Medical imaging system and method for imaging a subject

Abstract

The present invention is entitled "Medical Imaging System and Method for Imaging a Subject." The present invention discloses a medical imaging system comprising: at least one medical imaging device that provides image data of a subject; and a processing system programmed to generate a plurality of training images having simulated medical conditions by fusing pathological regions from a plurality of template source images to a plurality of target images. The processing system is further programmed to train a deep learning network model using the plurality of training images and to input the image data of the subject into the deep learning network model. The processing system is further programmed to generate a medical image of the subject based on the output of the deep learning network model.

Description

Technical Field

[0001] The field of the present disclosure relates generally to medical imaging systems and methods, and more particularly to techniques for generating simulated images of patient conditions / pathologies for machine learning-based applications. Background Art

[0002] In modern healthcare facilities, non-invasive medical imaging systems are commonly used to identify, diagnose and treat physical conditions. Medical imaging covers different non-invasive techniques for imaging and visualizing the internal structure and / or functional behavior (such as chemical or metabolic activity) of organs and tissues in a patient's body. Currently, there are multiple modalities of medical diagnostic and imaging systems, each of which typically operates on different physical principles to generate different types of images and information. These modalities include ultrasound systems, computed tomography (CT) systems, X-ray systems (including conventional imaging systems and digital or digitized imaging systems), positron emission tomography (PET) systems, single photon emission computed tomography (SPECT) systems and magnetic resonance (MR) imaging systems.

[0003] Similar to other technology domains, deep learning (DL) has also made significant progress in the field of medical imaging. DL is being used in many imaging modalities, including CT, PET, X-ray, SPECT, and MR imaging systems. Deep learning uses effective techniques to achieve state-of-the-art diagnostics. Generally, DL is a subset of machine learning, in which artificial neural network models learn from large amounts of training data. Applications of deep learning include medical image preprocessing (i.e., denoising and enhancement), medical image segmentation, and medical image object detection and recognition.

[0004] In DL-based medical imaging, it is expected that DL networks can work robustly under a range of patient conditions, such as varying degrees of pathology, implants, etc. The location and extent of the impact of such patient conditions on medical images can be substantial and potentially impact the performance of DL techniques. For example, in magnetic resonance imaging (MRI) systems, metal implants can disrupt signals near tissues and obscure relevant landmarks. Another example may include fracture conditions, which can distort the topology of bone structures or pathologies, such as BIRADS classification, which requires certain shapes for classification.

[0005] Therefore, there is a need for an improved magnetic resonance imaging system and method. Summary of the Invention

[0006] According to an embodiment of the present technology, a medical imaging system is provided. The system includes: at least one medical imaging device that provides image data of a subject; and a processing system that is programmed to generate a plurality of training images having simulated medical conditions. The processing system is further programmed to train a deep learning network model using the plurality of training images and to input the image data of the subject into the deep learning network model. The processing system is further programmed to generate medical images of the subject based on the output of the deep learning network model. The processing system generates the plurality of training images by fusing pathological regions from a plurality of template source images into a plurality of target images.

[0007] According to another embodiment of the present technology, a method for imaging a subject is presented. The method includes: generating image data of the subject using a medical imaging device; and generating a plurality of training images having simulated medical conditions by fusing pathological regions from a plurality of template source images to a plurality of target images. The method also includes: training a deep learning network model using the plurality of training images; and providing the image data of the subject as input to the deep learning network model. The method also includes generating a medical image of the subject based on the output of the deep learning network model. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0009] Figure 1 is a schematic diagram of an exemplary magnetic resonance imaging (MRI) system according to an embodiment of the present technology;

[0010] Figure 2A is Figure 1 An exemplary truncation artifact reduction / classifier system for use in an MRI system;

[0011] Figure 2B According to the embodiment of the present technology, Figure 2A A flowchart of an exemplary method implemented in a system;

[0012] Figure 3 It is available in Figure 2A Schematic diagram of an exemplary DL network used in the truncation artifact reduction / classifier system;

[0013] Figure 4 According to another embodiment of the present technology, a method for generating Figure 3 Schematic diagram of the method of training images of DL network;

[0014] Figure 5A 、 Figure 5B and Figure 5C is an exemplary simulated training image generated according to an embodiment of the present technology;

[0015] Figure 6 is a schematic diagram depicting experimental results of a deep learning network according to an embodiment of the present technology;

