System and method for reducing noise and artifacts in magnetic resonance imaging

Abstract

The present invention provides a computer-implemented method for reducing noise and artifacts in medical images. The method includes receiving a series of medical images along a first dimension, wherein the signal in the medical images has a higher correlation than the noise and artifacts in the medical images along the first dimension. The method also includes, for each of a plurality of pixels in the medical images, deriving a series of data points along the first dimension based on the series of medical images, inputting the series of data points into a neural network model, and outputting a signal component in the series of data points. The neural network model is configured to separate the signal component from the noise and artifact components in the series of data points. The method also includes generating a series of corrected medical images based on the output signal components.

Description

Background Art

[0001] The field of the present disclosure relates generally to systems and methods for reducing noise and artifacts, and more particularly, to systems and methods for reducing noise and artifacts in medical images using a neural network model.

[0002] Radial magnetic resonance imaging (MRI), in which k-space raw data is acquired via radial trajectories, is attractive in medical imaging due to its motion robustness and its ability to support accelerated imaging. Although acquisition noise is present in all MRIs, radial MRI is also plagued by artifacts in the form of streak artifacts. Artifacts can be caused by system imperfections such as gradient nonlinearity, poor fat suppression, motion, or undersampled data acquisition in accelerated imaging. In time-resolved imaging such as contrast-enhanced imaging, artifacts are exacerbated for increasing spatial and temporal resolutions due to increasing levels of undersampling. Summary of the invention

[0003] In one aspect, a computer-implemented method for reducing noise and artifacts in medical images is provided. The method includes receiving a series of medical images of a subject volume along a first dimension, wherein the medical images include signals, noise, and artifacts, and the signals have a higher correlation than the noise and artifacts in the first dimension. The method also includes, for each of a plurality of pixels in the medical images, deriving a series of data points along the first dimension based on the series of medical images, inputting the series of data points into a neural network model, and outputting a signal component in the series of data points. The neural network model is configured to separate the signal component in the series of data points from the noise and artifact components in the series of data points. The method also includes generating a series of corrected medical images based on the output signal components in the series of data points for each of the plurality of pixels.

[0004] In another aspect, a noise and artifact reduction system is provided. The noise and artifact reduction system includes a noise and artifact reduction computing device. The noise and artifact reduction computing device includes at least one processor electrically connected to at least one memory device. The at least one processor is programmed to receive a series of medical images of a subject volume along a first dimension, wherein the medical images include signals, noise, and artifacts, and the signals have a higher correlation than the noise and artifacts in the first dimension. The at least one processor is further programmed to derive a series of data points along the first dimension based on the series of medical images for each of a plurality of pixels in the medical images, input the series of data points into a neural network model, and output a signal component in the series of data points. The neural network model is configured to separate the signal component in the series of data points from the noise and artifact components in the series of data points. The at least one processor is further programmed to generate a series of corrected medical images based on the signal component output in the series of data points for each of a plurality of pixels. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Figure 1 is a schematic diagram of an exemplary magnetic resonance imaging (MRI) system.

[0006] Figure 2 is an exemplary noise and artifact reduction system.

[0007] Figure 3 is a flow chart of an exemplary method of reducing noise and artifacts.

[0008] Figure 4A yes Figure 3 A schematic diagram of the data flow in an exemplary embodiment of the method shown.

[0009] Figure 4B Shown Figure 3 Exemplary methods are shown and known methods are used to process magnetic resonance (MR) images.

[0010] Figure 4C Shown Figure 3 Another exemplary method and known method for processing MR images is shown.

[0011] Figure 5A It is a schematic diagram of the neural network model.

[0012] Figure 5B yes Figure 5A Schematic diagram of neurons in the neural network model shown.

[0013] Figure 6 is a schematic diagram of an exemplary neural network.

[0014] Figure 7is a block diagram of an exemplary computing device. DETAILED DESCRIPTION

[0015] The present disclosure includes systems and methods for reducing noise and artifacts in medical images. In the systems and methods described herein, signals in the medical images have higher correlation along the time dimension than noise and artifacts in the medical images. Artifacts are visual abnormalities in the medical images that are not present in the subject, unlike signals representing the subject's anatomical structure, and may be caused by imaging modalities such as pulse sequences, gradient nonlinearity, poor fat saturation, or motion. As used herein, a subject is a human (living or dead), an animal (living or dead), or a part of a human or an animal. Methodological aspects will be discussed in part as will be apparent and in part as will be explicitly discussed in the following description.

