A method and system for image reconstruction

Abstract

Embodiments of the present specification provide an image reconstruction method and system. The method includes obtaining an initial positron emission tomography (PET) image, the initial PET image being generated from PET data acquired by a long-axis sparse detector; obtaining an image optimization model; and generating a target image based on the initial PET image and the image optimization model. The target image has a higher quality than the initial PET image, approaching or reaching the quality level of a long-axis full-detector PET image.

Description

Technical Field

[0001] This specification relates to the field of medical imaging, and in particular to a method and system for reconstructing positron emission tomography (PET) images. Background Technology

[0002] In recent years, PET imaging technology has been widely used in clinical examinations and medical diagnosis. Existing long-axis sparse PET detectors possess the characteristics of long-axis PET detectors, allowing for a single scan covering the entire patient, thus improving imaging efficiency; however, the resulting image quality is relatively poor. Conventional image optimization methods cannot achieve the desired results. Therefore, a system and method are needed to obtain high-quality PET images. Summary of the Invention

[0003] This specification provides an image reconstruction method in one aspect. The method includes: acquiring an initial positron emission tomography (PET) image, the initial PET image being generated from PET data acquired by a long-axis sparse detector; acquiring an image optimization model; and generating a target image based on the initial PET image and the image optimization model, the target image having a higher quality than the initial PET image.

[0004] In some embodiments, generating a target image based on the initial PET image and the image optimization model includes: standardizing the initial PET image; inputting the standardized initial PET image into the image optimization model to obtain a model output image; and performing inverse standardization on the model output image to obtain the target image.

[0005] In some embodiments, the standardization of the initial PET image includes: standardizing the initial PET image based on the mean and standard deviation of the initial PET image.

[0006] In some embodiments, the inverse normalization of the model output image includes: performing inverse normalization on the model output image based on the mean and standard deviation of the initial PET image.

[0007] In some embodiments, the image optimization model is generated by: acquiring long-axis full-detector PET sample data; generating long-axis sparse detector PET sample images based on the long-axis full-detector PET sample data; reconstructing long-axis full-detector PET sample images based on the long-axis full-detector PET sample data; and training an initial image optimization model using the long-axis sparse detector PET sample images and the long-axis full-detector PET sample images as training samples to obtain the image optimization model.

[0008] In some embodiments, generating a target image based on the initial PET image and the image optimization model includes: acquiring a computed tomography (CT) image that matches the initial PET image; and generating the target image based on the initial PET image, the CT image, and the image optimization model.

[0009] In some embodiments, generating a target image based on the initial PET image, CT image, and the image optimization model includes: standardizing the initial PET image and CT image; inputting the standardized initial PET image and CT image into the image optimization model to obtain a model output image; and performing inverse standardization on the model output image to obtain the target image.

[0010] In some embodiments, the image optimization model is generated by: acquiring long-axis full-detector PET sample data; generating long-axis sparse detector PET sample images based on the long-axis full-detector PET sample data; reconstructing long-axis full-detector PET sample images based on the long-axis full-detector PET sample data; acquiring CT sample images that match the long-axis full-detector PET sample images; and training an initial image optimization model using the long-axis sparse detector PET sample images, CT sample images, and long-axis full-detector PET sample images as training samples to obtain the image optimization model.

[0011] Another aspect of this specification provides an image reconstruction system. The system includes a processor, characterized in that the processor is configured to perform the image reconstruction method as described in any of the preceding claims.

[0012] This specification also provides an image reconstruction method. The image reconstruction method includes acquiring positron emission tomography (PET) data of a scanned object within a defined scanning range, the defined scanning range corresponding to a detector with a first density; acquiring a neural network-based optimization model; acquiring target PET data based on the PET data and the neural network-based optimization model, the target PET data corresponding to a detector with a second density within the defined scanning range, the first density being less than the second density; and reconstructing a target image of the scanned object based on the target PET data. Attached Figure Description

[0013] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0014] Figure 1This is an exemplary flowchart of an image reconstruction method according to some embodiments of this specification.

[0015] Figure 2 This is an exemplary flowchart of an image optimization model generated according to some embodiments of this specification.

[0016] Figure 3 This is an exemplary flowchart of an image reconstruction method according to other embodiments of this specification.

[0017] Figure 4 This is an exemplary flowchart of a generative image optimization model shown in other embodiments of this specification.

[0018] Figure 5 This is an exemplary block diagram of an imaging system according to some embodiments of this specification.

[0019] Figure 6 This is an exemplary structural diagram of an imaging device according to some embodiments of this specification.

