Deep learning enhanced nuclear medicine single photon emission computed tomography method

Abstract

The present application relates to the technical field of medical image, more particularly, to a method for enhancing nuclear medicine single photon emission computed tomography by using deep learning, obtaining projection data of nuclear medicine single photon emission computed tomography;Based on the projection data, a deep learning enhanced count rate sparse back projection algorithm neural network model is constructed;The projection data is used to train the deep learning enhanced count rate sparse back projection algorithm neural network model;According to the trained deep learning enhanced count rate sparse back projection algorithm neural network model, the projection data is reconstructed;Output the reconstructed three-dimensional nuclear medicine image, wherein the three-dimensional nuclear medicine image includes forward projection three-dimensional nuclear medicine image and positive-negative projection three-dimensional nuclear medicine image, when processing low dose, low count rate nuclear medicine image, noise can be effectively suppressed, and the signal-to-noise ratio and spatial resolution of the image are improved.

Description

Technical Field

[0001] This invention relates to the field of medical imaging technology, and more specifically, to a method for enhancing nuclear medicine single-photon emission computed tomography using deep learning. Background Technology

[0002] Single-photon emission computed tomography (SPECT) in nuclear medicine is an important medical imaging technique widely used in the diagnosis of heart disease, oncology, and neurological diseases. However, traditional SPECT imaging techniques face many challenges, mainly including poor image quality, low reconstruction efficiency, and high radiation dose.

[0003] In recent years, with the development of computer technology and image processing algorithms, the quality of SPECT imaging has improved to some extent. Traditional reconstruction algorithms, such as filtered back projection (FBP) and maximum likelihood expectation-maximization (MLEM), have improved image quality to a certain degree. However, these methods still have significant shortcomings when processing nuclear medicine images with low doses and low count rates, often resulting in reconstructed images with high noise and low resolution, making it difficult to meet the needs of clinical diagnosis.

[0004] Furthermore, while iterative reconstruction algorithms improve image quality, their high computational complexity and long reconstruction time make them unsuitable for real-time clinical diagnosis. Additionally, traditional methods often require increased injection doses of radioactive tracers to obtain high-quality images, which undoubtedly increases the patient's radiation exposure risk.

[0005] Recently, deep learning technology has made significant progress in the field of medical image processing, bringing new opportunities to SPECT imaging. However, directly applying deep learning to SPECT image reconstruction still faces many challenges. First, the special properties of nuclear medicine images, such as low signal-to-noise ratio and sparse sampling, make it difficult to directly apply traditional deep learning models. Second, how to effectively combine the advantages of deep learning with the physical models of traditional reconstruction algorithms to achieve higher quality and more efficient image reconstruction remains an urgent problem to be solved.

[0006] Given the shortcomings of the existing technologies, there is an urgent need for new methods that can comprehensively utilize the advantages of deep learning technology and traditional reconstruction algorithms to improve the quality and reconstruction efficiency of SPECT images while reducing radiation dose. Summary of the Invention

[0007] The purpose of this invention is to provide a method for enhancing nuclear medicine single-photon emission computed tomography (SPECT) using deep learning, aiming to solve the aforementioned technical problems. Specifically, this invention, through innovative deep learning model design and optimization strategies, combined with the physical model of traditional reconstruction algorithms, achieves a significant improvement in SPECT image quality, a substantial increase in reconstruction efficiency, and an effective reduction in radiation dose.

[0008] This invention provides a method for enhancing nuclear medicine single-photon emission computed tomography using deep learning, comprising:

[0009] The acquisition steps include:

[0010] Acquire projection data from single-photon emission computed tomography (SPECT) imaging in nuclear medicine;

[0011] The processing steps include:

[0012] Based on the projection data, a deep learning enhanced count rate sparse backprojection algorithm neural network model is constructed. The deep learning enhanced count rate sparse backprojection algorithm neural network model includes an image cascade layer and a backprojection layer. The image cascade layer includes multiple two-dimensional convolutional layers, pooling layers, and two-dimensional deconvolutional layers. The backprojection layer uses the count rate sparse backprojection algorithm to calculate the reconstructed image data.