[0016] Figure 7 is a diagram according to an embodiment of the present technology. Figure 1 a schematic diagram of a medical image generated by an MRI system; and

[0017] Figure 8 is a flow chart depicting a method for imaging a subject in accordance with an embodiment of the present technology. DETAILED DESCRIPTION

[0018] One or more specific embodiments will be described below. In order to provide a concise description of these embodiments, not all features of an actual implementation may be described in the 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 goals, such as complying with system-related and business-related constraints that may vary from implementation to implementation. Furthermore, it should be understood that such development efforts may be complex and time-consuming, but remain a routine task of design, fabrication, and manufacturing for those of ordinary skill having the benefit of this disclosure.

[0019] When introducing elements of various embodiments of the present embodiment, the articles "a," "an," "the," and "said" are intended to mean that there are one or more of these 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 thus the additional numerical values, ranges, and percentages are within the scope of the disclosed embodiments. Furthermore, the terms "circuit," "circuitry," and "controller" may include a single component or multiple components that are active and / or passive and are connected or otherwise coupled together to provide the described functionality.

[0020] The technology presented includes systems and methods for removing artifacts from medical images or classifying images in the presence of artifacts using deep learning models. As used herein, a subject is a person (or patient), an animal, or a pseudo-animal. Unlike signals representing the anatomy or structure of the subject, an artifact is a visual anomaly in a medical image that is not present on the subject but may be caused by a condition of the subject (such as a metal implant or a bone fracture in the human body). Methodological aspects will be apparent in part and discussed explicitly in part in the following description.

[0021] Typically, the location and extent of the impact of patient conditions (e.g., metal implants or fractures) on medical images can be significant and potentially impact the performance of DL techniques. One way to mitigate this problem is to collect such relevant patient data to expose the training of DL networks to learn the intended task regardless of the presence of such patient conditions. Some of these conditions occur rarely, thus limiting the number of examples available for training the network.

[0022] Another approach is to use biophysical models, which may not exactly replicate the patient's condition and are computationally expensive. Deep learning-based image synthesis methods do exist. However, their applicability is hampered by the difficulty in controlling contrast, position, intensity variations, and other factors in the synthesized data; they also require large amounts of data to generate the model.

[0023] The embodiments presented herein relate to techniques for generating training images with simulated patient conditions for DL-based applications. It should also be noted that while the techniques herein are presented with respect to magnetic resonance imaging (MRI) systems, the techniques can also be used in other imaging systems using DL algorithms. Furthermore, while the techniques herein are presented in detail with respect to metal implants, the techniques can also be used for other medical conditions, such as fractures.

[0024] Figure 1 A schematic diagram of an exemplary MRI system 10 is shown. In the exemplary embodiment, the MRI system 10 includes a workstation 12 having a display 14 and a keyboard 16. The workstation 12 includes a processor 18, such as a commercially available programmable machine running a commercially available operating system. The workstation 12 provides an operator interface that allows scanning plans to be entered into the MRI system 10. The workstation 12 is coupled to a pulse sequence server 20, a data acquisition server 22, a data processing server 24, and a data storage server 26. The workstation 12 and each of the servers 20, 22, 24, and 26 communicate with each other.

[0025] In the exemplary embodiment, the pulse sequence server 20 operates the gradient system 28 and the radio frequency ("RF") system 30 in response to instructions downloaded from the workstation 12. The instructions are used to generate gradient waveforms and RF waveforms in an MR pulse sequence. The RF coil 38 and the gradient coil assembly 32 are used to execute the prescribed MR pulse sequence. The RF coil 38 is shown as a whole-body RF coil. The RF coil 38 can also be a local coil that can be placed near the anatomical structure to be imaged, or a coil array including multiple coils.

[0026] In the exemplary embodiment, gradient waveforms for performing a prescribed scan are generated and applied to the gradient system 28, which energizes the gradient coils in the gradient coil assembly 32 to generate magnetic field gradients G for position encoding the MR signals. x , G y and G z The gradient coil assembly 32 forms part of a magnet assembly 34 , which also includes a polarizing magnet 36 and an RF coil 38 .

