Data correction model training and correction method for photon counting spectral CT

Abstract

The application provides a data correction model training and correction method for photon counting spectral CT, comprising the following steps: training a first neural network model based on a plurality of first multi-channel distorted data and a plurality of corresponding first single-channel reference data under a single set of physical parameter conditions, so that the model is trained as a first photon energy data correction model for outputting first single-channel corrected data corresponding to the first multi-channel distorted data and the corresponding first single-channel reference data; training a second neural network model based on a plurality of second multi-channel distorted data and a plurality of corresponding second single-channel reference data under a plurality of sets of physical parameter conditions and a plurality of sets of physical parameters, so that the model is trained as a second photon energy data correction model for outputting second single-channel corrected data corresponding to the second multi-channel distorted data and the corresponding second single-channel reference data and the physical parameters, and the second neural network model is formed by adding a parameter fusion network to the first neural network model in advance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photon counting spectral CT imaging, and particularly relates to a photon energy data correction model training method, a correction method and an image reconstruction method for photon counting spectral CT. BACKGROUND

[0002] Computed tomography (CT) is the basis of radiological diagnosis and the key to major surgery and important medical decisions. CT mainly uses high-penetration X-ray beams to pass through a specific cross-section of the patient's body and is received by a detector. According to the different absorption degrees of different tissues to X-rays, a specific algorithm is used to reconstruct the cross-sectional information of the body to obtain a tomographic image. In the CT system, the detector technology is one of the most critical technologies, and the detector array is responsible for converting the received X-ray signals into electrical signals. The traditional energy integrating detector (EID) uses a scintillation crystal to absorb X-ray photons and emit visible light of corresponding intensity, which is detected by a photodiode and converted into an electrical signal. During the conversion process, the energy of all X-rays is accumulated, and the detector cannot collect information on the energy of individual photons. Therefore, the EID has certain defects, such as low energy resolution, image quality is greatly affected by electronic noise and scattered noise, and is prone to artifacts, etc.

[0003] With the continuous development of computer tomography technology, direct photon counting detectors (PCD) have received extensive attention and application. Unlike energy integrating detectors, photon counting detectors directly convert X-ray photons into electron-hole pairs, and under the drive of a bias voltage, electrons move to the anode and are absorbed by the anode, so that the number of X-ray photons can be directly counted, and the pulse height produced by the charge deposition on the anode is proportional to the absorbed photon energy. By setting a pulse detection threshold, low-energy noise can be effectively filtered out, and the signal-to-noise ratio can be improved. However, photon counting detectors also have their inherent defects. When X-ray photons fall on the edge of the detector pixel, the charge cloud generated may be recorded by the adjacent pixel, each pixel records a part of the actual photon energy, resulting in errors in the recording of photon energy, which is called the charge sharing effect. This will cause the measured X-ray spectrum to be distorted, which in turn affects the final tomographic reconstruction image. SUMMARY

[0004] In view of this, the present application provides a photon energy data correction model training method, a correction method and an image reconstruction method for photon counting spectral CT to eliminate or improve one or more defects in the prior art.

[0005] A first aspect of the present application provides a photon energy data correction model training method for photon counting spectral CT, the method comprising the following steps:

[0006] training the first neural network model based on the plurality of first multi-channel photon energy distortion data and the corresponding plurality of first single-channel photon energy reference data under a single set of physical parameter conditions, so as to train the first neural network model into a first photon energy data correction model for outputting corresponding first single-channel photon energy correction data for photon counting spectral CT according to the first multi-channel photon energy distortion data and the corresponding first single-channel photon energy reference data; and

[0007] training the second neural network model based on the plurality of second multi-channel photon energy distortion data and the corresponding plurality of second single-channel photon energy reference data under a plurality of sets of physical parameter conditions and the plurality of sets of physical parameters, so as to train the second neural network model into a second photon energy data correction model for outputting corresponding second single-channel photon energy correction data for photon counting spectral CT according to the second multi-channel photon energy distortion data and the corresponding second single-channel photon energy reference data and the physical parameters, wherein the second neural network model is formed by adding a parameter fusion network to the first neural network model in advance.

[0008] In some embodiments of the present application, the multi-channel photon energy distortion data is formed by combining the photon energy distortion data of a central pixel channel and the photon energy distortion data of a plurality of channels near the central pixel channel generated under a single set or a plurality of sets of physical parameter conditions.

[0009] In some embodiments of the present application, the first neural network model comprises a first feature extraction network, a three-dimensional filtering network, a dimension reduction processing network, a two-dimensional filtering network and a second feature extraction network.

[0010] The first feature extraction network comprises a plurality of feature extraction networks connected in sequence, for extracting energy features in the input first multi-channel photon energy distortion data, and outputting corresponding first multi-channel photon energy features, wherein each feature extraction network comprises a three-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence.

[0011] The three-dimensional filtering network comprises a plurality of residual blocks and a first channel injection layer connected in sequence, for filtering and denoising the extracted first multi-channel photon energy features, wherein each residual block comprises a three-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence, and the first channel injection layer comprises a pooling layer and a plurality of sets of three-dimensional convolution layers and nonlinear activation functions connected in sequence, and along the direction from input to output of the three-dimensional filtering network, the input of the first three-dimensional convolution layer in the first residual block and the output of the last nonlinear activation function in the first channel injection layer are added to serve as the output of the three-dimensional filtering network.

[0012] The dimension reduction processing network comprises a three-dimensional convolution layer, a normalization layer, a nonlinear activation function and a dimension reduction layer connected in sequence, and is used for reducing the channel dimension of the first multi-channel photon energy feature after filtering and denoising, and outputting a first single-channel photon energy feature;

[0013] The two-dimensional filtering network comprises a plurality of residual blocks and a second channel injection layer connected in sequence, and is used for filtering and denoising the output first single-channel photon energy feature, wherein each residual block comprises a two-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence, and the second channel injection layer comprises a pooling layer and a plurality of groups of two-dimensional convolution layers and nonlinear activation functions connected in sequence, and the input of the first two-dimensional convolution layer in the first residual block and the output of the last nonlinear activation function in the second channel injection layer are added and taken as the output of the two-dimensional filtering network along the direction from the input to the output of the two-dimensional filtering network.

[0014] The second feature extraction network comprises a plurality of two-dimensional convolution layers connected in sequence, and is used for extracting deeper energy features in the filtered and denoised first single-channel photon energy feature, and outputting first single-channel photon energy correction data.

[0015] In some embodiments of the present application, the second neural network model comprises a first feature extraction network, a first parameter fusion network, a three-dimensional filtering network, a second parameter fusion network, a dimension reduction processing network, a third parameter fusion network, a two-dimensional filtering network, a fourth parameter fusion network, a second feature extraction network and a fifth parameter fusion network, wherein each parameter fusion network comprises a nonlinear activation function and a fully connected layer connected in sequence.

[0016] The first feature extraction network comprises a plurality of feature extraction networks connected in sequence, and is used for extracting energy features in the input second multi-channel photon energy distortion data, and outputting second multi-channel photon energy features, wherein each feature extraction network comprises a three-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence.

