CN116172599A

CN116172599A - PET radioactivity distribution acquisition method and PET system

Info

Publication number: CN116172599A
Application number: CN202310193038.4A
Authority: CN
Inventors: 李楠
Original assignee: Sinounion Healthcare Inc
Current assignee: Sinounion Healthcare Inc
Priority date: 2023-02-27
Filing date: 2023-02-27
Publication date: 2023-05-30

Abstract

The invention relates to a PET radioactivity distribution acquisition method and a PET system, wherein the method comprises the following steps: for detection data used for medical image reconstruction, a first PET image which is not subjected to attenuation correction and a second PET image which is subjected to approximate attenuation correction are obtained, according to a pre-constructed mapping from a PET prior image alpha to a linear attenuation coefficient image mu and a known unknown number which is a log likelihood function of a PET radioactivity distribution x and a linear attenuation coefficient distribution mu, a log likelihood function with a constraint term is obtained, the constraint term is the mapping from the PET prior image alpha to the linear attenuation coefficient image mu, an iterative algorithm is adopted to maximize the log likelihood function of the constraint term, and when an end condition is reached, an estimated value of the PET radioactivity distribution x is obtained. The method can ensure the rapid convergence of the iterative process and increase the stability, the quantification and the accuracy of the reconstruction algorithm.

Description

PET radioactivity distribution acquisition method and PET system

Technical Field

The invention relates to the technical field of medical imaging, in particular to a PET radioactivity distribution acquisition method and a PET system.

Background

Positron emission tomography PET (Positron Emission Tomography) is one of the most advanced clinical examination imaging techniques in the current nuclear medicine field. The positron emitted by the radionuclide annihilates with an electron in the human body to generate two photons with opposite directions and 511keV energy. Photon pairs are attenuated in the body before they are acquired by the PET system, and photons on the body surface have greater detection efficiency for events than those inside the body. If such radiation attenuation effects are not corrected, attenuation artifacts can be produced in the reconstructed image, where the body edge image is too bright and the internal tissue image is too dark. The PET system is generally integrated with other modal systems (such as CT, MRI and the like) to acquire anatomical structure imaging of a patient, so that on one hand, the nuclide distribution condition can be accurately positioned, and the accuracy of focus positioning is improved; on the other hand, the tissue density distribution of the patient is correspondingly obtained and used for attenuation correction in PET image reconstruction, so that the accurate distribution of the radiopharmaceuticals in the patient can be obtained. The PET functional imaging and anatomical structure imaging of other modes are finally integrated, and the advantages of functional imaging and anatomical imaging are compatible, so that the purposes of early finding focus and diagnosing diseases are achieved, and the method has more advantages for diagnosis and treatment guidance of tumor, heart and brain diseases.

However, in multi-modality acquisition applications, attenuation information matching the PET image is sometimes not accurately obtained, resulting in attenuation correction errors, and ultimately, significant artifacts on the PET image. In order to accurately correct attenuation artifacts and widen the application scene of PET imaging, the key point is whether attenuation information can be directly extracted from PET acquired data without depending on other modality imaging. The existing algorithm needs to go through multiple iterations to approach the ideal value, which results in excessively long iterative convergence operation time, and generally needs to be matched with higher level computing resources (such as a high-performance GPU), thus increasing the use cost. In addition, the iterative algorithm cannot ensure that the calculation result converges to the globally optimal solution, but may converge to the locally optimal solution. To avoid this, many constraints and protections need to be added to the iterative algorithm, and adjustment parameters need to be set, which reduces the stability and robustness of the algorithm.

Disclosure of Invention

First, the technical problem to be solved

In view of the above-mentioned drawbacks and deficiencies of the prior art, the present invention provides a method for acquiring a PET radioactivity distribution and a PET system.

