• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    The present invention relates to a method for generating diagnostic images with reduced dose of contrast agent using a deep learning network (DLN) [114] that has been trained using zero contrast [100] and low contrast images [102] as input for DLN and full contrast images [104] as fundamental reference images. Before training, images are pre-processed [106, 110, 118] to co-register and normalize them. The trained DLN [114] is then used to predict a synthesis image of contrast agent for full dose [116] from acquired images of zero dose and low dose.
    公开号:BR112020007105A2
    申请号:R112020007105-6
    申请日:2018-10-09
    公开日:2020-09-24
    发明作者:Greg Zaharchuk；Enhao Gong；John M. Pauly
    申请人:The Board Of Trustees Of The Leland Stanford Junior University；
    IPC主号:
    专利说明:

    [001] [001] The present invention relates in general to medical diagnostic imaging. More specifically, it refers to imaging techniques that use contrast agents.
    [002] [002] Many types of medical imaging use contrast agents to increase the visualization of normal and abnormal structures. Examples include conventional angiography, fluoroscopy, computed tomography (CT), ultrasound, and magnetic resonance imaging (MRI). It is often desirable to reduce the dose contrast agent in order to increase the safety of said agents. However, the reduced dose reduces the desired image enhancements, so this was not possible.
    [003] [003] For example, MRI is a powerful imaging technique that provides unique information to distinguish different soft tissues and pathologies. Magnetic contrast agents with unique relaxation parameters are administered frequently to further increase the visibility of the pathology and the outline of the lesions. Gadolinium-based contrast agents (GBCAs) are widely used in MRI scans because of their paramagnetic properties, for applications such as angiography, neuroimaging and liver imaging.
    [004] [004] However, although GBCAs are designed to prevent toxic release and have been widely applied in contrast-enhanced MRI scans (CE-MRI) to assist in diagnosis, there are several side effects resulting from GBCA administration. Therefore, there are many reasons why it would be advantageous to reduce the GBCA dosage, while preserving the enhanced contrast in the magnetic resonance image.
    [005] [005] Similar problems are associated with the administration of contrast agents in other imaging modalities as well. Thus, it would be beneficial to be able to reduce the dose of contrast agents generally in diagnostic imaging techniques, without sacrificing the image enhancement benefits that contrast agents provide.
    [006] [006] The present invention provides techniques for enhancing image quality of diagnostic imaging modalities using a lower dose of contrast than is currently possible. This allows for new opportunities to improve the value of medical imaging. In addition to MRI, the techniques are generally applicable to a variety of diagnostic imaging techniques, including angiography, fluoroscopy, computed tomography (CT) and ultrasound.
    [007] [007] Surprisingly, the techniques of the present invention are capable of predicting a full dose contrast agent image synthesized from a low dose contrast agent image and a pre-dose image. The low dose can be any fraction of the full dose, but is preferably 1/10 or less of the full dose. Significantly and naively amplifying the contrast enhancement of a 1/10 low dose CE-MRI by a factor of ten results in poor image quality with generalized noise and ambiguous structures. Although the low contrast image cannot be used directly for diagnosis, or simply by amplifying its uptake, the techniques of the present invention are remarkably capable of recovering the full contrast signal and generating predicted full contrast images with high diagnostic quality. Specifically, in experimental tests, the method produced significant improvements over low-dose images of 10%, with PSNR gains above 5 dB, 11% increase in SSIM and improvements in image quality ratings and contrast enhancement.
    [008] [008] Modalities of the present invention use a deep learning network, such as a Convolutional Neural Network for image-to-image regression, with a pre-contrast image and a low-contrast image as an input and with a full-contrast image predicted as output. A residual learning approach is preferably used in forecasting.
    [009] [009] The method includes pre-processing to co-register and normalize between different images, so that they are directly comparable. This step is important, as there are different arbitrary acquisition and scale factors for each scan. Preferably, the medium signal is used for normalization. Pre-processed images are used to train the deep learning network to predict the full contrast image of pre-contrast and low-contrast images. The trained network is then used to synthesize full contrast images from clinical examinations of pre-contrast and low-contrast images. The techniques are generally applicable to any type of diagnostic imaging that uses a contrast agent. For example, imaging applications include fluoroscopy, MRI, tomography and ultrasound.
