Reusable whole body PET attenuation correction method based on attenuation map prediction

Abstract

The invention discloses a reusable whole-body PET (positron emission tomography) attenuation correction method based on attenuation map prediction. The method comprises the following steps: performing PET scanning on a target to obtain an uncorrected PET image; inputting the uncorrected PET image into a trained deep learning network to generate a corresponding attenuation correction coefficient graph; and correcting the uncorrected PET image by using the generated attenuation correction coefficient graph to obtain a corrected PET image. Wherein in the process of training the deep learning network, an attenuation correction coefficient graph obtained by performing first scanning on a target is used as prior information, and in combination with shifted position information provided by an uncorrected PET image obtained by second scanning, an attenuation correction coefficient graph after second scanning shifting is generated. According to the method, the anatomical precision of the attenuation correction coefficient graph is improved, and the details of the physical process in PET imaging are better reserved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and more specifically, to a reusable whole-body PET attenuation correction method based on attenuation map prediction. Background Art

[0002] Positron emission tomography (PET) is an advanced nuclear medicine imaging technology that can reveal the distribution of radioactive tracers in a patient's body. During the imaging process, the positrons released by the decay of radioactive isotopes annihilate with the negative electrons in the body within a range of 1-3 mm, interact with each other, and generate a pair of γ photons propagating in opposite directions. The PET scanner detects these photons, obtains projection data, and generates an image showing the metabolic level of tissues in the body through a reconstruction process. However, these images have significant noise and attenuation artifacts, and the spatial resolution is relatively low, making them unable to be directly applied to clinical diagnosis. Therefore, performing attenuation correction is a key step to ensure the accurate quantification of PET images. In a PET / CT combined scanning system, since computed tomography (CT) can provide accurate anatomical structure information and has good contrast, it is often used to generate an attenuation correction coefficient map (μ-map) to correct the attenuation artifacts in the PET image.

[0003] During CT scanning, patients are inevitably exposed to X-ray radiation, which has raised widespread concerns about radiation hazards. Especially with the advent of total-body PET / CT scanners, compared with traditional whole-body PET / CT, the amount of radioactive drugs used is reduced, but this also makes the radiation burden of the CT part more significant. Although attenuation correction in PET imaging usually uses ultra-low-dose attenuation correction CT (ACCT, 2.1 mSv), low-dose CT still contributes a significant amount of radiation, especially for patients who need to perform delayed imaging. Such patients are required to perform multiple PET / CT imaging within a short period of time, resulting in a large radiation dose. Therefore, it is particularly important to research and develop an attenuation correction method that can be repeatedly applied relying on a single CT scan. This method can not only ensure the imaging quality of attenuation correction but also effectively reduce radiation exposure, having important scientific value and broad application prospects, especially in the field of medical diagnosis.

[0004] In the prior art, the article "Learning CT-free attenuation-corrected total-body PET images through deep learning" published by Li et al. in European Radiology in 2024. This method uses a mature Cycle-GAN network to find the mapping relationship between unattenuated-corrected whole-body PET images and attenuated-corrected whole-body PET images (CTF-AC), and introduces two generator networks with the same encoder-decoder structure, as well as two corresponding discriminator networks, and introduces four different types of loss functions to constrain the training of the network. In addition, this method takes into account both the differences and similarities in the anatomical structures of human body parts, introduces part differences as a priori information into the network design, and directly generates attenuated-corrected PET images to achieve CT-free whole-body PET image attenuation correction.

[0005] After analysis, the prior art mainly has the following defects.

[0006] 1) During the PET / CT imaging process, patients need to receive radiation from both PET and CT simultaneously. Especially during whole-body scanning, the radiation dose of the CT part increases significantly. Although low-dose CT technology is adopted, the overall radiation burden is still relatively high, and long-term accumulation may pose potential hazards to patients' health.

