Magnetic resonance hardware contribution attenuation in positron emission tomography

Abstract

PET imaging (406) accounts for attenuation by MR hardware (110). A camera (112) captures MR hardware (110) as positioned on or by the patient (116). For example, optical imaging (402) is performed of MR local coils to be between or as positioned in the patient (116) between a transmission source and a PET detector. Image processing is used to determine (404) a location of the MR hardware (110). Attenuation of the MR hardware (110) is accounted for in attenuation correction for PET imaging (402) based on the determined location.

Description

[0001] Related applications This patent document claims the benefit of provisional U.S. patent application No. 62 / 706,806, filed on September 11, 2020, with the filing date under 35 U.S.SC §119(e), which is incorporated herein by reference. Technical Field

[0002] This embodiment relates to positron emission tomography (PET). In PET, attenuation information is used in reconstruction. Typically, attenuation is derived from computed tomography (CT) scans. Background Technology

[0003] In combined MRI and PET systems, MR hardware can attenuate emissions from PET, and CT may be unavailable, at least for detecting attenuation from the MR hardware. To acquire MR images, a local coil can be placed close to the patient's surface. In most MR acquisition sequences, the coil is not visible in the MR image. In combined PET and MR systems, the coil and the material housing the coil attenuate the 511 keV photons emitted from the patient and used to create the PET image.

[0004] For some rigid and non-deformable coils (e.g., head and neck coils), the material is physically attached to the patient stage in a fixed and 3D position, and thus an orientation is assumed. However, the actual position may differ slightly from the assumed position due to the tolerances of the patient stage positioning and the mechanism by which the coil is attached to the stage. Some MR coils are not placed in a known position or orientation. For example, flexible local coils are placed on the patient differently for each acquisition and are typically physically bent around the patient's body. The position, orientation, and deformation of the coil are unknown. For attenuation correction in PET imaging, MR invisible coils are ignored, although photon attenuation is performed. Due to the unknown positioning, attenuation correction is not performed on this material, and approximate errors of up to 10% in the PET active concentration may occur due to some coils. Summary of the Invention

[0005] For introduction, the preferred embodiments described below include methods, systems, and non-transient computer-readable media for PET imaging that take into account attenuation caused by MR hardware. A camera captures MR hardware, such as that positioned on or by a patient. For example, an MR local coil to be located, or such being located, between an emitter and a PET detector within the patient is optically imaged. Image processing is used to determine the position of the MR hardware. Based on the determined position, attenuation of the MR hardware is taken into account in attenuation correction for PET imaging.

[0006] In a first aspect, a method for PET imaging that takes into account attenuation caused by MR hardware is provided. An optical camera images a patient. The imaging captures images of the MR hardware and the patient. An attenuation coefficient map relative to the patient for the MR hardware is generated from the images. The PET imaging uses attenuation correction based at least in part on the attenuation coefficient map for the MR hardware.

[0007] In one embodiment, the patient is imaged outside the aperture of the combined PET and MR imaging system using the camera. The camera is oriented toward the patient bed of the combined PET and MR imaging system. For example, the patient is imaged while the patient is outside the aperture.

[0008] In another embodiment, one or more optical references on the MR hardware are imaged. The position of the MR hardware relative to the patient is determined from the optical references.

[0009] A template map of the attenuation of the MR hardware can be used. An attenuation coefficient map of the MR hardware, localized for the patient, is generated from the template map of the attenuation of the MR hardware. In one embodiment, the MR hardware is a flexible local coil. The template is adapted to reflect the localization of the flexible local coil on the patient. For example, the adaptation identifies the shape and position of a marker of the flexible local coil in the image. The three-dimensional position of the marker is determined from the shape and position. The template is fitted to the three-dimensional position. As another example, the three-dimensional position of the marker of the flexible local coil is predicted by a machine learning model receiving the image as input. The template is fitted to the three-dimensional position. In another embodiment, the machine learning model receiving the image as input directly outputs an attenuation coefficient map fitted to the patient.

[0010] More than one camera may be used. The patient is imaged using the optical camera and at least one additional optical camera. The attenuation coefficient map is generated from the image and at least one additional image from the at least one additional optical camera. In some embodiments, one or more three-dimensional locations of the MR hardware are identified from the image using color-coded markers and / or visible patterns on the MR hardware.

[0011] In one embodiment, at least a portion of the patient is reconstructed from the emission. The reconstruction uses attenuation correction, wherein an attenuation coefficient map for the MR hardware is included together with the attenuation for the portion of the patient.

[0012] In a second aspect, a medical imaging system includes: a positron emission tomography (PET) imager; and a magnetic resonance imaging (MRI) imager configured to image a patient using one or more local coils located on the patient. A camera is configured to capture images of the patient and the one or more local coils located on the patient. A processor is configured to correct for attenuation of the one or more local coils based on one or more locations identified respectively from the images. The correction is used for imaging performed by the PET imager.

[0013] In a further embodiment, the positron emission tomography (PET) imager and the magnetic resonance imaging (MRI) imager share an aperture for imaging the patient. The camera is mounted outside the aperture, and the images are captured when the patient is mostly outside the aperture.

[0014] In one embodiment, the one or more local coils include flexible local coils. The processor is configured to identify the one or more locations as three-dimensional locations based on markers on the one or more local coils.