(II) technical scheme

In order to achieve the above purpose, the main technical scheme adopted by the invention comprises the following steps:

in a first aspect, an embodiment of the present invention provides a method for acquiring a PET radioactivity distribution, including:

s10, acquiring a first PET image which is not subjected to attenuation correction and a second PET image which is subjected to approximate attenuation correction aiming at detection data used for reconstructing a medical image, wherein the second PET image is an image reconstructed after the attenuation correction is performed on the first PET image based on a linear attenuation correction coefficient empirical value of a designated area;

s20, according to a pre-constructed mapping from the PET prior image alpha to the linear attenuation coefficient image mu and a known unknown number which is a log likelihood function of the PET radioactivity distribution x and the linear attenuation coefficient distribution mu, obtaining a log likelihood function with a constraint term, wherein the constraint term is the mapping from the PET prior image alpha to the linear attenuation coefficient image mu;

the PET prior image alpha is a first PET image which is not subjected to attenuation correction and/or a second PET image which is subjected to approximate attenuation correction; the unknowns in the log likelihood function with the constraint term are as follows: x, μ and parameters of the map;

s30, maximizing the log-likelihood function of the constraint item by adopting an iterative algorithm according to the log-likelihood function with the constraint item, and obtaining an estimated value of the PET radioactivity distribution x when the end condition is reached;

the iterative algorithm maximizes the initial values of x, μ and the mapped parameters in the log likelihood function of the constraint term to specified values.

The end condition of this embodiment is to iterate until the log likelihood function converges, and the iteration time and noise effect can be considered in practical application, and the recommended iteration termination value is usually given empirically.

Optionally, before the step S20, a mapping from the PET a priori image α to the linear attenuation coefficient image μ is pre-constructed;

the mapping includes:

μ=f (θ|α) formula (1)

f represents the mapping of the training model, θ represents the parameters of the mapping, and α represents the input of the training model.

Optionally, the S20 includes:

for the known log-likelihood functions L and μ=f (θ|α), the log-likelihood functions with constraint terms are obtained as:

wherein ρ is a super parameter for adjusting the weight between the log likelihood function and the learning network penalty term, Δμ is the change step length of the linear attenuation coefficient image μ, x represents the PET radioactivity distribution, and y represents the detection data.

Alternatively, for equation (2), an alternating strategy is employed to maximize the log-likelihood function L with constraint terms based on a known solution algorithm _ρ Obtaining a log-likelihood function L meeting maximization constraint terms _ρ The required estimated value of x is taken as an output value;

wherein the alternating iteration strategy comprises: in the iterative process, only one variable is updated in each step, and the other two variables are fixed, so that the updating is alternately repeated;

specifically, the mapping parameter θ is an initial value of an accelerated convergence speed generated by pre-training of a CNN network, a Unet network or a GAN network as a training model;

a second image with an initial value of 0 or approximate attenuation correction of the attenuation coefficient distribution μ

The initial value of the PET radioactivity distribution x is a set value; namely, the numerical value of the full-space normal number distribution/the pixel value of the full space are 1000;

the known solution algorithm is MLEM, OSEM or MAP.

In this embodiment, in order to accelerate the convergence speed, the pre-training network may first generate a general network parameter as an iteration initial value of the network parameter, and then fine-tune for a specific scan patient.

The pre-training described above can be understood as using a large amount of training data to train a model to complete the mapping from a priori PET image to a linear attenuation coefficient image. The pre-trained network parameters may be used as the starting value for the network tuning process, allowing the model to converge faster.

Optionally, keeping the PET radioactivity distribution x constant for equation (2), maximizing the log likelihood function L with constraint terms for the unknown attenuation coefficient distribution μ and mapping parameters θ _ρ The method utilizes an alternating direction multiplier ADMM iterative algorithm to solve, and the process comprises the following three steps:

/>

where n represents the current iteration number and n iterations satisfy the end condition (i.e., the recommended iteration termination value); decoupling the constraint optimization problem into a network training sub-problem-corresponding formula (8) with an L2 norm as a loss function and a solution sub-problem-corresponding formula (9) with a penalty term mu in using an ADMM algorithm;

specifically, the linear attenuation coefficient distribution mu and the increment step delta mu are kept unchanged, and theta is updated based on a formula (8); iteratively updating μ based on equation (9) and then holding θ and Δμ unchanged; and (3) keeping mu and theta unchanged based on a formula (10), updating delta mu, and iteratively approaching the optimal solution.

