Deep learning super-resolution training for extremely low-field magnetic resonance imaging

Abstract

The present disclosure provides systems and methods for deep learning super-resolution training and / or image generation for low-field and ultra-low-field magnetic resonance imaging. In some aspects, the method includes acquiring a first image of a brain using a low-field strength magnetic resonance imaging system. The first image has a first resolution. The method further includes acquiring a deep learning brain model based on the high-field strength image. The deep learning brain model can be configured to be applied by a neural network comprising multiple layers. The method further includes applying the deep learning brain model to the first image to generate a second image of the brain. The second image has a second resolution, the second resolution being greater than the first resolution.

Description

[Background technology]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This application claims the benefit of and priority under 35 U.S.C. Section 120 to U.S. patent application Ser. No. 18 / 147,556, filed December 28, 2022, and entitled "DEEP LEARNING SUPER-RESOLUTION TRAINING FOR ULTRA LOW-FIELD MAGNETIC RESONANCE IMAGING," the disclosure of which is incorporated herein by reference in its entirety.

[0002] This disclosure relates to magnetic resonance imaging (MRI), medical imaging, medical intervention, and surgical intervention. MRI systems often involve large, complex machines that generate significantly high magnetic fields, resulting in significant constraints on the feasibility of certain surgical interventions. Limitations may include limited physical access to the patient by the surgeon and / or surgical robot and / or limitations on the use of certain electrical and mechanical components in the vicinity of the MRI scanning device. Such limitations are inherent in the underlying designs of many existing systems and are difficult to overcome. Summary of the Invention [Means for solving the problem]

[0003] In one aspect, the present disclosure describes a method. The method can include acquiring a first image of a brain using a low-field strength magnetic resonance imaging system. The first image can have a first resolution. The method can further include acquiring a deep learning brain model based on the high-field strength image. The deep learning brain model can be configured to be applied by a neural network comprising multiple layers. The method can further include applying the deep learning brain model to the first image to generate a second image of the brain. The second image can have a second resolution. The second resolution is greater than the first resolution.

[0004] In another aspect, the present disclosure describes a system. The system includes a processor and a memory. The memory can store machine-readable instructions. The processor can be configured to execute the machine-readable instructions. The machine-readable instructions, when executed, can implement a neural network. The neural network can be configured to obtain a high-field strength magnetic resonance model comprising multiple layers and to receive data representing a low-field image. The neural network can be further configured to transform the low-field image into a higher-resolution image based on the high-field strength magnetic resonance model and output the higher-resolution image. [Brief explanation of the drawings]

[0005] The various aspects described herein, both as to organization and method of operation, together with further objects and advantages thereof, may best be understood by reference to the following description taken in conjunction with the accompanying drawings, in which:

[0006] [Figure 1] FIG. 1 depicts components of an MRI scanning system according to at least one aspect of the present disclosure, including a dome-shaped housing for a magnetic array, the dome-shaped housing enclosing a region of interest therein, and further depicts the dome-shaped housing positioned to receive at least a portion of a patient's head reclined on a table within the region of interest.

[0007] [Figure 1A] FIG. 1A depicts a patient's head positioned within the region of interest of the MRI scanning system of FIG.

[0008] [Figure 2] 2 is a perspective view of an alternative dome-shaped housing for a magnetic array for use with the MRI scanning system of FIG. 1 having an access opening defined in the dome-shaped housing, in accordance with at least one aspect of the present disclosure.

[0009] [Figure 3] 3 is a perspective view of an alternative dome-shaped housing for a magnetic array for use with the MRI scanning system of FIG. 1 having an access opening and an adjustable gap defined within the dome-shaped housing, in accordance with at least one aspect of the present disclosure.

[0010] [Figure 4] FIG. 4 depicts a dome-shaped housing for use with an MRI scanning system having an access opening in the form of a centrally defined hole, in accordance with at least one aspect of the present disclosure.

[0011] [Figure 5] FIG. 5 is a cross-sectional view of the dome-shaped housing of FIG. 4 according to at least one aspect of the present disclosure.

[0012] [Figure 6] FIG. 6 depicts a control schematic for an MRI system in accordance with at least one aspect of the present disclosure.

[0013] [Figure 7] FIG. 7 is a flowchart depicting a method for acquiring imaging data from an MRI system according to at least one aspect of the present disclosure.

[0014] [Figure 8] FIG. 8 depicts an MRI scanning system and a robotic system according to at least one aspect of the present disclosure.

[0015] [Figure 9] FIG. 9 is a flowchart depicting a method for converting a low-resolution MRI image into a higher-resolution MRI image according to at least one aspect of the present disclosure.

[0016] [Figure 10]FIG. 10 is a flowchart depicting a method for converting low-resolution MRI images into super-resolution MRI images that includes implementing transfer learning to a high-field strength brain model based on a low-field strength dataset, according to at least one aspect of the present disclosure.

[0017] [Figure 11] FIG. 11 depicts an array of MRI images including low-field, low-resolution MRI brain images and higher-resolution MRI brain images converted from the low-field, low-resolution MRI brain images, according to at least one aspect of the present disclosure.

[0018] [Figure 12] FIG. 12 depicts an array of MRI images including a low-field low-resolution MRI phantom image and a super-resolution MRI phantom image converted from the low-field low-resolution MRI phantom image, according to at least one aspect of the present disclosure.

[0019] [Figure 13] FIG. 13 is a block diagram of a system for converting a low-resolution MRI image into a higher-resolution MRI image in accordance with at least one aspect of the present disclosure.

[0020] Corresponding reference characters indicate corresponding parts throughout the several views. The examples presented herein are illustrative of various disclosed embodiments and are in one form only, and such examples are not to be construed as in any way limiting the scope thereof. DETAILED DESCRIPTION OF THE INVENTION

[0021] (Detailed explanation) The applicant of the present application owns the following patent applications, each of which is incorporated herein by reference in its individual entirety: -International Patent Application No. PCT / US2022 / 72143, filed May 5, 2022, entitled "NEURAL INTERVENTIONAL MAGNETIC RESONANCE IMAGING APPARATUS"; -U.S. Patent Application No. 18 / 057,207, filed November 19, 2022, entitled "SYSTEM AND METHOD FOR REMOVING ELECTROMAGNETIC INTERFERENCE FROM LOW-FIELD MAGNETIC RESONANCE IMAGES"; -U.S. Patent Application No. 18 / 147,418, filed December 28, 2022, and entitled "MODULARIZED MULTI-PURPOSE MAGNETIC RESONANCE PHANTOM"; -U.S. Patent Application No. 18 / 147,452, filed December 28, 2022, and entitled "INTRACRANIAL RADIO FREQUENCY COIL FOR INTRAOPERATIVE MAGNETIC RESONANCE IMAGING."

[0022] Before describing various aspects of the interventional magnetic resonance imaging device in detail, it should be noted that the illustrative examples are not limited in application or use to the details of construction and arrangement of parts illustrated in the accompanying drawings and description. The illustrative examples may be implemented or incorporated in other aspects, variations, and modifications, and may be practiced or carried out in various ways. Furthermore, unless otherwise indicated, the terms and phrases employed herein are chosen for the purpose of describing the illustrative examples for the convenience of the reader, and not for purposes of limitation thereof. It should also be understood that one or more of the aspects, aspect expressions, and / or examples described below may be combined with any one or more of the other subsequently described aspects, aspect expressions, and / or examples.

[0023] Various aspects are directed to neurointerventional magnetic resonance imaging (MRI) devices that enable the integration of MRI with surgical intervention and guidance. This includes providing physical access to the patient's surrounding area as well as access to the patient's head using one or more access openings. In addition, neurointerventional MRI devices may enable the use of robotic guided instruments and / or conventional surgical tools. In various instances, neurointerventional MRI can be used intraoperatively to obtain scans of the patient's head and / or brain during surgical interventions, such as surgical procedures like brain biopsies or neurosurgery.

[0024] 1 depicts an MRI scanning system 100 including a dome-shaped housing 102 configured to receive a patient's head. The dome-shaped housing 102 may further include at least one access opening configured to allow access to the patient's head and allow for neurointervention. The space within the dome-shaped housing 102 forms a region of interest for the MRI scanning system 100. Target tissue within the region of interest is subjected to magnetization fields / pulses to obtain imaging data representative of the target tissue, as further described herein.

[0025] 1A , a patient can be positioned so that their head is positioned within a region of interest within the dome-shaped housing 102. The brain can be positioned entirely within the dome-shaped housing 102. In such cases, to facilitate intracranial intervention (e.g., neurosurgery) in coordination with MR imaging, the dome-shaped housing 102 can include one or more openings that provide access to the brain. The openings can be spaced around the periphery of the dome-shaped housing.