[0017] The first parameter fusion network is used for introducing a plurality of groups of physical parameters, fusing the second multi-channel photon energy features extracted by the first feature extraction network and the corresponding group of physical parameters, and outputting a first multi-channel fusion result.

[0018] The three-dimensional filtering network comprises a plurality of residual blocks and a first channel injection layer connected in sequence, and is configured to filter and denoise a first multi-channel fusion result output, wherein each residual block comprises a three-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence, the first channel injection layer comprises a pooling layer and a plurality of groups of three-dimensional convolution layers and nonlinear activation functions connected in sequence, and an input of a first three-dimensional convolution layer in a first residual block and an output of a last nonlinear activation function in the first channel injection layer are added together and taken as an output of the three-dimensional filtering network along a direction from an input to an output of the three-dimensional filtering network.

[0019] The second parameter fusion network is configured to introduce a plurality of groups of physical parameters, fuse a first multi-channel fusion result output by the three-dimensional filtering network after filtering and denoising and a corresponding group of physical parameters, and output a second multi-channel fusion result.

[0020] The dimension reduction processing network comprises a three-dimensional convolution layer, a normalization layer, a nonlinear activation function and a dimension reduction layer connected in sequence, and is configured to reduce a channel dimension of a second multi-channel fusion result output, and output a first single-channel fusion result.

[0021] The third parameter fusion network is configured to introduce a plurality of groups of physical parameters, fuse a first single-channel fusion result output by the dimension reduction processing network and a corresponding group of physical parameters, and output a second single-channel fusion result.

[0022] The two-dimensional filtering network comprises a plurality of residual blocks and a second channel injection layer connected in sequence, and is configured to filter and denoise a second single-channel fusion result output, wherein each residual block comprises a two-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence, the second channel injection layer comprises a pooling layer and a plurality of groups of two-dimensional convolution layers and nonlinear activation functions connected in sequence, and an input of a first two-dimensional convolution layer in a first residual block and an output of a last nonlinear activation function in the second channel injection layer are added together and taken as an output of the two-dimensional filtering network along a direction from an input to an output of the two-dimensional filtering network.

[0023] The fourth parameter fusion network is configured to introduce a plurality of groups of physical parameters, fuse a second single-channel fusion result output by the two-dimensional filtering network after filtering and denoising and a corresponding group of physical parameters, and output a third single-channel fusion result.

[0024] The second feature extraction network comprises a plurality of two-dimensional convolution layers connected in sequence, and is configured to extract deeper energy features in a third single-channel fusion result output.

[0025] The fifth parameter fusion network is used to introduce multiple groups of physical parameters, and fuse the deeper energy features extracted by the second feature extraction network and the corresponding group of physical parameters, to output second single-channel photon energy correction data.

[0026] In some embodiments of the present application, the physical parameters include part or all of the following: thickness of the photon counting detector, pixel, bias voltage, and number of photons emitted by the X-ray emission end.

[0027] In some embodiments of the present application, before training the second neural network model, the method further comprises: encoding the multiple groups of physical parameters by position encoding.

[0028] The second aspect of the present application provides a photon energy data correction method for photon counting spectral CT, which comprises the following steps:

[0029] Obtaining target multi-channel photon energy distortion data;

[0030] If the target multi-channel photon energy distortion data is obtained under the condition of a single group of physical parameters, inputting the target multi-channel photon energy distortion data into a first photon energy data correction model, so that the first photon energy data correction model outputs corresponding target single-channel photon energy correction data;

[0031] If the target multi-channel photon energy distortion data is obtained under the condition of multiple groups of physical parameters, inputting the target multi-channel photon energy distortion data into a second photon energy data correction model, so that the second photon energy data correction model outputs corresponding target single-channel photon energy correction data;

[0032] Wherein, the first photon energy data correction model and the second photon energy data correction model are obtained by pre-training based on the photon energy data correction model training method for photon counting spectral CT as described in the first aspect.

[0033] The third aspect of the present application provides a photon counting CT image reconstruction method, which comprises the following steps:

[0034] Image reconstruction is performed on the target single-channel photon energy correction data obtained by the photon energy data correction method for photon counting spectral CT as described in the second aspect, to obtain a photon counting CT reconstruction image corresponding to the target multi-channel photon energy distortion data.

[0035] The fourth aspect of the present application provides an electronic device, comprising: a computer device, the computer device comprising a processor and a memory, the memory storing computer instructions, the processor comprising a multi-core CPU and a multi-core GPU, for executing the computer instructions stored in the memory, when the computer instructions are executed by the processor, the device implements the steps of the photon energy data correction model training method for photon counting spectral CT as described in the first aspect, or, the steps of the photon energy data correction method for photon counting spectral CT as described in the second aspect, or, the steps of the photon counting CT image reconstruction method as described in the third aspect.

[0036] The fifth aspect of the present application provides a computer readable storage medium, storing a computer program, when the computer program is executed by a processor, the steps of the photon energy data correction model training method for photon counting spectral CT as described in the first aspect, or, the steps of the photon energy data correction method for photon counting spectral CT as described in the second aspect, or, the steps of the photon counting CT image reconstruction method as described in the third aspect are implemented.

[0037] The sixth aspect of the present application provides a computer program product, comprising computer instructions, when the computer instructions are executed by a processor, the steps of the photon energy data correction model training method for photon counting spectral CT as described in the first aspect, or, the steps of the photon energy data correction method for photon counting spectral CT as described in the second aspect, or, the steps of the photon counting CT image reconstruction method as described in the third aspect are implemented.

[0038] The photon energy data correction model training method, correction method and image reconstruction method for photon counting spectral CT of the present application, for charge sharing effect, based on single set of physical parameters and multiple sets of different physical parameters under the condition of multi-channel photon energy distortion data and the respective corresponding single-channel photon energy reference data, the corresponding neural network model is trained, and under the condition of multiple sets of physical parameters, the training method of parameter fusion is adopted, which can correct the nonlinear distortion spectrum caused by charge sharing effect, and can make the neural network model adapt to multiple physical parameters of the detector, improve the robustness of the model, and ensure good correction effect. Further, the image obtained after image reconstruction of the photon energy data corrected by the energy spectrum can obtain higher clarity.

[0039] Additional advantages, objects, and features of the application will be set forth in part by the description that follows, and will become apparent to those skilled in the art upon examination of the following detailed description and drawings in which illustrating the principles of the application. The objects and other advantages of the present application can be realized and attained by means of the structure particularly pointed out in the description and claims hereinafter.

[0040] Those skilled in the art will appreciate that the objects and advantages of the application can be realized and attained by means of the structure particularly pointed out in the appended claims. Understanding that these and other objects and advantages of the application can be achieved by the structures described above, the application is further described by the following detailed description and examples. BRIEF DESCRIPTION OF DRAWINGS

[0041] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description, serve to explain the principles of the application.

