Quality assurance workflow for low-field MRI prostate diagnostic system

Abstract

Quality assurance workflow for low-field MRI prostate diagnostic system. Systems and methods for performing quality assessment of medical imaging analysis tasks are provided. At least one low-field MRI (magnetic resonance imaging) quality assurance imaging data of a patient is received. Using one or more machine learning-based networks, a quality assessment of a medical imaging analysis task is performed based on the at least one low-field MRI quality assurance imaging data. A result of the quality assessment is output.

Description

[0001] Cross-reference to related applications

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 267,735, filed February 9, 2022, the disclosure of which is incorporated herein by reference in its entirety. Technical Field

[0003] This invention generally relates to automated low-field MRI (magnetic resonance imaging) diagnostic systems, and more specifically to a quality assurance workflow for low-field MRI prostate cancer diagnostic systems. Background Technology

[0004] Compared to high-field MRI (magnetic resonance imaging) systems commonly used for diagnostic imaging, low-field MRI systems use a relatively low-intensity MR field to provide low-cost imaging. These low-field MRI systems have a smaller footprint and do not require magnetic shielding, thus allowing them to be installed in more publicly accessible locations, such as pharmacies or medical clinics. However, due to the low-intensity MR field, the low-field MRI images produced by these systems are of relatively low quality and insufficient for radiologists to interpret. Recently, various methods have been proposed for automated analysis of low-field MRI images, such as for prostate cancer detection. However, patient movement and image artifacts can further degrade the quality of low-field MRI images, potentially leading to inaccurate automated analysis. Summary of the Invention

[0005] According to one or more embodiments, systems and methods are provided for quality assessment of low-field MRI quality assurance images for patients. The results of the quality assessment can be used, for example, to inform the patient that the results of an automated assessment of the low-field MRI diagnostic images may be inaccurate and that the patient should have the low-field MRI diagnostic images reacquired. Advantageously, the quality assessment according to the embodiments described herein enables an automated workflow for the automated analysis of low-field MRI diagnostic images, which can be implemented in pharmacies, medical clinics, or other publicly accessible locations without the need for guidance, examination, or intervention by radiologists or other clinicians.

[0006] A system and method are provided for quality assessment of medical imaging analysis tasks. The system receives at least one low-field MRI (magnetic resonance imaging) quality assurance image from a patient. Using one or more machine learning-based networks, a quality assessment of the medical imaging analysis task is performed based on at least one low-field MRI quality assurance image. The results of the quality assessment are output.

[0007] In one embodiment, quality assessment of a medical imaging analysis task includes using one or more machine learning-based networks to detect at least one of patient mislocalization or image artifacts based on at least one low-field MRI quality assurance imaging data. In one embodiment, the one or more machine learning-based networks are trained using a supervised method with labeled training data to classify at least one low-field MRI quality assurance imaging data as either high-quality or low-quality. In another embodiment, the one or more machine learning-based networks are trained using an unsupervised method with unlabeled training data to determine whether at least one low-field MRI quality assurance imaging data is outside the distribution of the unlabeled training data.

[0008] In one embodiment, a medical imaging analysis task is performed based on one or more low-field MRI diagnostic imaging data acquired from a patient. At least one low-field MRI quality assurance imaging data set includes first low-field MRI quality assurance imaging data acquired before acquiring one or more low-field MRI diagnostic imaging data sets and second low-field MRI quality assurance imaging data acquired after acquiring one or more low-field MRI diagnostic imaging data sets. A quality assessment of the medical imaging analysis task is performed based on the first and second low-field MRI quality assurance imaging data sets.

[0009] In one embodiment, the quality assessment of a medical imaging analysis task can be performed by: extracting a first set of features from first low-field MRI quality assurance imaging data using one or more machine learning-based networks; extracting a second set of features from second low-field MRI quality assurance imaging data using one or more machine learning-based networks; and calculating the distance between the first set of features and the second set of features. In one embodiment, the one or more machine learning-based networks include a Siamese neural network, and a first sub-network of the Siamese neural network is used to extract the first set of features from the first low-field MRI quality assurance imaging data, and a second sub-network of the Siamese neural network is used to extract the second set of features from the second low-field MRI quality assurance imaging data.

[0010] In one embodiment, one or more additional low-field MRI quality assurance imaging data of a patient are received between first low-field MRI quality assurance imaging data and second low-field MRI quality assurance imaging data. A quality assessment is performed based on the first low-field MRI quality assurance imaging data, the second low-field MRI quality assurance imaging data, and one or more additional low-field MRI quality assurance imaging data.

[0011] In one embodiment, at least one low-field MRI quality assurance imaging data includes at least one of raw k-space data, MR fingerprint data, or reconstructed images.

[0012] These and other advantages of the present invention will be apparent to those skilled in the art from the following detailed description and accompanying drawings. Attached Figure Description

[0013] Figure 1 A method for quality assessment of performing a medical imaging analysis task is illustrated according to one or more embodiments;

[0014] Figure 2 This illustrates a framework for training a twin network to compare first and second low-field MRI quality-assured imaging data, according to one or more embodiments.

[0015] Figure 3 This illustrates a workflow for automated detection of prostate cancer using a low-field MRI diagnostic system, according to one or more embodiments.

[0016] Figure 4 This document illustrates a workflow for the automated detection of prostate cancer using a low-field MRI diagnostic system based on additional low-field MRI quality assurance imaging data, according to one or more embodiments.

[0017] Figure 5 An exemplary artificial neural network is shown that can be used to implement one or more embodiments;

[0018] Figure 6 This illustrates a convolutional neural network that can be used to implement one or more embodiments; and

[0019] Figure 7 A high-level block diagram is shown that can be used to implement one or more embodiments of a computer. Detailed Implementation

[0020] This invention generally relates to methods and systems for quality assurance workflows in low-field MRI prostate diagnostic systems. Embodiments of the invention are described herein to provide an intuitive understanding of these methods and systems. Digital images typically consist of digital representations of one or more objects (or shapes). The digital representations of objects are generally described herein in terms of identifying and manipulating them. Such manipulation is a virtual manipulation performed in the memory or other circuitry / hardware of a computer system. Therefore, it should be understood that embodiments of the invention can be performed within a computer system using data stored within the computer system. Embodiments disclosed herein will be described with reference to the accompanying drawings, wherein like reference numerals denote like or similar elements.

