SPECT kidney scanning imaging method and system based on deep learning, medium and equipment
By constructing an image dataset and using a deep learning neural network to process fast static kidney images, the problem of low imaging efficiency of static kidney images is solved, and fast, high-quality kidney scanning imaging is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI RADIODYNAMIC HEALTHCARE TECH
- Filing Date
- 2024-10-24
- Publication Date
- 2026-04-24
AI Technical Summary
Existing technologies have low imaging efficiency when acquiring static images of the kidney, requiring the acquisition of multiple perspectives and sufficient signal quantity for each perspective, resulting in excessively long scanning times.
Using a deep learning-based approach, an image dataset was constructed, and a neural network model was used to process fast renal static images and standard renal static images, including normalization and post-processing, to output fast renal static images with the same quality as standard renal static images.
While maintaining image quality, it significantly improves the imaging efficiency of static kidney images and reduces scan time.
Smart Images
Figure CN121921602A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of kidney scanning technology, and more specifically to a deep learning-based SPECT kidney scanning imaging method, system, medium, and device. Background Technology
[0002] Single-Photon Emission Computed Tomography (SPECT), also known as renal cortical imaging, involves intravenously injecting an imaging agent that can be taken up by renal parenchymal cells and slowly excreted. After flowing through the kidneys with the bloodstream, the imaging agent is either secreted by the renal tubules (99mTc-DMSA) or filtered by the glomeruli (99mTc-GH). Part of it is reabsorbed by the proximal convoluted tubular epithelial cells and binds to sulfhydryl groups in the cytoplasm, thus remaining in the cortex for a relatively long time. Through planar or tomographic imaging, the renal cortex can be clearly visualized to understand the location, size, shape, and parenchymal function of the kidney, and space-occupying lesions can also be detected.
[0003] Currently, to obtain comprehensive and high-quality static images of the kidney, existing workflows typically require acquiring static images of the kidney from three perspectives, and each perspective must acquire sufficient signal. Then, the process of performing high-quality scanning on multiple perspectives requires a long scanning time, resulting in low imaging efficiency of static kidney images. Summary of the Invention
[0004] To address the shortcomings of existing technologies, the present invention aims to provide a deep learning-based SPECT kidney scanning imaging method, system, medium, and device. The goal is to optimize the data processing logic of kidney static images to achieve the acquisition of high-quality kidney static images through rapid scanning, thereby improving the imaging efficiency of kidney static images while ensuring imaging quality.
[0005] A first aspect of the present invention is to provide a deep learning-based SPECT kidney scanning imaging method, the method comprising: Acquire fast renal static images acquired using a fast scanning method and standard renal static images acquired using a standard scanning method, and construct an image dataset based on the fast renal static images and the standard renal static images; The fast kidney static images and standard kidney static images in the image dataset are normalized. A neural network model for kidney imaging is constructed, the normalized image dataset is imported into the neural network model, and the neural network model is trained. The neural network model performs post-processing and inverse normalization on the fast kidney static image and the standard kidney static image, and outputs a fast kidney static image with image quality corresponding to that of the standard kidney static image.
[0006] According to one aspect of the above technical solution, acquiring rapid renal static images based on a rapid scanning method and standard renal static images based on a standard scanning method includes the following acquisition methods: Single-plane simulated downsampling: Obtain a standard quality multi-plane static kidney image that meets preset requirements, and obtain a simulated fast scan image of one of the planes by using the Poisson downsampling method; Single-plane true downsampling: Acquire standard quality multi-plane static kidney images that meet preset requirements, and perform an additional scan on one of the planes by reducing the acquisition count or shortening the scan time; Multiplanar simulated downsampling: Obtain standard quality multiplanar static kidney images that meet preset requirements, and obtain simulated fast scan images of all planes by applying the Poisson downsampling method; Multiplanar true downsampling: Acquire standard-quality multiplanar static kidney images that meet preset requirements, and perform an additional scan by reducing the acquisition count or shortening the scan time for all planes.
[0007] According to one aspect of the above technical solution, the step of normalizing the fast kidney static images and standard kidney static images in the image dataset includes: The fast and standard static kidney images in the image dataset are normalized according to their maximum and minimum values, respectively. The input and output of the fast renal static image X are: ; The input and output of the standard static kidney image Y are: ; In the formula, the max and min functions take the maximum and minimum values of the image, respectively.
