Systems and methods for deep learning based accelerated magnetic resonance imaging for band extension coil sensitivity calibration

Abstract

Image reconstruction systems and methods include providing a sensitivity map of a coil of a magnetic resonance imaging (MRI) system to a neural network. The systems and methods also include providing interleaved k-space data to the neural network, where the interleaved k-space data includes partial k-space data interleaved with zero or synthetic k-space data to provide an extended field of view (FOV) that is different than a FOV utilized during acquisition of the partial k-space data, where the partial k-space data is obtained during a scan of a region of interest of the MRI system. The systems and methods also include outputting a final reconstructed MR image from the neural network based at least on the sensitivity map and the interleaved k-space data, where the final reconstructed MR image includes the FOV utilized during the acquisition of the partial k-space data.

Description

BACKGROUND

[0001] The subject matter disclosed herein relates to medical imaging, and more particularly, to systems and methods for deep learning based accelerated magnetic resonance imaging (MRI) for coil sensitivity calibration with extended field of view.

[0002] Non-invasive imaging techniques allow for obtaining images of internal structures or features of a patient / subject without performing an invasive procedure on the patient / subject. In particular, such non-invasive imaging techniques rely on various physical principles (such as differential transmission of X-rays through a target volume, reflection of sound waves within a volume, paramagnetism of different tissues and materials within a volume, decomposition of a target radioactive isotope within a body, etc.) to acquire data and construct images or otherwise represent the observed internal features of the patient / subject.

[0003] Recently, deep learning (DL) based reconstruction methods have been utilized in MRI to accelerate MR scans by exploiting prior information learned from historical MRI data. Most deep learning based reconstruction frameworks incorporate parallel imaging techniques (which require sensitivity information of a multi-channel coil array) to further increase the acceleration factor. Self-calibration is a widely used coil sensitivity calibration strategy (e.g., C3 calibration) that utilizes a fully sampled center k-space to estimate the sensitivity map. However, scans with limited field of view (FOV) can result in problematic sensitivity estimation, leading to degradation of the final reconstructed image. In particular, when MR scans are acquired in a limited FOV, in the phase encoding direction, tissue wrap effects (aliasing) can be amplified with iterative DL reconstruction, leading to degradation of the final reconstructed image. Additionally, certain reconstruction methods can increase computational time. Thus, an alternative method of providing sensitivity information is needed. SUMMARY

[0004] The following presents a summary of certain embodiments disclosed herein. It should be understood that these aspects are merely examples and are not intended to limit the scope of the disclosure. Indeed, the disclosure can encompass a variety of aspects that can not be set forth below.

[0005] In one embodiment, a DL-based image reconstruction system is provided. The system includes a memory that stores processor-executable routines. The system also includes a processing component configured to access the memory and execute the processor-executable routines, where the routines, when executed by the processing component, cause the processing component to perform actions. The actions include providing, to a neural network, a sensitivity map of a coil of a magnetic resonance imaging (MRI) system. The actions also include providing, to the neural network, interleaved k-space data, where the interleaved k-space data includes partial k-space data interleaved with zero or synthetic k-space data to provide an extended field of view (FOV) that is different than a FOV utilized during acquisition of the partial k-space data, where the partial k-space data is obtained during a scan of a region of interest of the MRI system. The actions also include outputting, from the neural network, at least one final reconstructed MR image based on at least the sensitivity map and the interleaved k-space data, where the at least one final reconstructed MR image includes the FOV utilized during acquisition of the partial k-space data.

[0006] In another embodiment, a DL-based image reconstruction method is provided. The method includes providing, via a processor, to a neural network, a sensitivity map of a coil of a magnetic resonance imaging (MRI) system. The method also includes providing, via the processor, to the neural network, interleaved k-space data, where the interleaved k-space data includes partial k-space data interleaved with zero or synthetic k-space data to provide an extended field of view (FOV) that is different than a FOV utilized during acquisition of the partial k-space data, where the partial k-space data is obtained during a scan of a region of interest of the MRI system. The method also includes outputting, via the processor, from the neural network, at least one final reconstructed MR image based on at least the sensitivity map and the interleaved k-space data, where the at least one final reconstructed MR image includes the FOV utilized during acquisition of the partial k-space data.