[0021] The embodiments disclosed herein provide quality assessments for medical imaging analysis tasks based on one or more low-field MRI quality assurance imaging data acquired for a patient. The results of the quality assessment provide an indication of the accuracy of the medical imaging analysis task. Advantageously, the embodiments described herein can be implemented in a low-field MRI diagnostic system to achieve a fully automated workflow for performing medical imaging analysis tasks automatically, without the need for guidance, examination, or intervention from a radiologist or other clinician. For example, the low-field MRI diagnostic system can be installed as a kiosk in a publicly accessible location, such as a pharmacy or medical clinic, enabling patients to acquire low-field MRI diagnostic images themselves for automated analysis (e.g., prostate cancer detection). If the results of the quality assessment indicate that the accuracy of the automated analysis may be low, the patient can be notified and prompted to reacquire low-field MRI diagnostic images for repeated automated analysis.

[0022] Figure 1 A method 100 for performing a medical imaging analysis task according to one or more embodiments is illustrated, wherein the medical imaging analysis task is performed based on one or more low-field MRI diagnostic medical images of a patient. The steps of method 100 may be performed by, for example... Figure 7 The computer 702 or one or more suitable computing devices are used to perform the operation.

[0023] exist Figure 1 At step 102, at least one low-field MRI quality assurance imaging data of the patient is received. This at least one low-field MRI quality assurance imaging data is acquired from a low-field MRI imaging device. Compared to typical MRI imaging devices used in hospitals and imaging centers (generally in the range of 1.5 to 3 T (Tesla), a low-field MRI imaging device is an MRI imaging device that uses a relatively low-intensity magnetic field to generate low-field MRI imaging data of the patient. In one example, the low-field MRI imaging device uses a magnetic field of less than or equal to 0.1 T.

[0024] In one embodiment, at least one low-field MRI quality-assured imaging data belongs to the patient's prostate. However, at least one low-field MRI quality-assured imaging data may belong to any other anatomical object of the patient, such as, for example, other organs, bones, lesions, or any other anatomical object of interest. At least one low-field MRI quality-assured imaging data may include raw k-space data represented as a digital array representing spatial frequencies, MR fingerprint data quantifying one or more properties of a material or tissue, and / or reconstructed images generated from the raw k-space data (e.g., by applying a Fourier transform to the raw k-space data).

[0025] The at least one low-field MRI quality assurance imaging data can be received directly from the low-field MRI imaging device during imaging data acquisition, or it can be received by loading previously acquired imaging data from the storage device or memory of the computer system or by receiving imaging data that has been transmitted from a remote computer system.

[0026] At step 104, a quality assessment for a medical imaging analysis task is performed using one or more machine learning-based networks based on at least one low-field MRI quality assurance imaging dataset. The one or more machine learning-based networks are trained during a previous offline training phase. Once trained, the one or more machine learning-based networks are applied during the inference phase, for example, to perform step 106.

[0027] A medical imaging analysis task is performed based on one or more low-field MRI diagnostic medical images of a patient. In one embodiment, the medical imaging analysis task includes the detection of cancer (e.g., prostate cancer). However, the medical imaging analysis task can be any other suitable medical imaging analysis task performed on one or more low-field MRI diagnostic medical images, such as, for example, anatomical landmark detection, segmentation, classification, etc. Any suitable method can be used to perform the medical imaging analysis task. For example, a machine learning-based prostate cancer detection network can be used to automate the detection of prostate cancer. In one embodiment, the medical imaging analysis task is performed according to a known method.

[0028] In one embodiment, quality assessment is performed by detecting incorrect patient positioning and image artifacts in at least one low-field MRI quality assurance imaging data. The at least one low-field MRI quality assurance imaging data may be acquired before acquiring one or more low-field MRI diagnostic medical images or at any other suitable time (e.g., after acquiring one or more low-field MRI diagnostic medical images or between acquiring one or more low-field MRI diagnostic medical images). In this embodiment, a machine learning-based network is trained during a previous offline training phase to identify incorrect patient positioning and image artifacts in at least one low-field MRI quality assurance imaging data (e.g., due to metal inside the patient or wearable devices on the patient). The machine learning-based network can be trained using supervised or unsupervised methods. In supervised methods, low-field MRI training imaging data is acquired from a large group of individuals in various positions (e.g., poses) and configurations, and the image quality of the training imaging data is manually labeled or scored. The machine learning-based network is trained to classify at least one low-field MRI quality assurance imaging data as high or low quality based on the labeled training imaging data. In unsupervised methods, a machine learning-based out-of-distribution (OOD) detection network is trained to determine whether at least one low-field MRI quality-assured imaging data is outside the distribution of the training data and therefore does not have the desired quality. Training the OOD detection network requires no labels or diagnostics. In one embodiment, the OOD detection network includes an autoencoder (or a variant thereof) architecture that only memorizes key features of the data in the distribution (i.e., the high-quality training images). It is assumed that low-quality images have a distribution significantly different from those in the high-quality distribution so that the autoencoder can recognize them. The OOD detection network can have any other suitable machine learning-based architecture.

[0029] In one embodiment, a quality assessment is performed additionally or alternatively to detect patient motion during the acquisition of one or more low-field MRI diagnostic imaging data. In this embodiment, at least one low-field MRI quality assurance data includes first low-field MRI quality assurance data and second low-field MRI quality assurance data. The first and second low-field MRI quality assurance data are acquired at different times. For example, the first low-field MRI quality assurance data may be acquired before acquiring one or more low-field MRI diagnostic medical images, and the second low-field MRI quality assurance data may be acquired after acquiring one or more low-field MRI diagnostic imaging data. A machine learning-based network is trained during a previous offline training phase to compare the first and second low-field MRI quality assurance imaging data to determine whether the patient remains stationary. In one embodiment, the machine learning-based network is implemented as a Siamese neural network for improved feature extraction for image similarity comparison. Figure 2An exemplary framework for training a twin network to compare first and second low-field MRI quality-assured imaging data is shown.