[0020] Figure 7 It is an image optimization model based on some embodiments shown in this specification. Detailed Implementation

[0021] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0022] It should be understood that the terms “system,” “device,” “unit,” and / or “module” used herein are one way to distinguish different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0023] As indicated in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0024] Flowcharts are used in this specification to illustrate the operations performed by the system according to embodiments of this specification. The related descriptions are provided to aid in a better understanding of the medical imaging methods and / or systems. It should be understood that preceding or subsequent operations are not necessarily performed precisely in sequence. Instead, steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0025] This application provides an image reconstruction method in several embodiments. The method includes acquiring an initial PET image generated from PET data acquired by a long-axis sparse detector; acquiring an image optimization model; and generating a target image based on the initial PET image and the image optimization model. The target image has higher quality than the initial PET image, approaching or reaching the quality level of a long-axis full-detector PET image. The image optimization model is a machine learning model. By optimizing the initial PET image generated from the PET data acquired by the long-axis sparse detector using the image optimization model, the quality of the target image is high (higher signal-to-noise ratio and resolution), approaching or reaching the quality level of a long-axis full-detector PET image. Furthermore, the long-axis sparse detector has lower cost due to the smaller number of detectors, and the reconstruction speed is faster due to the smaller amount of PET data acquired. Moreover, the long-axis sparse detector possesses the properties of a long-axis detector, allowing for full-body image reconstruction in a single scan, eliminating the need for multiple scans and stitching, thus improving imaging speed.

[0026] Figure 1 This is an exemplary flowchart of an image reconstruction method according to some embodiments of this specification.

[0027] The entity performing the image reconstruction method 100 may include a processing device. In some embodiments, the processing device may be a standalone electronic device, integrated into a medical scanning device (e.g., a long-axis PET scanner described below), or located on a cloud server. For example, the processing device may be the control panel of the medical scanning device, a personal computer, a laptop computer, a smartphone, a tablet computer, or a portable wearable device. In some embodiments, the image reconstruction method 100 may be performed by an imaging system 500.

[0028] In some embodiments, the image reconstruction method 100 may include:

[0029] Step 110: Acquire an initial PET image. In some embodiments, step 110 may be performed by the acquisition module 510.

[0030] The initial PET image is generated by a long-axis PET scanner by scanning the target object and acquiring PET data. A long-axis PET scanner refers to a PET scanner whose detector's axial field of view exceeds a certain threshold. PET scanning is a molecular imaging device for functional metabolic imaging. PET scanning uses positron-emitting radionuclides as tracers. By observing the uptake of the tracer by the region of interest (ROI), the functional metabolic state of the RIO can be understood. The PET image generated based on the PET data (e.g., the initial PET image) can provide detailed functional and metabolic molecular information about the RIO. The threshold can be, for example, 1 meter (m), 1.2m, 1.4m, 1.6m, 1.8m, 2m, etc. The threshold can be determined by the user or set by the system. For example, a long-axis PET scanner has an axial field of view of approximately 2m, and a single scan can cover all organs of the human body.

[0031] The detector of the long-axis PET scanning device includes multiple detector units. These detector units are arranged along the axial direction of the long-axis PET scanning device. In this embodiment, the detector of the long-axis PET scanning device is a long-axis sparse detector. The detector crystal arrangement of the long-axis sparse detector is sparser than that of a long-axis full detector with the same axial field of view, with spacing between adjacent detectors. The number of detector crystals in the long-axis sparse detector is less than the number of detector crystals in a long-axis full detector with the same axial field of view. A long-axis full detector refers to multiple detector crystals arranged adjacently and filling the inner wall of the detector. A long-axis sparse detector can be understood as a detector obtained by removing some crystals from each detector unit in a certain regular or irregular manner, based on a long-axis full detector, while maintaining the advantages of the long axis, but with a relatively sparser crystal distribution. For example, the number of detector crystals in a long-axis sparse detector can be 4 / 5, 3 / 4, 2 / 3, 1 / 2, 1 / 3, 1 / 4, etc., of the number of detector crystals in a long-axis full detector with the same axial field of view. In some embodiments, the detector units of the long-axis sparse detector are uniformly arranged. The distance between adjacent detector units is the same, but the crystal distribution within each detector unit is relatively sparse. The crystal distribution pattern within each detector unit can be the same or different.

[0032] In some embodiments, the target object may include biological and / or non-biological objects. For example, the target object may include the human body or specific parts thereof, such as the chest cavity, abdomen, limbs, etc., or combinations thereof. As another example, the target object may be a human-made component of living or non-living organic and / or inorganic matter, such as a phantom.

[0033] PET data is generated by scanning a target object using a long-axis sparse detector (hereinafter referred to as a long-axis sparse detector). In some embodiments, because the long-axis sparse detector has a large axial field of view, a single scan of the target object can cover the entire target object. For example, the axial field of view of the long-axis sparse detector is approximately 2m, and for a target object, such as a patient who is 1.8m tall, a single scan can cover all organs of the patient.

[0034] The initial PET image can be generated by reconstructing the PET data. Typical PET reconstruction methods may include Filtered Back Projection (FBP) algorithm, Maximum Likelihood Estimation (MLEM), Least Squares (LS) algorithm, Bayesian-based Maximum A posteriori (MAP) algorithm, Ordered Subset EM (OSEM) algorithm, Generalized Spatial Update Expectation Maximization (SAGE) algorithm, Block Iteration EM (BI-EM) algorithm, etc.

