PET image reconstruction method and pet-ct system

Abstract

The application discloses a PET image reconstruction method and a PET-CT system, and belongs to the technical field of medical imaging. The method comprises the following steps: acquiring an anatomical structure image and PET data of a scanning object, the anatomical structure image covering a partial region of the scanning object, and the axial range of the anatomical structure image being smaller than the axial range corresponding to the PET data; determining an extended attenuation image according to the anatomical structure image, the axial range of the extended attenuation image being larger than the axial range of the anatomical structure image; and reconstructing the PET data according to the extended attenuation image to obtain a PET image of the scanning object. The application solves the technical problem of the insufficient axial range of the anatomical structure image in the prior art when the PET image is reconstructed, and improves the accuracy of the PET image reconstruction.

Description

Technical Field

[0001] This application relates to the fields of image processing and network modeling, and in particular to a PET image reconstruction method and a PET-CT system. Background Technology

[0002] With the rapid development and widespread adoption of whole-body PET (Positron Emission Tomography) and ultra-long axial PET, these types of PET are increasingly accepted and valued in clinical practice. In actual equipment use, single-site scans are usually much smaller than the axial range of whole-body PET or ultra-long axial PET. For example, the uEXPLORER has a whole-body PET axial length of 194 cm, but a single-site scan, such as the brain, usually only requires 20-30 cm. Therefore, partial axial field-of-view (AFOV) reconstruction is widely used in clinical practice.

[0003] Partial axial AFOV reconstruction specifically refers to the method of reconstruction and correction using partial data within the axial field of view. This reconstruction method is a subset of whole-body PET reconstruction and is also a forced trade-off. In the existing technology, if AFOV images are needed, they need to be cropped from the whole-body PET reconstruction images. This method results in slow image output speed and waste of resources due to the large amount of data processed.

[0004] In existing technologies, single-site scanning typically only scans the area of ​​computed tomography (CT) or magnetic resonance imaging (MRI) with a length equal to that of the PET scan. However, the attenuation correction used in PET reconstruction relies on an anatomical image generated by CT or MRI with a range greater than or equal to that of the PET image to obtain more attenuation data for attenuation correction and improve image accuracy. Axial positions outside the scanned area lack corresponding attenuation image data, making it impossible to use related data for reconstruction, thus affecting the accuracy of the reconstructed image. Summary of the Invention

[0005] To address the technical problem that current partial axial field of view (AFOV) reconstruction methods in whole-body PET and ultra-long axial PET equipment only use attenuated images corresponding to CT images for attenuation correction and image reconstruction, resulting in limited sensitivity and inability to fully utilize all axial field of view data of the PET equipment, this invention provides a PET image reconstruction method to increase the data used for AFOV reconstruction, improve its sensitivity, and reduce noise.

[0006] The technical solution adopted by the present invention to solve its technical problem is: to provide a PET image reconstruction method, comprising: acquiring an anatomical structure image and PET data of a scanned object, wherein the axial range of the anatomical structure image is smaller than the axial range corresponding to the PET data; determining an extended attenuation image based on the anatomical structure image, wherein the axial range of the extended attenuation image is larger than the axial range of the anatomical structure image; and reconstructing the PET data based on the extended attenuation image to obtain a PET image of the scanned object.

[0007] Preferably, determining the extended attenuation image based on the anatomical structure image includes: determining the extended attenuation image based on the anatomical structure image using a first data extension model, wherein the first data extension model is a trained machine learning model.

[0008] Furthermore, determining the extended attenuation image using a first data expansion model based on the anatomical structure image includes: acquiring a contour image of the scanned object; and determining the extended attenuation image using a first data expansion model based on the attenuation image and the contour image of the scanned object, wherein the contour image can display the contour region of the scanned object.

[0009] Optionally, the method further includes: determining a preliminary anatomical structure image based on the PET data; determining a first predicted attenuation image using a second data expansion model based on the preliminary anatomical structure image and the contour image of the scanned object; and merging and stitching the anatomical structure image with the first predicted attenuation image to determine the expanded attenuation image.

[0010] Preferably, the method further includes: determining a preliminary anatomical structure image based on the PET data; and determining the extended attenuation image using a third data extension model based on the anatomical structure image, the contour image of the scanned object, and the preliminary anatomical structure image.

[0011] Furthermore, determining the extended attenuation image based on the anatomical structure image includes: determining a second predicted attenuation image based on the anatomical structure image; and cropping the second predicted attenuation image to obtain the extended attenuation image.

