A system and method for generating realistic contrast images using deep learning based on synthetic magnetic resonance imaging data.

Abstract

This invention provides a computer-implemented method for generating an artifact-corrected reconstructed contrast image from magnetic resonance imaging (MRI) data. The method includes inputting a synthetic contrast image derived from multi-latency multi-echo (MDME) scan data, or both MDME scan data acquired during a first scan of an object of interest using an MDME sequence and a composite image, into a trained deep neural network. The composite image is derived from both the MDME scan data and contrast scan data acquired during a second scan of the object of interest using a contrast MRI sequence. The method further includes generating the artifact-corrected reconstructed contrast image based on the synthetic contrast image or both the MDME scan data and the composite image using the trained deep neural network. The method further includes outputting the artifact-corrected reconstructed contrast image from the trained deep neural network.

Description

Background Technology

[0001] The topics disclosed in this article relate to medical imaging, and more specifically to systems and methods for generating realistic contrast images based on deep learning using synthetic magnetic resonance imaging data.

[0002] Non-invasive imaging techniques allow for the acquisition of images of a patient's or subject's internal structure or features without the need for invasive procedures. Specifically, such non-invasive imaging techniques rely on various physical principles (such as differential transmission of X-rays through a target volume, sound wave reflection within the volume, paramagnetism of different tissues and materials within the volume, and the disintegration of the target radionuclide within the body) to acquire data and construct images or otherwise represent the observed internal features of the patient or subject.

[0003] In conventional MRI, generating multiple image contrasts for an object of interest requires separate scans for each type of contrast, each lasting several minutes. Recently, a technique has been developed to synthetically generate multiple image contrasts from MRI data acquired in a single scan, thereby reducing scan time. However, some of these synthetically generated MR contrast images lack the quality of conventionally acquired MR contrast images. For example, synthetically generated MR contrast images may include artifacts (e.g., artificial brightening in certain areas of the imaging structure), which interfere with the diagnostic value of the synthetically generated MR contrast images. Attempts to address these issues have primarily focused on utilizing deep learning-based techniques, which still involve generating fully synthetic MR contrast images. Summary of the Invention

[0004] The following provides an overview of some embodiments disclosed herein. It should be understood that these aspects are provided merely to give the reader a brief overview of these particular embodiments, and are not intended to limit the scope of this disclosure. In fact, this disclosure may cover aspects that may not be shown below.

[0005] In one embodiment, a computer-implemented method is provided for generating an artifact-corrected reconstructed contrast image from magnetic resonance imaging (MRI) data. The method includes inputting a synthetic contrast image derived from multi-latency multi-echo (MDME) scan data, or both MDME scan data acquired during a first scan of an object of interest using an MDME sequence and a composite image, into a trained deep neural network, wherein the composite image is derived from both the MDME scan data and contrast scan data acquired during a second scan of the object of interest using a contrast MRI sequence. The method further includes generating the artifact-corrected reconstructed contrast image based on the synthetic contrast image or both the MDME scan data and the composite image using the trained deep neural network. The method further includes outputting the artifact-corrected reconstructed contrast image from the trained deep neural network.

[0006] In another embodiment, a deep learning-based artifact correction system is provided for generating artifact-corrected reconstructed contrast images from magnetic resonance imaging (MRI) data. The system includes memory encoding processor-executable routines. The system also includes a processing unit configured to access the memory and execute processor-executable routines, wherein these routines, when executed by the processing unit, cause the processing unit to perform actions. These actions include inputting a synthetic contrast image derived from multi-latency multi-echo (MDME) scan data, or both MDME scan data acquired during a first scan of the object of interest using an MDME sequence and a composite image, into a trained deep neural network, wherein the composite image is derived from both the MDME scan data and contrast scan data acquired during a second scan of the object of interest using a contrast MRI sequence, wherein the image intensity of the MDME scan data is normalized to the image intensity of the contrast scan data before generating the composite image. The actions also include generating an artifact-corrected reconstructed contrast image using the trained deep neural network based on the synthetic contrast image or both the MDME scan data and the composite image. The actions further include outputting the artifact-corrected reconstructed contrast image from the trained deep neural network.

