Diffusion model conditioned on multi-domain medical images with missing domains

Abstract

Systems and methods for performing a medical imaging analysis task conditioned on multi-domain medical images with missing modalities are provided. One or more medical images are received, each in a different domain, and a domain code is received that defines a presence of the different domains in a predefined set of domains. One or more weights are determined based on the domain code. One or more parameters of a machine learning based encoder are updated based on the one or more weights. Features are extracted from the one or more medical images using the machine learning based encoder with the one or more updated parameters. The medical imaging analysis task is performed based on the extracted features. A result of the medical imaging analysis task is output.

Description

Technical Field

[0001] This invention generally relates to AI / ML (artificial intelligence / machine learning) based medical imaging analysis, and particularly to diffusion models conditioned on multi-domain medical images with missing domains. Background Technology

[0002] Diffusion models are a type of generative AI model that generates data by simulating a diffusion process involving adding noise to the data and then reversing the process to generate new data. Diffusion models have recently attracted attention due to their broad applicability. However, the applicability of diffusion models to medical imaging is challenging due to the inherent complexity and heterogeneity of medical image data.

[0003] Different types of medical images provide different types of information. For example, CT (computed tomography) images are more effective at capturing bone, air, and blood contrast, while T1, T2, and proton density-weighted MR (magnetic resonance) images are more effective at capturing tissue properties. However, conventional diffusion models typically accept only a single medical image as input for medical imaging analysis and cannot utilize the information provided by multiple medical images from different domains. Furthermore, differences in acquisition protocols across clinical sites often lead to the unavailability of medical images in certain domains. Conventional diffusion models cannot account for such unavailability of medical images. Summary of the Invention

[0004] According to one or more embodiments, a system and method are provided for performing a medical imaging analysis task conditioned on multi-domain medical images with missing modalities. The system receives 1) one or more medical images, each in a different domain, and 2) domain codes defining the presence of different domains in a predefined set of domains. One or more weights are determined based on the domain codes. One or more parameters of a machine learning-based encoder are updated based on the one or more weights. Features are extracted from the one or more medical images using the machine learning-based encoder with the one or more updated parameters. The medical imaging analysis task is performed based on the extracted features. The results of the medical imaging analysis task are output.

[0005] In one embodiment, the domain code also defines the absence of one or more fields from a predefined set in that different domain. Each position of the domain code is associated with a corresponding field in the predefined set. One or more weights are determined by projecting the domain code onto one or more weights using a linear projector.

[0006] In one embodiment, the weight parameters and bias parameters of the machine learning-based encoder are updated. One or more parameters are updated by determining the dot product of one or more parameters of the machine learning-based encoder with a corresponding weight from one or more weights.

[0007] In one embodiment, one or more all-zero tensors from one or more domains of a predefined set of domains that do not exist in different domains are received. One or more medical images are concatenated with one or more all-zero tensors. Features are extracted from the concatenation results.

[0008] In one embodiment, one or more masks of at least one pathology or organ are received. One or more medical images are linked with one or more masks. Features are extracted from the linking results.

[0009] In one embodiment, the medical imaging analysis task includes medical image synthesis.

[0010] These and other advantages of the invention will be apparent to those skilled in the art from the following detailed description and accompanying drawings. Attached Figure Description

[0011] Figure 1 A method for performing a medical imaging analysis task according to one or more embodiments is illustrated; Figure 2 A workflow for extracting features from one or more medical images according to one or more embodiments is shown in order to perform a medical imaging analysis task; Figure 3 A network architecture of DFNControlNet conditioned on multi-domain medical images with missing domains, according to one or more embodiments, is shown. Figure 4 A network architecture of DNFNDDPM / LDM conditioned on multi-domain medical images with missing domains, according to one or more embodiments, is shown. Figure 5 An exemplary artificial neural network is shown that can be used to implement one or more embodiments; Figure 6 A convolutional neural network that can be used to implement one or more embodiments is shown; Figure 7 A data flow diagram is shown using a generative adversarial network that can be used to implement one or more embodiments; Figure 8 A schematic structure is shown that can be used to implement a recursive machine learning model of one or more embodiments; and Figure 9A high-level block diagram of a computer that can be used to implement one or more embodiments is shown. Detailed Implementation

[0012] This invention generally relates to methods and systems for diffusion models conditioned on multi-domain medical images with missing domains. Embodiments of the invention are described herein to provide a visual understanding of the 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 is to be understood that embodiments of the invention can be performed within a computer system using data stored within the computer system. Furthermore, references to image pixels herein can be equivalently used to refer to voxels of an image, and vice versa.

[0013] The embodiments described herein provide a novel image synthesis framework for efficient processing of various multi-to-one medical image synthesis tasks by using diffusion models conditioned on multi-domain medical images with missing domains. The image synthesis framework utilizes a Dynamic Filter Network (DFN) to leverage information provided by medical images from multiple domains and dynamically adapts to missing domain scenarios when certain domains are missing from the input medical image, thereby eliminating the need to train multiple models to manage different scenarios. Advantageously, the image synthesis framework maximizes the utilization of all available data (even in the presence of missing domains) while minimizing computational cost and training effort.

[0014] Figure 1 A method 100 for performing a medical imaging analysis task according to one or more embodiments is illustrated. The steps and sub-steps of method 100 may be performed by one or more suitable computing devices (such as, for example, Figure 7 The computer (702) is used to execute this. Figure 2 A workflow 200 for extracting features from one or more medical images according to one or more embodiments is illustrated to perform a medical imaging analysis task. It will be described together. Figure 1 and Figure 2 .

[0015] exist Figure 1At step 102, 1) one or more medical images, each in a different domain, and 2) a domain code defining the presence of different domains in a predefined set of domains. This predefined set of domains represents the domains to be processed by a machine learning-based encoder (utilized at step 108). In one embodiment, the domain code also defines the absence of one or more domains in the predefined set within the different domains. At step 102, a zero tensor for each of the one or more absent domains may be received.

[0016] Domain codes can be represented in any suitable form and are used to define the presence of the distinct domain in a predefined set of domains and / or the absence of one or more domains from the predefined set in the distinct domain. In one embodiment, a domain code is... vector, where This represents the number of domains in the predefined set of domains. Each position in the vector is associated with a corresponding domain in this predefined set of domains. The value at each position can be defined (e.g., by one-hot encoding), where a value of 1 defines the presence of the input medical image in the associated domain, and a value of 0 defines the absence of the input medical image in the associated domain. Other methods for encoding the presence and / or absence of medical images in this predefined set of domains are also anticipated.

