Systems and methods for low-field MR / PET imaging

Abstract

A system and method for PET attenuation correction using low-field MR image data includes receiving a first image dataset and a low-field magnetic resonance (MR) image dataset. An attenuation correction map is generated from the low-field MR image data using a first trained neural network. At least one attenuation correction procedure is applied to the first image dataset based on the attenuation correction map to generate at least one clinically attenuated corrected image.

Description

Technical Field

[0001] This application generally relates to nuclear imaging, and more particularly to hybrid magnetic resonance (MR) and positron emission tomography (PET) imaging scanners. Background Technology

[0002] Multimodal imaging systems utilize multiple modalities to perform diagnostic scans, such as magnetic resonance imaging (MR / MRI), computed tomography (CT), positron emission tomography (PET), and / or single-photon emission computed tomography (SPECT). These modalities are combined to provide complementary and / or overlapping clinical information. For example, MR scans generally provide soft tissue morphological data and offer higher resolution for soft tissue structure and functional characteristics. PET scans generally have lower resolution but provide more useful information about the functional status of body tissues and systems, such as the cardiovascular system. PET scans are superior in indicating the presence of tumors or reduced blood flow to certain organs or regions of the body. By performing both methods in a single device and imaging session, the complementary advantages of two or more imaging modalities can be provided simultaneously.

[0003] Current hybrid (e.g., multimodal) imaging systems, such as PET / MR or PET / CT systems, are extremely expensive. For example, PET / MR imaging systems are typically only available (i.e., economically feasible) in research settings. Therefore, improvements that can reduce hardware system costs, as well as installation and / or operating costs, are desirable. Summary of the Invention

[0004] In some embodiments, a computer-implemented method is disclosed. The computer-implemented method includes the steps of: receiving a first image dataset and a low-field magnetic resonance (MR) image dataset; generating an attenuation correction map from the low-field MR image data using a first trained neural network; and applying at least one attenuation correction procedure to the first image dataset based on the attenuation correction map to generate at least one clinical attenuation-corrected image.

[0005] In some embodiments, a system is disclosed. The system includes a first imaging modality, a low-field MR imaging modality, and a computer. The computer is configured to receive a first image dataset from the first imaging modality and a low-field magnetic resonance (MR) image dataset from the low-field MR imaging modality; generate an attenuation correction map from the low-field MR image data using a first trained neural network; and apply at least one attenuation correction procedure to the first image dataset based on the attenuation correction map to generate at least one clinical attenuation-corrected image.

[0006] In some embodiments, a non-transitory computer-readable medium is disclosed. The non-transitory computer-readable medium stores instructions configured to cause a computer system to perform the following steps: receiving a first image dataset and a low-field magnetic resonance (MR) image dataset; generating an attenuation correction map from the low-field MR image data using a first trained neural network; and applying at least one attenuation correction procedure to the first image dataset based on the attenuation correction map to generate at least one clinical attenuation-corrected image. Attached Figure Description

[0007] The features and advantages of the present invention will be more fully disclosed or will become apparent in the following detailed description of preferred embodiments, which are to be described in conjunction with the appendix. Figure 1 The figures are taken into consideration, where similar figures refer to similar parts. The accompanying figures are schematic and are not intended to show actual dimensions or scale.

[0008] Figure 1 The illustration shows a nuclear imaging system according to some embodiments.

[0009] Figure 2 A block diagram of a computer system according to some embodiments is shown.

[0010] Figure 3 The illustration shows a multimodal imaging system according to some embodiments, which has concentric and linearly aligned PET imaging rings and MR imaging rings.

[0011] Figure 4 The illustration shows a multimodal imaging system with concentric and linearly aligned PET imaging rings and an open-aperture MR imaging system.

[0012] Figure 5 The illustration shows a multimodal imaging system with a PET imaging ring and a spatially separated MR imaging ring, according to some embodiments.

[0013] Figure 6 The illustration shows a multimodal imaging system with a PET imaging ring and a spatially separated open-aperture MR imaging system, according to some embodiments.

[0014] Figure 7 This is a flowchart illustrating a method for attenuation correction of PET image data using low-field MR image data according to some embodiments.

[0015] Figure 8 The illustration shows some embodiments. Figure 7 A flowchart illustrating the process of each step of the method.

[0016] Figure 9 An embodiment of an artificial neural network according to some embodiments is illustrated.

[0017] Figure 10This is a flowchart illustrating a method for attenuation correction of PET image data using low-field MR image data and a previous image dataset, according to some embodiments.

[0018] Figure 11 The illustration shows some embodiments. Figure 10 A flowchart illustrating the process of each step of the method. Detailed Implementation

[0019] This description of exemplary embodiments is intended to be read in conjunction with the accompanying drawings, which are to be considered an integral part of the entire written description. In the description, relative terms should be interpreted as referring to orientations as described therein or as shown in the drawings discussed. These relative terms are for ease of description and do not require the device to be constructed or operated in a particular orientation. Terms relating to attachment, coupling, and the like, such as “connection” and “interconnection,” refer to a relationship in which structures are directly or indirectly operatively connected or attached to each other through intermediate structures, including physical, electrical, optical, or other attachments or relationships, unless otherwise explicitly described.

[0020] In various embodiments, systems and methods for attenuation correction of image data (such as PET image data) using low-field MR image data are disclosed. At least one trained neural network is configured to receive low-field MR image data and spatially register the low-field MR image data with PET image data. The registered low-field MR image data is fed to a second trained neural network (or an additional hidden layer of a first neural network) to generate an attenuation (mu) map. The mu map is applied to the PET image data to perform attenuation correction. The attenuated PET image data can be output for storage and / or used to generate one or more clinical PET images.