[0008] According to one aspect of the above technical solution, the step of normalizing the fast kidney static images and standard kidney static images in the image dataset includes: The rapid kidney static image and the standard kidney static image are normalized according to the maximum and minimum values of the image in the rapid kidney static image. Input and output are: .
[0009] According to one aspect of the above technical solution, the neural network model is a convolutional neural network, a GAN network, or a transformer-based image generation network that takes a downsampled single fast kidney static image as input and outputs a fast kidney static image with image quality corresponding to the standard kidney static image.
[0010] According to one aspect of the above technical solution, it is characterized in that, ; In the formula, F represents the neural network model architecture. These are the parameters of the neural network F.
[0011] According to one aspect of the above technical solution, the neural network model includes a multi-plane feature extraction module and a target plane optimization module; The multi-plane feature extraction module consists of multiple convolutional or densely connected modules. By extracting the spatial information of the three input planes, a three-dimensional feature matrix of the scanned object is formed. The three-dimensional feature matrix and the static image of the kidney on the plane to be optimized are input into the target plane optimization module to obtain a fast static image of the kidney optimized by deep learning.
[0012] A second aspect of the present invention is to provide a deep learning-based SPECT kidney scanning imaging system, applied to the method described in the above-mentioned technical solution, the system comprising: The image acquisition module is used to acquire fast kidney static images acquired based on a fast scanning method and standard kidney static images acquired based on a standard scanning method, and to construct an image dataset based on the fast kidney static images and the standard kidney static images; The normalization processing module is used to normalize the fast kidney static images and standard kidney static images in the image dataset. A neural network construction module is used to construct a neural network model for kidney imaging, import the normalized image dataset into the neural network model, and train the neural network model. The image optimization module is used to perform post-processing and inverse normalization processing on the fast kidney static image and the standard kidney static image through the neural network model, and output a fast kidney static image with image quality corresponding to the standard kidney static image.
[0013] A third aspect of the present invention is to provide a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods described in the above-described technical solutions.
[0014] A fourth aspect of the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the method described in the above-described technical solutions.
[0015] Compared with existing technologies, the SPECT kidney scanning imaging method, system, medium, and device based on deep learning as shown in this invention have the following advantages: This invention acquires rapid renal static images using a fast scanning method and standard renal static images using a standard scanning method. An image dataset is constructed based on these two images. The rapid and standard renal static images in the dataset are then normalized. A neural network model for renal imaging is constructed, and the normalized image dataset is imported into the neural network model for training. Finally, the neural network model performs post-processing and de-normalization on the rapid and standard renal static images, outputting rapid renal static images with image quality corresponding to those of the standard renal static images. This invention only requires acquiring rapid renal static images using a fast scanning method and then simultaneously importing both rapid and standard renal static images into a neural network model for processing. This allows for the output of rapid renal static images with image quality corresponding to those of the standard renal static images, significantly improving the imaging efficiency of SPECT renal scans while maintaining image quality, thus facilitating rapid subsequent treatment. Attached Figure Description
[0016] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart illustrating a deep learning-based SPECT kidney scanning imaging method in one embodiment of the present invention. Figure 2 This is a structural block diagram of a deep learning-based SPECT kidney scanning imaging system according to an embodiment of the present invention. Detailed Implementation
[0017] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be more thorough and complete.
[0018] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.
[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.
[0020] Example 1 Please see Figure 1 The first embodiment of the present invention provides a deep learning-based SPECT kidney scanning imaging method, the method comprising steps S10-S40: Step S10: Obtain fast kidney static images acquired using a fast scanning method and standard kidney static images acquired using a standard scanning method, and construct an image dataset based on the fast kidney static images and the standard kidney static images.
