Attenuation correction for medical imaging

Abstract

A frame for attenuation correction. The attenuation map may be generated by applying the non-attenuation corrected emission image to one or more trained artificial neural networks. Attenuation correction may be performed on the non-attenuation-corrected emission image by using the attenuation map.

Description

Technical Field

[0001] This disclosure relates generally to medical image data processing, and more particularly to a framework for attenuation correction. Background Technology

[0002] Since X-rays were first used to identify anatomical abnormalities, the field of medical imaging has seen significant advancements. Medical imaging hardware has evolved in the form of newer machines such as medical resonance imaging (MRI) scanners and computed tomography (CAT) scanners. Digital medical images are constructed using the raw image data obtained from such scanners. Digital medical images are typically either two-dimensional (“2-D”) images composed of pixel elements or three-dimensional (“3-D”) images composed of volume elements (“voxels”). Because a large amount of image data is generated in any given scan, there has been, and continues to be, the need to develop image processing techniques that can automate some or all of the process to determine the presence of anatomical abnormalities in the scanned medical images.

[0003] Multimodal imaging plays a crucial role in accurately identifying diseased and normal tissues. It provides combined benefits by fusing images acquired from different modalities. For example, the complementarity between anatomical (e.g., computed tomography (CT), magnetic resonance (MR)) and molecular (e.g., positron emission tomography (PET), single-photon emission computed tomography (SPECT)) imaging modalities has led to the widespread use of PET / CT and SPECT / CT imaging.

[0004] Series PET / CT (or SPECT / CT) involves a series of multiple scans (or multiple scans) to assess the state of a region of interest over a period of time. For cardiac PET / CT, two scans are typically performed in a single imaging session. The first scan is usually a "rest" scan performed while the patient is at rest; the second scan is usually a "stress" scan performed after the patient has been given a pharmacological stress agent. The typical workflow involves first acquiring a low-dose CT scan for attenuation correction. Subsequently, a radiotracer (e.g., 82Rb, 13NH3) is administered to the patient, and list-pattern PET data is acquired simultaneously. After a short wait for the radiotracer to clear, the patient is given a stress agent. Once the stress agent has increased the heart rate, a second injection is performed, along with the acquisition of "stress" list-pattern PET data. In many sites, a second CT scan is performed after the "stress" scan to provide more accurate attenuation correction for the "stress" images.

[0005] Mismatches can occur between CT and PET image data due to respiratory movements and / or spontaneous patient movement. In particular, stressors can cause cardiac movement and generate some patient discomfort. Both of these effects can contribute to mismatches between PET and CT image data. Additional problems can arise because the CT field of view (FOV) is limited by the size of the CT detector. Even with algorithms that extend CT images beyond the detector FOV, artifacts can still occur for large patients, impairing attenuation correction.

[0006] Image reconstruction is typically performed without attenuation correction. The registration procedure moves the CT image in the x, y, and / or z dimensions to produce an optimal match between the non-attenuation-corrected (NAC) image and the CT image. The NAC and CT images are displayed as overlapping images, and the operator determines whether the registration is satisfactory. If not, the procedure may allow the operator to manually move the CT image to the appropriate position. Because more movement is expected due to stress, some centers choose to perform a second CT scan after the stress scan. The same workflow is typically applied to automated registration followed by manual intervention. Summary of the Invention

[0007] This paper describes a framework for attenuation correction. Attenuation maps can be generated by applying an unattenuated emission image to one or more trained artificial neural networks. Attenuation correction can then be performed on the unattenuated emission image using the attenuation map. Attached Figure Description

[0008] A more complete understanding of the present disclosure and its many accompanying aspects will be readily obtained when considered in conjunction with the accompanying drawings and by referring to the following detailed description, as the present disclosure and its many accompanying aspects become better understood.