[0021] Figure 1An embodiment of a nuclear imaging system 2 is illustrated. The nuclear imaging system 2 includes a scanner disposed in a first gantry 16a for at least a first modality 12. The first modality 12 may include any suitable modality, such as, for example, computed tomography (CT) modality, positron emission tomography (PET) modality, single-photon emission computed tomography (SPECT) modality, etc. The first modality 12 may include a long-axis field-of-view (FOV) scanner or a short-axis FOV scanner. A patient 17 may lie on a movable patient bed 18 movable relative to the first gantry 16a, and / or may lie on a fixed bed that maintains a fixed position while the gantry 16a moves relative to the patient bed 18. In some embodiments, the nuclear imaging system 2 includes a scanner disposed in a second gantry 16b for a second modality 14. The second modality 14 may be any suitable imaging modality, such as, for example, MR modality, CT modality, PET modality, SPECT modality, and / or any other suitable imaging modality. The second modality 14 may include a long-axis FOV scanner or a short-axis FOV scanner. Each of the first mode 12 and / or the second mode 14 may include one or more detectors 50 configured to detect annihilation photons, gamma rays, magnetic resonance and / or other nuclear imaging events.

[0022] Scan data from the first mode 12 and / or the second mode 14 are stored in one or more computer databases 40 and processed by one or more computer processors 60 of the computer system 30. Figure 1 The graphical depiction of computer system 30 is provided for illustrative purposes only, and computer system 30 may include one or more separate computing devices, such as those described above. Figure 2 As described. Scan data may be provided by the first mode 12, the second mode 14, and / or as a separate dataset, such as from a memory coupled to the computer system 30. The computer system 30 may include one or more processing electronics for processing signals received from one of the plurality of detectors 50.

[0023] Figure 2 The illustration depicts a computer system 30 configured to perform one or more processes according to some embodiments. System 30 is a representative device and may include a processor subsystem 72, an input / output subsystem 74, a memory subsystem 76, a communication interface 78, and a system bus 80. In some embodiments, one or more components of system 30 may be combined or omitted, such as, for example, omitting the input / output subsystem 74. In some embodiments, system 30 may include... Figure 2 Other components not shown. For example, system 30 may also include, for example, an electrical subsystem. In other embodiments, system 30 may include... Figure 2Examples of the components shown. For example, system 30 may include multiple memory subsystems 76. For the sake of brevity and clarity, and not limitation, Figure 2 One of each component is shown in the image.

[0024] The processor subsystem 72 may include any processing circuitry operable to control the operation and performance of the system 30. In various aspects, the processor subsystem 72 may be implemented as a general-purpose processor, a chip multiprocessor (CMP), a special-purpose processor, an embedded processor, a digital signal processor (DSP), a network processor, an input / output (I / O) processor, a media access control (MAC) processor, a radio baseband processor, a coprocessor, a microprocessor or other processing device such as a Complex Instruction Set Computer (CISC) microprocessor, a Reduced Instruction Set Computing (RISC) microprocessor, and / or a Very Long Instruction Word (VLIW) microprocessor. The processor subsystem 72 may also be implemented by a controller, a microcontroller, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a programmable logic device (PLD), and so on.

[0025] In various aspects, the processor subsystem 72 can be configured to run an operating system (OS) and various applications. Examples of OS include, for example, operating systems known by trade names such as Apple OS, Microsoft Windows OS, Android OS, and Linux OS, as well as any other proprietary or open-source OS. Examples of applications include, for example, network applications, local applications, data input / output applications, user interaction applications, etc.

[0026] In some embodiments, system 30 may include a system bus 80, which couples various system components including a processing subsystem 72, an input / output subsystem 74, and a memory subsystem 76. System bus 80 may be any of several types of bus architectures, including a memory bus or memory controller, a peripheral bus or external bus, and / or a local bus using various available bus architectures, including but not limited to a 9-bit bus, Industry Standard Architecture (ISA), Microchannel Architecture (MSA), Extended ISA (EISA), Intelligent Drive Electronics (IDE), VESA Local Bus (VLB), Peripheral Component Interconnect Card International Association Bus (PCMCIA), Small Computer Interface (SCSI) or other proprietary buses, or any custom bus suitable for computing device applications.

[0027] In some embodiments, the input / output subsystem 74 may include any suitable mechanism or component that enables a user to provide input to the system 30 and enables the system 30 to provide output to the user. For example, the input / output subsystem 74 may include any suitable input mechanism, including but not limited to buttons, keypads, keyboards, click wheels, touchscreens, motion sensors, microphones, cameras, etc.

[0028] In some embodiments, the input / output subsystem 74 may include a visual peripheral output device for providing a user-visible display. For example, the visual peripheral output device may include a screen, such as a liquid crystal display (LCD) screen. As another example, the visual peripheral output device may include a portable display or projection system for providing content display on a surface remote from system 30. In some embodiments, the visual peripheral output device may include an encoder / decoder (also referred to as a codec) for converting digital media data into analog signals. For example, the visual peripheral output device may include a video codec, an audio codec, or any other suitable type of codec.

[0029] Visual peripheral output devices may include a display driver, circuitry for driving the display driver, or both. A visual peripheral output device may be operable to display content under the direction of the processor subsystem 72. For example, a visual peripheral output device may be able to play media playback information, application screens of applications implemented on system 30, information about ongoing communication operations, information about incoming communication requests, or device operation screens, to name just a few.