[0012] Optionally, the step of cropping the second predicted attenuation image to obtain the extended attenuation image includes: determining the cropping range of the second predicted attenuation image according to the scanning protocol of the PET data and / or the image center of the anatomical structure image, and cropping the second predicted attenuation image according to the cropping range to obtain the extended attenuation image.

[0013] The present invention also provides a PET image reconstruction method, comprising: acquiring a CT image of a scanned object; acquiring PET data of the scanned object; wherein a first axial range corresponding to the CT image is smaller than a second axial range corresponding to the PET data; inputting the CT image into a trained neural network model to acquire a predicted attenuation map, wherein a third axial range corresponding to the predicted attenuation map is larger than the first axial range; and reconstructing the PET data based on the predicted attenuation map to obtain a PET image of the scanned object.

[0014] Preferably, the method of reconstructing the PET data using the predicted attenuation map to obtain a PET image of the scanned object includes: acquiring an attenuation image of a CT image of the scanned object; merging the predicted image and the attenuation image of the CT image to obtain a target attenuation image; merging the predicted image and the attenuation image of the CT image to obtain the target attenuation image; and reconstructing the PET data based on the target attenuation image to obtain a PET image of the scanned object.

[0015] The present invention also provides a PET-CT system, comprising: a PET scanner, a CT scanner, and an integrated assembly thereof; a computer operatively coupled to the CT scanner and the PET scanner, the computer being configured to: control the CT scanner to perform scanning of a region of interest of a scanned object to obtain CT images; control the PET scanner to perform scanning of the region of interest to obtain PET data, wherein the axial field-of-view (AFOV) of the PET scanner is greater than that of the CT scanner; input the CT images into a trained neural network model to obtain a predicted attenuation map, the predicted attenuation map containing attenuation estimates of at least one peripheral organ located outside the axial field-of-view boundary of the CT scanner and inside the axial field-of-view boundary of the PET scanner; and iteratively reconstruct the PET data based on the predicted attenuation map to obtain a PET image of the scanned object.

[0016] The aforementioned PET image reconstruction method and PET-CT system determine the extended attenuation image based on anatomical structure images and use this extended attenuation image to reconstruct PET data. This enables effective attenuation correction for areas within the axial field of view of the PET scanner but outside the axial field of view of the CT scanner, improving the quality and accuracy of PET image reconstruction. By using machine learning models to predict the extended attenuation image, there is no need to increase the CT scan range, avoiding the problem of increasing the patient's radiation dose. It can predict attenuation images with a larger axial range using limited anatomical structure image information, improving the system's imaging efficiency. By combining multiple input data (such as contour images, preliminary anatomical structure images, etc.), the accuracy of the extended attenuation image is further improved, thus achieving more accurate PET image reconstruction. Attached Figure Description

[0017] Figure 1 This is an application environment diagram of a PET image reconstruction method in one embodiment.

[0018] Figure 2 This is a flowchart of a method for reconstructing a PET image based on an extended attenuation image in one embodiment.

[0019] Figure 3 This is a flowchart of a method for reconstructing a PET image based on an extended attenuation image in one embodiment.

[0020] Figure 4 This is a flowchart of a method for reconstructing a PET image using an extended attenuation image, as shown in another embodiment.

[0021] Figure 5 This is a flowchart of a method for reconstructing a PET image using an extended attenuation image, as shown in another embodiment.

[0022] Figure 6 This is a flowchart of a method for reconstructing a PET image using an extended attenuation image, as shown in another embodiment.

[0023] Figure 7 This is a block diagram of a system structure for reconstructing PET images using extended attenuation images, as shown in one embodiment.

[0024] Figure 8 This is an internal structural diagram of a computer device in one embodiment.

[0025] Figure 9 This is a schematic diagram of the associated LOR for full AFOV and partial AFOV reconstruction in one embodiment.

[0026] Figure 10 This is a schematic diagram illustrating the generation of a predicted attenuation image based on CT images and localization images, as an example of an embodiment.

[0027] Figure 11This is a schematic diagram illustrating the generation of a predicted attenuation image based on CT images, localization images, and external camera images, as one embodiment.

[0028] Figure 12 This is a schematic diagram illustrating the generation of a predicted attenuation image based on an MLTR attenuation image and a localization image, as an example of an embodiment.

[0029] Figure 13 This is a schematic diagram illustrating the generation of a predicted attenuation image based on a CT image, a localization image, and an MLTR attenuation image, as one embodiment. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0031] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0032] The PET image reconstruction method provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown. Figure 1 This diagram illustrates the application environment of the PET image reconstruction method in this embodiment. Terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on another network server. Specifically, after the medical data acquisition device acquires anatomical structure image data and PET data of the object being tested, it can send the anatomical structure image data and PET data to server 104 via a network. Server 104 can analyze and process the attenuated image data and PET data to obtain a reconstructed PET image, which is then sent to terminal 102. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and other smart display devices. Server 104 can be implemented using a standalone server or a server cluster composed of multiple servers.