[0021] In this embodiment, acquiring rapid renal static images based on a rapid scanning method and standard renal static images based on a standard scanning method includes the following acquisition methods: Single-plane simulated downsampling: Obtain a standard quality multi-plane static kidney image that meets preset requirements, and obtain a simulated fast scan image of one of the planes by using the Poisson downsampling method; Single-plane true downsampling: Acquire standard quality multi-plane static kidney images that meet preset requirements, and perform an additional scan on one of the planes by reducing the acquisition count or shortening the scan time; Multiplanar simulated downsampling: Obtain standard quality multiplanar static kidney images that meet preset requirements, and obtain simulated fast scan images of all planes by applying the Poisson downsampling method; Multiplanar true downsampling: Acquire standard-quality multiplanar static kidney images that meet preset requirements, and perform an additional scan by reducing the acquisition count or shortening the scan time for all planes.
[0022] Considering that the above single-plane or multi-plane simulated downsampling only takes into account ideal signal attenuation and absorption, in actual acquisition, the signal may be affected by detector absorption and leakage. Therefore, this embodiment adopts an optimized low-count image simulation method to optimize the ideal Poisson downsampling method. The steps are as follows: 1. Generate a two-dimensional Gaussian matrix gaussian_matrix to determine the probability distribution of the sampling locations; 2. Traverse the image pixels. For each pixel (i, j), sample it multiple times based on the pixel value at that position. If the pixel value at that position is n, then sample it n times. 3. Optimized Poisson sampling process: For each sampling, a random number between 0 and 1 is generated. If this random number is less than the sampling probability p, sampling is performed. The sampling position is selected based on the probability distribution of the Gaussian matrix. Specifically, a position is randomly selected from the Gaussian matrix, and its offset (d_row, d_column) from the center of the image is obtained. Then, the sampling results are accumulated and added to the corresponding positions (i+d_row, j+d_column) in the output image.
[0023] Step S20: Normalize the fast kidney static images and standard kidney static images in the image dataset.
[0024] In some feasible embodiments, the step of normalizing the fast kidney static images and standard kidney static images in the image dataset includes: The fast and standard static kidney images in the image dataset are normalized according to their maximum and minimum values, respectively. The input and output of the fast renal static image X are: ; The input and output of the standard static kidney image Y are: ; In the formula, the max and min functions take the maximum and minimum values of the image, respectively.
[0025] More preferably, the step of normalizing the fast kidney static images and standard kidney static images in the image dataset includes: The rapid kidney static image and the standard kidney static image are normalized according to the maximum and minimum values of the image in the rapid kidney static image. Input and output are: .
[0026] Step S30: Construct a neural network model for kidney imaging, import the normalized image dataset into the neural network model, and train the neural network model.
[0027] Step S40: The fast kidney static image and the standard kidney static image are post-processed and denormalized using the neural network model to output a fast kidney static image with image quality corresponding to that of the standard kidney static image.
[0028] In this embodiment, the neural network model is a convolutional neural network, a GAN network, or a transformer-based image generation network that takes a downsampled single fast kidney static image as input and outputs a fast kidney static image with image quality corresponding to the standard kidney static image.
[0029] in, ; In the formula, F represents the neural network model architecture. These are the parameters of the neural network F.
[0030] Furthermore, the neural network model includes a multi-plane feature extraction module and a target plane optimization module; The multi-plane feature extraction module consists of multiple convolutional or densely connected modules. By extracting the spatial information of the three input planes, a three-dimensional feature matrix of the scanned object is formed. The three-dimensional feature matrix and the static image of the kidney on the plane to be optimized are input into the target plane optimization module to obtain a fast static image of the kidney optimized by deep learning.
[0031] More specifically, in this embodiment, the optimization of SPECT kidney static images using deep learning includes the following steps: (1) Construct the training dataset; (2) Preprocessing and normalization of training data; (3) Neural network model construction; (4) Train the neural network using the training data; (5) Perform post-processing and inverse normalization on the processed data to obtain the output results.
[0032] The first step involves acquiring matched pairs of rapid scan and standard-quality static kidney image data. Specific acquisition methods include the following: (1) Single-plane simulated downsampling: Obtain standard quality multi-plane static kidney images that meet clinical diagnostic needs, and obtain the corresponding simulated fast scan image for one of the planes by using the Poisson downsampling method.
[0033] (2) Single-plane true downsampling: Obtain standard quality multi-plane static kidney images that meet clinical diagnostic needs, and perform an additional scan by reducing the acquisition count or shortening the scan time for one of the planes.
