performed in a specific order and can still achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0175] In this article, multiple embodiments are described, but for the sake of simplicity, the description of each embodiment is not exhaustive, and the same or similar features or parts between the embodiments may be omitted. In this article, "one embodiment", "some embodiments", "example", "specific example", or "some examples" are meant to be applicable to at least one embodiment or example according to the present application, rather than all embodiments. The above terms do not necessarily mean to refer to the same embodiment or example. In the absence of contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples.

[0176] The exemplary systems and methods of the present application have been specifically shown and described with reference to the above embodiments, which are merely examples of the best modes for implementing the present systems and methods. It will be appreciated by those skilled in the art that various changes may be made to the embodiments of the systems and methods described herein when implementing the present systems and / or methods without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A calibration method, characterized in that: Including: Inputting the uncorrected PET image into the trained correction network model to obtain a simulated MRI image, where the correction network model is trained based on a reversible generative network using PET training images and corresponding MRI label images; Correcting the uncorrected PET detection data corresponding to the uncorrected PET image using the simulated MRI image to obtain corrected PET detection data.

2. The calibration method according to claim 1, characterized in that: The correction network model is trained based on a reversible generative network using PET training images and corresponding MRI label images, and includes: Obtaining a plurality of PET training images and corresponding plurality of MRI label images; Taking the plurality of PET training images as input, the plurality of MRI label images and the plurality of PET training images as labels, and the simulated MRI image and the PET generated image as inverse outputs, and iteratively training the reversible generative network until a preset training completion condition is reached, and storing at least part of the reversible generative network models obtained by batch and / or generation iteration training; Determining the optimal model among the stored reversible generative network models as the correction network model.

3. The calibration method according to claim 2, characterized in that: The reversible generative network includes a plurality of sequentially connected reversible modules, and each reversible module is characterized by a reversible bijective function.

4. The calibration method according to claim 2, characterized in that: Each reversible module includes a reversible 1x1 convolution for shunting, a plurality of coupling layers in the first shunt path, and a plurality of coupling layers in the second shunt path, where the coupling layers in different shunt paths are interconnected by a transformation function, and the first and second shunt paths are spliced with each other after passing through the coupling layers.

5. The calibration method according to claim 4, characterized in that: The coupling layer is an affine coupling layer and is represented by the following formula: n 1:d =m 1:d n d+1:D =m d+1:D ⊙exp(s(m 1:d ))+t(m 1:d ) where D is a given dimension, d is a natural number and d < D, m is the input, n is the output, s and t respectively represent a scale transformation function and a translation transformation function, and ⊙ represents a Hadamard product.

6. The calibration method according to claim 3, characterized in that: The reversible generative network includes a first channel replication layer upstream in the forward direction of the plurality of sequentially connected reversible modules and a second channel replication layer upstream in the reverse direction of the plurality of sequentially connected reversible modules.

7. The calibration method according to claim 3, characterized in that: The reversible generative network includes a first normalization layer upstream in the forward direction of the plurality of sequentially connected reversible modules and a second normalization layer upstream in the reverse direction of the plurality of sequentially connected reversible modules.

8. The calibration method according to any one of claims 2 to 7, characterized in that: Each iteration includes: Forwardly inputting the PET training image into the reversible generative network to output the corresponding simulated MRI image; Reversely inputting the simulated MRI image into the reversible generative network to output the corresponding PET generated image; Determining a forward loss based on the simulated MRI image and the MRI label image; Determining a reverse loss based on the PET generated image and the PET training image; Updating the parameters of the reversible generative network based on the total loss determined by the forward loss and the reverse loss.

9. The calibration method according to any one of claims 2 to 7, characterized in that: The determining the optimal model among the stored reversible generative network models as the correction network model includes: Selecting a plurality of preferred models among the stored reversible generative network models based on a first judgment criterion; Selecting the optimal model among the plurality of preferred models based on a second judgment criterion, where the first judgment criterion is different from the second judgment criterion.

10. The calibration method according to claim 9, characterized in that: The selecting a plurality of preferred models from the stored reversible generative network models based on the first judgment criterion comprises: Determining a first similarity index between the simulated MRI image and the MRI label image corresponding to the stored reversible generative network model; Based on the first similarity index, the multiple preferred models are selected.