[0035] In some embodiments, the acquisition module 510 may acquire an initial PET image from a processing device that reconstructs the PET data. In some embodiments, the reconstructed initial PET image is stored in a storage device (e.g., Figure 6 The initial PET image can be acquired from the storage device (610).

[0036] Step 120: Obtain the image optimization model. In some embodiments, step 120 may be performed by the acquisition module 510.

[0037] The image optimization model is used to optimize the initial PET image, improving image quality and resulting in an optimized image with lower image noise (or higher signal-to-noise ratio), higher resolution, and better contrast. In some embodiments, the image optimization model is a pre-trained machine learning model.

[0038] Exemplary machine learning models may include neural network models (e.g., deep learning models), generative adversarial networks (GANs), deep belief networks (DBNs), stacked autoencoders (SAEs), logistic regression (LR) models, support vector machines (SVMs), decision tree models, naive Bayes models, random forest models or restricted Boltzmann machines (RBMs), gradient boosting decision tree (GBDT) models, LambdaMART models, adaptive augmentation models, hidden Markov models, perceptron neural network models, Hopfield network models, etc., or any combination thereof. Exemplary deep learning models may include deep neural network (DNN) models, convolutional neural network (CNN) models, recurrent neural network (RNN) models, feature pyramid network (FPN) models, etc. Exemplary CNN models may include V-Net models, U-Net models, FB-Net models, Link-Net models, etc., or any combination thereof. In some embodiments, the image optimization model is a three-dimensional U-Net model. The model structure of the three-dimensional U-Net model includes multiple connection layers, convolutional layers, pooling layers, and upsampling layers, specifically as follows: Figure 7 As shown. In some embodiments, the image optimization model can be a neural network model, and therefore the image optimization model is also referred to as a neural network-based image optimization model.

[0039] In some embodiments, the image optimization model is obtained by training with long-axis sparse detector PET sample images and long-axis full detector PET sample images as sample pairs. Specifically, the long-axis sparse detector PET sample images are the training input images, and the long-axis full detector PET sample images are the training target images (also known as the gold standard). The long-axis sparse detector PET sample images refer to images reconstructed based on long-axis sparse detector PET sample data. The long-axis sparse detector PET sample data is generated by reconstructing PET data acquired by the long-axis full detector (also known as PET sample data acquired by the long-axis full detector, used to train the image optimization model) after performing a sparse operation (downsampling) to simulate the PET data acquired by the long-axis sparse detector. The long-axis full detector PET sample images refer to PET images reconstructed based on PET sample data acquired by the long-axis full detector. For a detailed description of generating the image optimization model, please refer to other parts of this specification, for example... Figure 2 Its description will not be repeated here.

[0040] In some embodiments, the acquisition module 510 may acquire the image optimization model from the processing device that generated the image optimization model. In some embodiments, the image optimization model is stored in a storage device (e.g., Figure 6The image optimization model can be obtained from the storage device (610).

[0041] Step 130: Based on the initial PET image and the image optimization model, generate the target image. In some embodiments, step 130 may be performed by the optimization module 520.

[0042] The initial PET image is optimized using an image optimization model to obtain the target image, which has higher quality than the initial PET image. As mentioned above, the image optimization model is trained using long-axis sparse detector PET sample images and long-axis full detector PET sample images as training sample pairs. Therefore, the quality of the target image is close to or reaches the quality level of long-axis full detector PET images with the same axial field of view.

[0043] In some embodiments, the initial PET image may be standardized (also known as normalized) before image optimization. In some embodiments, the initial PET image may be standardized based on the mean and / or standard deviation of the initial PET image. The mean of the initial PET image refers to the average value of the pixel values ​​of all pixels in the initial PET image. The standard deviation of the initial PET image refers to the standard deviation of the pixel values ​​of all pixels in the initial PET image. For example, the initial PET image may be standardized based on equation (1):

[0044]

[0045] Where I′ is the initial PET image after standardization (also known as the standardized PET image), I is the initial PET image, μ is the mean of the initial PET image, and σ is the standard deviation of the initial PET image.

[0046] The standardized PET image is input into the image optimization model to obtain the model output image (i.e., the PET image output by the image optimization model). The model output image is a high-quality image corresponding to the standardized PET image. The noise in the model output image is suppressed, and the structural information is well recovered, approaching the image quality level of the training target image.

[0047] Since the model output image corresponds to a standardized PET image, further inverse standardization is required to ensure quantitative consistency of the PET images. Inverse standardization can be considered the inverse or reduction operation of the standardization process described above. The inversely standardized model output image (also called the target image) corresponds to the initial PET image. In some embodiments, inverse standardization can be performed on the model output image based on the mean and / or standard deviation of the initial PET image. For example, inverse standardization can be performed on the model output image based on equation (2):

[0048] I target =I out *σ+μ, (2)

[0049] Among them, I target For the target image, I out Output an image for the model.