    [010] [010] In one aspect, the invention provides a method for training a diagnostic imaging device to perform medical diagnostic images with reduced dose of contrast agent. The method includes a) taking diagnostic images of a set of individuals to produce a set of images comprising, for each individual in the set of individuals, i) a full contrast image acquired with a dose of total contrast agent administered to the individual, ii) a low contrast image acquired with a low dose of contrast agent administered to the object, in which the low dose of contrast agent is less than the total dose of contrast agent and iii) a low contrast image acquired without the contrast agent dose administered to the individual; b) pre-process the set of images to co-register and normalize the set of images to adjust the differences in acquisition and scaling between different scans; and c) training a deep learning network (DLN) with the pre-processed image set by applying zero contrast images from the image set and low contrast images from the image set as input to the DLN and using a cost function to compare the DLN output with full contrast images from the image set to train DLN parameters using backpropagation. The cost function can be, for example, a WEM loss function or a mix loss function with non-local structural similarities.
    [011] [011] In another aspect, the present invention provides an imaging method for medical diagnosis with reduced dose of contrast agent. The method includes a) performing an individual's imaging diagnosis to produce a low contrast image acquired with a low dose of contrast agent administered to the individual, where the low dose of contrast agent is less than the total dose of contrast agent. contrast, and a zero contrast image acquired without any dose of the contrast agent administered to the individual; b) pre-process the low contrast image and the zero contrast image to co-register and normalize the images to adjust the differences in acquisition and scaling; and c) applying the low contrast image and the zero contrast image as input to a deep learning network (DLN) to generate an individual synthesized full dose contrast agent image as an output from the DLN; where the DLN has been trained by applying zero contrast and low contrast images as input and full contrast images as fundamental true reference images.
    [012] [012] A low dose of contrast agent is preferably less than 10% of the total dose of contrast agent. Diagnostic imaging can be angiography, fluoroscopy, computed tomography (CT), ultrasound, or magnetic resonance imaging. For example, performing diagnostic imaging may include performing MRI imaging where the total dose of contrast agent is at most 0.1 mmol / kg contrast with Gadolinium for MRI. In one embodiment, the DLN is a convolutional encoder-decoder (CNN) neural network that includes bypass concatenation connections and residual connections.
    [013] [013] In another aspect, the present invention provides a medical diagnostic imaging method that includes a) performing an individual image diagnosis to produce a first image acquired with a first image acquisition sequence and a second image acquired with a second distinct image acquisition sequence from the first image acquisition sequence, where zero dose of contrast agent is administered during imaging diagnosis; b) pre-process the first image and the second image to co-register and normalize the images to adjust the differences in acquisition and scaling; c) apply the first image and the second image as input to a deep learning network (DLN) to generate, as an output from the DLN, an individual synthesized total dose contrast agent image; in which DLN was trained by applying zero contrast images with different image sequences as input and full contrast images acquired with a dose of full contrast agent as fundamental true reference images.
    [014] [014] Figure 1 is a flow chart showing a processing pipeline according to an embodiment of the present invention;
    [015] [015] Figure 2 illustrates a workflow of a protocol and procedure for the acquisition of images used for training in accordance with a modality of the present invention;
    [016] [016] Figure 3 is a schematic overview of an image processing flow according to an embodiment of the present invention;
    [017] [017] Figure 4 is an illustration of a signal model for synthesizing enhanced dose contrast enhanced MRI image according to an embodiment of the present invention;
    [018] [018] Figure 5 shows a detailed architecture of the deep learning model (DL) according to an embodiment of the present invention;
    [019] [019] Figure 6 is a set of images of a patient with intracranial metastatic disease, showing the predicted image synthesized from a pre-contrast image and a low dose image, compared with a full dose image according to a modality of the present invention;
    [020] [020] Figure 7 is a set of images of a patient with a brain neoplasm, showing a synthesized full dose image predicted from a pre-contrast image and a low dose image, compared to the full dose image according to an embodiment of the present invention;
    [021] [021] Figure 8 is a set of images of a patient with a programmable shunt and intracranial hypotension, showing noise suppression capabilities and consistent contrast enhancement capabilities according to one embodiment of the present invention.