[0013] Training the deep learning-enhanced count rate sparse backprojection algorithm neural network model using the projection data includes: constructing a neural network structure for the projection data of a three-dimensional nuclear medicine image; constructing an objective function, wherein the objective function simultaneously considers forward projection and backprojection; fixing the parameters of the backprojection layer and iteratively updating the parameters in the image cascade layer;

[0014] The projection data is reconstructed based on the trained deep learning-enhanced counting rate sparse back projection algorithm neural network model;

[0015] Output steps, including:

[0016] Output the reconstructed three-dimensional nuclear medicine image, wherein the three-dimensional nuclear medicine image includes a forward-projected three-dimensional nuclear medicine image and a forward-backward-projected three-dimensional nuclear medicine image.

[0017] Preferably, the iterative update step includes:

[0018] Perform forward data transmission and calculate the loss at each layer;

[0019] Calculate the gradient of the loss with respect to the parameters;

[0020] Update parameters using backpropagation.

[0021] Preferably, the step of acquiring projection data from nuclear medicine single-photon emission computed tomography includes:

[0022] Initialize the three-dimensional nuclear medicine image;

[0023] Convert the initialized nuclear medicine images into three-dimensional models;

[0024] Three-dimensional tomographic data is emitted using a photon detection probe to determine the detection event and the projection data.

[0025] Preferably, the step of initializing the three-dimensional nuclear medicine image uses a sine wave for initialization, specifically as follows:

[0026] The initial values ​​for the forward projection and reverse projection are generated using a sine function, where the initial value for the forward projection is the amplitude of the sine curve, and the initial value for the reverse projection is the phase of the sine curve.

[0027] Preferably, the step of converting the initialized nuclear medicine image into a three-dimensional model employs a spherical coordinate back projection method, specifically including:

[0028] Convert the pixel coordinates of a two-dimensional image to three-dimensional coordinates in a spherical coordinate system;

[0029] Construct a three-dimensional model based on the three-dimensional coordinates in the spherical coordinate system.

[0030] Preferably, the following steps are also included:

[0031] Construct a deep learning model based on Generative Adversarial Network (GAN), wherein the GAN includes a generator and a discriminator;

[0032] The GAN was used to enhance the quality of the reconstructed three-dimensional nuclear medicine images.

[0033] Preferably, the generator includes a deep generation subnet and a shallow generation subnet, wherein:

[0034] The input to the deep generative subnet is the preprocessed nuclear medicine scan image, and the output is the corresponding deep learning enhancement result;

[0035] The input to the shallow generation subnet is the preprocessed nuclear medicine scan image, and the output is the learning target of the corresponding deep learning method.

[0036] Preferably, the method further includes the following steps:

[0037] An iterative algorithm is used to solve the counting data optimization model, wherein the optimization model considers the counting data obtained from the back-projected 3D image, the forward-projected 3D image, and the sampling probe.

[0038] Based on the solution results of the optimization model, the reconstructed three-dimensional nuclear medicine image is optimized.

[0039] The beneficial effects of this invention are mainly reflected in the following aspects:

[0040] The method of this invention has several significant advantages and beneficial effects. First, by using a deep learning-enhanced count rate sparse backprojection algorithm neural network model, this method effectively integrates the feature extraction capabilities of deep learning with the physical model constraints of traditional algorithms, significantly improving the quality of image reconstruction. Particularly when processing low-dose, low-count-rate nuclear medicine images, this method demonstrates significant advantages, effectively suppressing noise and improving the signal-to-noise ratio and spatial resolution of the image.

[0041] Secondly, this invention employs an innovative training strategy that considers both forward and back-projection processes, achieving efficient optimization of model parameters. This strategy not only improves the model's convergence speed but also enhances the consistency and accuracy of the reconstructed images. By fixing the back-projection layer parameters and optimizing the image cascade layer parameters, this invention fully leverages the advantages of deep learning while maintaining the accuracy of the physical model.

[0042] Furthermore, the image enhancement model based on Generative Adversarial Networks (GANs) introduced in this invention further improves the visual quality of the reconstructed images. This method can effectively remove artifacts in images, enhance detailed features, and make the reconstructed images closer to real anatomical structures, providing more reliable imaging evidence for clinical diagnosis.

[0043] It is worth noting that the method of this invention significantly improves reconstruction efficiency while enhancing image quality. Through the parallel computing capabilities of deep learning models, this method greatly reduces reconstruction time, making real-time or near-real-time SPECT image reconstruction possible, which is of great significance for clinical diagnosis and treatment decisions.