[0050] It should be noted that the above description of method 100 is for illustrative purposes only and does not limit the scope of this specification. Those skilled in the art can make various modifications and changes to method 100 under the guidance of this specification. However, these modifications and changes are still within the scope of this specification. For example, the initial PET image can be preprocessed before image optimization. Exemplarily, the preprocessing may include image enhancement (e.g., point spread function (PSF), time-of-flight (TOF), etc.), image denoising, and image correction (e.g., image registration, attenuation correction, scattering correction, detector efficiency normalization correction, random correction, decay correction).

[0051] In some embodiments, the image reconstruction method 100 can also obtain a target image by recovering the acquired PET data in the data domain. Specifically, PET data of the scanned object within a set scanning range and a neural network-based optimization model can be acquired. The set scanning range corresponds to a detector with a first density. The first density refers to the number of detector crystals per unit area. In some embodiments, the first density can also be represented by the number of detector crystals within the set scanning range. The neural network-based optimization model is used to optimize the PET data to obtain target PET data, which has a higher quality than the PET data. Based on the PET data and the neural network-based image optimization model, the target PET data can be acquired. The target PET data corresponds to a detector with a second density within the set scanning range. The first density is less than the second density. The second density refers to the number of detector crystals per unit area corresponding to the target PET data. In some embodiments, the second density can also be represented by the number of detector crystals within the set scanning range corresponding to the target PET data.

[0052] Similar to the image optimization model described above, the neural network-based optimization model can be obtained by training with long-axis sparse detector PET sample data and long-axis full detector PET sample data as training sample pairs. Therefore, the quality of the target PET data can approach or reach the quality level of long-axis full detector PET data with the same axial field of view.

[0053] After obtaining the target PET data, the target image of the scanned object can be reconstructed based on the target PET data.

[0054] Figure 2 This is an exemplary flowchart illustrating the generation of an image optimization model according to some embodiments of this specification. In some embodiments, process 200 may be executed by acquisition module 510 and model training module 530.

[0055] Process 200 can correspond to process 100. The method for generating an image optimization model described in process 200 generates... Figure 1 The image optimization model described herein. In some embodiments, the initial image optimization model can be trained based on long-axis sparse detector PET sample images and long-axis full detector PET sample images to generate the image optimization model. Specifically:

[0056] Step 210: Obtain PET sample data for the long-axis full detector.

[0057] Long-axis full-field detector PET sample data refers to PET data acquired by a long-axis full-field detector for training an initial image optimization model. In some embodiments, a long-axis full-field detector can be used to scan one or more sample objects to acquire PET data. The acquired PET data is the long-axis full-field detector PET sample data. For example, a long-axis full-field detector with an axial field of view of approximately 2m can be used to scan multiple patients to acquire PET data for those multiple patients. The PET data for those multiple patients is the long-axis full-field detector PET sample data.

[0058] Step 220: Generate long-axis sparse detector PET sample images based on long-axis full detector PET sample data.

[0059] By performing a sparsity operation on the PET sample data of the long-axis full detector, sparsed PET data collected by the long-axis full detector is generated, thereby simulating the PET data collected by the long-axis sparse detector. In this embodiment, the sparsed PET data collected by the long-axis full detector is determined as the PET sample data of the long-axis sparse detector.

[0060] In some embodiments, the sparsity operation may include downsampling, deletion, filtering, etc. For example, PET data acquired by a long-axis sparse detector can be generated by downsampling the long-axis full-detector PET sample data. The downsampling rate can be, for example, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, etc. During the downsampling process, the long-axis full-detector PET sample data can be sampled uniformly or non-uniformly. For example, when the downsampling rate is 50%, the next or next set of PET data in the long-axis full-detector PET sample data can be sampled at intervals of one or more PET data sets to generate the long-axis sparse detector PET sample data. As another example, PET data acquired by a long-axis sparse detector can be generated by partially deleting the long-axis full-detector PET sample data. The amount of data deleted can be, for example, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, etc., of the long-axis full-detector PET sample data. During the data deletion process, the PET sample data of the long-axis full detector can be deleted uniformly or non-uniformly.

[0061] Reconstructing the PET sample data of the long-axis sparse detector can generate PET sample images of the long-axis sparse detector. Typical PET image reconstruction methods include... Figure 1 The steps described in step 110 will not be repeated here. Based on any of the above PET image reconstruction methods, the PET sample data of the long-axis sparse detector is reconstructed to generate the PET sample image of the long-axis sparse detector.

[0062] Step 230: Based on the long-axis full detector PET sample data, reconstruct the long-axis full detector PET sample image.

[0063] The long-axis full-detector PET sample image is obtained by directly reconstructing the long-axis full-detector PET sample data. In some embodiments, the reconstruction protocol used to reconstruct the long-axis full-detector PET sample image and the long-axis sparse detector PET image is the same. The reconstruction protocol may include parameters such as reconstruction algorithm, number of subsets, number of iterations, filtering method, pixel size, and layer thickness.

[0064] Step 240: Using the PET sample images of the long-axis sparse detector and the PET sample images of the long-axis full detector as training samples, train the initial image optimization model to obtain the image optimization model.