[0017] In one example, such as Figure 2 As shown in workflow 200, the predefined set of domains is domains A, B, C, and D. One or more medical images are medical images 202 in domains A, B, and D, and the domain code is domain code 206. Each position of domain code 206 is associated with a corresponding domain in the predefined set of domains. Therefore, the first position of domain code 206 is associated with domain A, the second position of domain code 206 is associated with domain B, the third position of domain code 206 is associated with domain C, and the fourth position of domain code 206 is associated with domain D. As shown in workflow 200, a medical image in domain C is absent or missing in the input. Therefore, domain code 206 defines a value of 1 for the first, second, and fourth positions to define the presence of a medical image in domains A, B, and D, and defines a value of 0 for the third position to define the absence of a medical image in domain C.

[0018] As used herein, a domain of medical imaging refers to the modality of the medical image and the protocol used to acquire the medical image in that modality. One or more medical image modalities may include, for example, MRI (Magnetic Resonance Imaging), CT (Computed Tomography), US (Ultrasound), X-ray, single-photon emission computed tomography (SPECT), positron emission tomography (PET), or any other medical imaging modality or combination of medical imaging modalities. Protocols used to acquire medical images (such as, for example, T1-weighted, T2-weighted, proton density-weighted MRI images, contrast and non-contrast images, CT images captured using low kV and high kV, or low and high resolution medical images) may include, for example, acquisition sequences or techniques for acquiring medical images. Therefore, different domains can be entirely different medical imaging modalities or different image protocols within the same overall imaging modality. One or more medical images may be represented in an image space (e.g., as pixel or voxel values ​​in spatial coordinates) or a latent space (e.g., as a low-dimensional compressed representation of one or more medical images represented as feature vectors). One or more medical images in the image space can be 2D (two-dimensional) images and / or 3D (three-dimensional) volumes.

[0019] It can receive one or more medical images and / or domain codes, for example, by directly from the image acquisition device (e.g., when acquiring one or more medical images). Figure 7 The image acquisition device 714 receives one or more medical images from the storage device or memory of a computer system (e.g., ...). Figure 7 The computer 702 loads one or more medical images and / or domain codes from its storage device 712 or memory 710, or from a remote computer system (e.g., Figure 7 The computer (702) receives one or more medical images and / or domain codes. Such a computer system or remote computer system may include one or more patient databases, such as, for example, EHR (Electronic Health Record), EMR (Electronic Medical Record), PHR (Personal Health Record), HIS (Health Information System), RIS (Radiology Information System), PACS (Picture Archiving and Communication System), LIMS (Laboratory Information Management System), or any other suitable database or system.

[0020] exist Figure 1 At step 104, one or more weights are determined based on the domain code. These one or more weights are used to weight the machine learning-based encoder (in...). Figure 1 One or more parameters are used in step 108. One or more weights can be represented as a vector (or in any other suitable form). In one embodiment, one or more weights comprise a weight vector. and bias vector It is used to update the weight and bias parameters of the machine learning-based encoder.

[0021] One or more weights can be determined using any suitable method. In one embodiment, a learnable linear projector is used to determine one or more weights. For example, such as... Figure 2 As shown in workflow 200, weights 210 are determined by linear projector 208 based on domain code 206. Linear projector is a linear transformation learned during the training process. The linear projector projects or maps the domain code to one or more weights using a set of parameters optimized during training.

[0022] exist Figure 1 At step 106, one or more parameters of the machine learning-based encoder are updated based on one or more weights. In one embodiment, the machine learning-based encoder is a convolutional layer of a neural network (such as, for example, a DFN). However, the machine learning-based encoder can be any other suitable encoder.

[0023] In one embodiment, one or more parameters of the machine learning-based encoder include weight parameters. and bias parameters For example, such as Figure 2 As shown in workflow 200, the weight parameters of encoder 212 are updated based on weight 210. and bias parameters In one embodiment, one or more parameters are updated by calculating a dot product. For example, this can be done by calculating weight parameters. and weight vector The dot product (i.e., The weight parameters can be updated by calculating the bias parameters. and bias vector The dot product (i.e., The bias parameters are updated using this method. Therefore, the domain code controls the behavior of the machine learning-based encoder. One or more parameters can be updated using any other suitable method.

[0024] exist Figure 1 At step 108, a machine learning-based encoder with one or more updated parameters is used to extract features from one or more medical images. The one or more medical images are concatenated along the channel dimension (along with all-zero tensors representing medical images without domains). The machine learning-based encoder (with one or more updated parameters) receives the concatenation results as input and generates features as output. These features are low-dimensional compressed representations of the one or more medical images, represented as feature vectors. In one example, such as... Figure 2As shown in workflow 200, encoder 212 extracts feature map 214 from the concatenation results of medical image 202. Therefore, each domain combination of one or more medical images dynamically updates the machine learning-based encoder based on one or more specific missing domains defined by domain codes, thereby enabling the extraction of relevant features.

[0025] exist Figure 1 At step 110, a medical imaging analysis task is performed based on the extracted features. The medical imaging analysis task can be performed using a machine learning-based task network (such as, for example, a decoder network). The machine learning-based task network receives the extracted features as input and generates the result of the medical imaging analysis task as output. In one embodiment, the medical imaging analysis task is image synthesis for generating a synthetic image from one or more medical images. The synthetic image can reside in a predefined set of domains that are not present in the different domains of the (one or more medical images). The medical imaging analysis task may additionally or alternatively include any other suitable tasks, such as, for example, detection, segmentation, classification, quantization, etc.

[0026] exist Figure 1 At step 112, the results of the medical imaging analysis task are output. For example, the results of the medical imaging analysis task can be output in the following way: on the display device of the computer system (e.g., ...). Figure 9 The results are displayed on the I / O 908 of the computer 902, and stored in the computer system's memory or storage device (e.g., Figure 9 The results are stored on the memory 910 or storage device 912 of the computer 902, or by transmitting the results to a remote computer system (e.g., Figure 9 Computer 902).

[0027] In some embodiments, Figure 1 and Figure 2 A machine learning-based encoder can be implemented in a diffusion model to perform medical imaging analysis tasks conditioned on multi-domain medical images with missing domains, such as... Figure 3 and Figure 4 As shown in the figure.

[0028] Figure 3 A network architecture 300 of DFNControlNet conditioned on multi-domain medical images with missing domains, according to one or more embodiments, is illustrated. Medical images 302 in domains A, B, and D are received and concatenated with a zero tensor 304 of the absent or missing domain C. Domain codes 306 define the presence of one or more medical images 302 in domains A, B, and C, and the absence of a medical image in domain C. A linear projector (...) Figure 3(Not shown in the diagram), one or more weights are determined according to domain code 306 to update the parameters of zero DFN layers 308 and 310. Zero DFN layers 308 and 310 can be implemented as... Figure 1 Step 108: Machine learning-based encoder or Figure 2 The encoder 212. The zero-DFN layer 308 receives the concatenated medical image as input and generates a first feature set as output. The first feature set is mixed with noise. The inputs are combined into 314 and fed into a trainable copy 312, which is a trainable copy of neural network block 316. Neural network block 316 can be, for example, a ResNet block, a conv-bn-relu block, a multi-head attention block, a transformer block, etc. The output of trainable copy 312 is fed into a zero-DFN layer 310, which generates a second feature set as output. Neural network block 316 receives noise. 314 is used as input, and the output of neural network block 316 is combined with the second feature set to generate the result. 318. Advantageously, the network architecture 300 is conditioned on multi-domain medical images and dynamically adjusts its behavior to adapt to different missing domain scenarios.