[0007] In another embodiment, a non-transitory computer-readable medium includes processor-executable code that, when executed by a processor, causes the processor to perform actions. The actions include providing, to a neural network, a sensitivity map of a coil of a magnetic resonance imaging (MRI) system. The actions also include providing, to the neural network, interleaved k-space data, where the interleaved k-space data includes partial k-space data interleaved with zero or synthetic k-space data to provide an extended field of view (FOV) that is different than a FOV utilized during acquisition of the partial k-space data, where the partial k-space data is obtained during a scan of a region of interest of the MRI system. The actions also include outputting, from the neural network, at least one final reconstructed MR image based on at least the sensitivity map and the interleaved k-space data, where the at least one final reconstructed MR image includes the FOV utilized during acquisition of the partial k-space data. BRIEF DESCRIPTION OF DRAWINGS

[0008] These and other features, aspects, and advantages of the present disclosure will become better understood when the following detailed description is read with reference to the accompanying drawings, in which like characters represent like parts throughout the drawings, wherein:

[0009] Figure 1 An embodiment of a magnetic resonance imaging (MRI) system suitable for use with the technology disclosed is shown;

[0010] Figure 2 is a schematic diagram illustrating the utilization of a neural network for DL-based MRI reconstruction with extended-FOV coil sensitivity calibration in accordance with aspects of the present disclosure;

[0011] Figure 3 is a flowchart of a method for DL-based MRI reconstruction with external coil sensitivity calibration in accordance with aspects of the present disclosure; and

[0012] Figure 4 A comparison of MR images utilizing external calibration reconstruction to MR images utilizing self-calibration reconstruction is shown. DETAILED DESCRIPTION

[0013] One or more specific embodiments will be described below. To provide a context for the various embodiments, a brief, general description of a suitable environment in which the embodiments can be implemented will be described now. In the interest of clarity, not all features of an actual implementation are described in this specification. It will of course be appreciated that in the development of any such actual implementation, numerous implementation-specific decisions must be made to achieve the developer's specific goals, such as compliance with system-related and business-related constraints, which will vary from one implementation to another. Moreover, it will be appreciated that such a development effort might be complex and time-consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill having the benefit of this disclosure.

[0014] When introducing elements of various embodiments of the present inventive subject matter, the articles "a," "an," "the," and "said" are intended to mean that there are one or more of the elements. The terms "comprising," "including," and "having" are intended to be inclusive and mean that there can be additional elements other than the listed elements. Additionally, any numerical examples in the following discussion are intended to be non-limiting, and thus additional numerical values, ranges, and percentages are within the scope of the disclosed embodiments.

[0015] While various aspects of the discussion below are provided in the context of medical imaging, it should be understood that the disclosed technology is not limited to such medical contexts. Rather, the examples and explanations provided in such medical contexts are merely for ease of explanation by providing examples of real-world implementations and applications. However, the disclosed technology can also be used in other contexts, such as image reconstruction for non-destructive inspection of manufactured parts or goods (i.e., quality control or quality review application scenarios) and / or non-invasive inspection of packages, boxes, luggage, etc. (i.e., security or screening application scenarios). Generally, the disclosed technology can be used in any imaging or screening context or image processing or photography field in which a set or class of acquired data undergoes a reconstruction process to generate an image or volume.

[0016] The deep learning (DL) methods discussed herein can be based on artificial neural networks, and thus can encompass one or more of the following: deep neural networks, fully interconnected networks, convolutional neural networks (CNNs), unfolded neural networks, perceptrons, encoders-decoders, recurrent networks, wavelet filter banks, u-nets, generative adversarial networks (GANs), or other neural network architectures. The neural networks can include shortcuts, activations, batch normalization layers, and / or other features. These techniques are referred to herein as DL techniques, but the term can also be used specifically with reference to the use of deep neural networks, which are neural networks with multiple layers.

[0017] As discussed herein, DL techniques (which can also be referred to as deep machine learning, hierarchical learning, or deep structured learning) are a branch of machine learning techniques that employ mathematical representations of data and artificial neural networks for learning and processing such representations. For example, DL methods can be characterized as using one or more algorithms to extract or model highly abstracted concepts of a class of data of interest. This can be done using one or more processing layers, where each layer generally corresponds to a different level of abstraction and thus can employ or utilize different aspects of the initial data or the output of a previous layer (i.e., a hierarchical or cascading structure of layers) as the target of the process or algorithm of a given layer. In the context of image processing or reconstruction, this can be characterized as different layers corresponding to different levels of features or resolutions in the data. Generally, a processing from one representation space to the next can be considered a “stage” of the process. Each stage of the process can be performed by a separate neural network or by a different portion of a larger neural network.