[0065] In some embodiments, the PET sample images of the long-axis sparse detector can be used as input images for model training, and the PET sample images of the long-axis full detector can be used as target images for model training. Based on the PET sample images of the long-axis sparse detector and the PET sample images of the long-axis full detector, the initial image optimization model is iteratively optimized (e.g., updating the parameters of the initial image optimization model, such as weights) until the iteration termination condition is met, resulting in the trained image optimization model (i.e., the image optimization model). For example, the initial image optimization model can be a U-net model, a GAN model, etc. The iteration termination condition includes the value of the objective function (also known as the loss function) falling below a threshold during a certain iteration, the completion of a preset number of iterations, or the convergence of the objective function (the value of the objective function in the previous iteration is lower than the value in the current iteration by a preset value), etc. In some embodiments, optimization algorithms such as Adaptive Moment Estimation (ADAM) and Stochastic Gradient Descent (SGD) can be used to continuously update the weights of the initial image optimization model during training until the objective function tends to converge, resulting in a stable model, which is the image optimization model.

[0066] The objective function reflects the difference between the long-axis full-detector PET sample image and the image output by the initial image optimization model during the iteration process. The objective function may include a focus loss function, a cross-entropy loss function, a logarithmic loss function, a mean squared difference (MSE) loss function, and a structural similarity index (SSIM) loss function. In some embodiments, to achieve the highest resolution while suppressing noise and maintaining good structural contrast information, the objective function employs a hybrid loss function of MSE and SSIM.

[0067] In some embodiments, before training the initial image optimization model, the long-axis sparse detector PET sample images and the long-axis full detector PET sample images can be standardized respectively. In some embodiments, the long-axis sparse detector PET sample images and the long-axis full detector PET sample images can be standardized based on the mean and / or standard deviation of the long-axis sparse detector PET sample images and the long-axis full detector PET sample images. For example, the long-axis sparse detector PET sample images and the long-axis full detector PET sample images can be standardized based on equation (1).

[0068] In some embodiments, before training the initial image optimization model, the long-axis sparse detector PET sample images and the long-axis full detector PET sample images may be preprocessed separately. Exemplarily, the preprocessing may include image enhancement (e.g., point spread function (PSF), time of flight (TOF), etc.), image denoising, and image correction (e.g., image registration, attenuation correction, scattering correction, detector efficiency normalization correction, stochastic correction, decay correction).

[0069] It should be noted that the above description of method 200 is for illustrative purposes only and does not limit the scope of this specification. Those skilled in the art can make various modifications and changes to method 200 under the guidance of this specification. However, these modifications and changes remain within the scope of this specification.

[0070] Figure 3 This is an exemplary flowchart of an image reconstruction method according to other embodiments of this specification.

[0071] In some embodiments, process 300 may be based on process 100, further acquiring a computed tomography (CT) image matching the initial PET image, and generating a target image based on the initial PET image, the CT image, and the image optimization model. Specifically:

[0072] Step 310: Obtain the initial PET image.

[0073] Step 320: Obtain the image optimization model.

[0074] In some embodiments, steps 310 and 320 are related to Figure 1 Steps 110 and 120 of the described process 100 are the same or similar, and will not be repeated here.

[0075] Step 330: Obtain a CT image that matches the initial PET image.

[0076] The CT image matched with the initial PET image is generated by a CT scanning device by scanning CT data of the same target object (e.g., a human body) in the initial PET image. The CT scanning device can scan a slice of a certain thickness in a part of the human body to provide attenuation information and precise anatomical localization of the region of interest (e.g., lesion). The CT image is generated by reconstructing the CT data. Typical CT reconstruction algorithms include Fourier transform (FT) algorithms, back projection (BP) algorithms, filtered back projection (FBP) algorithms, iterative reconstruction (IR) algorithms, etc.

[0077] In some embodiments, the CT scanning device is part of a PET-CT scanning device. A PET-CT scanning device combines a PET scanning device and a CT scanning device to scan the target object. The PET-CT scanning device uses the same bed and the same image processing workstation to fuse PET images (e.g., the initial PET image) and CT images (e.g., CT images matched with the initial PET image). The PET image provides detailed functional and metabolic molecular information about the region of interest, while the CT image provides attenuation information and precise anatomical localization of the region of interest. A single imaging session can obtain tomographic images of the site in all directions, simultaneously reflecting the pathophysiological changes and morphological structure of the region of interest, providing an understanding of the overall condition, and significantly improving diagnostic accuracy.

[0078] Step 340: Generate a target image based on the initial PET image, CT image, and the image optimization model.

[0079] In some embodiments, before generating the target image based on the initial PET image, CT image, and the image optimization model, the initial PET image and / or the CT image are resampled so that the resampled initial PET image and / or the CT image have the same scale (e.g., pixel size). For example, if the pixel size of the CT image is 1.2 mm * 1.2 mm and the pixel size of the initial PET image is 4 mm * 4 mm, the CT image can be resampled to a pixel size of 4 mm * 4 mm. As another example, if the pixel size of the CT image is 1.2 mm * 1.2 mm and the pixel size of the initial PET image is 4 mm * 4 mm, both the initial PET image and the CT image can be resampled to a pixel size of 6 mm * 6 mm.