[0009] Figure 1 A block diagram illustrating an exemplary system is shown; Figure 2 An exemplary method for attenuation correction is shown; Figure 3 The illustration shows an exemplary application of this framework for cardiac scanning; and Figure 4 An exemplary application of this framework for a series of physiologically gated images is shown. Detailed Implementation

[0010] In the following description, numerous specific details, such as examples of specific components, devices, methods, etc., are set forth to provide a thorough understanding of how this framework is implemented. However, it will be apparent to those skilled in the art that these specific details need not be used to practice the implementation of this framework. In other instances, well-known materials or methods are not described in detail to avoid unnecessarily obscuring the implementation of this framework. While this framework allows for various modifications and alternatives, specific embodiments thereof are shown by way of example in the accompanying drawings and will be described in detail herein. However, it should be understood that there is no intention to limit the invention to the specific forms disclosed; rather, it is intended to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Furthermore, for ease of understanding, certain method steps are described as separate steps; however, these separately described steps should not be construed as necessarily depending on the order in which they are performed.

[0011] Unless otherwise stated, as will be apparent from the following discussion, terms such as “segmentation,” “generation,” “registration,” “determination,” “alignment,” “positioning,” “processing,” “computation,” “selection,” “estimation,” “detection,” “tracking,” etc., can refer to the actions and processes of a computer system or similar electronic computing device that manipulate and transform data represented as physical (e.g., electronic) quantities within the registers and memory of the computer system into other data similarly represented as physical quantities within the memory or registers of the computer system or other such information storage, transmission, or display devices. Embodiments of the methods described herein can be implemented using computer software. Sequences of instructions designed to implement these methods, if written in a programming language conforming to recognized standards, can be compiled to execute on various hardware platforms and interface with a wide variety of operating systems. Furthermore, the implementation of this framework is not described with reference to any particular programming language. It will be understood that a wide variety of programming languages ​​can be used.

[0012] For the sake of brevity, an image or a portion thereof (e.g., a region of interest (ROI) in an image) corresponding to an object (e.g., tissue, organ, tumor, etc. of a subject, such as a patient) can be referred to as an image of the object or a portion thereof (e.g., the ROI) or including the object, or the object itself. For example, an ROI corresponding to an image of the lungs or heart can be described as the ROI including the lungs or heart. As another example, an image of the chest or an image including the chest can be referred to as a chest image, or simply chest. For the sake of brevity, processing (e.g., extraction, segmentation) of a portion of an image corresponding to an object can be described as the object being processed. For example, extracting a portion of an image corresponding to the lungs from the remainder of an image can be described as the lungs being extracted.

[0013] This paper presents a framework for attenuation correction. According to one aspect, a deep artificial neural network (e.g., a convolutional neural network or CNN) is trained for attenuation correction (AC) of emission image data (e.g., PET or SPECT). In some implementations, elastic registration is performed by the trained neural network to match anatomical image data (e.g., CT image data) with NAC emission image data before performing attenuation correction (AC). In other implementations, synthetic (or pseudo) anatomical image data is generated by the trained neural network based on the NAC emission image data to be used as an AC map. Attenuation correction can then be performed on the NAC emission image data using the synthetic anatomical image data.

[0014] Both methods have the potential to improve current protocols using deep learning. They can be more robust, fully automated, and eliminate the need for CT scans following stress scans. Advantageously, they can be used to reduce dose accumulation and / or total scan time, thereby reducing associated risks. Furthermore, AC quantification errors in clinical protocols can be reduced. The efficiency of technical specialist workflows can be improved. These and other exemplary advantages and features will be described in more detail below.

[0015] Figure 1 This is a block diagram illustrating an exemplary system 100. System 100 includes a computer system 101 for implementing the framework described herein. In some implementations, computer system 101 operates as a standalone device. In other implementations, computer system 101 may (e.g., using a network) connect to other machines, such as medical imaging equipment 102 and workstation 103. In a networked deployment, computer system 101 may operate as a server (e.g., in a server-client user network environment, as a client user machine in a server-client user network environment, or as a peer machine in a peer-to-peer (or distributed) network environment).

