SPECT low-dose chordal graph recovery method based on Poisson distribution noise diffusion probability model

Abstract

The invention relates to the technical field of recovery and reconstruction of a low-dose SPECT chordal graph, in particular to an SPECT low-dose chordal graph recovery method based on a Poisson distribution noise diffusion probability model. The method comprises the following steps: firstly, providing a PDPM preliminary framework comprising a forward process and a reverse process, and perfecting on the preliminary framework, comprising two parts: in the first part, generating a data set for training by using a method based on an ideal reverse process instead of abandoning the forward process; according to the method, the problem of mismatching between low-dose chordal diagrams and normal-dose chordal diagrams in a preliminary frame forward process is solved; in the second part, a time prediction aggregation module TPAM is introduced, and the performance of model recovery is effectively improved through aggregation prediction across multiple time steps.

Description

Technical Field

[0001] The present invention relates to the technical field of restoration and reconstruction of low-dose SPECT chordograms, and more specifically, to a SPECT low-dose chordogram restoration method based on a Poisson distribution noise diffusion probability model. Background Art

[0002] Single Photon Emission Computed Tomography (SPECT) is a nuclear medicine imaging modality that uses gamma rays for diagnostic imaging. In this procedure, a radioactive tracer is injected into the patient's body through a vein and is selectively absorbed by different organs or tissues. When the tracer undergoes radioactive decay, it releases gamma photons, which are then detected for imaging. A gamma camera, acting as a detection imaging device, then rotates around the patient's body to capture the gamma rays and generate projection data from multiple angles, also known as sinograms. These sinograms are then processed using tomographic reconstruction algorithms to generate anatomical images representing the patient's internal body structures. Since the radiation released by the radioactive tracers used in SPECT studies may change cell structure or function and even increase the risk of cancer, it is critical to minimize the tracer dose during the SPECT procedure. However, reducing the tracer dose may exacerbate the photon attenuation effect during the generation of the sinogram. When gamma photons travel through tissue, they interact with atomic electrons, causing many photons to deviate from their original path. This deflection causes some photons to be diverted away from the detector, a phenomenon known as photon attenuation. In normal-dose SPECT chordograms, the effect of photon attenuation is negligible, but becomes apparent in low-dose SPECT chordograms. Therefore, compared with normal-dose chordograms, low-dose SPECT chordograms show higher noise levels, reduced spatial resolution, and may introduce more noticeable artifacts. These factors usually lead to a decrease in the quality of the reconstructed image, making the doctor's diagnosis more difficult.

[0003] Many traditional model-driven methods have been proposed for the restoration and reconstruction of low-dose SPECT chordograms, such as iterative reconstruction methods. However, the model parameters of this method are often determined by experience, so the generalization ability of projection data under various noise levels is usually poor. With the increasing popularity of deep learning, methods that can automatically learn discriminative features from training data have attracted the attention of researchers. Various low-dose reconstruction methods based on deep learning can be roughly divided into three categories: end-to-end methods, post-processing methods, and pre-processing methods based on the chordogram domain. Specifically, the end-to-end reconstruction method takes projection data or k-space data as input to directly generate high-quality reconstructed images. However, this requires the model to independently learn the underlying physical properties, and the model establishment and training are difficult; the post-processing method uses the reconstructed image at low dose as the input of the neural network to restore high-quality images, but this usually requires a large amount of data sets and complex network architectures to achieve better performance in practical applications, which can easily lead to inaccurate restored images; the pre-processing method based on the chordogram domain limits the denoising restoration problem to the chordogram domain, first using a neural network to reduce the noise of the chordogram, and then using traditional algorithms such as iterative reconstruction to reconstruct images from high-quality chordograms. The calculation process is complex and the accuracy is low. Therefore, how to restore low-dose SPECT chordograms to normal dose levels has become an urgent problem to be solved in SPECT clinical imaging. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art in restoring low-dose SPECT chordograms to normal dose levels, which are difficult and of low accuracy, and to provide a SPECT low-dose chordogram restoration method based on a Poisson distribution noise diffusion probability model, thereby achieving low-dose restoration of SPECT chordograms with high restoration accuracy.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for recovering a low-dose SPECT chordogram based on a Poisson distribution noise diffusion probability model is provided, comprising the following steps:

[0007] S1. Training: For a data set containing n normal-dose / low-dose SPECT chordogram data pairs, use the ideal reverse process Markov chain to generate the intermediate images of the ideal reverse process, use these intermediate images together with the low-dose chordogram as input data, and use the corresponding normal-dose chordogram as a label to form a training data set for the U-Net model, and use the training data set to train the U-Net model;

