Attenuation correction factor generation

Abstract

And generating an attenuation correction factor. A framework for medical image data processing. An attenuation correction factor (ACF) map is generated by applying raw transmission data directly to one or more artificial neural networks. A medical image may then be reconstructed from the ACF map.

Description

TECHNICAL FIELD

[0001] The present disclosure relates generally to medical image data processing, and more specifically, to a framework for generating attenuation correction factors. BACKGROUND

[0002] The field of medical imaging has witnessed significant progress since the first use of X-rays to determine anatomical abnormalities. Medical imaging hardware has progressed in the form of newer machines such as medical resonance imaging (MRI) scanners, computed axial tomography (CAT) scanners, and the like. Digital medical images are constructed using 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”). Due to the large amount of image data generated in any given scan, there has been and remains a need to develop image processing techniques that can automate some or all of these processes to determine the presence of anatomical abnormalities in the medical images of a scan.

[0003] Multi-modality (or hybrid) imaging plays an important role in accurately identifying diseased and normal tissue. Multi-modality imaging provides a synergistic benefit by fusing images acquired by 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] In order to produce quantitatively accurate reconstructed images from multi-modality image data, attenuation correction (AC) needs to be performed. Without attenuation correction, significant artifacts can appear in the reconstructed images. Attenuation correction can be performed by estimating an attenuation map (μ-map) that represents the spatial distribution of tissue attenuation coefficients within the PET field of view. The intensity in the attenuation map represents the linear attenuation coefficient (LAC) values. Attenuation coefficients are used to describe how different media (e.g., bone, soft tissue, air) interact with imaging radiation.

[0005] CT image volumes measured in PET / CT imaging can be directly converted into 511 keV attenuation coefficient (μ) values to correct for photon attenuation effects. Splitting-based attenuation maps for PET / MR imaging can be generated from multi-point MR Dixon sequences. For PET / MR, the generated attenuation maps are well known to be problematic due to tissue misclassification, truncation, and incomplete or incorrect bone atlas addition. For PET / CT, there is a growing interest in application-specific imaging protocols (e.g., neuro, cardiac) and / or population-specific imaging protocols (e.g., pediatric) that can not otherwise require an acquired CT scan. SUMMARY

[0006] Described herein is a framework for medical image data processing. Attenuation correction factor (ACF) maps are generated by applying raw emission data directly to one or more artificial neural networks. Medical images can then be reconstructed from the ACF maps. BRIEF DESCRIPTION OF DRAWINGS

[0007] A more complete appreciation of the present disclosure and its many attendant aspects will be readily understood by reference to the following detailed description, when considered in connection with the accompanying drawings, in which:

[0008] Figure 1 A block diagram illustrating an exemplary system is shown;

[0009] Figure 2 An exemplary method of medical image processing is shown;

[0010] Figure 3 An exemplary training data set and an exemplary test data set are shown; and

[0011] Figure 4 An exemplary convolutional neural network (CNN) is shown. DETAILED DESCRIPTION

[0012] In the following description, numerous specific details are set forth such as examples of specific components, devices, methods, etc., to provide a thorough understanding of implementations of the present framework. It will be apparent, however, to one skilled in the art that the specific detail need not be employed to practice the present framework. In other instances, well-known materials or methods have not been described in detail in order to avoid obscuring the present framework. While the present framework is susceptible to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that there is no intent to limit the present disclosure to the particular forms disclosed, but on the contrary, this disclosure is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure. Further, for ease of understanding, certain method steps are depicted as separate steps; however, these separately depicted steps should not be construed as necessarily order dependent in their performance, unless so admitted.