[0026] The MRI scanning system 100 may include an auxiliary cart (see, for example, auxiliary cart 540 in FIG. 6 ) that stores certain conventional MRI electrical and electronic components (e.g., a computer, a programmable logic controller, a power distribution unit, amplifiers, etc.). The MRI scanning system 100 may also include a magnet cart that holds the dome-shaped housing 102, gradient coils, and / or transmission coils, as described further herein. In addition, the magnet cart may, in various instances, be attached to a receive coil. Referring primarily to FIG. 1 , the dome-shaped housing 102 may further include an RF transmission coil, a gradient coil 104 (depicted on its exterior), and a shim magnet 106 (depicted on its interior). Alternative configurations for the gradient coil 104 and / or the shim magnet 106 are also envisioned. In various instances, the shim magnet 106 may be adjustably positioned within a shim tray within the dome-shaped housing 102, which may allow a technician to configure the magnetic flux density of the dome-shaped housing 102 in fine detail.

[0027] Various structural housings for receiving a patient's head and enabling neurointervention can be utilized with MRI scanning systems such as MRI scanning system 100. In one aspect, MRI scanning system 100 may be equipped with an alternative housing, such as dome-shaped housing 202 (FIG. 2) or a two-piece housing 302 (FIG. 3) configured to form a dome shape. Dome-shaped housing 202 defines multiple access openings 203. Two-piece housing 302 also defines multiple access openings 303 and further includes an adjustable gap 305 between the two parts of the housing.

[0028] In various instances, housing 202 and housing 302 may include a bonding agent 308, such as, for example, an epoxy resin, that holds the plurality of magnetic elements 310 in a fixed position. The plurality of magnetic elements 310 may be bonded to a structural housing 312, such as, for example, a plastic substrate. In various aspects, bonding agent 308 and structural housing 312 may be non-conductive or diamagnetic materials. Referring primarily to FIG. 3 , two-piece housing 302 includes two structural housings 312. In various aspects, the structural housing for receiving the patient's head may be formed from more than two sub-components. Access opening 303 in structural housing 312 provides direct access to the patient's head and is not blocked by structural housing 312, bonding agent 308, or magnetic elements 310. Access opening 303 may be positioned, for example, within an open space of housing 302.

[0029] There are many possible configurations of neurointerventional MRI devices that can achieve improved access for surgical intervention. Many configurations are based on two main designs, commonly known as Halbach cylinders and Halbach domes, described in the following article: "Design of sparse Halbach magnet arrays for portable MRI using a genetic algorithm," IEEE transactions on magnetics, 54(1), 5100112, by Cooley et al. (e.g., Cooley, CZ, Haskell, MW, Cauley, SF, Sappo, C., Lapierre, CD, Ha, CG, Stockmann, JP, and Wald, LL (2018)). The article by Cooley et al., "Design of sparse Halbach magnet arrays for portable MRI using a genetic algorithm," published in IEEE transactions on magnetics, 54(1), 5100112 (2018), is incorporated herein by reference in its entirety.

[0030] In various cases, a dome-shaped enclosure for an MRI scanning system such as system 100 can include, for example, a Halbach dome, which defines the dome shape and is configured based on several factors, including main magnetic field B strength, field size, field homogeneity, device size, device weight, and patient access for neurointervention. In various aspects, the Halbach dome includes an outer radius and an inner radius at the base of the dome. The Halbach dome may include an elongated cylindrical portion extending from the base of the dome. In one aspect, the elongated cylindrical portion includes outer and inner radii that are the same as the base of the dome and continues a predetermined length from the base of the dome at a constant radius. In another aspect, the elongated cylindrical portion includes outer and inner radii that are different from the base of the dome (see, e.g., FIGS. 2 and 3). In such cases, the different outer and inner radii of the elongated cylindrical portion can blend with the base radius within a transition region.

[0031] 4 illustrates an example Halbach dome 400 for an MRI scanning system (e.g., system 100, etc.) in accordance with at least one aspect of the present disclosure. For example, the example Halbach dome 400 defines an access opening in the form of a hole or access opening 403, where the dome 400 is configured to receive the head and brain B of a patient P within a region of interest therein, and the access opening 403 is configured to allow access to the patient P for neurointervention using medical instruments and / or robotically controlled surgical tools. The Halbach dome 400 can be established with a single access opening 403 on an upper side 418 of the dome 400, which allows access to the top of the skull while minimizing impact on the magnetic field. Additionally or alternatively, the dome 300 can be configured with multiple access openings around the periphery of the structure 416 of the dome 400, as shown in FIGS. 2 and 3 .

[0032] Diameter D of access opening 403 hole may be small (e.g., about 2.54 cm) or very large (substantially the outer diameter r of the dome 400). ext) As the access opening 403 becomes larger, the dome 400 begins to resemble, for example, a Halbach cylinder. The access opening 403 is not limited to being at the apex of the dome 400. The access opening 403 can be located anywhere on the surface or structure 416 of the dome 400. In various instances, the entire dome 400 can be rotated so that the access opening 403 can be positioned with a desired physical location on the patient P.

[0033] FIG. 5 shows the diameter D of the access opening 403. hole , the length L of the dome 400, and the outer radius r of the dome 400. ext and the inner radius r in 1 depicts the relative dimensions of the Halbach dome 400, including the outer radius r. The Halbach dome 400 comprises a plurality of magnetic elements arranged in a Halbach array to form a magnetic assembly. The magnetic elements have an outer radius r. ext and the inner radius r in In one aspect, exemplary dimensions are r in =19.3cm, r ext = 23.6 cm, L = 38.7 cm, and 2.54 cm ≤ D < 19.3 cm.

[0034] Based on the above example dimensions, the Halbach dome 400 with access opening 403 may be configured with a magnetic flux density B of approximately 72 mT and an overall mass of approximately 35 kg. It should be understood that the dimensions may be selected to achieve a desired magnetic flux density B, total weight of the Halbach dome 400 and / or magnet cart, and geometry of the neuro-interventional access opening 403 based on a particular application.

[0035] In various aspects, the Halbach dome 400 may be configured to define a plurality of access openings 403 located around the periphery of the structure 416 of the dome 400. These plurality of access openings 403 may be configured to allow access to the patient's head and brain B using instruments (e.g., surgical instruments) and / or a surgical robot.

[0036] In various aspects, the access opening 403 may be adjustable. The adjustable configuration may include, for example, adjusting the diameter D of the access opening 403. hole The access opening 403 may be provided with the ability to be adjusted using either a motor, mechanical assistance, or a manual system with a mechanical iris diaphragm arrangement to adjust the aperture 403. This would allow for a configuration of the dome without the access opening 403, to perform an imaging scan, and then adjust the dome 400 and its mechanical iris diaphragm arrangement to include the access opening 403, thus allowing surgical intervention through the access opening 403.

[0037] Halbach domes and their magnetic arrays for facilitating neurointervention are further described in International Patent Application No. PCT / US2022 / 72143, filed May 5, 2022, and entitled "NEURAL INTERVENTIONAL MAGNETIC RESONANCE IMAGING APPARATUS," which is incorporated herein by reference in its entirety.

[0038] Referring now to FIG. 6, a schematic diagram for an MRI system 500 is shown. For example, the MRI scanning system 100 (FIG. 1) and various dome-shaped housings and magnetic arrays therefor, as described further herein, can be incorporated into the MRI system 500, for example. The MRI system 500 includes a housing 502, which may be similar in many aspects to the dome-shaped housing 102 (FIG. 1), the dome-shaped housing 202 (FIG. 2), and / or the dome-shaped housing 302 (FIG. 3). The housing 502 is dome-shaped and configured to form a region of interest or field of view 552 therein. For example, the housing 502 can be configured to receive a patient's head, in various aspects of the present disclosure.

[0039] Housing 502 includes a magnet assembly 548 having a plurality of magnets (e.g., a Halbach array of magnets) arranged therein. In various aspects, a main magnetic field B generated by magnetic assembly 548 extends into a field of view 552 containing an object (e.g., a patient's head) being imaged by MRI system 500.

[0040] The MRI system 500 also includes an RF transmit / receive coil 550. The RF transmit / receive coil 550 is combined into an integrated transmit-receive (Tx / Rx) coil. In other cases, the RF transmit coil can be separate from the RF receive coil. For example, the RF transmit coil can be incorporated into the housing 502, and the RF receive coil can be positioned within the housing 502 to acquire imaging data.