[0018] The present disclosure provides systems and methods for improved DL-based accelerated MRI reconstruction. In particular, external coil sensitivity calibration is utilized to improve image reconstruction. In particular, an extended FOV calibration MRI acquisition from which a coil sensitivity map is extracted is performed (e.g., a first FOV large enough to cover an imaged subject of interest), while partial or undersampled k-space data is obtained in a separate acquisition of the subject of interest (e.g., an accelerated MRI scan utilizing parallel imaging techniques) (e.g., a second FOV smaller than the first FOV). The extended FOV calibration MRI acquisition and data acquisition utilizing the smaller FOV can occur in the same imaging scan or in separate imaging scans. In separate scans, the extended FOV calibration scan is considered an external calibration MRI scan. The partial k-space data is interleaved with zero or synthetic k-space data to extend the FOV (e.g., in the phase encoding dimension). The interleaved k-space data with the extended FOV and the coil sensitivity map are input into a neural network (e.g., an unrolled neural network) that performs image reconstruction and outputs one or more final reconstructed images. The neural network can generate one or more reconstructed images from the interleaved k-space data with the extended FOV and the coil sensitivity map, and adjust the FOV of the reconstructed images (to the FOV utilized during acquisition of the partial k-space data) to generate the final reconstructed images. The improved DL-based accelerated MRI reconstruction utilizing external coil sensitivity calibration can minimize or eliminate aliasing artifacts (i.e., tissue wraparound effects) that typically occur with iterative DL reconstruction while maintaining image quality.

[0019] In view of the above, Figure 1 The magnetic resonance imaging (MRI) system 100 is illustratively shown as including a scanner 102, scanner control circuitry 104, and system control circuitry 106. In accordance with the embodiments described herein, the MRI system 100 is generally configured to perform MR imaging.

[0020] The system 100 also includes remote access and storage systems or devices, such as a picture archiving and communication system (PACS) 108, or other devices, such as a remote radiology facility, enabling on-site or off-site access to data acquired by the system 100. In this manner, MR data can be acquired and then processed and evaluated on-site or off-site. While the MRI system 100 can include any suitable scanner or detector, in the illustrated embodiment, the system 100 includes a whole-body scanner 102 having a housing 120 through which a bore 122 will be formed. A diagnostic table 124 is moveable into the bore 122 so that a patient 126 can be positioned therein for imaging of selected anatomical structures within the patient.

[0021] The scanner 102 includes a series of associated coils for generating controlled magnetic fields for exciting spin magnetic material within the anatomy of a subject being imaged. In particular, a primary magnet coil 128 is provided for generating a primary magnetic field B0that is generally aligned with the bore 122. A series of gradient coils 130, 132, and 134 allow controlled gradient magnetic fields to be generated during an examination sequence for positionally encoding certain spin magnetic nuclei within the patient 126. A radio frequency (RF) coil 136 is configured to generate RF pulses for exciting certain spin magnetic nuclei within the patient. In addition to the coils that can be local to the scanner 102, the system 100 includes a set of receive coils 138 (e.g., an array of coils) that are configured for placement proximal to the patient 126 (e.g., against the patient). For example, the receive coils 138 can include a cervical / thoracic / lumbar (CTL) coil, a head coil, a single-sided spine coil, etc. Generally, the receive coils 138 are placed proximal to the patient 126 or overhead so as to receive weak RF signals (weak relative to the transmitted pulses generated by the scanner coils) generated by certain spin magnetic nuclei within the patient 126 as the patient 126 returns to its relaxed state.

[0022] The various coils of the system 100 are controlled by external circuitry to generate the required fields and pulses in a controlled manner and to read the emissions from spin magnetic material. In the illustrated embodiment, a main power supply 140 provides power to the primary field coil 128 to generate the primary magnetic field B0. A power input 44 (e.g., power from a utility or power grid), a power distribution unit (PDU), a power supply (PS), and a driver circuit 150 together provide power to cause the gradient field coils 130, 132, and 134 to pulse. The driver circuit 150 can include amplification and control circuitry for supplying current to the coils in accordance with a defined pulse sequence output by the scanner control circuitry 104.