[0013] Unless otherwise indicated, it will be understood that, as used in the following discussion, the terms "segmenting," "generating," "registering," "determining," "aligning," "positioning," "processing," "calculating," "selecting," "estimating," "detecting," "tracking," and the like can refer to the action and processes of a computer system or similar electronic

[0014] For brevity, an image or a portion of the image (e.g., a region of interest (ROI) in the image) corresponding to an object (e.g., a tissue, an organ, a tumor, etc. of a subject (e.g., a patient, etc.)) or a portion of the image (e.g., a ROI) including the object or the image or a portion of the image (e.g., a ROI) of the object itself can be referred to as the object or the image or the portion of the image (e.g., a ROI) of the object itself or including the object. For example, a ROI of an image corresponding to a lung or a heart can be described as the ROI including the lung or the heart. As another example, a chest or an image including the chest can be referred to as a chest image, or simply a chest. For brevity, a portion of an image corresponding to an object being processed (e.g., extracted, segmented) can be described as the object being processed. For example, a portion of an image corresponding to a lung being extracted from a remaining portion of the image can be described as the lung being extracted.

[0015] Deep learning techniques have demonstrated the efficacy of artificial intelligence (AI) for attenuation correction tasks. However, previous work has performed attenuation map (or mu map) synthesis in the image domain by transforming reconstructed PET images into pseudo-CT or attenuation coefficient map (or mu map) images.

[0016] A framework for attenuation correction is presented herein. According to one aspect, a deep artificial neural network (e.g., a convolutional neural network or CNN) is trained to directly generate maps of attenuation correction factors (ACFs) from raw emission data (e.g., PET projection data) without requiring any image reconstruction (e.g., pseudo-CT image or mu map reconstruction) to be performed. In some implementations, the present framework directly generates ACF maps in the PET data projection space (e.g., sinograms or histo-projections). Thus, the synthesized ACF maps advantageously already have the format required for the attenuation correction process in downstream image reconstruction.

[0017] The present framework advantageously is faster and more efficient than conventional image-based approaches because it does not require reconstructed images as input, but rather directly generates ACF maps from raw emission data. These and other example advantages and features will be described in greater detail in the following description.

[0018] Figure 1 FIG. 1 is a block diagram illustrating an example system 100. The system 100 includes a computer system 101 for implementing the framework as described herein. In some implementations, the computer system 101 operates as a standalone device. In other implementations, the computer system 101 can be connected (e.g., using a network) to other machines, such as a medical imaging device 102 and a workstation 103. In a networked deployment, the computer system 101 can operate in the capacity of a server (e.g., a server in a server-client user network environment, 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).

[0019] In one implementation, the computer system 101 includes a processor device or central processing unit (CPU) 104 coupled 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) via an input-output interface 121. The computer system 101 can further include support circuits, such as cache, power supplies, clock circuits, and communication buses. Various other peripheral devices, such as additional data storage and printing devices, can also be connected to the computer system 101.

[0020] This technology can be implemented in various forms of hardware, software, firmware, dedicated processors, or combinations thereof, either as part of microinstruction code, as part of an application or software product, or a combination thereof, executed via an operating system. In some embodiments, the technology described herein is implemented as computer-readable program code tangibly embodied in one or more non-transitory computer-readable media 105. Specifically, this technology can be implemented by processing module 107. Non-transitory computer-readable media 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 processor device 104 to process data provided by, for example, medical imaging device 102. Similarly, computer system 101 is a general-purpose computer system, which becomes a special-purpose computer system when the computer-readable program code is executed. The computer-readable program code is not intended to be limited to any particular programming language and its implementation. It will be understood that the teachings of the disclosure contained herein can be implemented using various 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.

[0021] Medical imaging device 102 acquires medical image data 132. This medical 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 this 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). For example, medical imaging device 102 can be a PET / CT, SPECT / CT, or PET / MR scanner. Alternatively, medical imaging device 102 can include a single modality (e.g., PET, SPECT).

[0022] Workstation 103 may include a computer and suitable peripherals, such as a keyboard and display device, and may operate in conjunction with the entire 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.

[0023] It should be further understood that, because some of the system components and method steps depicted in the accompanying drawings 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.

[0024] Figure 2 An exemplary method 200 for medical image processing is illustrated. It should be understood that the steps of method 200 can be performed in the order shown or a different order. Additional, different, or fewer steps may also be provided. Furthermore, method 200 can utilize… Figure 1 The system can be implemented through 100 different systems or combinations of the above.