[0030] In some embodiments, the communication interface 78 may include any suitable hardware, software, or a combination of hardware and software capable of coupling the system 30 to one or more networks and / or additional devices. The communication interface 78 may be configured to operate with any suitable technology for controlling information signals using a desired set of communication protocols, services, or operating procedures. The communication interface 78 may include appropriate physical connectors for connection to a corresponding communication medium, whether wired or wireless.

[0031] Communication tools include networks. In various aspects, networks can include local area networks (LANs) and wide area networks (WANs), including but not limited to the Internet, wired channels, wireless channels, communication devices including telephones, computers, wired, radio, optical, or other electromagnetic channels, and combinations thereof, including other devices and / or components capable of transmitting data and associated with data transmission. For example, communication environments include in-body communication, various devices, and various communication modes, such as wireless communication, wired communication, and combinations thereof.

[0032] Wireless communication modes include any communication mode between points (e.g., nodes) that at least partially utilize wireless technology, which includes various protocols and combinations of protocols associated with wireless transmissions, data, and devices. These points include, for example, wireless devices (such as wireless headsets), audio and multimedia devices and equipment (such as audio players and multimedia players), telephones (including mobile phones and cordless phones), and computers and computer-related devices and components (such as printers, network-connected machines), and / or any other suitable devices or third-party devices.

[0033] Wired communication modes encompass any communication mode between points utilizing wired technology, which includes various protocols and combinations of protocols associated with wired transmission, data, and devices. These points include, for example, devices such as audio and multimedia equipment and apparatus (e.g., audio players and multimedia players), telephones (including mobile phones and cordless phones), and computers and computer-related devices and components (e.g., printers, network-connected machines), and / or any other suitable devices or third-party devices. In various implementations, the wired communication module can communicate according to multiple wired protocols. Examples of wired protocols can include Universal Serial Bus (USB) communication, RS-232, RS-422, RS-423, RS-485 serial protocols, FireWire, Ethernet, Fibre Channel, MIDI, ATA, Serial ATA, PCI Express, T-1 (and variants), Industry Standard Architecture (ISA) parallel communication, Small Computer System Interface (SCSI) communication, or Peripheral Component Interconnect (PCI) communication, to name just a few.

[0034] Therefore, in various aspects, communication interface 78 may include one or more interfaces, such as, for example, wireless communication interface, wired communication interface, network interface, transmit interface, receive interface, media interface, system interface, component interface, switching interface, chip interface, controller, etc. For example, when implemented by a wireless device or within a wireless system, communication interface 78 may include a wireless interface that includes one or more antennas, transmitters, receivers, transceivers, amplifiers, filters, control logic, etc.

[0035] In various aspects, the communication interface 78 can provide data communication functionality according to multiple protocols. Examples of protocols may include various Wireless Local Area Network (WLAN) protocols, including the Institute of Electrical and Electronics Engineers (IEEE) 802.xx series protocols, such as IEEE 802.11a / b / g / n / ac, IEEE 802.16, IEEE 80.20, and so on. Other examples of wireless protocols may include various Wireless Wide Area Network (WWAN) protocols, such as GSM cellular wirelessphone system protocols with GPRS, CDMA cellular wirelessphone communication systems with 1xRTT, EDGE systems, EV-DO systems, EV-DV systems, HSDPA systems, and so on. Further examples of wireless protocols may include Wireless Personal Area Network (PAN) protocols, such as infrared protocols, protocols from the Bluetooth Special Interest Group (SIG) series of protocols (e.g., Bluetooth specification versions 5.0, 6, 7, legacy Bluetooth protocols, etc.), and one or more Bluetooth profiles, etc. Yet another example of wireless protocols may include near-field communication technologies and protocols, such as electromagnetic induction (EMI) technology. Examples of EMI technologies can include passive or active radio frequency identification (RFID) protocols and devices. Other suitable protocols may include ultra-wideband (UWB), digital office (DO), digital home, trusted platform module (TPM), ZigBee, and so on.

[0036] In some embodiments, at least one non-transitory computer-readable storage medium is provided, having embodied computer-executable instructions thereon, wherein, when executed by at least one processor, the computer-executable instructions cause at least one processor to perform embodiments of the methods described herein. This computer-readable storage medium may be embodied in a memory subsystem 76.

[0037] In some embodiments, the memory subsystem 76 may include any machine-readable or computer-readable medium capable of storing data, including both volatile / non-volatile memory and removable / non-removable memory. The memory subsystem 76 may include at least one non-volatile memory cell. The non-volatile memory cell is capable of storing one or more software programs. The software program may contain, for example, applications, user data, device data and / or configuration data, or combinations thereof, to name just a few. The software program may contain instructions executable by various components of system 30.

[0038] In various aspects, the memory subsystem 76 may include any machine-readable or computer-readable medium capable of storing data, including both volatile / non-volatile memory and removable / non-removable memory. For example, the memory may include read-only memory (ROM), random access memory (RAM), dynamic RAM (DRAM), double data rate DRAM (DDR-RAM), synchronous DRAM (SDRAM), static RAM (SRAM), programmable ROM (PROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory (e.g., NOR or NAND flash memory), content-addressable memory (CAM), polymer memory (e.g., ferroelectric polymer memory), phase-change memory (e.g., austenite memory), ferroelectric memory, silicon-oxide-nitride-oxide-silicon (SONOS) memory, disk storage (e.g., floppy disk, hard disk, optical disk, magnetic disk), or card (e.g., magnetic card, optical card), or any other type of medium suitable for storing information.