[0023] Another control circuit 152 is provided for regulating the operation of the RF coil 136. The circuit 152 includes a switching device for alternating between an active mode of operation and an inactive mode of operation in which the RF coil 136 respectively transmits and does not transmit a signal. The circuit 152 also includes an amplification circuit configured to generate RF pulses. Similarly, the receive coils 138 are connected to a switch 154 that is capable of switching the receive coils 138 between a receive mode and a non-receive mode. Thus, in the receive mode, the receive coils 138 are in resonance with RF signals emanating from spin magnetic nuclei within the patient 126, while in the non-receive mode, they are not in resonance with RF energy from the transmit coil (i.e., coil 136) so as to prevent unintended operation from occurring. Additionally, a receive circuit 156 is structured to receive data detected by the receive coils 138 and can include one or more multiplexing and / or amplification circuits.

[0024] It should be noted that while the scanner 102 and control / amplification circuitry are shown as being coupled by a single wire, in actual instances there can be many such wires. For example, separate wires can be used for control, data communication, power transmission, etc. In addition, appropriate hardware can be provided along each type of wire for proper handling of data and current / voltage. In fact, various filters, digitizers, and processors can be provided between the scanner and either or both of the scanner control circuitry 104 and the system control circuitry 106.

[0025] As shown, the scanner control circuitry 104 includes interface circuitry 158 that outputs signals used to drive the gradient field coils and RF coils and to receive data representative of the magnetic resonance signals produced in the examination. The interface circuitry 158 is coupled to control and analysis circuitry 160. Based on the protocol selected via the system control circuitry 106, the control and analysis circuitry 160 performs commands for driving the circuitry 150 and 152.

[0026] The control and analysis circuitry 160 is also used to receive the magnetic resonance signals and perform subsequent processing prior to transferring the data to the system control circuitry 106. The scanner control circuitry 104 also includes one or more memory circuitry 162 that stores configuration parameters, pulse sequence descriptions, examination results, etc. during operation.

[0027] Interface circuitry 164 is coupled to the control and analysis circuitry 160 for exchanging data between the scanner control circuitry 104 and the system control circuitry 106. In certain embodiments, the control and analysis circuitry 160, while shown as a single unit, can include one or more hardware devices. The system control circuitry 106 includes interface circuitry 166 that receives data from the scanner control circuitry 104 and transfers data and commands back to the scanner control circuitry 104. The control and analysis circuitry 168 can include a CPU on a general or application specific computer or workstation. The control and analysis circuitry 168 is coupled to memory circuitry 170 to store programming code for operating the MRI system 100, as well as to store processed image data for later reconstruction, display, and transfer. The programming code can execute one or more algorithms configured to perform reconstruction of acquired data as described below when executed by the processor. In certain embodiments, the memory circuitry 170 can store one or more neural networks for reconstruction of acquired data as described below. In certain embodiments, image reconstruction can occur on a separate computing device having processing circuitry and memory circuitry.

[0028] Additional interface circuits 172 can be provided for exchanging information with external systems, such as remote access and storage devices 108, for example, to exchange image data, configuration parameters, etc. Finally, system control and analysis circuitry 168 can be communicatively coupled to various peripheral devices for facilitating operator interfaces and producing hard copies of reconstructed images. In the illustrated embodiment, these peripheral devices include a printer 174, a display 176, and a user interface 178, including devices such as a keyboard, mouse, touch screen (e.g., integrated with display 176), etc.

[0029] Figure 2 is a schematic diagram illustrating the utilization of a neural network for DL-based MRI reconstruction for extended FOV coil sensitivity calibration (external coil sensitivity calibration when calibration data is acquired in a separate scan). The DL-based MRI reconstruction for extended FOV coil sensitivity calibration can be performed on the circuitry of the MRI system 100 in Figure 1 or on a separate computing device. As an alternative to the typical self-calibration approach, the coil sensitivity can be obtained with an extended FOV calibration scan (in a separate scan or the same scan). Unlike typical self-calibration, the extended FOV calibration acquisition can have a different prescribed FOV from the target sequence (i.e., an acquisition scan to obtain image data of a subject of interest). Thus, the calibration scan will have a large FOV to cover the image subject and avoid aliasing effects in the sensitivity calibration. Additionally, the external calibration scan parameters are optimized to maximize the image signal-to-noise ratio while minimizing other image artifact effects (e.g., chemical shift). Moreover, the scan time of the external calibration scan can be less than 10 seconds and thus does not significantly prolong the overall MR scan.