[0007] 2) The combined PET / CT scan needs to complete the imaging processes of PET and CT separately, resulting in an extended overall scanning time. This not only increases patients' discomfort but also reduces the utilization efficiency of the scanning equipment and affects the timeliness of clinical diagnosis. In addition, multiple CT scans will further increase the medical cost, bringing economic pressure to both patients and the medical system.

[0008] 3) Currently, only deep learning technology is used to construct an end-to-end network to directly synthesize attenuated-corrected PET images. However, this type of method cannot fully consider the physical attenuation process in PET imaging, resulting in deviations in the physical accuracy of the generated CTF-AC images, especially in complex anatomical structures or abnormal lesion areas.

[0009] 4) After generating the μ-map through deep learning, using a proven traditional attenuation correction method for correction can make full use of the accuracy and reliability of traditional methods based on physics. In contrast, the deep learning method of directly generating AC PET images may lack sufficient physical constraints, resulting in insufficient quantitative accuracy. Summary of the Invention

[0010] The object of the present invention is to overcome the defects of the above-mentioned prior art and provide a reusable whole-body PET attenuation correction method based on attenuation map prediction. The method comprises the following steps:

[0011] Perform a PET scan on the target to obtain an uncorrected PET image;

[0012] Input the uncorrected PET image into a trained deep learning network to generate a corresponding attenuation correction coefficient map;

[0013] Use the generated attenuation correction coefficient map to correct the uncorrected PET image to obtain a corrected PET image;

[0014] Wherein, during the process of training the deep learning network, the attenuation correction coefficient map obtained by the first scan of the target is used as prior information, and combined with the shifted body position information provided by the uncorrected PET image obtained by the second scan, a shifted attenuation correction coefficient map for the second scan is generated.

[0015] Compared with the prior art, the advantages of the present invention are that for patients undergoing delayed imaging scans, an advanced deep learning network is used to find the mapping relationship between the unattenuated corrected (NAC) PET image and the corresponding attenuation correction coefficient map (μ map) in the delayed scan stage. The μ map in the initial scan stage is introduced as prior information, and the attenuation correction in the delayed scan stage is achieved through the generated μ map, thereby reducing the number of CT scans.

[0016] Through the following detailed description of the exemplary embodiments of the present invention with reference to the accompanying drawings, other features and advantages of the present invention will become clear. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings incorporated in and constituting a part of this specification illustrate embodiments of the present invention and, together with the description, serve to explain the principles of the present invention.

[0018] Figure 1 is a flowchart of a reusable whole-body PET attenuation correction method based on attenuation map prediction according to an embodiment of the present invention;

[0019] Figure 2 is a schematic diagram of the overall process of delayed imaging scan according to an embodiment of the present invention;

[0020] Figure 3 is a schematic diagram of a reusable attenuation correction process based on μ map prediction according to an embodiment of the present invention;

[0021] Figure 4 is a schematic diagram of a reusable attenuation correction network framework based on μ map prediction according to an embodiment of the present invention;

[0022] Figure 5 It is a schematic diagram of a reference standard μ-map image according to an embodiment of the present invention;

[0023] Figure 6 It is a schematic diagram of a generated μ-map image according to an embodiment of the present invention. Detailed implementation manners

[0024] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that: Unless otherwise specifically stated, the relative arrangements, numerical expressions, and numerical values of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0025] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way a limitation on the present invention or its application or use.

[0026] Technologies, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the said technologies, methods, and devices should be regarded as part of the specification.

[0027] In all the examples shown and discussed herein, any specific values should be construed as merely exemplary and not as a limitation. Thus, other examples of the exemplary embodiments may have different values.

[0028] It should be noted that: Similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0029] Regarding the CT radiation problem faced by patients who need to perform delayed imaging or multi-tracer imaging and thus require multiple PET / CT scans, considering the lack of anatomical structure information in NAC PET imaging, the present invention introduces the μ-map obtained from the first-stage PET / CT scan as anatomical prior information to provide a large amount of reference information for the prediction of the second-stage μ-map. Furthermore, the first-stage μ-map is combined with the second-stage NAC PET and incorporated into a deep learning network to directly predict the second-stage whole-body μ-map, and further combine the second-stage NAC PET to achieve attenuation correction. The second-stage μ-image noise finally obtained by the present invention can be well suppressed, the image contrast is obvious, and the tissue structure is well preserved. Hereinafter, taking the attenuation correction of the whole-body PET image as an example for illustration.

