Method, system, device and medium for positron emission tomography parameter imaging

Abstract

The present invention discloses a method, system, device and medium for positron emission tomography parameter imaging, including: generating a first input function before an acquisition period, a second input function within an acquisition period and generating an average activity image, obtaining a first predicted image according to the average activity image, if the first predicted image does not meet a condition, splicing the first input function and the second input function to obtain a third input function, updating the weight parameters of a deep neural network according to the third input function, the first predicted image, a label image and a multiplicative variable, inputting the average activity image into the updated deep neural network to obtain a second predicted image, if the second predicted image does not meet a condition, updating the label image and the multiplicative variable according to the second predicted image, the label image and the multiplicative variable, iterating several times until the second predicted image meets a preset condition, and outputting the second predicted image. The present invention can improve the image quality of short-time scanning parameter imaging and is applied to the field of PET imaging technology.

Description

Technical Field

[0001] The present invention relates to the field of PET imaging technology, and in particular to a method, system, device and medium for positron emission tomography parameter imaging. Background Art

[0002] Positron emission tomography (PET) technology is widely used and has become an important imaging diagnostic tool in oncology, neurology, cardiology and pathology research. The clinical imaging process of PET is to first label the essential substances in life metabolism, such as glucose, protein, nucleic acid, fatty acid, etc., with short half-life radionuclides such as 18 F, then injected into the human body, the patient is scanned within the effective field of view of the PET detector, and then the computer system reconstructs the image to obtain the medical image. Currently, the main radioactive marker used in hospitals is fluorodeoxyglucose, referred to as 18 F-FDG has been widely used in the diagnosis, staging and monitoring of tumors. 18 F-FDG usually dynamically collects data for 1 hour, divides the scanning time into frames during image reconstruction, and obtains multi-frame reconstructed PET parameter images based on the time frame results.

[0003] Linear models in kinetic models can be estimated by graphical methods, such as Patlak and other methods. Graphical methods are widely used because of their linear characteristics, simple calculation process, and fast calculation speed. 18 F-FDG PET parametric imaging of dynamic scanning. The Patlak method is characterized by the need for the distribution of the drug in the body to reach a state of equilibrium before scanning, so the dynamic scanning time is usually 1 hour. During the scanning process, the patient's autonomous movement will affect the quality of the PET parametric image, causing image artifacts and affecting clinical diagnosis. It is very difficult for patients to remain motionless for 1 hour, which can easily cause discomfort to patients. Therefore, shortening the dynamic PET scanning time for parametric imaging is of great significance.

[0004] However, simply using short-time scanning data for Patlak parameter imaging will lead to increased image noise and quantitative deviation, resulting in lower image quality obtained by short-time scanning parameter imaging. Summary of the invention

[0005] In view of this, an object of the present invention is to provide a method, system, device and medium for positron emission tomography parametric imaging, which can improve the image quality obtained by short-time scanning parametric imaging.

[0006] In one aspect, the present invention provides a method for positron emission tomography parametric imaging, comprising the following steps:

[0007] Acquire scanning data within an acquisition period, generate a first input function before the acquisition period, generate a second input function within the acquisition period according to the scanning data, and generate an average activity image;

[0008] Inputting the average activity image into an initial deep neural network to obtain a first predicted image, and determining whether the first predicted image meets a preset condition;

[0009] If the first predicted image does not meet the preset condition, concatenate the first input function and the second input function to obtain a third input function, and update the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current multiplicative variable;

[0010] The average activity image is input into the deep neural network after updating the weight parameters to obtain a second predicted image; if the second predicted image does not meet the preset conditions, the current label image and the current multiplicative variable are updated according to the second predicted image, the second predicted image is used as the first predicted image, and the concatenation of the first input function and the second input function is returned until the second predicted image meets the preset conditions, and the second predicted image is output.

[0011] Optionally, generating a second input function within a collection period according to the scan data specifically includes:

[0012] Divide the scan data into frames according to the scan time to obtain a plurality of frames of scan data;

[0013] Performing image reconstruction according to the scanning data of each frame to obtain positron emission tomography images of several frames;

[0014] A second input function is obtained according to pixel values ​​of a plurality of the positron emission tomography images of the frames.

[0015] Optionally, generating the average activity image specifically includes:

[0016] The cumulative average of the positron emission tomography images of the plurality of frames is calculated to obtain the average activity image.