[0033] Specifically, in the process of single-site scanning using positron emission tomography (PET), medical data acquisition equipment requires attenuation correction (AC) of PET images to meet clinical needs. Since the attenuated images required for attenuation correction are provided by anatomical structure images from other modal imaging devices, and their scanning range is smaller than the required imaging range of the PET image, deviations occur during the correction process, affecting the image correction effect. If other modal devices are used to scan the object according to the required imaging range of the PET image, it will result in excessively high doses to the patient or excessively long imaging times, which is detrimental to clinical imaging. Therefore, data expansion of anatomical structure images is a necessary technical means to meet the needs of PET image correction.

[0034] In one embodiment, such as Figure 2 As shown, a method for reconstructing PET images using extended attenuation images is provided. This embodiment illustrates the application of this method to medical imaging equipment, which includes, but is not limited to, PET systems, or multimodal combined systems such as PET-CT and PET-MR. The following description uses a PET-CT system as an example, but this method can be applied to single-modal and multimodal PET imaging systems. It is understood that this method can also be applied to... Figure 1 The terminal or server shown can also be applied to a system including both a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0035] Step 210: Obtain anatomical images and PET data of the scanned object.

[0036] In some embodiments, the anatomical structure image can be a structural image generated from CT or MRI scans that reflects the intensity attenuation of rays or signals as they pass through human tissue. Alternatively, it can be a three-dimensional image showing the internal structure, organs, and tissues of the scanned object. The anatomical structure image can be obtained by converting between a structural image and a three-dimensional image. The anatomical structure image contains anatomical information, which can ensure the accuracy of the distribution of radioactive tracers in PET images, ensure the accuracy of distinguishing lesions, reduce artifacts, and improve image quality.

[0037] In some embodiments, the anatomical images are obtained by CT or MR scans. The axial length of the anatomical images is less than the axial range of the corresponding PET data image, and the axial range direction is the bed entry direction. The anatomical images only cover a portion of the scanned object.

[0038] In some embodiments, the anatomical images can also be anatomical images generated by MLAA (Maximum Likelihood Attenuation and Activity). Specifically, PET data can be represented as a joint function of the radioactivity distribution and the attenuation distribution. The MLAA method estimates both the anatomical images and the activity images by establishing this joint model. MLAA uses the maximum likelihood estimation method and an iterative optimization algorithm to find the anatomical images and activity images that best match the PET measurement data.

[0039] In some embodiments, the PET data of the scanned object refers to PET data that can be used for imaging, obtained by scanning the object using a PET system, or a multimodal combined system such as PET-CT or PET-MR. The scanned object includes the human body and its various parts and tissue structures, such as the head, soft tissue, blood vessels, bones and joints, etc.; the scanned object can also be an animal, a phantom, etc.

[0040] PET data refers to data generated by positron emission tomography (PET), which is used for medical imaging to show physiological and metabolic activities in the human body. It generates images by injecting radioactive tracers to detect the gamma rays produced by the annihilation of positrons and electrons during their decay.

[0041] In this application, the imaging range of the anatomical structure image covers a portion of the scanned object, and the axial range of the anatomical structure image is smaller than the axial range of the corresponding PET data image. During AFOV imaging, the anatomical structure image scan only needs to scan a portion of the required imaging range, which reduces both the dose and scanning time. However, the actual amount of PET data obtained will exceed the imaging range, so the actual PET image range obtained using the PET data is larger than that of the anatomical structure image.

[0042] Step 220: Determine the extended attenuation image based on the anatomical structure image.

[0043] In some instances, the axial range of the extended attenuation image is greater than that of the anatomical structure image. During attenuation correction, the anatomical structure image data and the PET image data need to correspond in range. Therefore, when the imaging range of the PET image is larger than that of the anatomical structure image, the anatomical structure image needs to be extended to ensure that the data of the anatomical structure image and the PET image match. Thus, the axial range of the extended attenuation image is greater than or equal to the axial range of the PET image; similarly, the axial range of the extended anatomical structure image is greater than or equal to the axial range of the anatomical structure image.

[0044] In some embodiments, such as Figure 9The AFOV reconstruction and correction method shown involves partial AFOV reconstruction (the region of interest highlighted in red in the figure needs to be emphasized). The anatomical structure image only shows a portion of the area. According to PET imaging principles, theoretically, full FOV PET data can be obtained after drug injection. Therefore, to ensure effective use of PET data and the accuracy of the partially AFOV reconstructed PET image, the anatomical structure image needs to be expanded. This expansion attenuation image range should be greater than or equal to the required imaging range of the PET image, ensuring data correspondence between the expanded attenuation image and the PET image, thus improving the utilization rate of PET data.