[0029] Figure 4 A network architecture 400 of a multi-domain medical image conditioned on a denoised diffusion probability model / latent diffusion model (DFNDDPM / LDM) with missing domains, according to one or more embodiments, is shown. Medical images 402 in domains A, B, and D are received and concatenated with all-zero tensors 404 of the absent or missing domain C. Domain codes 406 define the presence of one or more medical images 402 in domains A, B, and C, and the absence of a medical image in domain C. The network architecture 400 includes a denoised UNet 408, which includes an encoder 410 and a decoder 412. The encoder 410 includes multiple DFN layers. Each DFN layer can be implemented as... Figure 1 Step 108: Machine learning-based encoder or Figure 2 Encoder 212. Using a linear projector ( Figure 4 (Not shown in the image), one or more weights are determined according to domain code 406 to update the parameters of the DFN layer of encoder 410. Linked medical images with noise The data is combined in group 414 and fed into encoder 410. The first DFN layer of encoder 410 receives the linked medical image and noise. The combination of features from encoder 414 is used as input, and the generated features are used as output. Each subsequent DFN layer receives the feature output of the previous DFN layer as input and generates features as output. Decoder 412 receives the features output from the last DFN layer of encoder 410 as input and generates the result. 418 is the output. In one embodiment, for example, when network architecture 400 is a DFN LDM, medical image 402 is first encoded by encoder 410 to extract features, and these features are concatenated along the channel dimension before being fed into the first DFN layer. Advantageously, network architecture 400 is implemented conditioned on multi-domain medical images.

[0030] In one embodiment, for a medical imaging analysis task that synthesizes medical images with pathology, in addition to multi-domain medical images, one or more masks of the pathology / injury and / or surrounding tissues or organs to be synthesized may also be received (at step 102). One or more masks are concatenated with one or more medical images along the channel dimension, and features are extracted from the concatenation results (at step 108). Method 100 then continues to perform the medical imaging analysis task based on the extracted features. If not all masks are always present, the domain codes may have corresponding positions defining the presence or absence of masks. In this way, according to the embodiments described herein, a multi-domain dataset with pathology can be constructed using an image synthesis framework.

[0031] The embodiments described herein are described with respect to the claimed system and the claimed method. Features, advantages, or alternative embodiments described herein may be assigned to other claimed objects, and vice versa. In other words, the claims and embodiments for the system may be improved using features described or claimed in the context of the corresponding method. In this case, the functional features of the method are implemented through the physical units of the system.

[0032] Furthermore, certain embodiments described herein are described with respect to methods and systems utilizing trained machine learning models, and with respect to methods and systems for providing trained machine learning models. Features, advantages, or alternative embodiments described herein may be assigned to other claimed objects, and vice versa. In other words, the claims and embodiments for providing trained machine learning models may be modified using features described or claimed in the context of utilizing trained machine learning models, and vice versa. Specifically, the dataset used in methods and systems for utilizing trained machine learning models may have the same properties and characteristics as the corresponding dataset used in methods and systems for providing trained machine learning models, and the trained machine learning model provided by the corresponding method and system may be used in methods and systems for utilizing trained machine learning models.

[0033] Generally, trained machine learning models mimic human cognitive functions associated with other human thought processes. Specifically, through training on training data, machine learning models can adapt to new situations and detect and infer patterns. Another term for a “trained machine learning model” is a “trained function.”

[0034] Generally, the parameters of a machine learning model can be adapted through training. Specifically, supervised training, semi-supervised training, unsupervised training, reinforcement learning, and / or active learning can be used. Furthermore, representation learning (an alternative term is "feature learning") can be used. Specifically, the parameters of a machine learning model can be iteratively adapted through several training steps. Specifically, within training, a certain cost function can be minimized. Specifically, within the training of a neural network, the backpropagation algorithm can be used.

[0035] In particular, machine learning models, such as Figure 1 The linear projector used in step 104 or the machine learning-based encoder used in step 108 Figure 2 Linear projector 208 or encoder 212, Figure 3 Zero-DFN layers 308 and 310, neural network block 316 or trainable copy 312, and / or Figure 4 The denoising UNet 408 can include, for example, neural networks, support vector machines, decision trees, and / or Bayesian methods, and / or can be based on machine learning models such as k-means clustering, Q-learning, genetic algorithms, and / or association rules. Specifically, the neural network can be, for example, a deep neural network, a convolutional neural network, or a convolutional deep neural network. Furthermore, the neural network can be, for example, an adversarial network, a deep adversarial network, and / or a generative adversarial network.

[0036] Figure 5An embodiment of an artificial neural network 500 that can be used to implement one or more machine learning models described herein is shown. Alternative terms for “artificial neural network” are “neural network,” “artificial neural network,” or “neural network.”

[0037] The artificial neural network 500 includes nodes 520, ..., 532 and edges 540, ..., 542, where each edge 540, ..., 542 is a directed connection from a first node 520, ..., 532 to a second node 520, ..., 532. Generally, the first nodes 520, ..., 532 and the second nodes 520, ..., 532 are different nodes 520, ..., 532; however, it is also possible that the first nodes 520, ..., 532 and the second nodes 520, ..., 532 are the same. For example, in... Figure 5 In the diagram, edge 540 is a directed connection from node 520 to node 525, and edge 542 is a directed connection from node 550 to node 532. Edges 540, ..., 542 from the first node 520, ..., 532 to the second node 520, ..., 532 are also represented as the "incoming edges" of the second node 520, ..., 532 and the "outgoing edges" of the first node 520, ..., 532.

[0038] In this embodiment, nodes 520, ..., 532 of the artificial neural network 500 can be arranged in layers 510, ..., 513, wherein these layers may include an inherent order introduced by edges 540, ..., 542 between nodes 520, ..., 532. Specifically, edges 540, ..., 542 may exist only between adjacent layers of nodes. In the illustrated embodiment, there exists an input layer 510 comprising only nodes 520, ..., 522 without incoming edges, an output layer 513 comprising only nodes 531, 532 without outgoing edges, and hidden layers 511, 512 between the input layer 510 and the output layer 513. Generally, the number of hidden layers 511, 512 can be arbitrarily chosen. The number of nodes 520, ..., 522 in the input layer 510 is typically related to the number of input values ​​of the neural network, and the number of nodes 531, 532 in the output layer 513 is typically related to the number of output values ​​of the neural network.