[0025] At 202, processing module 107 receives one or more trained artificial neural networks (ANNs). The one or more ANNs can be trained to directly generate attenuation correction factor (ACF) maps based on raw emission data. Raw emission data typically refers to non-reconstruction data (e.g., raw PET data), including but not limited to list pattern data, sine waves, tissue projection data, or tissue image data.

[0026] The intensity in the ACF plot represents a spatially varying correction factor (or linear attenuation coefficient) that takes into account the attenuation of radiation emitted within the patient's body (e.g., 511 keV photons). One or more artificial neural networks may include one or more deep neural networks, such as a single convolutional neural network (CNN) or a recurrent neural network. One or more artificial neural networks may include any architecture, such as a U-Net or a residual block network.

[0027] Figure 3An exemplary training dataset 302 and an exemplary test dataset 304 are illustrated. The training dataset 302 is used to train one or more ANNs and includes a corresponding raw emission dataset 306a, actual ACF maps 308a for various projection views or angles, and a synthesized ACF map 310a. The actual ACF map 308a can be derived from measured attenuation data. In some implementations, the actual ACF map 308a is derived by converting the corresponding anatomical image volume (e.g., CT or MR) to a 511 keV attenuation factor. The converted image volume can then be forward-projected for all views (i.e., in projection space) and exponentially converted to obtain the actual ACF map 308a. The synthesized ACF map 310a is generated by one or more ANNs based on the raw emission data 306a. The actual ACF map 308a is used as a ground truth during supervised training of one or more ANNs. More specifically, the weights of one or more ANNs can be adjusted based on the differences between the actual ACF diagram 308a and the synthesized ACF diagram 310a.

[0028] Test data 304 includes raw emission data 306b from various projected views or angles, actual AC graphs 308b, and synthetic ACF graphs 310b. The synthetic ACF graph 310b is generated by one or more trained ANNs based on the raw emission data 306b. Strong similarity is shown between the actual ACF graph 308b and the synthetic ACF graph 310b.

[0029] Back Figure 2 At position 204, processing module 107 receives raw emission data of the region of interest (ROI) of the subject or patient. The ROI can be any region identified for further research, such as the heart or lungs. The raw (or non-reconstructed) emission data can be acquired by medical imaging device 102. Medical imaging device 102 may include molecular imaging modalities (e.g., PET, SPECT) that directly acquire raw emission data of the ROI.

[0030] Emissions from one or more radionuclides (injected into the bloodstream of the subject) are detected from multiple viewing angles. The energy, location, time, flight time, and / or order of arrival of the emissions from the one or more radionuclides can be measured by the medical imaging device 102 and recorded as raw emission data. More specifically, in PET imaging, the signal is generated by the annihilation of emitted positrons with electrons in the surrounding medium or tissue. Positron annihilation can result in the generation of two 511 keV photons emitted almost back-to-back, which are simultaneously detected in time by the surrounding PET detectors in the medical imaging device 102 to form a line of response (LOR). TOF PET measures the difference in arrival time of these two photons, thereby locating the emission point along the LOR. This localization improves spatial resolution and reduces image noise.

[0031] In some implementations, the raw transmission data is organized into a sinusoid that captures the projection of the image from various angles. Each row in the sinusoid can represent the sum of events along the corresponding line of response (LOR). The sinusoid can be segmented by time-of-flight (TOF) information recorded with each event. In other implementations, the raw transmission data is organized into an organized projection or organized image data, which involves spatially locating each LOR event using time-of-flight (TOF) information. Angular compression of the data can be applied. Other formats of raw transmission data are also useful.

[0032] At 206, image processing module 117 generates an ACF map by directly applying the raw emission data as input to one or more trained artificial neural networks (ANNs). In some embodiments, at least one first neural network generates an ACF map for each PET projection window (bin). Multiple ACF maps can be generated by applying raw emission data from multiple projection windows to one or more trained ANNs. Time-of-flight (TOF) information can be incorporated into the projection windows so that all events occurring along a LOR are described not only by a single sine curve coordinate. The TOF information can be used to identify the location of events occurring along the LOR.

[0033] Figure 4 An exemplary convolutional neural network (CNN) 402 is shown. PET projection data 404 is directly applied to CNN 402 to generate an ACF map 406 without the need for any pseudo-CT or μ map.