[0016] In one implementation, computer system 101 includes a processor device or central processing unit (CPU) 104 coupled via an input-output interface 121 to one or more non-transitory computer-readable media 105 (e.g., computer storage or memory devices), a display device 108 (e.g., a monitor), and various input devices 110 (e.g., a mouse, touchpad, or keyboard). Computer system 101 may further include supporting circuitry such as caches, power supplies, clock circuits, and communication buses. Various other peripheral devices, such as additional data storage devices and printing devices, may also be connected to computer system 101.

[0017] This technology can be implemented in various forms of hardware, software, firmware, dedicated processors, or combinations thereof, as part of microinstruction code or as part of an application or software product, either or a combination thereof, which is executed via an operating system. In some implementations, the technology described herein is implemented as computer-readable program code tangibly embodied in one or more non-transitory computer-readable media 105. In particular, this technology can be implemented by a processing module 107. The non-transitory computer-readable medium 105 may include random access memory (RAM), read-only memory (ROM), floppy disk, flash memory, and other types of memory, or combinations thereof. The computer-readable program code is executed by a processor device 104 to process data provided by, for example, a medical imaging device 102. Accordingly, computer system 101 is a general-purpose computer system that becomes a dedicated computer system when executing the computer-readable program code. The computer-readable program code is not intended to be limited to any particular programming language or its implementation. It will be appreciated that the teachings of this disclosure contained herein can be implemented using a wide variety of programming languages ​​and their encodings. The same or different computer-readable media 105 may be used to store databases, including but not limited to image datasets, knowledge bases, individual subject data, medical records, subject diagnostic reports (or documents), or combinations thereof.

[0018] Medical imaging device 102 acquires image data 132. This image data 132 can be processed by processing module 107. Medical imaging device 102 can be a radiographic scanner (e.g., a nuclear medicine scanner) and / or suitable peripheral devices (e.g., a keyboard and display device) for acquiring, collecting, and / or storing such image data 132. Medical imaging device 102 can be a hybrid modality designed to acquire image data using at least one anatomical imaging modality (e.g., CT, MR) and at least one molecular imaging modality (e.g., SPECT, PET). The anatomical imaging modality focuses on extracting structural information, while the molecular imaging modality focuses on extracting functional information from molecules of interest. Medical imaging device 102 can be, for example, a PET / CT, SPECT / CT, or PET / MR scanner.

[0019] Workstation 103 may include a computer and suitable peripherals (such as a keyboard and display device) and may operate in conjunction with the overall system 100. For example, workstation 103 may communicate with medical imaging equipment 102, such that medical image data 132 can be presented or displayed at workstation 103. Workstation 103 may communicate directly with computer system 101 to display processed data and / or output results 144. Workstation 103 may include a graphical user interface to receive user input via input devices (e.g., keyboard, mouse, touchscreen, voice or video recognition interface, etc.) to manipulate the visualization and / or processing of data.

[0020] It should be further understood that, because some of the system components and method steps depicted in the accompanying figures can be implemented in software, the actual connections between system components (or process steps) can vary depending on how this framework is programmed. Given the teachings provided herein, those skilled in the art will be able to conceive of these and similar implementations or configurations of this framework.

[0021] Figure 2 An exemplary method 200 for attenuation correction is illustrated. It should be understood that the steps of method 200 may be performed in the order shown or a different order. Additional, different, or fewer steps may also be provided. Furthermore, method 200 can be used... Figure 1 The system can be implemented through 100 different systems or combinations thereof.

[0022] At 202, processing module 107 receives one or more trained artificial neural networks (ANNs). The one or more ANNs can be trained to improve and / or generate attenuation maps for attenuation correction of the emitted image. The one or more ANNs can include deep neural networks, such as convolutional neural networks (CNNs) or recurrent neural networks. The one or more ANNs can include any architecture, such as U-Net or residual block networks.