[0008] S2. Reasoning: Transform image x t and time t as the input of the neural network U-Net to obtain the prediction of the normal dose chord diagram Together with the prediction results of all time steps, they are input into the temporal prediction aggregation module TPAM. In TPAM, the prediction results from time step T to time step t are respectively combined with the image x t The restoration operator R(·) is input with time t, the image distribution is calculated, and the image generated after inverse transform sampling is used as the output; the restoration operator output corresponding to the prediction results from time step T to time t is averaged to generate the restored image x t-1 ;

[0009] S3. Iteration: In the inference of PDPM, from the low-dose SPECT chord diagram x T At the beginning, the process of step S2 is continuously iterated for a total of T-1 steps, and finally the U-Net prediction result obtained with x1 as input is used as the final restored image x0.

[0010] The present invention discloses a method for recovering low-dose SPECT chordograms based on a Poisson distribution noise diffusion probability model. The core part is a probability model based on Poisson noise diffusion, called Poisson Diffusion Probabilistic Model (PDPM). Considering the physical principle of low-dose SPECT chordogram formation, PDPM replaces the Gaussian noise used in the traditional diffusion model with Poisson noise, and uses low-dose and normal-dose SPECT chordograms as the starting point and end point of the recovery process. Specifically, the present invention is based on a preliminary PDPM framework including a forward process and a reverse process. The present invention improves the preliminary framework and includes two parts: the first part is to abandon the forward process and use a method based on an ideal reverse process to generate a data set for training. This method solves the mismatch problem between low-dose and normal-dose chordograms in the forward process of the preliminary framework; the second part is to introduce a temporal prediction aggregation module (TPAM), which effectively improves the performance of model recovery by aggregating predictions across multiple time steps. The PDPM proposed in the present invention is based on the noise properties generated during the formation of low-dose SPECT chord diagrams, and uses Poisson noise to perturb the image, which is consistent with the actual noise generation and propagation mechanism observed in SPECT imaging. Therefore, the model of the present invention is more consistent with the physical principle of low-dose SPECT chord diagram degradation than the traditional diffusion model. On the other hand, the TPAM introduced in the reverse process reduces the error between the image generated by the model reverse process and the actual image in the ideal reverse process by aggregating the prediction results within multiple time steps, so that the model performs better than the traditional diffusion model in terms of image restoration performance. In summary, compared with the traditional diffusion model, the PDPM proposed in the present invention is more suitable for low-dose SPECT chord diagram restoration tasks.

[0011] Preferably, the ideal reverse process comprises: distributing the reverse process image p θ (x t-1 |x t ) Replaced with normal dose SPECT chordogram x0.

[0012] Preferably, the Markov chain is expressed as:

[0013]

[0014] in,

[0015]

[0016] In the formula, x t-1 represents the restored image at step t-1, x t represents the restored image at step t, x0 represents the original image, i.e., the normal dose SPECT chord diagram, and t represents the time step.

[0017] Preferably, in step S1, the U-Net model is trained, and the training objective is expressed as:

[0018]

[0019] In the formula, is the prediction result of the neural network U-Net, and x0 is the normal dose SPECT chord diagram.

[0020] Preferably, in step S2, the recovery operator R(·) is expressed as:

[0021]

[0022] In the formula, represents the prediction result of the neural network U-Net at the tth time step, x t represents the restored image at the tth time step, t represents the time step, represents the restored image at the t-1th time step generated by the restoration operator.

[0023] Preferably, in step S2, the recovery operator R(·) uses the formula Calculate the image distribution p θ (x t-1 |x t ).

[0024] Preferably, in step S2, the recovery operator outputs corresponding to the prediction results from time steps T to t are averaged and processed using the following formula:

[0025]

[0026] The result calculated by the above formula is used as the restored image x t-1 .