[0015] In other embodiments, the processor is configured to generate an attenuation coefficient map by fitting the template attenuation coefficient map to the three-dimensional position.

[0016] In a third aspect, a method for PET imaging that takes into account attenuation caused by an MR local coil is provided. An optical image of an MR local coil, such as one located on a patient, is captured. The distribution of attenuation from the MR local coil is determined based on the optical image. A PET image is reconstructed from the emission, wherein the reconstruction includes attenuation correction using the distribution of attenuation from the MR local coil.

[0017] In one embodiment, markers on the MR local coil are located based on the optical image, and an attenuation template is fitted to the located markers. In other embodiments, distribution information is output from a machine learning model in response to input optical information. For example, spatial transformations of the template, such as deformed template and / or marker positions, are output. In still other embodiments, the deformation and position of the MR local coil, such as that located on the patient for MR imaging, are determined. The attenuation distribution is based on the deformation and position of the MR local coil.

[0018] This invention is defined by the appended claims, and the content of this section should not be construed as limiting those claims. Further aspects and advantages of the invention are discussed below in conjunction with preferred embodiments, and further aspects and advantages of the invention may be claimed later, independently or in combination. Attached Figure Description

[0019] The components and drawings are not necessarily drawn to scale; rather, the focus is on illustrating the principles of the invention. Furthermore, throughout the drawings, similar reference numerals indicate corresponding parts in different views.

[0020] Figure 1 This is a block diagram of a system according to one embodiment for taking into account the attenuation of MR hardware in PET imaging; Figure 2 This is a front view of an example medical imaging system including a camera for attenuation estimation; Figure 3 This is a top view of an example flexible local coil for MR; Figure 4 This is a flowchart of an embodiment of a method for PET imaging that takes into account attenuation performed by MR hardware; Figure 5 The illustration shows an example analytical approach for attenuation template fitting; Figure 6 A schematic diagram is shown showing how to correlate a camera with a marker for determining the marker's location; Figure 7 The diagram illustrates a deep learning-based approach for example of attenuation template fitting; and Figure 8 The illustration shows an example of a deep learning-based approach for determining the attenuation distribution of MR hardware. Detailed Implementation

[0021] Attenuation of MR hardware can be corrected by using the spatial distribution of attenuation coefficients. The precise three-dimensional (3D) position and orientation of the coil are determined to find the spatial distribution of attenuation by the MR hardware relative to PET / MR imaging coordinates. The spatial orientation and deformation of the MR hardware are determined from one or more optical cameras and provided for PET attenuation correction. The 3D position, orientation, and deformation of the MR coil are determined from camera images for accurate PET attenuation correction, thus providing attenuation correction for MR coils whose position is otherwise unknown for PET imaging. For example, one or more optical images (e.g., two-dimensional (2D) images) are used to create an attenuation correction map of deformable, PET-invisible, and MR-invisible MR coils in unknown positions and deformations preserved for camera imaging.

[0022] Optical cameras are low-cost, off-the-shelf devices, providing a cost-effective way to account for MR hardware attenuation. Similarly, markers used to assist image processing are simple and inexpensive to manufacture. MR hardware requires no modification other than attaching visible markers. The markers are invisible in PET or MR images and do not distort them. No MR or PET-detectable markers are required, thus avoiding sources of phase entanglement, image artifacts, or additional radiation sources.

[0023] Figure 1 A medical imaging system is illustrated. For example, a PET-MR system reconstructs data in PET while taking into account attenuation correction from the MR hardware. The MR hardware can attenuate photons in PET imaging, therefore an optical camera 112 is used to determine the distribution of attenuation contributed by intervening in the MR hardware. The medical imaging system performs… Figure 4 One method or another to reconstruct PET images while correcting for attenuation from MR hardware.

[0024] The PET-MR system includes an MR scanner or imager 104, a PET scanner or imager 102, a camera 112, and a bed 114 for a patient 116. A computed tomography (CT) system is not provided but may be included, such as being included in the same room or facility without sharing a common aperture 108. The processor 118, memory 120, and / or display 122 are part of the PET imager 102 and MR imager 104, or may be separate (e.g., a computer or workstation). Additional, different, or fewer components may be provided. For example, the system may be a computer without the MR imager 104 and / or the PET imager 102. As another example, the medical system includes a power supply, communication systems, and a user interface system.

[0025] Bed 114 is a wheeled bed, table, or other support for holding the subject of examination (such as a patient). A robot, gears, cables, tracks, and / or other devices move bed 114. This movement is along an axial dimension but may include other directions of movement, such as up and down. MR imager 104 and PET imager 102 share a housing (e.g., a frame) and bed 116. An aperture 108, formed for scanning the patient using MR and / or PET, is cylindrical or other shapes and is shared by PET imager 102 and MR imager 104. For PET and / or MR imaging and / or during PET and / or MR imaging, bed 114 moves the patient 116 into aperture 108.