[0080] In some embodiments, after resampling, the resampled initial PET and CT images are standardized. In some embodiments, the resampled initial PET images can be standardized based on the mean and standard deviation of the resampled initial PET images, and the resampled CT images can be standardized based on the mean and standard deviation of the resampled CT images. For example, it can be based on... Figure 1 The standardization process described in step 130 (e.g., formula (1) or similar) standardizes the resampled initial PET and CT images respectively, which will not be elaborated here.

[0081] The standardized initial PET image (also known as a standardized PET image) and the standardized CT image (also known as a standardized CT image) are input into the image optimization model to obtain the model output image (i.e., the PET image output by the image optimization model). In some embodiments, the standardized PET image and the standardized CT image are input into the image optimization model through two different channels. The axial noise of the model output image is suppressed, and the structural information is well recovered.

[0082] Since the model output image corresponds to the standardized PET and CT images, further inverse normalization processing is required. The inverse normalized model output image (also called the target image) corresponds to the initial PET image. In some embodiments, the model output image can be inverse normalized based on the mean and / or standard deviation of the initial PET image. For example, the model output image can be inverse normalized based on equation (2) or a similar method, which will not be elaborated here.

[0083] In some embodiments, the CT image may be replaced with other types of images, such as magnetic resonance (MR) images. For neural networks, more complete input information is more beneficial for network training, and the CT image (and / or MR image) can provide relatively unclear anatomical information from the PET image, which is more conducive to the generation of the target PET image.

[0084] It should be noted that the above description of method 300 is for illustrative purposes only and does not limit the scope of this specification. Those skilled in the art can make various modifications and changes to method 300 under the guidance of this specification. However, these modifications and changes are still within the scope of this specification. For example, the initial PET image and the CT image can also be preprocessed. Exemplarily, the preprocessing may include image enhancement (e.g., point spread function (PSF), time-of-flight (TOF), etc.), image denoising, and image correction (e.g., image registration, attenuation correction, scattering correction, detector efficiency normalization correction, random correction, decay correction). The preprocessing can be performed before or after the resampling of the initial PET image and the CT image, or before or after the normalization process, without specific limitations herein.

[0085] Figure 4 This is an exemplary flowchart of a generative image optimization model shown in other embodiments of this specification.

[0086] Process 400 can correspond to process 300. The method for generating an image optimization model described in process 400 generates... Figure 3 The image optimization model described herein. In some embodiments, process 400 may be based on process 200, further acquiring CT sample images matching the long-axis full-detector PET sample images, and using the long-axis sparse detector PET sample images, CT sample images, and long-axis full-detector PET sample images as training samples to train an initial image optimization model, thereby obtaining the image optimization model. Specifically:

[0087] Step 410: Obtain PET sample data for the long-axis full detector.

[0088] Step 420: Generate long-axis sparse detector PET sample images based on long-axis full detector PET sample data.

[0089] Step 430: Based on the long-axis full detector PET sample data, reconstruct the long-axis full detector PET sample image.

[0090] In some embodiments, steps 410-430 and Figure 2 Steps 210-230 of the described process 200 are the same or similar, and will not be repeated here.

[0091] Step 440: Obtain a CT sample image that matches the long-axis full-detector PET sample image.

[0092] The CT sample image matching the long-axis full-detector PET sample image is generated by a CT scanning device by scanning CT sample data of one or more sample objects identical to those in the long-axis full-detector PET sample image. The CT sample image can be generated by reconstructing the CT sample data.

[0093] In some embodiments, the CT scanning device is part of a PET-CT scanning device. A PET-CT scanning device combines a PET scanning device with a CT scanning device to scan the one or more sample objects. The detector of the PET scanning device is the long-axis full-detector.

[0094] Step 450: Using the long-axis sparse detector PET sample image, CT sample image, and long-axis full detector PET sample image as training samples, train the initial image optimization model to obtain the image optimization model.

[0095] In some embodiments, the long-axis sparse detector PET sample images and the CT sample images can be used as input images for model training, and the long-axis full detector PET sample images can be used as target images for model training. Based on the long-axis sparse detector PET sample images, CT sample images, and long-axis full detector PET sample images, an initial image optimization model is iteratively optimized (e.g., updating the parameters of the initial image optimization model, such as weights) until the iteration termination condition is met, resulting in a trained image optimization model (i.e., the image optimization model). For example, the initial image optimization model can be a U-net model, a GAN model, etc.

[0096] In some embodiments, before training the initial image optimization model, the long-axis sparse detector PET sample images, the CT images, and the long-axis full detector PET sample images can be standardized respectively. In some embodiments, the long-axis sparse detector PET sample images, the CT images, and the long-axis full detector PET sample images can be standardized based on the mean and / or standard deviation of the long-axis sparse detector PET sample images, the CT images, and the long-axis full detector PET sample images. For example, the long-axis sparse detector PET sample images, the CT images, and the long-axis full detector PET sample images can be standardized based on equation (1) or a similar method.