[0030] See Figure 1 As shown, the provided reusable whole-body PET attenuation correction method based on attenuation map prediction includes the following steps:

[0031] Step S1: Obtain a dataset that reflects the μ-map in the initial scan stage and the correspondence between the unattenuated-corrected PET image in the delayed scan stage and the μ-map in the delayed scan stage.

[0032] Figure 2 The overall process of delayed imaging scan is shown, including two scans, namely the initial scan (or the first scan) and the scan performed 2 hours later (or the second scan). Both scans include CT scan and PET scan. The CT scan generates a μ-map for correcting the corresponding NAC (unattenuated-corrected) PET image, and then generates an attenuation-corrected AC PET image. In the delayed scan stage, the CT scan can be used to update the μ-map, thereby providing more accurate attenuation correction information. This step is particularly important to avoid inaccurate correction caused by registration errors when the patient's body position changes or anatomical structures change during the delay period.

[0033] To reduce the number of CT scans in the actual imaging process, the present invention uses the two scan processes to construct a dataset for subsequent deep learning network training. This dataset reflects the μ-map in the initial scan stage and the correspondence between the unattenuated-corrected PET image in the delayed scan stage and the μ-map in the delayed scan stage.

[0034] Step S1: Construct a deep learning network for predicting the μ-map.

[0035] The deep learning network can be used to generate the μ-map in the delayed scan stage (i.e., the updated μ-map). During the training process, the μ-map in the initial scan stage and the unattenuated-corrected PET image in the delayed scan stage are used as the network inputs, and the μ-map in the delayed scan stage is used as the label for network learning.

[0036] Specifically, since the purpose of updating the μ-map is to prevent registration errors caused by the patient's body position change and anatomical structure change, a deep learning network for μ-map prediction is designed. Using the μ-map of the initial scan as a prior containing high-frequency anatomical information such as bones, and combining the shifted body position information provided by the NAC PET in the delayed scan stage, the shifted μ-map in the delayed scan stage is generated. See Figure 3 the reusable attenuation correction process based on μ-map prediction shown.

[0037] The deep learning network can adopt various types of neural networks, such as convolutional neural network, generative adversarial network, etc. In one embodiment, Figure 4The network framework is composed of a U-Net framework consisting of 7 weight-normalized convolutional modules. Each module is composed of 2 weight-normalized convolutional kernels with a kernel size of 3 and a stride of 1, two Swish activation functions, and two group normalization operations (GN). The network follows the U-Net standard framework. In the encoder stage, each module is downsampled by a 2D average pooling layer with a kernel size of 2 and a stride of 2. In the decoder stage, a transposed convolutional module with a kernel size of 2 and a stride of 2 is used for upsampling and restoration operations, and skip connections are made between the encoder and the decoder. In addition, a pair of numbers in each module respectively represent the number of input channels and output channels of the module. The number of channels ranges from 2 to 512 and then back to 64. Finally, after a standard 2D convolution with a kernel size of 1 and a stride of 1, a single-channel synthesized μ map, that is, the updated μ map, is output.

[0038] Step S3: Using the obtained dataset, train the deep learning network based on the set loss function to learn the mapping relationship between the non-attenuation-corrected PET image and the μ map in the same scanning stage.

[0039] In one embodiment, the overall loss function for training the deep learning network is the sum or weighted sum of 2 loss functions of the framework. For example, the 2 loss functions are respectively the pixel-level mean squared error loss (L mse ) and the perceptual loss (L lpips ). The design goal of the loss function is to balance the global consistency of the generated μ map and the accuracy of the local perceptual features to ensure that the μ map generated in the delayed imaging stage can accurately reflect the true anatomical information of the patient.