[0039] In one embodiment, the memory subsystem 76 may contain an instruction set in the form of a file for performing various methods, such as those described herein, including A / B testing and cache optimization. The instruction set may be stored in any acceptable machine-readable instruction form, including source code or various suitable programming languages. Some examples of programming languages ​​that may be used to store the instruction set include, but are not limited to, Java, C, C++, C#, Python, Objective-C, Visual Basic, or .NET programming. In some embodiments, a compiler or interpreter is included to translate the instruction set into machine-executable code for execution by the processing subsystem 72.

[0040] Figure 3-6 The illustration shows a multimodal imaging system (such as the one mentioned above). Figure 1 Various embodiments of the modal arrangements 90a-90d of the described imaging system. Each modal arrangement includes a first imaging mode 12 (such as a PET imaging mode) and a second imaging mode 14a-14d (such as a low-field MR imaging mode). Although the embodiments discussed herein include a PET imaging mode, it will be appreciated that the first imaging mode can be any suitable single imaging mode, such as PET, SPECT, CT, etc., and / or any suitable hybrid imaging mode, such as a PET / CT scanner, a SPECT / PET scanner, etc.

[0041] Low-field MR imaging systems include magnetic resonance imaging systems that have field strengths lower than those currently used in clinical settings. For example, in various embodiments, low-field MR modes may have field strengths of less than about 0.2T to about 1T, although it will be appreciated that any suitable field strength less than typical clinical field strengths may be applied, such as, for example, any field strength less than about 1.5T, less than about 3T, etc.

[0042] Using low-field MR imaging modalities 14a-14d allows for a reduction in the initial cost of multimodal imaging systems. For example, low-field MR imaging modalities 14a-14d have lower equipment costs (e.g., lack of dedicated cooling mechanisms due to lower field strength) and lower facility / installation cost requirements (e.g., low-field MR imaging modalities 14a-14d do not require the indoor shielding necessary for conventional MR imaging modalities; lower space requirements due to their smaller footprint compared to conventional MR imaging systems, etc.). Furthermore, low-field MR imaging modalities 14a-14d have lower operating costs. For example, conventional MR imaging modalities require the use of coolers or cryogenic cooling equipment, while low-field MR imaging modalities 14a-14d can forgo such cooling mechanisms.

[0043] The use of low-field MR allows for the use of open-design low-field MR imaging modalities 14b, 14d (e.g., as used in modal arrangements 90b, 90d, and as described below). Figure 4 and 6 (As discussed). Open-field MR imaging modalities 14b and 14d allow patient access during MR imaging for interventional and / or radiotherapy procedures, increasing patient comfort, clinical access, and patient outcomes. The lower field intensity of low-field MR imaging modalities 14a-14d provides lower energy deposition (e.g., specific absorption rate (SAR)) in patient tissues. Low-field MR imaging modalities 14a-14d offer reductions in chemical displacement, susceptibility, and flow / motion artifacts, and further provide improved imaging when used in conjunction with metallic implants.

[0044] As discussed below, using low-field MR imaging modalities 14a-14d eliminates the need for attenuation correction using other imaging modalities. For example, in some embodiments, low-field MR imaging modalities 14a-14d can be used instead of CT imaging modalities to provide a second image dataset for attenuation correction of PET image data captured using PET imaging modalities. Compared to CT imaging modalities, using MR imaging modalities eliminates the need for patients to receive CT radiation doses.

[0045] In some embodiments, the low-field MR imaging modality may be primarily suited for generating mutagrams that attenuate-corrected for PET image data. For example, in some embodiments, the magnetic field strength, gradient strength, receiver coil parameters (e.g., material, number, location, etc.) and / or low-field MR imaging modality parameters may be reduced to a level insufficient to generate clinically usable images but sufficient to generate mutagrams. Each simplified parameter may be further optimized independently or collectively for mutagram generation. In some embodiments, the sampling efficiency of the low-field MR imaging modality may be increased by performing longer scans (at lower field strengths) compared to conventional MR imaging modalities.

[0046] In some embodiments, the receiving elements in the low-field MR imaging modality (e.g., RF coil architecture, magnetic field gradient design, etc.) combined with the process disclosed herein using improved image reconstruction algorithms allow for acceptable image quality for PET image data attenuation correction using the low-field MR imaging modality (and associated low field intensity). Compared to conventional MR imaging modalities, the use of multiple receiving elements in the low-field MR imaging modality provides more efficient sampling through parallel imaging and simultaneous multi-slice (SMS) acquisition. Sampling in the low-field MR imaging modality can be further improved using compressed sensing techniques and higher bandwidth-time product RF pulses (because RF power deposition is proportional to the square of the magnetic field of the MR imaging modality).

[0047] Figure 3 The illustration shows a modal arrangement 90a including a PET imaging modality 12 and an MR imaging modality 14a arranged in a "washer-dryer" configuration, wherein the apertures of the PET imaging modality 12 and the apertures of the MR imaging modality 14a are coaxially aligned, and the imaging modalities 12 and 14a are spatially adjacent and in contact, allowing the patient to traverse each of the imaging modalities simultaneously (or sequentially) without moving the patient bed 18 and / or repositioning the patient 17.