[0017] Optionally, inputting the average activity image into a deep neural network after updating weight parameters to obtain a second predicted image specifically includes:

[0018] Inputting the average activity image into a deep neural network after updating weight parameters, and downsampling the average activity image several times by the deep neural network after updating weight parameters to obtain a downsampled average activity image;

[0019] The sampled average activity image is upsampled several times using bilinear interpolation to obtain a second predicted image.

[0020] Optionally, updating the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current multiplicative variable specifically includes:

[0021] Establishing a loss function according to the third input function, the first predicted image, the current label image and the current multiplicative variable;

[0022] The loss function is solved with the goal of minimizing the loss function to obtain updated weight parameters, and the weight parameters of the deep neural network are replaced with the updated weight parameters.

[0023] Optionally, updating the current label image and the current multiplicative variable according to the second predicted image, the current label image and the current multiplicative variable specifically includes:

[0024] De-noising the current label image to obtain a denoised current label image;

[0025] Establishing a label image update function according to the denoised current label image, the current label image, the second predicted image and the current multiplicative variable, and solving the label image update function to obtain an updated current label image;

[0026] An updated current multiplicative variable is obtained according to the second predicted image, the updated current label image and the current multiplicative variable.

[0027] On the other hand, the present invention provides a system for positron emission tomography parameter imaging, comprising a first module, a second module, a third module, a fourth module and a fifth module, wherein:

[0028] The first module is used to obtain scanning data within an acquisition period, generate a first input function before the acquisition period, generate a second input function within the acquisition period according to the scanning data, and generate an average activity image;

[0029] The second module is used to input the average activity image into an initial deep neural network to obtain a first predicted image, and determine whether the first predicted image meets a preset condition;

[0030] The third module is used for, if the first predicted image does not meet the preset condition, concatenating the first input function and the second input function to obtain a third input function, and updating the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current multiplicative variable;

[0031] The fourth module inputs the average activity image into the deep neural network after updating the weight parameters to obtain a second predicted image; if the second predicted image does not meet the preset conditions, the current label image and the current multiplicative variable are updated according to the second predicted image, the second predicted image is used as the first predicted image, and the concatenation of the first input function and the second input function is returned until the second predicted image meets the preset conditions, and the second predicted image is output.

[0032] In another aspect, the present invention provides a device for positron emission tomography parameter imaging, comprising:

[0033] at least one processor;

[0034] at least one memory for storing at least one program;

[0035] When the at least one program is executed by the at least one processor, the at least one processor implements the method described above.

[0036] In another aspect, the present invention provides a computer-readable storage medium having a processor-executable program stored therein. When the processor-executable program is executed by the processor, the program is used to perform the method as claimed in the preceding claim.

[0037] On the other hand, the present invention provides a positron emission tomography parameter imaging system, comprising a positron emission tomography device and a computer device connected to the positron emission tomography device; wherein:

[0038] The positron emission tomography device is used to scan the human body, obtain the scan data within the acquisition period, and upload the scan data within the acquisition period to the computer device;

[0039] The computer device comprises:

[0040] at least one processor;

[0041] at least one memory for storing at least one program;

[0042] When the at least one program is executed by the at least one processor, the at least one processor implements the method described above.

[0043] Implementation of the present invention includes the following beneficial effects: the present invention obtains scanning data within an acquisition period, generates a first input function before the acquisition period, generates a second input function within the acquisition period and generates an average activity image according to the scanning data, splices the first input function and the second input function to obtain a third input function; inputs the average activity image into a deep neural network to obtain a first predicted image, determines whether the first predicted image meets a preset condition, if the first predicted image does not meet the preset condition, updates the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current multiplicative variable, inputs the average activity image into the deep neural network after the weight parameters are updated to obtain a second predicted image, updates the current label image and the current multiplicative variable according to the second predicted image, the current label image and the current multiplicative variable, uses the second predicted image as the first predicted image, returns to execute to determine whether the first predicted image meets the preset condition, until the first predicted image meets the preset condition, outputs the first predicted image, first supplements the input function that has not been collected, splices the input function that has not been collected with the input function of the collected data, combines the input function with the deep neural network for several iterations, and can improve the image quality obtained by short-time scanning parameter imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of the steps of a method for positron emission tomography parameter imaging provided by the present invention;