[0016] While sparse or low-level reconstruction methods can be used to minimize artifacts, their use is limited due to high computational complexity. The systems and methods described herein use a temporal neural network to suppress artifacts in a time series of images. Unlike most other neural networks, the neural network used herein is trained in an unsupervised manner. The time series of images itself is used to train a neural network that learns to separate features or components in the time series data. The result of this process is a network that reduces noise and artifacts in the image. The trained neural network takes the time series data as input and provides as output data points of the time series with reduced noise and artifacts.

[0017] For the purpose of detailed description, the imaging modality of an MR system, an MR image (such as an MR image acquired by a radial acquisition trajectory), and a series of MR images along the time dimension are used herein only as examples. However, the systems and methods described herein are not limited to MR systems, MR images, or MR images acquired by a radial acquisition trajectory, or the time dimension. The systems and methods described herein can be applied to other imaging modalities, such as computed tomography (CT) and positron emission tomography (PET), to a series of data points along other dimensions (such as the spatial dimension) or across multiple coils, and to a series of images acquired with other MR acquisition trajectories.

[0018] In magnetic resonance imaging (MRI), a subject is placed in a magnet. When the subject is in a magnetic field generated by a magnet, the magnetic moments of nuclei such as protons attempt to align with the magnetic field, but precess around the magnetic field in a random order at the Larmor frequency of the nuclei. The magnetic field of the magnet is called B0 and extends in the longitudinal or z direction. During the acquisition of MRI images, a magnetic field in the xy plane and close to the Larmor frequency (called the excitation field B1) is generated by a radio frequency (RF) coil and can be used to rotate or "tilt" the net magnetic moment Mz of the nucleus from the z direction toward the transverse or xy plane. After the excitation signal B1 terminates, the nucleus emits a signal called an MR signal. In order to use the MR signals to generate an image of the subject, magnetic field gradient pulses (Gx, Gy, and Gz) are used. Gradient pulses are used to scan through k-space, the space or distance of spatial frequency in reverse. There is a Fourier relationship between the acquired MR signals and the image of the subject, so the image of the subject can be derived by reconstructing the MR signals.

[0019] Figure 1 A schematic diagram of an exemplary MRI system 10 is shown. In an 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 scan definitions 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.

[0020] In an exemplary embodiment, the pulse sequence server 20 responds to instructions downloaded from the workstation 12 to operate the gradient system 28 and the radio frequency ("RF") system 30. The instructions are used to generate gradient waveforms and RF waveforms in the MR pulse sequence. The RF coil 38 and the gradient coil assembly 32 are used to perform the defined 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.

[0021] In the exemplary embodiment, a gradient waveform for performing a delimiting scan is generated and applied to a gradient system 28, which excites the gradient coils in a gradient coil assembly 32 to generate a magnetic field gradient G for position encoding of 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 .

[0022] In an exemplary embodiment, the RF system 30 includes an RF transmitter for generating RF pulses used in an MR pulse sequence. The RF transmitter responds to the scan definition 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 can be applied to the RF coil 38 by the RF system 30. The response MR signal detected by the RF coil 38 is received by the RF system 30 and amplified, demodulated, filtered and digitized under the instruction of the command generated by the pulse sequence server 20. The RF coil 38 is described as a transmitter and receiver coil, so 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 that transmits RF pulses and a separate receiver coil that detects MR signals. The transmission channel of the RF system 30 can be connected to the RF transmission coil, and the receiver channel can be connected to a separate RF receiver coil. Typically, the transmission channel is connected to the entire body RF coil 38, and each receiver section is connected to a separate local RF coil.

[0023] In an 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 quadrature component and the Q quadrature component of the received MR signal. The magnitude of the received MR signal can then be determined as the square root of the sum of the squares of the I component and the Q component, as shown in the following equation (1):

[0024]

[0025] And the phase of the received MR signal can also be determined as shown in the following equation (2):

[0026]

[0027] In some embodiments, the pulse sequence server 20 also optionally receives subject data from a physiological acquisition controller 40. The controller 40 receives physiological signals from sensors connected to the subject, such as electrocardiogram ("ECG") signals from electrodes, or respiration signals from a respiratory monitoring device such as bellows. The physiological signals are typically used by the pulse sequence server 20 to synchronize or "gate" the performance of the scan with the heartbeat or respiration of the subject.