[0039] Specifically, (real) numbers can be assigned as values ​​to each node 520, ..., 532 of the neural network 500. Here, x (n) iThis represents the value of the i-th node 520, ..., 532 in the n-th layer 510, ..., 513. The values ​​of nodes 520, ..., 522 in the input layer 510 are equivalent to the input values ​​of the neural network 500, and the values ​​of nodes 531 and 532 in the output layer 513 are equivalent to the output values ​​of the neural network 500. Furthermore, each edge 540, ..., 542 may include a weight (which is a real number), specifically, a real number within the interval [-1, 1] or the interval [0, 1]. Here, w (m,n) i,j Let w represent the weight of the edge between the i-th node 520, ..., 532 of layer m (510, ..., 513) and the j-th node 520, ..., 532 of layer n (510, ..., 513). Furthermore, for the weight w... (n,n+1) i,j Define the abbreviation w (n) i,j .

[0040] Specifically, in order to calculate the output value of neural network 500, the input value is propagated through the neural network. Specifically, the values ​​of nodes 520, ..., 532 in the (n+1)th layer 510, ..., 513 can be calculated based on the values ​​of nodes 520, ..., 532 in the nth layer 510, ..., 513 using the following formula: .

[0041] In this paper, the function f is the transfer function (another term is "activation function"). Known transfer functions are step functions, sigmoid functions (e.g., logistic functions, generalized logistic functions, hyperbolic tangent functions, arctangent functions, error functions, smoothstep functions, or rectifier functions). Transfer functions are primarily used for normalization purposes.

[0042] Specifically, these values ​​are propagated layer by layer through a neural network, wherein the value of the input layer 510 is given by the input of the neural network 500, wherein the value of the first hidden layer 511 can be calculated based on the value of the input layer 510 of the neural network, wherein the value of the second hidden layer 512 can be calculated based on the value of the first hidden layer 511, and so on.

[0043] To set the value of the edge Neural networks 500 must be trained using training data. 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 the computed output data. Specifically, the training data and the computed output data include several values, the number of which is equal to the number of nodes in the output layer.

[0044] Specifically, the weights within 500 of the neural network are recursively adapted using a comparison between the calculated output data and the training data (backpropagation algorithm). Specifically, the weights are changed according to the following formula: Where γ is the learning rate, and in the case that the (n+1)th layer is not the output layer, based on Count The equation is calculated recursively as follows: And in the case that the (n+1)th layer is the output layer 513, the number The equation is calculated recursively as follows: in It is the first derivative of the activation function, and It is the comparison training value of the j-th node of the output layer 513.

[0045] A convolutional neural network (CNN) is a neural network that uses convolution operations instead of general matrix multiplication in at least one of its layers (so-called "convolutional layers"). Specifically, a convolutional layer performs a dot product of one or more convolutional kernels with the input data / image of the convolutional layer, where the entries for the one or more convolutional kernels are parameters or weights adapted through training. In particular, Frobenius inner products and ReLU activation functions can be used. A CNN may include additional layers such as pooling layers, fully connected layers, and normalization layers.

[0046] By using convolutional neural networks, input images can be processed very efficiently because convolutional operations based on different kernels can extract various image features. This allows relevant image features to be found during training by adapting the weights of the convolutional kernels. Furthermore, due to weight sharing within the convolutional kernels, fewer parameters need to be trained, preventing overfitting during training and allowing for faster training or more layers in the network, thus improving network performance.

[0047] Figure 6An embodiment of a convolutional neural network 600 that can be used to implement one or more machine learning models described herein is shown. In the shown embodiment, the convolutional neural network 600 includes an input node layer 610, a convolutional layer 611, a pooling layer 613, a fully connected layer 614, an output node layer 616, and hidden node layers 612, 614. Alternatively, the convolutional neural network 600 may include a plurality of convolutional layers 611, a plurality of pooling layers 613, and a plurality of fully connected layers 615, as well as other types of layers. The order of the layers can be arbitrarily chosen; typically, the fully connected layer 615 is used as the last layer before the output layer 616.

[0048] Specifically, within the convolutional neural network 600, nodes 620, 622, and 624 of node layers 610, 612, and 614 can be viewed as arranged as a d-dimensional matrix or a d-dimensional image. In particular, in the two-dimensional case, the values ​​of nodes 620, 622, and 624 indexed by i and j in the nth node layer 610, 612, and 614 can be represented as x(n)[i,j]. However, the arrangement of nodes 620, 622, and 624 in a node layer 610, 612, and 614 itself has no effect on the computations performed within the convolutional neural network 600, because these are given only by the weights and structure of the edges.

[0049] Convolutional layer 611 is a connection layer between the preceding node layer 610 (with node value x(n-1)) and the following node layer 612 (with node value x(n)). Specifically, convolutional layer 611 is characterized by the structure and weights of the input edges that form the convolution operation based on a certain number of kernels. In particular, the structure and weights of the edges of convolutional layer 611 are selected such that the value x(n) of node 622 in the following node layer 612 is computed as a convolution based on the value x(n-1) of node 620 in the preceding node layer 610. In the two-dimensional case, convolution* is defined as: .

[0050] Here, the kernel K is a d-dimensional matrix (in this embodiment, a two-dimensional matrix), which is typically small compared to the number of nodes 620, 622 (e.g., a 3×3 or 5×5 matrix). Specifically, this implies that the weights of the edges in convolutional layer 611 are not independent, but are chosen such that they produce the convolution equation. In particular, for a 3×3 kernel, there are only 9 independent weights (each entry in the kernel matrix corresponds to one independent weight), regardless of the number of nodes 620, 622 in the preceding node layer 610 and the following node layer 612.

[0051] Generally, convolutional neural networks 600 use node layers 610, 612, and 614 with multiple channels, especially due to the use of multiple kernels in convolutional layer 611. In those cases, the node layer can be considered as a (d+1)-dimensional matrix (the first dimension indexing the channels). Then, the action of convolutional layer 611 is defined as a two-dimensional example of the following: in Corresponding to the a-th channel of the preceding node layer 610, Corresponding to the b-th channel of the subsequent node layer 612, and This corresponds to one of the kernels. If convolutional layer 611 acts on the preceding node layer 610 with A channels and outputs the following node layer 612 with B channels, then there exist A·B independent d-dimensional kernels. .

[0052] Generally, in a convolutional neural network 600, an activation function is used. In this embodiment, ReLU (an abbreviation for "Rectified Linear Unit") is used, where... This makes the action of convolutional layer 611 in the two-dimensional example as follows: .

[0053] Other activation functions may also be used, such as ELU (an abbreviation for "Exponential Linear Unit"), LeakyReLU, Sigmoid, Tanh, or Softmax.