[0034] (3) Multiplanar simulation downsampling: Obtain standard quality multiplanar static kidney images that meet clinical diagnostic needs, and obtain corresponding simulated fast scan images for all planes using the Poisson downsampling method.
[0035] (4) Multiplanar real downsampling: Obtain standard quality multiplanar static kidney images that meet clinical diagnostic needs, and perform an additional scan by reducing the acquisition count or shortening the scan time for all planes.
[0036] Step 2: Normalization of static kidney images Normalization methods for kidney static images typically involve normalizing the image by taking the maximum and minimum values. For an input fast kidney static image X and a standard kidney static image Y, the normalized input and output are:
[0037]
[0038] The `max` and `min` functions retrieve the maximum and minimum values of the image, respectively. The advantage of this method is that it can be applied to images at different acceleration ratios. However, its disadvantages include the need for an additional acceleration factor to perform quantitative correction of the image, and the resulting image exhibits differences in stability and quantitative consistency compared to the standard image.
[0039] This embodiment also proposes an improved normalization method, which uses the maximum and minimum values of a fast renal static image X to normalize X and the corresponding standard renal static image Y. The input and output are:
[0040] This method does not require an additional acceleration factor to perform quantitative image correction, and the generated images show good quantitative consistency. Its drawback is that it can only achieve image optimization at a fixed magnification.
[0041] Step 3: Construct a kidney imaging enhancement neural network F Neural network models are typically convolutional neural networks, GAN networks, or transformer-based image generation networks that take a single downsampled fast static image of the kidney as input and a corresponding standard image as output.
[0042] Where F represents the neural network model architecture, These are the parameters of the neural network F.
[0043] Step 4: Train the neural network using the training data The training process of a neural network model is a process of minimizing a loss function using different training data:
[0044] Loss is the loss function, and common loss functions include one or more of the following: L1 loss, L2 loss, and SSIM loss.
[0045] During training, each set of multi-plane inputs can freely choose and change the combination of input directions, and each combination of directions and the final output constitute a training pair.
[0046] In this embodiment, an improved method for quantitatively optimizing renal loss is provided: In static kidney images, the quantitative difference between the left and right kidneys is of great concern to medical professionals. However, current deep learning reconstruction algorithms cannot effectively handle this difference. Therefore, this embodiment proposes a method for calculating a quantitative kidney optimization loss to accelerate reconstruction in anterior kidney imaging. The calculation method for this loss is as follows: 1. Kidney structure segmentation 1.1 The kidney structure is segmented using threshold segmentation, and its process is as follows: a. Select the maximum value of the static kidney image; b. Using a certain percentage of the maximum value as a threshold, segment out the suspected kidney region that is greater than the threshold; preferably, the percentage coefficient is 0.3; c. Obtain the connected components of the suspected kidney region, and select the two with the largest areas, corresponding to the left and right kidneys respectively; d. Due to the high granular noise in the static kidney image, the delineated boundaries are not smooth. Image expansion and image contraction are performed once each to obtain the kidney regions K_left and K_right with smooth edges.
[0047] 1.2. Kidney structure segmentation can also be obtained through deep learning algorithms.
[0048] 2. Obtaining the background region in static kidney images: To eliminate background signal interference from organs anterior and posterior to the kidney, the mean background signal needs to be calculated. The process is as follows: a. Calculate the x-coordinate range of the kidney region, i.e., the x-coordinate range x_range of the K_left and K_right regions; b. Delineate the background region several pixels below the kidney, i.e., the minimum Y value y_min -n of the K_left and K_right regions; n is generally chosen between 5 and 10. c. Obtain the background region: background = [x_range; y_min-nm:y_min-n]; m is the width of the background region, generally between 2 and 5; d. Calculate the mean value of the background signal mb = mean(background).
[0049] 3. Calculate the quantitative optimized loss of the left and right kidneys. The calculation expression is: Loss = Quant_(k_left)[ ] - Quant_(k_right)[Y'] Among them, Quant is a quantitative indicator for the left and right kidneys, which can be selected as the area of the region, the total count of the region, or the total count of the region minus the background area.