[0028] In an exemplary embodiment, the pulse sequence server 20 is also connected to a scan room interface circuit 42, which receives signals from sensors associated with the condition of the subject and the magnet system. Through the scan room interface circuit 42, a patient positioning system 44 receives commands for moving the subject to a desired position before and / or during scanning.

[0029] 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 a 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 the operating parameters of the RF system 30 or the gradient system 28, or to control the view order in which k-space is sampled.

[0030] 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 a two-dimensional or three-dimensional image, applying filters to the reconstructed image, performing back-projection image reconstruction on the acquired MR data, generating functional MR images, and calculating motion or flow images.

[0031] 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 In the example of FIG. 1 , the 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 in a disk storage device 48 or 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.

[0032] Figure 2is a schematic diagram of an exemplary noise and artifact reduction system 200. In an exemplary embodiment, the system 200 includes a noise and artifact reduction computing device 202 configured to reduce noise and artifacts in a medical image. The computing device 202 also includes a neural network model 204. The system 200 may include a second noise and artifact reduction computing device 203. The second noise and artifact reduction computing device 203 may be used to train the neural network model 204, and the noise and artifact reduction computing device 202 may then use the trained neural network model 204. The second noise and artifact reduction computing device 203 may be the same computing device as the noise and artifact reduction computing device 202, such that the training and use of the neural network model 204 are performed on one computing device. Alternatively, the second noise and artifact reduction computing device 203 may be a computing device separate from the noise and artifact reduction computing device 202, such that the training and use of the neural network model 204 are performed on a separate computing device. The noise and artifact reduction computing device 202 may be included in the workstation 12 of the MRI system 10 , or may be included on a separate computing device in communication with the workstation 12 .

[0033] In an exemplary embodiment, the system 200 also includes a metadata editor 206 configured to update metadata associated with the image. The system 200 may also include a user interface manager 208 configured to receive user input.

[0034] Figure 3 3 is a flow chart of an exemplary method 300 for reducing noise and artifacts in medical images. The method 300 may be performed in the noise and artifact reduction system 200. The method 300 includes receiving 302 a series of medical images of a volume in a subject along a first dimension. The series of medical images may be received from any suitable source such as an MRI system 10. The method 300 may also include receiving a series of raw data of a volume in a subject acquired by an imaging modality, wherein the series of raw data is along the first dimension, and reconstructing the series of raw data to derive the series of medical images. For example, the series of raw data is a series of k-space raw data of the abdominal cavity of a subject acquired by a dynamic contrast enhancement (DCE) scan using the MRI system 10. A DCE scan typically images a volume in a subject multiple times while a contrast agent is injected into the subject. A DCE scan may be used to study the dynamic response of a subject to achieve different contrasts of anatomical structures from images acquired without a contrast agent, or both. The series of raw data includes a series of raw data along a time domain or a temporal domain. That is, the series of raw data includes a series of raw data sets acquired at different time points. In this series, each set of raw data may be in two dimensions (2D), which is raw data of a slice, or in three dimensions (3D), which is raw data of multiple slices or raw data acquired by 3D acquisition.

[0035] In an exemplary embodiment, the medical image includes a signal, noise, and artifacts. In the time dimension, the signal has a higher correlation than the noise and artifacts. For each of a plurality of pixels in the medical image, the method 300 also includes reducing 303 the noise and artifacts of the medical image. Reducing 303 the noise and artifacts includes deriving 304 a series of data points along a first dimension based on the series of medical images, inputting 306 the series of data points into the neural network model 204, and outputting 308 the signal components in the series of data points. The neural network model 204 is configured to separate the signal components in the series of data points from the noise and artifact components in the series of data points. The plurality of pixels may be all or some of the pixels in the medical image. For example, if the series of medical images is a medical image with a resolution of 64×64×16 acquired at 12 different time points, the series of medical images will have 12 groups of medical images, each group having a 3D medical image with a resolution of 64×64×16 in the x, y, and z dimensions and a total of 65,536 pixels. The first dimension is the time dimension, and the sequence of data points will have 12 data points. Reducing 303 noise and artifacts may be repeated for all 65,536 pixels, or may be repeated for a portion of the 65,536 pixels. Since the time series of multiple pixels are independent of each other, reducing 303 noise and artifacts for each pixel may be calculated in parallel to increase the speed of the noise / artifact reduction process.