[0026] PET imager 102 is a nuclear imaging system. PET imager 102 includes a detector ring, such as a scintillation crystal coupled to an avalanche photodiode. Solid-state or semiconductor detectors can be used. The detectors detect gamma rays emitted indirectly by a positron emission tracker. A detector ring can be used to detect pairs of gamma rays generated by the same positrons. The gamma ray pairs travel approximately 180 degrees apart. If the directions of travel intersect the detector arrangement at two unknown points, an overlapping pair can be detected. To distinguish specific pairs, the overlap of the detected gamma rays is determined. Timing of the received data is used to pair the detected gamma rays. As cue data, the timing can also indicate the time of flight, thus providing general information about where the emission occurred along the response line. Taking into account the detectors involved in the detection of the event, the response line is determined based on the detected event.

[0027] The detected events are transmitted to memory 120 and / or processor 118. Processor 118 is connected to detector 16, such as via a superprocessor. Processor 118 is also connected to MR system 104 to receive information from the patient's MR scan before, during, and / or after the PET scan.

[0028] The patient's tissue and any other material between the detector and the detector may attenuate gamma radiation. By taking this attenuation into account, PET imaging or reconstruction is more accurate. Attenuation models of the patient are fitted to MR data, or attenuation determined from the patient's MR scans provide attenuation information from the patient. MR hardware between the patient and the PET detector may also cause attenuation. The processor 118 also takes into account attenuation caused by the MR hardware in PET reconstruction.

[0029] The MR imager 104 includes a primary magnet for generating a generally uniform primary magnetic field in the aperture 108. It also includes a whole-body coil and gradient coils. A whole-body coil may not be included. The MR imager 104 may include local coils, such as coils with a rigid or hard housing with or without hinges. The local coils are positioned on, against, or adjacent to the patient 116 (e.g., within 6 inches). In one embodiment, one or more flexible or “blanket” type local coils 110 are used. The local coil 110 is located above or below the patient and at least partially conforms to the patient 116 due to its flexibility. The local coil 110 may include a substrate, housing, electronics, cables, antennas, and / or other gamma-ray attenuating materials. Other MR hardware (such as cables) may be positioned between the patient and at least some of the detectors of the PET imager 102 in the aperture 108.

[0030] The MR imager 104 is configured to image the patient 116 using one or more local coils 110 (whether flexible or fixed) located on the patient 116. Radio frequency pulses are transmitted by the MR imager 104, such as by whole-body coils under spatial control of gradient coils. The local coils 110 are used to receive radio frequency signals generated as the spins of molecules shifted by the transmitted pulses realign with the main magnetic field.

[0031] The MR imager 104 is housed within the same enclosure as the PET imager 102, or is separated from and connected to the bed 114 via a common track. Completely separate MR imager 104 and PET imager 102 can be used. Local coil 110 or other MR hardware may be left on the patient 116 even during PET imaging with respect to separate imaging systems (e.g., not a combined system or without a shared enclosure and / or bed 114).

[0032] To account for attenuation of MR hardware (such as flexible local coil 110) in PET imaging, a marker is added to the MR hardware to make it detectable. For PET detection, the marker would need to emit radiation, which is undesirable. Because MR hardware is used in MR imaging, it is typically not detectable by the MR imager 104 to avoid imaging artifacts. Instead, the camera 112 detects the MR hardware.

[0033] Camera 112 is an optical camera, such as a CCD or solid-state camera. Camera 112 captures two-dimensional images or videos. In other embodiments, depth cameras, stereo camera systems, structured light, time-of-flight and / or three-dimensional imaging cameras are used.

[0034] Camera 112 is installed in the same room as PET imager 102 and / or MR imager 104. For example, and as Figure 1 and 2 As shown, camera 112 is attached to frame 200 or housing at one end of aperture 108 (e.g., above patient 116 and bed 114). Camera 112 captures images of patient 116 before being positioned in aperture 108 (e.g., when the patient is mostly outside aperture 108). For example, camera 112 is oriented downwards, with the field of view mostly or completely outside aperture 108, to capture patient 116 before beginning to move into aperture 108 or as patient 116 is moved into aperture 108. Camera 112 may be tilted to capture patient 116 within aperture 108.

[0035] In one embodiment, an optical camera 112 is mounted on a holder 200 outside the aperture 108. Multiple cameras 112 may be provided to stitch together a larger field of view and / or redundancy. The cameras 112 are used to create one or more 2D images of the MR hardware (e.g., local coil 110) on the patient 116 before, during, or after PET image acquisition. In one arrangement, the optical camera 112 is oriented to image the patient outside the aperture 108 and / or holder 200. In another arrangement, the optical camera 112 is oriented to image the patient within the aperture 108, thereby facilitating optical imaging during PET and / or MR acquisition.

[0036] Each camera 112 is calibrated to the patient table 114, gantry 200, or other part of the medical imaging system. The cameras are calibrated by capturing images at various orientations of a checkerboard or other markers placed flat on the patient table or bed 114 at a distance from the camera 112. The marker positions determined from each camera are combined to improve accuracy and coverage. Camera intrinsic parameters, extrinsic parameters, and / or lens distortion parameters are estimated as part of the calibration. Camera intrinsic parameters include the camera's focal length. The focal length is used to determine the spatial position of the markers placed on the MR coil 110 after calibration. The internal matrix, rotation matrix, and translation vector generated from the calibration are used to transform from image space coordinates to world space coordinates.