[0041] The housing 502 also includes one or more gradient coils 504 configured to generate gradient fields to facilitate imaging of objects within a field of view 552 generated by a magnet assembly 548 (e.g., enclosed by a dome-shaped housing and a dome-shaped array of magnetic elements therein). A shim tray adapted to receive shim magnets 506 can also be incorporated within the housing 502.

[0042] During the imaging process, a main magnetic field B0 extends into the field of view 552. The direction of the effective magnetic field (B1) changes in response to RF pulses and associated electromagnetic fields transmitted by the RF transmit / receive coil 550. For example, the RF transmit / receive coil 550 may be configured to selectively transmit RF signals or pulses to objects (e.g., tissue in a patient's brain) within the field of view 552. These RF pulses may modify the effective magnetic field experienced by spins in the sample tissue.

[0043] The housing 502 is in signal communication with an auxiliary cart 530, which is configured to provide power to the housing 502 and send / receive control signals to / from the housing 502. The auxiliary cart 530 includes a power distribution unit 532, a computer 542, a spectrometer 544, a transmit / receive switch 545, an RF amplifier 546, and a gradient amplifier 558. In various instances, the housing 502 can be in signal communication with multiple auxiliary carts, each of which can support one or more of the power distribution unit 532, the computer 542, the spectrometer 544, the transmit / receive switch 545, the RF amplifier 546, and / or the gradient amplifier 558.

[0044] The computer 542 is in signal communication with the spectrometer 544 and is configured to transmit and receive signals between the computer 542 and the spectrometer 544. When an object within the field of view 552 is excited with RF pulses from the RF transmit / receive coil 550, the precession of the object results in an induced current, i.e., an MR current, which is detected by the RF transmit / receive coil 550 and transmitted to the RF preamplifier 556. The RF preamplifier 556 is configured to boost or amplify the excitation data signals and transmit them to the spectrometer 544. The spectrometer 544 is configured to transmit the excitation data to the computer 542 for storage, analysis, and image construction. The computer 542 is configured to combine multiple stored excitation data signals, for example, to generate an image. In various instances, the computer 542 is in signal communication with at least one database 562 that stores a reconstruction algorithm 564 and / or a pulse sequence 566. The computer 542 is configured to utilize the reconstruction algorithm to generate an MR image 568.

[0045] From the spectrometer 544, the signal can also be relayed to an RF transmit / receive coil 550 within the housing 502 via an RF power amplifier 546 and a transmit / receive switch 545 positioned between the spectrometer 544 and the RF power amplifier 546. From the spectrometer 544, the signal can also be relayed to a gradient coil 560 within the housing 502 via a gradient power amplifier 558. For example, the RF power amplifier 546 is configured to amplify the signal and transmit it to the RF transmit coil 560, and the gradient power amplifier 558 is configured to amplify the gradient coil signal and transmit it to the gradient coil 560.

[0046] In various cases, the MRI system 500 can include a noise cancellation coil 554. For example, the auxiliary cart 530 and / or the computer 542 can be in signal communication with the noise cancellation coil 554. In other cases, the noise cancellation coil 554 can be optional. For example, some MRI systems disclosed herein may not include a supplemental / auxiliary RF coil for detecting and canceling electromagnetic interference (i.e., noise).

[0047] A flowchart depicting a process 570 for acquiring an MRI image is shown in FIG. 7. The flowchart can be implemented, for example, by the MRI system 500. In various instances, in block 572, a target object (e.g., a portion of a patient's anatomy) is positioned within a main magnetic field B within a region of interest (e.g., region of interest 552) such as in a dome-shaped housing (e.g., magnet assembly 548) of various MRI scanning devices described further herein. The main magnetic field B is configured to magnetically polarize hydrogen protons (H protons) in the target object (e.g., all organs and tissues), known as net longitudinal magnetization M. This is proportional to the tissue's proton density (PD) and evolves exponentially in time, with a time constant known as the tissue's longitudinal relaxation time T. The T value of individual tissues depends on several factors, including, for example, their microscopic structure, the water and / or lipid content therein, and the strength of the polarizing magnetic field. For these reasons, the T1 value of a given tissue sample depends on the age and state of health.

[0048] In block 574, a time-varying oscillating magnetic field B1, i.e., an excitation pulse, is applied to the magnetically polarized target object using an RF coil (e.g., RF transmit / receive coil 550). The carrier frequency of the pulsed B1 field is set to the resonance frequency of 1H protons, which diverts the longitudinal magnetization from its equilibrium longitudinal direction, resulting in a rotated magnetization vector, which can generally have a transverse magnetization component as well as a longitudinal magnetization component, depending on the diversion angle used. Typical B1 pulses include inversion pulses (i.e., 180-degree pulses) and 90-degree pulses. The 180-degree pulse reverses the direction of the magnetization of 1H protons in the longitudinal axis. The 90-degree pulse rotates the magnetization of 1H protons by 90 degrees so that the magnetization is in the transverse plane. The MR signal is a time-varying current proportional to the transverse component of the magnetization and is detected using a suitable RF coil. These MR signals decay exponentially in time, with a time constant known as the transverse relaxation time, T2, which also depends on, for example, the microscopic tissue structure, water / lipid content, and the strength of the magnetic field used.

[0049] In block 576, the MR signals are spatially encoded by exposing the target object to an additional magnetic field (known as a gradient field) generated by a gradient coil (e.g., gradient coil 560). The gradient field, which varies linearly in space, is applied for short-duration periods in pulsed form, with spatial variation in each direction. The end result is the generation of multiple spatially encoded MR signals, which can be detected in block 577 and reconstructed to form an MR image depicting a slice of the object. An RF receive coil (e.g., RF transmit / receive coil 550) can be configured to detect the spatially encoded RF signals. The slice may be oriented in the transverse, sagittal, coronal, or any oblique plane.

[0050] In block 578, the spatially encoded signals for each slice of the scanned region are mathematically digitized and spatially decoded using a computer reconstruction program (e.g., by computer 542) to generate an image depicting the internal anatomical structures of the subject. In various instances, the reconstruction program may utilize an (inverse) Fourier transform to convert the spatially encoded data (k-space data) back to geometrically decoded data.

[0051] FIG. 8 depicts a graphical illustration of a robotic system 680 that can be used for neurointervention using an MRI scanning system 600. The robotic system 680 includes a computer system 696 and a surgical robot 682. The MRI scanning system 600 can be similar to the MRI system 500 and can include a dome-shaped housing with an access opening and a magnetic array, as further described herein. For example, the MRI system 500 can include one or more access openings defined in a Halbach array of magnets in a permanent magnet assembly to provide access to one or more anatomical portions of a patient being imaged during a medical procedure. In various cases, the robotic arm and / or instruments of the surgical robot 682 are configured to extend through the access openings in the permanent magnet assembly to reach the patient or target site. Each access opening can provide access to the patient and / or surgical site. For example, in the case of multiple access openings, the multiple access openings can allow access from different directions and / or proximal locations.

[0052] According to various embodiments, the robotic system 680 is configured to be installed outside the MRI system 600. As shown in FIG. 8 , the robotic system 680 can include a robotic arm 684 configured for movement with one or more degrees of freedom. According to various embodiments, the robotic arm 684 includes one or more mechanical arm sections including a hollow shaft 686 and an end effector 688. The hollow shaft 686 and the end effector 688 are configured to be moved, rotated, and / or pivoted through various ranges of motion via one or more motion controllers 690. The double-headed curved arrows in FIG. 8 represent example rotational movements produced by the motion controllers 690 at various joints in the robotic arm 684.

[0053] According to various embodiments, the robotic arm 684 of the robotic system 682 is configured to access various anatomical portions of interest through or around the MRI scanning system 600. According to various embodiments, the access opening is designed to take into account the size of the robotic arm 684. For example, the access opening defines a perimeter configured to accommodate the robotic arm 684, hollow shaft 686, and end effector 688 therethrough. In various instances, the robotic arm 684 is configured to access various anatomical portions of a patient from around the side of the magnetic imaging device 600. The hollow shaft 686 and / or the end effector 688 can be adapted to receive a robotic instrument 692 (e.g., a biopsy needle having a cutting edge 694 for collecting a biopsy sample from a patient, etc.).

[0054] The reader will understand that robotic system 682 can be used in combination with various dome-shaped and / or cylindrical magnetic housings described further herein. Also, robotic system 682 and robotic instrument 692 in FIG. 8 are exemplary. Alternative robotic systems can be utilized in conjunction with the various MRI systems disclosed herein. Hand-held surgical instruments and / or additional imaging devices (e.g., endoscopes) and / or systems can also be utilized in conjunction with the various MRI systems disclosed herein.