[0045] In some embodiments, the axial range of the extended attenuation image can be extended to the entire range of the scanned object; for example, an anatomical image of a partial scanned object obtained in a partial scan (such as a head and chest scan) is extended to the whole body of the scanned object, and then the extended anatomical image is converted into an extended attenuation image.

[0046] The extended attenuation image can be obtained by interpolation or extrapolation on the attenuation image obtained from the anatomical structure image. The extended attenuation image is obtained by predicting the extension from the anatomical structure image, and the axial range of the extended attenuation image is larger than the axial range of the anatomical structure image. The anatomical structure image can be obtained during the current scan or a historical image from the patient's recent scans.

[0047] The extent of the extended attenuation image is determined by the PET image scanning protocol. The axial scanning field of view of the extended attenuation image is greater than or equal to the axial scanning field of view of the PET image to be corrected. The extent of the extension equals the axial scanning field of view of the extended attenuation image minus the axial scanning field of view of the anatomical structure image. The axial scanning field of view can be expressed in length, which can be the length range from head to toe or from chest to abdomen; the axial scanning field of view can also be expressed in width, which is the width range from left to right. The direction of the extension is the bed entry direction (Z-axis). It can extend only one side of the anatomical structure image along the Z-axis or extend both sides; it can also extend the width range to one or both sides; or it can extend both the length and width ranges simultaneously.

[0048] In some embodiments, the anatomical structure image is used as a reference to extend outwards by a certain range (e.g., ±10cm) to ensure that the anatomical structure image covers the entire area of ​​the PET image corresponding to the PET data.

[0049] In some embodiments, for AFOV local scans (such as the brain or heart), the extent of coverage may be appropriately reduced; for PET whole-body scans, the extent of coverage needs to cover the entire body.

[0050] Step 230: Reconstruct the PET data based on the extended attenuation image to obtain a PET image of the scanned object.

[0051] In this embodiment, the MLEM algorithm is used to reconstruct the PET data, with 5 iterations. During the reconstruction process, an extended attenuation image is used to correct the attenuation of the PET data. Since the axial range of the extended attenuation image matches the axial range of the PET data, accurate attenuation correction can be performed on all PET data, thereby obtaining high-quality PET images.

[0052] In this process, the pixel values ​​in the extended attenuation image are converted into attenuation coefficients. These attenuation coefficients are used to compensate for the signal loss of γ photons when they pass through the tissue during PET imaging, thereby improving the quantitative accuracy and signal-to-noise ratio of PET images and ensuring the accurate reflection of functional metabolic information.

[0053] Attenuation correction is performed on each Line of Response (LOR) using extended attenuation images to compensate for signal loss as gamma photons pass through tissue. The LOR is a virtual straight line connecting two detectors, representing the propagation path of a pair of gamma photons from the annihilation event point to the detector. In attenuation correction, the LOR is used to calculate the attenuation path of gamma photons as they pass through human tissue. Anatomical images generated from CT or MR scans can be used to determine the attenuation coefficient at each point on the LOR, which is then integrated and summed to correct for signal loss.

[0054] Specifically, in LOR attenuation correction, the attenuation factor is used to calculate the total attenuation on each LOR. The specific steps are as follows:

[0055] 1. Determine the LOR path. For each LOR (a straight line connecting two detectors), determine its path in the extended attenuation image. Each point on the LOR path corresponds to a pixel value (i.e., the attenuation coefficient μ value) in the extended attenuation image.

[0056] 2. Calculate the total attenuation. Integrate the attenuation coefficient along the LOR path to calculate the total attenuation of γ photons as they pass through the tissue. The formula is: Total attenuation = ∫μ(x)dx.

[0057] Where μ(x) is the attenuation coefficient at a point on the LOR path, and the integral represents the total attenuation along the LOR.

[0058] 3. Calculate the attenuation factor. The attenuation factor (AF) is calculated based on the total attenuation. The attenuation factor represents the degree of signal attenuation after gamma photons pass through tissue. The formula is as follows: AF = e -∫μ(x)dx .

[0059] The attenuation factor AF is a value between 0 and 1. The smaller the AF, the more severe the attenuation.

[0060] 4. Correct the coincidence event count. Divide the coincidence event count on each LOR by the corresponding attenuation factor to obtain the corrected count. The formula is:

[0061] Where, N measured N is the raw count measured by the detector. corrected This is the corrected count.