Δμ is an intermediate variable added to solve for μ and θ, and represents the amount of change in μ. This change in μ derives from equation (2) because a separate constraint on Δμ (i.e., L2-Norm for Δμ) is required in the log-likelihood function construction, ensuring that the magnitude of μ change is not too great.

Optionally, when the medical image is a PET image, before S10, the method further includes:

s00, acquiring a training sample for training a model based on the reconstructed medical image and the matched associated image;

wherein each training sample comprises: the reconstructed PET image/simulated PET image comprises an approximate linear attenuation correction coefficient corresponding to the PET image and other mode images corresponding to the PET image, wherein the other mode images are used for acquiring a real linear attenuation correction coefficient, and the real linear attenuation correction coefficient is used for verifying whether a trained model is converged or not;

s01, training a model based on a training sample to obtain a trained model;

mapping the parameter theta in the trained model enables the loss function phi value of the optimized training model to be minimum.

Optionally, training the model based on the training samples includes:

inputting the approximate linear attenuation correction coefficient of each training sample into a model to obtain output, and comparing the output with the real linear attenuation correction coefficients obtained by other mode images in the training sample by means of a loss function phi;

and/or the number of the groups of groups,

inputting reconstructed unattenuated PET images of each training sample into a model to obtain output, and comparing the output with real linear attenuation correction coefficients obtained by other mode images in the training sample by means of a loss function phi;

and/or the number of the groups of groups,

summing the reconstructed unattenuated PET image of each training sample with the approximate linear attenuation correction coefficient, inputting the summed image into a model to obtain an output, and comparing the output with the true linear attenuation correction coefficient obtained by other mode images in the training sample by means of a loss function phi;

the loss function phi is one or more of L1 norm, L2 norm and KL divergence, and is used for the similarity between each output of the model and the true linear attenuation correction coefficient to which the output belongs in scale training.

In a second aspect, embodiments of the present invention also provide a PET system, comprising: a memory and a processor; the memory has stored therein computer program instructions, and the processor executes the computer program instructions stored in the memory, specifically performing the method of any of the first aspects.

In a third aspect, embodiments of the present invention also provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs a method of acquiring a PET radioactivity distribution according to any of the first aspects.

(III) beneficial effects

In order to improve the convergence speed and the result stability of the linear attenuation coefficient iterative algorithm, in the embodiment of the invention, the mapping for mapping the PET prior image into the linear attenuation coefficient image can be constructed in advance, and the mapping parameters and the linear attenuation coefficient are added into the iterative solving process of the maximum likelihood function L of the PET radioactivity distribution. The current algorithm iterates by taking the more accurate linear attenuation coefficient estimation as a starting point so as to achieve the purposes of optimizing a convergence path and converging to a global optimal solution as soon as possible, and the stability, the quantification and the accuracy of the algorithm are improved.

Further, the mapping parameters suitable for the current PET scanning data are obtained in a self-adaptive mode by adopting the deep learning network mapping, so that more accurate linear attenuation coefficient image estimation is obtained, the method is applied to a maximum likelihood function L iterative optimization algorithm, compared with an initial value (a full imaging visual field uniform linear attenuation coefficient image or a linear attenuation coefficient image obtained by converting other modal images) used by an original algorithm, the method is independent of other modal images, and meanwhile, a convergence path can be effectively optimized, and the reliability is improved.

In addition, attenuation information extracted by deep learning is derived from PET images, mismatch between multi-mode images does not exist, and motion and truncation artifacts can be avoided in the convergence adjustment process.