[0055] In various aspects of the present disclosure, the MRI systems described herein may comprise low-field MRI (LF-MRI) systems. In such cases, the main magnetic field B generated by the permanent magnet assembly may be, for example, 0.1 T to 1.0 T. In other cases, the MRI systems described herein may comprise ultra-low-field MRI (ULF-MRI) systems. In such cases, the main magnetic field B generated by the permanent magnet assembly may be, for example, 0.03 T to 0.1 T.

[0056] Higher magnetic fields, such as those above 1.0 T, can preclude the use of certain electrical and mechanical components near MRI scanning equipment. For example, the presence of surgical instruments and / or surgical robotic components made of metal, particularly ferrous metals, can be dangerous near higher magnetic fields because such instruments can be attracted toward the magnetization source. Also, higher magnetic fields often require specially designed rooms with additional precautions and shielding to limit magnetic interference. Despite the limitations for high-field MRI systems, low-field and very-low-field MRI systems present various challenges to obtaining high-quality images with sufficient resolution to achieve desired imaging objectives.

[0057] LF and ULF-MRI systems generally define relatively poor overall magnetic field homogeneity compared to higher field MRI systems. For example, as further described herein, a dome-shaped enclosure for an array of magnets can comprise a Halbach array of permanent magnets that generates a magnetic field B0 with a homogeneity of 1,000 ppm to 10,000 ppm within a region of interest in various aspects of the present disclosure.

[0058] Images acquired using an LF-MRI system or a ULF-MRI system can be referred to as low-field images. Images acquired using a high-field MRI system can be referred to as high-field images. A training dataset for machine learning consisting of low-field images can be referred to as a low-field dataset. A training dataset for machine learning consisting of high-field images can be referred to as a high-field dataset.

[0059] In various cases, the intrinsic MRI signal is proportional to the field strength. As a result, the signal-to-noise (SNR) associated with LF-MRI and ULF-MRI systems can theoretically be up to 20 times lower than the SNR for high-field MRI systems. For example, for a 70 mT MRI system, the SNR can be approximately 5% of the SNR for a 1.5 T MRI system.

[0060] To improve spatial resolution, MRI systems can increase the number of k-space lines. However, increasing the number of k-space lines may further reduce the SNR and / or increase the scan time required for each scan. In various LF-MRI and ULF-MRI systems and methods, further reducing the SNR and / or increasing the scan time may be undesirable.

[0061] In various instances, deep learning can be employed to obtain higher spatial resolution without SNR reduction and / or increased scan time. Deep learning techniques are further described in the 2018 article by Muhammad Haris, Greg Shakhnarovich, and Norimichi Ukita, "Deep Back-Projection Networks for Super-Resolution" (e.g., Haris, Muhammad, and Shakhnarovich, Greg, and Ukita, Norimichi (2018) Deep Back-Projection Networks for Super-Resolution, 1664-1673. 10.1109 / CVPR.2018.00179), which is incorporated herein by reference in its entirety.

[0062] Such deep learning techniques typically utilize high-quality MRI training datasets. High-quality MRI training datasets can include a large number of MRI images. For example, to ensure statistical significance and robust analysis, MRI training datasets employed for deep learning studies can include at least several hundred 3D MRI volumes, each of which can include at least 20 2D slices. Thus, a typical high-quality MRI training dataset used for deep learning training can include over 2,000 2D slices. Increasing the size of the dataset can increase the variance of the data, thus ensuring greater statistical significance and generating more accurate models. However, in various cases, collecting a large number of high-resolution images using certain types of MRI systems may be impractical. For example, it may be impractical to obtain a high-quality training dataset of images acquired via non-clinical and / or pre-FDA-approved MRI systems (e.g., certain low-field MRI systems) that are in the early stages of research and development.

[0063] Instead of relying exclusively on a high-quality MRI training dataset based on low-field images, low-field, low-resolution images can be converted into higher-resolution images using a pre-trained model, where the model is trained using a high-field dataset (e.g., using a high-quality training dataset based on high-field images). For example, the pre-trained model can be based on images acquired from a high-field MRI system, such as a clinically available 1.5T MRI system. In various instances, for example, pre-trained models that have been trained using images from a 1.5T MRI system can be commonly available.

[0064] In such cases, a low-field, low-resolution image can be converted into a higher-resolution image (such as a high-resolution or super-resolution image), for example, without sacrificing SNR and / or requiring a longer scan time. As used herein, a "high-resolution" image can sometimes refer to an image with a spatial resolution equal to or greater than 1.5 mm × 1.5 mm. As used herein, a "low-resolution" image can sometimes refer to an image with a spatial resolution equal to or less than 3.0 × 3.0 mm. As used herein, a "normal-resolution" image can sometimes refer to an image with a spatial resolution of 1.5 mm × 1.5 mm to 3.0 × 3.0 mm. Generally, higher image resolution can correlate with smaller pixel size (e.g., an image with a spatial resolution of 1.5 mm × 1.5 mm is higher resolution than an image with a spatial resolution of 3.0 mm × 3.0 mm). As used herein, a "super-resolution" image may refer to an image that has been generated from an original image dataset, where the super-resolution image has a higher resolution than the resolution of the original image dataset. For example, a super-resolution image may be a high-resolution image generated from a low-resolution image dataset. As another example, a super-resolution image may be a normal-resolution image generated from a low-resolution dataset. As yet another example, a super-resolution image may be an ultra-high-resolution image generated from a high-resolution image dataset.

[0065] In various cases, a portion of the pre-trained model (e.g., a subset of neural layers, etc.) can be retrained based on a low-field dataset that is smaller than the high-field dataset used to generate the pre-trained model.

[0066] In such cases, the pre-trained model can be customized and / or fine-tuned to adjust for differences between the high-field model (i.e., based on high-field images) and the low-field model (i.e., based on low-field images). In various cases, retraining a pre-trained model or portions / layers thereof is referred to as transfer learning.

[0067] Referring now to FIG. 9 , a flowchart 1000 is shown. The flowchart 1000 depicts a method for converting a low-resolution MRI image into a higher-resolution MRI image by applying a deep learning model to the low-resolution MRI image. As described further herein, the flowchart 1000 utilizes a high-field deep learning model to improve the resolution of the low-field image. In various instances, the flowchart 1000 and / or portions of the flowchart 1000 can be implemented by a computing device, such as computing device 1310 ( FIG. 13 ). The first MRI image can be accessible by the system 1300 of FIG. 13 , for example.

[0068] In block 1002, a first MRI image is acquired. The first MRI image can be acquired from an LF-MRI system or a ULF-MRI system. In various cases, the first MRI image in block 1002 can be acquired from, for example, MRI scanning system 100 (FIG. 1) and / or MRI system 500 (FIG. 6). The first MRI image can be, for example, an image of a brain or a portion of a brain.

[0069] In various cases, the first MRI image acquired in block 1002 can be acquired from an LF-MRI system or a ULF-MRI system. In some cases, the MRI system used to acquire the first MRI image in block 1002 can be configured to generate a low magnetic field strength of less than 1 T within the field of view. In various cases, the low magnetic field strength can be, for example, less than 100 mT, and in some cases, about 70 mT.

[0070] In some cases, the first MRI image can be stored, for example, in a memory (such as memory 1314 of computing device 1310 (FIG. 13)) and / or can be transmitted, for example, from a remote computing device (such as a remote network and / or cloud storage device 1340 (FIG. 13)).

[0071] The resolution of the first MRI image can depend, for example, on the field strength of the MRI system. The resolution of the first MRI image can be a low-resolution image having a spatial resolution equal to or lower than 3.0 mm x 3.0 mm (e.g., a low-resolution image having a spatial resolution in the range of 3.0 mm x 3.0 mm to 5.0 mm x 5.0 mm, etc.). The reader will understand that the aforementioned exemplary ranges are based on acquiring the first MRI image, for example, in block 1002, using an LF-MRI system or an ULF-MRI system.

[0072] At block 1004, a deep learning model may be obtained. The deep learning model may be stored, for example, in a memory (such as memory 1314 of computing device 1310 (FIG. 13)) and / or may be transmitted, for example, from a remote computing device (such as a remote network and / or cloud storage device 1340 (FIG. 13)). The deep learning model may be accessible, for example, to system 1300 of FIG. 13. The deep learning model is based on high-field images.