[0045] Figure 2 is a flow chart of steps for obtaining a second input function provided by the present invention;

[0046] Figure 3 It is a flowchart of the steps of obtaining updated weight parameters of a deep neural network provided by the present invention;

[0047] Figure 4 It is a flowchart of the steps of obtaining PET imaging that satisfies the conditions by iteratively performing a deep neural network provided by the present invention;

[0048] Figure 5 It is a structural schematic diagram of a positron emission tomography parameter imaging system provided by the present invention;

[0049] Figure 6 It is a structural schematic diagram of a positron emission tomography parameter imaging device provided by the present invention;

[0050] Figure 7 It is a structural schematic diagram of another positron emission tomography parameter imaging system provided by the present invention. DETAILED DESCRIPTION

[0051] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0052] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0053] When the following description relates to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims. In the description of the present application, it is necessary to understand that the terms "first / S100", "second / S200", "third / S300", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0054] like Figure 1 As shown, Figure 1 The present invention provides a method for positron emission tomography parameter imaging, comprising the following steps:

[0055] S100, acquiring scanning data within an acquisition period, generating a first input function before the acquisition period, generating a second input function within the acquisition period according to the scanning data, and generating an average activity image.

[0056] Specifically, before imaging, the human body is scanned to obtain scanning data within the acquisition period. The conventional dynamic scanning method of PET is to collect data for 1 hour immediately after the injection of the tracer. In order to shorten the acquisition time, the present invention collects scanning data for another period of time after a period of time after the injection of the drug. For example, after t1 minute of drug injection, t2 minutes of data are collected. Generally, the collection can be started 45 to 60 minutes after the injection of the drug.

[0057] The collected scanning data is not a complete 1 hour, and the input function of the missing part of the t1 minute time period needs to be supplemented. The first supplemented input function is shown in formula (1):

[0058]

[0059] in, is the first input function, A1, A2, A3, λ1, λ2 and λ3 are all parameters of the input function, and their values ​​can be set according to actual conditions, for example, A1=851.1225, A2=21.8798, A3=20.8113, λ1=-4.1339, λ2=-0.1191, λ3=-0.0104.

[0060] First, the scan data of t2 minutes (scan data within the acquisition period) is dynamically divided into frames for image reconstruction, and the frames are divided according to the scanning time to reconstruct multiple frames of PET parameter images. The second input function is obtained based on the multiple frames of PET parameter images. The average activity image z is obtained by calculating the cumulative average of multiple frames of PET parameter images.

[0061] S200: Input the average activity image into an initial deep neural network to obtain a first predicted image, and determine whether the first predicted image meets a preset condition.

[0062] The preset condition may be, but is not limited to, a certain number of iterations.

[0063] The average activity image is input into a deep neural network, which downsamples the average activity image several times, for example, 5 times, and upsamples the downsampled average activity image several times, for example, 5 times, using bilinear interpolation to obtain a first predicted image.

[0064] Among them, the deep neural network is based on the encoder-decoder architecture. The encoder part includes a convolution layer of a preset dimension, such as a 3×3 convolution layer, for downsampling, followed by batch normalization (BN) and leaky linear rectifier unit (LReLU), followed by another convolution layer of a preset dimension, followed by a BN layer and an LReLU layer. After each downsampling and upsampling, the number of feature channels is fixed to 128. The decoder part includes bilinear interpolation upsampling, followed by a convolution layer of a preset dimension, followed by BN and LReLU.

[0065] In this embodiment, five downsampling and upsampling operations can be used to enable the network to capture features of different levels of the average activity image, and better extract and understand the features of the input average activity image. Due to the presence of checkerboard artifacts, the present invention uses bilinear interpolation instead of deconvolution upsampling. In addition, the present invention uses skip connections to connect the encoder path and the decoder path in series to reduce the number of training parameters and include structures of different feature scales.

[0066] S300: If the first predicted image does not meet the preset conditions, concatenate the first input function and the second input function to obtain a third input function, and update the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current multiplicative variable.

[0067] Specifically, the first input function generated in the first step without collecting data before the collection period The first input function is spliced ​​with the second input function of the scan data within t2 minutes. During the splicing, the first input function and the second input function are equal at the time when data collection starts. The third input function obtained by splicing is:

[0068]

[0069] Among them, C p (t) represents the third input function, ω is the proportional constant, where at time t1,

[0070] Specifically, a loss function is established according to the third input function, the first predicted image, the current label image and the current multiplicative variable, the loss function is solved with the goal of minimizing the loss function to obtain updated weight parameters, and the weight parameters of the deep neural network are replaced with the updated weight parameters.