[0030] Figure 2 A framework 200 is illustrated according to one or more embodiments for training a Siamese network to compare first and second low-field MRI quality-assured imaging data. The steps of framework 200 are performed during a prior offline training phase to train the Siamese network. Once trained, the Siamese network can be applied during an online inference phase, such as, for example, in... Figure 1 Step 106.

[0031] A Siamese network is trained using a pair of low-field MRI training imaging data 202 and 204, which may include raw k-space data, MR fingerprint data, and / or reconstructed images. The Siamese network comprises two identical subnetworks 206 and 208 with the same architecture and parameters (i.e., shared weights). Each Siamese subnetwork 206 and 208 receives the low-field MRI training imaging data 202 and 204 as input and extracts features 210 and 212 as output, respectively. The Siamese subnetworks 206 and 208 can be implemented using a CNN (Convolutional Neural Network) for extracting features 210 and 212. Depending on the size of the training data, different types of CNNs, such as ResNet or DenseNet, can be used. If reconstructed images of the low-field MRI training imaging data 202 and 204 are not available, other types of networks (e.g., transformer networks) can be used to extract features directly from the raw k-space data. The distance or similarity 214 between features 210 and 212 can be calculated based on various distance functions, such as Euclidean distance or cosine similarity. The Siamese network is trained based on loss function 216. In one example, loss function 216 can be a triple loss function as defined in equation (1):

[0032]

[0033] Where A is the anchor (reference) image / measurement, P is the positive example (another image / measurement from the same patient without motion), and N is the negative example (an image / measurement from another case or from the same case but with image quality issues). A Siamese network is trained to search for an optimal model that results in the distance between the encoded features of anchor A and positive example P being less than or equal to the distance between the encoded features of anchor A and negative example N.

[0034] exist Figure 1 At step 106, the results of the quality assessment are output. For example, the results can be output by displaying the results on a display device of the computer system, storing the results on the memory or storage device of the computer system, or by transmitting the results to a remote computer system.

[0035] In one embodiment, if the results of a quality assessment indicate that the outcome of a medical imaging analysis task may be inaccurate (e.g., due to incorrect patient posture, image artifacts, or patient movement), a notification (e.g., an audible or visual prompt) may be presented to the user. This notification may prompt the user to reacquire one or more low-field MRI diagnostic medical images to repeat the medical imaging analysis task.

[0036] In one embodiment, Figure 1 Method 100 can be performed by a low-field MRI diagnostic system. The low-field MRI diagnostic system may include a low-field MRI image acquisition device and one or more computing devices (e.g., Figure 7 The computer 702) is used to execute Figure 1 The method comprises 100 steps. This low-field MRI diagnostic system can be installed as a kiosk in a pharmacy, medical clinic, or any other publicly accessible location, enabling a patient to acquire their own low-field MRI diagnostic images for automated medical imaging analysis tasks (e.g., prostate cancer detection). Advantageously, the quality assessment of the medical imaging analysis tasks performed (e.g., based on…) Figure 1 Method 100 can, for example, inform the patient that the results of a medical imaging analysis task may be inaccurate and prompt the patient to reacquire low-field MRI diagnostic images to re-perform the medical imaging analysis task. Therefore, quality assessment enables an automated workflow for performing medical imaging analysis tasks using a low-field MRI diagnostic system without the need for guidance, examination, or intervention from a radiologist or other clinician. Figure 3 and Figure 4 An exemplary workflow for the automated detection of prostate cancer using a low-field MRI diagnostic system is shown in the figure.

[0037] Figure 3 A workflow 300 for automated detection of prostate cancer using a low-field MRI diagnostic system, according to one or more embodiments, is illustrated. While workflow 300 is executed for automated prostate cancer detection, it can also be executed for any other medical imaging analysis task.

[0038] At step 302, the patient is positioned within the low-field MRI imaging apparatus. The patient may be positioned in a shaped seat to limit changes in patient position.

[0039] At step 304, a first quality assurance (Q / A) scan is acquired. This is done before acquiring all other diagnostic images and measurements in workflow 300 (i.e., before...). Figure 3 The prostate localization scan was acquired at step 306, and in Figure 3Step 310 (Cancer detection scan acquired) acquires the first quality assurance scan. Acquisition of the first quality assurance scan can be performed very quickly because it does not require diagnostic information. Therefore, the spatial resolution of the first quality assurance scan may be relatively low, no contrast is needed within the prostate, and image reconstruction is not necessarily performed.

[0040] At step 306, a prostate localization scan is acquired. The goal of the prostate localization scan is to locate the prostate and measure glandular volume. The prostate localization scan can be a T2-weighted sequence or a DWI (diffusion-weighted imaging) image with a low b-value sequence to estimate glandular volume.

[0041] At step 308, the prostate localization scan is processed. In one embodiment, the AI ​​(artificial intelligence) system can infer the prostate bounding box (i.e., the imaginary rectangle used as a reference point), gland mask (i.e., the precise outline of the gland) or gland volume from the prostate localization scan from the original k-space or from the reconstructed image.

[0042] In one embodiment, T2 or DWI with low b-value scans can be used for prostate localization. A segmentation network implemented as, for example, U-Net or a similar architecture can be applied to compute gland segmentation. The segmentation network receives a reconstructed image with relatively high spatial resolution as input. If multiple sequences (e.g., T2 or DWI) are used, the segmentation network can be applied to different sequences, and Dice coefficients between segments are computed. If the Dice coefficient is significantly less than 1 (e.g., <0.8), this indicates significant inter-contrast motion, which may affect cancer detection.

[0043] In step 310, a cancer detection scan is acquired based on the processed prostate localization scan. The goal of the cancer detection scan is to detect abnormalities in the prostate and assess the risk of malignant disease.

[0044] At step 312, in one embodiment, a second quality assurance scan is acquired. All other diagnostic images and measurements are acquired in workflow 300 (i.e., in...). Figure 3 The prostate localization scan was acquired at step 306, and in Figure 3 After the cancer detection scan (taken at step 310), a second quality assurance scan is taken.

[0045] At step 314, the cancer detection scan is processed to detect cancer in the scan and determine a malignancy risk score 316. An AI system can be implemented to infer the risk score 316 associated with the overall prostate abnormality (and malignancy risk), which is calculated from the acquired signal confined to the glandular region. The risk score 316 can be calculated based on the raw signal (e.g., k-space or MRF (magnetic resonance fingerprint)) or reconstructed images at various spatial resolutions.