[0030] As Figure 2 depicted, the extended FOV calibration coil sensitivity can be utilized in combination with a neural network-based (e.g., unrolled neural network) reconstruction. As depicted, one or more initial images (MR images) 180, coil sensitivity maps 182, and extended FOV k-space data 184 are fed as inputs to a DL image reconstruction network 186 (e.g., neural network).

[0031] Expected under-sampled or partial k-space data 188 (e.g., multi-phase k-space data) is acquired from an MRI scan of an object of interest (e.g., a portion of a patient) at a first FOV. In certain embodiments, the k-space data can be acquired during a cine imaging sequence to capture motion (e.g., of a heart during different phases of a cardiac cycle). In certain embodiments, the MRI scan is an accelerated scan with parallel imaging techniques (e.g., where signals from individual coils are amplified, digitized, and processed along separate channels simultaneously) to reduce scan time. Extended FOV k-space data 184 is derived by interleaving the under-sampled k-space data 188 with zero (e.g., in the phase encoding dimension) or synthetic k-space data to extend the FOV such that it is different from (and larger than) the FOV utilized during the scan to acquire the under-sampled k-space data 188. One or more initial images 180 are reconstructed by coil-combined reconstruction of the extended FOV k-space data 184.

[0032] The coil sensitivity maps 182 are extracted from data acquired during an external calibration scan (at a second FOV) separate from the accelerated scan of the object of interest. For example, the external calibration scan can occur before or after the accelerated scan. In certain embodiments, the coil sensitivity maps 182 are extracted from calibration data acquired (at a second FOV) during the same scan. The second FOV is different from (and larger than) the first FOV (utilized during the acquisition of the k-space data). The coil sensitivity maps 182 are extracted from the calibration data according to both the location of the target scan (e.g., the accelerated scan) and the extended FOV of the extended FOV k-space data 184. The coil sensitivity maps 182 include a calibration FOV that matches the extended FOV. For all channels, regions outside of a central k-space region of the calibration FOV are padded with zero.

[0033] As depicted, the DL image reconstruction network 186 is an unrolled neural network. In particular, according to certain aspects, image reconstruction is represented as an unrolled (i.e., non-cyclic) process in which all computational steps occur in a fixed sequence without repetition, and in which data fidelity computation and / or image update computation are incorporated into the neural network. Due to the unrolled nature of the iterative steps, different neural networks or models can be employed at different stages or steps of the process in order to optimize the performance of the reconstruction process. Alternatively, the same neural network or model can be used at multiple (or all) of the repeated steps (e.g., data fitting steps and / or image update steps) of the unrolled process. In contrast, unrolled (i.e., cyclic) iterative processes repeat the same operations at each iteration. The network 186 can be trained by retrospectively under-sampling thousands of fully-sampled images. The fully-sampled images can be used as ground truth. In certain embodiments, the network 186 can utilize different types of neural networks.

[0034] The DL image reconstruction network 186 includes a plurality of step or iteration blocks 189 (e.g., Step 1 through Step N as depicted). The number of step blocks 189 can vary. The step blocks 189 are connected from one step block to the next step block by direct connections. Each step or iteration block 189 includes a data consistency layer or unit 190 and an image domain neural network layer or regularization unit 192. The data consistency layer 190 is configured to preserve fidelity of the coil data in the output of the respective step or iteration block. The image domain neural network layer 192 is configured to generate regularization information based on the respective output from the respective proceeding iteration block or step. The regularization information represents additional information needed to reduce generalization error in the output of each iteration block or step.