[0071] S400. Input the average activity image into the deep neural network after updating the weight parameters to obtain a second predicted image. If the second predicted image does not meet the preset conditions, update the current label image and the current multiplicative variable according to the second predicted image, the current label image and the current multiplicative variable, use the second predicted image as the first predicted image, return to execute the splicing of the first input function and the second input function, until the second predicted image meets the preset conditions, and output the second predicted image.

[0072] Specifically, the average activity image is input into the deep neural network after updating the weight parameters to obtain a second predicted image, the current label image is denoised to obtain a denoised current label image, a label image update function is established based on the denoised current label image, the current label image, the second predicted image and the current multiplicative variable, the label image update function is solved to obtain the updated current label image, and the updated current multiplicative variable is obtained based on the second predicted image, the updated current label image and the current multiplicative variable.

[0073] After updating the weight parameters of the deep neural network, the current label image and the current multiplicative variable, the current label image before updating in step S300 is replaced by the updated current label image, and the current multiplicative variable before updating in step S300 is replaced by the updated current multiplicative variable, and the second predicted image is used as the first predicted image, and the execution of step S300 is returned until the second predicted image meets the preset conditions, and the second predicted image is output as the final PET imaging.

[0074] Taking 2 iterations as an example, in the first iteration, the current label image is K 1 and the current multiplicative variable u 1 , input the average activity image into the deep neural network to obtain the first predicted image F(θ 1 |z), F(θ 1 |z) does not meet the preset conditions, according to the third input function, K 1 、u 1 and F(θ 1 |z) updates the weight parameters of the deep neural network. In the second iteration, the average activity image is input into the updated weight parameter deep neural network to obtain the second predicted image F(θ 2 |z), F(θ 2 |z), does not meet the conditions, according to F(θ 2 |z)K 1 and u 1 Update K 1 and u 1 , get K 2 and u 2 , then, according to K 2 and u 2 、F(θ 2 |z) Update θ 2 , get a new deep neural network and perform the third iteration.

[0075] In some embodiments, Figure 2 As shown, Figure 2 The method is a flow chart of steps for obtaining a second input function, wherein the second input function within a collection period is generated according to the scan data and specifically includes:

[0076] S110, dividing the scan data into frames according to the scan time to obtain a plurality of frames of scan data.

[0077] Specifically, for example, the acquisition time period is 15 minutes, and 15 minutes of scanning data is obtained. The dynamic frame division is 1 minute / frame, and the 15 minutes of scanning data is divided into frames at 1 minute / frame to obtain 15 frames of 1 minute of scanning data.

[0078] S120, performing image reconstruction according to the scanning data of each frame to obtain positron emission tomography images of several frames.

[0079] Specifically, image reconstruction is performed according to the scanning data of each frame to obtain a frame-by-frame positron emission tomography image. For example, image reconstruction is performed according to the scanning data of the first frame of 1 minute to obtain a positron emission tomography image of the first frame.

[0080] S130: Obtain a second input function according to pixel values ​​of the plurality of frame-based positron emission tomography images.

[0081] Specifically, a second input function is established from pixel values ​​at the descending aorta of the reconstructed dynamic PET parameter images of several frames, and the second input function is used to describe the activity of the tracer in the plasma.

[0082] In some embodiments, Figure 3 As shown, Figure 3 The invention is a flowchart of steps for obtaining weight parameters for updating a deep neural network. In step S300, the weight parameters of the deep neural network are updated according to the third input function, the first predicted image, the current label image and the current multiplicative variable, specifically including:

[0083] S310: Establish a loss function according to the third input function, the first predicted image, the current label image and the current multiplicative variable.

[0084] Specifically, the PET imaging principle of the present invention combining deep neural network (deep learning) and regularized denoising is first analyzed:

[0085] The third input function uses the conventional Patlak method for parameter imaging. The conventional Patlak method fitting function is shown in formula (3):

[0086]

[0087] Among them, C T (t) represents the activity value of the target tissue, C p (t) represents the third input function of the artery, which is the tracer concentration in arterial blood, V E represents the blood volume fraction, t represents the total time of dynamic collection, and t * Indicates the time point when the drug distribution in the body reaches dynamic equilibrium, K i It is the tissue uptake rate constant. After the flow reaches equilibrium, the ratio of drug solubility in tissue to that in plasma can be used as a quantitative evaluation indicator.