[0048] Figure 4 The diagram illustrates the relationship between... Figure 3 A similar modal arrangement 90b is illustrated, but the low-field MR imaging mode 14b is an open-aperture imaging mode. Open-aperture imaging modes provide lower field intensity and lower image quality. Although the image quality of the low-field open-aperture MR imaging mode 14b may be insufficient for diagnostic imaging, it provides adequate quality for attenuation correction and registration of another imaging mode, such as PET imaging mode 12.

[0049] Figure 5 and 6The illustration shows modal arrangements 90c-90d, where PET imaging modality 12 and MR imaging modalities 14c-14d are spatially separated. Spatially separating imaging modalities 12, 14c-14d allows the patient to be imaged first in a first imaging modality (such as PET imaging modality 12) and secondly in a second imaging modality (such as MR imaging modalities 14c-14d). In other embodiments, separate imaging modalities can be used in locations where space is scarce, where separate imaging modalities are switched for various imaging procedures within the same space.

[0050] In some embodiments, low-field MR imaging modalities—such as Figure 3-6 One of the low-field MR imaging modalities 14a-14d illustrated in the figure is configured to provide a low-field MR image dataset for attenuation correction of PET image data obtained from PET imaging modal 12. Figure 7 This is a flowchart 200 illustrating a method for attenuation correction of PET image data using low-field MR image data according to some embodiments. Figure 8 The illustration shows some embodiments. Figure 7 The process diagram for each step of the method is shown in Figure 250. Any of the modal arrangements 90a-90d can be used to implement any of the following methods and / or processes.

[0051] At step 202, a first image dataset (such as PET image data 252) is obtained using a first imaging modality 12 (such as, for example, a PET imaging modality). At step 204, a low-field MR image dataset is obtained using low-field MR imaging modalities 14a-14d. PET image data 252 and / or low-field MR image data 254 can be obtained directly from the corresponding imaging modalities 12, 14a-14d, and / or can be stored in and obtained from non-transient storage (such as, for example, a computer database 40).

[0052] At step 206, image registration module 256 performs image registration to align (e.g., spatially align) the low-field MR image data with the PET image data. Image registration module 256 generates a set of PET spatially registered low-field MR image data 258. Image registration module 256 may include a trained neural network and / or a conventional image registration algorithm. For example, in some embodiments, the trained neural network may include a neural network trained using a training dataset, including pre-registration and / or alignment datasets. The trained neural network may be configured to apply any suitable image registration process, such as, for example, anatomical landmarks, deep reinforcement learning processes, image synthesis using generative adversarial networks (independently and / or followed by conventional registration), unsupervised learning methods, and / or any other suitable process.

[0053] In some embodiments, a neural network is trained using one or more training methods configured to compensate for gradient nonlinearity and magnetic inhomogeneity in low-field MR imaging modalities. For example, in some embodiments, gradient and magnetic nonlinearity can be directly incorporated into the reconstruction process (as part of the image registration process), such as by using MR fingerprinting to model inhomogeneity and mitigating the need for uniform gradient, RF, and B0 fields. Compensation for gradient and / or magnetic nonlinearity provides a reduction in the dictionary and higher efficiency at lower magnetic field strengths. Embodiments including trained neural networks and / or conventional image registration algorithms can be used with any of the foregoing embodiments and / or any of the following embodiments.

[0054] Figure 9 An embodiment of an artificial neural network 100 according to some embodiments is shown. Alternative terms for "artificial neural network" are "neural network," "artificial neural network," "neural network," or "trained function." The artificial neural network 100 includes nodes 120-132 and edges 140-142, wherein each edge 140-142 is a directed connection from a first node 120-132 to a second node 120-132. Generally, the first node 120-132 and the second node 120-132 are different nodes 120-132, although the first node 120-132 and the second node 120-132 may also be the same. For example, in... Figure 2 In the diagram, edge 140 is a directed connection from node 120 to node 123, while edge 142 is a directed connection from node 130 to node 132. The edge 140-142 from the first node 120-132 to the second node 120-132 is also labeled as the "incoming edge" of the second node 120-132 and the "outgoing edge" of the first node 120-132.

[0055] In this embodiment, nodes 120-132 of the artificial neural network 100 can be arranged in layers 110-113, wherein the layers can include an inherent order introduced by edges 140-142 between nodes 120-132. Specifically, edges 140-142 may exist only between adjacent layers of nodes. In the illustrated embodiment, there is an input layer 110 containing only nodes 120-122 without incoming edges, an output layer 113 containing only nodes 131 and 132 without outgoing edges, and hidden layers 111 and 112 between the input layer 110 and the output layer 113. Generally, the number of hidden layers 111 and 112 can be arbitrarily chosen. The number of nodes 120-122 in the input layer 110 is typically related to the number of input values ​​of the neural network, while the number of nodes 131 and 132 in the output layer 113 is typically related to the number of output values ​​of the neural network 500.

[0056] Specifically, (real) numbers can be assigned as values ​​to each node 120-132 of the neural network 100. Here, x (n) i The values ​​of nodes 120-132 in the nth layer (110-113) are labeled. The values ​​of nodes 120-122 in the input layer 110 are equivalent to the input values ​​of the neural network 100, and the values ​​of nodes 131 and 132 in the output layer 113 are equivalent to the output values ​​of the neural network 100. Furthermore, each edge 140-142 may include a weight as a real number; specifically, the weight is a real number within the interval [-1, 1] or the interval [0, 1]. Here, w... (m,n) i,j This indicates the weight of the edge between the i-th node (120-132) of layer m (110-113) and the j-th node (120-132) of layer n (110-113). Furthermore, let w be the weight. (n,n+1) i,j Define the abbreviation w (n) i,j .