11. The calibration method according to claim 9, characterized in that: The selecting the optimal model from the multiple preferred models based on the second judgment criterion includes: Performing a first correction on the uncorrected PET detection data corresponding to the uncorrected PET image using the simulated MRI images corresponding to the multiple preferred models to obtain a first corrected PET image; Performing a second correction on the uncorrected PET detection data corresponding to the uncorrected PET image using the MRI label images corresponding to the multiple preferred models to obtain a second corrected PET image; Determine a second similarity index between the first corrected PET image and the second corrected PET image corresponding to the plurality of preferred models; Based on the second similarity indicator, the optimal model is selected.

12. The calibration method according to claim 10 or 11, characterized in that: The similarity index includes one or more of peak signal-to-noise ratio, structural similarity, root mean square error, mean square error or mean absolute error.

13. The calibration method according to any one of claims 2 to 7, characterized in that: The step of acquiring a plurality of PET training images and a plurality of corresponding MRI label images includes: Acquire multiple raw PET images; Acquire a corresponding plurality of original MRI images; registering the plurality of original PET images with the plurality of original MRI images; Based on the registration, the sizes of the multiple original MRI images are adjusted to be consistent with the sizes of the multiple original PET images, so as to obtain multiple MRI label images and multiple PET training images.

14. The calibration method according to claim 13, characterized in that: The acquiring of multiple original PET images comprises: Performing a first detection on a plurality of objects to obtain first detection data; Reconstructing the first detection data to obtain the multiple original PET images; The obtaining of the corresponding multiple original MRI images comprises: performing a second detection on the plurality of objects to obtain second detection data; The plurality of second detection data are reconstructed to obtain the plurality of original MRI images.

15. The calibration method according to claim 2, characterized in that: The preset training completion condition is that a predetermined number of generations / batches / iterations or a total loss reaches stable convergence or a combination of convergence and a predetermined number of generations / batches / iterations.

16. The calibration method according to claim 1, characterized in that: Correcting the uncorrected PET detection data corresponding to the uncorrected PET image using the simulated MRI image to obtain corrected PET detection data includes: Converting the simulated MRI image into an attenuation factor; The uncorrected PET detection data corresponding to the uncorrected PET image is corrected using the attenuation factor to obtain corrected PET detection data.

17. The calibration method according to claim 1, characterized in that: The uncorrected PET detection data includes the uncorrected PET sinogram.

18. An imaging method, characterized in that: include: Acquiring uncorrected PET detection data obtained by detecting the target object; Performing a first image reconstruction on the uncorrected PET detection data to obtain a reconstructed uncorrected PET image; Correcting the uncorrected PET detection data according to the correction method according to any one of claims 1 to 17 to obtain corrected PET detection data; A second image reconstruction is performed on the corrected PET detection data to obtain a corrected PET image as an imaging image.

19. A calibration device, characterized in that: include: The correction network model is configured to input the uncorrected PET image into the trained correction network model to obtain a simulated MRI image, wherein the correction network model is based on a reversible generative network and is trained using the PET training image and the corresponding MRI label image; The correction unit is configured to correct the uncorrected PET detection data corresponding to the uncorrected PET image using the simulated MRI image to obtain corrected PET detection data.

20. The calibration device according to claim 19, characterized in that: The correction unit is further configured to convert the simulated MRI image into an attenuation factor; The uncorrected PET detection data is corrected using the attenuation factor to obtain corrected PET detection data.

21. An imaging system, characterized in that: include: an acquisition unit configured to acquire uncorrected PET detection data obtained by detecting the target object; a first reconstruction unit configured to perform a first image reconstruction on the uncorrected PET detection data to obtain a reconstructed uncorrected PET image; The correction device of claim 19 or 20 is configured to input the uncorrected PET image into a trained correction network model to obtain a simulated MRI image; the correction network model is based on a reversible generative network and is trained using PET training images and corresponding MRI label images; the uncorrected PET detection data is corrected using the simulated MRI image to obtain corrected PET detection data; The second reconstruction unit is configured to perform a second image reconstruction on the corrected PET detection data to obtain a corrected PET image as an imaging image.

22. The imaging system according to claim 21, characterized in that The acquisition unit is a PET detection device, which is configured to perform radiation detection on a target object to obtain the detection data; the PET detection device includes a plurality of scintillation crystal arrays, a plurality of photoelectric conversion units coupled to the plurality of scintillation crystal arrays, and a readout circuit module electrically connected to the plurality of photoelectric conversion units.

23. An electronic device, characterized in that: include: A processor and a memory storing a computer program, wherein the processor is configured to perform the method according to any one of claims 1 to 18 when running the computer program.

24. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which is configured to perform the method of any one of claims 1 to 18 when executed.

Images