[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 inputting a synthetic contrast image derived from multi-latency multi-echo (MDME) scan data, or both MDME scan data and a composite image acquired during a first scan of the object of interest using an MDME sequence, into a trained deep neural network. The composite image is derived from both the MDME scan data and contrast scan data acquired during a second scan of the object of interest using a contrast MRI sequence. The image intensity of the MDME scan data is normalized to the image intensity of the contrast scan data before generating the composite image. The actions also include generating an artifact-corrected reconstructed contrast image based on the synthetic contrast image or both the MDME scan data and the composite image using the trained deep neural network. The actions further include outputting the artifact-corrected reconstructed contrast image from the trained deep neural network. Attached Figure Description

[0008] These and other features, aspects, and advantages of this disclosure will be better understood when the following detailed description is read with reference to the accompanying drawings, in which the same symbols denote the same parts throughout the drawings, wherein:

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

[0010] Figure 2 This is a flowchart of a method for DL-based MRI reconstruction of true contrast images according to various aspects of this disclosure;

[0011] Figure 3 It is a schematic diagram of a grafting operation for generating composite data, based on various aspects of this disclosure;

[0012] Figure 4 This is a schematic diagram illustrating the use of an artifact correction network (using only center k-space data from contrast data) according to various aspects of this disclosure;

[0013] Figure 5 This is a schematic diagram illustrating the use of an artifact correction network (using random k-space data from contrast data) according to various aspects of this disclosure.

[0014] Figure 6 A comparison of MR contrast images generated using different techniques is shown; and

[0015] Figure 7 A comparison of MR contrast images generated using different techniques is shown. Detailed Implementation

[0016] One or more specific implementations will be described below. To provide a concise description of these implementations, not all features of an actual implementation will be described in this specification. It should be understood that in the development of any such actual implementation, as in any engineering or design project, many implementation-specific decisions must be made to achieve the developer's specific objectives, such as complying with system-related and business-related constraints that may differ from implementation to implementation. Furthermore, it should be understood that such development efforts may be complex and time-consuming, but remain routine tasks of design, fabrication, and manufacturing for those skilled in the art who benefit from this disclosure.

[0017] When describing the elements of various embodiments of the subject matter of this invention, the articles “a,” “an,” “the,” and “the” are intended to indicate the presence of one or more of the stated elements. The terms “comprising,” “including,” and “having” are intended to be inclusive and mean that additional elements may be present in addition to the listed elements. Furthermore, any numerical examples in the following discussion are intended to be non-limiting, and therefore the additional numerical values, ranges, and percentages are within the scope of the disclosed embodiments.

[0018] While the various aspects discussed below are provided in the context of medical imaging, it should be understood that the techniques disclosed in this invention are not limited to such medical contexts. In fact, the examples and explanations provided in such medical contexts are merely for the purpose of facilitating explanation by providing examples of real-world implementations and applications. However, the techniques disclosed in this invention 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 inspection applications) and / or non-invasive inspection of packages, boxes, suitcases, etc. (i.e., security screening or screening applications). Generally speaking, the techniques disclosed in this invention can be used in any imaging or screening context or in image processing or photography field, where a set or class of acquired data undergoes a reconstruction process to generate an image or volume.

[0019] The deep learning (DL) methods discussed in this paper can be based on artificial neural networks and therefore may 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), dense neural networks, or other neural network architectures. Neural networks may include shortcuts, activations, batch normalization layers, and / or other features. These techniques are referred to as DL techniques in this paper, but the term may also be used specifically with reference to the use of deep neural networks, which are neural networks with multiple layers.

[0020] As discussed in this paper, deep learning (DL) techniques (also known 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 used to learn and process such representations. For example, DL methods can be characterized as using one or more algorithms to extract or model a class of highly abstract concepts from data of interest. This can be accomplished using one or more processing layers, where each layer typically corresponds to a different level of abstraction and may therefore take or utilize different aspects of the initial data or the output of the previous layer (i.e., the hierarchical or cascaded structure of the layers) as the target of the process or algorithm for a given layer. In the context of image processing or reconstruction, this can be characterized as different layers corresponding to different feature levels or resolutions in the data. Generally, the processing from one representation space to the next level of representation space can be viewed as a “stage” of a process. Each stage of the process can be performed by a single neural network or by different parts of a larger neural network.