[0040] In one embodiment, the pixel-level mean squared error loss L mse is used to measure the pixel-level difference between the generated μ map and the true μ map (μ), and is defined as:

[0041]

[0042] where N is the total number of pixels in the image; and μ i respectively represent the values of the i-th pixel in the generated and true μ maps. By minimizing L mse , the network can generate prediction results consistent with the target μ Figure 1 at the global pixel level. This loss function is especially suitable for capturing large-scale anatomical structure information to ensure that the generated μ map is overall close to the true anatomical information.

[0043] In one embodiment, the perceptual loss L lpips optimizes the visual quality and structural features of the image by comparing the differences between the generated and true μ maps in the feature space, and is defined as:

[0044]

[0045] Among them, φ l (·) represents the features extracted in the feature space of the l-th layer; H l , W l , C l are the height, width, and number of channels of the feature map respectively. Compared with the pixel-level loss, the perceptual loss can better retain high-level semantic information and local details. Especially when the patient's body position changes greatly during the delay stage, it ensures that the generated μ-map is more visually consistent with the true value of the μ-map.

[0046] During the deep learning grid training process, the μ-map extracted from the initial scan stage and the unattenuated corrected PET image in the delayed scan stage in the delayed scan dataset are used as the network input, and the μ-map in the delayed scan stage is used as the network learning label. The Adam optimizer can be used for optimization, and the cosine annealing learning rate scheduler is used for learning rate adjustment. By training the network, the mapping relationship between the unattenuated corrected whole-body PET image and the μ-map in the same scan stage can be obtained.

[0047] To verify the performance of the trained deep learning network, model verification can be carried out on the same type of delayed scan dataset until a model that meets the accuracy requirements is obtained.

[0048] Step S4, for the actually acquired unattenuated corrected PET image, use the trained deep learning network for correction.

[0049] The trained deep learning network can be applied to actual image correction. For example, the model application process includes: performing a PET scan on the target to obtain an uncorrected PET image; inputting the uncorrected PET image into the trained deep learning network to generate a corresponding μ-map; using the generated μ-map to correct the uncorrected PET image, and then obtaining a corrected PET image.

[0050] It should be understood that the present invention can be applied not only to the scenario of delayed imaging, but also to the scenario where the patient needs to perform multi-tracer imaging.

[0051] To further verify the effect of the present invention, model verification was carried out on a large number of delayed imaging datasets. The experimental results show that the image quality of the generated μ-map is significantly improved, as Figure 5 and Figure 6 shown, where Figure 5 is the reference standard μ-map image, Figure 6 is the μ-map image generated by using the present invention.

[0052] In summary, compared with the prior art, the present invention has the following advantages:

[0053] 1) For whole-body PET images, the present invention proposes a reusable PET attenuation correction method based on μ-map prediction, which can predict subsequent μ-maps based on the μ-map obtained after a single CT scan.

[0054] 2) The present invention takes into account the characteristic that NAC PET itself lacks high-frequency anatomical information such as human bones, and incorporates the prior knowledge of the high-frequency and anatomical information of the original μ-map into the network structure, providing great reference for the network in predicting the μ-map from NAC PET in the next stage, and significantly improving the image quality.

[0055] 3) Compared with the prior art, the μ-map generated by the present invention can retain high-frequency anatomical information such as bones in the initial scanning stage and fully combine the body position information in the delayed scanning stage. This combination not only improves the anatomical accuracy of the μ-map, but also ensures the physical accuracy of the generated AC PET image by subsequently combining traditional attenuation correction methods. Compared with the end-to-end method of directly generating AC PET images by Cycle-GAN, the present invention can better retain the details of the physical process in PET imaging, thus significantly improving the reliability of quantitative analysis.