[0088] Among them, t *Generally, the 20th minute is taken. If the dynamic collection time t is 60 minutes, then the time period used for Patlak calculation is 20-60 minutes.

[0089] In the present invention, Where J represents the number of pixels in each frame of the PET parameter image. The present invention rewrites the PET imaging formula (3) as shown in formula (4):

[0090]

[0091] Wherein, diag(·) represents a diagonal matrix, I represents an all-1 matrix, and when Patlak estimation is performed on all dynamic frame PET parameter images obtained in step S120, PET imaging is performed according to formula (5):

[0092]

[0093] in, T represents the frame number of dynamic PET parameter images.

[0094] In the present invention, it is desired to obtain high-quality PET imaging, so the depth image prior and the regularization denoising term are combined in formula (5), as shown in formula (6):

[0095]

[0096] Wherein, F(·) represents the output of the deep neural network, θ represents the unknown weight parameter in the deep neural network, z represents the input of the deep neural network, K represents the label image, in the present invention, z is the average activity image of the dynamic PET parameter image, and f(·) represents the regularized denoising term, which can adopt Gaussian, non-local mean filtering, or denoising network and other methods.

[0097] In order to solve formula (6), we rewrite it into the augmented Lagrangian form, as shown in formula (7):

[0098]

[0099] Among them, u represents the Lagrange multiplicative variable, and λ and ω are selected parameters.

[0100] The solution of formula (7) is equivalent to the solution of formula (6). The purpose of the present invention is to solve equation (7) to obtain the final PET imaging that meets the conditions. Steps S300 to S400 are the process of solving formula (7) after multiple iterations. In the first iteration, the current label image K of the first iteration is 1According to formula (3), the current multiplicative variable is the Lagrange multiplicative variable, which is a number close to 0 and greater than 0.

[0101] So, if Figure 4 As shown, Figure 4 It is a flowchart of the steps of deep neural network iteration to obtain PET imaging that meets the conditions. According to formula (7), the loss function at the nth iteration can be obtained, as shown in formula (8):

[0102]

[0103] in, represents the solution of θ when v takes the minimum value. The superscript n represents the nth iteration, and n+1 represents the next iteration. θ n+1 is the updated weight parameter, K n represents the current label image, F(θ n |z) represents the first predicted image, Y and X are obtained according to the third input function, as shown in the above formula (4), u n is the current Lagrange multiplicative variable.

[0104] S320, solving the loss function with the goal of minimizing the loss function to obtain updated weight parameters, and replacing the weight parameters of the deep neural network with the updated weight parameters.

[0105] Specifically, solve the above formula (8) to obtain θ m+1 As the updated weight parameter, and using θ n+1 Replace θ n Get a new deep neural network.

[0106] Among them, θ m+1 As the weight parameter of the deep neural network at the n+1th iteration.

[0107] In some embodiments, the average activity image is input into a deep neural network after updating weight parameters to obtain a second predicted image, specifically comprising:

[0108] S410, inputting the average activity image into the deep neural network after updating the weight parameters, and the deep neural network after updating the weight parameters downsamples the average activity image several times to obtain a downsampled average activity image.

[0109] The structure of the deep neural network after updating the weight parameters is the same as the deep neural network structure in step S200, except that the weight parameters are different. In this step, the weight parameters are θ n+1 , downsampling the average activity image several times to obtain the downsampled average activity image, which is the same as the upsampling process in the above step S300.

[0110] S420 , up-sample the sampled average activity image several times by using bilinear interpolation to obtain a second predicted image.

[0111] Specifically, the sampled average activity image is upsampled several times using bilinear interpolation to obtain the second predicted image, which is the same as the downsampling process of the above step S300.

[0112] Among them, the second predicted image F(θ m+1 |z) as the first predicted image when step S300 updates the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current multiplicative variable at the n+1th iteration.

[0113] In some embodiments, in step S400, updating the current label image and the current multiplicative variable according to the second predicted image, the current label image and the current multiplicative variable specifically includes:

[0114] S410: De-noise the current label image to obtain a denoised current label image.