    [022] [022] Modalities of the present invention provide an imaging diagnostic technique based on deep learning to significantly reduce contrast agent dosage levels while maintaining diagnostic quality for clinical images.
    [023] [023] Illustrations of the protocol and procedure of a modality of the present invention are shown in Figure 1 and Figure 2. Although the modalities described below focus on MRI imaging for purposes of illustration, the principles and techniques of the present invention described in that document they are not limited to MRI, but are generally applicable to various imaging modalities that make use of contrast agents.
    [024] [024] Figure 1 is a flow chart showing a processing pipeline for an embodiment of the present invention. The deep learning network is trained using images of multiple contrasts 100, 102, 104 acquired from readings from a multiplicity of individuals with a wide range of clinical indications. Images are pre-processed to perform image co-registration 106, to produce multi-contrast images 108 and data augmentation 110 to produce normalized image fragments of multiple contrasts 112. Rigid or non-rigid co-registration can be used to adjust multiple slices or image volumes to combine pixels and voxels with each other. As there may be arbitrary differences in scale between different volumes, normalization is used to match the intensity of each image / volume. Brain and anatomy masks are optionally used to extract the important regions of interest in each image / volume. Reference images 104 are also processed to perform co-registration and normalization 118.
    [025] [025] Figure 2 illustrates the workflow of the protocol and procedure for acquiring images used for training. After a pre-contrast (zero dose) image 200 is acquired, a low dose (e.g., 10%) of contrast is administered and a low dose image 202 is acquired. An additional dose (for example, 90%) of contrast is then administered to total a total dose of 100%, and a full dose image of 204 is then acquired.
    [026] [026] In an MRI modality, images are acquired with 3T MRI readers (GE Healthcare, Waukesha, WI, USA) using neuro clinical protocol with high resolution 3D image, T1-weighted, with image of inversion recovery and preparation fast gradient gradient (IR-FSPGR) in 3Q.
    [027] [027] For other imaging modalities, similar settings are used to include at least one set of images without enhancement acquired without injecting contrasts. And, optionally, at least one set of images with low level enhancement that is acquired with low contrast injection.
    [028] [028] Figure 3 is a schematic overview of the image processing flow. The detailed workflow steps are applied to the multiple contrast images acquired for each individual. At acquisition stage 300, readings are performed as described above to acquire a zero contrast image 306, low contrast image 308 and full contrast image 310. These multiple contrast images then pass through preprocessing stage 302. The images are then used in a deep learning training stage 304 to train a deep learning network to synthesize a full dose image
    [029] [029] After co-registration and normalization, the remaining signal differences between images of different contrast levels are, in theory, related only to contrast captures and non-structural image noise, as shown in Figure 4, which is an illustration of the signal model to synthesize the MRI image with full dose contrast 412. Compared to the reference high quality contrast uptake 406 between pre-contrast 400 and 404 full dose CE-MRI, the uptake of low dose contrast 408 between pre-contrast 400 and low dose 402 is noisy, but includes contrast information. Using a deep learning method, we learned how to remove noise to generate high-quality predicted contrast capture 410 and then combined this with the 400 pre-contrast reading to synthesize a full-dose 412 CE-MRI image.
    [030] [030] After pre-processing, a deep learning network is trained using the fundamental 100% full dose CE-MRI images as the fundamental reference images. Non-contrast MRI (zero dose) and 10% low dose EC-MRI are supplied to the network as inputs, and the network output is an approximation of the full dose CE-MRI. During training, this network implicitly learns the removal of guided noise from the capture of noisy contrast extracted from the difference signal between low-dose and non-contrast (zero-dose) images, which can be scaled to generate the enhancement of the contrast of an image of total dose.