[0062] In the above correction method, the PET image is corrected by using an expanded attenuated image, which compensates for the data mismatch problem of using anatomical structure images for PET correction, avoids repeated scanning, saves time, reduces dose damage to the human body, and improves the accuracy of PET image correction. The reconstruction method can be analytical reconstruction or iterative reconstruction. The axial range of the PET image obtained after reconstruction is greater than or equal to the axial range of the anatomical structure; or, the axial range of the PET image covers the entire scanned object.

[0063] In some embodiments, such as Figure 3 As shown in step 310, anatomical images, PET data, and contour images of the scanned object are acquired.

[0064] Contour images are images that reveal the imaging area and structural contours of the scanned object. They can be localization images from CT, MR, or PET scans, including frontal and / or lateral views. Localization images are obtained through a localization scan before the imaging scan.

[0065] In some embodiments, the contour image may also be a camera image containing spatial structure obtained by a depth camera, infrared camera, lidar, structured light scanner, or stereo vision camera.

[0066] In some embodiments, the contour image may simultaneously include a positioning image and / or a camera image.

[0067] The axial range of the anatomical structure image is smaller than that of the contour image, while the axial range of the extended attenuation image is greater than or equal to that of the contour image.

[0068] Step 320: Based on the anatomical structure image and the contour image of the scanned object, determine the extended attenuation image using the first data extension model;

[0069] like Figure 10The specific embodiment shown involves inputting CT images, anteroposterior localization images, and lateral localization images into a machine learning model, outputting a predicted attenuation image. This predicted attenuation image is an image with anatomical structures and an expanded field of view. After cropping the predicted attenuation image, attenuation correction is performed on a PET image to obtain a PET image. Utilizing the fact that the scanning field of view of the localization image is larger than that of the CT image, the anteroposterior and lateral localization images, obtained by scanning the human body from two different angles (front and side), can display the human body contour and approximate internal structures, used for preliminary localization and understanding of human morphology. The predicted attenuation image obtained using localization images from two different orientations and CT images is more accurate.

[0070] like Figure 11 The specific embodiment shown utilizes CT images, anteroposterior and lateral localization images, and images from an external camera. A machine learning model is used to obtain a predicted attenuation image, which is then cropped and reconstructed using PET. The external camera can provide further image information, offering more reference information to the machine learning model, improving the accuracy of the predicted attenuation image, and laying a good foundation for subsequent image analysis and diagnosis.

[0071] like Figure 13 The specific embodiment demonstrated utilizes anteroposterior localization images, CT images, and MLTR attenuation images generated by MLAA (Multi-Layer Assessment and Analysis) to input into a machine learning model to generate predicted attenuation images. These predicted attenuation images are then cropped and reconstructed using PET scans. MLTR attenuation images are the raw data recorded by the detector during a PET scan, containing information such as the time and location of photon events. The MLTR attenuation images obtained after MLAA reconstruction reflect the attenuation characteristics of different parts of the human body to radiation. By leveraging attenuation images generated from PET data, CT images from CT scans, and predicted attenuation images from anteroposterior localization images, the advantages of different types of image data are fully utilized, enabling the model to acquire more comprehensive features and improving the understanding of the internal structure of the human body.

[0072] In this embodiment, the model type can be a Feedforward Neural Network (FNN), a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), a Transformer, a Generative Adversarial Network (GAN), a Graph Neural Network (GNN), a deep reinforcement learning model, etc. The model is trained using supervised learning methods, including gradient descent, stochastic gradient descent (SGD), batch gradient descent, etc.

[0073] Specifically, taking gradient descent as an example, the label data used for training employs anatomical structure images generated from traditional CT scans. The contour images and anatomical structure images are input into the network together. The model outputs predicted anatomical structure images, and the loss function is adjusted to update the network parameters based on the error between the predicted attenuated images and the labels, until the model training is complete, resulting in the first data-expanded model. By training the network model together with contour and anatomical structure images, the contour images provide spatial information for model training, enriching the model structure.

[0074] By inputting both the contour image and the anatomical structure image into the first data expansion model, the resulting expanded attenuation image becomes more accurate.

[0075] The first type of data expansion model can be: Feedforward Neural Network (FNN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Transformer, Generative Adversarial Network (GAN), Graph Neural Network (GNN), deep reinforcement learning models, etc. Supervised learning methods are used to train the model, including gradient descent, stochastic gradient descent (SGD), batch gradient descent, etc.