[0115] Specifically, for the current label image K n Denoise and obtain the denoised current label image f(K n ).

[0116] S420, establishing a label image update function according to the denoised current label image, the current label image, the second predicted image and the current multiplicative variable, and solving the label image update function to obtain an updated current label image.

[0117] Specifically, as shown in formula (9):

[0118]

[0119] Among them, K n+1 The updated current label image is used as the current label image of the n+1th iteration step S300, K n is the current label image of this iteration, F(θ n+1 |z) is the second predicted image output by the deep neural network after updating the weight parameters, u n is the current multiplicative variable at the nth iteration, f(K n ) The denoised current label image.

[0120] The solution process of formula (9) is as follows:

[0121] Formula (9) is solved by setting the derivative with respect to K to zero using the fixed-point strategy, Find the derivative of H(K) with respect to K Available here is a convex function. Taking the derivative of formula (9) and setting it to zero, we get:

[0122] λ(K n+1 -f(K n ))+ω(K n+1 -F(θ n+1 |z)-u n )=0 (10)

[0123] Therefore, formula (9) is solved as follows:

[0124]

[0125] Through the above solution, a parameter image with low noise and accurate quantitativeness can be obtained during a short-time whole-body dynamic scan.

[0126] S430 , obtaining an updated current multiplicative variable according to the second predicted image, the updated current label image and the current multiplicative variable.

[0127] Specifically, the updated current multiplicative variable u is obtained n+1 As shown in formula (12):

[0128] u n+1 =u n -K n+1 +F(θ n+1 |z) (12)

[0129] Among them, u n+1 It will be used as the current multiplicative variable of step S300 at the n+1th iteration.

[0130] The implementation of the present invention includes the following beneficial effects:

[0131] The present invention first supplements the input function that has not been collected, splices the collected input function with the input function of the collected data, and combines the input function with the deep neural network for several iterations, which can improve the image quality obtained by short-time scanning parameter imaging.

[0132] The present invention, through steps S410-S430, iterates to obtain the final PET imaging, denoises the label image, updates the label image according to the label image after denoising in this iteration and the deep neural network input to obtain the label image of the next iteration, and considers the label image when updating the weight parameters. Each iteration can reduce the noise of PET imaging obtained by the deep neural network after updating the weight parameters. After several iterations, the noise of the obtained PET imaging is reduced, the quantitative deviation is reduced, and the image quality obtained by short-time scanning parameter imaging is improved.

[0133] The deep neural network model of the present invention can be run on a GPU (NVIDIA RTX 4090, memory is 24GB). The Adam optimizer implemented by PyTorch 1.1 is used on the Pycharm platform, and the mean square error between the label image and the network output is used as the loss function. Considering that the deep image prior (DIP) method is prone to overfitting, the quality of the network output will drop rapidly after a certain number of iterations. The learning rate used in the present invention decays with the number of iterations to alleviate the overfitting phenomenon. Typically, the initial learning rate is set to decay once every 1000 iterations. By decaying the learning rate, the network output can avoid a sudden drop due to an excessively large learning rate.

[0134] In order to compromise the convergence speed and performance of the proposed method, this paper adopts a fixed point strategy to solve formula (11) and sets the hyperparameters λ = 0.2 and ω = 0.2.

[0135] like Figure 5 As shown, Figure 5 The present invention is a structural schematic diagram of a positron emission tomography parameter imaging system. The present invention also provides a positron emission tomography parameter imaging system, including a first module, a second module, a third module and a fourth module, wherein:

[0136] The first module is used to obtain scanning data within an acquisition period, generate a first input function before the acquisition period, generate a second input function within the acquisition period according to the scanning data, and generate an average activity image;

[0137] The second module is used to input the average activity image into an initial deep neural network to obtain a first predicted image, and determine whether the first predicted image meets a preset condition;

[0138] The third module is used for, if the first predicted image does not meet the preset condition, concatenating the first input function and the second input function to obtain a third input function, and updating the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current multiplicative variable;

[0139] The fourth module is used to input the average activity image into the deep neural network after updating the weight parameters to obtain a second predicted image. If the second predicted image does not meet the preset conditions, the current label image and the current multiplicative variable are updated according to the second predicted image, the second predicted image is used as the first predicted image, and the first input function and the second input function are returned to execute the splicing until the second predicted image meets the preset conditions, and the second predicted image is output.