[0097] In some embodiments, before training the initial image optimization model, the long-axis sparse detector PET sample images, the CT images, and the long-axis full detector PET sample images may be preprocessed respectively. Exemplarily, the preprocessing may include image enhancement (e.g., point spread function (PSF), time of flight (TOF), etc.), image denoising, and image correction (e.g., image registration, attenuation correction, scattering correction, detector efficiency normalization correction, stochastic correction, decay correction).

[0098] Figure 5 This is an exemplary block diagram of an imaging system according to some embodiments of this specification.

[0099] like Figure 5 As shown, the imaging system 500 may include an acquisition module 510, an optimization module 520, and a model training module 530. In some embodiments, the imaging system 500 may be implemented by an imaging device 600 (such as a processor 620).

[0100] The acquisition module 510 can acquire an initial PET image, a CT image matched with the initial PET image, an image optimization model, and model training samples (e.g., long-axis full-detector PET sample data and CT sample images matched with long-axis full-detector PET sample images).

[0101] The optimization module 520 can optimize the initial PET image based on an image optimization model to generate a target image. The target image has higher quality than the initial PET image, approaching or reaching the quality level of a long-axis full-detector PET image. In some embodiments, the optimization module 520 can generate the target image based on the initial PET image and the image optimization model. In some embodiments, the optimization module 520 can generate the target image based on the initial PET image, a CT image matching the initial PET image, and the image optimization model.

[0102] The model training module 530 can train an initial image optimization model based on model training samples to generate an image optimization model. In some embodiments, the model training module 530 can use long-axis sparse detector PET sample images and long-axis full detector PET sample images as training samples to train the initial image optimization model and obtain the image optimization model. In some embodiments, the model training module 530 can use long-axis sparse detector PET sample images, CT sample images, and long-axis full detector PET sample images as training samples to train the initial image optimization model and obtain the image optimization model.

[0103] It should be noted that the above description of the imaging system and its modules is for convenience only and should not limit this application to the scope of the embodiments described. It is understood that those skilled in the art, after understanding the principle of the system, may arbitrarily combine the modules or construct subsystems connected to other modules without departing from this principle. In some embodiments, Figure 5 The acquisition module 510, optimization module 520, and model training module 530 disclosed herein can be different modules within a single system, or a single module can implement the functions of two or more of the aforementioned modules. For example, the modules can share a single storage module, or each module can have its own separate storage module. Such variations are all within the scope of protection of this application.

[0104] Figure 6 This is an exemplary structural diagram of an imaging device 600 according to some embodiments of this specification. Figure 6 As shown, the imaging device 600 may include a memory 610, a processor 620, and a communication bus. The memory 610 and the processor 620 can communicate with each other via the communication bus. The processor 620 can be used to execute the dynamic imaging method provided in any of the above embodiments of this application.

[0105] In some embodiments, the processor 620 may be implemented as a central processing unit, server, terminal device, or any other possible processing device. In some embodiments, the aforementioned central processing unit, server, terminal device, or other processing device may be implemented on a cloud platform. In some embodiments, the aforementioned central processing unit, server, or other processing device may be interconnected with various terminal devices, and the terminal devices may perform information processing tasks or partial information processing tasks.

[0106] In some embodiments, memory 610 (or a computer-readable storage medium) may store data and / or instructions (such as computer instructions). In some embodiments, memory 610 may store computer instructions that processor 620 (or a computer) can read to execute the image reconstruction method provided in any embodiment of this specification. In some embodiments, the storage device may include mass storage, removable storage, volatile read-write storage, read-only storage (ROM), and any combination thereof. In some embodiments, the storage device may be implemented on a cloud platform.

[0107] The beneficial effects that the embodiments of this specification may bring include, but are not limited to: (1) by using a long-axis sparse detector, the quality of the reconstructed target image is high (higher signal-to-noise ratio and resolution), approaching or reaching the quality level of a long-axis full-detector PET image; (2) the long-axis sparse detector has a lower cost due to the smaller number of detectors, and at the same time, the reconstruction speed is fast due to the smaller amount of PET data collected; (3) the long-axis sparse detector has the properties of a long-axis detector, and can cover the whole body of the patient in one scan, and can reconstruct the whole body image of the patient in one scan without multiple scans and stitching.

[0108] The basic concepts have been described above. Obviously, for those skilled in the art, the detailed disclosure above is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

[0109] Furthermore, this specification uses specific terms to describe embodiments thereof. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Moreover, certain features, structures, or characteristics in one or more embodiments of this specification can be appropriately combined.

[0110] Furthermore, unless expressly stated in the claims, the order of processing elements and sequences, the use of numbers and letters, or other names described in this specification are not intended to limit the order of the processes and methods described herein. Although various examples have been discussed in the foregoing disclosure of some embodiments of the invention that are currently considered useful, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments; rather, the claims are intended to cover all modifications and equivalent combinations that conform to the spirit and scope of the embodiments described herein. For example, while the system components described above can be implemented using hardware devices, they can also be implemented solely using software solutions, such as installing the described system on existing servers or mobile devices.