[0044] Finally, the method of this invention achieves the goal of maintaining high-quality image reconstruction while reducing radiation dose through optimized data acquisition and processing strategies. This not only reduces the risk of radiation exposure for patients but also expands the application scope of SPECT technology, especially for patients requiring multiple follow-up examinations and special populations sensitive to radiation, which has important clinical value.

[0045] In summary, the method for enhancing nuclear medicine single-photon emission computed tomography using deep learning provided by this invention has made significant progress in many aspects, such as image quality, reconstruction efficiency, and radiation dose control, opening up new avenues for the clinical application and further development of SPECT technology. Attached Figure Description

[0046] Figure 1This is a flowchart of the method of the present invention.

[0047] Figure 2 This is a flowchart of the neural network model construction process of the present invention.

[0048] Figure 3 This is a flowchart of the model training process of the present invention.

[0049] Figure 4 This is a flowchart of the image reconstruction process of the present invention.

[0050] Figure 5 This is a flowchart of the GAN optimization process of the present invention. Detailed Implementation

[0051] Please refer to Figures 1-5 This invention provides a method for enhancing nuclear medicine single-photon emission computed tomography (SPECT) using deep learning. This method improves image quality and reconstruction efficiency in nuclear medicine SPECT through deep learning technology. The specific embodiments of this invention will be described in detail below.

[0052] First, the method of the present invention includes an acquisition step, a processing step, and an output step. In the acquisition step, projection data from nuclear medicine single-photon emission computed tomography (SPECT) is acquired. This projection data is typically two-dimensional projection images acquired by a SPECT scanner at different angles.

[0053] In the processing steps, a deep learning-enhanced count rate sparse backprojection algorithm neural network model is constructed based on the acquired projection data. This model combines the feature extraction capabilities of deep learning with the advantages of the traditional count rate sparse backprojection algorithm. Subsequently, the neural network model is trained using the acquired projection data. During training, the model learns how to extract useful features from the projection data and optimizes the reconstruction process.

[0054] After training, the trained deep learning-enhanced counting rate sparse backprojection algorithm neural network model is used to reconstruct the projection data. This step converts the two-dimensional projection data into a three-dimensional nuclear medicine image.

[0055] In the output step, the reconstructed three-dimensional nuclear medicine image is output. Notably, the method of this invention simultaneously outputs both orthographic projection and orthographic-reverse projection three-dimensional nuclear medicine images. These two types of images provide complementary information, aiding physicians in making more accurate diagnoses.

[0056] Next, we will detail the structure of the deep learning-enhanced counting rate sparse backprojection algorithm neural network model. This model consists of two main parts: an image cascade layer and a backprojection layer.

[0057] The image cascaded layer consists of multiple 2D convolutional layers, pooling layers, and 2D deconvolutional layers. This structural design enables the model to effectively extract and process image features. Specifically, the 2D convolutional layers are used to extract local features, the pooling layers are used to reduce the spatial dimensionality of the feature map and improve the model's translation invariance, and the 2D deconvolutional layers are used to restore the spatial resolution of the feature map.

[0058] The backprojection layer uses a count rate sparse backprojection algorithm to calculate the reconstructed image data. This layer leverages the advantages of traditional algorithms to effectively handle the low count rate problem in nuclear medicine imaging. By combining deep learning and traditional algorithms, the method of this invention can improve the quality of the reconstructed image while maintaining computational efficiency.

[0059] During training, this invention employs an innovative training strategy. First, a neural network structure is constructed from the projection data of three-dimensional nuclear medicine images. This structural design allows the model to learn the reconstruction process directly from the projection data, without requiring pre-reconstructed images.

[0060] Secondly, an objective function is constructed that considers both forward and back projections. The objective function can be expressed as:

[0061] ,

[0062] in, , and These represent image similarity loss, projection data consistency loss, and temporal consistency loss, respectively. , and These are the corresponding weighting coefficients. This objective function design comprehensively considers various factors in the reconstruction process, which helps improve the quality of the reconstructed image.

[0063] During training, the parameters of the backprojection layer are fixed, and the parameters in the image cascade layers are iteratively updated. This strategy effectively balances the advantages of deep learning and traditional algorithms. Parameter updates employ backpropagation, and the specific process includes forward data propagation, loss calculation, gradient calculation, and parameter update.