[0042] Figure 1 A flowchart of a photon energy data correction model training method for photon counting spectral CT according to an embodiment of the present application;

[0043] Figure 2 A structure diagram of a first neural network model corresponding to a single set of physical parameter conditions according to an embodiment of the present application;

[0044] Figure 3 A structure diagram of a second neural network model corresponding to a plurality of sets of physical parameter conditions according to an embodiment of the present application;

[0045] Figure 4 A flowchart of a photon energy data correction method for photon counting spectral CT according to an embodiment of the present application;

[0046] Figure 5 A flowchart of a photon counting CT image reconstruction method according to an embodiment of the present application;

[0047] Figure 6 An effect comparison diagram of a reconstructed image obtained in a single set of parameter mode according to an embodiment of the present application;

[0048] Figure 7 An effect comparison diagram of a reconstructed image obtained in a plurality of sets of parameter mode according to an embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the embodiments and drawings. Herein, the illustrative embodiments of the present application and the descriptions thereof are used to explain the present application, but are not used to limit the present application.

[0050] It is also noted herein that, while the above describes example embodiments, there are several variations and modifications which can be made to the disclosed solution without departing from the scope of the application as defined by the appended claims.

[0051] It should be emphasized that the term "comprises / comprising" when used in this text, means the presence of the stated features, elements, steps or components but does not preclude the presence or addition of one or more other features, elements, steps or components.

[0052] It is also noted herein that, if not otherwise specified, the term "connected" in this text can mean not only direct connection, but also indirect connection in the presence of an intermediate.

[0053] In the following, embodiments of the application will be described with reference to the accompanying drawings. In the drawings, like reference numerals designate like or similar parts or like or similar steps.

[0054] Due to the charge sharing effect of the photon counting detector, the measured X-ray spectrum (energy spectrum) is distorted, which further affects the final computed tomography reconstruction image. For the charge sharing effect of the photon counting detector, the embodiment of the application provides a photon energy data correction model training method, a correction method and an image reconstruction method for photon counting spectral CT. By using a pre-designed neural network model to correct the distorted photon energy data caused by the charge sharing effect, the nonlinear distorted spectrum caused by the charge sharing effect can be corrected, so that the image reconstructed based on the corrected photon energy data can obtain higher clarity.

[0055] Figure 1 A flowchart of a photon energy data correction model training method for photon counting spectral CT according to an embodiment of the application is shown in FIG. 1. As shown in FIG. 1, the method comprises the following steps: Figure 1

[0056] In step S110, a first neural network model is trained based on a plurality of first multi-channel photon energy distortion data and corresponding first single-channel photon energy reference data obtained under a single set of physical parameter conditions, so as to train the first neural network model into a first photon energy data correction model for outputting corresponding first single-channel photon energy correction data for photon counting spectral CT according to the first multi-channel photon energy distortion data and the corresponding first single-channel photon energy reference data.

[0057] ​In step S120, the second neural network model is trained based on the plurality of second multi-channel photon energy distortion data and the corresponding plurality of second single-channel photon energy reference data obtained under the plurality of sets of physical parameter conditions, and the plurality of sets of physical parameters, to output the corresponding second single-channel photon energy correction data for photon counting spectral CT according to the second multi-channel photon energy distortion data and the corresponding second single-channel photon energy reference data, wherein the second neural network model is formed by adding a parameter fusion network to the first neural network model in advance.

[0058] In some embodiments, the multi-channel photon energy distortion data is formed by combining the photon energy distortion data of the central pixel channel and the photon energy distortion data of a plurality of channels near the central pixel channel generated under a single set or multiple sets of physical parameter conditions.

[0059] Specifically, the photon energy distortion data of the central pixel channel and the multiple channels near the central pixel channel generated under the single set or multiple sets of physical parameter conditions can be obtained by software simulation under the one or more sets of physical parameter conditions of the photon counting detector that produces the charge sharing effect. Here, the channel can refer to a row or a column of the pixel array in the detector, and the central pixel channel refers to the row or column in which the central pixel unit is located. Each central pixel channel contains multiple central pixel units. When the charge sharing effect occurs, the number of photons actually received by the central pixel unit decreases and the photon energy decreases because the photons fall on the edge of the central pixel unit and are recorded by the surrounding multiple pixel units. Therefore, the number of photons in the high energy region decreases, the number of photons in the low energy region increases, and the energy spectrum of the incident X-rays is distorted. Therefore, under the ideal state in which the detector does not produce the charge sharing effect, the true photon energy received by the central pixel unit should be the sum of the photon energies received by the central pixel unit and the multiple pixel units around it under the state in which the detector produces the charge sharing effect. In this embodiment, according to the principle that the charge sharing effect of the photon counting CT causes the energy spectrum distortion, the distorted photon energy data, i.e., the photon energy distortion data, received or recorded by each row and each column of pixel units in the pixel array under multiple different angles is obtained by software simulation under the one or more sets of physical parameter conditions of the detector. In the process of specific implementation, the open-source photon counting detector toolkit is used to simulate a real photon counting CT scanner. By introducing the charge sharing effect generated by the interaction between the incident photons and the detector material, and introducing a certain noise, the photon energy distortion data is obtained. According to the energy of the incident X-rays, the distorted photon energy data is divided into multiple energy windows corresponding to multiple energy thresholds of the detector, such as 20, 70, 120, etc. The distorted photon energy data received or recorded by all pixel units in each row or each column is taken as a set, and multiple sets of photon energy distortion data form a photon energy distortion data set, i.e., the data set contains multiple single-channel photon energy distortion data, and each single-channel photon energy distortion data represents the photon energy distortion data of different energy windows under the charge sharing effect at different angles recorded by all pixel units in each channel. Therefore, each single-channel photon energy distortion data in the photon energy distortion data set contains four dimensions, and the data size can be represented as (R, W, A, C) = (1, W, A, C), where R represents the number of rows or columns (the number of channels) of the photon counting detector, W represents the number of energy windows into which the photons received by each pixel unit are divided according to the energy size, i.e., the number of energy thresholds; A represents the number of angles divided in the angle range detected by the detector, and C represents the number of pixel units in each channel.

[0060] In one specific example, the charge sharing effect considers the case that when a center pixel receives an X-ray photon, the photon falls on the edge of the center pixel, and is received by a 3x3 pixel array centered on the center pixel (that is, the center pixel is located at the position of the second row and the second column of the 3x3 pixel array), that is, the charge sharing of the 3x3 pixel array containing the center pixel and its surrounding pixels is considered, wherein the number of photons and the photon energy of the center pixel as the high-energy region decrease, and the number of photons and the photon energy of the surrounding pixels of the center pixel as the low-energy region increase.

[0061] It can be seen that the number of channels of the multi-channel photon energy distortion data formed by merging and expanding the single-channel photon energy distortion data is related to the number of pixel units in which the received or recorded photons are distorted when the incident photons produce a charge sharing effect. In one specific example, for the charge sharing of a 3x3 pixel array centered on a center pixel, the photon energy distortion data of the channel in which the center pixel producing the charge sharing effect is expanded into three-channel photon energy distortion data. After the photon received by the center pixel produces a charge sharing effect, the three-channel photon energy distortion data is formed by merging all the photon energy distortion data recorded by the channel in which the center pixel is located and the channel in which each of the eight pixel units around the center pixel is located in the 3x3 pixel array recording the photon energy, and the size of the three-channel photon energy distortion data is (R, W, A, C) = (3, W, A, C).