[0076] Specifically, taking gradient descent as an example, the label data used for training adopts anatomical structure images generated by traditional CT as labels. The imaging range of the anatomical structure image input to the model is smaller than the imaging range corresponding to the attenuation image used as the label. The model outputs the predicted attenuation image, and the loss function is adjusted to update the network parameters based on the error between the predicted attenuation image and the label until the model training is completed, thus obtaining the first data expansion model.

[0077] By inputting anatomical structure images into the first data expansion model, expanded attenuated images can be obtained directly without adjusting the imaging range and rescanning, thus reducing the scan dose for patients.

[0078] Steps 330 and 230 are the same, and will not be repeated here.

[0079] In some embodiments, such as Figure 4 The flowchart shown is for the method of reconstructing PET images using extended attenuation images. Steps 410 and 210 are the same and will not be described again here.

[0080] Step 420: Determine preliminary anatomical structure images based on PET data;

[0081] Preliminary attenuation images can be generated from MLAA (Maximum Likelihood Attenuation Map) reconstruction of PET data to produce MLTR (Maximum Likelihood Attenuation Map) anatomical images. MLTR anatomical images reflect the degree of attenuation of gamma rays in different tissues, usually expressed as a linear attenuation coefficient. This method can directly generate anatomical images from PET data, avoiding the additional radiation from CT / MR.

[0082] Step 430: Based on the preliminary anatomical structure and the contour image of the scanned object, determine the first predicted attenuation image using the second data expansion model. For example... Figure 12 This demonstrates how attenuation images generated from orthogonal localization images and MLAA are input into a network model to produce predicted attenuation images. The predicted attenuation images and CT attenuation images are then stitched together to obtain an extended attenuation image.

[0083] In this embodiment, the model type can be a Feedforward Neural Network (FNN), a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), a Transformer, a Generative Adversarial Network (GAN), a Graph Neural Network (GNN), a deep reinforcement learning model, etc. The model is trained using supervised learning methods, including gradient descent, stochastic gradient descent (SGD), batch gradient descent, etc.

[0084] The contour image and the preliminary attenuation image are fed together into the second data expansion model to determine the first predicted attenuation image.

[0085] Step 440: Merge and stitch the anatomical structure image with the first predicted attenuation image to determine the extended attenuation image.

[0086] The merging and stitching method uses weighted fusion to fuse the overlapping regions of the anatomical structure image and the first predicted attenuation image.

[0087] Steps 450 and 230 are the same, and will not be repeated here.

[0088] In some embodiments, such as Figure 5The flowchart shown is for the method of reconstructing PET images using extended attenuation images. Steps 510 and 310 are the same, and steps 520 and 420 are the same, so they will not be described again here.

[0089] Step 530: Based on the anatomical structure image, the contour image of the scanned object, and the preliminary anatomical structure image, determine the extended attenuation image using the third data extension model;

[0090] In this embodiment, the model type can be a Feedforward Neural Network (FNN), a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), a Transformer, a Generative Adversarial Network (GAN), a Graph Neural Network (GNN), a deep reinforcement learning model, etc. The model is trained using supervised learning methods, including gradient descent, stochastic gradient descent (SGD), batch gradient descent, etc.

[0091] The contour image, anatomical structure image, and preliminary attenuation image are input together into a third data expansion model to determine the expanded attenuation image.

[0092] Steps 540 and 230 are the same, and will not be repeated here.

[0093] In some embodiments, such as Figure 6 The flowchart shown is for the method of reconstructing PET images using extended attenuation images. Steps 610 and 210 are the same and will not be described again here.

[0094] As in step 620, a second predicted attenuation image is determined based on the anatomical structure image;

[0095] The optional second predicted attenuation image may be the extended attenuation image obtained in steps 320, 440, and 530.

[0096] Step 630: Crop the second predicted attenuation image to obtain an extended attenuation image;

[0097] To increase the data used for AFOV reconstruction, improve its sensitivity, and reduce noise, while ensuring the accuracy of the attenuation image affects the quantitative reconstruction, the extended attenuation image will be cropped and truncated to ensure that the centers of the attenuation image, the PET image corresponding to the PET data, and the extended attenuation image are consistent.

[0098] In some embodiments, the specific truncation method is as follows: after confirming the expanded attenuated image, the maximum expansion length is determined by the Mean-square-error (MSE) of the VOI of the axial edge of the reconstructed image, and the relevant formula is as follows:

[0099]

[0100] Where N is the number of pixels in the VOI, y is the ROI (Region of Interest) in the image, and i is each element in the ROI. For axial edges, on the one hand, attenuating image expansion can effectively reduce image noise and lower the MSE value; however, since the expanded part is output from the network and has some error compared to the standard attenuated image, it introduces some reconstruction image error, increasing the MSE value. The final expansion length should correspond to the case with the minimum MSE value.