[0057] Specifically, the input values ​​are propagated through a neural network to calculate the output value of neural network 100. Specifically, the values ​​of nodes 120-132 in the (n+1)th layer 110-113 can be calculated based on the values ​​of nodes 120-132 in the nth layer 110-113 using the following formula:

[0058]

[0059] In this paper, the function f is the transfer function (another term is the "activation function"). The transfer function is known to be a step function, a sigmoid function (e.g., the logistic function, the generalized logistic function, the hyperbolic tangent, the arctangent function, the error function, the smoothing step size function), or a correction function. The transfer function is primarily used for normalization purposes.

[0060] Specifically, these values ​​are propagated layer by layer through the neural network, where the value of input layer 110 is given by the input of neural network 100, the value of first hidden layer 111 can be calculated based on the value of input layer 110 of the neural network, the value of second hidden layer 112 can be calculated based on the value of first hidden layer 111, and so on.

[0061] To set the value w of the edge (m,n) i,j The neural network 100 must be trained using training data. Specifically, the training data includes training input data and training output data (denoted as t). i For the training step, the neural network 100 is applied to the training input data to generate computed output data. Specifically, the training data and the computed output data include multiple values, the number of which is equal to the number of nodes in the output layer.

[0062] Specifically, the comparison between the calculated output data and the training data is used to recursively adapt the weights within the neural network 100 (backpropagation algorithm). Specifically, the weights change according to the following formula:

[0063]

[0064] Where γ is the learning rate, and the number δ(n) j It can be based on δ (n+1) j The recursive calculation is as follows:

[0065]

[0066] If the (n+1)th layer is not an output layer, and

[0067]

[0068] If the (n+1)th layer is the output layer 113, then f' is the first derivative of the activation function, and y (n+1) j It is the comparison training value of the j-th node of the output layer 113.

[0069] In some embodiments, at optional step 208, the PET spatially registered MR image dataset is output for one or more future reconstructions, for training or retraining a neural network, and / or for any other suitable process. For example, in some embodiments, the low-field MR image data lacks the definition required to generate clinical images, but can be used to assist in generating other clinical images, such as generating mu maps for attenuation correction of PET image data, as discussed below. In other embodiments, the low-field MR image data can be associated with clinician-approved and labeled reconstructed images—such as attenuation-corrected PET reconstructed images with one or more anomalies labeled by the clinician. The labeled low-field MR images can be used to further refine or train a neural network configured to provide image registration using one or more training or reinforcement processes, as discussed above. Optional outputs of the PET spatially registered MR image data can be integrated into any of the foregoing embodiments or any of the following embodiments.

[0070] At step 210, a mu map 262 is generated from the set of low-field MR image data 258 spatially registered in PET space via a mu map generation process 260. The mu map 262 can be generated using a second trained neural network. In some embodiments, the mu map generation process 260 is configured to apply a maximum likelihood reconstruction (MLAA) algorithm for activity and attenuation, although it will be understood that the mu map 262 can be generated according to any one or more suitable attenuation map generation processes. In some embodiments, the trained neural network can utilize templates, spectral information, direct segmentation of MR data, and / or segmentation of images to generate mu maps. The trained neural network and / or conventional mu map generation processes can be used with any of the foregoing embodiments and / or any of the following embodiments.

[0071] In some embodiments, a single trained network may be configured to perform image registration and mutagram generation simultaneously and / or sequentially. For example, in some embodiments, a first set of hidden layers may be configured to perform registration of low-field MR image data with PET image data, and a second set of hidden layers may be configured to generate mutagrams from low-field MR image data registered in PET space. In some embodiments, a single set of hidden layers performs image registration and mutagram generation simultaneously. A single trained neural network may be used with any of the foregoing embodiments and / or any of the following embodiments.

[0072] At step 212, the generated mu image 262 is used for attenuation correction 264 of the PET image data 252 to generate at least one attenuated PET image 266. Attenuation correction can be performed using any suitable procedure, such as, for example, one or more known attenuation correction procedures. In some embodiments, a neural network can be trained to perform attenuation correction based on the generated mu image 262. It will be appreciated that any suitable attenuation correction procedure 264 can be applied to correct the PET image data 252. The attenuation correction procedure 264 generates at least one attenuated PET image 266 and / or an attenuated PET image dataset. At step 214, at least one attenuated PET image 266 and / or the attenuated PET image dataset can be output for storage on a storage device (such as a computer database 40) for later retrieval and use, and / or can be provided for clinical and / or diagnostic procedures.

[0073] Figure 10 This is a flowchart 200a illustrating a method for attenuation correction of PET image data using low-field MR image data and previous image data according to some embodiments. Figure 11 The illustration shows some embodiments. Figure 10 The process flow 250, flowchart 200a, and process flow 250a for each step of the method are similar to those described above. Figure 7-8The flowchart 200 and process flow 250 discussed herein, and similar descriptions, are not repeated herein. At step 218, a collection of previous image data 270 is received. The collection of previous (e.g., existing) image data 270 may include, for example, conventional MR image data and / or CT image data of a patient generated during one or more previous imaging studies. The previous image dataset may be received from a storage device, such as, for example, a computer database 40.