[0035] The image domain neural network layer 192 can be formed from a fully convolutional residual network (e.g., a 3D spatio-temporal convolutional neural network). Each image domain neural network 192 can be composed of a plurality of 3D convolutional layers (e.g., each 3D convolutional layer having 3x3x3 kernels) that utilize circular padding along the phase-encoding and temporal directions to enforce circular boundary conditions in two dimensions. An initial convolution expands one or more images into feature maps that propagate through the network until a final convolution that reassembles the feature maps into the original number of input images. Prior to each convolutional layer are ReLU pre-activation layers that operate individually on each image / feature channel.

[0036] As depicted, the one or more initial images 180, the coil sensitivity map 182, and the extended FOV k-space data 184 are fed into the data consistency layer 190 of Step 1, which is coupled to and provides one or more inputs to the image domain neural network layer 192 via a single image channel. In certain embodiments, the data consistency layer 190 and the image domain neural network layer 192 are coupled via multiple image channels. One or more output images (as well as the coil sensitivity map 182 and the extended FOV k-space data 184) are generated by the image domain neural network layer 192 and provided to the data consistency layer 190 of the next step or iteration block 189. This process occurs at each subsequent step. The final step 189 or iteration block (i.e., Step N) outputs one or more reconstructed images 194. The FOV of the one or more reconstructed images 194 (which are at an extended FOV) is changed to the FOV utilized during acquisition of the undersampled k-space data 188 to generate one or more final reconstructed images 196 (MR images). Each reconstructed image 194 generated by the network 186 is a multi-channel output, which is combined into a single channel in the final reconstructed image 196. The final reconstructed image 196 is free of aliasing artifacts.

[0037] Figure 3 is a flowchart of a method 198 for DL-based MRI reconstruction with external coil sensitivity calibration.Figure 1 One or more components of the MRI system 100 can be used to perform method 198. One or more steps of method 198 can be performed simultaneously or in conjunction with... Figure 3 The different sequences of execution are shown. Method 198 includes performing a separate external calibration scan (box 200) using an MRI system to obtain calibration data 202. Method 198 also includes performing an accelerated scan (e.g., using parallel imaging techniques) (box 204) on an object of interest (e.g., a portion of a patient's body) using an MRI system to obtain partial or undersampled k-space data (e.g., multiphase k-space data). The separate external calibration scan can occur before or after the accelerated scan. In some embodiments, the external calibration scan and the accelerated scan can utilize different acquisition sequences. For example, the external calibration scan can utilize a 3D GRE-based coil calibration sequence, while the accelerated scan can utilize a 2D accelerated cine sequence or a balanced steady-state gradient echo sequence. In some embodiments, calibration data can be acquired during the same scan. In either case, the FOV used to acquire the calibration data is greater than the FOV utilized during the accelerated portion of the scan.

[0038] Method 198 includes expanding the FOV of the undersampled k-space data 206 to generate expanded FOV k-space data 208 (box 210). The expanded FOV k-space data 208 is formed by interleaving the undersampled k-space data 206 with zero or synthetic k-space data. The expanded FOV of the expanded FOV k-space data 208 is larger than the FOV used to acquire the undersampled k-space data but smaller than the FOV of the calibration scan. Method 198 also includes reconstructing one or more initial images 214 from the undersampled k-space data 206.

[0039] Method 198 further includes extracting a coil sensitivity map 216 (box 218) from calibration data 202. The coil sensitivity map 216 is extracted from external calibration data 202 based on both the location of the target scan (e.g., accelerated scan) and the extended FOV of the extended FOV k-space data 208. In some embodiments, the coil sensitivity map is extracted from calibration data acquired during the target scan. The coil sensitivity map 216 includes an external calibration FOV that matches the extended FOV. For all channels, the area outside the central k-space region of the calibration FOV is zero-padded.

[0040] Method 198 further includes providing a sensitivity map 216, extended FOV k-space data 208, and one or more initial images 212 to a reconstruction network 220 (e.g., an expanded neural network), which generates one or more reconstructed images 222 (box 224) based on the sensitivity map 216, extended FOV k-space data 208, and one or more initial images 212. The reconstruction network 220 is as follows... Figure 2The reconstructed image 222 has an extended FOV. The method 198 further includes adjusting the FOV of the reconstructed image 222 to generate a final reconstructed image 226 (block 228). The final reconstructed image 226 has a FOV utilized in the undersampling of the k-space data 206. The final reconstructed image 226 is free of aliasing artifacts.