[0037] Once calibrated, camera 112 can be used to determine the spatial distribution of the MR hardware relative to PET imager 102. Processor 118 applies image processing to detect the spatial distribution of the MR hardware from one or more images captured from camera 112. In one embodiment, markers are added to or included on the MR hardware to aid detection. The markers are colored and / or patterned. For example, colored markers are placed at fixed or known locations on the surface of MR coil 110. The entire MR coil 110 can be patterned. Structured patterns on the markers can improve the accuracy of determining the central location of the markers. The markers can be two-dimensional or three-dimensional, such as flat stickers, spheres, or cylinders.

[0038] Figure 3 A top view of one embodiment of a flexible coil 110 for use with a patient is shown. The local coil 110 includes a substrate having a top surface 302, a rigid housing 304 for electronics and / or cables, color-coded circular markers 306, and holes 308. Any number and / or positioning of the markers 306 can be used. Other local coils 110 with additional marker arrangements can be used.

[0039] Processor 118 is a general-purpose processor, digital signal processor, graphics processing unit, application-specific integrated circuit, artificial intelligence processor, field-programmable gate array, digital circuit, analog circuit, a combination thereof, or other device now known or developed later for determining the spatial distribution of MR hardware relative to the patient (or patient table 114) and / or reconstructing the MR hardware in PET imaging while taking into account attenuation of the MR hardware. Processor 118 can be a single device, multiple devices, or a network. For more than one device, parallel or sequential partitioning of processing can be used. Different devices comprising processor 118 can perform different functions, such as one processor for determining the spatial distribution of attenuation from the MR hardware and another processor for reconstructing the object (i.e., the patient) space using attenuation correction. In one embodiment, processor 118 is a control processor or other processor for a medical imaging system or PET imager 102. In other embodiments, processor 118 is part of a separate workstation or computer.

[0040] Processor 118 operates according to stored instructions to perform the various actions described herein. Processor 118 is configured by software, firmware, and / or hardware to perform... Figure 4 Any or all of the actions. The operation and configuration of processor 118 are described in general below. Figure 4 The example implementation is described in more detail in the following discussion.

[0041] Processor 118 is configured to reconstruct the activity distribution in PET imaging using emissions detected along the response lines. Time-of-flight (TOF) can be used. Emissions detected along the response lines are used to reconstruct the activity distribution. To account for differences in attenuation due to different problems and any obstacles along different response lines, attenuation correction factors or integrated attenuation along the corresponding response lines are used in the reconstruction. For some of the response lines, attenuation may be caused by MR hardware (such as local coil 110). Similarly, the reconstruction can account for scattering using attenuation. Scattering from MR hardware is taken into account in the reconstruction.

[0042] Processor 118 is configured to correct the attenuation of one or more local coils 110 based on one or more locations identified from images captured by camera 112. Processor 118 is configured to identify the one or more locations as three-dimensional (3D) locations based on markers on the one or more local coils. The 3D location indicates the position, deformation, and / or orientation of the MR hardware. Processor 118 generates an attenuation coefficient map (attenuation map or attenuation correction factor map) by fitting a template attenuation coefficient map to the 3D location detected by camera 112.

[0043] Processor 118 uses events (e.g., response line events or PET data), attenuation data, parameter data, attenuation correction factor (ACF), and / or other information stored in memory 120 for processing. For processing, data bypasses memory 120, is temporarily stored in memory 120, or is loaded from memory 120.

[0044] Detected events, response line information (e.g., sine wave), time steps, cue data, attenuation information, ACF, reconstructed images, or other data are stored in memory 120. The data is stored in any format. Memory 120 is a buffer, cache, RAM, removable media, hard disk drive, magnetic, optical, database, or other memory now known or to be developed later. Memory 120 is a single device or a group of two or more devices. Memory 120 is part of the PET imager 102 or a remote workstation or database, such as a PACS memory.

[0045] Additionally or alternatively, memory 120 is a non-transient computer-readable storage medium having processing instructions. Memory 120 stores data representing instructions executable by a programmed processor 118 for determining the spatial distribution of MR hardware, determining the spatial distribution of attenuation from the MR hardware, and / or reconstructing in PET based on attenuation correction using the MR hardware. Instructions for implementing the processes, methods, and / or techniques discussed herein are provided on a non-transient computer-readable storage medium or memory (such as cache, buffer, RAM, removable media, hard disk drive, or other computer-readable storage media). Computer-readable storage media include various types of volatile and non-volatile storage media. Functions, actions, or tasks illustrated in the figures or described herein are performed in response to one or more sets of instructions stored in or on a computer-readable storage medium. Functions, actions, or tasks are independent of a particular type of instruction set, storage medium, processor, or processing strategy and can be performed by software, hardware, integrated circuits, firmware, microcode, etc., operating individually or in combination. Similarly, processing strategies can include multiprocessing, multitasking, parallel processing, etc. In one embodiment, the instructions are stored on a removable media device for reading by a local or remote system. In other embodiments, the instructions are stored at a remote location for transmission over a computer network or telephone line. In still other embodiments, the instructions are stored within a given computer, CPU, GPU, or system.