[0023] In some implementations, one or more ANNs include at least one first neural network, at least one second neural network, or a combination thereof. At least one first neural network is trained using a training set of co-registration pairs of real anatomical images (e.g., CT, MR) and non-attenuation-corrected emission images (e.g., PET, SPECT). For example, co-registration pairs of CT and PET images can be used to train at least one first neural network. The registered real anatomical images in the training set provide the underlying ground truth for training. The at least one first neural network is trained to perform elastic registration of real anatomical images with non-attenuation-corrected emission images to generate an attenuation map that matches the emission distribution.

[0024] At least one second neural network can be trained using a training set of paired unattenuated corrected emission images and linear attenuation coefficient maps (μ maps or attenuation maps) based on clinical subject data. The linear attenuation coefficient maps provide the ground truth for training and can be computed from anatomical image data (e.g., CT data). The at least one second neural network is trained to generate synthetic anatomical images based on the unattenuated corrected emission images. The synthetic anatomical images are output as attenuation maps.

[0025] At 204, processing module 107 receives a non-attenuation-corrected (NAC) emission image of a region of interest (ROI) for the subject or patient. The ROC can be any region identified for further research, such as the heart or lungs. The NAC emission image can be acquired by medical imaging device 102. Medical imaging device 102 may include a molecular imaging modality that directly acquires NAC emission images of the ROC. To generate the emission image, the molecular imaging modality can detect emissions generated by a radioactive isotope injected into the bloodstream of the subject. In some implementations, the emission image is a PET or SPECT image. Other types of molecular imaging modalities are also useful.

[0026] In some implementations, processing module 107 also receives real anatomical images of the region of interest. These real anatomical images can be acquired by the anatomical imaging modality of medical imaging device 102. The real anatomical images can be, for example, real CT or MR images. Other types of anatomical imaging modalities are also useful.

[0027] At 206, the image processing module 117 generates an attenuation map by applying at least a non-attenuation-corrected (NAC) emission image to one or more trained artificial neural networks (ANNs). As previously discussed, the one or more trained ANNs may include at least one first neural network, at least one second neural network, or a combination thereof.

[0028] In some implementations, at least one first neural network generates an attenuation map by performing elastic registration of a real anatomical image with an NAC emission image to generate a registered anatomical image, which is output as the attenuation map. Elastic registration is performed to spatially align the real anatomical image with the NAC emission image, thereby generating the registered anatomical image. In some implementations, at least one second neural network generates the attenuation map by constructing a synthetic (or pseudo) anatomical image of the region of interest based on the NAC emission image as the attenuation map output.

[0029] In some implementations, a control unit is provided before the first and second neural networks to select either the first or second neural network for generating the attenuation map. The control unit may include combinational digital circuitry, a processor, a neural network, or any other type of computational circuitry. When both the real anatomical image and the NAC emission image are provided as input to the control unit, the control unit may send or transmit the real anatomical image and the NAC emission image to at least one first neural network to generate the attenuation map. When only the NAC emission image (without the real anatomical image) is provided as input to the control unit, the control unit may send or transmit the input NAC emission image to at least one second neural network to generate the attenuation map. If the real anatomical image is provided but unavailable, the control unit may also send the input NAC emission image to at least one second neural network for attenuation map generation.

[0030] At 208, image processing module 117 performs attenuation correction on the NAC emission image using an attenuation map generated by one or more trained ANNs to generate an attenuated emission image. More specifically, a correction factor may be determined based on the attenuation map and used to correct the attenuation of the NAC emission image, thereby producing the attenuated emission image. The attenuated emission image may be displayed, for example, at workstation 103.

[0031] In some implementations, steps 204 to 208 can be repeated multiple times over time to generate a set of attenuation-corrected emission images using a single real anatomical image. For example, a real anatomical image of the region of interest can be acquired once, while NAC emission images of the region of interest can be acquired at predetermined intervals (e.g., at different respiratory stages or cardiac locations).