[0056] The present invention may be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0057] A computer-readable storage medium may be a tangible device that can retain and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanically encoded device such as a punch card or raised structures in grooves having instructions stored thereon, and any suitable combination of the foregoing. The computer-readable storage medium as used herein is not construed as being a transitory signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0058] The computer-readable program instructions described herein can be downloaded to various computing / processing devices from a computer-readable storage medium or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0059] The computer program instructions for performing the operations of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, Python, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, by using the state information of the computer-readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field-programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer-readable program instructions to implement various aspects of the present invention.

[0060] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0061] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that the instructions, when executed by the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in one or more boxes of the flowchart and / or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that causes a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises a manufacture including instructions for implementing various aspects of the functions / acts specified in one or more boxes of the flowchart and / or block diagram.

[0062] The computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process such that the instructions executed on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in one or more boxes of the flowchart and / or block diagram.

[0063] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram may represent a module, a segment of code, or a portion of an instruction, and the module, segment of code, or portion of an instruction may include one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions noted in the boxes may occur out of the order noted in the figures. For example, two consecutive boxes may in fact be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each box in the block diagrams and / or flowcharts, and combinations of boxes in the block diagrams and / or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or acts, or by a combination of dedicated hardware and computer instructions. As is well known to those skilled in the art, implementation by hardware, implementation by software, and implementation by a combination of software and hardware are equivalent.

[0064] The embodiments of the present invention have been described above. The above description is exemplary and not exhaustive, and is also not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, the practical application, or the improvement of the technology in the market, or to enable other ordinary skill in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A reusable whole-body PET attenuation correction method based on attenuation map prediction, comprising the following steps: Perform a PET scan on the target to obtain an uncorrected PET image; Input the uncorrected PET image into a trained deep learning network to generate a corresponding attenuation correction coefficient map; Use the generated attenuation correction coefficient map to correct the uncorrected PET image to obtain a corrected PET image; Wherein, during the training of the deep learning network, the attenuation correction coefficient map obtained by the first scan of the target is used as prior information, and combined with the shifted body position information provided by the uncorrected PET image obtained by the second scan, to generate the shifted attenuation correction coefficient map of the second scan.

2. The method according to claim 1, wherein The deep learning model is constructed based on the U-Net network, including an encoder and a decoder, and there is a skip connection between the encoder and the decoder.

3. The method according to claim 1, characterized in that, The overall loss function for training the deep learning network includes the pixel-level mean square error loss L mse and the perceptual loss L lpips , respectively set to: where N is the total number of pixels in the image; represents the value of the i-th pixel in the generated attenuation correction coefficient map, μ i represents the value of the i-th pixel in the true attenuation correction coefficient map, φ l (·) represents the feature extracted in the feature space of the l-th layer; H l , W l , C l are the height, width, and number of channels of the feature map extracted in the feature space of the L-th layer, respectively, and L represents the number of layers.

4. The method according to claim 2, wherein The deep learning network includes multiple weight-normalized convolutional modules. Each weight-normalized convolutional module consists of 2 weight-normalized convolutional kernels with a kernel size of 3 and a stride of 1, two Swish activation functions, and two group normalization operations. In the encoder stage, each weight-normalized convolutional module is downsampled through a 2D average pooling layer with a kernel size of 2 and a stride of 2. In the decoder stage, a transposed convolutional module with a kernel size of 2 and a stride of 2 is used for upsampling and restoration operations.

5. The method according to claim 4, wherein The deep learning network uses a standard 2D convolutional layer with a kernel size of 1 and a stride of 1 to output a single-channel synthetic attenuation correction coefficient map.

6. The method according to claim 1, characterized in that The uncorrected PET image is a whole-body PET image.

7. The method according to claim 1, wherein During the training of the deep learning network, the Adam optimizer is used for optimization, and the cosine annealing learning rate scheduler is used to adjust the learning rate.

8. The method according to claim 3, characterized in that The overall loss function is a weighted sum of the pixel-level mean square error loss and the perceptual loss.

9. A computer-readable storage medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer device, comprising a memory and a processor, wherein a computer program capable of running on the processor is stored on the memory, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images