[0021] This disclosure provides systems and methods for generating realistic (i.e., equivalent to conventional) contrast images from synthetic MRI data based on deep learning. Specifically, artifact-corrected contrast images (i.e., realistic contrast images) are reconstructed using a trained deep neural network (e.g., a deep learning module or a Distance Between Simulation and Observational Indices (DISO) model). Composite data or composite images can be generated by grafting a portion of k-space data (unacquired k-space) from a synthetic contrast image or image onto contrast data or contrast images (contrast images only), which contain at least a central portion of k-space data with or without other k-space regions remote from the center. Scans for acquiring data for the synthetic contrast image and the contrast images can occur in a single consecutive imaging session (i.e., where scans occur sequentially in response to a single initiation signal or input (e.g., a click)). Before grafting, the image intensity of the synthetic contrast image can be normalized to the image intensity of the contrast images. The artifact-corrected reconstructed contrast image output by the trained deep neural network has fewer artifacts than a synthetic contrast image of the same contrast type. Furthermore, the artifact-corrected reconstructed image has similar diagnostic quality to conventionally acquired contrast images. Therefore, the disclosed implementation enables the generation of true contrast images while still utilizing data acquired using MDME sequences in a single-click multi-contrast imaging scheme.

[0022] Considering the above, Figure 1 A magnetic resonance imaging (MRI) system 100 is schematically shown as including a scanner 102, scanner control circuitry 104, and system control circuitry 106. According to the embodiments described herein, the MRI system 100 is typically configured to perform MR imaging.

[0023] System 100 also includes: a remote access and storage system or device, such as an image archiving and communication system (PACS) 108; or other devices, such as a remote radiology device, enabling on-site or remote access to data acquired by system 100. This allows MR data to be acquired and then processed and evaluated on-site or remotely. While MRI system 100 may include any suitable scanner or detector, in the illustrated embodiment, system 100 includes a whole-body scanner 102 with a housing 120 through which an aperture 122 is formed. A diagnostic table 124 is movable into the aperture 122, allowing a patient 126 to be positioned therein for imaging of selected anatomical structures within the patient's body.

[0024] Scanner 102 includes a series of associated coils for generating a controlled magnetic field used to excite gyromagnetic material within the anatomical structures of the subject being imaged. Specifically, a primary magnet coil 128 is provided to generate a primary magnetic field B0 generally aligned with aperture 122. A series of gradient coils 130, 132, and 134 allow the generation of a controlled gradient magnetic field during the examination sequence for positional encoding of certain gyromagnetic nuclei within patient 126. Radio frequency (RF) coil 136 is configured to generate radio frequency pulses for exciting certain gyromagnetic nuclei within patient 126. In addition to the coils that may be located locally on scanner 102, system 100 also includes a set of receiving coils 138 (e.g., coil array) configured for placement proximal to patient 126 (e.g., against patient). For example, receiving coils 138 may include cervical / thoracic / lumbar (CTL) coils, head coils, single-sided spinal coils, etc. Generally, the receiving coil 138 is placed near or above the patient 126 in order to receive weak RF signals generated by certain ferromagnetic nuclei in the patient's body when the patient 126 returns to its relaxed state (weak in relation to the transmission pulse generated by the scanner coil).

[0025] The various coils of system 100 are controlled by external circuitry to generate the desired fields and pulses in a controlled manner and to read out emissions from the gyromagnetic material. In an illustrated embodiment, a main power supply 140 powers the primary field coil 128 to generate a primary magnetic field Bo. Power input 44 (e.g., power from a utility or grid), a power distribution unit (PDU), a power supply (PS), and driver circuitry 150 may together provide power to cause the gradient field coils 130, 132, and 134 to generate pulses. Driver circuitry 150 may include amplification and control circuitry for supplying current to the coils according to a defined sequence of digitized pulses output by scanner control circuitry 104.

[0026] Another control circuit 152 is provided for regulating the operation of the RF coil 136. Circuit 152 includes a switching device for alternating between an active operating mode and a passive operating mode, wherein the RF coil 136 transmits a signal and does not transmit a signal, respectively. Circuit 152 also includes an amplifier circuit configured to generate RF pulses. Similarly, a receiving coil 138 is connected to a switch 154 capable of switching the receiving coil 138 between a receiving mode and a non-receiving mode. Thus, in receiving mode, the receiving coils 138 resonate with the RF signal generated by the release of the vortex nucleus within the patient 126, while in non-receiving mode, they do not resonate with the RF energy from the transmitting coil (i.e., coil 136) to prevent unintended operation. Additionally, the receiving circuit 156 is configured to receive data detected by the receiving coil 138 and may include one or more multiplexing and / or amplification circuits.