[0050] It is worth noting that even when scanning very slowly, the static images of the kidney have high image granularity and noise. Therefore, using adversarial generative loss can often achieve better image results.
[0051] Specifically, the neural network model includes a multi-plane feature extraction module and a target plane optimization module.
[0052] The multi-plane feature extraction module consists of multiple convolutional or densely connected units. It extracts spatial information from the three input planes to form a 3D feature matrix of the scanned object. This 3D feature matrix and the image of the plane to be optimized are then input into the target plane optimization module to obtain a fast scan optimized by deep learning. The target plane optimization module contains multiple layers of convolutional or densely connected units and symmetrical upsampling and downsampling units.
[0053] The neural network model shown in this embodiment uses a multi-plane feature optimization network. The input of the multi-plane feature optimization network is a multi-plane static kidney image, and the output is a single-plane static kidney image or a multi-plane static kidney image.
[0054] When the input data is a single-plane downsampled static image of the kidney, the input of the neural network model is a downsampled fast static image of the kidney and standard static images of the kidney from two other perspectives, and the output is a fast static image of the kidney corresponding to the image quality of the standard static image of the kidney.
[0055] When the input data is a multi-plane downsampled static image of the kidney, the input of the neural network model is a fast static image of the kidney in three planes, and the output is an optimized image in three planes.
[0056] Compared with existing technologies, the SPECT kidney scanning imaging method based on deep learning shown in this embodiment has the following advantages: By acquiring rapid renal static images using a fast scanning method and standard renal static images using a standard scanning method, an image dataset is constructed based on these two images. The rapid and standard renal static images in the image dataset are then normalized. A neural network model for renal imaging is constructed, and the normalized image dataset is imported into the neural network model for training. Finally, the neural network model performs post-processing and de-normalization on the rapid and standard renal static images, outputting rapid renal static images with image quality corresponding to that of the standard renal static images. This embodiment only requires acquiring rapid renal static images using a fast scanning method and then simultaneously importing both rapid and standard renal static images into the neural network model for processing to output rapid renal static images with image quality corresponding to that of the standard renal static images. Therefore, this embodiment can significantly improve the imaging efficiency of SPECT renal scanning while ensuring image quality, thereby facilitating rapid subsequent treatment.
[0057] Example 2 Please see Figure 2 The second embodiment of the present invention provides a deep learning-based SPECT kidney scanning imaging system, which is applied to the method described in the above embodiments. The system includes: an image acquisition module 10, a normalization processing module 20, a neural network construction module 30, and an image optimization module 40.
[0058] Image acquisition module 10 is used to acquire fast kidney static images acquired based on a fast scanning method and standard kidney static images acquired based on a standard scanning method, and to construct an image dataset based on the fast kidney static images and the standard kidney static images; Normalization processing module 20 is used to normalize the fast kidney static images and standard kidney static images in the image dataset; The neural network construction module 30 is used to construct a neural network model for kidney imaging, import the normalized image dataset into the neural network model, and train the neural network model. The image optimization module 40 is used to perform post-processing and inverse normalization processing on the fast kidney static image and the standard kidney static image through the neural network model, and output a fast kidney static image with image quality corresponding to the standard kidney static image.
[0059] Compared with existing technologies, the SPECT kidney scanning imaging system based on deep learning shown in this embodiment has the following advantages: By acquiring rapid renal static images using a fast scanning method and standard renal static images using a standard scanning method, an image dataset is constructed based on these two images. The rapid and standard renal static images in the image dataset are then normalized. A neural network model for renal imaging is constructed, and the normalized image dataset is imported into the neural network model for training. Finally, the neural network model performs post-processing and de-normalization on the rapid and standard renal static images, outputting rapid renal static images with image quality corresponding to that of the standard renal static images. This embodiment only requires acquiring rapid renal static images using a fast scanning method and then simultaneously importing both rapid and standard renal static images into the neural network model for processing to output rapid renal static images with image quality corresponding to that of the standard renal static images. Therefore, this embodiment can significantly improve the imaging efficiency of SPECT renal scanning while ensuring image quality, thereby facilitating rapid subsequent treatment.
[0060] Example 3 A third embodiment of the present invention provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods described in the above embodiments.