[0111] Similarly, it should be noted that, in order to simplify the description disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments in this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0112] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of range in some embodiments of this specification are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

[0113] For each patent, patent application, patent application publication, and other material, such as articles, books, specifications, publications, and documents, referenced in this specification, the entire contents of which are incorporated herein by reference. This excludes historical application documents that are inconsistent with or conflict with the content of this specification, as well as documents that limit the broadest scope of the claims in this specification (currently or subsequently appended to this specification). It should be noted that in the event of any inconsistency or conflict between the descriptions, definitions, and / or terminology used in the supplementary materials to this specification and the content of this specification, the descriptions, definitions, and / or terminology used in this specification shall prevail.

[0114] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. An image reconstruction method for a long-axis sparse detector, characterized in that, include: Acquire initial positron emission tomography (PET) images, which are generated from PET data acquired by a long-axis sparse detector; The image optimization model is obtained, and the image optimization model is generated in the following way: Acquire long-axis full-detector PET sample data; Based on the long-axis full detector PET sample data, a long-axis sparse detector PET sample image is generated. There is a gap between adjacent detector crystals of the long-axis sparse detector, and the number of detector crystals of the long-axis sparse detector is less than the number of detector crystals of the long-axis full detector with the same axial field of view. Based on the long-axis full detector PET sample data, the long-axis full detector PET sample image is reconstructed; Using the PET sample images of the long-axis sparse detector and the PET sample images of the long-axis full detector as training samples, an initial image optimization model is trained to obtain the image optimization model; and Based on the initial PET image and the image optimization model, a target image is generated, which has a higher quality than the initial PET image.

2. The image reconstruction method according to claim 1, wherein generating the target image based on the initial PET image and the image optimization model includes: The initial PET image is then standardized. The standardized initial PET image is input into the image optimization model to obtain the model output image; as well as The target image is obtained by performing inverse normalization on the model output image.

3. The image reconstruction method according to claim 2, wherein the standardization processing of the initial PET image includes: The initial PET images are standardized based on their mean and standard deviation.

4. The image reconstruction method according to claim 2, wherein the inverse normalization processing of the model output image includes: Based on the mean and standard deviation of the initial PET image, the model output image is subjected to inverse normalization.

5. The image reconstruction method according to claim 1, wherein generating the target image based on the initial PET image and the image optimization model comprises: Acquire a computed tomography (CT) image that matches the initial PET image; The target image is generated based on the initial PET image, CT image, and the image optimization model.

6. The image reconstruction method according to claim 5, wherein generating the target image based on the initial PET image, CT image, and the image optimization model comprises: The initial PET and CT images are standardized. The standardized initial PET and CT images are input into the image optimization model to obtain the model output image; and The target image is obtained by performing inverse normalization on the output image of the model.

7. The image reconstruction method according to claim 5, wherein the image optimization model generation process further includes: Acquire CT sample images that match the long-axis full-detector PET sample images; The step of training an initial image optimization model using the long-axis sparse detector PET sample images and the long-axis full detector PET sample images as training samples, and obtaining the image optimization model includes: Using the PET sample images, CT sample images, and PET sample images of the long-axis sparse detector as training samples, an initial image optimization model is trained to obtain the image optimization model.

8. The image reconstruction method according to claim 1, wherein generating a long-axis sparse detector PET sample image based on the long-axis full detector PET sample data comprises: Long-axis sparse detector PET sample data is generated by performing sparse operations on the long-axis full detector PET sample data. The PET sample data of the long-axis sparse detector is reconstructed to generate the PET sample image of the long-axis sparse detector.

9. An image reconstruction system for a long-axis sparse detector, characterized in that, The device includes a processor, characterized in that the processor is configured to perform the image reconstruction method as described in any one of claims 1-8.

10. An image reconstruction method for a long-axis sparse detector, characterized in that, include: Acquire positron emission tomography (PET) data of the scanned object within a set scanning range, wherein the set scanning range corresponds to a detector with a first density, and the first density represents the number of detector crystals per unit area; A neural network-based image optimization model is obtained, which is generated in the following way: Acquire long-axis full-detector PET sample data; Based on the long-axis full detector PET sample data, generate long-axis sparse detector PET sample images; Based on the long-axis full detector PET sample data, the long-axis full detector PET sample image is reconstructed; Using the PET sample images of the long-axis sparse detector and the PET sample images of the long-axis full detector as training samples, an initial image optimization model is trained to obtain the image optimization model. Based on the PET data and the neural network-based optimization model, target PET data is obtained. This target PET data corresponds to a detector with a second density set within a defined scanning range. The first density is less than the second density, and the second density represents the number of detector crystals per unit area corresponding to the target PET data. Reconstruct the target image of the scanned object based on the target PET data.

Images