[0036] In an exemplary embodiment, the method 300 further includes generating 310 a series of corrected medical images based on the signal components output in the series of data points for each pixel in the plurality of pixels. Generating 310 the series of corrected medical images may be performed after noise and artifact reduction 303 is completed for all pixels in the plurality of pixels (e.g., Figure 3 303 ), or may be performed within a loop of reducing 303 noise and artifacts for each of a plurality of pixels. When generating 310 a series of corrected medical images, for each of a plurality of pixels, the value of the medical image at the pixel is replaced with the signal component. If the reduction 303 of noise and artifacts is not repeated for all pixels of the medical image, a corrected image may be derived by replacing the value with the output signal component only at the pixel where the reduction 303 of noise and artifacts is performed. Since the neural network model 204 is configured to separate the signal from the noise and artifacts, the corrected medical image has less noise and artifacts than the initially received image.

[0037] Figure 4A A data flow 400 in an exemplary embodiment of method 300 is shown. Figure 4B and Figure 4CA comparison of an image processed using method 300 and an image processed using a known method is shown.

[0038] In an exemplary embodiment, a series of raw k-space data or raw data 402 is received. The series of raw data 402 may be acquired by time-resolved imaging, DCE scanning, diffusion-weighted imaging, parametric mapping, functional MRI, or using multiple acquisitions (NEX). The series of raw data 402 is reconstructed 404 by a first reconstruction pass. The first reconstruction pass may be a direct inversion of the raw data 402 into a medical image, such as using an inverse Fourier transform, without any systematic correction or adjustment to the raw data 402, or a reconstructed image, such as conventional adjustments performed by the system 10, such as filtering the raw data to reduce Gibbs ringing caused by step discontinuities. To further increase computational speed, the reconstructed spatial resolution may be reduced from the resolution of the acquired raw data 402 in the first reconstruction pass. For example, if the imaging resolution of the raw data 402 is 256×256, the raw data 402 may be reconstructed into an image 406 having a resolution of 128×128 or less in the first reconstruction pass. After the first reconstruction pass, a first reconstructed image 406 of the raw data 402 is exported. The first reconstructed image 406 is used to train 407 the neural network model 204. Because the neural network model 204 is trained by the reconstructed images 406 of the same subject, the trained neural network model is patient-specific. Advantageously, using the first reconstructed images 406 to train the neural network model 204 eliminates the need for a large training data set in typical supervised learning. All data required to train the neural network model 204 is extracted from the acquired measurements themselves. In addition, because the trained neural network model is patient-specific, the trained neural network model is adapted to the specific patient and is a form of adaptive reconstruction / processing. Therefore, the neural network model can be robust to modeling mismatch, which can occur when using an offline training strategy.

[0039] In an exemplary embodiment, the raw data 402 may also be reconstructed 408 by a preferred reconstruction to derive a preferred reconstructed image 410. The preferred reconstruction may be a reconstruction with adjustments performed by the system 10. With the adjustments, the preferred reconstructed image 410 has better image quality than the first pass reconstructed image 406. The preferred reconstructed image 410 is input 412 to the trained neural network model 204 and inferred 414 by the neural network model 204 to derive a corrected image 416. The corrected image 416 has reduced noise and artifacts. The corrected image 416 may be displayed on the operator display 14 or the display 46 of the MRI system 10.

[0040] In some embodiments, reconstruction 408 using preferred reconstruction may not be performed. Instead, the raw data 402 is reconstructed 404 by a first-pass reconstruction, and the first-pass reconstructed image 406 is input into and inferred by the trained neural network model 204. The neural network model 204 may be pre-trained. In one example, once the raw data 402 is received, the neural network model 204 may be trained during the scan while waiting for the raw data 402 to be reconstructed 408 by the preferred reconstruction. In another example, once a portion of the raw data 402 representing a time series is received, the neural network model 204 may be trained during the scan while waiting for the rest of the scan to end. In the absence of system adjustments, the first-pass reconstruction is faster than the preferred reconstruction. Because the series of data points along the time dimension are independent for different pixels, the training 407 and inference 414 may be performed in parallel for different pixels. In addition, because the series of data points along the time dimension have far fewer data points than a typical image, the speed of training 407 and inference 414 performed by the neural network model 204 is much faster than that of a neural network model for image processing. Furthermore, the systems and methods described herein are compatible with any reconstruction and post-processing options, as the preferred reconstruction is not limited to a specific reconstruction or post-processing option, and the corrected image 416 can be input into any post-processing module for further post-processing.