[0061] Example 54 A fourth embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, it implements the methods described in the above embodiments.
[0062] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0063] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.
Claims
1. A deep learning-based SPECT kidney scanning imaging method, characterized in that, The method includes: Acquire fast renal static images acquired using a fast scanning method and standard renal static images acquired using a standard scanning method, and construct an image dataset based on the fast renal static images and the standard renal static images; The fast kidney static images and standard kidney static images in the image dataset are normalized. A neural network model for kidney imaging is constructed, the normalized image dataset is imported into the neural network model, and the neural network model is trained. The neural network model performs post-processing and inverse normalization on the fast kidney static image and the standard kidney static image, and outputs a fast kidney static image with image quality corresponding to that of the standard kidney static image.
2. The SPECT kidney scanning imaging method based on deep learning according to claim 1, characterized in that, Acquiring rapid static kidney images based on fast scanning methods and standard static kidney images based on standard scanning methods, including the following acquisition methods: Single-plane simulated downsampling: Obtain a standard quality multi-plane static kidney image that meets preset requirements, and obtain a simulated fast scan image of one of the planes by using the Poisson downsampling method; Single-plane true downsampling: Acquire standard quality multi-plane static kidney images that meet preset requirements, and perform an additional scan on one of the planes by reducing the acquisition count or shortening the scan time; Multiplanar simulated downsampling: Obtain standard quality multiplanar static kidney images that meet preset requirements, and obtain simulated fast scan images of all planes by applying the Poisson downsampling method; Multiplanar true downsampling: Acquire standard-quality multiplanar static kidney images that meet preset requirements, and perform an additional scan by reducing the acquisition count or shortening the scan time for all planes.
3. The SPECT kidney scanning imaging method based on deep learning according to claim 1, characterized in that, The steps for normalizing the fast kidney static images and standard kidney static images in the image dataset include: The fast and standard static kidney images in the image dataset are normalized according to their maximum and minimum values, respectively. The input and output of the fast renal static image X are: ； The input and output of the standard static kidney image Y are: ； In the formula, the max and min functions take the maximum and minimum values of the image, respectively.
4. The deep learning-based SPECT kidney scanning imaging method according to claim 1, characterized in that, The steps for normalizing the fast kidney static images and standard kidney static images in the image dataset include: The rapid kidney static image and the standard kidney static image are normalized according to the maximum and minimum values of the image in the rapid kidney static image. Input and output are: 。 5. The deep learning-based SPECT kidney scanning imaging method according to claim 4, characterized in that, The neural network model is a convolutional neural network, a GAN network, or a transformer-based image generation network that takes a downsampled single fast kidney static image as input and outputs a fast kidney static image with image quality corresponding to the standard kidney static image.
6. The deep learning-based SPECT kidney scanning imaging method according to claim 5, characterized in that, ； In the formula, F represents the neural network model architecture. These are the parameters of the neural network F.
7. The deep learning-based SPECT kidney scanning imaging method according to any one of claims 1-6, characterized in that, The neural network model includes a multi-plane feature extraction module and a target plane optimization module; The multi-plane feature extraction module consists of multiple convolutional or densely connected modules. By extracting the spatial information of the three input planes, a three-dimensional feature matrix of the scanned object is formed. The three-dimensional feature matrix and the static image of the kidney on the plane to be optimized are input into the target plane optimization module to obtain a fast static image of the kidney optimized by deep learning.
8. A deep learning-based SPECT kidney scanning imaging system, characterized in that, The system, applicable to the method of any one of claims 1-7, comprises: The image acquisition module is used to acquire fast kidney static images acquired based on a fast scanning method and standard kidney static images acquired based on a standard scanning method, and to construct an image dataset based on the fast kidney static images and the standard kidney static images; The normalization processing module is used to normalize the fast kidney static images and standard kidney static images in the image dataset. A neural network construction module is used to construct a neural network model for kidney imaging, import the normalized image dataset into the neural network model, and train the neural network model. The image optimization module is used to perform post-processing and inverse normalization processing on the fast kidney static image and the standard kidney static image through the neural network model, and output a fast kidney static image with image quality corresponding to the standard kidney static image.
9. A readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method described in any one of claims 1-7.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and operable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1-7.