[0074] At step 206a, the set of low-field MR image data 254 and previous image data 270 is registered to the image space of PET image data 252 by a first trained neural network 256a. The first neural network 256a can be configured to perform the registration of the low-field MR image data 254 with the previous image dataset in any suitable order. For example, in some embodiments, the set of low-field MR image data 254 can be registered to the image space of the set of previous image data 270 in a first step, and the image space of the previous image dataset can be registered to the image space of the set of PET image data 252 in a second step. As another example, in some embodiments, the set of previous image data 270 can be registered to the image space of the set of low-field MR image data 254 in a first step, and the image space of the set of low-field MR image data 254 can be registered to the image space of the set of PET image data 252 in a second step. After registering the low-field MR image data 254 and the previous image data 270, process 200a continues with a process 200 similar to that discussed above. The second trained neural network 260a is configured to use each of the sets of PET spatially registered low-field MR data 258 and PET spatially registered MR / CT (e.g., previous) data 272. It will be appreciated that registration using previous image data can be included in any of the foregoing and / or following embodiments.

[0075] At optional step 220, a set of previous image data 272 registered to PET space is output. The set of previous image data 272 may include conventional MR image data and / or CT image data registered to the image space of a set of PET image data 252. In some embodiments, the set of previous image data 272 registered to PET space is used to generate one or more clinical images in the image space of the PET image data 252 for comparison with and / or in combination with the reconstructed PET images generated at step 214.

[0076] In a first embodiment, a computer-implemented method is disclosed. The computer-implemented method includes the steps of: receiving a first image dataset and a low-field magnetic resonance (MR) image dataset; generating an attenuation correction map from the low-field MR image data using a first trained neural network; applying at least one attenuation correction procedure to the first image dataset based on the attenuation correction map to generate an attenuated-corrected image dataset; and generating at least one clinical image from the attenuated-corrected image dataset.

[0077] In any subsequent embodiment, the computer-implemented method of the first embodiment may further include the step of registering the low-field MR image dataset to the image space of the first image dataset before generating the attenuation correction map. A second trained neural network is configured to register the low-field MR image dataset to the image space of the first image dataset.

[0078] In any subsequent embodiment, the computer-implemented method of the first embodiment may further include the steps of: receiving a previous image dataset not obtained simultaneously with the first image dataset and the low-field MR image dataset, and, before registering the low-field MR image dataset to the image space of the first image dataset, registering the image space of the first of the previous image dataset or the low-field MR image dataset to the image space of the second of the previous image dataset or the low-field MR image dataset. Generating an attenuation correction map from each of the low-field MR image data and the previous image dataset. The previous image dataset includes either conventional MR image data or computed tomography (CT) image data.

[0079] In any subsequent embodiment, the first image dataset includes a positron emission tomography (PET) image dataset and / or a single-photon emission computed tomography (SPECT) image dataset. The low-field MR image dataset is generated from a low-field MR imaging mode having a field intensity of less than 1 Tesla. The low-field MR image dataset is generated from one of an open-aperture low-field MR imaging mode or a closed-aperture low-field MR imaging mode. The low-field MR image dataset is generated from one of a low-field MR imaging mode coaxially aligned and spatially adjacent to the first imaging mode configured to generate the first image dataset, or a low-field MR imaging mode spatially separated from the first imaging mode.

[0080] In any subsequent embodiment, the first trained neural network is trained using a training dataset comprising MR image data pre-registered to the corresponding PET image space, and is trained using at least one of anatomical landmarks, deep reinforcement learning, image synthesis using generative adversarial networks, unsupervised learning, a combination thereof, or any other neural network architecture.

[0081] In a second embodiment, a system is disclosed. The system includes a first imaging mode, a low-field MR imaging mode, and a computer. The computer is configured to implement the computer-implemented method of the first embodiment.

[0082] In a third embodiment, a non-transitory computer-readable medium storing instructions configured to cause a computer system to perform steps of a first embodiment of a computer-implemented method is disclosed.

[0083] In a fourth embodiment, a computer-implemented method is disclosed. This computer-implemented method includes the steps of receiving a first image dataset, a low-field magnetic resonance (MR) image dataset, and a previous imaging dataset. The previous image dataset includes image data acquired not simultaneously with the first image dataset and the low-field MR image dataset; generating an attenuation correction map from the low-field MR image dataset and the previous image dataset using a first trained neural network; applying at least one attenuation correction procedure to the first image dataset based on the attenuation correction map to generate an attenuated-corrected image dataset; and generating at least one clinical image from the attenuated-corrected image dataset.

[0084] In any subsequent embodiment, the computer-implemented method of the fourth embodiment may further include the step of registering the image space of a first image dataset or a low-field MR image dataset to the image space of a second image dataset or a previous image dataset. Subsequently, the image space of the low-field MR image dataset and the image space of the previous image data are registered to the image space of the first image dataset. Registration is performed before the second neural network generates the attenuation correction map.

[0085] In any subsequent embodiment, the first image dataset includes a positron emission tomography (PET) image dataset and / or a single-photon emission computed tomography (SPECT) image dataset. The low-field MR image dataset is generated from a low-field MR imaging modality with a field intensity of less than 1 Tesla. The low-field MR image dataset is generated from one of an open-aperture low-field MR imaging modality or a closed-aperture low-field MR imaging modality. The low-field MR image dataset is generated from one of a low-field MR imaging modality coaxially aligned with and spatially adjacent to the first imaging modality configured to generate the first image dataset, or a low-field MR imaging modality spatially separated from the first imaging modality. The prior image dataset includes one of conventional MR image data or computed tomography (CT) image data.