[0046] At step 318, a measurement quality assessment is performed based on one or more of the first quality assurance scan (acquired at step 304) and / or the second quality assurance scan (acquired at step 312). In one example, according to Figure 1 Method 100 performs a measurement quality assessment. In one embodiment, the measurement quality assessment may be performed using only one of a first quality assurance scan (acquired at step 304) or a second quality assurance scan (acquired at step 312) to detect incorrect patient positioning (its location at step 302) and image artifacts. In another embodiment, the measurement quality assessment may be performed additionally or alternatively using both the first quality assurance scan (acquired at step 304) and the second quality assurance scan (acquired at step 312) to identify significant patient movement during the acquisition of diagnostic scans / measurements that may affect cancer detection. This comparison may be performed in image space or raw data space, thus image reconstruction is optional.

[0047] In one embodiment, in addition to the first low-field MRI quality-assured imaging data and the second low-field MRI quality-assured imaging data, Figure 1 The at least one low-field MRI quality assurance imaging data received at step 102 may also include one or more additional low-field MRI quality assurance imaging data. For example, additional low-field MRI quality assurance imaging data may be acquired after acquiring some or all diagnostic images (e.g., localization scans, detection scans, etc.) for continuous monitoring of patient movement. Figure 4 An exemplary workflow is shown for the automated detection of prostate cancer using supplemental low-field MRI quality assurance imaging data with a low-field MRI diagnostic system.

[0048] Figure 4 A workflow 400 for automatically detecting prostate cancer using a low-field MRI diagnostic system based on additional low-field MRI quality assurance imaging data, according to one or more embodiments, is shown. Figure 4 The workflow 400 is similar to Figure 3 The workflow is 300, but it incorporates the acquisition of additional low-field MRI quality assurance imaging data. For example... Figure 4As shown, workflow 400 includes step 402, in which an additional quality assurance scan is acquired between the first quality assurance scan (acquired at step 304) and the second quality assurance scan (acquired at step 312). The additional quality assurance scan is acquired after acquiring the prostate localization scan (at step 306) but before acquiring the cancer detection scan (at step 310). Advantageously, the additional quality assurance scan allows for continuous monitoring of patient movement. This ensures that the entire imaging protocol is not discarded due to movement occurring somewhere between the start and end. For example, if patient movement is detected between the additional quality assurance scan (acquired at step 402) and the second quality assurance scan (acquired at step 312), the prostate localization scan (acquired at step 306) and processing (performed at step 308) can be retained, and only the cancer detection scan acquisition step 310 and the cancer detection scan processing step 314 are re-executed.

[0049] To perform quality assessment based on first low-field MRI quality assurance imaging data, second low-field MRI quality assurance imaging data, and one or more additional low-field MRI quality assurance imaging data, a Siamese network can be used in real time to compare each consecutive pair (e.g., to compare the first quality assurance scan acquired at step 304 with the additional quality assurance scan acquired at step 402, and to compare the additional quality assurance scan acquired at step 402 with the second quality assurance scan acquired at step 312) or to compare the first quality assurance scan with each subsequent quality assurance scan (e.g., to compare the first quality assurance scan acquired at step 304 with the additional quality assurance scan acquired at step 402, and to compare the first quality assurance scan acquired at step 304 with the additional quality assurance scan acquired at step 402).

[0050] The embodiments described herein are for both the claimed system and the claimed method. Features, advantages, or alternative embodiments described herein can be assigned to other claims, and vice versa. In other words, the system claims can be modified using features described or claimed in the context of the method. In this case, the functional characteristics of the method are embodied by the objective unit providing the system.

[0051] Furthermore, certain embodiments described herein are directed to methods and systems utilizing trained machine learning-based networks (or models), and to methods and systems for training machine learning-based networks. Features, advantages, or alternative embodiments described herein can be assigned to other claimed objects, and vice versa. In other words, the claims for methods and systems for training machine learning-based networks can be modified with features described or claimed in the context of methods and systems utilizing trained machine learning-based networks, and vice versa.

[0052] Specifically, the trained machine learning-based networks used in the embodiments described herein can be adapted by methods and systems for training machine learning-based networks. Furthermore, the input data of the trained machine learning-based network can include advantageous features and embodiments of the training input data, and vice versa. Similarly, the output data of the trained machine learning network can include advantageous features and embodiments of the output training data, and vice versa.

[0053] Generally speaking, trained machine learning-based networks mimic human cognitive functions associated with other human thought processes. Specifically, by training on training data, trained machine learning-based networks can adapt to new environments and detect and infer patterns.

[0054] Generally, the parameters of machine learning-based networks can be tuned through training. Specifically, supervised training, semi-supervised training, unsupervised training, reinforcement learning, and / or active learning can be used. Additionally, representation learning (an alternative term is "feature learning") can be used. Specifically, the parameters of a trained machine learning-based network can be iteratively tuned through several training steps.

[0055] Specifically, the trained machine learning-based network can include neural networks, support vector machines, decision trees, and / or Bayesian networks, and / or the trained machine learning-based network can be based on k-means clustering, Q-learning, genetic algorithms, and / or association rules. Specifically, the neural network can be a deep neural network, a convolutional neural network, or a convolutional deep neural network. Furthermore, the neural network can be an adversarial network, a deep adversarial network, and / or a generative adversarial network.

[0056] Figure 5 An embodiment of an artificial neural network 500 according to one or more embodiments is shown. Alternative terms for "artificial neural network" are "neural network," "artificial neural network," or simply "neural network." The artificial neural network 500 can be used to implement the machine learning networks described herein, such as, for example, in... Figure 1 Step 106 utilizes one or more machine learning-based networks and Figure 2Subnets 206 and 208 of the twin network.

[0057] The artificial neural network 500 includes nodes 502-522 and edges 532, 534, ..., 536, where each edge 532, 534, ..., 536 is a directed connection from a first node 502-522 to a second node 502-522. Generally, the first node 502-522 and the second node 502-522 are different nodes 502-522, but it is also possible that the first node 502-522 and the second node 502-522 are the same. For example, in... Figure 5 In the diagram, edge 532 is a directed connection from node 502 to node 506, and edge 534 is a directed connection from node 504 to node 506. Edges 532, 534, ..., 536 from the first node 502-522 to the second node 502-522 are also represented as the "input edges" of the second node 502-522 and the "output edges" of the first node 502-522.