    [031] [031] This model is an encoder-decoder convolutional neural network with 3 encoder steps 500, 502, 504 and 3 decoder steps 506, 508, 510. In each step, there are 3 convolutional layers connected by 3 x 3 Conv-BN -ReLU. The coding steps are connected in sequence by 2x2 maximum association, and the decoder steps are connected in sequence by 2 x 2 sampling for more. Ignoring the concatenated connections 512, 514, 516 combines symmetrical layers to prevent loss of resolution. Residual connections 518, 520, 522, 524, 526, 528, 530 allow the model to synthesize a full dose image by predicting the enhancement signal 540 from a difference between the pre-dose image 542 and the image of low dose 544. The cost function 532 compares the predicted total dose image 534 and the fundamental reference true total dose image 536, which allows optimization of network parameters via error backlog 538.
    [032] [032] The network architecture described above is just an illustrative example. Other architectures are possible. For example, the network may have a different number of layers, image size on each layer, and variable connections between layers. The function in each layer can be different from linear or non-linear functions. The output layer can have a different so-called activation function, which is mapped to a given range of output intensity. There may be a multiple number of concatenated networks, the so-called recurring network, which further improves the capacity that a single network can achieve.
    [033] [033] In one test, the network was trained on about 300 2D slices of the 3D volumes co-recorded in each patient, excluding the slices at the base of the brain that had low SNR and no valuable information for anatomy or contrast. Standard rigid image transformations are used to further increase the data set in training to ensure the robustness of the model.
    [034] [034] In training, the descent of the stochastic gradient (SGD) was used for each subset of training data sets and back propagation is used to optimize the network parameters in relation to a cost function that compares the MRI images of total dose predicted and true. In one embodiment, the absolute mean error cost (WEM) function, also known as L1 loss, is used in training. The training takes 200 times with the SGD and ADAM method for optimization and 10% of the training data set with random permutations was used for validation to optimize hyperparameters and choose the best model among all iterations. SGD and ADAM are also examples of operators to solve the network training optimization problem. There are many other options, including, for example, RMSprop and Adagrad. Basically, the optimizers allow for faster and smoother convergence.
    [035] [035] The loss functions can include, for example, a map of functions of the fundamental image and the expected image pair for defined loss values. This includes pixel or voxel loss based on pixel / voxel differences that generally use L1 (mean absolute error) or L2 (mean square error) loss. In addition, the loss may be based on regions of a certain size, considering the similarity of the structures, for example, structural similarities of the SSIM. Or the loss can be calculated based on other previously trained networks or simultaneously, the so-called perceptual losses or adverse losses, respectively. In addition, the loss can be any arbitrary weighted combination of many loss functions.
    [036] [036] In a clinical setting, the trained deep learning network is used to synthesize a full dose image from zero and low dose images. Co-registered and normalized images without contrast (zero dose) and low dose are loaded from DICOM files and inserted into the trained network. With efficient routing, which takes about 0.1 s per 512 by 512 image, the synthesized total dose image is generated. The process is conducted for each 2D slice to generate the entire 3D volume and stored in a new DICOM folder for further evaluation.
    [037] [037] In tests of the technique, the synthesized full-dose CE-MRI images contain consistent contrast uptake and similar enhancement.
    [038] [038] As shown in Figure 6, in a patient with intracranial metastatic disease, the predicted image 606 synthesized from the pre-contrast image 600 and the low-dose image 602 have similar enhancements of contrast enhancement in the lesions as the image of total dose 604. The lesions show better visibility on the synthesized full contrast image 606, although they cannot be reliably appreciated on the low dose CE-MRI image 602. In addition, the synthesized CE-MRI image 606 shows a similar outline of a metastatic lesion in the right posterior cerebellum compared to the true full-contrast EC-MRI image 604.
    [039] [039] For another case in a patient with cerebral neoplasia, shown in Figure 7, the synthesized full-dose CE-MRI image 706 predicted from the pre-contrast image 700 and the low dose image 702 shows contrast uptake similar to that of the actual full dose CE-MRI image 704. In addition, the image quality of the synthesized total dose image 706 is better compared to the acquired total dose CE-MRI 704, which has motion and motion artifacts more serious serration distortion. Surprisingly, this demonstrates that the synthesized results have additional advantages over the original full dose CE-MRI for better removal of noise and artifacts, for example, suppression of motion artifacts and jagged distortion.