[0041] Figure 4B Four stages of a series of images acquired by a DCE scan with a radial imaging trajectory are shown. The top row of images 418 are images without any filtering in the time dimension. There is a large amount of noise and artifacts 424 in image 418. The middle row of images 420 are images processed with a low-level filter in the time dimension to remove the linear component of the noise and artifacts. The amount of noise and artifacts 424 in image 420 is less than the amount of noise and artifacts 424 in image 418. The bottom row of images 422 are images processed with the systems and methods disclosed herein. The noise and artifacts 424 in image 422 are significantly less than the noise and artifacts in image 418 or image 420. Images 418, 420, 422 are reconstructed using a preferred reconstruction, which includes system adjustments.

[0042] Figure 4CImages 468, 470, 472 are shown that are also acquired by DCE scanning with radial imaging tracks but reconstructed by direct inversion or first-pass reconstruction (i.e., reconstruction by inverse Fourier transform of the raw data). Similarly, the bottom row of images 472 are images processed with neural network filtering, while the middle row of images 470 are images processed with low-level filtering, and the top row of images 468 are images without any temporal filtering. Although the images are reconstructed by first-pass reconstruction, the amount of noise and artifacts 424 in image 472 is also much less than the amount of noise and artifacts in image 468 or image 470.

[0043] Figure 5A An exemplary artificial neural network model 204 is shown. The exemplary neural network model 204 includes neuron layers 502, 504-1 to 504-n, and 506, including an input layer 502, one or more hidden layers 504-1 to 504-n, and an output layer 506. Each layer may include any number of neurons, i.e., Figure 5A Where q, r and n can be any positive integer. It should be understood that Figure 5A The depicted structures and configurations of neural networks of varying structures and configurations implement the methods and systems described herein.

[0044] In an exemplary embodiment, the input layer 502 may receive different input data. For example, the input layer 502 includes a first input a1 representing a training image, a second input a2 representing a pattern identified in the training image, a third input a3 representing an edge of the training image, etc. The input layer 502 may include thousands or more inputs. In some embodiments, the number of elements used by the neural network model 204 changes during the training process, and if, for example, during the execution of the neural network, it is determined that some neurons have less relevance, they are bypassed or ignored.

[0045] In an exemplary embodiment, each neuron in the hidden layers 504-1 to 504-n processes one or more inputs from the input layer 502 and / or one or more outputs from a neuron in one of the previous hidden layers to generate a decision or output. The output layer 506 includes one or more outputs, each indicating a label, a confidence factor, a weight describing an input, and / or an output image. However, in some embodiments, the outputs of the neural network model 204 are obtained from the hidden layers 504-1 to 504-n in addition to or instead of the outputs from the output layer 506.

[0046] In some embodiments, each layer has a discrete identifiable function with respect to the input data. For example, if n equals 3, the first layer analyzes the first dimension of the input, the second layer analyzes the second dimension of the input, and the last layer analyzes the third dimension of the input. The dimensions may correspond to aspects that are considered to be of strong certainty, then to those aspects that are considered to be of medium importance, and finally to those aspects that are considered to be less relevant.

[0047] In other embodiments, the layers are not clearly delineated in terms of the functionality they perform.For example, two or more of the hidden layers 504-1 through 504-n may share decisions related to labels, with no single layer making independent decisions regarding labels.

[0048] Figure 5B Depicted is a diagram corresponding to Figure 5A Example neuron 550 of a neuron labeled "1, 1" in hidden layer 504-1 of FIG. For each input to neuron 550 (e.g., Figure 5A The input layer 502 in the input layer is weighted so that the input a1 to a p corresponds to the weights w1 to w2 determined during the training process of the neural network model 204 p .

[0049] In some embodiments, some inputs lack explicit weights, or have weights below a threshold. The weights are applied to a function α (labeled by reference numeral 510), which may be a summation and may produce a value z1 that is input to a function labeled f 1，1 (z1) is a function 520. Function 520 is any suitable linear or nonlinear function. Figure 5B As depicted, function 520 produces a plurality of outputs that may be provided to neurons of subsequent layers, or used as outputs of neural network model 204. For example, the outputs may correspond to index values ​​of a label list, or may be calculated values ​​used as inputs to subsequent functions.

[0050] It should be understood that the structure and function of the depicted neural network model 204 and neurons 550 are for illustration purposes only, and other suitable configurations exist. For example, the output of any given neuron may depend not only on the values ​​determined by past neurons, but also on future neurons.