[0058] In this embodiment, nodes 502-522 of the artificial neural network 500 can be arranged in layers 524-530, wherein the layers can include an inherent order introduced by edges 532, 534, ..., 536 between nodes 502-522. Specifically, edges 532, 534, ..., 536 can only exist between adjacent layers of nodes. Figure 5 In the illustrated embodiment, the following layers exist: an input layer 524 comprising only nodes 502 and 504 without input edges, an output layer 530 comprising only node 522 without output edges, and hidden layers 526 and 528 located between the input layer 524 and the output layer 530. Generally, the number of hidden layers 526 and 528 can be arbitrarily chosen. The number of nodes 502 and 504 in the input layer 524 is typically related to the number of input values ​​of the neural network 500, and the number of nodes 522 in the output layer 530 is typically related to the number of output values ​​of the neural network 500.

[0059] Specifically, (real) numbers can be assigned as values ​​to each node 502-522 of the neural network 500. In this paper, x (n) i This represents the value of the i-th node 502-522 in the n-th layer 524-530. The values ​​of nodes 502-522 in the input layer 524 are equivalent to the input values ​​of neural network 500, and the value of node 522 in the output layer 530 is equivalent to the output value of neural network 500. Furthermore, each edge 532, 534, ..., 536 may include a weight as a real number, specifically a real number within the interval [-1, 1] or the interval [0, 1]. In this paper, w... (m，n) i，jThis represents the weight of the edge between the i-th node 502-522 in layer m (524-530) and the j-th node 502-522 in layer n (524-530). Additionally, the abbreviation w... (n) i，j Defined for weight w (n，n+1) i，j .

[0060] Specifically, to calculate the output value of neural network 500, the input value is propagated through the neural network. Specifically, the values ​​of nodes 502-522 in layer (n+1) 524-530 can be calculated based on the values ​​of nodes 502-522 in layer n 524-530 using the following formula:

[0061]

[0062] In this paper, the function f is the transfer function (another term is the "activation function"). Known transfer functions are step functions, sigmoid functions (e.g., logic functions, generalized logic functions, hyperbolic tangent functions, arctangent functions, error functions, smooth step functions), or rectifier functions. Transfer functions are primarily used for standardization purposes.

[0063] Specifically, the value is propagated layer by layer through the neural network, where the value of the input layer 524 is given by the input of the neural network 500, the value of the first hidden layer 526 can be calculated based on the value of the input layer 524 of the neural network, the value of the second hidden layer 528 can be calculated based on the value of the first hidden layer 526, and so on.

[0064] To set the value w for the edge (m，n) i，j Training data is required to train the neural network 500. Specifically, the training data includes training input data and training output data (denoted as t). i For the training step, neural network 500 is applied to the training input data to generate computational output data. Specifically, the training data and the computational output data include multiple values, the number of which is equal to the number of nodes in the output layer.

[0065] Specifically, the comparison between the calculated output data and the training data is used to recursively adjust the weights within the neural network (backpropagation algorithm). Specifically, the weights change according to the following formula:

[0066]

[0067] Where γ is the learning rate, and if the (n+1)th layer is not an output layer, then the number δ (n) j It can be based on δ (n+1) j Recursively calculate as

[0068]

[0069] Furthermore, if the (n+1)th layer is an output layer 530, then

[0070]

[0071] Where f' is the first derivative of the activation function, and y (n+1) j It is the comparison training value of the j-th node of the output layer 530.

[0072] Figure 6 A convolutional neural network 600 according to one or more embodiments is illustrated. The convolutional neural network 600 can be used to implement the machine learning networks described herein, such as, for example, in... Figure 1 Step 106 utilizes one or more machine learning-based networks and Figure 2 Subnets 206 and 208 of the twin network.

[0073] exist Figure 6 In the illustrated embodiment, the convolutional neural network 600 includes an input layer 602, a convolutional layer 604, a pooling layer 606, a fully connected layer 608, and an output layer 610. Alternatively, the convolutional neural network 600 may include several convolutional layers 604, several pooling layers 606, and several fully connected layers 608, as well as other types of layers. The order of the layers can be arbitrarily chosen; typically, the fully connected layer 608 is used as the last layer before the output layer 610.

[0074] Specifically, within the convolutional neural network 600, nodes 612-620 of layer 602-610 can be considered as arranged as a d-dimensional matrix or a d-dimensional image. Specifically, in the two-dimensional case, the values ​​of nodes 612-620 indexed by i and j in the nth layer 602-610 can be represented as x. (n) [i，j] However, the arrangement of nodes 612-620 in layer 602-610 has no impact on the computations performed within the convolutional neural network 600, since these are given solely by the structure and weights of the edges.

[0075] Specifically, the convolutional layer 604 is characterized by the structure and weights of the input edges that form the convolution operation based on a certain number of kernels. Specifically, the structure and weights of the input edges are chosen such that the value x of node 614 of the convolutional layer 604... (n) k The value x is calculated as the value of node 612 based on the previous layer 602. (n-1) convolution x (n) k =K k *x (n-1)Where convolution* is defined in the two-dimensional case as

[0076]

[0077] This article discusses the k-th kernel K. k It is a d-dimensional matrix (a two-dimensional matrix in this embodiment), which is typically small compared to the number of nodes 612-618 (e.g., a 3×3 or 5×5 matrix). Specifically, this means that the weights of the input edges are not independent, but are chosen such that they produce the convolution equation. Specifically, for the kernel, which is a 3×3 matrix, there are only 9 independent weights (each entry of the kernel matrix corresponds to an independent weight), regardless of the number of nodes 612-620 in the corresponding layers 602-610. Specifically, for convolutional layer 604, the number of nodes 614 in the convolutional layer is equal to the number of nodes 612 in the previous layer 602 multiplied by the number of kernels.