[0046] Medical imaging systems may include a display 122. For example, a processor 118 reconstructs a scanned patient or object from response-line PET data and attenuation data. The reconstructed object or reconstructed activity distribution is used for three-dimensional rendering, multi-planar reconstruction, or two-dimensional imaging of the function of the object or patient 116. The image is displayed on the display 122. The display 122 may be a CRT, LCD, plasma screen, projector, printer, or other output device for displaying PET images of the activity distribution. The PET images may be qualitative or quantitative images.

[0047] Figure 4 An embodiment of a method for PET imaging that takes into account attenuation caused by MR hardware (such as local coils) is illustrated. The distribution of the MR hardware relative to the patient and / or the PET imager and the attenuation caused by the MR hardware are determined based on image processing of one or more optical images of the patient and the MR hardware. PET reconstruction includes attenuation correction for the distribution of attenuation from the MR hardware, resulting in more accurate PET imaging.

[0048] Figure 4 The method is implemented by camera 112, processor 118, PET imager 102, medical imaging system, and / or another component or system. For example, the camera performs action 402, processor 118 performs action 404, and PET imager 102 performs action 406 with or without processor 118. The resulting PET image is displayed on display 122. Other components may perform actions such as: the workstation performs both actions 404 and 406 based on information provided by camera 112 and PET imager 102 or memory 120.

[0049] Additional, different, or fewer actions can be performed. For example, actions for placing a patient on a bed and positioning the MR hardware relative to the patient are included. As another example, actions for calibrating a camera to a PET-MR imager are provided. The actions are performed in the order shown or in a different order. Action 404 can be included as part of action 406.

[0050] In action 402, one or more cameras image the patient. These cameras are located outside the scanning aperture of the combined PET and MR imaging system, such as near the aperture's inlet / outlet, and are connected to the imaging system housing. The cameras may be located within or inside the aperture of the imaging system housing. The cameras are oriented toward the patient bed of the combined PET and MR imaging system. Alternatively, images are captured before the patient is transferred to the bed of one or more PET and / or MR imaging systems. Optical imaging of the patient with MR hardware is performed while the patient is outside, inside, or partially inside and outside the aperture.

[0051] The one or more cameras capture one or more images of the patient and MR hardware (such as one or more MR local coils) located on the patient. The MR hardware positioned for MR imaging will be on the patient during PET imaging, so the one or more images are captured using MR hardware positioned for both PET and MR imaging.

[0052] Images are captured before, during, and / or after PET and / or MR imaging. For example, as the bed moves the patient into the aperture for MR and PET imaging, a camera captures a series of images (e.g., video) of the patient. The images are stitched together or processed into patches representing different parts of the patient. As another example, multiple cameras capture the patient from different angles when they are outside the aperture or mostly outside the aperture, allowing representations of the patient from different views to be stitched together. In other examples, multiple cameras have overlapping or identical fields of view, and the images are processed separately for the combined (e.g., averaged) results. A single camera with a single image showing the patient and at least a portion of the MR flexible local coil as MR hardware is used as the camera for the following purposes. Figure 4 An example from further discussion.

[0053] The captured image is an optical image. Images in infrared or other frequencies can be used. The wavelength or the detected signal is different from the wavelength or the detected signal used for PET or MR imaging.

[0054] The captured image represents the patient and the MR local coil. To facilitate detection of the MR local coil, one or more optical references are placed on it. For example, a colored or patterned (cross-line) circular or other shaped sticker is placed at a known or specific location on the MR local coil. Cylindrical or other shaped 3D markers can be used. Where the local coil can be bent or folded and / or due to the three-dimensional surface of the local coil, the markers are displaced relative to each other in three dimensions. The image captures or represents the relative placement of the markers to each other on the MR local coil.

[0055] In action 404, the processor generates an attenuation coefficient map relative to the patient or patient table for the MR hardware from the image. The relative placement of the markers (e.g., color-coded markers or visible patterns) and / or the distortion of the markers in the image are used to determine the spatial distribution of the local coils on the patient, compared to a known distribution and shape of the markers on the local coils. This spatial distribution is based on the optical image to provide a distribution of attenuation from the MR local coils. Any deformation and position of the MR local coils, as located on the patient for MR imaging, are used to generate the attenuation coefficient map. The deformation and position of the MR local coils or markers captured in the image indicate the distribution of attenuation. In an alternative embodiment, the deformation and position of the MR local coils are determined from an image without markers.

[0056] The position of the MR local coil (or other MR hardware) in three dimensions is determined relative to the patient. The MR local coil is placed on or against the patient. The PET imager images the emissions from the patient. By detecting the position relative to the patient, the 3D positioning of the MR local coil in the same coordinate system (e.g., world coordinates) as the PET imager is determined from the images (or multiple images).

[0057] The processor generates an attenuation coefficient map (e.g., attenuation distribution in three dimensions) for an MR local coil, such as one located on a patient, from the spatial distribution of the local coils. The model of the local coils or the measured attenuation is deformed, oriented, and / or positioned to account for the local coils' positioning, orientation, and / or deformation on the patient. For example, the processor generates an attenuation coefficient map of the MR hardware, such as one positioned in 3D for the patient, from a template map of the MR hardware's attenuation.