[0064] Preferably, in one embodiment of the present invention, the data forward transmission process can be represented as follows:

[0065] ,

[0066] in, Indicates the first The output of the layer, For activation function, and These are the weights and bias parameters, respectively. The gradient of the loss with respect to the parameters can be expressed as:

[0067] ,

[0068] ,

[0069] The parameter update uses stochastic gradient descent, and the update formula is as follows:

[0070] ,

[0071] ,

[0072] in, The learning rate is typically between 0.001 and 0.1.

[0073] The method of this invention employs an innovative data acquisition approach when acquiring projection data from single-photon emission computed tomography (SPECT) images in nuclear medicine. First, the three-dimensional nuclear medicine image is initialized. The initialization method uses a sine wave, which provides a good initial estimate and accelerates the convergence process.

[0074] Next, the initialized nuclear medicine images are converted into three-dimensional models. This step uses a spherical coordinate back-projection method, which can effectively handle geometric distortion problems in SPECT imaging.

[0075] Finally, three-dimensional tomographic data is emitted using a photon detection probe to determine the detected events and projection data. This photon-count-based data acquisition method can better handle low-dose imaging problems, improving image quality while reducing radiation dose.

[0076] The method of this invention, by combining deep learning and traditional algorithms, not only improves the image quality of nuclear medicine single-photon emission computed tomography (SPECT) but also enhances reconstruction efficiency. This method has significant advantages in processing low-dose, low-count-rate nuclear medicine images, providing more reliable imaging evidence for clinical diagnosis. In a preferred embodiment of this invention, the iterative update process of the deep learning-enhanced count-rate sparse backprojection algorithm neural network model is further refined. This process includes three main steps: data forward propagation, loss gradient calculation, and parameter update.

[0077] First, the forward pass step calculates the loss of the data at each layer of the neural network. The specific calculation formula is as follows:

[0078] ,

[0079] in, Indicates the first Layer loss, For activation function, This is the weight matrix. This is the output of the previous layer. This is a bias term. Preferably, the present invention uses ReLU (Rectified Linear Unit) as the activation function, and its expression is:

[0080] ,

[0081] The RelU function was chosen based on its wide application and excellent performance in deep learning, especially its advantages in solving the vanishing gradient problem.

[0082] Next, calculate the loss against the parameters. and The gradient.

[0083] ,

[0084] ,

[0085] in, and These represent the back-projected image and the front-projected image, respectively. and These are the model parameters. This gradient calculation method can effectively capture the influence of model parameters on the final reconstruction result, thereby guiding the direction of parameter optimization.

[0086] Finally, the values ​​of θ and ϕ are updated using backpropagation. The update formula is as follows:

[0087] ,

[0088] Grad ,

[0089] in, The learning rate is an important hyperparameter. In this invention, the initial learning rate is set to 0.001, and a learning rate decay strategy is adopted, reducing the learning rate to 90% of its original value every 100 iterations. This strategy enables the model to quickly approach the optimal solution in the early stages of training, while fine-tuning the parameters in the later stages of training improves the model's convergence performance.

[0090] The method of this invention employs an innovative data acquisition and preprocessing approach when acquiring projection data from single-photon emission computed tomography (SPECT) images in nuclear medicine. First, the three-dimensional nuclear medicine image is initialized. Initialization uses a sine wave method, the specific implementation of which is as follows:

[0091]

[0092] in, For the initial 3D image, For spatial coordinates, For amplitude, , , Here are the angular frequencies in each direction. For phase. In a preferred embodiment of the invention, Set to 1, (in (Image side length) This initialization method provides a good starting point and accelerates the model's convergence process.

[0093] Next, the initialized nuclear medicine image is converted into a three-dimensional model. This invention employs a spherical coordinate back projection method, the specific implementation of which is as follows:

[0094] ,

[0095] in, The coordinates are in the Cartesian coordinate system. These correspond to spherical coordinates. This transformation method can effectively handle geometric distortion problems in SPECT imaging and improve the accuracy of reconstructed images.

[0096] Finally, three-dimensional tomographic data is emitted using a photon detection probe to determine the detection events and projection data. In this step, the invention employs photon transmission simulation based on the Monte Carlo method. Specifically, for each emitted photon, its path is determined by the following probability distribution:

[0097] ,

[0098] in, For the photon's travel distance The probability of subsequent interaction. This is a linear attenuation coefficient. This method allows for a more realistic simulation of photon transmission within human tissue, resulting in more accurate projection data.