[0027] It should be noted that although the scanner 102 and the control / amplification circuit described above are shown connected by a single wire, in practical instances, many such wires may exist. For example, separate wires may be used for control, data communication, power transmission, etc. Furthermore, appropriate hardware may be provided along each type of wire for proper processing of data and current / voltage. In practice, various filters, digitizers, and processors may be provided between the scanner and either or both of the scanner control circuit 104 and system control circuit 106.

[0028] As shown in the figure, the scanner control circuit 104 includes an interface circuit 158 ​​that outputs signals for driving the gradient field coil and the RF coil, and for receiving data representing the magnetic resonance signals generated in the examination sequence. The interface circuit 158 ​​is connected to a control and analysis circuit 160. Based on a defined scheme selected via the system control circuit 106, the control and analysis circuit 160 executes commands for driving circuits 150 and 152.

[0029] The control and analysis circuit 160 is also used to receive magnetic resonance signals and perform subsequent processing before transmitting the data to the system control circuit 106. The scanner control circuit 104 also includes one or more memory circuits 162 that store configuration parameters, pulse sequence descriptions, inspection results, etc. during operation.

[0030] Interface circuitry 164 is coupled to control and analysis circuitry 160 for exchanging data between scanner control circuitry 104 and system control circuitry 106. In some embodiments, control and analysis circuitry 160, while shown as a single unit, may include one or more hardware devices. System control circuitry 106 includes interface circuitry 166 that receives data from scanner control circuitry 104 and transmits data and commands back to scanner control circuitry 104. Control and analysis circuitry 168 may include a CPU on a general-purpose or application-specific computer or workstation. Control and analysis circuitry 168 is coupled to memory circuitry 170 to store programming code for operating the MRI system 100 and to store processed image data for subsequent reconstruction, display, and transmission. The programming code may execute one or more algorithms configured to perform reconstruction of the acquired data as described below when executed by a processor. In some embodiments, memory circuitry 170 may store one or more neural networks for reconstruction of the acquired data as described below. In some embodiments, image reconstruction may occur on a separate computing device having processing circuitry and memory circuitry.

[0031] Additional interface circuitry 172 may be provided for exchanging image data, configuration parameters, etc., with external system components such as remote access and storage devices 108. Finally, system control and analysis circuitry 168 may be communicatively coupled to various peripheral devices to facilitate the operator interface and generate hard copies of the reconstructed images. In the illustrated embodiment, these peripheral devices include a printer 174, a display 176, and a user interface 178, which includes devices such as a keyboard, mouse, and touchscreen (e.g., integrated with display 176).

[0032] Figure 2 This is a flowchart of a DL-based MRI reconstruction method 180 for a true contrast image (e.g., an artifact-corrected reconstructed contrast image) that is equivalent to a conventional contrast image. Figure 1 One or more components of the MRI system 100 can be used to perform method 180. One or more steps of method 180 can be performed simultaneously or in conjunction with... Figure 2The different sequences of execution are shown. Method 180 includes performing sequential scans on the same object of interest or subject (e.g., for regions such as the brain) to acquire MDME scan data (e.g., for all spin echo sequences) and contrast scan data (box 182). The raw MDME scan data and contrast scan data are k-space data. Specifically, quantitative scanning is performed using MDME sequences. Multiple image contrasts can be generated from the MDME scan data acquired in a single scan. Specifically, MR signals acquired with MDME sequences can be used to synthesize computer parametric maps (e.g., T1 maps, T2 maps, proton density maps, etc.) based on a predefined model that describes the behavior of MR signals pixel-wise (or voxel-wise). Synthetic images (e.g., T1-weighted and T2-weighted images, T1-FLAIR and T2-FLAIR images, STIR images, DIR images, and / or proton density-weighted images) are then generated pixel-wise (or voxel-wise) based on the model using the parametric maps and operator-specified parameters (e.g., TE, TR, delay). Synthetic contrast images are comparable to those generated using operator-specified parameters in actual or conventional MR scans. However, these synthetic contrast images may include artifacts (e.g., artificial brightening), which make them less than ideal for diagnostic purposes.