Drawings

FIG. 1 is a flow chart of a method for obtaining a PET radioactivity distribution according to an embodiment of the present invention;

FIG. 2 (a) is a schematic diagram of a PET image without attenuation correction;

FIG. 2 (b) is a schematic illustration of an approximately attenuation corrected PET image;

FIG. 2 (c) is a schematic illustration of a PET image attenuation corrected using the method of the present invention;

fig. 2 (d) is a schematic diagram of an estimation result of a linear attenuation coefficient prior distribution obtained by using a deep learning network.

Detailed Description

The invention will be better explained by the following detailed description of the embodiments with reference to the drawings.

In the multi-modality acquisition application of the prior art, the attenuation information matched with the PET data cannot be obtained accurately sometimes, which leads to attenuation correction errors, so that extra artifacts are generated on the PET image.

The concrete explanation is as follows:

first, in PET multi-modality imaging, there may be positional misalignment of images of different modalities. In long PET scans, the patient's body may move (e.g., the arm, head, etc. may move during the scan time), which may also cause the PET image to be mismatched with images of other modalities, creating significant attenuation artifacts.

Second, there is a possibility that false positives occur. To eliminate the possibility of false positives, an additional scan is required to delay the scan, which increases the patient's X-ray radiation dose.

Furthermore, the scan range of PET is typically greater than that of other modalities (such as CT or MRI). When scanning a patient of relatively large body weight, other modality imaging is likely not to provide a sufficiently large imaging range, resulting in a truncation of the attenuation image. The application of such incomplete attenuation information in PET attenuation correction also produces attenuation artifacts.

Finally, the application conditions of other modalities also restrict the application of PET imaging, for example, patients with dentures or cardiac pacemakers cannot do MR examination, and thus affect the application of PET/MR. In addition, CT imaging requires extremely high radiation protection requirements, and MR imaging requires strict nuclear magnetic resonance shielding, which leads to high scanning protection requirements of multi-mode imaging and is not easy to popularize.

In order to be able to correct attenuation artefacts accurately, the application of PET imaging is widened, and attenuation information is extracted directly from PET acquisition data independent of other modality imaging. Meanwhile, the operation speed and the result stability of the linear attenuation coefficient iterative algorithm are improved, and the invention provides a method for acquiring the PET radioactivity distribution by means of the information of the mapping from the PET image to the linear attenuation coefficient image.

For a better understanding of the embodiments of the present invention, some words are described below:

the PET reconstructed image is PET radioactivity distribution/PET radioactivity distribution image x;

the attenuation correction coefficient, the linear attenuation correction coefficient, the linear attenuation coefficient image and the linear attenuation correction coefficient image all represent one meaning, and different descriptions are adopted in different embodiments.

Example 1

As shown in fig. 1, an embodiment of the present invention provides a method for acquiring a PET radioactivity distribution, where the main execution subject of the method of the embodiment may be a control device/electronic device of the method for acquiring a PET radioactivity distribution, and the control device may be integrated in an acquisition device of a PET system or in a separate computer processing device, and the method for acquiring a PET radioactivity distribution includes the following steps:

In this embodiment, the method can iterate to the convergence of the log-likelihood function, and can combine the iteration time and noise with the iteration termination value of experience to determine the final end condition when the method is applied specifically.

It will be appreciated that, prior to step S20 described above, a mapping from the PET a priori image α to the linear attenuation coefficient image μmay be constructed;

the mapping includes:

μ=f (θ|α) formula (1)

The training model of the present embodiment may be a CNN network, a Unet network, or a GAN network.

Compared with the traditional method for PET reconstruction by combining other mode images, the attenuation correction information in the PET reconstruction process of the embodiment comes from PET data, so that when the PET multi-mode images are not matched due to respiration or heartbeat and patient movement, the image can be still subjected to attenuation correction, and image artifacts are eliminated; if the attenuation images obtained by other modes have artifacts (such as PET/CT scanning patients with cardiac pacemakers or metal dental braces in vivo and CT images have obvious metal artifacts), accurate attenuation correction can still be performed.