[0034] Back Figure 2At 208, image processing module 117 performs image reconstruction using an ACF map generated by one or more trained ANNs. The ACF map can be used to reconstruct attenuation-corrected medical images (e.g., PET images). The ACF map can be directly applied within the image reconstruction algorithm to correct for attenuation. In some implementations, attenuation-corrected medical images can be reconstructed using direct image reconstruction methods such as filtered backprojection. Other types of image reconstruction methods can also be used, such as iterative methods (e.g., maximum likelihood expectation maximization or MLEM) or ordered subset expectation maximization (OSEM). For example, the reconstructed image can be displayed at workstation 103.

[0035] The following is a list of non-limiting illustrative embodiments disclosed herein:

[0036] Illustrative Example 1. An image processing system includes: 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 together with the computer-readable program code to perform steps including: receiving one or more artificial neural networks; receiving raw emission data of a region of interest; generating an attenuation correction factor (ACF) map by directly applying the raw emission data to one or more artificial neural networks; and reconstructing a medical image using the ACF map.

[0037] Illustrative Example 2. The image processing system of Illustrative Example 1, wherein the medical images include positron emission tomography (PET) or single-photon emission computed tomography (SPECT) images.

[0038] Illustrative Example 3. An image processing system of any one of Illustrative Examples 1-2, wherein one or more artificial neural networks include a single convolutional neural network.

[0039] Illustrative Example 4. An image processing system of any one of Illustrative Examples 1-3, wherein one or more artificial neural networks include a U-shaped network or a residual block network.

[0040] Illustrative Example 5. An image processing system of any one of Illustrative Examples 1-4, wherein one or more artificial neural networks are trained using the corresponding raw emission dataset and the actual ACF graph.

[0041] Illustrative Example 6. An image processing system of any one of Illustrative Examples 1-5, wherein the raw transmission data includes one or more sine waves separated by time-of-flight information.

[0042] Illustrative Example 7. An image processing system of any one of Illustrative Examples 1-6, wherein the raw emission data includes tissue projection or tissue image data.

[0043] Illustrative Example 8. An image processing system of any one of Illustrative Examples 1-7, wherein the processor device operates together with computer-readable program code to generate multiple ACF graphs by directly applying raw emission data on multiple projection windows to one or more artificial neural networks.

[0044] Illustrative Example 9. An image processing method includes: receiving one or more artificial neural networks; receiving raw emission data of a region of interest; generating an attenuation correction factor (ACF) map by directly applying the raw emission data to one or more artificial neural networks; and reconstructing a medical image using the ACF map.

[0045] Illustrative Example 10. Image processing method of Illustrative Example 9, wherein the medical images include positron emission tomography (PET) or single-photon emission computed tomography (SPECT) images.

[0046] Illustrative Example 11. An image processing method of any one of Illustrative Examples 9-10, wherein one or more artificial neural networks include a single convolutional neural network.

[0047] Illustrative Example 12. An image processing method of any one of Illustrative Examples 9-11, wherein one or more artificial neural networks include a U-shaped network or a residual block network.

[0048] Illustrative Example 13. The image processing method of any one of Illustrative Examples 9-12 further includes: training one or more artificial neural networks using the corresponding original emission dataset and the actual ACF graph.

[0049] Illustrative Example 14. An image processing method of any one of Illustrative Examples 9-13, wherein the raw emission data includes one or more sine waves.

[0050] Illustrative Example 15. An image processing method of any one of Illustrative Examples 9-14, wherein the raw emission data includes tissue projection or tissue image data.

[0051] Illustrative Example 16. An image processing method of any one of Illustrative Examples 9-15, wherein generating an ACF image includes: directly applying raw emission data on a plurality of projection windows to one or more artificial neural networks to generate a plurality of ACF images.

[0052] Illustrative Example 17. One or more non-transitory computer-readable media embodying instructions that can be executed by a machine to perform operations including: receiving one or more artificial neural networks; receiving raw emission data of a region of interest; generating an attenuation correction factor (ACF) map by directly applying the emission projection data to one or more artificial neural networks; and reconstructing a medical image using the ACF map.