[0027] The present invention also provides a SPECT low-dose chordogram recovery system based on a Poisson distribution noise diffusion probability model, comprising:

[0028] Training module: used to generate the intermediate images of the ideal reverse process using the ideal reverse process Markov chain for a data set containing n normal-dose / low-dose SPECT chordogram data pairs, and use these intermediate images together with the low-dose chordogram as input data and the corresponding normal-dose chordogram as labels to form the training data set of the U-Net model, and use the training data set to train the U-Net model;

[0029] Reasoning module: used to convert image x t and time t as the input of the neural network U-Net to obtain the prediction of the normal dose chord diagram Together with the prediction results of all time steps, they are input into the temporal prediction aggregation module TPAM. In TPAM, the prediction results from time step T to time step t are respectively combined with the image x t The restoration operator R(·) is input with time t, the image distribution is calculated, and the image generated after inverse transform sampling is used as the output; the restoration operator output corresponding to the prediction results from time step T to time t is averaged to generate the restored image x t-1 ;

[0030] Iteration module: used in the inference of PDPM from the low-dose SPECT chord diagram x T At the beginning, the process of step S2 is continuously iterated for a total of T-1 steps, and finally the U-Net prediction result obtained with x1 as input is used as the final restored image x0.

[0031] Preferably, in the training module, the Markov chain is expressed as:

[0032]

[0033] in,

[0034]

[0035] In the formula, x t-1 represents the restored image at step t-1, x t represents the restored image at step t, x0 represents the original image, i.e., the normal dose SPECT chord diagram, and t represents the time step;

[0036] The U-Net model is trained, and the training objective is expressed as:

[0037]

[0038] In the formula, is the prediction result of the neural network U-Net, and x0 is the normal dose SPECT chord diagram.

[0039] Preferably, in the inference module, the recovery operator R(·) is expressed as:

[0040]

[0041] The recovery operator R(·) uses the formula Calculate the image distribution p θ (x t-1 |x t ), and the image generated after inverse transform sampling is used as output; the recovery operator output corresponding to the prediction result from time step T to time t is averaged and processed using the following formula:

[0042]

[0043] The result calculated by the above formula is used as the restored image x t-1 .

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention discloses a method for restoring low-dose SPECT chordograms based on a Poisson distribution noise diffusion probability model. The proposed PDPM is based on the noise properties generated during the formation of low-dose SPECT chordograms, and uses Poisson noise to perturb the image, which is consistent with the actual noise generation and propagation mechanism observed in SPECT imaging. Therefore, the model of the present invention is more consistent with the physical principle of low-dose SPECT chordogram degradation than the traditional diffusion model. On the other hand, the TPAM introduced in the reverse process reduces the error between the image generated by the model reverse process and the actual image in the ideal reverse process by aggregating the prediction results within multiple time steps, so that the model performs better in image restoration performance than the traditional diffusion model. In summary, compared with the traditional diffusion model, the PDPM proposed in the present invention is more suitable for low-dose SPECT chordogram restoration tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Schematic diagram of the model framework of a SPECT low-dose chordogram recovery method based on the Poisson distribution noise diffusion probability model. DETAILED DESCRIPTION

[0047] The present invention is further described below in conjunction with specific implementation methods. The accompanying drawings are only used for exemplary descriptions and are only schematic diagrams, not actual drawings, and cannot be understood as limiting this patent; in order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0048] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right" and the like indicate directions or positional relationships based on the directions or positional relationships shown in the drawings, it is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and cannot be understood as limitations on this patent. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0049] Embodiment 1

[0050] This embodiment is the first embodiment of a method for restoring a low-dose SPECT chordogram based on a Poisson distribution noise diffusion probability model. This embodiment aims to solve the problem of poor image quality in a low-dose SPECT chordogram due to the influence of photon attenuation during the imaging process, and to achieve low-dose restoration of the SPECT chordogram.

[0051] This embodiment proposes a SPECT low-dose chordogram recovery method based on a Poisson distribution noise diffusion probability model. The core part of the method is a probability model based on Poisson noise diffusion, called a Poisson Diffusion Probabilistic Model (PDPM). Considering the physical principle of low-dose SPECT chordogram formation, PDPM replaces the Gaussian noise used in the traditional diffusion model with Poisson noise, and uses low-dose and normal-dose SPECT chordograms as the starting point and end point of the recovery process. Specifically, this embodiment first proposes a preliminary PDPM framework that includes a forward process and a reverse process. This preliminary framework strictly derives the changes in the probability distribution of the image in the forward and reverse processes using mathematical formulas, laying a foundation for the physical principle of the method of this embodiment. Then the preliminary framework was improved, which included two parts: the first part was to abandon the forward process and use the method based on the ideal reverse process to generate the training data set. This method solved the mismatch between the low-dose and normal-dose chord diagrams in the forward process of the preliminary framework; the second part was to introduce the Temporal Prediction Aggregation Module (TPAM), which effectively improved the performance of model recovery by aggregating predictions across multiple time steps. The following is a detailed description of the proposed PDPM preliminary framework and the improvements to the preliminary framework.