Example two

For a better understanding of the method described in the first embodiment, the following description will specifically exemplify the two methods with reference to the formulas. The method of the embodiment ensures that the radioactivity distribution estimated value is quickly and stably converged. The following method is described by fusing the training process and the use process together. The method comprises the following specific steps:

the following steps 01 to 03 are all conventional steps, and the present embodiment is not modified.

The invention provides a method for acquiring accurate linear attenuation coefficient image estimation from PET priori image mapping by utilizing a deep learning network and training network parameters/mapping parameters in real time in the iterative solving process of linear attenuation coefficients, which is used for adjusting a convergence path of radioactivity distribution and ensuring rapid and stable convergence of radioactivity distribution solving. The method comprises the following specific steps:

step 01:the PET acquisition process can be modeled as the following formula:

y= [ y ] in formula (1) _1t ,y _2t ,…,y _it ,…,y _NT ]' means that the detected data is the detection data,

the average value of the detection data is represented by N, the size of the detection data sinogram, T, the size of the time-of-flight TOF discrete space, i, the index (index) of the detection data sinogram response line LOR (line of response), and T, the index of the time-of-flight TOF discrete space. The prime superscript indicates a matrix transpose operation. x= [ x ] ₁ ,x ₂ ,…,x _j ,…,x _M ]' represents an unknown activity distribution image, M represents the size of the activity distribution image space, j represents the index of the variation of the activity distribution image space, and represents the point source of the corresponding spatial position. Mu= [ mu ] ₁ ,μ ₂ ,…,μ _k ,…,μ _K ]' represents an unknown linear attenuation coefficient image, K represents the size of the linear attenuation coefficient image space, K represents the index of variation of the linear attenuation coefficient image space, and represents the point source of the corresponding spatial position. A= [ a ] _ijt ]The probability that the spatial point source j is detected by the response line LORi and the time of flight TOF is t in the PET system is expressed in a mathematical form as a system matrix, and the physical characteristic of the system is reflected, namely, l= [ l ] _ik ]The linear attenuation coefficient matrix represents the track crossing length of the LOR i when passing through the spatial position point source k. r= [ r ] _1t ,r _2t ,…,r _it ,…,r _NT ]' represents the average of random noise and scattered noise.

Step 02:the PET detection data obey poisson distribution, and unknowns are PET radioactivity distribution x and linear attenuation coefficient distribution mu. The log-likelihood function of the probe data is expressed as:

step 03:taking equation (1) into equation (2), ignoring terms that are not related to unknowns, the log-likelihood function can be written as:

the above formula (3) is a log-likelihood function, which is also an objective function described below.

Step 04:constructing a deep neural network to realize mapping from the PET prior image alpha to the linear attenuation coefficient image mu:

μ＝f(θ|α) (4)

the PET prior image alpha of the same patient is used as the input of the deep neural network, and the network parameter theta, namely the mapping parameter, is trained and updated in the problem optimization process.

Mapping the inputted PET prior image alpha to select PET reconstructed image which is not subjected to attenuation correction

Simultaneous selection of approximately linear attenuation coefficient image guess +.>

Approximate attenuation correction image +.>

The conventional reconstruction process only uses the acquired data, in this embodiment, the prior image is an additional image obtained to assist in reconstruction, in this embodiment, a PET reconstructed image is used without attenuation correction

Approximate attenuation correction image +.>

Here, the image without attenuation correction

Although there are significant attenuation artifacts, structural information of different tissues is still preserved, such as the patient's edge range can still be determined, although the patient's edge is highlighted; pulmonary uptake, although not contrast correct, can be done with a pulmonary contour delineation. Approximately attenuation corrected image->

Since the linear attenuation image used as attenuation correction is an approximation, attenuation artefacts remain unavoidable, such as an unclear lung contour, but segmentation of the liver organ is relative to +.>

More accurate. Thus the unattenuated corrected PET image +.>

And approximately attenuation corrected image->

As the two-channel mapping input, two PET prior images are used together, and the two PET prior images are mutually verified to achieve the purpose of more accurately estimating the linear attenuation image, namely

Without loss of generality, the mapping input may also select only unattenuated PET images

Or an approximately attenuation corrected image

Or choose to unattenuated PET image +.>

And approximately attenuation corrected image->

The weighted summation is used as network input and is selected according to actual needs.