[0033] As part of sequential scanning, for a specific contrast type, accelerated contrast scanning (e.g., acceleration factors of 2X, 3X, 4X, 5X, etc.) is performed using MRI contrast sequences (e.g., T1-weighted and T2-weighted, T1-FLAIR and T2-FLAIR, STIR, DIR, and / or proton density-weighted)). Accelerated scanning can be performed using parallel imaging techniques (e.g., where signals from individual coils are amplified, digitized, and processed simultaneously along individual channels) to reduce scan time. Contrast scan data is undersampled or partially k-space data. Contrast scan data from accelerated scanning may be low-frequency-only contrast scan data. In some embodiments, contrast scan data includes only the centerline or center k-space used to provide contrast information. For example, for acceleration factors of 2X, 3X, 4X, and 5X, the contrast scan data includes 50%, 33%, 25%, and 20% of the center-filled k-space, respectively. In other embodiments, contrast scan data includes the centerline or center k-space and k-space outside of other random lines or regions of the center k-space. For example, in regions outside the central k-space, undersampled contrast scan data is interleaved or zero-padding (e.g., in the phase-encoded dimension).

[0034] The scans used to acquire MDME scan data and contrast scan data can occur within a single continuous imaging session (i.e., where scans (accelerated and quantitative scans) occur sequentially in response to a single start signal or input (e.g., a click). A quantitative scan can be performed first, followed by an accelerated scan, and vice versa. In some embodiments, the acquisition of MDME scan data and contrast scan data can occur in separate scans (i.e., using separate start signals).

[0035] Method 180 further includes normalizing the image intensity range of the MDME scan data to the image intensity of the contrast scan data (box 184). Method 180 further includes grafting portions of the MDME scan data into the contrast scan data to form or generate composite data (box 186). Grafting serves as a structure-sharing operation. Different contrasts derived from the MDME data share structural information (e.g., high-frequency information). Since the scan data originate from the same subject, the contrasts derived from the MDME data also share structural information with the contrasts derived from the accelerated scan. High-frequency information (from regions outside the central region) from the MDME scan data (e.g., reference data) is grafted into zero-filled regions of the contrast scan data to form composite data (e.g., the grafted k-space). Normalization is performed prior to grafting to minimize grafting artifacts due to gain mismatch.

[0036] Method 180 further includes converting the MDME scan data and composite data (box 188). Specifically, the MDME scan data is converted into a synthetic contrast image 190 (e.g., a two-dimensional (2D) image), and the composite data is converted into a composite image 192. For example, the k-space data is converted into image data via Fourier transform (e.g., inverse fast Fourier transform (IFFT)). The composite image 192 and the synthetic contrast image 190 (or MDME scan data) are input into a trained deep neural network 194 (e.g., an artifact correction network or an artifact prediction network) (box 196). Furthermore, the synthetic contrast image 190 corresponds to the type of contrast scan sequence used to provide contrast scan data to the composite image 192. For example, if the regular contrast scan for acquiring the contrast scan data is T2-FLAIR, then the composite image 192 will be a synthetic T2-FLAIR image. The trained deep neural network 194 is trained (e.g., via supervised learning) to predict artifacts to be removed in order to generate a real or conventional contrast image that approximates the ground truth image (i.e., a conventionally acquired contrast image, such as a T2-FLAIR image). Method 180 further includes generating a real (e.g., conventional) contrast image 198 (box 200).

[0037] The trained deep neural network 194 (e.g., the DISO model) can be a dense neural network. The layers of a dense neural network are fully connected (e.g., dense) through neurons within the network layers. Each neuron in a layer receives input from all neurons present in the previous layer. Linear operations are used, where each input is connected to each output via weights. In some implementations, each dense block may include three layers. The output of the final layer (the true contrast image 198) constitutes the network output (e.g., one or more convolutional kernel parameters, convolutional kernels, etc.), which is compared to the composite image to compute some loss or error function, which is backpropagated to guide network training. The loss or error function measures the difference between the network output (e.g., convolutional kernels or kernel parameters) and the training target (e.g., the composite image 192). In some implementations, the loss function may be an exponent (e.g., DISO) that considers absolute error (AE), correlation coefficient, and root mean square error (RMSE) of non-center. In some implementations, the loss function may be defined by other metrics associated with the specific task in question (e.g., the structural similarity index (SSIM)), such as the softmax function.