[0027] In an exemplary embodiment, the RF system 30 includes an RF transmitter for generating RF pulses used in MR pulse sequences. The RF transmitter responds to the scan scheme and direction from the pulse sequence server 20 to generate RF pulses with a desired frequency, phase, and pulse amplitude waveform. The generated RF pulses are applied by the RF system 30 to an RF coil 38. The responsive MR signals detected by the RF coil 38 are received by the RF system 30 and amplified, demodulated, filtered, and digitized according to commands generated by the pulse sequence server 20. The RF coil 38 is depicted as both a transmitter and receiver coil, such that the RF coil 38 transmits RF pulses and detects MR signals. In one embodiment, the MRI system 10 may include a transmitter RF coil for transmitting RF pulses and a separate receiver coil for detecting MR signals. The transmit channels of the RF system 30 may be connected to the RF transmit coil, and the receiver channels may be connected to separate RF receiver coils. Typically, the transmit channels are connected to the whole-body RF coil 38, and each receiver segment is connected to a separate local RF coil.

[0028] In the exemplary embodiment, the RF system 30 also includes one or more RF receiver channels. Each RF receiver channel includes an RF amplifier that amplifies the MR signals received by the RF coil 38 to which the channel is connected; and a detector that detects and digitizes the I and Q quadrature components of the received MR signals. The magnitude of the received MR signal can then be determined as the square root of the sum of the squares of the I and Q components, as shown in the following equation (1):

[0029]

[0030] And the phase of the received MR signal can also be determined as shown in the following equation (2):

[0031]

[0032] In an exemplary embodiment, the digitized MR signal samples generated by the RF system 30 are received by the data acquisition server 22. The data acquisition server 22 can operate in response to instructions downloaded from the workstation 12 to receive real-time MR data and provide buffer memory so that no data is lost due to data overflow. In some scans, the data acquisition server 22 only passes the acquired MR data to the data processing server 24. However, in scans where information derived from the acquired MR data is needed to control further execution of the scan, the data acquisition server 22 is programmed to generate the required information and transmit it to the pulse sequence server 20. For example, during a pre-scan, MR data is acquired and used to calibrate the pulse sequence executed by the pulse sequence server 20. In addition, navigator signals can be acquired during the scan and used to adjust operating parameters of the RF system 30 or gradient system 28, or to control the order in which views are sampled in k-space.

[0033] In an exemplary embodiment, the data processing server 24 receives MR data from the data acquisition server 22 and processes the MR data according to instructions downloaded from the workstation 12. Such processing may include, for example, Fourier transforming the raw k-space MR data to produce two-dimensional or three-dimensional images, applying filters to reconstructed images, performing backprojection image reconstruction on the acquired MR data, removing artifacts from the MR data, classifying MR images in the presence of artifacts, generating functional MR images, and computing motion or flow images.

[0034] In an exemplary embodiment, the image reconstructed by the data processing server 24 is transmitted back to the workstation 12 and stored there. In some embodiments, the real-time image is stored in a database memory cache ( Figure 1 (not shown), real-time images can be output from the database memory cache to the operator display 14 or a display 46 located near the magnet assembly 34 for use by the attending physician. Batch mode images or selected real-time images can be stored on a disk storage device 48 or in a host database on the cloud. When such images have been reconstructed and transferred to the storage device, the data processing server 24 notifies the data storage server 26. The operator can use the workstation 12 to archive images, generate films, or send images to other facilities via the network.

[0035] As mentioned earlier, the MR data acquired from the data acquisition server 22 may include artifacts due to pathological conditions of the subject, such as metal implants or bone fractures in the human body. Directly removing these artifacts using deep learning provides superior performance to conventional methods.

[0036] Figure 2A is Figure 1Schematic diagram of an exemplary artifact reduction / classifier system 200 for use in an MRI system 10 of the present invention. In an exemplary embodiment, the system 200 includes a computing device 202 configured to classify an image or reduce artifacts therein. The computing device 202 includes a DL network model 204. The system 200 also includes a training image generator 206 to generate training images according to an embodiment of the present technology. The training images from the training image generator 206 are used to train the DL network model 204. In one embodiment of the present technology, a training image generator is presented to generate these training images with simulated medical conditions. In one embodiment, the training images are images with simulated artifacts that are the result of the patient's medical condition. The computing device 202 can then use the trained DL network model 204 to generate medical images classified as having a certain medical condition or medical images with reduced artifacts. The computing device 202 can be included in the workstation 12 of the MRI system 10, or can be included on a separate computing device in communication with the workstation 12. Furthermore, in another embodiment, the training image generator 206 may be included on a separate computing device and may generate training images prior to actual operation of the MRI system 10 .