[0140] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0141] like Figure 6 As shown, Figure 6 The present invention is a schematic diagram of the structure of a device for positron emission tomography parameter imaging. The present invention also provides a device for positron emission tomography parameter imaging, including:

[0142] at least one processor;

[0143] at least one memory for storing at least one program;

[0144] When the at least one program is executed by the at least one processor, the at least one processor implements the method steps of positron emission tomography parameter imaging described in the above method embodiment.

[0145] Among them, the memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. The memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a remote memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0146] It can be seen that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0147] In addition, the embodiment of the present application also discloses a computer program product or a computer program, and the computer program product or the computer program is stored in a computer-readable storage medium. The processor of the computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device performs the above method. Similarly, the contents in the above method embodiment are all applicable to the storage medium embodiment, and the functions specifically implemented by the storage medium embodiment are the same as those in the above method embodiment, and the beneficial effects achieved are also the same as those achieved by the above method embodiment.

[0148] The present invention also provides a computer-readable storage medium, which stores a program executable by a processor. The program executable by the processor is used to implement the above method when executed by the processor.

[0149] It is understood that all or some steps and systems in the disclosed method above can be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, magnetic tape, disk storage or other magnetic storage device, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0150] like Figure 7 As shown, Figure 7 is a structural schematic diagram of another positron emission tomography parameter imaging system. The present invention also provides a positron emission tomography parameter imaging system, including a positron emission tomography device and a computer device connected to the positron emission tomography device; wherein,

[0151] The positron emission tomography device is used to scan the human body, obtain the scan data within the acquisition period, and upload the scan data within the acquisition period to the computer device;

[0152] The computer device comprises:

[0153] at least one processor;

[0154] at least one memory for storing at least one program;

[0155] When the at least one program is executed by the at least one processor, the at least one processor implements the method of positron emission tomography parameter imaging.

[0156] Specifically, the positron emission tomography device may be an existing positron emission tomography scanner; and the computer device may be different types of electronic devices, including but not limited to desktop computers, laptops and other terminals.

[0157] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0158] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0159] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A method for positron emission tomography parametric imaging, characterized in that: The following steps are involved: Acquire scanning data within an acquisition period, generate a first input function before the acquisition period, generate a second input function within the acquisition period according to the scanning data, and generate an average activity image; Inputting the average activity image into an initial deep neural network to obtain a first predicted image, and determining whether the first predicted image meets a preset condition; If the first predicted image does not meet the preset conditions, the first input function and the second input function are concatenated to obtain a third input function, and the weight parameters of the deep neural network are updated according to the third input function, the first predicted image, the current label image and the current Lagrangian multiplicative variable; wherein the current label image is a positron emission tomography parameter image calculated by the Patlak method; Input the average activity image into the deep neural network after updating the weight parameters to obtain a second predicted image; if the second predicted image does not meet the preset conditions, update the current label image and the current Lagrangian multiplicative variable according to the second predicted image, the current label image and the current Lagrangian multiplicative variable, use the second predicted image as the first predicted image, return to execute the splicing of the first input function and the second input function until the second predicted image meets the preset conditions, and output the second predicted image; Wherein, updating the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current Lagrangian multiplicative variable specifically includes: Establishing a loss function according to the third input function, the first predicted image, the current label image and the current Lagrangian multiplicative variable; Solving the loss function with the goal of minimizing the loss function to obtain updated weight parameters, and replacing the weight parameters of the deep neural network with the updated weight parameters; Wherein, the loss function is: in, represents the solution of θ when v takes the minimum value; the superscript n represents the nth iteration, and n+1 represents the next iteration; θ n+1 is the updated weight parameter; K n represents the current label image; F(θ n |z) represents the first predicted image, u n is the current Lagrangian multiplicative variable, z is the average activity image of the dynamic PET parameter image, PET stands for positron emission tomography; Y and X are obtained according to the third input function, t represents the total time of dynamic acquisition, t * represents the time point when the drug distribution in the body reaches dynamic equilibrium, T represents the number of frames of dynamic PET parameter images, diag(·) represents a diagonal matrix, I represents a matrix of all 1s, C T (t) represents the activity value of the target tissue, C p (t) is the third input function, J represents the number of pixels in each frame of the PET parameter image; ω is the proportional constant.