[0038] Before utilizing the artifact correction network 194, method 180 includes training neural network 202 (box 206) with training data 204. Training the artifact correction network 194 is similar to utilizing network 194. The training dataset (and test dataset) includes multiple sets of individually acquired contrast images (for the same contrast type such as T2-FLAIR or different contrast types) and corresponding MDME data for multiple subjects. Specifically, for each imaging subject, contrast images (e.g., for a specific contrast type) and corresponding MDME data are utilized. Both pathological and non-pathological cases are utilized. A composite image is generated from the corresponding contrast data and MDME data, and this composite image is fed to neural network 202 for training. As described above, supervised learning is used to train network 206.

[0039] Figure 3This is a schematic diagram depicting the grafting operation used to generate composite data. As depicted, in this case, a separately acquired (e.g., using MRI contrast scans) undersampled contrast dataset or image 208 (e.g., for a specific contrast type, such as T2-FLAIR) includes only the central k-space 210, with other regions 212 being zero-filled. Multiple synthetic contrast images 214 (e.g., 2D images) are also depicted, whose contrast type (e.g., T2-FLAIR) is the same as that of contrast datasets or images derived from MDME data previously acquired from the same subject as contrast dataset 208. It should be noted that the contrast types of contrast dataset 208 and the synthetic contrast images do not necessarily have to be the same. In some embodiments, all contrast types can be utilized. The synthetic contrast image 214 shares structural information with the contrast dataset 208. During grafting, the zero-filled regions 212 are filled with the corresponding k-spaces 216 from those regions of the synthetic contrast image 214 (or MDME data) to achieve the sharing of structural information and the formation of composite data or composite image 218.

[0040] Figure 4This is a schematic diagram depicting the use of an artifact correction network or artifact prediction network 220 (utilizing center-only k-space data from the contrast data). As depicted, fully sampled reference MDME data 222 is acquired using an MDME sequence during quantitative scanning. As depicted, individually undersampled contrast data 224 (contrast-only data) is acquired using an accelerated MRI contrast scan with a specific contrast type (e.g., T2-FLAIR). As depicted, contrast data 224 includes center-only k-space 226 lateral to zero-fill regions 228. Contrast data 224 and reference MDME data 222 have the same contrast type (e.g., T2-FLAIR). In some embodiments, contrast data 224 and reference MDME data 222 do not necessarily have the same contrast type. In some embodiments, all contrast types can be utilized. Prior to grafting, image intensity normalization 230 from reference MDME data 222 to contrast data 224 is performed. Grafting 232 is then performed. During grafting 232, k-space data from reference MDME data 222 outside the central region (as indicated by grafting region 234) is grafted into the previous zero-filled region 228 to form composite data 236. Then, transformation 238 is performed on the normalized reference MDME data 222 and composite data 236 to form a reference MDME or reference contrast image and a composite image, respectively, input to the artifact correction network 220. The output of the artifact correction network 220 is compared with the composite image to determine an error or loss function 240, which can be backpropagated to network 220, and used to help generate the final image 242 (e.g., an artifact-corrected reconstructed contrast image).

[0041] Figure 5This is a schematic diagram depicting the use of an artifact correction network or artifact prediction network 220 (utilizing only center k-space data from the contrast data). As depicted, fully sampled reference MDME data 222 is acquired using an MDME sequence during quantitative scanning. As depicted, individually undersampled contrast data 224 is acquired using an accelerated MRI contrast scan with a specific contrast type (e.g., T2-FLAIR). As depicted, contrast data 224 includes a center k-space 226 and a randomly sampled k-space 244 outside the center k-space 226, with zero-fill regions 228 in between. Contrast data 224 and reference MDME data 222 have the same contrast type (e.g., T2-FLAIR). In some embodiments, contrast data 224 and reference MDME data 222 do not necessarily have the same contrast type. In some embodiments, all contrast types can be utilized. Prior to grafting, image intensity normalization 230 from reference MDME data 222 to contrast data 224 is performed. Grafting 232 is then performed. During grafting 232, k-space data from reference MDME data 222 outside the central region (as indicated by grafting region 234) is grafted into the previous zero-filled region 228 to form composite data 236. Then, transformation 238 is performed on the normalized reference MDME data 222 and composite data 236 to form a reference MDME or reference contrast image and a composite image, respectively, input to the artifact correction network 220. The output of the artifact correction network 220 is compared with the composite image to determine an error or loss function 240, which can be backpropagated to network 220, and used to help generate the final image 242 (e.g., an artifact-corrected reconstructed contrast image).