[0054] In the shown embodiment, the input layer 610 includes 36 nodes 620 arranged as a two-dimensional 6×6 matrix. The first hidden node layer 612 includes 72 nodes 622 arranged as two two-dimensional 6×6 matrices, each of which is the result of convolving the values ​​of the input layer with a 3×3 kernel within the convolutional layer 611. Equivalently, the nodes 622 of the first hidden node layer 612 can be interpreted as being arranged as a three-dimensional 2×6×6 matrix, where the first dimension corresponds to the channel dimension.

[0055] The advantage of using convolutional layers 611 is that it can take advantage of the spatial local correlation of the input data by enforcing a local connectivity pattern between nodes in neighboring layers, and in particular by connecting each node only to a small region of nodes in the previous layer.

[0056] Pooling layer 613 is the connecting layer between the preceding node layer 612 (with node value x(n-1)) and the following node layer 614 (with node value x(n)). Specifically, pooling layer 613 can be characterized by the activation function and edge structure and weights that form the pooling operation based on the nonlinear pooling function f. For example, in the two-dimensional case, the value x(n) of node 624 in the following node layer 614 can be calculated based on the value x(n-1) of node 622 in the preceding node layer 612 as follows: In other words, by using pooling layer 613, the number of nodes 622 and 624 can be reduced by replacing a number of adjacent nodes 622 in the preceding node layer 612 with a single node 622 from the subsequent node layer 614, the replacement being calculated based on the values ​​of the number of adjacent nodes. Specifically, the pooling function f can be a maximum value function, an average value function, or an L2 norm function. Specifically, for pooling layer 613, the weights of the incoming edges are fixed and are not modified through training.

[0057] The advantage of using pooling layer 613 is that it reduces the number of nodes 622 and 624 and the number of parameters. This results in a reduction in the computational cost of the network and better control over overfitting.

[0058] In the illustrated embodiment, pooling layer 613 is a max-pooling layer that replaces four adjacent nodes with only one node, the value of which is the maximum of the values ​​of the four adjacent nodes. 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.

[0059] Generally, the last layer of the convolutional neural network 600 is a fully connected layer 615. The fully connected layer 615 is the connection layer between the preceding node layer 614 and the following node layer 616. The fully connected layer 613 can be characterized by the fact that most, in particular all, edges exist between nodes 614 of the preceding node layer 614 and nodes 616 of the following node layer, and the weight of each of these edges can be adjusted individually.

[0060] In this embodiment, the nodes 624 of the preceding node layer 614 of the fully connected layer 615 are displayed as a two-dimensional matrix, and additionally as unrelated nodes (indicated as a row of nodes, where the number of nodes is reduced for better presentation). This operation is also referred to as "flattening". In this embodiment, the number of nodes 626 in the following node layer 616 of the fully connected layer 615 is less than the number of nodes 624 in the preceding node layer 614. Alternatively, the number of nodes 626 can be equal to or greater than the number of nodes 624.

[0061] Furthermore, in this embodiment, a Softmax activation function is used within the fully connected layer 615. By applying the Softmax function, the sum of the values ​​of all nodes 626 in the output layer 616 is 1, and all values ​​of all nodes 626 in the output layer 616 are real numbers between 0 and 1. In particular, if the input data is classified using a convolutional neural network 600, the values ​​of the output layer 616 can be interpreted as the probability that the input data falls into one of the different categories.

[0062] Specifically, a convolutional neural network 600 can be trained based on the backpropagation algorithm. To prevent overfitting, regularization methods can be used, such as dropout of nodes 620, ..., 624, random pooling, use of artificial data, and weight decay based on L1 or L2 norm or maximum norm constraints.

[0063] According to one aspect, a machine learning model may include one or more Residual Networks (ResNets). Specifically, a ResNet is an artificial neural network that includes at least one jump or skip connection for hopping at at least one layer of the artificial neural network. Specifically, a ResNet may be a convolutional neural network that includes one or more skip connections that skip one or more convolutional layers respectively. According to some examples, a ResNet may be represented as an m-layer ResNet, where m is the number of layers in the corresponding architecture, and according to some examples, it may take values ​​of 34, 50, 101, or 152. According to some examples, such an m-layer ResNet may correspondingly include (m-2) / 2 skip connections.

[0064] Skip connections can be viewed as bypassing a process where the output of a previous layer is directly fed into a layer following that layer via one or more bypassed layers. Instead of directly fitting the desired mapping, the bypassed layer must then fit a residual mapping that "balances" the output of the directly fed layer.

[0065] Fitting residual mappings is computationally easier to optimize than directional mappings. More importantly, this mitigates the vanishing / exploding gradient problem during optimization when training machine learning models: if a bypassed layer encounters this problem, its contribution can be skipped by regularizing the output of the direct feed. Therefore, the advantage of using ResNet is that much deeper networks can be trained.

[0066] Generative adversarial models (GA models) consist of a generator function and a discriminator function. The generator function creates synthetic data, and the discriminator function distinguishes between synthetic and real data. By training the generator and / or discriminator functions, the generator function is configured to create synthetic data that is incorrectly classified as real by the discriminator function, while the discriminator function is configured to distinguish between real data and synthetic data generated by the generator function. In game theory, generative adversarial models can be interpreted as zero-sum games. The training of the generator and / or discriminator functions is primarily based on minimizing a cost function.

[0067] By using a GA (Generative Approach) model, synthetic data with the same characteristics as the training dataset can be generated based on the training dataset. Training a GA model can be based on unannotated data (unsupervised learning), resulting in low workload in training the GA model.

[0068] Figure 7 A data flow graph is shown according to one or more embodiments of an embodiment for using a generative adversarial network to create synthetic output data G(x) 708 based on input data x 702 that is indistinguishable from real output data y 704. The synthetic output data G(x) 708 has the same structure as the real output data y 704, but its content is not derived from real-world data.

[0069] The Generative Adversarial Network (GAN) consists of a generator function G 706 and a classifier function C 710, which are trained together. The generator function G 706 is tasked with providing real synthetic output data G(x) 708 based on the input data x 702, and the classifier function C 710 is tasked with distinguishing between the real output data y 704 and the synthetic output data G(x) 708. Specifically, the output of the classifier function C 710 is a real number between 0 and 1, corresponding to the probability that the input value is real data, such that an ideal classifier function would compute an output value of C(y) 714 ≈ 1 for the real data y 704 and C(G(x)) 712 ≈ 0 for the synthetic data G(x) 708.

[0070] During training, the parameters of the generator function G 706 are adapted so that the synthetic output data G(x) 708 has the same characteristics as the real output data y 704, making the classifier function C 710 unable to distinguish between real and synthetic data. Simultaneously, the parameters of the classifier function C 710 are adapted so that it distinguishes between real and synthetic data in the best possible way. Here, training relies on pairs including input data x 702 and corresponding real output data y 704. Within a single training step, the generator function G 706 is applied to the input data x 702 to generate the synthetic output data G(x) 708. Furthermore, the classifier function C 710 is applied to the real output data y 704 to generate the first classification result C(y) 714. Additionally, the classifier function C 710 is applied to the synthetic output data G(x) 708 to generate the second classification result C(G(x)) 712.