[0058] The model or template attenuation coefficient map can be derived from actual measurements of attenuation caused by the MR local coil or the type of MR local coil. For example, to generate a template, markers for the MR local coil temporarily include small attenuation objects (e.g., ball bearings or other objects detectable for computed tomography or other attenuation measurements) to facilitate marker positioning and orientation. The local coil is laid flat or laid in another known location in three dimensions. The attenuation coefficient map is created by transmission measurements, such as performing a rotating rod radiation source for scanning the local coil, lutetium oxysilicate (LSO) radiation from a PET scan, CT scan, or some combination of several measurements. The location of the markers is identified in the attenuation coefficient map by positioning small, opaque objects (e.g., ball bearings) on each marker. For circular markers, only a single opaque object is required at the center of the circle. In other markers, more than one object may be used per marker to create the attenuation template. Alternatively, simulation or known material properties can be used to create the model or template attenuation coefficient map.

[0059] A template attenuation map of the coil is created once. Then, if the local coil is to be used for a patient, the measurement opacity can be removed. Other local coils of the same type have markers in the same location, but without the measurement opacity used to create the template. Then, for each patient imaging instance, the template is adapted to create an attenuation map of the coil on the patient. Templates or models with known attenuation for local coils are modified to reflect the local coil's positioning in three dimensions (e.g., position on the patient, orientation on the patient, and / or deformation of the local coil around the patient) for PET and MR imaging.

[0060] Various approaches can be used to adapt the attenuation coefficient map. In the analytical approach, one or more two-dimensional optical images of the coil on the patient can be preprocessed, such as filtered or normalized. The spatial location of each marker is calculated based on known parameters of the marker shape. The template attenuation coefficient map is fitted to the calculated marker location. If the goodness of fit of the markers is below a threshold, further improvement of the fit can be performed using a data-driven PET consistency criterion method. The data-driven PET consistency criterion method treats various PET-detected events as point sources and fits the attenuation to the PET measurement results.

[0061] Figure 5 An embodiment of the analysis approach is illustrated. Markers on the MR local coil are located from the optical image 502 via image processing 504. The processor identifies the shape and position of the markers on the flexible local coil in the image. A shape recognition algorithm is used to identify the markers in the image. As a result of image processing 504, the marker position 506 in three dimensions is determined from the shape and position of the markers in the image. For example, for circular markers or straight lines, the Hough Transform is used. The voting space of the Hough Transform can be extended to consider the non-circular appearance of a two-dimensional (2D) circle projected onto the plane when the marker is not orthogonal to the camera axis (i.e., the projected shape of the circle would be elliptical). For 3D markers, if they are imaged outside the optical axis, the projected appearance in the image plane is also elliptical; however, the semi-axis has a length more similar to the length of the 2D marker projection. The center of the ellipse is the same as the center of the imaged 2D circle only when the circular plane is parallel to the image plane. The same applies to 3D spheres when the sphere is aligned with the camera axis. Determine both the center of the marker and the diameter of the marker in the image in order to find the spatial location of the marker in world space.

[0062] Figure 6 A schematic diagram of the calculation for 3D position determination based on a pinhole camera model is shown. F I is the diameter of the marker, measured in millimeters. FLet f be the diameter of the marker in pixels, and f be the focal length of camera 112. Using this prior information and the measured diameter in pixels found in the image, the distance (D) of the marker from the camera plane in camera coordinates can be found. The x and y image coordinates of the marker are found by finding the center pixel of the marker in the image frame. An inner matrix K is used to transform from image coordinates to camera coordinates, and then a rotation matrix and translation vector are used to transform from camera coordinates to world coordinates to find the 3D position of the marker relative to the world origin, as shown in the following equation: P camera =D*[xy1]*K -1 P world =(P camera -t)*R -1 Other models can be used to determine the 3D location of a marker from 2D camera images. In other embodiments, 3D imaging, such as a depth camera, can be used.

[0063] Return to Figure 5 The template attenuation coefficient diagram 508 of the local coil is fitted 510 to the located marker. Any non-rigid fitting can be used, such as a fitting based on mutual information or intensity. This fitting will wind or fit the local coil into the 3D position. For hard local coils, a rigid fitting can be used.

[0064] The attenuation coefficient map can be integrated with a constrained fitting model. Using a mathematical model of the coil that imposes known constraints on possible deformation states, an attenuation coefficient map template is fitted to a marker point or location. In an alternative embodiment, the template attenuation coefficient map is fitted without being constrained by the model. In other embodiments, multiple templates are created at different orientations and deformation locations. The template that best matches the marker location is used for deformation of the marker or for further deformation to the marker. The best-fit location of the selected template provides the position. The end result is an attenuation or mutagenesis representation of the attenuation of a local coil, such as on or placed in 3D by a patient (Figure 516).

[0065] If the goodness of fit is below a predefined threshold, a data-driven PET approach based on consistency criteria can be used to improve the fit. PET data from PET scans is used in the data-driven fitting from PET data to template images 514 512.

[0066] In other approaches, the processor applies a machine learning model to output the distribution of marker positions or attenuation in 3D. Distribution information is output from the machine learning model in response to input optical information.