Since PET scanning is always matched to other modality imaging, the approximate linear attenuation coefficient distribution is approximated

Other modality images may be based. Taking a PET/CT imaging system as an example, a high signal-to-noise ratio image obtained by the CT system can be utilized, and CT values are converted into photon linear attenuation coefficient distribution images under 511KeV energy by a bilinear method; taking a PET/MR imaging system as an example, MR images are segmented for different regions (such as soft tissue, fat, lung, air, etc.), and then corresponding theoretical linear attenuation coefficient values are directly assigned (such as selecting soft tissue region to assign 0.0975 cm) ^-1 Assignment of fat area to 0.0864cm ^-1 The lung area was assigned 0.0224cm ^-1 Air zone assigned 0). In the linear attenuation coefficient distribution->

During the calculation, interpolation of the initialized distribution image can be selected to reduce partial volume effect.

Step 05:based on equations (3) and (4), the solution of the radioactivity distribution image x and the linear decay coefficient image μ becomes a constrained maximum likelihood function optimization problem, namely:

this constraint optimization problem can be solved in this embodiment using the augmented lagrangian equation (augmented Lagrangian), with equation (5) written as a log-likelihood function with constraint terms:

where ρ is a fixed value of the hyper-parameter used to adjust the weights between the log-likelihood function and the learning network penalty term, typically chosen prior to training. Δμ is a change step of the linear attenuation coefficient image μ, which belongs to an intermediate variable, and represents a change amount of μ. This change in μ derives from equation (2) because a separate constraint on Δμ (i.e., L2-Norm for Δμ) is required in the log-likelihood function construction, ensuring that the magnitude of μ change is not too great.

Step 06:the log-likelihood function L with constraint terms in equation (6) _ρ For unknowns x and μ to be a very complex function, an iterative algorithm is required to gradually approach the optimal solution. For unknown PET radioactivity distribution x, keeping linear attenuation coefficient distribution mu and learning network parameter theta as constants, and maximizing log likelihood function L of constraint term _ρ Obtaining an iteratively updated radioactivity distribution:

n in the formula represents the current iteration number. The iterative initial value of the PET radioactivity distribution x may select the full-space normal number distribution. Since the PET image characterizes the radioactivity distribution, the value of which cannot be negative, the total initial value of this embodiment can be set to a pixel value of full space, e.g., 1000.

In this embodiment, the solution of the PET radioactivity distribution x may be selected from a maximum likelihood maximum expected value algorithm MLEM (Maximum Likelihood Expectation Maximization) or an acceleration algorithm ordered subset maximum expected value algorithm OSEM (Ordered Subset Expectation Maximization) or a maximum a posteriori estimation algorithm MAP (Maximum aPosteriori) commonly used in the industry, which is not limited in this embodiment, and may be selected according to actual needs.

It can be understood that in the implementation process, the linear attenuation coefficient distribution μ and the training network parameter θ are kept as constants, the log likelihood function is maximized to solve the PET radioactivity distribution x, then the log likelihood function which keeps the PET activity distribution x as a constant and maximizes the constraint term is selected to solve the linear attenuation coefficient distribution μ and the training deep learning network parameter θ alternately. The operation is performed alternately, the attenuation correction is continuously corrected to approach the real attenuation condition, and finally the estimated values of x and mu meeting the requirement of the maximized objective function are obtained.

Step 07:for solving the equation of the above equation (6), the present embodiment proposesFor a specific computational solution process.