[0037] Figure 2B 2 is a flow chart of an exemplary method 250. The method 250 can be implemented on the artifact reduction / classifier system 200. In an exemplary embodiment, the method includes acquiring 252 a plurality of template source images and a plurality of target images. The plurality of template source images include representative examples of patient medical conditions. For example, the plurality of template source images can include images of patients with metal implants in the knee or other body parts. The plurality of template source images can also include images of patients with fractured bones. In addition, the plurality of target images are selected from a group of patient images without the medical condition. The template source images and the target images can both be stored in a historical image database of the same patient or in a historical image database of various pathologies of different patients, from which these images can be acquired when needed.

[0038] The method 250 further includes deriving 254 a segmentation mask of the pathological region from the plurality of template source images. The method 250 further includes fusing 256 the segmentation mask to the plurality of target images to generate a plurality of training images having the medical condition. Finally, the DL network model is executed at step 258 to generate a medical image of the patient with reduced artifacts or a medical image classified as having a specific medical condition. In one embodiment, the method steps 252, 254, and 256 may be performed in Figure 2A The training image generator 206 is implemented in FIG.

[0039] Figure 32 is a representation of an exemplary DL network model 300 that can be used as the DL network model 204 in an embodiment of 200. The exemplary DL network model 300 includes layers 320, 340, 360, and 380. Layers 320 and 340 are connected using a neural connection 330. Layers 340 and 360 are connected using a neural connection 350. Layers 360 and 380 are connected using a neural connection 370. Data flows forward from input layer 320 to output layer 380 via inputs 312, 314, 316 and reaches output 390. Inputs 312, 314, 316 can be source images, and output 390 can be a target image.

[0040] Layer 320 is the input layer, which Figure 3 The example includes multiple nodes 322, 324, 326. Layers 340 and 360 are hidden layers and are Figure 3 The example includes nodes 342, 344, 346, 348, 362, 364, 366, 368. The DL network model 300 may include more or fewer hidden layers 340 and 360 than shown. Layer 380 is the output layer and is Figure 3 380. The example DL network model 300 includes a node 382 having an output 390. Each input 312 to 316 corresponds to a node 322 to 326 of the input layer 320, and each node 322 to 326 of the input layer 320 has a connection 330 to each node 342 to 348 of the hidden layer 340. Each node 342 to 348 of the hidden layer 340 has a connection 350 to each node 362 to 368 of the hidden layer 360. Each node 362 to 368 of the hidden layer 360 has a connection 370 to the output layer 380. The output layer 380 has an output 390 to provide the output from the exemplary DL network model 300.

[0041] Among the connections 330, 350, and 370, certain exemplary connections 332, 352, and 372 may be assigned increased weights, while other exemplary connections 334, 354, and 374 may be assigned smaller weights in the DL network model 300. For example, input nodes 322 to 326 are activated by receiving input data via inputs 312 to 316. Nodes 342 to 348 and 362 to 368 of hidden layers 340 and 360 are activated by data flowing forward through the network model 300 via connections 330 and 350, respectively. After the data processed in the hidden layers 340 and 360 is sent via connection 370, node 382 of the output layer 380 is activated. When the output node 382 of the output layer 380 is activated, the node 382 outputs an appropriate value based on the processing completed in the hidden layers 340 and 360 of the DL network model 300.

[0042] Figure 4is a diagram illustrating a method 400 for generating training images for a DL network model according to another embodiment of the present technology. The method 400 may be Figure 2A 4 is implemented in the training image generator 206 of . The method 400 includes obtaining a template metal volume at step 402. The template metal volume includes a template source image that includes a representative example of a metal implant in a patient's knee as shown in the knee image 404. For example, the template metal volume 402 is obtained from a database of historical images of patients with metal implants (such as screws located in their knees). At step 406, the method includes segmenting the metal region (i.e., a segmentation mask) from the template metal volume 402. For example, based on the geometry of the screw, a segmentation algorithm (such as a "connected pixel algorithm") can be used to segment or detect the metal region in the knee. Typically, the segmentation algorithm determines which pixels of the image 404 belong to which objects.