[0042] Figure 6 A comparison of MR contrast images (e.g., from the same subject) generated using different techniques is shown. Image 246 is a T2-FLAIR contrast image synthesized from MDME data. Image 248 is a conventional T2-FLAIR contrast image acquired separately. Image 250 is a T2-FLAIR contrast image generated using the deep learning-based techniques disclosed above. Images 248 and 250 have similar contrast in pathological regions 252 and 254. Furthermore, cortical thickening artifacts and brightening artifacts present in image 246 are corrected for or absent in image 250.

[0043] Figure 7A comparison of MR contrast images (e.g., from the same subject) generated using different techniques is shown. Image 256 is a T2-FLAIR contrast image synthesized from MDME data. Image 258 is a conventional T2-FLAIR contrast image acquired separately. Image 260 is a T2-FLAIR contrast image generated using the deep learning-based techniques disclosed above. Images 258 and 260 have similar contrast in the highlighted regions 262 and 264.

[0044] The technical effects of the disclosed subject matter include providing systems and methods for generating realistic (i.e., equivalent to conventional) contrast images from synthetic MRI data based on deep learning. Specifically, a trained deep neural network is used to reconstruct artifact-corrected contrast images (i.e., realistic contrast images). Composite data or composite images can be generated by grafting a portion of k-space data (unacquired k-space) from one or more synthetic contrast images onto contrast data or contrast images (contrast images only) containing at least the central portion of k-space data. Synthetic contrast images are generated from MRI scans of an object of interest (e.g., a region such as a patient's brain) using a multi-time-lag multi-echo (MDME) sequence. Contrast data or images are acquired during scans of the same object of interest using the contrast MRI sequence. Scans for acquiring data for both the synthetic contrast images and the contrast images can occur in a single consecutive imaging session (i.e., where scans occur sequentially in response to a single initiation signal or input (e.g., a click)). Before grafting, the image intensity of the synthetic contrast image can be normalized to the image intensity of the contrast image. The artifact-corrected reconstructed contrast image output by the trained deep neural network has fewer artifacts than a synthetic contrast image of the same contrast type. Furthermore, the artifact-corrected reconstructed images exhibit diagnostic quality similar to conventionally acquired contrast images. Therefore, the disclosed embodiment enables the generation of true-contrast images while still utilizing data acquired using MDME sequences in a single-shot multi-contrast imaging protocol.

[0045] Referring to the technology presented herein and protected by the claims, and applying it to physical objects and concrete examples of practical nature, which explicitly improves the present art, it is therefore not abstract, intangible, or purely theoretical. Furthermore, if any claim appended to the end of this specification contains one or more elements designated as “means for [performing]…” or “steps for [performing]…”, such elements are intended to be interpreted pursuant to Section 35, Section 112(f) of the USC. However, for any claim containing elements designated in any other manner, such elements are not intended to be interpreted pursuant to Section 35, Section 112(f) of the USC.

[0046] This written description uses examples to disclose the subject matter, including best practices, and also enables those skilled in the art to practice the subject matter, including making and using any device or system and performing any included methods. The patent scope of this subject matter is defined by the claims and may include other examples that would 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 are not indistinguishable from the literal language of the claims, or if they include equivalent structural elements that have minor differences from the literal language of the claims.

Claims

1. A computer-implemented method for generating artifact-corrected reconstructed contrast images from magnetic resonance imaging (MRI) data, comprising: Both the synthetic contrast image and the composite image derived from multi-latency multi-echo MDME scan data are input into a trained deep neural network. The synthetic contrast image is derived from the MDME scan data acquired during the first scan of the object of interest using the MDME sequence, and the composite image is derived from both the MDME scan data and the contrast scan data acquired during the second scan of the object of interest using the contrast MRI sequence. The trained deep neural network is used to generate the artifact-corrected reconstructed contrast image based on both the synthesized contrast image and the composite image; and The reconstructed contrast image with artifact correction is output from the trained deep neural network.