[0062] The single-channel photon energy reference data in step S110 or step S120 is a set of photon energy reference data received or recorded by all pixel units in each row or each column at different angles under an ideal state in which the photon counting detector does not produce a charge sharing effect under one or more sets of physical parameter conditions of the detector. In the process of acquiring the photon energy reference data, the open source photon counting detector toolkit is used to simulate the real photon counting CT scanner, a set or multiple sets of physical parameters of the detector are set, and no charge sharing effect and noise are introduced. In this ideal state, the photon energy reference data received or recorded by the pixel units in each row and each column at different angles is acquired. In one specific example, in the ideal state, the photon is received and recorded by the center pixel in the 3x3 pixel array, and it is considered that the incident photon is not received by the multiple pixels around the center pixel except the center pixel, that is, the photon is not recorded by any other pixel, and the photon energy spectrum is not distorted.

[0063] In some embodiments, the physical parameters include part or all of the following: the thickness of the photon counting detector, the pixel, the bias voltage, and the number of photons emitted by the X-ray emission end.

[0064] Specifically, the number of photons emitted by the emission end is the number of photons emitted by the emission end per unit time. The single set of physical parameter conditions can be a condition formed by taking a set of corresponding values of the thickness, the pixel, and the number of photons emitted by the emission end of the photon counting detector, such as a thickness value a, a pixel value b, and a number of photons emitted by the emission end e, which can be represented by a vector (a, b, e); or a condition formed by taking a set of corresponding values of the thickness, the pixel, the bias voltage, and the number of photons emitted by the emission end of the photon counting detector, such as a thickness value a, a pixel value b, a number of photons emitted by the emission end e, and a bias voltage value d, which can be represented by a vector (a, b, e, d); and there are various cases. The multiple sets of physical parameter conditions can be conditions formed by taking multiple sets of corresponding values of the thickness, the pixel, and the number of photons emitted by the emission end of the photon counting detector, such as a thickness value a1, a pixel value b1, and a number of photons emitted by the emission end e1, and a thickness value a2, a pixel value b2, and a number of photons emitted by the emission end e2, and so on, that is, vectors (a1, b1, e1) and (a2, b2, e2), and so on; or conditions formed by taking multiple sets of corresponding values of the thickness, the pixel, the bias voltage, and the number of photons emitted by the emission end of the photon counting detector, such as a thickness value a1, a pixel value b1, a number of photons emitted by the emission end e1, and a bias voltage value d1, and a thickness value a2, a pixel value b2, a number of photons emitted by the emission end e2, and a bias voltage value d2, and so on, that is, vectors (a1, b1, e1, d1) and (a2, b2, e2, d2), and so on; and there are also various cases.

[0065] In some embodiments, the first neural network model comprises: a first feature extraction network, a three-dimensional filtering network, a dimension reduction processing network, a two-dimensional filtering network, and a second feature extraction network connected in sequence.

[0066] The first feature extraction network comprises a plurality of feature extraction networks connected in sequence, and is configured to extract energy features in the input first multi-channel photon energy distortion data, and output first multi-channel photon energy features.

[0067] The three-dimensional filtering network comprises a plurality of residual blocks and a first channel injection layer connected in sequence, and is configured to filter and denoise the extracted first multi-channel photon energy features, wherein each residual block comprises a three-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence, the first channel injection layer comprises a pooling layer and a plurality of groups of three-dimensional convolution layers and nonlinear activation functions connected in sequence, and along a direction from input to output of the three-dimensional filtering network, an input of a first three-dimensional convolution layer in a first residual block and an output of a last nonlinear activation function in the first channel injection layer are added to serve as an output of the three-dimensional filtering network.

[0068] The dimension reduction processing network comprises a three-dimensional convolution layer, a normalization layer, a nonlinear activation function and a dimension reduction layer connected in sequence, and is configured to reduce a channel dimension of the filtered and denoised first multi-channel photon energy features, and output a first single-channel photon energy feature.

[0069] The two-dimensional filtering network comprises a plurality of residual blocks and a second channel injection layer connected in sequence, and is configured to filter and denoise the output first single-channel photon energy feature, wherein each residual block comprises a two-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence, the second channel injection layer comprises a pooling layer and a plurality of groups of two-dimensional convolution layers and nonlinear activation functions connected in sequence, and along a direction from input to output of the two-dimensional filtering network, an input of a first two-dimensional convolution layer in a first residual block and an output of a last nonlinear activation function in the second channel injection layer are added to serve as an output of the two-dimensional filtering network.

[0070] The second feature extraction network comprises a plurality of two-dimensional convolution layers connected in sequence, and is configured to extract deeper energy features in the filtered and denoised first single-channel photon energy feature, and output first single-channel photon energy correction data.

[0071] As Figure 2As shown, in one specific example, the first feature extraction network includes two feature extraction networks, three-channel photon energy distortion data (3, 3, 360, 1632) obtained under a set of physical parameter conditions is input into the first feature extraction network, energy features in the three-channel photon energy distortion data are extracted by a three-dimensional convolution layer in the first feature extraction network for the first time, and then group normalization and introduction of nonlinearity are performed through a normalization layer and a nonlinear activation function in sequence. The second feature extraction is performed by a three-dimensional convolution layer in the second feature extraction network, and then the normalized and nonlinearly activated output three-channel energy feature data (64, 3, 360, 1632) is obtained. The three-dimensional filtering network can alleviate the gradient disappearance problem in the deep network, is composed of two residual blocks, and the three-channel energy feature data (64, 3, 360, 1632) is input into the three-dimensional filtering network for filtering and removing noise. The pooling layer in it can adaptively compress the features, the input three-channel energy feature data (64, 3, 360, 1632) and the output of the data after sequentially passing through two residual blocks and a channel injection layer are added, and the filtered three-channel energy feature data (64, 3, 360, 1632) is output as the final output. Then the filtered three-channel energy feature data (64, 3, 360, 1632) is input into the dimension reduction processing network for channel dimension reduction operation, the three-channel energy feature data (64, 3, 360, 1632) is converted into single-channel energy feature data (64, 1, 360, 1632), and the dimension with a dimension of 1 is removed, and the single-channel energy feature data (64, 360, 1632) is output. The two-dimensional filtering network can alleviate the gradient disappearance problem in the deep network, is composed of two residual blocks, and then the single-channel energy feature data (64, 360, 1632) is input into the two-dimensional filtering network for filtering and removing noise. The pooling layer in it can adaptively compress the features, the input single-channel energy feature data (64, 360, 1632) and the output of the data after sequentially passing through two residual blocks and a channel injection layer are added, and the filtered single-channel energy feature data (64, 360, 1632) is output as the final output. Finally, the filtered single-channel energy feature data (64, 360, 1632) is input into the second feature extraction network for deeper energy feature extraction, and the final single-channel energy feature data (3, 360, 1632) is output by the second feature extraction network, that is, the corrected single-channel photon energy data (1, 3, 360, 1632).

[0072] In some embodiments, before training the second neural network model, the photon energy data correction model training method further includes: encoding the plurality of sets of physical parameters by position encoding.