Keeping the PET radioactivity distribution x constant for equation (6), maximizing the log likelihood function L with constraint terms for the unknown attenuation coefficient distribution mu and the learning network parameters theta _ρ The method utilizes an alternating direction multiplier ADMM iterative algorithm to solve, and the process comprises the following three steps:

n in the formula represents the current iteration number. After using the ADMM algorithm, the constraint optimization problem is decoupled into a network training sub-problem (8) with an L2 norm as the loss function and a solving sub-problem (9) with a decay coefficient distribution μ of penalty terms.

Only one variable is updated at each step in the iterative process, while the other two variables are fixed, and the updating is repeated alternately.

Specifically: firstly, keeping the linear attenuation coefficient distribution mu and the increment step delta mu unchanged, and updating the network parameter theta through a training learning network, as shown in a formula (8); then, maintaining the network parameter theta and the linear attenuation coefficient increase step delta mu unchanged, and iteratively updating the linear attenuation coefficient distribution mu, as shown in a formula (9); finally, the linear attenuation coefficient distribution mu and the network parameter theta are kept unchanged, and the linear attenuation coefficient increasing step delta mu is updated as shown in a formula (10). And the three variables are iterated to approach the optimal solution, so that accurate estimation of linear attenuation coefficient distribution is obtained.

In order to accelerate the convergence speed, a pre-training network can be used for generating general network parameters as iteration initial values of the network parameters, and then fine adjustment is carried out for a specific scanning patient. The iterative initial value of the attenuation coefficient distribution mu may be all 0,or is

The mapping network of the present embodiment may select a CNN network, a Unet network, a GAN network, or other networks. The loss function can also select L1 norm, KL divergence and the like without losing generality, and can also be selected according to actual needs by weighting and summing a plurality of loss functions. The loss function at this point may be the loss function used in training the network.

The pre-training refers to using a large amount of training data to train a model to complete the mapping from the prior PET image to the linear attenuation coefficient image. In this embodiment, for the patient actually scanned, the network parameters need to be fine-tuned to be more suitable for the actual situation, and the pre-trained network parameters can be used as the initial value of the network fine-tuning process, which ensures the network convergence speed.

Step 08:the PET image quality is greatly different under the influence of different fields, different devices and different scanning parameters, so that the embodiment needs to train a universal learning network/mapping network in advance as an initial value of current network parameter training, and then fine-tune training parameters according to actual PET acquisition data, thereby ensuring that a sexual attenuation coefficient result accords with the actual acquisition data and greatly reducing the requirement of training network data generalization.

Specifically, in order to be able to sufficiently extract PET image features, a mapping of the PET image to a linear attenuation coefficient image using a pre-built deep learning network G is selected.

For example, in training the deep learning network G, the PET images in the training samples may include unattenuated PET images, for example PET/CT

And approximately attenuation corrected PET image +.>

Two PET images are input as two channels, and linear attenuation generated by mapping PET imagesCoefficient image μ _DL As output, with a true linear attenuation image μ obtained by CT scanning _CT By comparing, the loss function phi is minimized by optimizing the training network parameters theta, and finally, the PET image can be translated into an accurate linear attenuation coefficient image, namely:

the superscript of (1) indicates the result of the solution to this value.

The training samples in the training data may be from analog simulation or from actual acquisition. All PET images of the training data need to be preprocessed, and the linear attenuation coefficient images in each sample are matched with the PET images through preprocessing screening, so that truncation or motion artifacts do not exist. That is, the data input in the training is a PET reconstructed image, the data is output as a linear attenuation coefficient image mapped by the PET image, the learning target is a linear attenuation coefficient image obtained by the CT image actually acquired, the network training is to optimize network parameters, and the network output is ensured to be similar to the actual result.

It is understood that the deep learning network G may select a CNN network, a Unet network, a GAN network, or other networks. Typically the deep learning network input may select only unattenuated PET images

Or approximately attenuation corrected image->

Or choose to unattenuated PET image +.>

And approximately attenuation corrected image->

The summation is taken as a network input. The loss function phi can be used to scale mu _DL Sum mu _CT The L1 norm, L2 norm, KL divergence, etc. may be selected, or a plurality of loss functions may be weighted and summed.