    [040] [040] In the case of a patient with programmable deviation and intracranial hypotension, shown in Figure 8, the results also demonstrate better ability to suppress noise and consistent ability to improve contrast. The engorgement of the dural sinuses and the pachymeningeal enhancement are clearly seen in the synthesized image 806, predicted from the pre-contrast image 800 and the low dose image 802, and the fundamental full dose image 804, although there is less noise in the regions of no enhancement to the synthesized image. Thus, the method generates diagnostic quality contrast from low-dose acquisition, also demonstrating improvement in noise suppression and contrast enhancement.
    [041] [041] Tests of the method by the inventors demonstrated that the synthesized CE-MRI images improved significantly (both p <0.001) compared to the low dose image in all quantitative metrics with 11.0% improvements in SSIM and gains more than 5.0 dB in the PSNR. Compared with the true acquired total dose CE-MRI, the proposed method is not significantly different (both p> 0.05) for the overall image quality and the clarity of the enhanced contrast. In comparison, low contrast images of 10% are significantly (p <0.001) worse in enhanced contrast clarity, with a decrease of more than 3 on a 5-point scale.
    [042] [042] As Figure 7 and Figure 8 show, and as verified by the classifications of neuroradiologists, the synthesized total dose images show significant improvements (p <0.05) over the total dose acquired CE-MRI in the suppression of artifacts in the regions of non-improvement, which was an additional advantage of the DL method.
    [043] [043] Compared to the actual full dose CE-MRI, no significant difference (p> 0.05) in the contrast enhancement information can be identified by neuro-radiologists who have not seen the source of the image. The synthesized full dose images also seem to suppress image artifacts better (p <0.05). This is because the high intensity signal of the actual full dose enhanced contrast inevitably increases the motion / jagged distortion artifacts that are best suppressed when estimating the contrast of low intensity contrast enhancements. The impact of being able to reduce the contrast dose by 90% and retain diagnostic information can have a major impact on patient well-being and imaging costs.
    [044] [044] The illustrative mode described above uses a 2D CNN. It is possible to obtain additional performance gains in 3D CNN models, considering the spatial information correlated in adjacent thin slices. Thus, the implementation of the present invention illustrated with a 2D CNN is only an illustration, and 3D implementations are envisaged within the scope of the present invention. More generally, 2D algorithms (single image slice process) are provided,
    [045] [045] In addition, the inventors predict that, with more data sets, more complicated but powerful models, such as generative models, can be trained to further improve contrast synthesis. Finally, although the loss of the WEM is used as a cost function to train the DL model in the illustrative modality above, other loss functions are foreseen, such as loss of mixture functions with non-local structural similarities or the use of the Generative Adversarial Network ( GAN) as a data-driven cost function.
    [046] [046] In the illustrative mode described above, a fixed level of gadolinium contrast is used with a 90% reduction from the original clinical use level.
    [047] [047] More generally, the term total dose is used in this document to refer to a standard dosage that allows for a certain defined visual quality.
    [048] [048] However, further low dose reduction is expected to be feasible using alternative network structures and loss function, using larger training data sets with various pathologies, in addition to other complementary information, such as multiple contrast MRI , multiple imaging modalities and information on the patient's clinical history. It is also envisaged that the technique may make use of a fragment-based method, in which the size of the input image may be different from 128 x 128 x 1. It can be any size x × y × z that uses a fragment of size x × y and consider the depth / thickness of z.
    [049] [049] It is also envisaged that low dose levels greater than 10% can be used. In general, it is beneficial to minimize the dose subject to the restriction that the synthesized image retains the quality of the diagnosis. The resulting minimum low dose may vary depending on the contrast agent and the imaging modality. It is also possible to base the training on more than two images with different dose levels and it is not necessary to include a zero dose image. In general, if the training data set has a sufficient number of tests paired with two distinct dose levels, the network can be trained to automatically generalize to improve the lowest low dose to the highest high dose, where the highest level high dose is a total or lower dose, and the lowest dose level is lower than the high dose, that is, 0 ≤ d low <dal ≤ 1.