[0073] In various cases, the deep learning brain model can be based on MRI images acquired using an MRI system configured to generate a high magnetic field of greater than 1 T within the field of view. In various cases, the high magnetic field strength can be, for example, about 1.5 T or about 3 T within the field of view.

[0074] In various cases, the deep learning model can be a pre-trained model. In other cases, the deep learning model can be obtained by training a model based on a high-field dataset including high-field high-resolution images and high-field low-resolution images. For example, the high-field low-resolution images can be augmented based on the high-field high-resolution images. In response to comparing the augmented high-field low-resolution images and the corresponding high-field high-resolution images, the model can be trained based on the augmentation and comparison.

[0075] In various aspects, a deep learning model is implemented via a neural network. For example, a deep learning model may include multiple neural layers forming a neural network. A neural network is a computational model used in machine learning. A neural network is generally composed of nodes organized within neural layers. The nodes are configured to perform a function on provided inputs to produce an output value. A neural network requires a training cycle to learn the parameters (i.e., weights) used to map inputs to desired outputs. The mapping occurs via a function. Thus, the weights are weights for the mapping function of the neural network.

[0076] At block 1008, the deep learning model may be applied to the first MRI image to generate a second MRI image. The resolution of the second MRI image may exceed the resolution of the first MRI image. In some cases, the second MRI image may have a two-fold improved resolution compared to the first image. For example, if the first image has a spatial resolution of 3.0 mm x 3.0 mm, the second MRI image may have a spatial resolution of 1.5 mm x 1.5 mm. In other cases, the second MRI image may have a more than two-fold improved resolution compared to the first image.

[0077] In various cases, the amount of resolution improvement achieved by generating the second MRI image can be controlled based on training of the deep learning model. For example, in some cases, the amount of resolution improvement achieved can be optimized based on the feasibility of performing clinical analysis on the second MRI image having a given resolution and the computational cost of achieving the given resolution. In some cases, configuring the deep learning model to generate the second MRI image to have a two-fold improved resolution compared to the first image allows the second MRI image to be suitable for clinical analysis while also minimizing the computational cost required to generate the second MRI image. In some cases, configuring the deep learning model to generate the second MRI image to have a two-fold improved resolution compared to the first MRI image allows for near-real-time analysis of the second MRI image, for example, because of the minimized computational cost required to generate the second MRI image.

[0078] In various cases, in response to generating the second MRI image at block 1008, the flowchart can be configured to output the second MRI image. The second MRI image can be provided to, for example, a clinician and / or a patient. In various cases, the second MRI image can be displayed on a graphical user interface, such as a computer screen.

[0079] As further described herein, converting low-resolution MRI images into higher-resolution MRI images by applying a deep learning model, as set forth in flowchart 1000, can further include implementing a transfer learning technique to fine-tune the deep learning model. Transfer learning can be utilized to retrain at least one layer of the deep learning model based on a low-field dataset. In various instances, a subset of neural layers of a neural network implementing the deep learning model can be retrained. Retraining at least one neural layer of the deep learning model can include accessing a low-field dataset including low-field high-resolution images and low-field low-resolution images. The low-field low-resolution images can be augmented based on the low-field high-resolution images. In response to comparing the augmented low-field low-resolution images and the corresponding low-field high-resolution images, at least one neural layer can be retrained based on the augmentation and comparison.

[0080] 10, another flowchart 1010 is depicted for converting low-resolution MRI images to higher-resolution MRI images. Flowchart 1010 includes a model training subroutine 1014 based on a high-field dataset 1014a and a transfer learning subroutine 1016 based on a low-field dataset 1016a. At block 1018, the transfer learning training model is applied to the input from block 1012 to generate a super-resolution prediction. The super-resolution prediction from block 1018 can be output to a display at block 1020. In various instances, flowchart 1010 and / or portions of flowchart 1010 can be implemented by a computing device, such as computing device 1310 (FIG. 13).

[0081] The model training subroutine 1014 includes utilizing high-field data sets 1014a obtained by one or more high-field MRI systems. The high-field data sets 1014a include high-field high-resolution imaging data 1014b and high-field low-resolution imaging data 1014c. In various cases, the high-field data sets 1014a can be based on MRI images acquired using an MRI system configured to generate a high magnetic field of greater than 1 T within the field of view. In various cases, the high-field data sets 1014a can be based on MRI images acquired using an MRI system configured to generate a high magnetic field strength of about 1.5 T or about 3 T within the field of view, for example.

[0082] In block 1014d, the high-field dataset 1014a is preprocessed. In some cases, the preprocessing may include distortion correction (e.g., correction of geometric distortions resulting from field inhomogeneity and / or gradient nonlinearity), spatial normalization (e.g., spatial normalization to compensate for image intensity resulting from low-frequency intensity inhomogeneity), and / or head masking (e.g., head masking to remove extraneous artifacts from outside the head). Then, in block 1014e, the high-field dataset 1014a is augmented. For example, the high-field low-resolution imaging data 1014c may be augmented based on the high-field high-resolution imaging data 1014b. In some cases, the augmentation may include image cropping, image rotation, image flipping, image filtering, image translation, image shearing, and / or image scaling. In response to comparing the augmented high-field low-resolution imaging data and the corresponding high-field high-resolution imaging data, a model may be trained in block 1014f. In various instances, the model training in block 1014f can be iterative with the application of deep learning techniques in block 1014g. The deep learning techniques can include gradual upsampling and / or iterative upsampling and downsampling techniques. In one aspect of the present disclosure, the deep learning techniques applied in block 1014f can include deep backprojection networks, such as those described in the 2018 article "Deep Backprojection Networks for Super-Resolution" by Muhammad Haris, Greg Shakhnarovich, and Norimichi Ukita (incorporated herein by reference in its entirety).

[0083] At block 1014h, the model can be evaluated to determine if a pre-trained model exists. Upon completion of the model training subroutine 1014, the flowchart 1010 can proceed to applying a transfer learning subroutine 1016.

[0084] In other aspects of the present disclosure, a pre-trained model can be obtained and provided to computing device 1310 (FIG. 13). For example, the pre-trained model can be publicly available. In some cases, the pre-trained model can be downloaded, for example, from a remote network and / or cloud storage device (such as network 1340 (FIG. 13)).

[0085] The pre-trained model from the model training subroutine 1014 is provided to a transfer learning subroutine 1016. The transfer learning subroutine 1016 includes utilizing a low-field dataset 1016a obtained by one or more low-field MRI systems. The low-field dataset 1016a includes low-field high-resolution imaging data 1016b and low-field low-resolution imaging data 1016c. In various cases, the low-field dataset 1016a may be based on MRI images acquired using an MRI system configured to generate a low-field field below 1 T within the field of view. In various cases, the low-field dataset 1016a may be based on MRI images acquired using an MRI system configured to generate a low-field field of about 100 mT or about 70 mT within the field of view, for example.

[0086] In block 1016d, the low-field data set 1016a is pre-processed. Then, in block 1014e, the low-field data set 1016b is augmented. For example, the low-field low-resolution imaging data can be augmented based on the low-field high-resolution imaging data. In response to comparing the augmented low-field low-resolution imaging data 1016c and the corresponding low-field high-resolution imaging data 1016b, one or more layers of the model from the model training subroutine 1014 can be re-trained in block 1016f.

[0087] The model training subroutine 1014 and the transfer learning subroutine 1016 can be stored, for example, in a memory (such as memory 1314 of computing device 1310 (FIG. 13)) and / or can be transmitted, for example, from a remote computing device (such as a remote network and / or cloud storage device 1340).

[0088] In various aspects, the deep learning model from the model training subroutine 1014 is implemented and / or applied to the input via a neural network. For example, the deep learning model can be implemented via a neural network including multiple neural layers. A subset of the neural network can be retrained using the transfer learning subroutine 1016. For example, less than 10 percent or less than 5 percent of the neural layers can be retrained using the transfer learning subroutine 1016. In various cases, between 4 and 10 neural layers can be retrained using the transfer learning subroutine 1016. In yet other cases, a single neural layer (e.g., the final layer in a neural network) can be retrained using the transfer learning subroutine 1016.

[0089] In block 1018, a super-resolution prediction is computed. For example, block 1018 may receive input from block 1012 and apply the transfer learning training model from model training subroutine 1014 and transfer learning subroutine 1016 to the input. Block 1012 may, for example, be similar in many aspects to block 1002 (FIG. 9). For example, the input may include imaging data corresponding to low-field MRI images.