[0073] Specifically, the plurality of sets of physical parameters of the detector are positionally encoded by the following formula respectively:

[0074]

[0075]

[0076] posA=sin(X extend ⊙InvF)

[0077] posB=cos(X extend ⊙InvF)

[0078] pos=[posA,posB]

[0079] wherein n represents the number of physical parameters in the set of physical parameters of the photon counting detector; c represents the dimension size of the data output after encoding, which is consistent with the size of the first dimension of the data output by each network (the first feature extraction network, the three-dimensional filtering network, the dimension reduction processing network, the two-dimensional filtering network, or the second feature extraction network) to be fused in the neural network model with parameter fusion (i.e. related to the size of the convolution kernel of each convolution layer in each network); A is a constructed vector, and the value of an element in the vector is 2k, wherein k∈[0,2,4,…,c×n / 2); InvF is an inverse frequency matrix in position encoding; X is a physical parameter vector composed of the physical parameters of the set of photon counting detectors, and the length of the vector is n; the physical parameter vector X is extended to obtain X extend , and the length of X extend is consistent with the length of the inverse frequency matrix InvF; posA is a sine encoding matrix, posB is a cosine encoding matrix, represents the operation of multiplying the elements at corresponding positions of the matrices, and pos is a physical parameter encoding matrix finally obtained after encoding.

[0080] In some embodiments, the second neural network model comprises: a first feature extraction network, a first parameter fusion network, a three-dimensional filtering network, a second parameter fusion network, a dimension reduction processing network, a third parameter fusion network, a two-dimensional filtering network, a fourth parameter fusion network, a second feature extraction network, and a fifth parameter fusion network. Each parameter fusion network comprises a nonlinear activation function and a fully connected layer connected in sequence.

[0081] The first feature extraction network comprises a plurality of feature extraction networks connected in sequence, for extracting energy features in the input second multi-channel photon energy distortion data, and corresponding outputting second multi-channel photon energy features, wherein each feature extraction network comprises a three-dimensional convolution layer, a normalization layer, and a nonlinear activation function connected in sequence.

[0082] The first parameter fusion network is configured to introduce a plurality of groups of physical parameters, and fuse the second plurality of channel photon energy features extracted by the first feature extraction network and the corresponding groups of physical parameters, to output a first plurality of channel fusion result;

[0083] The three-dimensional filtering network comprises a plurality of residual blocks and a first channel injection layer connected in sequence, and is configured to filter and denoise the output first plurality of channel fusion result, wherein each residual block comprises a three-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence, and the first channel injection layer comprises a pooling layer and a plurality of groups of three-dimensional convolution layers and nonlinear activation functions connected in sequence, and along a direction from input to output of the three-dimensional filtering network, an input of a first three-dimensional convolution layer in a first residual block and an output of a last nonlinear activation function in the first channel injection layer are added to serve as an output of the three-dimensional filtering network;

[0084] The second parameter fusion network is configured to introduce a plurality of groups of physical parameters, and fuse the first plurality of channel fusion result filtered and denoised by the three-dimensional filtering network and the corresponding groups of physical parameters, to output a second plurality of channel fusion result;

[0085] The dimension reduction processing network comprises a three-dimensional convolution layer, a normalization layer, a nonlinear activation function and a dimension reduction layer connected in sequence, and is configured to reduce a channel dimension of the output second plurality of channel fusion result, to output a first single-channel fusion result;

[0086] The third parameter fusion network is configured to introduce a plurality of groups of physical parameters, and fuse the first single-channel fusion result output by the dimension reduction processing network and the corresponding groups of physical parameters, to output a second single-channel fusion result;

[0087] The two-dimensional filtering network comprises a plurality of residual blocks and a second channel injection layer connected in sequence, and is configured to filter and denoise the output second single-channel fusion result, wherein each residual block comprises a two-dimensional convolution layer, a normalization layer and a nonlinear activation function connected in sequence, and the second channel injection layer comprises a pooling layer and a plurality of groups of two-dimensional convolution layers and nonlinear activation functions connected in sequence, and along a direction from input to output of the two-dimensional filtering network, an input of a first two-dimensional convolution layer in a first residual block and an output of a last nonlinear activation function in the second channel injection layer are added to serve as an output of the two-dimensional filtering network;

[0088] The fourth parameter fusion network is configured to introduce a plurality of groups of physical parameters, and fuse the second single-channel fusion result filtered and denoised by the two-dimensional filtering network and the corresponding groups of physical parameters, to output a third single-channel fusion result;

[0089] The second feature extraction network comprises a plurality of two-dimensional convolution layers connected in sequence, and is configured to extract deeper energy features in the output third single-channel fusion result.

[0090] The fifth parameter fusion network is configured to introduce a plurality of groups of physical parameters, and fuse the deeper energy features extracted by the second feature extraction network and the corresponding group of physical parameters, so as to output second single-channel photon energy correction data.

[0091] As shown in Figure 3 In one specific example, on the basis of the network structure shown in Figure 2 The first to fifth parameter fusion networks are added, that is Figure 2 The outputs of the respective networks and the outputs of the respective parameter fusion networks are added together and input into the next network. Each parameter fusion network comprises a nonlinear activation function and a fully connected layer connected in sequence, wherein the nonlinear activation function uses a SiLU function, and the data dimension output by the fully connected layer is consistent with the output data dimension of the corresponding added network. That is, in Figure 3In the neural network model shown, on the basis of the three-channel photon energy distortion data (3, 3, 360, 1632) obtained under the input of multiple sets of physical parameter conditions, the pre-encoding of each set of physical parameters under the corresponding conditions is also input. The three-channel photon energy distortion data (3, 3, 360, 1632) is input into the first feature extraction network for twice feature extraction, and the first feature extraction network outputs three-channel energy feature data (64, 3, 360, 1632). The physical parameter encoding matrix (1, 3x64) obtained by pre-encoding a set of physical parameters under the corresponding conditions is input into the first and second parameter fusion networks respectively, and both output the physical parameter encoding matrix (1, 64), and (1, 64) is transformed into (64, 3, 360, 1632), or (1, 64) can be transformed into (64, 3, 360, 1632) by the full connection layer in the first and second parameter fusion networks. After the physical parameter encoding matrix (64, 3, 360, 1632) passing through the first parameter fusion network and the corresponding three-channel energy feature data (64, 3, 360, 1632) are added, the first three-channel fusion result is obtained and input into the three-dimensional filtering network, and the three-dimensional filtering network outputs the filtered first three-channel fusion result (64, 3, 360, 1632). In turn, after the physical parameter encoding matrix (64, 3, 360, 1632) passing through the second parameter fusion network and the corresponding filtered first three-channel fusion result (64, 3, 360, 1632) are added, the second three-channel fusion result is obtained and input into the dimension reduction processing network for dimension reduction processing, and the dimension reduction processing network outputs the first single-channel fusion result (64, 1, 360, 1632). The dimension 1 is omitted, and the first single-channel fusion result (64, 360, 1632) is output. The physical parameter encoding matrix (1, 3x64) obtained by pre-encoding the set of physical parameters under the above condition is input into the third and fourth parameter fusion networks respectively, and both output the physical parameter encoding matrix (1, 64), and (1, 64) is transformed into (64, 360, 1632), or (1, 64) can be transformed into (64, 360, 1632) by the full connection layer in the third and fourth parameter fusion networks. After the physical parameter encoding matrix (64, 360, 1632) passing through the third parameter fusion network and the corresponding first single-channel fusion result (64, 360, 1632) are added, the second single-channel fusion result is obtained and input into the two-dimensional filtering network, and the two-dimensional filtering network outputs the filtered second single-channel fusion result (64, 360, 1632).The third single-channel fusion result after energy feature extraction (3, 360, 1632) is output by the second feature extraction network; the physical parameter coding matrix (1, 3x64) obtained by pre-encoding the set of physical parameters under the above conditions is input into the fifth parameter fusion network, and a physical parameter coding matrix (1, 3) is output, which is transformed into (3, 360, 1632), or the data dimension can also be transformed by the full connection layer in the fifth parameter fusion network, and (1, 3) is transformed into (3, 360, 1632); the physical parameter coding matrix (3, 360, 1632) after the fifth parameter fusion network and the corresponding third single-channel fusion result after energy feature extraction (3, 360, 1632) are added, and the result is the corrected single-channel photon energy data (1, 3, 360, 1632).