Example verification as shown in fig. 2 (a) to 2 (d), fig. 2 (a) is a schematic diagram of a PET image without attenuation correction, fig. 2 (b) is a schematic diagram of a PET image with approximate attenuation correction, fig. 2 (c) is a schematic diagram of a PET image with attenuation correction by the method of the present invention, and fig. 2 (d) is an estimation result of a linear attenuation coefficient a priori distribution obtained by using a learning network. From this, the result of fig. 2 (c) is superior to the first two PET images, and the convergence speed is high, the reliability is high, and no artifact is generated.

The method of the embodiment has no problem of attenuation image truncation, and is convenient for doctors to scan patients with large weight; the attenuation correction iterative process is regulated by linear attenuation coefficient priori estimation generated by the deep learning network, and the quantification performance and the tissue distribution are more accurate than those of the original iterative algorithm, so that the stability and the convergence speed of the attenuation correction algorithm are greatly improved; the linear attenuation coefficient generated by the deep learning network is applied to an attenuation coefficient iterative algorithm, so that the problem of PET acquisition data generalization is solved, the construction difficulty of the learning network is simplified, and the stability of the network is improved; the PET acquisition is independent of other modes, can be applied to single PET scanning, reduces the scanning environment requirement, and expands the application occasions.

In addition, the embodiment of the invention also provides a PET system, which comprises: a memory and a processor; the memory stores computer program instructions, and the processor executes the computer program instructions stored in the memory, specifically executes the above-mentioned method for acquiring the radioactivity distribution of PET, and the like.

It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The use of the terms first, second, third, etc. are for convenience of description only and do not denote any order. These terms may be understood as part of the component name.

Furthermore, it should be noted that in the description of the present specification, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with the embodiment or example being included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art upon learning the basic inventive concepts. Therefore, the appended claims should be construed to include preferred embodiments and all such variations and modifications as fall within the scope of the invention.

It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, the present invention should also include such modifications and variations provided that they come within the scope of the following claims and their equivalents.

Claims

1. A method of acquiring a PET radioactivity distribution, comprising:

2. The method according to claim 1, characterized in that:

before the step S20, a mapping from the PET prior image alpha to the linear attenuation coefficient image mu is constructed;

the mapping includes:

μ=f (θ|α) formula (1)

3. The method according to claim 2, characterized in that: the S20 includes:

4. A method according to claim 3, characterized in that:

for equation (2), the alternative strategy is used to maximize the log likelihood function L with constraint terms based on the known solution algorithm _ρ Obtaining a log-likelihood function L meeting maximization constraint terms _ρ The required estimated value of x is taken as an output value;

The initial value of the PET radioactivity distribution x is a set value;

the known solution algorithm is MLEM, OSEM or MAP.

5. A method according to claim 3, characterized in that:

keeping the PET radioactivity distribution x constant for equation (2), maximizing the log likelihood function L with constraint terms for the unknown attenuation coefficient distribution μ and mapping parameters θ _ρ The method utilizes an alternating direction multiplier ADMM iterative algorithm to solve, and the process comprises the following three steps:

wherein n represents the current iteration number, and n iterations meet the ending condition; decoupling the constraint optimization problem into a network training sub-problem-corresponding formula (8) with an L2 norm as a loss function and a solution sub-problem-corresponding formula (9) with a penalty term mu in using an ADMM algorithm;

6. The method according to claim 4, wherein:

when the medical image is a PET image, the method further includes, before S10:

s01, training a model based on a training sample to obtain a trained model;

7. The method of claim 6, wherein training the model based on the training samples comprises:

and/or the number of the groups of groups,

8. A PET system, comprising: a memory and a processor; the memory has stored therein computer program instructions, and the processor executes the computer program instructions stored in the memory, in particular to perform a method of acquiring a PET radioactivity distribution according to any one of the preceding claims 1 to 7.

9. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when run by a processor, performs a method of acquiring a PET radioactivity distribution according to any one of claims 1 to 7.