[0051] The neural network model 204 may include a convolutional neural network (CNN), a deep learning neural network, a reinforcement or enhanced learning module or program, or a combined learning module or program that learns in two or more fields or aspects of interest. Supervised and unsupervised machine learning techniques can be used. In supervised machine learning, the processing element may be provided with an exemplary input and its associated output, and may attempt to discover a general rule that maps the input to the output so that when a subsequent new input is provided, the processing element may accurately predict the correct output based on the discovered rules. The neural network model 204 may be trained using an unsupervised machine learning program. In unsupervised machine learning, the processing element may need to find its own structure in an unlabeled exemplary input. Machine learning may involve identifying and recognizing patterns in existing data in order to facilitate prediction of subsequent data. A model may be created based on an exemplary input so that an effective and reliable prediction of a new input is made.

[0052] In addition or alternatively, a machine learning program can be trained by inputting sample data sets or certain data such as images, object statistics and information into the program. The machine learning program can use a deep learning algorithm that can focus primarily on pattern recognition and can be trained after processing multiple examples. The machine learning program can include Bayesian program learning (BPL), speech recognition and synthesis, image or object recognition, optical character recognition, and / or natural language processing, alone or in combination. The machine learning program can also include natural language processing, semantic analysis, automated reasoning, and / or machine learning.

[0053] Based on these analyses, the neural network model 204 can learn how to identify features and patterns that can then be applied to analyze image data, model data, and / or other data. For example, the model 204 can learn to identify features in a series of data points.

[0054] Figure 6 An exemplary neural network 600 included in the neural network model 204 is shown. The neural network 600 is an autoencoder network. The neural network 600 includes a bottleneck layer 602, which has fewer neurons 550 than each of the input layer 604, the output layer 606, and the other intermediate layers 608. The bottleneck layer 602 forces the neural network 600 to learn the characteristics of the input data. The neural network 600 is used as an example here for illustration purposes only. The neural network 600 may have a different configuration or may be a different type of neural network.

[0055] In operation, a series of data points 610 along the time dimension is input into the input layer 604. Another series of data points 612 is output from the output layer 606. The series of data points 612 has the same number of data points as the series of data points 610, but with reduced noise and artifacts.

[0056] In an exemplary embodiment, the neural network 600 is an unsupervised neural network. Unsupervised neural networks do not require a large amount of labeled training data, which may be challenging for the acquisition of medical images, especially for the same patient or the same application. During the training of the neural network 600, the neural network 600 takes a series or multiple series of training data points as input and outputs a series of data points. The neural network 600 is adjusted by reducing the loss function defined by the user to below a threshold level. Because the signal in a series of data points has a higher correlation along the time dimension than the noise and artifacts, the neural network 600 separates the signal component from the noise and artifact components in the series of data points. In some embodiments, the neural network 600 is trained by an analysis model of medical images. The analysis model can be a model of the series of medical images or the raw data acquired by the imaging modality. For example, a simulator such as a Bloch simulator or an extended phase graph method can be used to simulate MR signals and / or images under predefined experimental conditions. An exponential model can also be used to estimate the MR signals and / or images. The simulated MR signals and / or images are used to train the neural network 600 instead of the original k-space data 402 and / or the first reconstructed image 406 (such as Figure 4A shown).

[0057] For illustration purposes only, a series of data points 610 along the time dimension is used as an example. The series of data points 610 can also be a series of data points across the coil. For example, a series of raw data 402 of a volume of a subject is acquired by multiple coils. A series of medical images 406, 410 of the volume are reconstructed from the series of raw data 402. The 1D series of data points 610 of a specific pixel in the medical images 406, 410 are data points at the specific pixel across multiple coils.

[0058] In another example, the medical images 406, 410 may be segmented into multiple segments, such as foreground, background, or a specific organ, such as the liver. Noise and artifact reduction may be performed individually or only for one or more segments.

[0059] The workstation 12 and noise and artifact reduction computing devices 202, 203 described herein may be any suitable computing device 800 and software implemented therein. Figure 7 800. In an exemplary embodiment, the computing device 800 includes a user interface 804 that receives at least one input from a user. The user interface 804 may include a keyboard 806 that enables the user to enter relevant information. The user interface 804 may also include, for example, a pointing device, a mouse, a stylus, a touch-sensitive panel (e.g., a touchpad and a touch screen), a gyroscope, an accelerometer, a position detector, and / or an audio input interface (e.g., including a microphone).