[0101] Optionally, the expanded attenuation image can be obtained by directly generating the truncated expanded attenuation image using either the first or second data expansion model. Specifically, this involves adding an expansion length to the loss function of the first or second data expansion model; for example, explicitly adding the loss of the MSE value to the network's loss function. Simultaneously, during model training, the expansion length is used as a training parameter, and the optimal expanded image is iteratively output based on this fused loss. The truncated expanded attenuation image is then directly output through either the first or second data expansion model.

[0102] Steps 640 and 230 are the same, and will not be repeated here.

[0103] It should be understood that, although Figure 1-6 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order requirement for the execution of these steps; they can be executed in other orders. Furthermore, Figure 1-6 At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0104] In one embodiment, such as Figure 7 As shown, an image reconstruction apparatus is provided, comprising: an acquisition module, a determination module, a correction module, and a reconstruction module, wherein:

[0105] Image acquisition module 710 is used to acquire anatomical images and PET data of the scanned object;

[0106] In some embodiments, the anatomical structure image can be a structural image generated from CT, MR, or MRI scans that reflects the intensity attenuation of rays or signals as they pass through human tissue. Alternatively, it can be a three-dimensional image showing the internal structure, organs, and tissues of the scanned object. The anatomical structure image can be obtained by converting between a structural image and a three-dimensional image. The anatomical structure image contains anatomical information, which can ensure the accuracy of the distribution of radioactive tracers in PET images, ensure the accuracy of lesion tissue differentiation, reduce artifacts, and improve image quality.

[0107] In some embodiments, the anatomical images are obtained by CT or MR scans. The axial length of the anatomical images is less than the axial range of the corresponding PET data image, and the axial range direction is the bed entry direction. The anatomical images only cover a portion of the scanned object.

[0108] In some embodiments, the anatomical images may also be anatomical images generated by MLAA (Maximum Likelihood Attenuation and Activity).

[0109] In some embodiments, the image acquisition module 810 is further configured to acquire a contour image by performing a positioning scan using a medical imaging device, such as CT, PET, MR, or XR. A contour image is an image that reflects the imaging area and structural contour of the scanned object.

[0110] The imaging range of anatomical structure images covers a portion of the scanned object, and the axial range of the anatomical structure images is smaller than the axial range of the corresponding PET data images. In this application, during AFOV imaging, the anatomical structure image scanning only needs to scan a portion of the required imaging range, which reduces the dose and scanning time. However, the actual amount of PET data obtained is greater than the imaging range, so the actual PET image range corresponding to the PET data is larger than that of the anatomical structure images.

[0111] The image determination module 720 is used to perform the above method steps 220, 320, 440, 530, and 630 to determine the extended anatomical structure image;

[0112] In some instances, the axial range of the extended attenuation image is larger than that of the anatomical structure image. During attenuation correction, the attenuation image data and PET image data need to correspond in range. Therefore, when the imaging area of ​​the PET image is larger than that of the anatomical structure image, the anatomical structure image needs to be extended to ensure that the attenuation image data corresponding to the anatomical structure image matches the data of the PET image. Thus, the axial range of the extended attenuation image is greater than or equal to the axial range of the PET image; similarly, the axial range of the extended attenuation image is greater than or equal to the axial range of the anatomical structure image.

[0113] The extended attenuation image can be obtained by interpolation or extrapolation on the attenuation image corresponding to the anatomical structure image. The extended attenuation image is obtained by predicting the extension from the anatomical structure image. The axial range of the extended attenuation image is larger than that of the anatomical structure image. The anatomical structure image can be obtained during the current scan or from the patient's recent scan history.

[0114] The image reconstruction module 730 reconstructs the PET data based on the extended attenuation image to obtain a PET image of the scanned object.

[0115] This is used to correct PET data based on the extended attenuation image. Pixel values ​​in the extended attenuation image are converted into attenuation coefficients, which are used to compensate for signal loss when gamma photons pass through tissue during PET imaging, thereby improving the quantitative accuracy and signal-to-noise ratio of PET images and ensuring accurate reflection of functional metabolic information. Attenuation correction is performed on each Line of Response (LOR) using the extended attenuation image to compensate for signal loss when gamma photons pass through tissue. The LOR is a virtual straight line connecting two detectors, representing the propagation path of a pair of gamma photons from the annihilation event point to the detector. In attenuation correction, the LOR is used to calculate the attenuation path of gamma photons as they pass through human tissue. The attenuation coefficient at each point on the LOR can be determined by the attenuation corresponding to the anatomical structure image generated by CT or MR, thereby correcting for signal loss.