[0090] In various cases, the super-resolution prediction can be output in block 1020 to a display, such as display 1330 (FIG. 13). The output in block 1020 can be displayed on a graphical user interface, such as a computer screen. The super-resolution prediction can include an MRI image having improved spatial resolution compared to the MRI image corresponding to the input provided in block 1012. In some cases, the super-resolution prediction output to the display in block 1020 can have a resolution improved by a factor of two for each of the three axes. For example, the input received from block 1020 can include imaging data having a spatial resolution of 3 mm×3 mm×6 mm, and the super-resolution prediction output to the display in block 1020 can have a spatial resolution of 1.5 mm×1.5 mm×3.0 mm. Thus, the image can be improved from a low-resolution MRI image (e.g., 3 mm×3 mm) to a high-resolution MRI image (e.g., 1.5 mm×1.5 mm).

[0091] Exemplary MRI images are shown in Figures 11 and 12. Figure 11 depicts a first array of MRI brain images 1100 including image 1102, image 1104, image 1106, and image 1108 acquired in different imaging planes through the brain. Figure 11 also depicts a second array of corresponding brain images 1110 including image 1112, image 1114, image 1116, and image 1108, respectively. The first array of MRI brain images 1100 is of lower resolution than the second array of MRI brain images 1110, which has been reconstructed into a higher resolution image without sacrificing SNR or scan time efficiency using a deep learning model pre-trained based on high-field MRI data.

[0092] 12 depicts a first array of MRI phantom images 1200 including image 1202, image 1204, image 1206, and image 1208 at different imaging planes through the phantom. FIG. 12 also depicts a second array of corresponding phantom images 1210 including image 1212, image 1214, image 1216, and image 1208, respectively. The first array of MRI phantom images 1200 is at a lower resolution than the second array of MRI phantom images 1210, which has been reconstructed into super-resolution images without sacrificing SNR or scan time efficiency using a deep learning model pre-trained on high-field MRI data.

[0093] 13 , a system 1300 for converting low-resolution MRI images into higher-resolution MRI images is shown. The system 1300 includes a computing device 1310 having a memory 1314, a processor 1312, and an application 1316. The computing device 1310 may communicate with one or more other computing devices via a network 1340. The computing device 1310 may be implemented as, for example, a server, a desktop computer, a laptop computer, and / or a mobile device (such as a tablet device or a mobile phone device). In various instances, the computing device 1310 may represent multiple computing devices in communication with each other (such as multiple servers in communication with each other).

[0094] Processor 1312 may represent two or more processors on computing device 1310 executing in parallel and utilizing corresponding instructions stored using memory 1314. Memory 1314 represents a non-transitory computer-readable storage medium. Memory 1314 may represent, for example, one or more different types of memory utilized by computing device 1310. In addition to storing instructions that enable processor 1312 to implement application 1316 and the computer-readable instructions stored in memory 1314, memory 1314 may be used to store, for example, data, imaging data, pre-trained models, algorithms, and / or subroutines.

[0095] Application 1316 may be accessed directly by a user of computing device 1310. In other implementations, application 1316 may run on computing device 1310 as a component of a cloud network, and a user accesses application 1316 from another computing device via a network, such as network 1340. Application 1316 allows a user to convert low-resolution MRI images into higher-resolution MRI images. Application 1316 may also include additional functionality, such as noise reduction and / or further editing and / or manipulation of the images.

[0096] The application 1316 includes a neural network 1318 with multiple layers 1320a, 1320b, etc. The neural network 1318 can be configured to convert low-resolution MRI images into higher-resolution MRI images. For example, the neural network 1318 can be trained, and / or a pre-trained neural network can be provided to the application 1316. The neural network 1318 is configured to apply a deep learning model to input (e.g., input imaging data representing low-field low-resolution MRI images) and generate output (e.g., output imaging data representing low-field high-resolution MRI images). As described further herein, the deep learning model can be based on a high-field model based on a high-field dataset, or a high-field model that is updated via transfer learning based on a smaller low-field dataset. Various methods for converting low-resolution images into higher-resolution images are further described herein.

[0097] The system 1300 also includes an input device 1302, which may comprise, for example, a user interface. The user interface may include additional elements and components, such as other tools used for image editing and manipulation, and graphic designs for use as part of the application 1316.

[0098] In various instances, the input device 1302 can be incorporated into the computing device 1310. In other instances, the input device 1302 can be separate from the computing device 1310. In either event, the input device 1302 is configured to receive inputs related to imaging data, pre-trained models, algorithms, and / or subroutines thereof that are accessed by the processor to transform low-resolution MRI images (e.g., provided to the computing device 1310 via the input device 1302) into higher-resolution MRI images (e.g., output from the computing device 1310 via the display 1330). The input device 1302 may access images from memory 1314 and / or from other storage locations (either locally on the computing device 1310 or remotely on another computing device accessed across a network 1340).

[0099] In various instances, in response to receiving input from input device 1302, processor 1312 of computing device 1310 can implement application 1316 to apply neural network 1318 and its layers 1320a, 1320b, etc. to the input imaging data. For example, machine-readable instructions stored in memory 1314 can invoke application 1316 via processor 1312 to convert low-resolution MRI images into higher-resolution MRI images, e.g., according to flowcharts 1000 (FIG. 9) and 1100 (FIG. 10).

[0100] example Various additional aspects of the subject matter described herein are presented in the following numbered examples.

[0101] Example 1: A method including: acquiring a first image of a brain using a low-field strength magnetic resonance imaging system, the first image having a first resolution; acquiring a deep learning brain model based on the high-field strength image, the deep learning brain model configured to be applied by a neural network having a plurality of layers; and applying the deep learning brain model to the first image to generate a second image of the brain, the second image having a second resolution, the second resolution being greater than the first resolution.

[0102] Example 2: The method of Example 1, wherein obtaining a deep learning brain model based on high field intensity images includes obtaining a pre-trained model.

[0103] Example 3: The method described in Example 1, wherein obtaining a deep learning brain model based on high field intensity images includes accessing a high field dataset including high field intensity high resolution images and high field intensity low resolution images, enhancing the high field intensity low resolution images based on the high field intensity high resolution images, and training a deep learning brain model based on the enhanced high field intensity low resolution images.

[0104] Example 4: The method of any one of Examples 1-3, further comprising performing transfer learning to fine-tune the deep learning brain model with respect to the first image of the brain.

[0105] Example 5: The method of any one of Examples 1-3, further comprising retraining at least one layer of the deep learning brain model using the low-field dataset.

[0106] Example 6: The method of any one of Examples 1-3, further comprising retraining a subset of layers of the deep learning brain model using a low-field dataset.

[0107] Example 7: The method of any one of Examples 1-3, further comprising accessing a low-field dataset including low-field intensity high-resolution images and low-field intensity low-resolution images, enhancing the low-field intensity low-resolution images based on the low-field intensity high-resolution images, and retraining at least one layer of the deep learning brain model based on the enhanced low-field intensity low-resolution images.

[0108] Example 8: The method of any one of Examples 1-7, further comprising outputting a second image of the brain to a display.

[0109] Example 9: The method of any one of Examples 1-7, further comprising displaying a second image of the brain.

[0110] Example 10: The method of any one of Examples 1-9, wherein acquiring a first image of the brain using a low-field strength magnetic resonance imaging system includes generating a low magnetic field strength below 100 mT, and the deep learning brain model is based on magnetic resonance imaging images acquired with a high magnetic field strength above 1 T.

[0111] Example 11: A system comprising: a processor; and a memory storing machine-readable instructions, the processor configured to execute the machine-readable instructions, which, when executed, implement a neural network, the neural network configured to: obtain a high-field strength magnetic resonance model comprising a plurality of layers; receive data representing a low-field image; transform the low-field image into a higher resolution image based on the high-field strength magnetic resonance model; and output the higher resolution image.

[0112] Example 12: The system of Example 11, wherein the high field strength magnetic resonance model comprises a pre-trained deep learning model.

[0113] Example 13: The system of Example 12, wherein the pre-trained deep learning model is trained using a high-field dataset including high-field intensity high-resolution images and high-field intensity low-resolution images.

[0114] Example 14: The system described in Example 11, wherein the neural network is further configured to acquire a high-field dataset including high-field intensity high-resolution images and high-field intensity low-resolution images, enhance the high-field intensity low-resolution images based on the high-field intensity high-resolution images, and train a high-field intensity magnetic resonance model based on the enhanced high-field intensity low-resolution images.

[0115] Example 15: The system of any one of Examples 11-14, wherein the neural network is further configured to retrain at least one layer of the high field strength magnetic resonance model using the low field data set.