[0060] In addition, in an exemplary embodiment, the computing device 800 includes a display interface 817 that presents information such as input events and / or verification results to the user. The display interface 817 may also include a display adapter 808 coupled to at least one display device 810. More specifically, in an exemplary embodiment, the display device 810 may be a visual display device such as a cathode ray tube (CRT), a liquid crystal display (LCD), a light emitting diode (LED) display, and / or an "electronic ink" display. Alternatively, the display interface 817 may include an audio output device (e.g., an audio adapter and / or a speaker) and / or a printer.

[0061] The computing device 800 also includes a processor 814 and a memory device 818. The processor 814 is coupled to the user interface 804, the display interface 817, and the memory device 818 via the system bus 820. In an exemplary embodiment, the processor 814 communicates with the user, such as by prompting the user via the display interface 817 and / or by receiving user input via the user interface 804. The term "processor" generally refers to any programmable system, including system and microcontrollers, reduced instruction set computers (RISC), complex instruction set computers (CISC), application specific integrated circuits (ASICs), programmable logic circuits (PLCs), and any other circuits or processors capable of performing the functions described herein. The above examples are exemplary only and are therefore not intended to limit the definition and / or meaning of the term "processor" in any way.

[0062] In the exemplary embodiment, the memory device 818 includes one or more devices that enable information such as executable instructions and / or other data to be stored and retrieved. In addition, the memory device 818 includes one or more computer-readable media, such as, but not limited to, dynamic random access memory (DRAM), static random access memory (SRAM), a solid state disk, and / or a hard disk. In an exemplary embodiment, the memory device 818 stores, but is not limited to, application source code, application object code, configuration data, additional input events, application status, assertion statements, verification results, and / or any other type of data. In an exemplary embodiment, the computing device 800 may also include a communication interface 830 coupled to the processor 814 via the system bus 820. In addition, the communication interface 830 is communicatively coupled to a data acquisition device.

[0063] In an exemplary embodiment, processor 814 may be programmed by encoding operations using one or more executable instructions and providing the executable instructions in memory device 818. In an exemplary embodiment, processor 814 is programmed to select a plurality of measurements received from a data collection device.

[0064] In operation, a computer executes computer executable instructions embodied in one or more computer executable components stored on one or more computer readable media to implement various aspects of the present invention described and / or shown herein. Unless otherwise specified, the execution or implementation order of the operations in the embodiments of the present invention shown and described herein is not required. That is, unless otherwise specified, these operations can be performed in any order, and embodiments of the present invention may include more or fewer operations than those disclosed herein. For example, it is conceivable that performing or implementing a particular operation before, simultaneously with, or after another operation is within the scope of various aspects of the present invention.

[0065] At least one technical effect of the systems and methods described herein includes (a) reduction of noise and artifacts with reduced computational complexity; (b) patient-specific trained neural networks; and (c) compatibility with any reconstruction and post-processing options.

[0066] Exemplary embodiments of systems and methods for reducing noise and artifacts are described in detail above. These systems and methods are not limited to the specific embodiments described herein, but rather the components of the systems and / or operations of the methods may be used independently and separately from other components and / or operations described herein. In addition, the components and / or operations described may also be defined in other systems, methods, and / or devices, or used in conjunction with other systems, methods, and / or devices, and are not limited to practice with only the systems described herein.

[0067] Although specific features of various embodiments of the present invention may be shown in some drawings but not in other drawings, this is for convenience only. According to the principles of the present invention, any feature of a drawing may be referenced and / or claimed in combination with any feature of any other drawing.

[0068] This written description uses examples to disclose the invention, including the best mode, and also to enable those 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. Such other examples are intended to fall within the scope of the claims if they have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements with insignificant differences from the literal language of the claims.

Claims

1. A computer-implemented method for reducing noise and artifacts in a medical image, the method comprising: receiving a series of medical images of a volume of a subject along a first dimension, wherein the medical images include a signal, noise, and an artifact, the signal having a higher correlation than the noise and the artifact in the first dimension; For each pixel of a plurality of pixels in the medical image, deriving a series of data points along the first dimension based on the series of medical images; inputting the series of data points into a neural network model, wherein the neural network model is configured to separate signal components in the series of data points from noise and artifact components in the series of data points; as well as outputting the signal components in the series of data points; as well as A series of corrected medical images are generated based on the signal components output in the series of data points for each pixel in the plurality of pixels. The method of claim 1 , wherein the first dimension is a time dimension.