    [050] [050] The inventors also envisage the synthesis of high dose images from zero dose images, without the need for contrast dose images. The zero dose images in this case include several MR images acquired using different sequences (for example, T1w, T2w, FLAIR, DWI) that show different appearance / intensity of different tissues. The trained network can predict a
    Total contrast T1w synthesized for use in diagnosis. So, according to this method, the deep learning network is trained with a first image acquired with a first image acquisition sequence and a second image acquired with a second image acquisition sequence distinct from the first image acquisition sequence, in which the zero dose of contrast agent is administered during the diagnostic image when these images are acquired.
    In addition, images with other sequences can also be used. For example, in one mode, there are five images with five different sequences acquired. The training uses a fundamental reference T1w image acquired with the total dose of contrast agent. The images are pre-processed to co-register and normalize them to adjust the differences in acquisition and scaling. Normalization may include normalization of intensity within each contrast using the histogram corresponding to the median. Pre-processing can also include polarization field correction to correct the polarization field distortion via the N4ITK algorithm. Pre-processing may also include the use of a trained classifier (VGG16) to filter out more abnormal slices. The deep learning network is preferably a deep 2.5D convolutional adversary network. Using the trained network, a first image and a second image obtained with different sequences are used as input to the deep learning network (DLN), which outputs a synthesized image of the individual's full-dose contrast agent.
    权利要求:
    Claims (8)
    [1]
    1. Method for training an imaging diagnostic device to perform an image for medical diagnosis with a reduced dose of contrast agent, the method characterized by the fact that it comprises the steps of: a) performing an image diagnosis of a group of individuals to produce a set of images comprising, for each individual in the set of individuals, (i) a full contrast image acquired with a total dose of contrast agent administered to the individual, (ii) a low contrast image acquired with a low contrast agent dose administered to the individual, where the low dose of contrast agent is less than the total dose of contrast agent, and (iii) a zero contrast image acquired without any dose of contrast agent administered to the individual; b) pre-process the set of images to co-register and normalize the set of images to adjust the differences in acquisition and scaling between different readings; c) training the deep learning network (DLN) with the pre-processed image set by applying zero contrast images from the image set and low contrast images from the image set as input to the DLN and using a cost function to compare the DLN output with full contrast images from the set of images to train the DLN parameters using backpropagation.
    [2]
    2. Method, according to claim 1, characterized by the fact that the cost function is a function of loss of WEM or a function of loss of mixture with non-local structural similarities.
    [3]
    3. Imaging method for medical diagnosis with reduced dose of contrast agent, the method characterized by the fact that it comprises: a) performing an image diagnosis of an individual to produce a low contrast image acquired with a low dose of contrast agent administered to the individual, where the low dose of contrast agent is less than the total dose of contrast agent, and an image of zero contrast acquired without any dose of contrast agent administered to the individual; b) pre-process the low contrast image and the zero contrast image to co-register and normalize the images to adjust the differences in acquisition and scaling; c) apply the low contrast image and the zero contrast image as an input to a deep learning network (DLN) to generate the individual synthesized full dose contrast agent image as an output of the DLN; in which DLN was trained to apply zero contrast and low contrast images as input and full contrast images as fundamental reference images.
    [4]
    4. Method according to claim 3, characterized by the fact that the low dose of contrast agent is less than 10% of the total dose of contrast agent.
    [5]
    5. Method, according to claim 3, characterized by the fact that performing diagnostic imaging includes performing angiography, fluoroscopy, computed tomography (CT), ultrasound, or magnetic resonance imaging.
    [6]
    6. Method, according to claim 3, characterized by the fact that performing diagnostic imaging comprises performing magnetic resonance imaging, and the total dose of contrast agent is at most 0.1 mmol / kg of contrast with Gadolinium for MRI.
    [7]
    7. Method, according to claim 3, characterized by the fact that DLN is a convolutional encoder-decoder neural network (CNN) that includes bypassing concatenated connections and residual connections.
    [8]
    8. Medical diagnostic imaging method, the method characterized by the fact that it comprises: a) performing an individual image diagnosis to produce a first image acquired with a first image acquisition sequence and a second image acquired with a second image sequence distinct image acquisition from the first image acquisition sequence, where zero dose of contrast agent is administered during imaging diagnosis;
    b) pre-process the first image and the second image to co-
    register and normalize the images to adjust the differences in acquisition and scaling;
    c) apply the first image and the second image as an entrance to a deep learning network (DLN) to generate the individual's synthesized total dose contrast agent image as an output of the DLN;
    where DLN was trained to apply zero contrast images with different image sequences as input and full contrast images acquired with the total dose of contrast agent as fundamental reference images.