[0078] If the nodes 612 of the preceding layer 602 are arranged as a d-dimensional matrix, then using multiple kernels can be interpreted as adding another dimension (denoted as the "depth" dimension), so that the nodes 614 of the convolutional layer 604 are arranged as a (d+1)-dimensional matrix. If the nodes 612 of the preceding layer 602 have already been arranged as a (d+1)-dimensional matrix including the depth dimension, then using multiple kernels can be interpreted as extending along the depth dimension, so that the nodes 614 of the convolutional layer 604 are also arranged as a (d+1)-dimensional matrix, where the size of the (d+1)-dimensional matrix with respect to the depth dimension is a multiple of the number of kernels in the preceding layer 602.

[0079] The advantage of using convolutional layer 604 is that it can take advantage of the spatial local correlation of the input data by implementing a local connection pattern between nodes of adjacent layers, specifically by each node being connected to only a small region of the nodes of the previous layer.

[0080] exist Figure 6 In the illustrated embodiment, the input layer 602 comprises 36 nodes 612 arranged in a two-dimensional 6×6 matrix. The convolutional layer 604 comprises 72 nodes 614 arranged in two two-dimensional 6×6 matrices, each of which is the result of the convolution of the input layer values ​​with the kernel. Equivalently, the nodes 614 of the convolutional layer 604 can be interpreted as arranged in a three-dimensional 6×6×2 matrix, where the last dimension is the depth dimension.

[0081] Pooling layer 606 can be characterized by the structure and weights of the input edges and the activation function of its nodes 616, which forms a pooling operation based on a nonlinear pooling function f. For example, in the two-dimensional case, the value x of node 616 of pooling layer 606 is... (n) It can be based on the value x of node 614 of the previous layer 604. (n-1) To calculate as follows:

[0082] x (n) [i, j] = f(x) (n-1) [id1, jd2], ..., x (n-1) [id1+d1-1,jd2+d2-1]).

[0083] In other words, by using pooling layer 606, the number of nodes 614 and 616 can be reduced. This is achieved by replacing the number of neighboring nodes 614 (d1, d2) in the previous layer 604 with a single node 616, whereby the single node 616 is calculated as a function of the values ​​of the number of neighboring nodes in the pooling layer. Specifically, the pooling function f can be a maximum function, average, or L2 norm. Specifically, for pooling layer 606, the weights of the input edges are fixed and are not modified through training.

[0084] The advantage of using pooling layer 606 is that it reduces the number of nodes 614 and 616, as well as the number of parameters. This results in a reduction in the computational cost of the network and controls overfitting.

[0085] exist Figure 6 In the illustrated embodiment, pooling layer 606 is max pooling, replacing four adjacent nodes with only one node, where the value is the maximum of the four adjacent node values. Max pooling is applied to each d-dimensional matrix of the previous layer; in this embodiment, max pooling is applied to each of two two-dimensional matrices, thereby reducing the number of nodes from 72 to 18.

[0086] The fully connected layer 608 can be characterized by the fact that most, in particular all, of the edges between nodes 616 of the previous layer 606 and nodes 618 of the fully connected layer 608 exist, and the weight of each edge can be adjusted individually.

[0087] In this embodiment, the nodes 616 of the preceding layer 606 of the fully connected layer 608 are all displayed as a two-dimensional matrix, and are also displayed as unrelated nodes (indicated as a row of nodes, where the number of nodes is reduced for better presentation). In this embodiment, the number of nodes 618 in the fully connected layer 608 is equal to the number of nodes 616 in the preceding layer 606. Alternatively, the number of nodes 616 and 618 can be different.

[0088] Furthermore, in this embodiment, the value of node 620 in output layer 610 is determined by applying the Softmax function to the value of node 618 in the previous layer 608. By applying the Softmax function, the sum of the values ​​of all nodes 620 in output layer 610 is 1, and all values ​​of all nodes 620 in output layer 610 are real numbers between 0 and 1.

[0089] The convolutional neural network 600 may also include ReLU (rectified linear unit) layers or activation layers with nonlinear transfer functions. Specifically, the number and structure of nodes in the ReLU layer are the same as those in the previous layer. Specifically, the value of each node in the ReLU layer is calculated by applying a rectification function to the value of the corresponding node in the previous layer.

[0090] The inputs and outputs of different convolutional neural network blocks can be connected using summation (residual / dense neural networks), element-wise multiplication (attention), or other differentiable operators. Therefore, if the entire pipeline is differentiable, the convolutional neural network architecture can be nested rather than sequential.

[0091] Specifically, a convolutional neural network 600 can be trained based on the backpropagation algorithm. To prevent overfitting, regularization methods can be used, such as dropping nodes 612-620, random pooling, using artificial data, and weight decay based on L1 or L2 norm or maximum norm constraints. Different loss functions can be combined to train the same neural network to reflect the joint training objective. A subset of neural network parameters can be excluded from the optimization to retain weights pre-trained on another dataset.

[0092] The systems, apparatus, and methods described herein can be implemented using digital circuitry or using one or more computers employing well-known computer processors, memory units, storage devices, computer software, and other components. Typically, a computer includes a processor for executing instructions and one or more memories for storing instructions and data. A computer may also include or be coupled to one or more mass storage devices, such as one or more disks, internal hard disks and removable disks, magneto-optical disks, optical disks, etc.

[0093] The systems, apparatus, and methods described herein can be implemented using computers operating in a client-server relationship. Typically, in such systems, the client computer is located remotely to the server computer and interacts via a network. The client-server relationship can be defined and controlled by computer programs running on the respective client and server computers.

[0094] The systems, apparatus, and methods described herein can be implemented in a network-based cloud computing system. In such a system, a server or other processor connected to the network communicates with one or more client computers via the network. For example, a client computer may communicate with the server via a web browser application residing on and running on the client computer. The client computer may store data on the server and access the data via the network. The client computer may send data requests or online service requests to the server via the network. The server may perform the requested service and provide data to one or more client computers(s). The server may also send data adapted to cause the client computer to perform specified functions, such as performing calculations, displaying specified data on a screen, etc. For example, the server may send requests adapted to cause the client computer to perform one or more steps or functions of the methods and workflows described herein, including... Figure 1 Or one or more of steps or functions in steps 3-4. Certain steps or functions of the methods and workflows described herein include... Figure 1 One or more of steps or functions, such as 3-4, may be executed by a server or by another processor in a network-based cloud computing system. Certain steps or functions of the methods and workflows described herein include... Figure 1 One or more of steps 3-4 can be performed by a client computer in a web-based cloud computing system. The steps or functions of the methods and workflows described herein include... Figure 1 One or more of steps 3-4 can be performed by a server and / or by a client computer in a web-based cloud computing system in any combination.