[0041] Figure 4 A comparison of MR images reconstructed with external calibration versus MR images reconstructed with self-calibration is shown. Top row MR images 230 of a heart are reconstructed with the above-described external calibration based unrolling neural network based reconstruction. Bottom row MR images 232 of the heart are reconstructed with a deep learning based neural network that utilizes self-calibration (e.g., ESPRIT calibration that derives coil sensitivity maps from data acquired with a smaller or truncated FOV). Each row of MR images 230, 232 includes a corresponding magnified image of a region 234 (e.g., dashed outline) of the heart. The scans are performed with a 1.5T scanner with a 30-channel abdominal coil array. The scans include two sequences, a 3D Cartesian gradient echo (GRE) based coil calibration sequence and a separate 2D accelerated cine sequence. The 3D GRE sequence is used for coil calibration. The scan FOV of the 3D GRE sequence is 50 centimeters (cm). The spatial resolution of the 3D GRE sequence is 1.5 cm. The readout bandwidth of the 3D GRE sequence is 62.5 kilohertz (kHz). The 2D accelerated cine sequence is a 2D Cartesian FIESTA sequence with 12x acceleration and variable density k-t sampling for acquiring short axis cine images. The scan FOV of the 2D accelerated cine sequence is 36 cm. The spatial resolution of the 2D accelerated sequence is 1.4 millimeters. The readout bandwidth of the 2D accelerated sequence is 15 kHz. The image capture is of a cardiac cycle from systole to diastole. In the region indicated by arrow 236 in the bottom row of images 232, there are aliasing artifacts. In the same region indicated by arrow 236 in the top row of images 230, there are no aliasing artifacts. Additionally, the quality of the two rows of images 230, 232 is the same. Thus, image reconstruction with external calibration produces images of similar quality (without aliasing artifacts) compared to ESPRIT based reconstruction.

[0042] A technical effect of the disclosed subject matter involves providing a system and method for DL based MRI reconstruction with extended FOV coil sensitivity calibration (e.g., external coil sensitivity calibration when derived from a separate scan). Improved DL based accelerated MRI reconstruction with extended FOV coil sensitivity calibration can minimize or eliminate aliasing artifacts (i.e., tissue wraparound effect) that typically occur with iterative DL reconstruction while maintaining image quality.

[0043] Reference to the technology presented in this specification and to the examples given herein is not, and should not be taken as, an acknowledgement or any form of suggestion that the technology is not prior art or that it would be a prior art to a presently claimed invention.

[0044] This written description uses examples to disclose the subject matter, including the best mode, and also to enable any person skilled in the art to practice the subject matter, including making and using any devices or systems and performing any incorporated methods. The patentable scope of the subject matter is defined by the claims, and can include other examples that occur to those skilled in the art. Such other examples are intended to fall within the scope of the claims if they have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements with insubstantial differences from the literal languages of the claims.

Claims

1. A deep learning-based image reconstruction system, the image reconstruction system comprising: A memory that encodes processor-executable routines; A processing unit configured to access the memory and execute the processor-executable routine, wherein the processor-executable routine, when executed by the processing unit, causes the processing unit to: The sensitivity map of the coils of the magnetic resonance imaging (MRI) system is provided to the neural network, wherein the sensitivity map includes an extended calibrated field of view (FOV). Interleaved k-space data is provided to the neural network, wherein the interleaved k-space data includes partial k-space data interleaved with zero or synthesized k-space data to provide an extended field of view (FOV) different from that used during the acquisition of the partial k-space data, wherein the partial k-space data is acquired during scanning of the region of interest of the MRI system. and At least one final reconstructed MR image is output from the neural network based on at least the sensitivity map and the interleaved k-space data, wherein the at least one final reconstructed MR image includes the FOV utilized during the acquisition of the portion of the k-space data.

2. The image reconstruction system according to claim 1, wherein outputting the at least one final reconstructed MR image using the neural network comprises: At least one reconstructed MR image is generated via the neural network based at least on the sensitivity map and the interleaved k-space data, wherein the at least one reconstructed MR image is located at the extended FOV; as well as The at least one reconstructed MR image is adjusted to the FOV utilized during the acquisition of the partial k-space data to generate the at least one final reconstructed MR image.