[0099] The method of this invention significantly improves the data quality of nuclear medicine single-photon emission computed tomography (SPECT) by employing these innovative data acquisition and preprocessing techniques. This not only provides more reliable input for subsequent image reconstruction but also helps reduce radiation dose and decrease the risk of radiation exposure for patients.

[0100] In summary, the deep learning-enhanced single-photon emission computed tomography (SPECT) method for nuclear medicine proposed in this invention achieves significant progress in improving image quality, reducing radiation dose, and enhancing reconstruction efficiency through innovative model structure design, optimized training strategies, and advanced data acquisition and preprocessing techniques. This provides new possibilities for the application of nuclear medicine imaging in clinical diagnosis, and promises to bring patients a more accurate and safer diagnostic experience. In another embodiment of this invention, to further improve the quality of the reconstructed image, a deep learning model based on generative adversarial networks (GANs) is introduced. This model includes two main components: a generator and a discriminator, used to enhance the quality of the reconstructed three-dimensional nuclear medicine image.

[0101] The generator consists of a deep generation subnet and a shallow generation subnet. The input to the deep generation subnet is the preprocessed nuclear medicine scan image, and its output is the corresponding deep learning enhancement result. The input to the shallow generation subnet is also the preprocessed nuclear medicine scan image, but its output is the learning objective of the corresponding deep learning method. This dual-subnet structure design can simultaneously capture both deep and shallow features of the image, thereby achieving a more comprehensive image enhancement effect.

[0102] Preferably, the deep generation subnet adopts a U-Net structure, where the encoder part consists of multiple convolutional layers and downsampling layers, and the decoder part consists of corresponding deconvolutional layers and upsampling layers. Specifically, the encoder structure can be represented as follows:

[0103] ,

[0104] in, Indicates the first One convolutional layer, The input image is used. The decoder structure can then be represented as:

[0105] ,

[0106] in, Indicates the first The shallow generation subnet uses a deconvolutional layer. This structural design effectively preserves image details while achieving high-quality image reconstruction. The shallow generation subnet employs a residual network structure, containing multiple residual blocks. The structure of each residual block can be represented as:

[0107]

[0108] in, For input, This is a residual mapping. This structural design effectively alleviates the gradient vanishing problem in deep network training and improves the model's learning ability. The discriminator adopts a PatchGAN-based design, which can distinguish the authenticity of local regions of an image. The output of the discriminator can be expressed as:

[0109] ,

[0110] in, is the sigmoid activation function, used to map the output to the interval [0,1], representing the probability that the discriminator considers the input image to be a real image.

[0111] ,

[0112] in, To combat the losses, To rebuild the losses, and These are the corresponding weighting coefficients. In a preferred embodiment of the present invention, Set to 0.001, Set to 1.0. The discriminator's loss function uses the standard binary classification cross-entropy loss:

[0113] ,

[0114] in, Represents the true data distribution. Indicates noise distribution. and These represent the generator and the discriminator, respectively.

[0115] The method of this invention also introduces an iterative algorithm to solve the counting data optimization model, further improving the quality of the reconstructed image. This optimization model considers three aspects: the back-projected 3D image, the forward-projected 3D image, and the counting data obtained from the sampling probe. Its objective function can be expressed as:

[0116] ,

[0117] in, The 3D image to be reconstructed For the system matrix, For observation data, For regularization terms, Here is the regularization parameter. In a preferred embodiment of the invention, total variational regularization is used, i.e.:

[0118] ,

[0119] in, and They represent , and Gradient operator for direction.

[0120] To solve this optimization problem, this invention employs the Alternating Direction Multiplier Method (ADMM). Specifically, auxiliary variables are introduced. The original problem is transformed into:

[0121] ,

[0122] in, This represents the gradient operator. The iterative steps of ADMM are as follows:

[0123] 1. Update

[0124] ,

[0125] 2. Update

[0126] ,

[0127] 3. Update dual variables

[0128] ,

[0129] in, For penalty parameters, This indicates the number of iterations. In an embodiment of the invention, The initial value is set to 0.1, and an adaptive strategy is used for adjustment.