[0086] In any subsequent embodiment, the first trained neural network is trained using a training dataset comprising MR image data pre-registered to the corresponding PET image space, and is trained using at least one of anatomical landmarks, deep reinforcement learning, image synthesis using generative adversarial networks, unsupervised learning, a combination thereof, or any other neural network architecture.

[0087] In a fifth embodiment, a system is disclosed. The system includes a first imaging mode, a low-field MR imaging mode, and a computer. The computer is configured to implement the computer-implemented method of the fourth embodiment.

[0088] In a sixth embodiment, a non-transitory computer-readable medium storing instructions configured to cause a computer system to perform the steps of a fourth embodiment of a computer-implemented method is disclosed.

[0089] Although the subject matter has been described with reference to exemplary embodiments, it is not limited thereto. Rather, the appended claims should be interpreted broadly to include other variations and embodiments that may be made by those skilled in the art.

Claims

1. A computer-implemented method, comprising: Receive the first image dataset and the low-field magnetic resonance (MR) image dataset; Attenuation correction maps are generated from low-field MR image data using a first trained neural network, wherein the first trained neural network is trained to compensate for gradient nonlinearity and magnet inhomogeneity in the low-field MR imaging modality. At least one attenuation correction process is applied to the first image dataset based on the attenuation correction map to generate at least one clinical attenuation-corrected image.

2. The computer-implemented method of claim 1, comprising registering a low-field MR image dataset to the image space of a first image dataset before generating an attenuation correction map.

3. The computer-implemented method of claim 2, wherein the second trained neural network is configured to register a low-field MR image dataset to the image space of the first image dataset.

4. The computer-implemented method according to claim 2, comprising: Receive a previous image dataset that was not obtained simultaneously with the first image dataset and the low-field MR image dataset; Before registering the low-field MR image dataset to the image space of the first image dataset, the image space of the first of the previous image dataset or the low-field MR image dataset is registered to the image space of the second of the previous image dataset or the low-field MR image dataset, and the attenuation correction map is generated from each of the low-field MR image data and the previous image dataset.

5. The computer-implemented method of claim 4, wherein the prior image dataset comprises either conventional MR image data or computed tomography (CT) image data.

6. The computer-implemented method of claim 1, wherein the first image dataset comprises a positron emission tomography (PET) image dataset.

7. The computer-implemented method of claim 1, wherein the first image dataset comprises a single-photon emission computed tomography (SPECT) image dataset.

8. The computer-implemented method of claim 1, wherein the low-field MR image dataset is generated by a low-field MR imaging modality having at least one parameter optimized to obtain image data for mu image generation.

9. The computer-implemented method of claim 1, wherein the low-field MR image dataset is generated from a low-field MR imaging mode having a field intensity of less than 1 Tesla.

10. The computer-implemented method of claim 1, wherein the low-field MR image dataset is generated from an aperture-based low-field MR imaging modality.

11. The computer-implemented method of claim 1, wherein the low-field MR image dataset is generated by a low-field MR imaging mode that is coaxially aligned with and spatially adjacent to a first imaging mode configured to generate the first image dataset.

12. The computer-implemented method of claim 1, wherein the first trained neural network is trained using a training dataset comprising MR image data pre-registered to the corresponding PET image space.

13. The computer-implemented method of claim 10, wherein the first trained neural network is trained using at least one of anatomical landmarks, deep reinforcement learning, image synthesis using generative adversarial networks, unsupervised learning, or a combination thereof.

14. A system comprising: First imaging mode; Low-field MR imaging modality; and Computer, the computer is configured to: Receive a first image dataset from a first imaging mode, and receive a low-field magnetic resonance (MR) image dataset from a low-field MR imaging mode; Attenuation correction maps are generated from low-field MR image data using a first trained neural network, wherein the first trained neural network is trained to compensate for gradient nonlinearity and magnet inhomogeneity in the low-field MR imaging modality. At least one attenuation correction process is applied to the first image dataset based on the attenuation correction map to generate at least one clinical attenuation-corrected image.

15. The system of claim 14, wherein the computer is further configured to register the low-field MR image dataset to the image space of the first image dataset before generating the attenuation correction map.

16. The system of claim 15, wherein the second trained neural network is configured to register the low-field MR image dataset to the image space of the first image dataset.

17. The system of claim 14, wherein the computer is configured to: Receive a previous image dataset that was not obtained simultaneously with the first image dataset and the low-field MR image dataset; Before registering the low-field MR image dataset to the image space of the first image dataset, the image space of the first of the previous image dataset or the low-field MR image dataset is registered to the image space of the second of the previous image dataset or the low-field MR image dataset, and the attenuation correction map is generated from each of the low-field MR image data and the previous image dataset.

18. The system of claim 14, wherein the first image dataset comprises one of a positron emission tomography (PET) image dataset and a single-photon emission computed tomography (SPECT) image dataset.

19. The system of claim 14, wherein the low-field MR image dataset is generated by a low-field MR imaging modality having at least one parameter optimized to obtain image data for mu map generation.

20. A non-transitory computer-readable medium storing instructions configured to cause a computer system to perform the following steps: Receive the first image dataset and the low-field magnetic resonance (MR) image dataset; Attenuation correction maps are generated from low-field MR image data using a first trained neural network, wherein the first trained neural network is trained to compensate for gradient nonlinearity and magnet inhomogeneity in the low-field MR imaging modality. At least one attenuation correction process is applied to the first image dataset based on the attenuation correction map to generate at least one clinical attenuation-corrected image.

Images