3. The image reconstruction system of claim 2, wherein adjusting the at least one reconstructed MR image to the FOV utilized during the acquisition of the partial k-space data comprises combining the multi-channel outputs of the at least one reconstructed MR image into a single-channel output for the at least one final reconstructed MR image.

4. The image reconstruction system according to claim 1, wherein the extended FOV is set in the phase encoding dimension.

5. The image reconstruction system according to claim 1, wherein the neural network comprises an unfolded neural network.

6. The image reconstruction system of claim 1, wherein the scanning includes accelerated scanning utilizing parallel imaging.

7. The image reconstruction system of claim 1, wherein, when executed by the processing unit, the processor executes a routine that causes the processing unit to extract the sensitivity map from calibration data acquired during the scan or a separate external calibration scan, based on both the location of the scan and the extended FOV.

8. The image reconstruction system of claim 7, wherein the FOV of the scan or the separate external calibration scan used to acquire the calibration data is greater than the FOV utilized during the acquisition of the partial k-space data.

9. The image reconstruction system of claim 8, wherein for all channels, the region outside the central k-space region of the extended calibration FOV is filled with zeros.

10. A deep learning-based image reconstruction method, the image reconstruction method comprising: Sensitivity maps of the coils of a magnetic resonance imaging (MRI) system are provided to a neural network via a processor, wherein the sensitivity maps include an extended calibrated field of view (FOV). Interleaved k-space data is provided to the neural network via the processor, wherein the interleaved k-space data includes partial k-space data interleaved with zero or synthesized k-space data to provide an extended FOV different from the field of view (FOV) utilized during the acquisition of the partial k-space data, wherein the partial k-space data is acquired during the scanning of the region of interest of the MRI system. as well as At least one final reconstructed MR image is output from the neural network via the processor, based at least on the sensitivity map and the interleaved k-space data, wherein the at least one final reconstructed MR image includes the FOV utilized during the acquisition of the portion of the k-space data.

11. The image reconstruction method of claim 10, wherein outputting the at least one final reconstructed MR image from the neural network via the processor comprises: At least one reconstructed MR image is generated via the neural network based at least on the sensitivity map and the interleaved k-space data, wherein the at least one reconstructed MR image is located at the extended FOV; as well as The processor adjusts the at least one reconstructed MR image to the FOV utilized during the acquisition of the partial k-space data to generate the at least one final reconstructed MR image.

12. The image reconstruction method of claim 11, wherein adjusting the at least one reconstructed MR image to the FOV utilized during the acquisition of the partial k-space data comprises combining the multi-channel outputs of the at least one reconstructed MR image into a single-channel output for the at least one final reconstructed MR image.

13. The image reconstruction method according to claim 10, wherein the extended FOV is set in the phase encoding dimension.

14. The image reconstruction method according to claim 10, wherein the neural network includes an unfolded neural network.

15. The image reconstruction method of claim 10, wherein the individual scans include accelerated scans utilizing parallel imaging.

16. The image reconstruction method of claim 10, comprising extracting the sensitivity map from calibration data acquired during the scan or a separate external calibration scan via the processor based on both the location of the scan and the extended FOV.

17. The image reconstruction method of claim 16, wherein the FOV of the scan or the separate external calibration scan used to acquire the calibration data is greater than the FOV utilized during the acquisition of the partial k-space data.

18. The image reconstruction method of claim 17, wherein for all channels, the region outside the central k-space region of the extended calibration FOV is filled with zeros.

19. A non-transitory computer-readable medium, the computer-readable medium comprising processor-executable code, the processor-executable code causing the processor, when executed by a processor, to: The sensitivity map of the coils of the magnetic resonance imaging (MRI) system is provided to the neural network, wherein the sensitivity map includes an extended calibrated field of view (FOV). Interleaved k-space data is provided to the neural network, wherein the interleaved k-space data includes partial k-space data interleaved with zero or synthesized k-space data to provide an extended field of view (FOV) different from that used during the acquisition of the partial k-space data, wherein the partial k-space data is acquired during scanning of the region of interest of the MRI system; and At least one final reconstructed MR image is output from the neural network based on at least the sensitivity map and the interleaved k-space data, wherein the at least one final reconstructed MR image includes the FOV utilized during the acquisition of the portion of the k-space data.

20. The non-transitory computer-readable medium of claim 19, wherein the scanning includes accelerated scanning utilizing parallel imaging.

Images