[0052] (1) Preliminary framework of PDPM:

[0053] The preliminary framework of PDPM consists of two parts: the forward process and the reverse process. The forward process simulates the degradation process of the chord from normal dose to low dose by gradually subtracting Poisson random noise from the data; the reverse process restores the low-dose chord to the corresponding normal-dose chord by learning to transform the forward process. Given a set of normal-dose SPECT chords x0~q(x0), in the forward process, Poisson random noise is gradually subtracted from the image by the following formula:

[0054] x t :=x t-1 -∈ t ,t=1,2,…,T,

[0055] Among them, x0 and x T They represent the starting and ending images in the forward process, namely the normal-dose SPECT chord diagram and the low-dose SPECT chord diagram, respectively. t is the image after the forward process has been performed for t steps. ∈ t is a random Poisson noise with mean λ, represented by ∈ t ~P(λ). The specific definition of the Markov chain of the forward process is:

[0056]

[0057] in,

[0058]

[0059] The reverse process of PDPM is defined as a Markov chain with learnable parameters θ:

[0060]

[0061] in,

[0062]

[0063] Here, An approximation of x0 generated by a neural network with learnable parameters θ. In PDPM, U-Net is used as this trainable neural network, and since x t-1 The image distribution p θ (x t-1 |x t ) does not follow any known distribution, so Inverse Transform Sampling is used to generate the image x by sampling from the image distribution t-1 .

[0064] (2) Improvement of the preliminary framework 1: Method for generating PDPM training dataset.

[0065] In the training phase of the model, a method based on the ideal reverse process is used to generate a dataset for training U-Net. Specifically, for a pair of normal-dose / low-dose SPECT chordograms (x N ,x L ), set x0 = x N and x T =x L Based on the formula of the reverse process, the following Markov chain is used to generate the image

[0066]

[0067] in,

[0068]

[0069] In fact, the reverse process image distribution p θ (x t-1 |x t ) Replaced with the normal dose chord diagram x0, which is why the image generation process represented by this Markov chain is called the ideal inverse process. Similar to the inverse process in PDPM, the inverse transform is used to generate the image from the image distribution These images were then used as a dataset for training the U-Net in PDPM. Compared to the method of generating training data using a forward process, this improved method avoids the mismatch problem between the low-dose and normal-dose chord diagrams mentioned earlier.

[0070] (3) The second improvement of the preliminary framework: introducing TPAM into the reverse process of PDPM.

[0071] In the reverse process of the model, i.e. the inference phase, a time prediction aggregation module (TPAM) is designed to alleviate the The actual x in the reverse process is t-1 Specifically, in the reverse process before improvement, Calculated The image distribution is generated by inverse transform sampling The whole process is represented by the recovery operator R(·), which is defined as follows:

[0072]

[0073] After introducing TPAM to improve the model, when TPAM uses the recovery operator to generate When using Calculate and combine the previous U-Net prediction, that is Final The calculation is done by the following formula:

[0074]

[0075] In this way, TPAM can effectively alleviate the negative impact of accumulated errors in the reverse process, thereby further improving the recovery performance of the model.

[0076] like Figure 1 As shown in FIG. 1 , it is a schematic diagram of the model framework proposed in the present invention. The overall architecture of the model includes a recovery process with a low-dose SPECT chord diagram as the starting point and a normal-dose SPECT chord diagram as the end point, and the number of time steps is T. In the single-step recovery process, the U-Net model is used to predict the estimation of the high-dose chord diagram, and then the time prediction aggregation module TPAM is used to generate the recovery result of this step.

[0077] The specific steps include:

[0078] Step 1: Training.

[0079] For a dataset containing n normal-dose / low-dose SPECT chordogram data pairs Using ideal reverse process Markov chain Generate intermediate images of the ideal reverse process, use these intermediate images together with the low-dose chord map as input data, and the corresponding normal-dose chord map as labels to form the training data set of the U-Net model, expressed as The U-Net model is trained using this training dataset with the following training objectives:

[0080]

[0081] in, is the prediction result (i.e., output) of the neural network U-Net; x0 is the normal dose SPECT chord diagram.

[0082] Step 2: Reasoning.

[0083] From the image x t Restore the generated image x t-1 As an example, the single-step recovery process of t and time t as the input of the neural network U-Net to obtain the prediction of the normal dose chord diagram The image x t The prediction result of this step Together with the prediction results of all previous time steps, that is, In TPAM, the prediction results from time step T to time step t are respectively t and t input recovery operator R(·), which uses the formula Calculate the image distribution p θ (x t-1 |x t ), and the image generated after inverse transform sampling is used as the output. The output of the recovery operator corresponding to the prediction results from time step T to t is averaged, that is, the following formula is used for processing:

[0084]

[0085] The result calculated by the above formula is used as the restored image x t-1 .