[0092] In steps S110 and S120, the obtained multiple multi-channel photon energy distortion data and the corresponding multiple first single-channel photon energy reference data constitute a photon energy data set, which is divided into a training data set and a test data set according to a certain proportion, and the neural network model is iteratively trained based on the training data set and the test data set by minimizing the loss function until the loss function converges, so that the model has the ability of photon energy distortion correction and energy spectrum correction. In the iterative training process, the model performance under different initial learning rates is compared, and the appropriate initial learning rate is selected, and the learning rate is dynamically adjusted in a demand-adjusted manner.

[0093] In the iterative training process of the second neural network model with parameter fusion, the multi-channel photon energy distortion data and the corresponding single-channel photon energy reference data, and the pre-encoded physical parameters under the corresponding conditions are input into the second neural network model with parameter fusion, the loss function is calculated, and the loss function is minimized by using the back propagation method. The loss function specifically uses the MSE function, as shown in the following formula:

[0094]

[0095] L = 1 / N åN i=1 (yi - y^i)2 MSELoss is a loss value, X is a physical parameter of a photon counting detector, S is multi-channel photon energy distortion data, f is corrected single-channel photon energy data (single-channel photon energy correction data) output by a model, y is single-channel photon energy reference data corresponding to S, i is a sequence number of input multi-channel photon energy distortion data, single-channel photon energy reference data corresponding to the multi-channel photon energy distortion data or a group of physical parameters, i = 1, 2, …, n, n is a total number participating in mean square error calculation. Then, the gradient value of the current model parameter is calculated by back propagation In each iteration training process, the gradient descent method is used, the learning rate of the model is combined, and the Adam optimization algorithm is used to adjust the parameters of the model, gradually reduce the gradient value, until the loss function of the model reaches convergence, so that the loss function value of the model reaches the minimum value, so that the output of the model infinitely approximates the corresponding single-channel photon energy reference data, and then the training of the neural network model is stopped, and finally the trained parameter fusion neural network model for photon counting CT imaging is obtained.

[0096] Figure 4 It is an embodiment of the present application. As shown in the flowchart of a photon energy data correction method for photon counting spectral CT, the present application also provides a photon energy data correction method for photon counting spectral CT, which comprises the following steps: Figure 4 As shown in the flowchart of a photon energy data correction method for photon counting spectral CT, the present application also provides a photon energy data correction method for photon counting spectral CT, which comprises the following steps:

[0097] Step S410, obtaining target multi-channel photon energy distortion data.

[0098] The target multi-channel photon energy distortion data in this step is obtained by obtaining the real photon energy distortion data generated by the charge sharing effect of the photon counting detector, and merging the photon energy distortion data of each center pixel channel and the photon energy distortion data of a plurality of channels near the corresponding center pixel channel.

[0099] Step S420, if the target multi-channel photon energy distortion data is obtained under a single set of physical parameters, the target multi-channel photon energy distortion data is input into a first photon energy data correction model, so that the first photon energy data correction model outputs corresponding target single-channel photon energy correction data for photon counting spectral CT.

[0100] Step S430, if the target multi-channel photon energy distortion data is obtained under a plurality of sets of physical parameters, the target multi-channel photon energy distortion data is input into a second photon energy data correction model, so that the second photon energy data correction model outputs corresponding target single-channel photon energy correction data for photon counting spectral CT.

[0101] The first photon energy data correction model and the second photon energy data correction model are obtained by training the photon energy data correction model training method for photon counting spectral CT in any one of the preceding embodiments.

[0102] The single-channel photon energy correction data output by the trained neural network model, i.e., the photon energy data correction model, is the corrected photon energy data of the central pixel channel contained in the input multi-channel photon energy distortion data after correction. Accordingly, the energy spectrum of the distorted X-ray photons is also corrected, and the number of counting errors of lower energy photons is concentrated in the higher energy region. In one specific example, the corrected single-channel photon energy data is the photon energy correction data of the central pixel channel in 3x3, i.e., the corrected photon energy data of all pixels in the second row or column in 3x3. Accordingly, the photon number of the surrounding 8 pixels of the center pixel in 3x3 is concentrated to the center pixel, and the distorted energy spectrum can be effectively corrected.

[0103] Figure 5 A flowchart of a photon counting CT image reconstruction method according to an embodiment of the present application is shown in FIG. 5. As shown in FIG. 5, the present embodiment also provides a photon counting CT image reconstruction method, which includes the following steps: Figure 5

[0104] In step S510, the target single-channel photon energy correction data obtained by the photon energy data correction method for photon counting spectral CT in any one of the preceding embodiments is subjected to image reconstruction to obtain a photon counting CT reconstruction image corresponding to the target multi-channel photon energy distortion data.

[0105] In the present embodiment, the filtered back-projection algorithm is used to directly obtain a noise-removed reconstruction image for photon counting CT from the plurality of corrected single-channel photon energy data obtained by the photon energy data correction method in any one of the preceding embodiments. The pixel size of the detector, the number of rows and columns of the detector array, and other parameters used in the algorithm are consistent with the physical parameters of the detector set when obtaining the photon energy distortion data corresponding to the reconstruction image.

[0106] ​The performance of the photon energy data correction model trained by the method is tested and evaluated by the image reconstruction method. Specifically, the first photon energy data correction model and the second photon energy data correction model are tested by using the test data sets obtained under the conditions of single set of physical parameters and multiple sets of physical parameters, respectively. First, the multi-channel photon energy distortion data in the test data sets under the conditions of single set of physical parameters and multiple sets of physical parameters are corrected and the energy spectrum is corrected by the first photon energy data correction model and the second photon energy data correction model, respectively. Then, the single-channel photon energy correction data output by the first photon energy data correction model and the second photon energy data correction model are respectively reconstructed by the image reconstruction method to obtain the first photon counting CT reconstructed image and the second photon counting CT reconstructed image. Second, the ideal state photon energy reference data in the test data sets under the conditions of single set of physical parameters and multiple sets of physical parameters are directly reconstructed by the image reconstruction method to obtain the first reconstructed reference image and the second reconstructed reference image. Finally, the performance of the first photon energy data correction model and the second photon energy data correction model is evaluated by comparing the first photon counting CT reconstructed image and the first reconstructed reference image, and the second photon counting CT reconstructed image and the second reconstructed reference image. Specifically, a plurality of evaluation indexes are used, including the peak signal-to-noise ratio (PSNR) of the image, the normalized mean square error (NMSE), and the structural similarity index (SSIM), and the formulas are as follows:

[0107]

[0108]

[0109]

[0110]

[0111] wherein, F MSE is the mean square error, I(k,j) and K(k,j) represent monochrome images with a size of m×p, I(k,j) is the photon counting CT reconstructed image, K(k,j) is the reconstructed reference image, k and j represent the kth row and jth column in the image data matrix, k=0,2,…,m-1, j=0,2,…,p-1; F NMSE is the normalized mean square error, which is a normalized form of the mean square error; F PSNR is the peak signal-to-noise ratio, MAX represents the maximum value of each pixel of the monochrome image; F SSIM is the structural similarity index, μ I and μ K are the average values of the images I and K, respectively, δ I 2 and δ K 2respectively, the variance of the images I and K, δ Ik is the covariance of the images I and K, C1, C2 are constants. NMSE the value of F ranges between 0 and 1, NMSE The smaller the value of F is, the better the performance of the photon energy data correction model is. PSNR The greater the value of F is, the closer the photon counting CT reconstructed image and the reconstructed reference image are, i.e., the higher the quality of the photon counting CT reconstructed image is. SSIM The value of F ranges between 0 and 1, and is used to measure the similarity between the photon counting CT reconstructed image and the reconstructed reference image. SSIM The greater the value of F is, the higher the similarity and the smaller the difference between the photon counting CT reconstructed image and the reconstructed reference image are.

[0112] In one specific test case, the test results are shown in Tables 1-3. In Tables 1-3, the evaluation index data corresponding to the noise is obtained based on the distorted image obtained after image reconstruction based on the photon energy distortion data obtained in the case of having noise and generating the charge sharing effect (e.g., the distorted image in Figure 6 or Figure 7 The parameter fusion mode or the multi-group parameter mode corresponds to the second photon energy data correction model; and the single-group parameter mode corresponds to the first photon energy data correction model.

[0113] Table 1

[0114]

[0115] Table 2

[0116]

[0117] Table 3

[0118]

[0119] As can be seen from Tables 1-3, in the case of obtaining data under the same input conditions, the correction performance of the second photon energy data correction model in the parameter fusion mode is basically consistent with that of the first photon energy data correction model in the single-group parameter mode, and each evaluation index is obviously improved. After energy spectrum correction, the average value of each index of multiple energy windows is obtained, and it can be known that in the two modes, the PSNR value can reach more than 30, the NMSE value is close to 0, and the SSIM value is close to 1, which also indicates that in the two modes, through energy spectrum correction, the photon counting CT reconstructed image obtained finally is close to the reconstructed reference image obtained in the ideal state, and the image similarity is very high.

[0120] Correspondingly, the reconstructed images obtained in the two modes are as follows: Figures 6-7As shown in the figure, the distorted image is a distorted image obtained by directly reconstructing the photon energy distorted data, the corrected image is a photon counting CT reconstructed image obtained by reconstructing the corrected single-channel photon energy data (single-channel photon energy corrected data), and the reference image is a reconstructed reference image obtained by reconstructing the photon energy reference data in an ideal state. It can be known by selecting a region of interest to measure the average value of pixels that the ROI image (image of a region of interest) in the photon counting CT reconstructed image obtained by energy spectrum correction is close to the average value of the reconstructed reference image, so it can be concluded that the quality of the reconstructed image can be effectively improved in both modes. It should be noted that the second photon energy data correction model in the parameter fusion mode or the multi-group parameter mode has shown better correction performance for the photon energy distorted data obtained under different conditions of different groups of physical parameters, and the quality of the corresponding reconstructed image has been effectively improved. The first photon energy data correction model in the single-group parameter mode has shown better correction performance basically consistent with the above performance for the photon energy distorted data obtained under a single condition (single group of physical parameters), and the correction effect of the photon energy distorted data obtained under other conditions has decreased, so the second photon energy data correction model in the parameter fusion mode or the multi-group parameter mode has better and more extensive applicability.

[0121] Corresponding to the above method, the application also provides an electronic device, which comprises a computer device, the computer device comprising a processor and a memory, the memory storing computer instructions, the processor comprising a multi-core CPU and a multi-core GPU, and the computer instructions being executed by the processor to implement the steps of the above method.

[0122] The embodiment of the application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by the processor to implement the steps of the above method. The computer readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0123] The embodiment of the application also provides a computer program product, which comprises computer instructions, and the computer instructions are executed by the processor to implement the steps of the above method.

[0124] Those of ordinary skill in the art will appreciate that the various illustrative components, systems and methods described in connection with the embodiments disclosed herein can be implemented as hardware, software, or both. The particular implementation is dependent on the specific application and design constraints imposed on the overall system. Skilled persons can implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application. When implemented in hardware, for example, the hardware can comprise an electronic circuit, an Application Specific Integrated Circuit (ASIC), a suitable firmware, a plug-in, a functional card, etc. When implemented in software, the elements of the application are the program or code segments to perform a specific task. The program or code segments can be stored in a machine-readable medium, or transmitted by a carrier wave as data signals over a transmission medium or communication link.

[0125] It is to be understood that the application is not limited to the particular configurations and processes described herein and shown in the drawings. For simplicity, detailed descriptions of known methods and apparatuses are omitted so as not to obscure the disclosure. In the above-described embodiments, several specific steps are described and illustrated as examples. However, the method processes of the present application are not limited to the specific steps described and illustrated, and the order of the steps can be changed, or other steps can be added, or replaced, or eliminated, depending on the application.

[0126] In the present application, features described and / or illustrated with respect to one embodiment can be used in the same or a similar way in one or more other embodiments, and / or in combination with or instead of features of other embodiments.

[0127] The above description is only preferred embodiments of the present application, and is not intended to limit the present application. The embodiments of the present application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the present application.

Claims

1. A method for training a photon energy data correction model for photon counting spectral CT, characterized in that, The method includes: A first neural network model is trained based on multiple first multi-channel photon energy distortion data and corresponding multiple first single-channel photon energy reference data obtained under a single set of physical parameters. This first neural network model is then trained to become a first photon energy data correction model that outputs corresponding first single-channel photon energy correction data for photon counting spectral CT based on the first multi-channel photon energy distortion data and the corresponding first single-channel photon energy reference data. Based on multiple second multi-channel photon energy distortion data and corresponding multiple second single-channel photon energy reference data obtained under multiple sets of physical parameters, a second neural network model is trained to train the second neural network model into a second photon energy data correction model for outputting corresponding second single-channel photon energy correction data for photon counting energy spectrum CT based on the second multi-channel photon energy distortion data and corresponding second single-channel photon energy reference data and physical parameters. The second neural network model is formed by adding a parameter fusion network to the first neural network model in advance.

2. The method according to claim 1, characterized in that, The multi-channel photon energy distortion data is formed by merging the photon energy distortion data of the central pixel channel generated under one or more sets of physical parameters with the photon energy distortion data of multiple channels near the central pixel channel.