[0043] At step 408, a target normal volume is obtained. The target normal volume 408 includes a target image that includes a representative example of a patient's knee without the medical condition as shown in knee image 410. Step 412 includes processing the segmented metal regions. In one embodiment, processing the segmented metal regions includes performing image registration, landmark matching, histogram matching, or a combination thereof between the segmented metal regions and the target normal volume. Image registration involves transforming different sets of images into a coordinate system. Typically, the segmented metal regions can come from template source images of different people (adults or children, i.e., different knee sizes). Therefore, image registration is used for scale matching to accommodate the size difference between the template source image and the target image.

[0044] In addition, the segmented metal regions may be located at different locations within the patient's body. Therefore, in one embodiment, landmark matching of the segmented metal regions is used to align the segmented metal regions with the target image. As will be appreciated by those skilled in the art, landmark matching can be performed using small deformation image matching or large deformation image matching algorithms. Finally, histogram matching can be used to normalize the segmented metal regions to compensate for variations in the imaging system sensor, atmospheric conditions, or intensity variations. As will be appreciated by those skilled in the art, histogram matching involves transforming the segmented metal regions to match their histogram to the histogram of the target image.

[0045] Method 400 also includes enhancing metal regions at landmark points or regions of interest in the target image at step 414. In one embodiment, the regions of interest in the target image can be determined using an attention map derived from another machine learning network or based on regions most likely to have a medical condition (e.g., a surgical implant). In another embodiment, the regions of interest in the target image can be determined using ground truth labels from multiple target images or using an atlas-based approach. Furthermore, the enhancement process includes performing a series of transformations on the segmented metal regions (i.e., the segmentation mask). The series of transformations includes rotating, resizing, and elastically deforming the segmented metal regions. Typically, there are not many available template source images with medical conditions. For example, if there are 100 template source images, only 5 or 10 of these 100 template source images may have some medical conditions, such as metal implants or fractures. Furthermore, the enhancement process can include pasting a segmentation mask from one region in the multiple template source images to different regions in the multiple target images. In other words, in one embodiment, different locations of metal implants can be simulated. Thus, the enhancement step is performed to expand the training images of the deep learning network model by simulating various patient conditions with different segmentation masks. For example, knee screws can be enhanced at 5-6 landmark points in the target image, or screws of different sizes can be enhanced in the target image. Thus, based on a single template source image of a medical condition, it is possible to generate 20 to 30 training images with simulated medical conditions. Finally, at step 416, the segmented metal regions are fused with the target image. This fusion operation ensures that the enhanced metal regions are seamlessly blended or pasted with the target image, as shown in image 418. This fusion process may include performing contrast equalization of the multiple template source images and the multiple target images to achieve a perceived smoothness between the multiple target images and the pasted segmented mask.

[0046] Figures 5A to 5C are exemplary simulated training images generated according to embodiments of the present technology. Figure 5A A sample axial image of a knee 500 is shown. Specifically, image 502 corresponds to a fusion of a metal region that is scaled to completely obscure the femoral condyle landmark. Image 504 shows a metal region positioned away from the landmark. Image 506 shows randomly positioned metal, and image 508 shows a fusion of metal above the tibial landmark.

[0047] Figure 5B A coronal sample 510 of a knee image is shown. Figure 5B , image 512 corresponds to randomly placed metal, and image 514 corresponds to metal placed below the meniscus. Additionally, image 516 shows metal placed on the meniscus, where the aspect ratio of the metal is preserved. Finally, image 518 shows metal placed above the meniscus.

[0048] Figure 5C A sagittal sample of a knee image 520 is shown. Figure 5C In FIG, image 522 shows randomly placed metal in the knee. Image 524 corresponds to metal placed below the meniscus, and image 526 shows metal placed above the meniscus and made to scale the entire region of interest. In addition, image 528 shows metal placed above the meniscus.

[0049] Figure 6 Schematic diagram 600 depicting experimental results of a deep learning network model according to an embodiment of the present technology is shown. The experiments were performed on test image samples of actual clinical subjects damaged by metal implants. Figure 6 , graphs 602 and 604 show the results of the deep learning network model trained as described above without simulated training images and with simulated training images. The horizontal axis 606 in both graphs 602 and 604 shows the true or actual labels of the test samples, while the vertical axis 608 shows the predicted labels of the deep learning network model for the test samples.