    类似技术:
    公开号 | 公开日 | 专利标题
    BR112020007105A2|2020-09-24|method for training a diagnostic imaging device to perform a medical diagnostic imaging with a reduced dose of contrast agent
    Torrado-Carvajal et al.2016|Fast patch-based pseudo-CT synthesis from T1-weighted MR images for PET/MR attenuation correction in brain studies
    Xi et al.2015|Simultaneous CT-MRI reconstruction for constrained imaging geometries using structural coupling and compressive sensing
    Willemink et al.2014|Computed tomography radiation dose reduction: effect of different iterative reconstruction algorithms on image quality
    Goubran et al.2013|Image registration of ex-vivo MRI to sparsely sectioned histology of hippocampal and neocortical temporal lobe specimens
    Xiang et al.2018|Ultra-fast t2-weighted mr reconstruction using complementary t1-weighted information
    Yu et al.2020|Medical image synthesis via deep learning
    Kim et al.2021|Deep learning–based image reconstruction for brain CT: improved image quality compared with adaptive statistical iterative reconstruction-Veo |
    Huang et al.2019|Arterial spin labeling images synthesis from sMRI using unbalanced deep discriminant learning
    Jun et al.2019|Parallel imaging in time‐of‐flight magnetic resonance angiography using deep multistream convolutional neural networks
    Taguchi et al.2018|“X-map 2.0” for edema signal enhancement for acute ischemic stroke using non–contrast-enhanced dual-energy computed tomography
    Roy et al.2017|Synthesizing CT from ultrashort echo-time MR images via convolutional neural networks
    Seith et al.2016|Comparison of Positron Emission Tomography Quantification Using Magnetic Resonance–and Computed Tomography–Based Attenuation Correction in Physiological Tissues and Lesions: A Whole-Body Positron Emission Tomography/Magnetic Resonance Study in 66 Patients
    Mostafapour et al.2021|Feasibility of deep learning-guided attenuation and scatter correction of whole-body 68Ga-PSMA PET studies in the image domain
    Oh et al.2020|Semantic segmentation of white matter in FDG-PET using generative adversarial network
    Nguyen et al.2020|Applying artificial intelligence to mitigate effects of patient motion or other complicating factors on image quality
    Roy et al.2016|Patch based synthesis of whole head MR images: Application to EPI distortion correction
    Bauer et al.2021|Generation of annotated multimodal ground truth datasets for abdominal medical image registration
    Gong et al.2020|Parameter-transferred Wasserstein generative adversarial network | for low-dose PET image denoising
    Catana2020|Attenuation correction for human PET/MRI studies
    Guo et al.2020|Anatomic and molecular mr image synthesis using confidence guided cnns
    Kunz et al.2017|Wavelet-based angiographic reconstruction of computed tomography perfusion data: Diagnostic value in cerebral venous sinus thrombosis
    Farrag et al.2021|Evaluation of fully automated myocardial segmentation techniques in native and contrast‐enhanced T1‐mapping cardiovascular magnetic resonance images using fully convolutional neural networks
    Vey et al.2019|The role of generative adversarial networks in radiation reduction and artifact correction in medical imaging
    Matsubara et al.2021|Prediction of an oxygen extraction fraction map by convolutional neural network: validation of input data among MR and PET images
    同族专利:
    公开号 | 公开日
    AU2018346938A1|2020-04-23|
    CA3078728A1|2019-04-18|
    CN111601550A|2020-08-28|
    US20190108634A1|2019-04-11|
    US20210241458A1|2021-08-05|
    KR20200063222A|2020-06-04|
    JP2020536638A|2020-12-17|
    WO2019074938A1|2019-04-18|
    EP3694413A1|2020-08-19|
    SG11202003232VA|2020-05-28|
    US10997716B2|2021-05-04|
    EP3694413A4|2021-06-30|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US7305111B2|2004-01-30|2007-12-04|University Of Chicago|Automated method and system for the detection of lung nodules in low-dose CT images for lung-cancer screening|