[0116] Example 16: The system described in any one of Examples 11-14, wherein the neural network is further configured to acquire a low-field dataset including low-field strength high-resolution images and low-field strength low-resolution images, enhance the low-field strength low-resolution images based on the low-field strength high-resolution images, and retrain at least one layer of the high-field strength magnetic resonance model based on the enhanced low-field strength low-resolution images.

[0117] Example 17: The system of Example 16, wherein the high-field data set is larger than the low-field data set.

[0118] Example 18: The system of any one of Examples 11-14, wherein the high-field strength magnetic resonance model is trained using high-field images acquired at high magnetic field strengths exceeding 1 T.

[0119] Example 19: The system of Example 18, wherein the low-field images are acquired at low magnetic field strengths below 0.3 T.

[0120] Example 20: The system of Example 18, wherein the low-field images are acquired at low magnetic field strengths below 100 mT.

[0121] Although various aspects disclosed herein are directed to brain imaging and / or neurological intervention, the reader will understand that the various systems and methods disclosed herein may, in various instances, be used to image other portions of a patient's anatomy and / or different structures.

[0122] While several embodiments have been illustrated and described, it is not the applicant's intention to restrict or limit the scope of the appended claims to such details. Numerous modifications, variations, changes, substitutions, combinations, and equivalents to those embodiments may be implemented and will occur to those skilled in the art without departing from the scope of the present disclosure. Furthermore, the structure of each element associated with the described embodiments can alternatively be described as a means for providing the function performed by that element. Furthermore, where a material is disclosed for a component, other materials may also be used. It is therefore to be understood that the foregoing description and the appended claims are intended to cover all such modifications, combinations, and variations that fall within the scope of the disclosed embodiments. The appended claims are intended to cover all such modifications, variations, changes, substitutions, modifications, and equivalents.

[0123] The foregoing detailed description sets forth various aspects of the present devices and / or processes through the use of block diagrams, flowcharts, and / or examples. To the extent that such block diagrams, flowcharts, and / or examples contain one or more functions and / or operations, it will be understood by those skilled in the art that each function and / or operation within such block diagrams, flowcharts, and / or examples can be individually and / or collectively implemented by a wide range of hardware, software, firmware, or virtually any combination thereof. Those skilled in the art will recognize that some aspects of the embodiments disclosed herein may equivalently be implemented, in whole or in part, as one or more computer programs running on one or more computers (e.g., as one or more programs running on one or more computer systems), as one or more programs running on one or more processors (e.g., as one or more programs running on one or more microprocessors), as firmware, or in integrated circuits, as virtually any combination thereof, and that designing circuitry and / or writing code for the software and / or firmware would be clearly within the skill of one of ordinary skill in the art in light of this disclosure. Additionally, those skilled in the art will understand that the subject mechanisms described herein can be distributed as one or more program products in a variety of forms, and that the illustrative embodiments of the subject matter described herein apply regardless of the particular type of signal-bearing medium used to actually effect the distribution.

[0124] The instructions used to program the logic to implement various disclosed aspects can be stored in memory within the system, such as dynamic random access memory (DRAM), cache, flash memory, or other storage device. Additionally, the instructions can be distributed over a network or using other computer-readable media. Thus, a machine-readable medium can include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computer), including, but not limited to, floppy diskettes, optical disks, compact disks, read-only memories (CD-ROMs), and magneto-optical disks, read-only memories (ROMs), random access memories (RAMs), erasable programmable read-only memories (EPROMs), electrically erasable programmable read-only memories (EEPROMs), magnetic or optical cards, flash memories, or tangible machine-readable storage devices used in transmitting information via the Internet via electrical, optical, acoustical, or other forms of propagated signals (e.g., carrier waves, infrared signals, digital signals, etc.). Thus, non-transitory computer-readable media includes any type of tangible machine-readable medium suitable for storing or transmitting electronic instructions or information in a form readable by a machine (e.g., a computer).

[0125] As used in any aspect of this specification, the term "control circuit" may refer to, for example, hardwired circuitry, programmable circuitry (e.g., a computer processor including one or more individual instruction processing cores, processing units, processors, microcontrollers, microcontroller units, controllers, digital signal processors (DSPs), programmable logic devices (PLDs), programmable logic arrays (PLAs), or field programmable gate arrays (FPGAs)), state machine circuitry, firmware that stores instructions executed by the programmable circuitry, and any combination thereof. Control circuits may collectively or individually be embodied as circuitry that forms part of a larger system (e.g., an integrated circuit (IC), an application-specific integrated circuit (ASIC), a system-on-chip (SoC), a desktop computer, a laptop computer, a tablet computer, a server, a smartphone, etc.). Thus, as used herein, a "control circuit" includes, but is not limited to, electrical circuitry having at least one discrete electrical circuit, electrical circuitry having at least one integrated circuit, electrical circuitry having at least one application-specific integrated circuit, electrical circuitry forming a general-purpose computing device configured by a computer program (e.g., a general-purpose computer configured, at least in part, by a computer program to perform the processes and / or devices described herein, or a microprocessor configured, at least in part, by a computer program to perform the processes and / or devices described herein), electrical circuitry forming a memory device (e.g., a form of random access memory), and / or electrical circuitry forming a communications device (e.g., a modem, a communications switch, or an optoelectronic appliance). Those skilled in the art will recognize that the subject matter described herein can be implemented in analog or digital form or some combination thereof.

[0126] As used in any aspect of this specification, the term "logic" may refer to apps, software, firmware, and / or circuitry configured to perform any of the foregoing operations. Software may be embodied as a software package, code, instructions, instruction sets, and / or data recorded on a non-transitory computer-readable storage medium. Firmware may be embodied as hard-coded (e.g., non-volatile) code, instructions or instruction sets, and / or data in a memory device.

[0127] When used in any aspect of this specification, the terms "component," "system," "module," and the like may refer to a controlled circuit computer-related entity that is either hardware, a combination of hardware and software, software, or software in execution.

[0128] As used in any aspect of this specification, an "algorithm" refers to a self-consistent sequence of steps leading to a desired result, and a "step" refers to manipulations on physical quantities and / or logical states which may, but need not, take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated. It is common usage to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like. These and similar terms can be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities and / or states.

[0129] The network may include a packet-switched network. The communication devices may be capable of communicating with each other using a selected packet-switched network communication protocol. One exemplary communication protocol may include an Ethernet communication protocol, which may enable communication using Transmission Control Protocol / Internet Protocol (TCP / IP). The Ethernet protocol may conform to or be compatible with the Ethernet standard published by the Institute of Electrical and Electronics Engineers (IEEE), published in December 2008 and entitled "IEEE 802.3 Standard," and / or later versions of this standard. Alternatively, or in addition, the communication devices may be capable of communicating with each other using an X.25 communication protocol. The X.25 communication protocol may conform to or be compatible with standards promulgated by the International Telecommunication Union Telecommunication Standardization Sector (ITU-T). Alternatively, or in addition, the communication devices may be capable of communicating with each other using a frame relay communication protocol. The frame relay communication protocol may conform to or be compatible with standards promulgated by the Consultative Committee for International Telegraph and Telephone (CCITT) and / or the American National Standards Institute (ANSI). Alternatively, or in addition, the transceivers may be capable of communicating with each other using an Asynchronous Transfer Mode (ATM) communication protocol. The ATM communication protocol may conform to or be compatible with the ATM standard published by the ATM Forum, published in August 2001 and entitled "ATM-MPLS Network Interworking 2.0," and / or later versions of this standard. Of course, different and / or later-developed connection-oriented network communication protocols are equally contemplated herein.

[0130] Unless otherwise specifically stated as is apparent from the foregoing disclosure, throughout the foregoing disclosure, discussions using terms such as "processing," "computing," "calculating," "determining," "displaying," or the like, should be understood to refer to the actions and processes of a computer system or similar electronic computing device that manipulates and converts data represented as physical (electronic) quantities in the computer system's registers and memory into other data that is similarly represented as physical quantities in the computer system's memory or registers or other such information storage, transmission, or display device.

[0131] One or more components may be referred to herein as being "configured to," "configurable to," "operable / operative to," "adapted / adaptable," "able to," "conformable / conformed to," etc. Those skilled in the art will recognize that, unless the context requires otherwise, "configured to" can generally encompass active and / or inactive and / or standby state components.