[0032] Figure 3 An exemplary application of this framework for cardiac scanning is illustrated. The first and second rows of images show coronal sections of the heart of the test subject, while the third and fourth rows show coronal projections of the heart of the test subject. Relatively good PET-CT image alignment exists in these images. The first row of images shows coronal sections of different μ maps used for attenuation correction. Synthetic CT images (attenuation maps or μ maps) 304 are generated by one or more trained ANNs. Image 306 is a difference image obtained by subtracting the baseline ground truth original μ map 302 from the synthetic CT image 304, shown as a percentage of the original value in μ map 302 obtained using the default AC method. Distorted CT images (attenuation maps or μ maps) 307 are generated by one or more trained ANNs. Image 308 is a difference image obtained by subtracting the baseline ground truth original μ map 302 from the distorted CT image 307.

[0033] The second row of images shows coronal sections of the corresponding reconstructed PET images with attenuation correction. PET image 310 is the base ground truth image with attenuation correction using the original μ image 302. Attenuation correction is performed on PET image 314 using the synthetic CT image 304. Image 316 is a difference image obtained by subtracting the base ground truth image 310 from PET image 314. Attenuation correction is performed on PET image 317 using the distorted CT image 307. Image 318 is a difference image obtained by subtracting the base ground truth image 310 from PET image 317.

[0034] The third row of images shows the coronal projections of the different μ maps used for attenuation correction. The synthetic CT image (attenuation map or μ map) 324 is generated by one or more trained ANNs. Image 326 is a difference image obtained by subtracting the base ground truth original μ map 320 from the synthetic CT image 324, shown as a percentage of the original values ​​in μ map 320 obtained using the default AC method. The distorted CT image (attenuation map or μ map) 327 is generated by one or more trained ANNs. Image 328 is a difference image obtained by subtracting the base ground truth original μ map 320 from the distorted CT image 327.

[0035] The fourth row of images shows the coronal projection of the corresponding reconstructed PET images with attenuation correction. PET image 330 is the base ground truth image with attenuation correction using the original μ image 320. Attenuation correction is performed on PET image 334 using the synthetic CT image 324. Image 336 is the difference image obtained by subtracting the base ground truth image 330 from PET image 334. Attenuation correction is performed on PET image 337 using the distorted CT image 327. Image 338 is the difference image obtained by subtracting the base ground truth image 330 from PET image 337.

[0036] Both methods for generating attenuation maps provide reasonable attenuation correction for the heart compared to the original CT image. In cases where the CT image is not aligned with the PET image, activity at certain locations within the myocardium may be undercorrected. This can lead to a "false alarm defect" in the reconstructed image. Both methods for attenuation correction presented in this paper have the potential to avoid this. Furthermore, CT synthesis is well-positioned to address the CT truncation problem.

[0037] Figure 4An exemplary application of this framework for a series of physiologically gated images is illustrated. The first image set (402, 404, 406) illustrates attenuation correction using a synthetic CT image (attenuation map or μ-map) 402. The synthetic CT image 402 is generated by one or more trained ANNs based on PET images acquired at different respiratory phases during a cardiac scan. Image 404 is a PET image reconstructed from the synthetic CT image 402 using attenuation correction. Image 406 shows the PET image 404 superimposed on the synthetic CT image 402.

[0038] The second image set (408, 410, 412) illustrates attenuation correction using a distorted (or registered) CT image (attenuation map or μ-map) 408. The distorted CT image 408 is generated by one or more trained ANNs based on single CT and PET images acquired at different respiratory phases during a cardiac scan. Image 410 is a PET image reconstructed from the distorted CT image 408 using attenuation correction. Image 412 shows the PET image 410 superimposed on the distorted CT image 408.

[0039] While this framework has been described in detail with reference to exemplary embodiments, those skilled in the art will appreciate that various modifications and substitutions can be made therein without departing from the spirit and scope of the invention as set forth in the appended claims. For example, within the scope of this disclosure and the appended claims, elements and / or features of different exemplary embodiments may be combined with and / or substituted for each other.