[0095] The systems, apparatus, and methods described herein can be implemented using computer program products tangibly embodied in an information carrier, for example, executed by a programmable processor in a non-transitory machine-readable storage device; and the methods and workflow steps described herein include Figure 1 One or more of steps or functions in 3-4 may be implemented using one or more computer programs executable by such a processor. A computer program is a set of computer program instructions that can be used directly or indirectly in a computer to perform a specific activity or produce a specific result. Computer programs can be written in any form of programming language, including compiled or interpreted languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0096] exist Figure 7A high-level block diagram of an example computer 702, which can be used to implement the systems, apparatus, and methods described herein, is depicted. Computer 702 includes a processor 704 operatively coupled to a data storage device 712 and a memory 710. Processor 704 controls the overall operation of computer 702 by executing computer program instructions that define these operations. The computer program instructions may be stored in data storage device 712 or other computer-readable medium and loaded into memory 710 when execution is required. Therefore, Figure 1 The methods and workflow steps or functions of 3-4 may be defined by computer program instructions stored in memory 710 and / or data storage device 712, and controlled by processor 704 that executes the computer program instructions. For example, the computer program instructions may be implemented as computer-executable code programmed by a person skilled in the art to perform... Figure 1 Or 3-4 methods and workflow steps or functions. Therefore, by executing computer program instructions, processor 704 performs... Figure 1 Alternatively, methods and workflow steps or functions may be described in 3-4. Computer 702 may also include one or more network interfaces 706 for communicating with other devices via a network. Computer 702 may also include one or more input / output devices 708 that enable users to interact with computer 702 (e.g., monitor, keyboard, mouse, speakers, buttons, etc.).

[0097] Processor 704 may include both general-purpose and special-purpose microprocessors, and may be the sole processor or one of multiple processors in computer 702. For example, processor 704 may include one or more central processing units (CPUs). Processor 704, data storage device 712, and / or memory 710 may include one or more application-specific integrated circuits (ASICs) and / or one or more field-programmable gate arrays (FPGAs), supplemented or incorporated therein.

[0098] Both data storage device 712 and memory 710 include tangible, non-transitory, computer-readable storage media. Both data storage device 712 and memory 710 may include high-speed random access memory, such as dynamic random access memory (DRAM), static random access memory (SRAM), double data rate synchronous dynamic random access memory (DDR RAM), or other random access solid-state storage devices, and may include non-volatile memory, such as one or more disk storage devices, such as internal hard disks and removable disks, magneto-optical disk storage devices, optical disk storage devices, flash memory devices, semiconductor storage devices, such as erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), optical disc read-only memory (CD-ROM), digital universal disc read-only memory (DVD-ROM), or other non-volatile solid-state storage devices.

[0099] Input / output device 708 may include peripheral devices such as printers, scanners, displays, etc. For example, input / output device 708 may include display devices such as cathode ray tube (CRT) or liquid crystal display (LCD) monitors for displaying information to a user, keyboards, and pointing devices such as mice or trackballs, through which the user can provide input to computer 702.

[0100] Image acquisition device 714 can be connected to computer 702 to input image data (e.g., medical images) to computer 702. It is possible to implement image acquisition device 714 and computer 702 as a single device. It is also possible for image acquisition device 714 and computer 702 to communicate wirelessly via a network. In a possible embodiment, computer 702 can be remotely located relative to image acquisition device 714.

[0101] One or more computers, such as computer 702, may be used to implement any or all of the systems and apparatus discussed herein.

[0102] Those skilled in the art will recognize that actual computer or computer system implementations may have other structures and may include other components, and Figure 7 A high-level representation of some components of such a computer for illustrative purposes.

[0103] The foregoing detailed description should be understood as illustrative and exemplary in each respect, and not restrictive, and the scope of the invention disclosed herein is not determined by the detailed description, but by the claims interpreted with the maximum breadth permitted by patent law. It should be understood that the embodiments shown and described herein are merely illustrative of the principles of the invention, and various modifications can be made by those skilled in the art without departing from the scope and spirit of the invention. Various other combinations of features can be implemented by those skilled in the art without departing from the scope and spirit of the invention.

Claims

1. A computer-implemented method, comprising: Receive at least one low-field MRI (magnetic resonance imaging) quality assurance imaging data from the patient; Quality assessment of a medical imaging analysis task is performed using one or more machine learning-based networks based on at least one low-field MRI quality assurance imaging data to detect incorrect patient localization, wherein at least one of the one or more machine learning-based networks was trained during a prior offline training phase to identify incorrect patient localization, wherein the training imaging data is low-field MRI training imaging data acquired from a large group of individuals in various locations and configurations; and Output the results of the quality assessment.

2. The computer-implemented method of claim 1, wherein the quality assessment of performing a medical imaging analysis task based on the at least one low-field MRI quality assurance imaging data using one or more machine learning-based networks further comprises: Using one or more machine learning-based networks, image artifacts are detected based on at least one low-field MRI quality-assured imaging dataset.

3. The computer-implemented method according to claim 2, wherein, At least one of the one or more machine learning-based networks is trained using a supervised method with labeled training data to classify the at least one low-field MRI quality-assured imaging data as either high-quality or low-quality.

4. The computer-implemented method of claim 2, wherein at least one of the one or more machine learning-based networks is trained using an unsupervised method with unlabeled training data to determine whether the at least one low-field MRI quality assurance imaging data is outside the distribution of the unlabeled training data.

5. The computer-implemented method according to claim 1, wherein: The medical imaging analysis task is performed based on one or more low-field MRI diagnostic imaging data acquired from the patient. The at least one low-field MRI quality assurance imaging data includes first low-field MRI quality assurance imaging data acquired before acquiring one or more low-field MRI diagnostic imaging data and second low-field MRI quality assurance imaging data acquired after acquiring one or more low-field MRI diagnostic imaging data, and Quality assessment of medical imaging analysis tasks performed using one or more machine learning-based networks on at least one low-field MRI quality assurance imaging data includes: A quality assessment was performed based on the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data.