[0071] The parameters of the adaptation generation function G 706 and the classifier function C 710 are based on minimizing the cost function using the backpropagation algorithm. In this embodiment, the cost function K of the classifier function C 710 is... C yes BCE is defined as The binary cross-entropy. By using this cost function, the true output data is incorrectly classified as synthetic output data (generated by...). Both incorrectly classifying synthetic output data as real output data (represented by C(G(x)) 712 ≈ 1) increase the cost function K to be minimized. C Furthermore, the cost function K of the generator function G706 G yes By using this cost function, the correctly classified synthetic output data (indicated by C(G(x)) 712 ≈ 0 results in a cost function K to be minimized. G The increase.

[0072] Specifically, a recursive machine learning model is a machine learning model whose output depends not only on the input values ​​and the parameters of the machine learning model adapted through the training process, but also on the hidden state vector, which is based on previous inputs used on the recursive machine learning model. In particular, the recursive machine learning model may include additional stored states or additional structures that incorporate time delays or include feedback loops.

[0073] Specifically, the basic structure of a recurrent machine learning model can be a neural network, which can be represented as a recurrent neural network. Such a recurrent neural network can be described as an artificial neural network where the connections between nodes form a directional graph along a time series. Specifically, a recurrent neural network can be interpreted as a directional acyclic graph. Specifically, the recurrent neural network can be a finite-spiking recurrent neural network or an infinite-spiking recurrent neural network (where a finite-spiking network can be unfolded and replaced with a strictly feedforward neural network, and an infinite-spiking network cannot be unfolded and replaced with a strictly feedforward neural network).

[0074] Specifically, recurrent neural networks can be trained based on the BPTT algorithm (an acronym for "backward propagation by time"), the RTRL algorithm (an acronym for "real-time recursive learning"), and / or genetic algorithms.

[0075] By using a recursive machine learning model, input data including sequences of variable length can be used. Specifically, this implies that the method cannot be used only with a fixed number of input datasets (and requires different training for all other numbers of input datasets used as input), but can be used with any number of input datasets. This means that, independent of the number of input datasets contained in different sequences, the entire training dataset can be used within training, and the training data is not reduced to training data corresponding to a fixed number of consecutive input datasets.

[0076] Figure 8 The schematic structure of a recursive machine learning model F is illustrated using both recursive representation 802 and unfolded representation 804. This recursive machine learning model F can be used to implement one or more machine learning models described in this paper. The recursive machine learning model takes several input datasets x, x1, ..., x2. N Take 806 as input and create corresponding output datasets y, y1, ..., y2. N A set of 808. Furthermore, the output depends on the so-called hidden vectors h1, h2, ..., h3. N 810, the hidden vectors implicitly include information about the input dataset previously used as input to the recursive machine learning model F 812. This is achieved by using these hidden vectors h, h1, ..., h... N 810 can take advantage of the order of the input dataset.

[0077] In a single processing step, the recursive machine learning model F812 takes the hidden vector h created in the previous step. n-1 and the input dataset x n As input. Within this step, the recursive machine learning model F generates the updated hidden vector h. n and the output dataset yn As output. In other words, a processing step calculates... Alternatively, by splitting the recursive machine learning model F812 into a part F(y) that computes the output data and a part F(h) that computes the hidden vectors, one processing step computes... and For the first processing step, h0 can be randomly selected or filled with all entries that are zero. The parameters of the recurrent machine learning model F 812, which was previously trained on the training dataset, remain unchanged between the different processing steps.

[0078] In particular, the output data and hidden vectors of the processing step depend on all the previous input datasets used in the previous steps. and .

[0079] The systems, apparatus, and methods described herein can be implemented using digital circuitry or using one or more computers employing 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.

[0080] The systems, apparatus, and methods described herein can be implemented using computers operating in a client-server relationship. Typically, in such a system, the client computer is located remotely from 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.

[0081] The systems, apparatuses, and methods described herein can be implemented within a network-based cloud computing system. In such a system, a server or another processor connected to the network communicates with one or more client computers via the network. Client computers can communicate with the server via, for example, a web browser application residing and operating on the client computer. Client computers can store data on the server and access that data via the network. Client computers can transmit requests for data or for online services to the server via the network. The server can perform the requested service and provide data to one or more client computers(s). The server can also transmit data suitable for causing the client computer to perform specified functions (e.g., perform calculations, display specified data on a screen, etc.). For example, the server can transmit requests suitable for causing the client computer to perform one or more steps or functions of the methods and workflows described herein, including... Figure 1-3 One or more of the steps or functions. Certain steps or functions of the methods and workflows described herein (including...) Figure 1-3 One or more of the steps or functions described herein 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 (including...) Figure 1-3 One or more of the steps in the methods and workflows described herein can be performed by a client computer in a web-based cloud computing system. Figure 1-3 One or more of the steps can be performed by a server and / or by a client computer in a web-based cloud computing system in any combination.

[0082] The systems, apparatuses, and methods described herein can be implemented using a computer program product tangibly embodied in an information carrier (e.g., embodied in a non-transitory machine-readable storage device) for execution by a programmable processor; and the methods and workflow steps described herein (including Figure 1-3 One or more of the steps or functions of a computer program can be implemented using one or more computer programs that can be executed 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 certain activity or produce a certain result. A computer program 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.

[0083] Figure 9 A high-level block diagram of an example computer 902, which can be used to implement the systems, apparatus, and methods described herein, is depicted. Computer 902 includes a processor 904 operatively coupled to a data storage device 912 and a memory 910. Processor 904 controls this operation by executing computer program instructions that define the overall operation of computer 902. The computer program instructions may be stored in the data storage device 912 or other computer-readable medium and loaded into memory 910 when execution of the computer program instructions is desired. Therefore, Figure 1-3 The methods and workflow steps or functions can be defined by computer program instructions stored in memory 910 and / or data storage device 912, and can be controlled by processor 904 that executes the computer program instructions. For example, the computer program instructions can be implemented to be programmed by those skilled in the art to perform... Figure 1-3 The methods, workflow steps, or functions are computer-executable code. Therefore, by executing computer program instructions, processor 904 performs... Figure 1-3The computer 902 may include methods and workflow steps or functions. The computer 902 may also include one or more network interfaces 906 for communicating with other devices via a network. The computer 902 may also include one or more input / output devices 908 (e.g., monitor, keyboard, mouse, speaker, buttons, etc.) that enable a user to interact with the computer 902.