2. The method according to claim 1, characterized in that The step of generating a second input function within a collection period according to the scan data specifically includes: Divide the scan data into frames according to the scan time to obtain a plurality of frames of scan data; Performing image reconstruction according to the scanning data of each frame to obtain positron emission tomography images of several frames; A second input function is obtained according to pixel values ​​of a plurality of the positron emission tomography images of the frames.

3. The method according to claim 2, characterized in that The generating of the average activity image specifically comprises: The cumulative average of the positron emission tomography images of the plurality of frames is calculated to obtain the average activity image.

4. The method according to claim 1, characterized in that: The step of inputting the average activity image into a deep neural network after updating weight parameters to obtain a second predicted image comprises the following steps: Inputting the average activity image into a deep neural network after updating weight parameters, and downsampling the average activity image several times by the deep neural network after updating weight parameters to obtain a downsampled average activity image; The sampled average activity image is upsampled several times using bilinear interpolation to obtain a second predicted image.

5. The method according to any one of claims 1 to 4, characterized in that: The updating of the current label image and the current Lagrangian multiplicative variable according to the second predicted image, the current label image and the current Lagrangian multiplicative variable specifically includes: De-noising the current label image to obtain a denoised current label image; Establishing a label image update function according to the denoised current label image, the current label image, the second predicted image and the current Lagrangian multiplicative variable, and solving the label image update function to obtain an updated current label image; An updated current Lagrangian multiplicative variable is obtained according to the second predicted image, the updated current label image and the current Lagrangian multiplicative variable.

6. A positron emission tomography parameter imaging system, characterized in that: It includes a first module, a second module, a third module, a fourth module and a fifth module, wherein: The first module is used to obtain scanning data within an acquisition period, generate a first input function before the acquisition period, generate a second input function within the acquisition period according to the scanning data, and generate an average activity image; The second module is used to input the average activity image into an initial deep neural network to obtain a first predicted image, and determine whether the first predicted image meets a preset condition; The third module is used for, if the first predicted image does not meet the preset conditions, concatenating the first input function and the second input function to obtain a third input function, and updating the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current Lagrangian multiplicative variable; wherein the current label image is a positron emission tomography parameter image calculated by the Patlak method; The fourth module inputs the average activity image into the deep neural network after updating the weight parameters to obtain a second predicted image; if the second predicted image does not meet the preset conditions, the current label image and the current Lagrangian multiplicative variable are updated according to the second predicted image, the current label image and the current Lagrangian multiplicative variable, the second predicted image is used as the first predicted image, and the first input function and the second input function are returned to be spliced ​​until the second predicted image meets the preset conditions, and the second predicted image is output; Wherein, updating the weight parameters of the deep neural network according to the third input function, the first predicted image, the current label image and the current Lagrangian multiplicative variable specifically includes: Establishing a loss function according to the third input function, the first predicted image, the current label image and the current Lagrangian multiplicative variable; Solving the loss function with the goal of minimizing the loss function to obtain updated weight parameters, and replacing the weight parameters of the deep neural network with the updated weight parameters; Wherein, the loss function is: in, represents the solution of θ when v takes the minimum value; the superscript n represents the nth iteration, and n+1 represents the next iteration; θ n+1 is the updated weight parameter; K n represents the current label image; F(θ n |z) represents the first predicted image, u n is the current Lagrangian multiplicative variable, z is the average activity image of the dynamic PET parameter image, PET stands for positron emission tomography; Y and X are obtained according to the third input function, t represents the total time of dynamic acquisition, t * represents the time point when the drug distribution in the body reaches dynamic equilibrium, T represents the number of frames of dynamic PET parameter images, diag(·) represents a diagonal matrix, I represents a matrix of all 1s, C T (t) represents the activity value of the target tissue, C p (t) is the third input function, J represents the number of pixels in each frame of the PET parameter image; ω is the proportional constant.

7. A device for positron emission tomography parameter imaging, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 5 when executed by the processor.

9. A system for positron emission tomography parameter imaging, characterized in that: The invention comprises a positron emission tomography device and a computer device connected to the positron emission tomography device; wherein: The positron emission tomography device is used to scan the human body, obtain the scan data within the acquisition period, and upload the scan data within the acquisition period to the computer device; The computer device comprises: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 5.

Images