[0130] Through this iterative optimization method, this invention can effectively address the low signal-to-noise ratio and sparse sampling problems in nuclear medicine imaging, significantly improving the quality of reconstructed images. Combining the deep learning-enhanced count rate sparse backprojection algorithm neural network model and the GAN-based image enhancement model introduced earlier, the method proposed in this invention achieves multiple innovations and breakthroughs in the field of nuclear medicine single-photon emission computed tomography.

[0131] This comprehensive optimization method not only improves the quality and efficiency of image reconstruction but also effectively reduces radiation dose, providing patients with a safer and more accurate diagnostic experience. Furthermore, the method of this invention has good adaptability and scalability, allowing for flexible adjustment and optimization according to specific application scenarios and needs, providing new ideas and possibilities for the further development of nuclear medicine imaging technology.

[0132] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for enhancing nuclear medicine single-photon emission computed tomography using deep learning, characterized in that, include: The acquisition steps include: Acquire projection data from single-photon emission computed tomography (SPECT) imaging in nuclear medicine; The processing steps include: Based on the projection data, a deep learning enhanced count rate sparse backprojection algorithm neural network model is constructed. The deep learning enhanced count rate sparse backprojection algorithm neural network model includes an image cascade layer and a backprojection layer. The image cascade layer includes multiple two-dimensional convolutional layers, pooling layers, and two-dimensional deconvolutional layers. The backprojection layer uses the count rate sparse backprojection algorithm to calculate the reconstructed image data. Training the deep learning-enhanced count rate sparse backprojection algorithm neural network model using the projection data includes: constructing a neural network structure for the projection data of a three-dimensional nuclear medicine image; constructing an objective function, wherein the objective function simultaneously considers forward projection and backprojection; fixing the parameters of the backprojection layer and iteratively updating the parameters in the image cascade layer; The projection data is reconstructed based on the trained deep learning-enhanced counting rate sparse back projection algorithm neural network model; Output steps, including: Output the reconstructed three-dimensional nuclear medicine image, wherein the three-dimensional nuclear medicine image includes a forward-projected three-dimensional nuclear medicine image and a forward-backward-projected three-dimensional nuclear medicine image.

2. The method according to claim 1, characterized in that, The iterative update steps include: Perform forward data transmission and calculate the loss at each layer; Calculate the gradient of the loss with respect to the parameters; Update parameters using backpropagation.

3. The method according to claim 1, characterized in that, The steps for acquiring projection data from nuclear medicine single-photon emission computed tomography (SPECT) include: Initialize the three-dimensional nuclear medicine image; Convert the initialized nuclear medicine images into three-dimensional models; Three-dimensional tomographic data is emitted using a photon detection probe to determine the detection event and the projection data.

4. The method according to claim 3, characterized in that, The step of initializing the three-dimensional nuclear medicine image uses a sine wave for initialization, and the specific method is as follows: The initial values ​​for the forward projection and reverse projection are generated using a sine function, where the initial value for the forward projection is the amplitude of the sine curve, and the initial value for the reverse projection is the phase of the sine curve.

5. The method according to claim 3, characterized in that, The step of converting the initialized nuclear medicine image into a three-dimensional model employs a spherical coordinate back projection method, specifically including: Convert the pixel coordinates of a two-dimensional image to three-dimensional coordinates in a spherical coordinate system; Construct a three-dimensional model based on the three-dimensional coordinates in the spherical coordinate system.

6. The method according to claim 1, characterized in that, It also includes the following steps: Construct a deep learning model based on Generative Adversarial Network (GAN), wherein the GAN includes a generator and a discriminator; The GAN was used to enhance the quality of the reconstructed three-dimensional nuclear medicine images.

7. The method according to claim 6, characterized in that, The generator includes a deep generation subnet and a shallow generation subnet, wherein: The input to the deep generative subnet is the preprocessed nuclear medicine scan image, and the output is the corresponding deep learning enhancement result; The input to the shallow generation subnet is the preprocessed nuclear medicine scan image, and the output is the learning target of the corresponding deep learning method.

8. The method according to claim 1, characterized in that, The method further includes the following steps: An iterative algorithm is used to solve the counting data optimization model, wherein the optimization model considers the counting data obtained from the back-projected 3D image, the forward-projected 3D image, and the sampling probe. Based on the solution results of the optimization model, the reconstructed three-dimensional nuclear medicine image is optimized.

Images