[0084] Processor 904 may include both general-purpose microprocessors and special-purpose microprocessors, and may be the sole processor of computer 902 or one of multiple processors. For example, processor 904 may include one or more central processing units (CPUs). Processor 904, data storage device 912 and / or memory 910 may include one or more application-specific integrated circuits (ASICs) and / or one or more field-programmable gate arrays (FPGAs), supplementing or incorporating the foregoing.

[0085] Data storage device 912 and memory 910 each include a tangible, non-transitory computer-readable storage medium. Data storage device 912 and memory 910 may each 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 memory devices, and may include non-volatile memory, such as one or more disk storage devices (e.g., internal hard disks and removable disks), magneto-optical disk storage devices, optical disk storage devices, flash memory devices, semiconductor memory devices (e.g., erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), digital versatile disc read-only memory (DVD-ROM) discs), or other non-volatile solid-state memory devices.

[0086] Input / output device 908 may include peripheral devices such as printers, scanners, displays, etc. For example, input / output device 908 may include display devices (such as cathode ray tube (CRT) or liquid crystal display (LCD) monitors) for displaying information to a user, a keyboard, and pointing devices such as mice or trackballs through which the user can provide input to computer 902.

[0087] Image acquisition device 914 can be connected to computer 902 to input image data (e.g., medical images) into computer 902. It is possible to implement image acquisition device 914 and computer 902 as a single device. It is also possible for image acquisition device 914 and computer 902 to communicate wirelessly via a network. In a possible embodiment, computer 902 can be remotely located relative to image acquisition device 914.

[0088] Any or all of the systems, apparatuses, and methods discussed herein may be implemented using one or more computers (such as computer 902).

[0089] Those skilled in the art will recognize that the actual implementation of a computer or computer system may have other structures and may include other components, and for illustrative purposes, Figure 9 It is a high-level representation of some components in such a computer.

[0090] Individuals with male or female gender identity are included in the term, independent of the use of grammatical terms.

[0091] The foregoing detailed descriptions should be understood in every respect as illustrative and exemplary, not restrictive, and the scope of the invention disclosed herein is not determined by these detailed descriptions, but rather by the claims as interpreted under the full breadth permitted by patent law. It is to be understood that the embodiments shown and described herein are merely illustrative descriptions 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.

[0092] The following is a list of non-limiting illustrative embodiments disclosed herein: Illustrative Example 1. A computer-implemented method comprising: receiving 1) one or more medical images, each in a different domain, and 2) domain codes defining the presence of the different domains in a predefined set of domains; determining one or more weights based on the domain codes; updating one or more parameters of a machine learning-based encoder based on the one or more weights; extracting features from the one or more medical images using the machine learning-based encoder having the one or more updated parameters; performing a medical imaging analysis task based on the extracted features; and outputting the results of the medical imaging analysis task.

[0093] Illustrative Example 2. The computer-implemented method according to Illustrative Example 1, wherein the domain code further defines the non-existence of one or more domains of a predefined set in different domains.

[0094] Illustrative Example 3. A computer-implemented method according to any one of Illustrative Examples 1-2, wherein each position of a field code is associated with a corresponding field in a predefined set of fields.

[0095] Illustrative Example 4. A computer-implemented method according to any one of Illustrative Examples 1-3, wherein determining one or more weights based on domain codes comprises: projecting the domain codes onto one or more weights using a linear projector.

[0096] Illustrative Example 5. A computer-implemented method according to any one of Illustrative Examples 1-4, wherein updating one or more parameters of a machine learning-based encoder based on one or more weights includes: updating the weight parameters and bias parameters of the machine learning-based encoder.

[0097] Illustrative Example 6. A computer-implemented method according to any one of Illustrative Examples 1-5, wherein updating one or more parameters of a machine learning-based encoder based on one or more weights comprises: determining the dot product of one or more parameters of the machine learning-based encoder with a corresponding weight among one or more weights.

[0098] Illustrative Example 7. A computer-implemented method according to any one of Illustrative Examples 1-6, wherein: receiving 1) one or more medical images, each in a different domain, and 2) a domain code defining the presence of the different domains in a predefined set of domains, includes: receiving one or more all-zero tensors of one or more domains in the predefined set of domains that are not present in the different domains; and extracting features from the one or more medical images using a machine learning-based encoder having one or more updated parameters includes: concatenating the one or more medical images with one or more all-zero tensors, and extracting features from the concatenation result.

[0099] Illustrative Example 8. A computer-implemented method according to any one of Illustrative Examples 1-7, wherein: receiving 1) one or more medical images, each of the one or more medical images being in a different domain, and 2) domain codes, the domain codes defining the presence of different domains in a predefined set of domains, includes: receiving one or more masks of at least one pathology or organ; and extracting features from the one or more medical images using a machine learning-based encoder having one or more updated parameters, including: concatenating the one or more medical images with one or more masks, and extracting features from the concatenation result.

[0100] Illustrative Example 9. A computer-implemented method according to any one of Illustrative Examples 1-8, wherein the medical imaging analysis task includes medical image synthesis.

[0101] Illustrative Example 10. An apparatus comprising: a construct for receiving: 1) one or more medical images, each in a different domain; and 2) a domain code defining the presence of the different domains in a predefined set of domains; a component for determining one or more weights based on the domain code; a component for updating one or more parameters of a machine learning-based encoder based on the one or more weights; a component for extracting features from the one or more medical images using the machine learning-based encoder having one or more updated parameters; a component for performing a medical imaging analysis task based on the extracted features; and a component for outputting the results of the medical imaging analysis task.

[0102] Illustrative Example 11. The apparatus according to Illustrative Example 10, wherein the domain code further defines the absence of one or more domains of the predefined domain set in the different domains.

[0103] Illustrative Example 12. The apparatus according to any one of Illustrative Examples 10-11, wherein each position of the field code is associated with a corresponding field in a predefined set of fields.

[0104] Illustrative Example 13. The apparatus according to any one of Illustrative Examples 10-12, wherein the component for determining one or more weights based on a domain code includes: a component for projecting the domain code onto one or more weights using a linear projector.

[0105] Illustrative Example 14. The apparatus according to any one of Illustrative Examples 10-13, wherein the component for updating one or more parameters of a machine learning-based encoder based on one or more weights includes: a component for updating weight parameters and bias parameters of the machine learning-based encoder.

[0106] Illustrative Example 15. A non-transitory computer-readable storage medium includes instructions that, when executed by a computer, cause the computer to perform operations including: receiving 1) one or more medical images, each in a different domain, and 2) domain codes defining the presence of the different domains in a predefined set of domains; determining one or more weights based on the domain codes; updating one or more parameters of a machine learning-based encoder based on the one or more weights; extracting features from the one or more medical images using the machine learning-based encoder having the one or more updated parameters; performing a medical imaging analysis task based on the extracted features; and outputting the results of the medical imaging analysis task.

[0107] Illustrative Example 16. A non-transitory computer-readable storage medium according to Illustrative Example 15, wherein the field code further defines the absence of one or more fields of the predefined field set in the different fields.