[0050] In Figure 602, out of a total of 226 femur image test samples, the network accurately classified (or predicted) 210 images as femur images, while the remaining 16 images were incorrectly classified as tibia images (9) or noise (7). In contrast, Figure 604 accurately classified 211 images as femur images. Furthermore, out of a total of 105 tibia image test samples, the network in Figure 602 accurately classified 59 images as tibia images, while the network in Figure 604 accurately classified 71 images as tibia images. Furthermore, out of a total of 466 irrelevant or noise image test samples, the network in Figure 602 accurately classified 294 images as noise images, while the network in Figure 604 accurately classified 389 images as tibia images. Similarly, out of a total of 225 coronal image test samples, the network in Figure 602 accurately classified 209 images as coronal images, while the network in Figure 604 accurately classified 210 images as coronal images. Finally, out of a total of 277 sagittal image test samples, the network in Figure 602 accurately classified 263 images as sagittal images, while the network in Figure 604 accurately classified 268 images as sagittal images.

[0051] Overall, the experiments showed that the DL network model was trained without simulated training images, i.e., Figure 602 accurately classified 1035 images out of a total of 1299 images, i.e., an accuracy of 79.67%. In contrast, the DL network model was trained with simulated training images, i.e., Figure 604 accurately classified 1149 images out of a total of 1299 images, i.e., an accuracy of 88.45%. Therefore, according to an embodiment of the present technology, the accuracy was improved by approximately 9% using a DL network model trained with simulated training images.

[0052] Figure 7 A diagram depicting an embodiment of the present technology is shown, wherein an MRI system (eg, Figure 1 Schematic diagram 700 of a medical image generated by a system 10). Typically, Figure 7 Medical images 702, 704, 706, 708, 710, and 712 are shown, corresponding to the sagittal coverage mask, the meniscus sagittal plane, the meniscus coronal plane, the femoral condylar medial-lateral plane (FCIL), the tibial plane, and the femoral coronal plane (FCP), respectively. As can be seen from these medical images, the present technology is able to very accurately depict landmarks (horizontal lines) even in the presence of metal artifacts. Note that in images 702 and 710, the arrows point to the metal areas (i.e., the white halos), while in image 708, the arrows point to the phantom areas (i.e., the dark areas).

[0053] Figure 8 A flow chart depicting a method 800 for imaging a subject according to an embodiment of the present technology is shown. At step 802, the method includes generating image data of the subject using a medical imaging device. In one embodiment, the medical imaging device includes an MRI system. The image data generated by the MRI system may include artifacts due to medical conditions present in the subject's body. Therefore, the method includes processing the image data to reduce the artifacts or classify the images even in the presence of these artifacts.

[0054] At step 804, the method includes generating a plurality of training images having simulated medical conditions by fusing pathological regions from a plurality of template source images to a plurality of target images. The template source images may be obtained from a database of historical images of patients having metallic implants (such as screws or fractures in their knees). Furthermore, the target images are selected from a set of images of patients without the medical condition. Furthermore, in method 800, fusing the pathological regions includes: deriving segmentation masks of the pathological regions from the plurality of template source images; and processing the segmentation masks. In one embodiment, processing the segmentation masks includes performing image registration, landmark matching, histogram matching, or a combination thereof between the segmentation masks and the target images. Furthermore, fusing includes an enhancement process in which the segmentation masks are enhanced at landmark points or regions of interest in the target images. The enhancement process includes transforming the segmentation masks, which undergo a series of transformations before being applied to the regions of interest in the plurality of target images. In one embodiment, the regions of interest in the plurality of target images may be determined using an attention map derived from another machine learning network or based on regions most likely to have a medical condition (e.g., a surgical implant). In another embodiment, the regions of interest in the plurality of target images may be determined using ground truth labels on the plurality of target images or using an atlas-based approach.The series of transformations includes rotating, resizing, and elastically deforming the segmented metal regions.

[0055] Method 800 also includes training a deep learning network model using a plurality of training images at step 806. At step 808, the image data from step 802 is provided as input to the trained deep learning network model. Finally, at step 810, a medical image of the subject is generated based on the output of the deep learning network model.

[0056] Advantages of this technique include providing users with the flexibility to synthesize patient conditions at locations of interest driven by tasks or deep learning features that guide the task. This technique also overcomes the computational complexity of synthesizing such data using biophysical models (e.g., susceptibility simulation for metals in MR) or the data complexity (various patient conditions and correspondences) and inflexibility (image intensity, fusion ratio, etc.) of deep learning-based synthesis.