[0086] Step 3: Iterate.

[0087] In the reasoning of PDPM, from the low-dose SPECT chord diagram x T Start, and iterate the above process continuously for a total of T-1 steps. Finally, the U-Net prediction result obtained with x1 as input is As the final restored image x0.

[0088] Compared with other methods for low-dose SPECT chordogram restoration, the diffusion model works by processing images step by step, which gives it an advantage in solving the key challenge of processing images in medical imaging without losing the important information contained therein. The step-by-step processing method of the diffusion model enables the model to process only a small amount of noise data at each step while minimizing the loss of important information. Therefore, the use of the diffusion model for low-dose SPECT chordogram restoration can not only improve the accuracy of image restoration, but also effectively preserve key features such as chord edges and tissue features that are crucial to medical diagnosis.

[0089] Compared with the traditional diffusion model based on Gaussian noise, the PDPM proposed in this embodiment uses Poisson noise to perturb the image based on the noise properties generated during the formation of the low-dose SPECT chord diagram, which is consistent with the actual noise generation and propagation mechanism observed in SPECT imaging. Therefore, the model of this embodiment is more consistent with the physical principle of low-dose SPECT chord diagram degradation than the traditional diffusion model. On the other hand, the TPAM introduced in the reverse process reduces the error between the image generated by the model's reverse process and the actual image in the ideal reverse process by aggregating the prediction results within multiple time steps, so that the model performs better in image restoration performance than the traditional diffusion model. In summary, compared with the traditional diffusion model, PDPM is more suitable for low-dose SPECT chord diagram restoration tasks.

[0090] Embodiment 2

[0091] This embodiment is a first embodiment of a SPECT low-dose chordogram recovery system based on a Poisson distribution noise diffusion probability model, comprising:

[0092] Training Modules:

[0093] Used to analyze a dataset containing n normal-dose / low-dose SPECT chordogram data pairs Using ideal reverse process Markov chain Generate intermediate images of the ideal reverse process, use these intermediate images together with the low-dose chord map as input data, and the corresponding normal-dose chord map as labels to form the training data set of the U-Net model, expressed as The U-Net model is trained using this training dataset with the following training objectives:

[0094]

[0095] in, is the prediction result (i.e., output) of the neural network U-Net; x0 is the normal dose SPECT chord diagram.

[0096] Reasoning module:

[0097] Used to convert x t and time t as the input of the neural network U-Net to obtain the prediction of the normal dose chord diagram The image x t The prediction result of this step Together with the prediction results of all previous time steps, that is, In TPAM, the prediction results from time step T to time step t are respectively t and t input recovery operator R(·), which uses the formula Calculate the image distribution p θ (x t-1 |x t ), and the image generated after inverse transform sampling is used as the output. The output of the recovery operator corresponding to the prediction results from time step T to t is averaged, that is, the following formula is used for processing:

[0098]

[0099] The result calculated by the above formula is used as the restored image x t-1 .

[0100] Iteration module:

[0101] Used in the inference of PDPM, from the low-dose SPECT chordogram x T Start, and iterate the above process continuously. Finally, the U-Net prediction result obtained with x1 as input is As the final restored image x0.

[0102] Embodiment 3

[0103] This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method described in the first embodiment are implemented.

[0104] In the specific contents of the above-mentioned specific implementation methods, the various technical features can be combined in any non-contradictory manner. In order to make the description concise, not all possible combinations of the above-mentioned technical features are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for recovering low-dose chordograms of SPECT based on a Poisson distribution noise diffusion probability model, characterized in that: The following steps are involved: S1. Training: For a data set containing n normal-dose / low-dose SPECT chordogram data pairs, use the ideal reverse process Markov chain to generate the intermediate images of the ideal reverse process, use these intermediate images together with the low-dose chordogram as input data, and use the corresponding normal-dose chordogram as a label to form a training data set for the U-Net model, and use the training data set to train the U-Net model; S2. Reasoning: Transform image x t and time t as the input of the neural network U-Net to obtain the prediction of the normal dose chord diagram Together with the prediction results of all time steps, they are input into the temporal prediction aggregation module TPAM. In TPAM, the prediction results from time step T to time step t are respectively combined with the image x t The restoration operator R(·) is input with time t, the image distribution is calculated, and the image generated after inverse transform sampling is used as the output; the restoration operator output corresponding to the prediction results from time step T to time t is averaged to generate the restored image x t-1 ; S3. Iteration: In the inference of PDPM, from the low-dose SPECT chord diagram x T At the beginning, the process of step S2 is continuously iterated for a total of T-1 steps, and finally the U-Net prediction result obtained with x1 as input is used as the final restored image x0.