[0067] Figure 7An example is shown in which the machine learning model outputs a marker position 712. The 3D position 712 of the marker of the flexible local coil is predicted by a machine learning model that takes an image 710 captured by a receiving camera as input. Various machine learning models, such as neural networks, can be used. For example, a convolutional neural network (CNN) 708 or a fully connected neural network is trained using deep learning 706 to output the 3D spatial position 712 of the marker, taking into account the input optical image 710 (such as a normalized or filtered optical image). To train 706 this network 708, multiple pairs of ground-case marker positions 702 (e.g., determined automatically or manually from CT images) and corresponding 2D optical images 704 are used. In the prediction phase, when a patient is being scanned and no ground-case marker position is known, the marker position 712 is determined by the trained network 708 from the 2D optical image 710.

[0068] Such as about Figure 5 As discussed, a template attenuation coefficient map or a local coil model is fitted to the 3D location. The 3D location, output by the machine learning model, is used to fit the template attenuation coefficient map. For example, a mathematical model of a coil that imposes constraints on possible deformation states is fitted to the 3D location. Further improvements can be provided (e.g., PET consistency criterion methods), such as if the goodness of fit is below a threshold.

[0069] Figure 8 An example approach is shown where the attenuation coefficient map is adapted to the output of a machine learning model. Instead of outputting 3D locations, the machine learning model directly outputs a 3D distribution of attenuation fitted to the patient or adapted attenuation coefficient map 814 in response to receiving image 812 as input. Various machine learning models can be used, such as neural networks (e.g., CNN 808 or fully connected networks). CNN 808 creates the attenuation coefficient map 814 of the coils on the patient directly from the 2D optical image 812. Template attenuation coefficient map 810 and / or other inputs may or may not be provided to CNN 808 as input. During the training phase, multiple pairs of real-world attenuation images of the coils (e.g., CT image 802) and corresponding 2D optical images 804 are used. Each pair of training data inputs represents the coils on the patient table at different locations. Template attenuation coefficient map 810 may or may not be provided as input. The CNN is trained using deep learning 806. During the prediction phase, when a patient is to be scanned and no real-world attenuation image is known, a 2D optical image 812 is input to a CNN 808 with or without a template attenuation coefficient map 810 to deform the template attenuation image or to output the template attenuation image to a patient-specific instance, thereby providing the predicted attenuation coefficient map 814 in 3D.

[0070] Different machine learning models can be trained for different situations. For example, different machine learning models can be provided for different types of local coils. In this case, the template attenuation coefficient map 810 is not used. In another example, CNN808 is trained 806 to output attenuation coefficient maps 814 for different types of local coils. Therefore, the template attenuation coefficient map 810 for local coils for a given patient can be used as input, but does not have to be used.

[0071] Refer again Figure 4 In action 406, the PET imager uses attenuation correction to perform PET imaging. Attenuation correction includes attenuation correction for attenuation caused by MR hardware or local coils. Attenuation coefficient maps, adapted to placement (deformation, orientation, and / or positioning) on ​​the patient, are used in PET imaging. Attenuation correction may include attenuation correction factors specific to the patient tissue.

[0072] Acquire PET sine wave data. For example, acquire time-of-flight (TOF) data for emissions detected along multiple response lines. This acquisition is performed by scanning using a PET scanner. In an alternative embodiment, this acquisition is performed by transferring from or uploading from memory.

[0073] Gamma rays are detected by one or more detector rings or other sets of detectors. The patient ingests or is injected with a radiopharmaceutical. The radiopharmaceutical includes isotopes. The isotopes decay over time, resulting in the generation of positrons, which form gamma radiation, or photons. Response line events from the patient are detected. Acquisition occurs within any time period. For example, acquisition occurs within 1, 10, 100, or other numbers of minutes. The detected gamma rays are examined for overlap to define the response line, and the time difference or relative timing for overlap detection is recorded as cue data. Any time window can be used for overlap processing, such as a 0.2 ms overlap time window. Each detected emission event corresponds to a line or portion of a line passing through the patient. The volume can be reconstructed by detecting emission events from different angles surrounding the patient.

[0074] Reconstructing patient image objects from detected emission data targeting response lines. For example, reconstructing object space from PET TOF (such as from cue or sine wave data). Other PET detections and reconstructions can be used. Reconstructing the activity distribution in three dimensions. The activity distribution is used for imaging, such as volume rendering, multiplanar reconstruction, or planar imaging.

[0075] Any reconstruction can be used. In one embodiment, the reconstruction is a Poisson iterative reconstruction, such as maximum likelihood reconstruction. OSEM, FORE, or other reconstructions can be used. The reconstruction estimates the object or patient space from the response line. The detected events are used to iteratively determine the object space using forward, backward, or forward and backward projections.

[0076] Reconstruction includes attenuation correction. Some of the response circuitry passes through the MR hardware. The contribution of attenuation from the MR hardware is used in the reconstruction (i.e., at least partially based on a correction factor for the MR hardware). The 3D distribution of attenuation from the MR hardware is included in the attenuation correction factor for the response circuitry passing through the 3D distribution. The distribution of attenuation from the MR local coil is used in the attenuation correction for PET reconstruction. In one embodiment, attenuation correction is used to reconstruct at least a portion of the patient from the emission from the patient. An attenuation coefficient map for the MR hardware is included in the attenuation correction along with the attenuation for the portion of the patient.