3. The method according to claim 1, further comprising: receiving a series of raw data of the volume within the subject acquired by an imaging modality, wherein the series of raw data is along the first dimension; as well as The series of raw data is reconstructed to derive the series of medical images. 4 . The method according to claim 3 , wherein the series of raw data comprises a plurality of raw data of the volume acquired by a plurality of coils, and the series of data points derived along the first dimension are data points at specific pixels of the plurality of coils.

5. The method according to claim 3, further comprising: reconstructing the series of raw data using a first-pass reconstruction to derive a series of first-pass reconstructed images; For each pixel of a plurality of pixels in the first-pass reconstructed images, deriving a plurality of series of training data points based on the series of first-pass reconstructed images by deriving a series of training data points along the first dimension based on the series of first-pass reconstructed images; as well as The neural network model is trained using the plurality of series of training data points.

6. The method according to claim 5, further comprising reconstructing the series of raw data using a preferred reconstruction to derive a series of preferred reconstructed images, wherein the series of preferred reconstructed images have higher image quality than the series of first-pass reconstructed images, wherein: Deriving a series of data points further comprises deriving the series of data points along the first dimension for each pixel of the plurality of pixels based on the series of preferred reconstructed images; and Inputting the series of data points also includes inputting the series of data points into a trained neural network model.

7. The method according to claim 3, wherein: reconstructing the series of raw data using a first-pass reconstruction to derive a series of first-pass reconstructed images; Deriving a series of data points further comprises deriving the series of data points along the first dimension for each pixel of the plurality of pixels based on the series of first-pass reconstructed images; and Inputting the series of data points also includes inputting the series of data points into a trained neural network model.

8. The method of claim 1, wherein the neural network model is an unsupervised neural network model.

9. The method of claim 8, wherein the neural network model is trained using a series of unlabeled training images.

10. The method of claim 8, further comprising training the neural network model with an analysis model representing the series of medical images.

11. A noise and artifact reduction system, the noise and artifact reduction system comprising a noise and artifact reduction computing device, the noise and artifact reduction computing device comprising at least one processor electrically coupled to at least one memory device, and the at least one processor being programmed to: receiving a series of medical images of a volume of a subject along a first dimension, wherein the medical images include a signal, noise, and an artifact, the signal having a higher correlation than the noise and the artifact in the first dimension; For each pixel of a plurality of pixels in the medical image, deriving a series of data points along the first dimension based on the series of medical images; inputting the series of data points into a neural network model, wherein the neural network model is configured to separate signal components in the series of data points from noise and artifact components in the series of data points; as well as outputting the signal components in the series of data points; as well as A series of corrected medical images are generated based on the signal components output in the series of data points for each pixel in the plurality of pixels.

12. The system of claim 11, wherein the first dimension is a time dimension.

13. The system of claim 11, wherein the at least one processor is further programmed to: receiving a series of raw data of the volume within the subject acquired by an imaging modality, wherein the series of raw data is along the first dimension; and The series of raw data is reconstructed to derive the series of medical images. 14 . The system of claim 13 , wherein the series of raw data comprises a plurality of raw data of the volume acquired by a plurality of coils, and the series of data points derived along the first dimension are data points at specific pixels of the plurality of coils.

15. The system of claim 13, wherein the at least one processor is further programmed to: reconstructing the series of raw data using a first-pass reconstruction to derive a series of first-pass reconstructed images; For each of the plurality of pixels in the first-pass reconstructed images, deriving a plurality of series of training data points based on the series of first-pass reconstructed images by deriving a series of training data points along the first dimension based on the series of first-pass reconstructed images; and The neural network model is trained using the plurality of series of training data points.

16. The system of claim 15, wherein the at least one processor is further programmed to: reconstructing the series of raw data using preferred reconstruction to derive a series of preferred reconstructed images, wherein the series of preferred reconstructed images have higher image quality than the series of first-pass reconstructed images; deriving the series of data points along the first dimension for each pixel of the plurality of pixels based on the series of preferred reconstructed images; and The series of data points are input into a trained neural network model.

17. The system of claim 13, wherein the at least one processor is further programmed to: reconstructing the series of raw data using a first-pass reconstruction to derive a series of first-pass reconstructed images; deriving the series of data points along the first dimension for each pixel of the plurality of pixels based on the series of first-pass reconstructed images; and The series of data points are input into a trained neural network model.

18. The system of claim 11, wherein the neural network model is an unsupervised neural network model.

19. The system of claim 18, wherein the neural network model is trained using a series of unlabeled training images.

20. The system of claim 18, the at least one processor being further programmed to train the neural network model with an analysis model representative of the series of medical images.

Images