2. The computer-implemented method according to claim 1, wherein the second scan is an accelerated scan, thereby acquiring partial k-space data.

3. The computer-implemented method of claim 1, wherein the first scan and the second scan are performed in a single continuous imaging session in response to a single initiation input.

4. The computer-implemented method of claim 1, wherein the first scan and the second scan are performed separately.

5. The computer-implemented method of claim 1, wherein the contrast scan data includes at least the central portion of k-space data.

6. The computer-implemented method of claim 5, wherein the contrast scan data includes random k-space data outside the central portion of the k-space data.

7. The computer-implemented method of claim 5, comprising grafting k-space data from the MDME scan data onto a zero-fill region outside the central portion of the k-space data of the contrast scan data to provide shared structural information, thereby generating composite data, wherein the composite image originates from the composite data.

8. The computer-implemented method according to claim 7, comprising normalizing the image intensity of the MDME scan data to the image intensity of the contrast scan data prior to grafting.

9. The computer-implemented method of claim 1, wherein the reconstructed contrast image for artifact correction comprises a T2-FLAIR image.

10. The computer-implemented method of claim 1, wherein the artifact-corrected reconstructed contrast image has fewer artifacts compared to a contrast image of the same type synthesized solely from the MDME scan data.

11. The computer-implemented method of claim 1, further comprising using supervised learning to train a neural network to generate the trained deep neural network, wherein the training data for the supervised learning comprises a composite image generated from both contrast scan data acquired during different scans of the object of interest using the contrast MRI sequence and corresponding MDME scan data acquired from separate scans of the object of interest using the MDME sequence.

12. A deep learning-based artifact correction system for generating artifact-corrected reconstructed contrast images from magnetic resonance imaging (MRI) data, comprising: A memory that encodes processor-executable routines; A processing unit configured to access the memory and execute processor-executable routines, wherein the routines, when executed by the processing unit, cause the processing unit to: Both a synthetic contrast image and a composite image derived from multi-latency multi-echo MDME scan data are input into a trained deep neural network. The synthetic contrast image is derived from the MDME scan data acquired during the first scan of the object of interest using an MDME sequence, and the composite image is derived from both the MDME scan data and the contrast scan data acquired during the second scan of the object of interest using a contrast MRI sequence. Before generating the composite image, the image intensity of the MDME scan data is normalized to the image intensity of the contrast scan data. The trained deep neural network is used to generate the artifact-corrected reconstructed contrast image based on both the synthesized contrast image and the composite image; and The reconstructed contrast image with artifact correction is output from the trained deep neural network.

13. The system of claim 12, wherein the second scan is an accelerated scan, thereby acquiring partial k-space data.

14. The system of claim 12, wherein the first scan and the second scan are performed in a single continuous imaging session in response to a single initiation input.

15. The system of claim 12, wherein the first scan and the second scan are performed separately.

16. The system of claim 12, wherein the contrast scan data includes at least the central portion of k-space data.

17. The system of claim 16, wherein the contrast scan data includes random k-space data outside the central portion of the k-space data.

18. The system of claim 16, wherein the routine, when executed by the processing unit, causes the processing unit to graft k-space data from the MDME scan data onto a zero-fill region outside the central portion of the k-space data of the contrast scan data to provide shared structural information, thereby generating composite data, wherein the composite image originates from the composite data.

19. The system of claim 12, wherein the artifact-corrected reconstructed contrast image comprises a T2-FLAIR image.

20. 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: Both the synthesized contrast image and the composite image derived from multi-delay multi-echo MDME scan data are input into a trained deep neural network, whereby... The composite contrast image is derived from the MDME scan data acquired during the first scan of the object of interest using the MDME sequence. The composite image is derived from both the MDME scan data and the contrast scan data acquired during the second scan of the object of interest using the contrast MRI sequence. Before generating the composite image, the image intensity of the MDME scan data is normalized to the image intensity of the contrast scan data. The trained deep neural network is used to generate a reconstructed contrast image with artifact correction based on both the synthetic contrast image and the composite image. as well as The reconstructed contrast image with artifact correction is output from the trained deep neural network.

Images