[0108] Illustrative Example 17. A non-transitory computer-readable storage medium according to any one of Illustrative Examples 15-16, wherein updating one or more parameters of a machine learning-based encoder based on one or more weights comprises: determining the dot product of one or more parameters of the machine learning-based encoder with a corresponding weight among one or more weights.

[0109] Illustrative Example 18. A non-transitory computer-readable storage medium according to any one of Illustrative Examples 15-17, wherein: receiving 1) one or more medical images, each in a different domain, and 2) a domain code defining the presence of the different domains in a predefined set of domains, includes: receiving one or more all-zero tensors of one or more domains of a predefined set of domains not present in the different domains; and extracting features from the one or more medical images using a machine learning-based encoder having one or more updated parameters includes: concatenating the one or more medical images with one or more all-zero tensors, and extracting features from the concatenation result.

[0110] Illustrative Example 19. A non-transitory computer-readable storage medium according to any one of Illustrative Examples 15-18, wherein: receiving 1) one or more medical images, each in a different domain, and 2) domain codes defining the presence of different domains in a predefined set of domains, includes: receiving one or more masks of at least one pathology or organ; and extracting features from the one or more medical images using a machine learning-based encoder having one or more updated parameters, including: concatenating the one or more medical images with one or more masks, and extracting features from the concatenation result.

[0111] Illustrative Example 20. A non-transitory computer-readable storage medium according to any one of Illustrative Examples 15-19, wherein the medical imaging analysis task includes medical image synthesis.

Claims

1. A computer-implemented method, comprising: Receive 1) one or more medical images, each in a different domain, and 2) domain codes, which define the presence of the different domains in a predefined set of domains; One or more weights are determined based on the domain code; Based on the one or more weights, update one or more parameters of the machine learning-based encoder; The machine learning-based encoder with one or more updated parameters is used to extract features from one or more medical images. Perform medical imaging analysis tasks based on the extracted features; as well as Output the results of the medical imaging analysis task.

2. The computer-implemented method according to claim 1, wherein, The domain code also defines the absence of one or more domains in the predefined domain set in the different domains.

3. The computer-implemented method according to claim 1, wherein, Each position of the domain code is associated with a corresponding domain in the predefined set of domains.

4. The computer-implemented method according to claim 1, wherein, Determining one or more weights based on the domain code includes: Using a linear projector, the domain code is projected onto the one or more weights.

5. The computer-implemented method according to claim 1, wherein, Updating one or more parameters of the machine learning-based encoder based on the one or more weights includes: Update the weight and bias parameters of the machine learning-based encoder.

6. The computer-implemented method according to claim 1, wherein, Updating one or more parameters of the machine learning-based encoder based on the one or more weights includes: Determine the dot product of one or more parameters of the machine learning-based encoder with a corresponding weight among the one or more weights.

7. The computer-implemented method according to claim 1, wherein: Receive 1) one or more medical images, each in a different domain, and 2) domain codes, the domain codes defining the presence of the different domains in a predefined set of domains, including: Receive one or more all-zero tensors from one or more of the predefined set of domains that do not exist in the different domains; and Extracting features from the one or more medical images using the machine learning-based encoder with the one or more updated parameters includes: Connect the one or more medical images to the one or more all-zero tensors, and Extract features from the link results.

8. The computer-implemented method according to claim 1, wherein: Receive 1) one or more medical images, each in a different domain, and 2) domain codes, the domain codes defining the presence of the different domains in a predefined set of domains, including: One or more masks that receive at least one pathological or organ condition; and Extracting features from the one or more medical images using the machine learning-based encoder with the one or more updated parameters includes: Link the one or more medical images with the one or more masks, and Extract features from the link results.

9. The computer-implemented method according to claim 1, wherein, The medical imaging analysis task includes medical image synthesis.

10. An apparatus comprising: Used to receive 1) one or more medical images, each in a different domain, and 2) domain codes, the domain codes defining the components of the presence of the different domains in a predefined set of domains; A component used to determine one or more weights based on the domain code; A component for updating one or more parameters of a machine learning-based encoder based on the one or more weights; A component for extracting features from the one or more medical images using the machine learning-based encoder with one or more updated parameters; Components for performing medical imaging analysis tasks based on the extracted features; as well as A component used to output the results of the medical imaging analysis task.

11. The apparatus according to claim 10, wherein, The domain code also defines the absence of one or more domains in the predefined domain set in the different domains.

12. The apparatus according to claim 10, wherein, Each position of the domain code is associated with a corresponding domain in the predefined set of domains.

13. The apparatus of claim 10, wherein the component for determining one or more weights based on the domain code comprises: A component for projecting the domain code onto the one or more weights using a linear projector.

14. The apparatus of claim 10, wherein the component for updating one or more parameters of the machine learning-based encoder based on the one or more weights comprises: A component for updating the weight and bias parameters of the machine learning-based encoder.

15. A non-transitory computer-readable storage medium comprising instructions that, when executed by a computer, cause the computer to perform operations, the operations including: Receive 1) one or more medical images, each in a different domain, and 2) domain codes, which define the presence of the different domains in a predefined set of domains; One or more weights are determined based on the domain code; Update one or more parameters of the machine learning-based encoder based on the one or more weights; Features are extracted from the one or more medical images using the machine learning-based encoder with one or more updated parameters; Perform medical imaging analysis tasks based on the extracted features; as well as Output the results of the medical imaging analysis task.

16. The non-transitory computer-readable storage medium according to claim 15, wherein, The domain code also defines the absence of one or more domains in the predefined domain set in the different domains.

17. The non-transitory computer-readable storage medium of claim 15, wherein updating one or more parameters of the machine learning-based encoder based on the one or more weights comprises: Determine the dot product of one or more parameters of the machine learning-based encoder with a corresponding weight among the one or more weights.

18. The non-transitory computer-readable storage medium according to claim 15, wherein: Receive 1) one or more medical images, each in a different domain, and 2) domain codes, the domain codes defining the presence of the different domains in a predefined set of domains, including: Receive one or more all-zero tensors from one or more of the predefined set of domains that do not exist in the different domains; and Extracting features from the one or more medical images using the machine learning-based encoder with the one or more updated parameters includes: Connect the one or more medical images to the one or more all-zero tensors, and Extract features from the link results.

19. The non-transitory computer-readable storage medium according to claim 15, wherein: Receive 1) one or more medical images, each in a different domain, and 2) domain codes, the domain codes defining the presence of the different domains in a predefined set of domains, including: One or more masks that receive at least one pathological or organ condition; and Extracting features from the one or more medical images using the machine learning-based encoder with the one or more updated parameters includes: Link the one or more medical images with the one or more masks, and Extract features from the link results.

20. The non-transitory computer-readable storage medium according to claim 15, wherein, The medical imaging analysis task includes medical image synthesis.

Images