2. The SPECT low-dose chordogram recovery method based on the Poisson distribution noise diffusion probability model according to claim 1, characterized in that: The ideal reverse process includes: distributing the reverse process image p θ (x t-1 |x t ) Replaced with normal dose SPECT chordogram x0.

3. The SPECT low-dose chordogram recovery method based on the Poisson distribution noise diffusion probability model according to claim 2, characterized in that: The Markov chain is expressed as: in, In the formula, x t-1 represents the restored image at step t-1, x t represents the restored image at step t, x0 represents the original image, i.e., the normal dose SPECT chord diagram, and t represents the time step.

4. The SPECT low-dose chordogram recovery method based on the Poisson distribution noise diffusion probability model according to claim 1, characterized in that: In step S1, the U-Net model is trained, and the training objective is expressed as: In the formula, is the prediction result of the neural network U-Net, and x0 is the normal dose SPECT chord diagram.

5. The SPECT low-dose chordogram recovery method based on the Poisson distribution noise diffusion probability model according to claim 1, characterized in that: In step S2, the recovery operator R(·) is expressed as: In the formula, represents the prediction result of the neural network U-Net at the tth time step, x t represents the restored image at the tth time step, t represents the time step, represents the restored image at the t-1th time step generated by the restoration operator.

6. The SPECT low-dose chordogram recovery method based on the Poisson distribution noise diffusion probability model according to claim 5, characterized in that: In step S2, the recovery operator R(·) uses the formula Calculate the image distribution p θ (x t-1 |x t ).

7. The SPECT low-dose chordogram recovery method based on the Poisson distribution noise diffusion probability model according to claim 6, characterized in that: In step S2, the recovery operator outputs corresponding to the prediction results from time step T to t are averaged and processed using the following formula: The result calculated by the above formula is used as the restored image x t-1 .

8. A SPECT low-dose chordogram recovery system based on a Poisson distribution noise diffusion probability model, characterized in that: include: Training module: used to generate the intermediate images of the ideal reverse process using the ideal reverse process Markov chain for a data set containing n normal-dose / low-dose SPECT chordogram data pairs, and use these intermediate images together with the low-dose chordogram as input data and the corresponding normal-dose chordogram as labels to form the training data set of the U-Net model, and use the training data set to train the U-Net model; Reasoning module: used to transform image x t and time t as the input of the neural network U-Net to obtain the prediction of the normal dose chord diagram Together with the prediction results of all time steps, they are input into the temporal prediction aggregation module TPAM. In TPAM, the prediction results from time step T to time step t are respectively combined with the image x t The restoration operator R(·) is input with time t, the image distribution is calculated, and the image generated after inverse transform sampling is used as the output; the restoration operator output corresponding to the prediction results from time step T to time t is averaged to generate the restored image x t-1 ; Iteration module: used in the inference of PDPM from the low-dose SPECT chord diagram x T At the beginning, the process of step S2 is continuously iterated for a total of T-1 steps, and finally the U-Net prediction result obtained with x1 as input is used as the final restored image x0.

9. The SPECT low-dose chordogram recovery system based on the Poisson distribution noise diffusion probability model according to claim 8, characterized in that: In the training module, the Markov chain is represented as: in, In the formula, x t-1 represents the restored image at step t-1, x t represents the restored image at step t, x0 represents the original image, i.e., the normal dose SPECT chord diagram, and t represents the time step; The U-Net model is trained, and the training objective is expressed as: In the formula, is the prediction result of the neural network U-Net, and x0 is the normal dose SPECT chord diagram.

10. The SPECT low-dose chordogram recovery system based on the Poisson distribution noise diffusion probability model according to claim 8, characterized in that: In the inference module, the recovery operator R(·) is expressed as: The recovery operator R(·) uses the formula Calculate the image distribution p θ (x t-i |x t ), and the image generated after inverse transform sampling is used as output; the recovery operator output corresponding to the prediction result from time step T to time t is averaged and processed using the following formula: The result calculated by the above formula is used as the restored image x t-1 .

Images