[0077] In PET imaging, images are displayed. The activity or emission distribution reconstructed as image objects is used to create PET images. Images are generated by reconstructing the object space and then rendering or imaging from the reconstructed objects. Images are PET images of patients, such as those showing the function or absorption of radiopharmaceuticals. Images benefit from attenuation correction for MR hardware. This provides more accurate imaging that is less sensitive to noise.

[0078] Although the invention has been described above with reference to various embodiments, it should be understood that many changes and modifications can be made without departing from the scope of the invention. Therefore, it is intended that the above detailed description be considered illustrative rather than restrictive, and it should be understood that the appended claims, including all equivalents, are intended to define the spirit and scope of the invention.

Claims

1. A method for positron emission tomography (PET) imaging (402) taking into account attenuation performed by magnetic resonance MR hardware (110), the method comprising: The patient (116) is imaged (402) using an optical camera (112), the image (402) capturing images of the patient (116) and the MR hardware (110); Generate (404) an attenuation coefficient map relative to the patient (116) for the MR hardware (110) from the image; and The PET imaging (402) is performed (406) using attenuation correction based at least in part on the attenuation coefficient map for the MR hardware (110). The generation (404) includes: an attenuation coefficient map that is fitted to the patient (116) by the output of a machine learning model that receives the image as input.

2. The method of claim 1, wherein imaging (402) comprises: The patient (116) is imaged (402) outside the aperture of the combined PET and MR imaging (402) system using the camera (112), which is oriented toward the patient (116) bed of the combined PET and MR imaging (402) system.

3. The method of claim 2, wherein imaging (402) comprises: The patient (116) is imaged (402) while outside the aperture.

4. The method of claim 1, wherein imaging (402) includes imaging (402) one or more optical references on the MR hardware (110), and wherein generating (404) includes determining the position of the MR hardware (110) relative to the patient (116).

5. The method of claim 1, wherein generating the (404) attenuation coefficient map comprises: (404) Attenuation coefficient map of MR hardware (110) located for the patient (116) is generated from the template map of attenuation of the MR hardware (110).

6. The method of claim 5, wherein the MR hardware (110) includes a flexible local coil, and wherein generating (404) includes adapting a template to reflect the positioning of the flexible local coil on the patient (116).

7. The method of claim 6, wherein the adaptation includes: Identify the shape and position of the marker of the flexible local coil in the image; determine the three-dimensional position of the marker from the shape and position; And fitting the template to the three-dimensional position.

8. The method of claim 6, wherein the adaptation comprises: predicting the three-dimensional position of the marker of the flexible local coil by a machine learning model receiving the image as input; and fitting a template to the three-dimensional position.

9. The method of claim 1, wherein imaging (402) of the patient (116) using an optical camera (112) comprises imaging (402) the patient (116) using the optical camera (112) and at least one additional optical camera (112), and wherein generating (404) comprises generating (404) the attenuation coefficient map from the image and at least one additional image from the at least one additional optical camera (112).

10. The method of claim 1, wherein generating (404) comprises: One or more three-dimensional locations of the MR hardware (110) are identified from the image using color-coded markers and / or visible patterns on the MR hardware (110).

11. The method of claim 1, wherein performing (406) comprises: Reconstructing at least a portion of the patient (116) from the emission, the reconstruction using attenuation correction, wherein an attenuation coefficient map for the MR hardware (110) is included together with the attenuation for the portion of the patient (116).

12. A medical imaging (402) system, comprising: Positron emission tomography (102); A magnetic resonance imaging (MRI) device (104) is configured to image the patient (116) using one or more local coils located on the patient (116); A camera (112) is configured to capture images of the patient (116) and the one or more local coils located on the patient (116); as well as The processor (118) is configured to correct the attenuation of the one or more local coils based on one or more locations identified from the image, the correction being used for imaging (402) by the positron emission tomography imager (102). The processor (118) responds to the input of optical information by applying a machine learning model to output information on the distribution of attenuation in 3D.

13. The medical imaging (402) system of claim 12, wherein the positron emission tomography (PET) imager (102) and the magnetic resonance imaging (MRI) imager (104) share an aperture for imaging (402) the patient (116), wherein the camera (112) is mounted outside the aperture, and the image is captured when the patient (116) is mostly outside the aperture.

14. The medical imaging (402) system of claim 12, wherein the one or more local coils include flexible local coils, and wherein the processor (118) is configured to identify the one or more locations as three-dimensional locations based on markers on the one or more local coils.

15. The medical imaging (402) system of claim 14, wherein the processor (118) is configured to generate an attenuation coefficient map from a template attenuation coefficient map to the three-dimensional position.

16. The method of claim 1, taking into account attenuation caused by a local MR coil, the method comprising: Capture (402) an optical image of the MR local coil located on the patient (116); The distribution of attenuation from the MR local coil is determined (404) based on the optical image; and Reconstructing (406) PET images from emission, the reconstruction includes attenuation correction using the distribution of attenuation from the MR local coil.

17. The method of claim 16, wherein determining the (404) distribution comprises: The marker on the MR local coil is located based on the optical image; And fitting the attenuation template to the located marker.

18. The method of claim 16, wherein determining (404) comprises: The deformation and position of the MR local coil located on the patient (116) for MR imaging (402) are determined, and the attenuation distribution is based on the deformation and position of the MR local coil.

Images