[0057] This written description uses examples to disclose the invention, including the best mode, and also to enable any person skilled in the art to practice the invention, including making and using any devices or systems and performing any included methods. The patentable scope of the invention is defined by the claims and may include other examples that occur to those skilled in the art. If such other examples have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements with insignificant differences from the literal language of the claims, such other examples are intended to fall within the scope of the claims.

Claims

1. A medical imaging system, comprising: at least one medical imaging device providing image data of a subject; A processing system programmed to: generating a plurality of training images having simulated medical conditions; training a deep learning network model using the plurality of training images; inputting the image data of the subject into the deep learning network model; as well as generating a medical image of the subject based on an output of the deep learning network model; wherein the processing system is programmed to generate the plurality of training images by fusing pathological regions from a plurality of template source images to a plurality of target images, The deep learning network model is trained to reduce artifacts corresponding to metal implants in the image data.

2. The medical imaging system of claim 1, wherein the plurality of template source images include representative examples of a medical condition of the subject, and wherein the plurality of target images are selected from a group of images free of the medical condition.

3. The medical imaging system of claim 2, wherein the processing system is programmed to determine the pathological region from the plurality of template source images by deriving a segmentation mask of the pathological region from the plurality of template source images.

4. The medical imaging system of claim 3 , wherein the processing system is programmed to fuse the pathological region by processing the segmentation mask, wherein the processing comprises performing image registration, landmark matching, histogram matching, or a combination thereof between the segmentation mask and the plurality of target images. 5 . The medical imaging system of claim 4 , wherein the processing system is programmed to fuse the pathological region by pasting the processed segmentation masks on the regions of interest of the plurality of target images.

6. The medical imaging system of claim 5, wherein the processing system is programmed to determine the region of interest based on ground truth labeling on the plurality of target images or using an atlas-based approach.

7. The medical imaging system of claim 5, wherein the processing system is programmed to determine the region of interest using an attention map derived from another machine learning network or based on regions most likely to have a medical condition. 8 . The medical imaging system of claim 5 , wherein the segmentation mask undergoes a series of transformations before being pasted over the regions of interest of the plurality of target images.

9. The medical imaging system of claim 8, wherein the series of transformations comprises rotating, resizing, and elastically deforming the segmentation mask.

10. The medical imaging system of claim 5, wherein the processing system is programmed to paste the segmentation mask from one region in the plurality of template source images to a different region in the plurality of target images.

11. The medical imaging system of claim 1 , wherein the processing system is programmed to perform contrast equalization of the plurality of template source images and the plurality of target images to obtain perceptual smoothness between the plurality of target images and the pasted segmented masks.

12. The medical imaging system of claim 1, wherein the medical condition comprises a metal implant or a bone fracture related condition.

13. A method for imaging a subject, the method comprising: generating image data of the subject using a medical imaging device; generating a plurality of training images having simulated medical conditions by fusing pathological regions from a plurality of template source images to a plurality of target images; training a deep learning network model using the plurality of training images; providing the image data of the subject as input to the deep learning network model; generating a medical image of the subject based on an output of the deep learning network model, The deep learning network model is trained to reduce artifacts corresponding to metal implants in the image data.

14. The method of claim 13, wherein the plurality of target images are selected from a group of images free of a medical condition, and wherein the plurality of template source images comprise representative examples of the medical condition of the subject. 15 . The method of claim 14 , wherein determining the pathological region from the plurality of template source images comprises deriving a segmentation mask of the pathological region from the plurality of template source images. 16 . The method of claim 15 , wherein fusing the pathological region comprises processing the segmentation mask, wherein the processing comprises performing image registration, landmark matching, histogram matching, or a combination thereof between the segmentation mask and the plurality of target images. The method according to claim 15 , wherein fusing the pathological region comprises pasting the segmentation mask on the region of interest of the plurality of target images.

18. The method of claim 17, wherein the region of interest is determined based on ground truth labeling on the plurality of target images or using an atlas-based approach.

19. The method of claim 17, wherein the segmentation mask undergoes a series of transformations before being pasted over the regions of interest of the plurality of target images.

20. The method of claim 19, wherein the series of transformations comprises rotating, resizing, and elastically deforming the segmentation mask.

Images