6. The computer-implemented method of claim 5, wherein performing quality assessment based on the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data comprises: A first set of features is extracted from the first low-field MRI quality assurance imaging data using one or more machine learning-based networks; A second set of features is extracted from the second low-field MRI quality assurance imaging data using one or more machine learning-based networks; as well as Calculate the distance between the first set of features and the second set of features.

7. The computer-implemented method according to claim 6, wherein: One or more machine learning-based networks include Siamese neural networks. Extracting a first set of features from first low-field MRI quality assurance imaging data using one or more machine learning-based networks includes: extracting a first set of features from the first low-field MRI quality assurance imaging data using a first subnetwork of a Siamese neural network, and Extracting a second set of features from the second low-field MRI quality assurance imaging data includes: using a second subnetwork of a Siamese neural network to extract a second set of features from the second low-field MRI quality assurance imaging data.

8. The computer-implemented method of claim 5, further comprising receiving one or more additional low-field MRI quality assurance imaging data of a patient acquired between the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data, wherein performing a quality assessment based on the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data includes: A quality assessment is performed based on the first low-field MRI quality assurance imaging data, the second low-field MRI quality assurance imaging data, and the one or more additional low-field MRI quality assurance imaging data.

9. The computer-implemented method of claim 1, wherein the at least one low-field MRI quality assurance imaging data comprises at least one of raw k-space data, MR fingerprint data, or reconstructed images.

10. An apparatus comprising: A device for receiving at least one low-field MRI (magnetic resonance imaging) quality assurance imaging data from a patient; An apparatus for performing quality assessment of a medical imaging analysis task using one or more machine learning-based networks based on at least one low-field MRI quality assurance imaging data to detect incorrect patient positioning, wherein at least one of the one or more machine learning-based networks is trained during a prior offline training phase to identify incorrect patient positioning, wherein the training imaging data is low-field MRI training imaging data of a large group of individuals in various locations and configurations. as well as A device for outputting the results of quality assessment.

11. The apparatus of claim 10, wherein the apparatus for performing a quality assessment task of medical imaging analysis using one or more machine learning-based networks based on at least one low-field MRI quality assurance imaging data further comprises: A device for detecting image artifacts using one or more machine learning-based networks based on at least one low-field MRI quality assurance imaging data.

12. The apparatus of claim 11, wherein at least one of the one or more machine learning-based networks is trained using a supervised method with labeled training data to classify the at least one low-field MRI quality-assured imaging data as either high-quality or low-quality.

13. The apparatus of claim 11, wherein at least one of the one or more machine learning-based networks is trained using an unsupervised method with unlabeled training data to determine whether the at least one low-field MRI quality assurance imaging data is outside the distribution of the unlabeled training data.

14. The apparatus according to claim 10, wherein: The medical imaging analysis task is performed based on one or more low-field MRI diagnostic imaging data acquired from the patient. The at least one low-field MRI quality assurance imaging data includes first low-field MRI quality assurance imaging data acquired before acquiring one or more low-field MRI diagnostic imaging data and second low-field MRI quality assurance imaging data acquired after acquiring one or more low-field MRI diagnostic imaging data, and A device for quality assessment of medical imaging analysis tasks using one or more machine learning-based networks based on at least one low-field MRI quality assurance imaging data includes: A device for performing quality assessment based on the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data.

15. A non-transitory computer-readable medium storing computer program instructions, which, when executed by a processor, cause the processor to perform operations, comprising: Receive at least one low-field MRI (magnetic resonance imaging) quality assurance imaging data from the patient; Quality assessment of a medical imaging analysis task is performed using one or more machine learning-based networks based on at least one low-field MRI quality assurance imaging data to detect incorrect patient localization, wherein at least one of the one or more machine learning-based networks was trained during a prior offline training phase to identify incorrect patient localization, wherein the training imaging data is low-field MRI training imaging data acquired from a large group of individuals in various locations and configurations; and Output the results of the quality assessment.

16. The non-transitory computer-readable medium of claim 15, wherein the quality assessment of performing a medical imaging analysis task based on the at least one low-field MRI quality assurance imaging data using one or more machine learning-based networks further comprises: Image artifacts are detected using one or more machine learning-based networks based on at least one low-field MRI quality assurance imaging data.

17. The non-transitory computer-readable medium according to claim 15, wherein: The medical imaging analysis task is performed based on one or more low-field MRI diagnostic imaging data acquired from the patient. The at least one low-field MRI quality assurance imaging data includes first low-field MRI quality assurance imaging data acquired before acquiring one or more low-field MRI diagnostic imaging data and second low-field MRI quality assurance imaging data acquired after acquiring one or more low-field MRI diagnostic imaging data, and Quality assessment of medical imaging analysis tasks performed using one or more machine learning-based networks on at least one low-field MRI quality assurance imaging data includes: A quality assessment was performed based on the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data.

18. The non-transitory computer-readable medium of claim 17, wherein performing quality assessment based on the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data comprises: A first set of features is extracted from the first low-field MRI quality assurance imaging data using one or more machine learning-based networks; A second set of features is extracted from the second low-field MRI quality assurance imaging data using one or more machine learning-based networks; as well as Calculate the distance between the first set of features and the second set of features.

19. The non-transitory computer-readable medium according to claim 18, wherein: The one or more machine learning-based networks include Siamese neural networks. Extracting a first set of features from first low-field MRI quality assurance imaging data using one or more machine learning-based networks includes: extracting a first set of features from the first low-field MRI quality assurance imaging data using a first subnetwork of a Siamese neural network, and Extracting a second set of features from the second low-field MRI quality assurance imaging data includes: using a second subnetwork of a Siamese neural network to extract a second set of features from the second low-field MRI quality assurance imaging data.

20. The non-transitory computer-readable medium of claim 17, further comprising receiving one or more additional low-field MRI quality assurance imaging data of a patient acquired between the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data, wherein performing a quality assessment based on the first low-field MRI quality assurance imaging data and the second low-field MRI quality assurance imaging data comprises: A quality assessment is performed based on the first low-field MRI quality assurance imaging data, the second low-field MRI quality assurance imaging data, and the one or more additional low-field MRI quality assurance imaging data.

Images