[0116] The reconstruction method can be analytical reconstruction or iterative reconstruction, and the axial range of the PET image obtained after reconstruction is greater than or equal to the axial range of the anatomical structure image; or, the axial range of the PET image covers the entire scanned object.

[0117] Specific limitations regarding the PET image reconstruction apparatus can be found in the limitations of the PET image reconstruction method described above, and will not be repeated here. Each module in the aforementioned PET image reconstruction apparatus can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0118] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores image data. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a PET image reconstruction method.

[0119] Those skilled in the art will understand that Figure 8 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement any of the PET image reconstruction methods described in the above embodiments.

[0120] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements any of the PET image reconstruction methods described above.

[0121] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0122] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0123] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A PET image reconstruction method, characterized in that, The method includes: Acquire anatomical structure images and PET data of the scanned object, wherein the anatomical structure images cover a portion of the scanned object, and the axial range of the anatomical structure images is smaller than the corresponding axial range of the PET data; Based on the anatomical structure image, an extended attenuation image is determined, wherein the axial range of the extended attenuation image is greater than the axial range of the anatomical structure image; The PET data is reconstructed based on the extended attenuation image to obtain a PET image of the scanned object.

2. The method according to claim 1, characterized in that, Determining the extended attenuation image based on the anatomical structure image includes: Based on the anatomical structure image, the expanded attenuation image is determined using a first data expansion model, wherein the first data expansion model is a trained machine learning model.

3. The method according to claim 2, characterized in that, The step of determining the expanded attenuation image based on the anatomical structure image using the first data expansion model includes: Obtain the contour image of the scanned object; Based on the anatomical structure image and the contour image of the scanned object, the extended attenuation image is determined using a first data extension model, and the contour image is capable of displaying the contour region of the scanned object.

4. The method according to claim 1 or 3, characterized in that, The method further includes: Preliminary anatomical images were determined based on the PET data; Based on the preliminary anatomical structure image and the contour image of the scanned object, a first predicted attenuation image is determined using a second data expansion model; The anatomical structure image is merged and stitched with the first predicted attenuation image to determine the extended attenuation image.

5. The method according to claim 1, characterized in that, The method further includes: Obtain the outline image of the scanned object; Preliminary anatomical images were determined based on the PET data; The extended attenuation image is determined using a third data extension model based on the anatomical structure image, the contour image of the scanned object, and the preliminary anatomical structure image.

6. The method according to claim 1, characterized in that, Determining the extended attenuation image based on the anatomical structure image includes: A second predicted attenuation image is determined based on the anatomical structure image; The second predicted attenuation image is cropped to obtain the extended attenuation image.

7. The method according to claim 6, characterized in that, The step of cropping the second predicted attenuation image to obtain the extended attenuation image includes: The truncated range of the second predicted attenuation image is determined according to the scanning protocol of the PET data and / or the image center of the anatomical structure image, and the second predicted attenuation image is truncated according to the truncated range to obtain the extended attenuation image.

8. A PET image reconstruction method, characterized in that, The method includes: Acquire CT images of the object being scanned; Acquire PET data of the scanned object; the first axial range corresponding to the CT image is smaller than the second axial range corresponding to the PET data; The CT image is input into a trained neural network model to obtain a predicted attenuation map, wherein the third axial range corresponding to the predicted attenuation map is greater than the first axial range. The PET data is reconstructed based on the predicted attenuation map to obtain a PET image of the scanned object.

9. The method according to claim 8, characterized in that, The method of reconstructing the PET data using the predicted attenuation map to obtain a PET image of the scanned object includes: obtaining an attenuation map of the CT image of the scanned object; The predicted image and the attenuated image of the CT image are merged to obtain the target attenuated image; The predicted image and the attenuation image of the CT image are merged to obtain the target attenuation image; The PET data is reconstructed based on the target attenuation image to obtain a PET image of the scanned object.

10. A PET-CT system, characterized in that, include: PET scanner; The CT scanner is integrated with the PET scanner. A computer operatively coupled to the CT scanner and the PET scanner, the computer being configured to: The CT scanner is controlled to perform a scan on the region of interest of the object to obtain CT images; The PET scanner is controlled to perform a scan on the region of interest to obtain PET data, and the axial field of view of the PET scanner is larger than that of the CT scanner. The CT image is input into a trained neural network model to obtain a predicted attenuation map, which includes an attenuation estimate of at least one peripheral organ located outside the axial field of view of the CT scanner and inside the axial field of view of the PET scanner. The PET data is iteratively reconstructed based on the predicted attenuation map to obtain a PET image of the scanned object.

Images