[0132] The terms "proximal" and "distal" are used herein with reference to a clinician manipulating a handle portion of a surgical instrument. The term "proximal" refers to the portion closest to the clinician, and the term "distal" refers to the portion located away from the clinician. It should be further understood that for convenience and clarity, spatial terms such as "vertical," "horizontal," "up," and "down" may be used herein with respect to the drawings. However, surgical instruments are used in many orientations and positions, and these terms are not intended to be limiting and / or absolute.

[0133] Those skilled in the art will generally recognize that terms used herein, particularly in the appended claims (e.g., the body of the appended claims), are generally intended as “open” terms (e.g., the term “including” should be interpreted as “including but not limited to,” the term “having” should be interpreted as “having at least,” the term “includes” should be interpreted as “includes but is not limited to,” etc.). It will be further understood by those skilled in the art that where a specific number of introduced claim recitations are intended, such intention will be expressly recited in the claim, and that in the absence of such recitation, no such intention exists. For example, as an aid to understanding, the following appended claims may contain the use of the introductory phrases “at least one” and “one or more” to introduce claim recitations. However, the use of such phrases should not be construed as implying that introducing a claim recitation with the indefinite article "a" or "an" limits any particular claim containing such an introduced claim recitation to claims containing only one such recitation, even when the same claim includes the introductory phrases "one or more" or "at least one" and an indefinite article such as "a" or "an" (e.g., "a" and / or "an" should typically be interpreted to mean "at least one" or "one or more"); the same applies to the use of a definite article used to introduce a claim recitation.

[0134] Additionally, even when a claim recitation incorporating a specific number is explicitly recited, those of skill in the art will recognize that such recitation should typically be interpreted to mean at least the number recited (e.g., a literal recitation of "two recitations" without other modifiers typically means at least two recitations or two or more than two recitations). Furthermore, in those instances where notation similar to "at least one of A, B, and C, etc." is used, such a construct is generally intended in the sense that one of skill in the art would understand the notation (e.g., "a system having at least one of A, B, and C" would include, without limitation, systems having A alone, B alone, C alone, A and B together, A and C together, B and C together, and / or A, B, and C together, etc.). In those instances where notation similar to "at least one of A, B, or C, etc." is used, generally, such constructs are intended in the sense that one of ordinary skill in the art would understand the notation (e.g., "a system having at least one of A, B, and C" would include, but not be limited to, systems having A alone, B alone, C alone, A and B together, A and C together, B and C together, and / or A, B and C together, etc.). It will be further understood by those of ordinary skill in the art that disjunctive words and / or phrases presenting two or more alternative terms, whether in the description, claims, or drawings, should typically be understood to contemplate the possibility of including one of the terms, either of the terms, or both terms, unless the context indicates otherwise.For example, the phrase "A or B" will typically be understood to include the possibilities of "A," or "B," or "A and B."

[0135] With respect to the appended claims, those skilled in the art will understand that the actions recited therein may generally be performed in any order. Also, while various operational flow diagrams are presented in a certain sequence, it should be understood that various actions may be performed in other orders than those depicted, or may be performed in parallel. Examples of such alternative orderings may include overlapping, interleaved, interrupted, reordered, incremental, prelude, supplemental, simultaneous, reverse, or other variant orderings, unless the context indicates otherwise. Furthermore, terms such as "responsive to," "related to," or other past tense adjectives generally are not intended to exclude such variants, unless the context indicates otherwise.

[0136] It should be noted that any reference to "one aspect," "an aspect," "an exemplification," "one exemplification," and the like means that a particular feature, structure, or characteristic described in connection with an aspect is included in at least one aspect. Thus, appearances of the phrases "in one aspect," "in an aspect," "in an exemplification," and "in one exemplification" in various places throughout this specification do not necessarily all refer to the same aspect. Furthermore, particular features, structures, or characteristics may be combined in any suitable manner in one or more aspects.

[0137] Any patent application, patent, non-patent publication, or other disclosure material referred to herein and / or listed in any Application Data Sheet is incorporated herein by reference to the extent the incorporated material does not contradict this specification. Accordingly, to the extent necessary, the disclosure as expressly set forth herein supersedes any conflicting material incorporated herein by reference. Any material, or portion thereof, that is considered to be incorporated herein by reference but that conflicts with existing definitions, descriptions, or other disclosure material set forth herein will be incorporated only to the extent that no conflict arises between the incorporated material and the existing disclosure material.

[0138] In summary, numerous benefits resulting from employing the concepts described herein have been described. The foregoing description of one or more embodiments has been presented for purposes of illustration and explanation. It is not intended to be exhaustive or to be limited to the precise embodiments disclosed. Modifications or variations are possible in light of the above teachings. One or more embodiments have been selected and described, along with various modifications as suited to particular uses, to illustrate the principles and practical applications, thereby enabling those skilled in the art to utilize various embodiments. It is intended that the claims submitted herewith define the overall scope.

Claims

1. 1. A method comprising: acquiring a first image of the brain using a low-field strength magnetic resonance imaging system, the first image comprising a first resolution; Obtaining a deep learning brain model based on high field intensity images, the deep learning brain model being configured to be applied by a neural network comprising multiple layers; applying the deep learning brain model to the first image to generate a second image of the brain, the second image having a second resolution, the second resolution being greater than the first resolution; A method comprising:

2. The method of claim 1 , wherein obtaining the deep learning brain model based on high-field intensity images comprises obtaining a pre-trained model.

3. obtaining the deep learning brain model based on high field intensity images, accessing a high-field dataset comprising high-field intensity high-resolution images and high-field intensity low-resolution images; enhancing the high field intensity low resolution image based on the high field intensity high resolution image; training the deep learning brain model based on the enhanced high-field intensity low-resolution images; The method of claim 1 , comprising:

4. 4. The method of claim 3, further comprising performing transfer learning to fine-tune the deep learning brain model on the first image of the brain.

5. 4. The method of claim 3, further comprising retraining at least one layer of the deep learning brain model with a low-field dataset.

6. 4. The method of claim 3, further comprising retraining a subset of the layers of the deep learning brain model with a low-field dataset.

7. accessing a low-field dataset comprising low-field intensity high-resolution images and low-field intensity low-resolution images; enhancing the low field strength low resolution image based on the low field strength high resolution image; retraining at least one layer of the deep learning brain model based on the enhanced low-field intensity low-resolution images; and The method of claim 3 further comprising:

8. The method of claim 3 , further comprising outputting the second image of the brain to a display.

9. The method of claim 3 , further comprising displaying the second image of the brain.

10. 4. The method of claim 3, wherein acquiring the first image of the brain using the low-field strength magnetic resonance imaging system includes generating a low magnetic field strength below 100 mT, and the deep learning brain model is based on magnetic resonance imaging images acquired with a high magnetic field strength above 1 T.

11. 1. A system comprising: a processor; a memory storing machine-readable instructions, the processor configured to execute the machine-readable instructions, the machine-readable instructions, when executed, implementing a neural network, the neural network comprising: obtaining a high field strength magnetic resonance model comprising a plurality of layers; receiving data representing a low-field image; converting the low-field image into a higher resolution image based on the high-field strength magnetic resonance model; outputting the higher resolution image; a memory configured to: A system comprising:

12. The system of claim 11 , wherein the high-field strength magnetic resonance model comprises a pre-trained deep learning model.

13. 13. The system of claim 12, wherein the pre-trained deep learning model is trained using a high-field dataset comprising high-field intensity high-resolution images and high-field intensity low-resolution images.

14. The neural network further comprises: acquiring a high-field dataset comprising high-field intensity high-resolution images and high-field intensity low-resolution images; enhancing the high field intensity low resolution image based on the high field intensity high resolution image; training the high field strength magnetic resonance model based on the enhanced high field strength low resolution images; The system of claim 11 configured to:

15. The system of claim 14 , wherein the neural network is further configured to retrain at least one layer of the high-field strength magnetic resonance model using a low-field data set.

16. The neural network further comprises: acquiring a low-field data set comprising a low-field intensity high-resolution image and a low-field intensity low-resolution image; enhancing the low field strength low resolution image based on the low field strength high resolution image; retraining at least one layer of the high-field strength magnetic resonance model based on the enhanced low-field strength low-resolution image; The system of claim 14 configured to:

17. The system of claim 16 , wherein the high-field data set is larger than the low-field data set.

18. The system of claim 14 , wherein the high-field strength magnetic resonance model is trained using high-field images acquired at high magnetic field strengths above 1 T.

19. 20. The system of claim 18, wherein the low-field images are acquired at a low magnetic field strength below 0.3 T.

20. 20. The system of claim 18, wherein the low-field images are acquired at a low magnetic field strength below 100 mT.

Images