Claims

1. An image processing system comprising: a non-transitory memory device for storing computer readable program code; and a processor device in communication with the non-transitory memory device, the processor device operating with the computer readable program code to perform steps comprising (i) receiving one or more artificial neural networks, (ii) receiving a non-attenuation corrected emission image of a region of interest, (iii) generating an attenuation map by applying at least the non-attenuation corrected emission image to the one or more artificial neural networks, and (iv) performing attenuation correction on the non-attenuation corrected emission image by using the attenuation map.

2. The image processing system of claim 1, wherein the one or more artificial neural networks comprise at least one first neural network, at least one second neural network, or a combination thereof.

3. The image processing system of claim 2, wherein the processor device operates with the computer readable program code to generate the attenuation map by applying a true anatomical image and the non-attenuation corrected emission image of the region of interest to the at least one first neural network.

4. The image processing system of claim 3, wherein the at least one first neural network performs an elastic registration of the true anatomical image and the non-attenuation corrected emission image to generate a registered anatomical image, wherein the registered anatomical image is output as the attenuation map.

5. The image processing system of claim 4, wherein the true anatomical image comprises a computed tomography (CT) or magnetic resonance (MR) image.

6. The image processing system of claim 2, wherein the at least one second neural network generates a synthetic anatomical image based on the non-attenuation corrected emission image, wherein the synthetic anatomical image is output as the attenuation map.

7. The image processing system of claim 1, wherein the non-attenuation corrected emission image comprises a positron emission tomography (PET) or single photon emission computed tomography (SPECT) image.

8. The image processing system of claim 1, wherein the one or more artificial neural networks comprise one or more convolutional neural networks.

9. The image processing system of claim 1, wherein the processor device operates with the computer readable program code to train the one or more artificial neural networks using a training set of pairs of true anatomical images and non-attenuation corrected emission images.

10. The image processing system of claim 1, wherein the processor device operates with the computer readable program code to train the one or more artificial neural networks using a training set of pairs of non-attenuation corrected emission images and linear attenuation coefficient maps.

11. An image processing method comprising: (i) receiving one or more artificial neural networks; (ii) receiving a non-attenuation corrected emission image of a region of interest; (iii) generating an attenuation map by applying at least the non-attenuation corrected emission image to the one or more artificial neural networks; and (iv) performing attenuation correction on the non-attenuation corrected emission image by using the attenuation map.

12. The image processing method of claim 11, further comprising applying a ground truth anatomical image of the region of interest to the one or more artificial neural networks.

13. The image processing method of claim 12, wherein the one or more artificial neural networks perform an elastic registration of a ground truth anatomical image to a non- attenuation corrected emission image to generate a registered anatomical image, wherein the registered anatomical image is output as an attenuation map.

14. The image processing method of claim 12, wherein the ground truth anatomical image comprises a computed tomography (CT) or magnetic resonance (MR) image.

15. The image processing method of claim 11, wherein the one or more artificial neural networks generate a synthetic anatomical image based on a non-attenuation corrected emission image, wherein the synthetic anatomical image is output as an attenuation map.

16. The image processing method of claim 11, wherein the non-attenuation corrected emission image comprises a positron emission tomography (PET) or single photon emission computed tomography (SPECT) image.

17. The image processing method of claim 11, further comprising training the one or more artificial neural networks using a training set of co-registered pairs of a ground truth anatomical image and a non-attenuation corrected emission image.

18. The image processing method of claim 11, further comprising training the one or more artificial neural networks using a training set of paired non-attenuation corrected emission images and linear attenuation coefficient maps.

19. One or more non-transitory computer-readable media embodying instructions executable by a machine to perform operations for image processing, the operations comprising: (i) receiving one or more artificial neural networks; (ii) receiving a non-attenuation corrected emission image of a region of interest; (iii) generating an attenuation map by applying at least the non-attenuation corrected emission image to the one or more artificial neural networks; and (iv) performing attenuation correction on the non-attenuation corrected emission image by using the attenuation map.

20. The one or more non-transitory computer-readable media of claim 19, wherein the one or more artificial neural networks comprise a U-Net or a residual block network.

Images