[0053] Illustrative Example 18. One or more non-transitory computer-readable media of Illustrative Example 17, wherein the operation further includes: training one or more artificial neural networks using corresponding raw emission datasets and actual ACF graphs.

[0054] Illustrative Example 19. One or more non-transitory computer-readable media of any one of Illustrative Examples 17-18, wherein the raw emission data includes one or more sine curves, tissue projections, or tissue image data.

[0055] Illustrative Example 20. One or more non-transitory computer-readable media of any one of Illustrative Examples 17-19, wherein generating an ACF graph includes: directly applying raw emission data on a plurality of projection windows to one or more artificial neural networks to generate a plurality of ACF graphs.

[0056] While this framework has been described in detail with reference to exemplary embodiments, those skilled in the art will understand 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: Non-transitory memory devices used for storing computer-readable program code; as well as A processor device communicating with the non-transitory memory device, the processor device operating together with the computer-readable program code to perform steps including the following: (i) Receive one or more artificial neural networks, (ii) Receive raw transmission data for the region of interest. (iii) Generating an attenuation correction factor (ACF) map by directly applying the raw transmission data to the one or more artificial neural networks, and (iv) Reconstructing medical images using the ACF plot.

2. The image processing system according to claim 1, wherein the medical image includes positron emission tomography (PET) or single-photon emission computed tomography (SPECT) images.

3. The image processing system of claim 1, wherein the one or more artificial neural networks comprise a single convolutional neural network.

4. The image processing system according to claim 1, wherein the one or more artificial neural networks include a U-shaped network or a residual block network.

5. The image processing system of claim 1, wherein the one or more artificial neural networks are trained using the corresponding original emission dataset and the actual ACF graph.

6. The image processing system of claim 1, wherein the raw transmission data comprises one or more sine waves separated by time-of-flight information.

7. The image processing system of claim 1, wherein the raw emission data includes tissue projection or tissue image data.

8. The image processing system of claim 1, wherein the processor device operates together with the computer-readable program code to generate the ACF graph by directly applying the raw emission data on the plurality of projection windows to the one or more artificial neural networks to generate a plurality of ACF graphs.

9. An image processing method, comprising: Receive one or more artificial neural networks; Receive raw transmission data for the region of interest; Attenuation correction factor (ACF) maps are generated by directly applying the raw emission data to the one or more artificial neural networks; and Medical images are reconstructed using the ACF diagram.

10. The image processing method according to claim 9, wherein the medical image includes a positron emission tomography (PET) or single-photon emission computed tomography (SPECT) image.

11. The image processing method of claim 9, wherein the one or more artificial neural networks comprise a single convolutional neural network.

12. The image processing method according to claim 9, wherein the one or more artificial neural networks include a U-shaped network or a residual block network.

13. The image processing method according to claim 9, further comprising: The one or more artificial neural networks are trained using the corresponding original emission dataset and the actual ACF graph.

14. The image processing method according to claim 9, wherein the raw emission data includes one or more sine waves.

15. The image processing method according to claim 9, wherein the raw emission data includes tissue projection or tissue image data.

16. The image processing method according to claim 9, wherein generating the ACF image comprises: The raw emission data on multiple projection windows are directly applied to the one or more artificial neural networks to generate multiple ACF graphs.

17. One or more non-transitory computer-readable media embodying instructions that are machine-executable to perform operations including: Receive one or more artificial neural networks; Receive raw transmission data for the region of interest; Attenuation correction factor (ACF) maps are generated by directly applying the raw emission data to the one or more artificial neural networks; and Medical images are reconstructed using the ACF diagram.

18. The one or more non-transitory computer-readable media of claim 17, wherein the operation further comprises: The one or more artificial neural networks are trained using the corresponding original emission dataset and the actual ACF graph.

19. The one or more non-transitory computer-readable media of claim 17, wherein the original emission data comprises one or more sine curves, tissue projections, or tissue image data.

20. One or more non-transitory computer-readable media according to claim 17, wherein generating the ACF diagram comprises: The raw emission data on multiple projection windows are directly applied to the one or more artificial neural networks to generate multiple ACF graphs.

Images