3. The method according to claim 1, characterized in that, The first neural network model includes a first feature extraction network, a three-dimensional filtering network, a dimensionality reduction processing network, a two-dimensional filtering network, and a second feature extraction network; The first feature extraction network includes multiple feature extraction networks connected in sequence, used to extract energy features from the input first multi-channel photon energy distortion data and output the first multi-channel photon energy features accordingly. Each feature extraction network includes a three-dimensional convolutional layer, a normalization layer and a non-linear activation function connected in sequence. The three-dimensional filtering network includes multiple residual blocks and a first channel injection layer connected in sequence, used to filter and denoise the extracted first multi-channel photon energy features. Each residual block includes a three-dimensional convolutional layer, a normalization layer, and a nonlinear activation function connected in sequence. The first channel injection layer includes a pooling layer connected in sequence, as well as multiple sets of three-dimensional convolutional layers and nonlinear activation functions connected in sequence. Along the input-to-output direction of the three-dimensional filtering network, the input of the first three-dimensional convolutional layer in the first residual block and the output of the last nonlinear activation function in the first channel injection layer are added together to obtain the output of the three-dimensional filtering network. The dimensionality reduction processing network includes a three-dimensional convolutional layer, a normalization layer, a nonlinear activation function, and a dimensionality reduction layer connected in sequence. It is used to reduce the channel dimension of the first multi-channel photon energy feature after filtering and denoising, and outputs the first single-channel photon energy feature accordingly. The two-dimensional filtering network includes multiple residual blocks and a second channel injection layer connected in sequence, used to filter and denoise the output first single-channel photon energy features. Each residual block includes a two-dimensional convolutional layer, a normalization layer, and a nonlinear activation function connected in sequence. The second channel injection layer includes a pooling layer connected in sequence, as well as multiple sets of two-dimensional convolutional layers and nonlinear activation functions connected in sequence. Along the input-to-output direction of the two-dimensional filtering network, the input of the first two-dimensional convolutional layer in the first residual block and the output of the last nonlinear activation function in the second channel injection layer are added together to obtain the output of the two-dimensional filtering network. The second feature extraction network includes multiple two-dimensional convolutional layers connected in sequence, used to extract deeper energy features from the filtered and denoised first single-channel photon energy features, and output the first single-channel photon energy correction data accordingly.

4. The method according to claim 1, characterized in that, The second neural network model includes a first feature extraction network, a first parameter fusion network, a three-dimensional filtering network, a second parameter fusion network, a dimensionality reduction processing network, a third parameter fusion network, a two-dimensional filtering network, a fourth parameter fusion network, a second feature extraction network, and a fifth parameter fusion network, wherein each parameter fusion network includes a non-linear activation function and a fully connected layer connected in sequence; The first feature extraction network includes multiple feature extraction networks connected in sequence, used to extract energy features from the input second multi-channel photon energy distortion data and output the second multi-channel photon energy features accordingly. Each feature extraction network includes a three-dimensional convolutional layer, a normalization layer and a non-linear activation function connected in sequence. The first parameter fusion network is used to introduce multiple sets of physical parameters, fuse the second multi-channel photon energy features extracted by the first feature extraction network with the corresponding set of physical parameters, and output the first multi-channel fusion result accordingly. The three-dimensional filtering network includes multiple residual blocks and a first channel injection layer connected in sequence, used to filter and denoise the output first multi-channel fusion result. Each residual block includes a three-dimensional convolutional layer, a normalization layer and a nonlinear activation function connected in sequence. The first channel injection layer includes a pooling layer connected in sequence, and multiple sets of three-dimensional convolutional layers and nonlinear activation functions connected in sequence. Along the input to output direction of the three-dimensional filtering network, the input of the first three-dimensional convolutional layer in the first residual block and the output of the last nonlinear activation function in the first channel injection layer are added together to obtain the output of the three-dimensional filtering network. The second parameter fusion network is used to introduce multiple sets of physical parameters, and fuse the first multi-channel fusion result after filtering and denoising output by the three-dimensional filtering network with the corresponding set of physical parameters to output the second multi-channel fusion result. The dimensionality reduction processing network includes a three-dimensional convolutional layer, a normalization layer, a nonlinear activation function, and a dimensionality reduction layer connected in sequence, which is used to reduce the channel dimension of the output second multi-channel fusion result, corresponding to the output first single-channel fusion result; The third parameter fusion network is used to introduce multiple sets of physical parameters, and fuse the first single-channel fusion result output by the dimensionality reduction processing network with the corresponding set of physical parameters to output the second single-channel fusion result. The two-dimensional filtering network includes multiple residual blocks and a second channel injection layer connected in sequence, used to filter and denoise the output second single-channel fusion result. Each residual block includes a two-dimensional convolutional layer, a normalization layer, and a nonlinear activation function connected in sequence. The second channel injection layer includes a pooling layer connected in sequence, as well as multiple sets of two-dimensional convolutional layers and nonlinear activation functions connected in sequence. Along the input to output direction of the two-dimensional filtering network, the input of the first two-dimensional convolutional layer in the first residual block and the output of the last nonlinear activation function in the second channel injection layer are added together to obtain the output of the two-dimensional filtering network. The fourth parameter fusion network is used to introduce multiple sets of physical parameters, and fuse the filtered and denoised second single-channel fusion result output by the two-dimensional filtering network with the corresponding set of physical parameters to output the third single-channel fusion result. The second feature extraction network includes multiple two-dimensional convolutional layers connected in sequence, used to extract deeper energy features from the output third single-channel fusion result; The fifth parameter fusion network is used to introduce multiple sets of physical parameters, fuse the deeper energy features extracted by the second feature extraction network with the corresponding sets of physical parameters, and output the second single-channel photon energy correction data.

5. The method according to any one of claims 1 to 4, characterized in that, The physical parameters include some or all of the following: the thickness of the photon counting detector, pixels, bias voltage, and the number of photons emitted by the X-ray emitting end; Before training the second neural network model, the method further includes encoding the multiple sets of physical parameters by positional encoding.

6. A method for correcting photon energy data in photon counting spectral CT, characterized in that, The method includes: Acquire multi-channel photon energy distortion data of the target; If the target multi-channel photon energy distortion data is obtained under a single set of physical parameters, then the target multi-channel photon energy distortion data is input into the first photon energy data correction model so that the first photon energy data correction model outputs the corresponding target single-channel photon energy correction data for photon counting energy spectrum CT. If the target multi-channel photon energy distortion data is obtained under multiple sets of physical parameters, then the target multi-channel photon energy distortion data is input into the second photon energy data correction model so that the second photon energy data correction model outputs the corresponding target single-channel photon energy correction data for photon counting spectral CT. The first photon energy data correction model and the second photon energy data correction model are obtained in advance based on the photon energy data correction model training method for photon counting spectral CT as described in any one of claims 1 to 5.

7. A method for reconstructing photon-counted CT images, characterized in that, The method includes: Image reconstruction is performed on the target single-channel photon energy correction data obtained by the photon energy data correction method for photon counting spectral CT as described in claim 6, to obtain the photon counting CT reconstructed image corresponding to the target multi-channel photon energy distortion data.

8. An electronic device comprising a processor, a memory, and computer instructions stored in the memory, characterized in that, The processor includes a multi-core CPU and a multi-core GPU for executing the computer instructions, and when the computer instructions are executed, the device implements the steps of the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method as described in any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1 to 7.

Images