    US9378548B2|2011-12-01|2016-06-28|St. Jude Children's Research Hospital|T2 spectral analysis for myelin water imaging|
    EP2890300B1|2012-08-31|2019-01-02|Kenji Suzuki|Supervised machine learning technique for reduction of radiation dose in computed tomography imaging|
    WO2015009830A1|2013-07-16|2015-01-22|Children's National Medical Center|Three dimensional printed replicas of patient's anatomy for medical applications|
    US9700219B2|2013-10-17|2017-07-11|Siemens Healthcare Gmbh|Method and system for machine learning based assessment of fractional flow reserve|
    US20150282719A1|2014-04-02|2015-10-08|University Of Virginia Patent Foundation|Systems and methods for medical imaging incorporating prior knowledge|
    US10335105B2|2015-04-28|2019-07-02|Siemens Healthcare Gmbh|Method and system for synthesizing virtual high dose or high kV computed tomography images from low dose or low kV computed tomography images|
    US10547873B2|2016-05-23|2020-01-28|Massachusetts Institute Of Technology|System and method for providing real-time super-resolution for compressed videos|
    US20180018757A1|2016-07-13|2018-01-18|Kenji Suzuki|Transforming projection data in tomography by means of machine learning|
    US10096109B1|2017-03-31|2018-10-09|The Board Of Trustees Of The Leland Stanford Junior University|Quality of medical images using multi-contrast and deep learning|
    US11100621B2|2017-10-20|2021-08-24|Imaging Biometrics, Llc|Simulated post-contrast T1-weighted magnetic resonance imaging|
    US10482600B2|2018-01-16|2019-11-19|Siemens Healthcare Gmbh|Cross-domain image analysis and cross-domain image synthesis using deep image-to-image networks and adversarial networks|
    US10665011B1|2019-05-31|2020-05-26|Adobe Inc.|Dynamically estimating lighting parameters for positions within augmented-reality scenes based on global and local features|US11100621B2|2017-10-20|2021-08-24|Imaging Biometrics, Llc|Simulated post-contrast T1-weighted magnetic resonance imaging|
    US11164067B2|2018-08-29|2021-11-02|Arizona Board Of Regents On Behalf Of Arizona State University|Systems, methods, and apparatuses for implementing a multi-resolution neural network for use with imaging intensive applications including medical imaging|
    EP3857515A1|2018-09-28|2021-08-04|Mayo Foundation for Medical Education and Research|Systems and methods for multi-kernel synthesis and kernel conversion in medical imaging|
    GB201912701D0|2019-09-04|2019-10-16|Univ Oxford Innovation Ltd|Method and apparatus for enhancing medical images|
    WO2021052850A1|2019-09-18|2021-03-25|Bayer Aktiengesellschaft|Generation of mri images of the liver|
    WO2021052896A1|2019-09-18|2021-03-25|Bayer Aktiengesellschaft|Forecast of mri images by means of a forecast model trained by supervised learning|
    WO2021061710A1|2019-09-25|2021-04-01|Subtle Medical, Inc.|Systems and methods for improving low dose volumetric contrast-enhanced mri|
    WO2021069343A1|2019-10-11|2021-04-15|Bayer Aktiengesellschaft|Acceleration of mri examinations|
    US10984530B1|2019-12-11|2021-04-20|Ping An TechnologyCo., Ltd.|Enhanced medical images processing method and computing device|
    EP3872754A1|2020-02-28|2021-09-01|Siemens Healthcare GmbH|Method and system for automated processing of images when using a contrast agent in mri|
    KR102316312B1|2021-02-01|2021-10-22|주식회사 클라리파이|Apparatus and method for contrast amplification of contrast-enhanced ct images based on deep learning|
    法律状态:
    2021-11-23| B350| Update of information on the portal [chapter 15.35 patent gazette]|
    优先权:
    申请号 | 申请日 | 专利标题
    US201762570068P| true| 2017-10-09|2017-10-09|
    US62/570,068|2017-10-09|
    PCT/US2018/055034|WO2019074938A1|2017-10-09|2018-10-09|Contrast dose reduction for medical imaging using deep learning|
    [返回顶部]