WO2021246810A1 - 오토인코더 및 다중 인스턴스 학습을 통한 뉴럴 네트워크 학습 방법 및 이를 수행하는 컴퓨팅 시스템 - Google Patents
오토인코더 및 다중 인스턴스 학습을 통한 뉴럴 네트워크 학습 방법 및 이를 수행하는 컴퓨팅 시스템 Download PDFInfo
- Publication number
- WO2021246810A1 WO2021246810A1 PCT/KR2021/006966 KR2021006966W WO2021246810A1 WO 2021246810 A1 WO2021246810 A1 WO 2021246810A1 KR 2021006966 W KR2021006966 W KR 2021006966W WO 2021246810 A1 WO2021246810 A1 WO 2021246810A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- state
- learning
- data
- patch
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0475—Generative networks
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
Definitions
- the present invention relates to a method for learning a neural network and a computing system for performing the same, and more particularly, to a method for learning a neural network that can improve the performance of a neural network even with a small number of training data using an autoencoder and a multi-instance learning technique and a computing system for performing the same.
- a type of machine learning e.g., a deep learning method using a convolutional neural network (CNN)
- CNN convolutional neural network
- FIGS. 1A and 1B multiple instance learning, which is one of the background technologies of the present invention, will be described with reference to FIGS. 1A and 1B .
- a bag which is a set of instances, is regarded as a learning unit. So, in individual instance training, you label the instances, whereas in multi-instance training you label the bag, not the instance.
- Multi-instance learning is similar to individual-instance learning except in terms of units of learning, but with the following limitations: In performing binary classification, it is assumed that if the bag is positive, at least one of the instances present in the bag is positive, and if the bag is negative, all instances in the bag are negative.
- multi-instance learning can be applied to, for example, a field of diagnosing a lesion from a pathological Whole Slide Image using a neural network.
- a neural network for diagnosing a lesion not a hole-slide-image, but an image patch divided into a certain size is used as training data. Because it is given to -slide-image.
- 1A is a diagram illustrating an example of learning data used for multi-instance learning.
- 1 shows M bags B 1 to B M , each containing N data instances.
- FIG. 1B is a diagram illustrating a pseudo code illustrating an example of a process of learning a neural network (NN) through a multi-instance learning technique.
- FIG. 1B shows a process of learning learning data by one epoch, and in the actual learning process, learning may be performed by a plurality of epochs. In FIG. 1B, it is assumed that learning proceeds with the learning data shown in FIG. 1A.
- multi-instance learning it is first performed in the process S10 of extracting training data instances (T 1 to T M ) from each bag (B 1 to B M ), and then to the extracted training data instances.
- a process (S20) of training the neural network (NN) is performed.
- the probability that the corresponding instance is positive is calculated (S11, S12).
- the neural network before training is completed is used in the process of extracting the training data instance, so if multiple data instances are extracted from one bag, the possibility of incorrect instances being extracted increases. will occur
- the technical task to be achieved by the present invention is to provide a method and system capable of properly learning a neural network even with a relatively small number of data by extracting a plurality of data instances from one data bag for training.
- an autoencoder for determining whether an input data instance is in a first state or a second state, and a possibility that the input data instance is in the first state or the second state can be output
- a neural network learning method performed in a computing system including a neural network, for each of a plurality of data bags labeled in either a first state or a second state, among data instances included in the data bag
- a neural network learning method including determining that it is an instance.
- the neural network learning system determines whether the data instance input to the autoencoder is in a first state based on a difference between the data instance input to the autoencoder and the output data output from the autoencoder It can be judged whether there are two states.
- the autoencoder is trained in advance only with data instances in the first state, and the possibilities for each data instance included in the data bag, and at least some of each data instance included in the data bag.
- the step of judging a part of each data instance included in the data bag as a learning instance based on the determination result of the autoencoder may include, when the data bag is labeled as the first state, from the data instance most likely to be in the second state.
- the autoencoder inputting the data into the autoencoder in the order in which the possibility of the second state is lowered, and determining that the upper part of the data instance determined by the autoencoder to be the first state is the instance for learning corresponding to the data bag; and when the data bag is labeled as the second state, it is input to the autoencoder in the order from the data instance most likely to be in the second state to the less likely to be in the second state, and determined to be in the second state by the autoencoder It may include determining that the data instance of the upper part of the data bag is a learning instance corresponding to the data bag.
- the method for learning a neural network may further include repeatedly performing the extraction step and the learning step for one or more epochs.
- each of the plurality of data bags is a whole-image
- a data instance included in each of the plurality of data bags includes a predetermined whole-image corresponding to the data bag. It may be characterized as each image patch divided by the size of .
- an autoencoder for determining whether an input patch is in a first state or a second state, wherein the patch is one of images obtained by dividing an image into a predetermined size; and a neural network capable of outputting a possibility that the input patch is in a first state or a second state.
- a method for learning a neural network performed in a computing system wherein a plurality of pieces labeled with either the first state or the second state are an extraction step of extracting a training patch which is a part of the patches constituting the training image with respect to each training image; and a learning step of learning the neural network based on a training patch corresponding to each of the plurality of training images, wherein the extraction step includes inputting each patch constituting the training image into the neural network being trained, calculating a probability for each patch constituting the learning image; And on the basis of the possibility of each patch constituting the learning image, and the determination result of the autoencoder for at least a part of each patch constituting the learning image, it is determined that a part of each patch constituting the learning image is a patch for learning There is provided a neural network learning method comprising the step of:
- the autoencoder is trained in advance with only the patch in the first state, the possibility of each patch constituting the learning image, and the autoencoder for at least a portion of each patch constituting the learning image
- the step of judging a part of each patch constituting the training image as a training patch based on the determination result may include, when the training image is labeled in the first state, from the patch most likely to be in the second state to the second state.
- the autoencoder inputting to the autoencoder in the order of decreasing probability, and determining that the upper partial patch determined to be in the first state by the autoencoder is a learning patch corresponding to the learning image; and when the learning image is labeled as the second state, it is input to the autoencoder in the order from the patch most likely to be in the second state to the less likely patch to be in the second state, and determined to be in the second state by the autoencoder It may include the step of determining that the patch of the upper part of the training patch corresponding to the training image.
- each of the learning images is either an image including a lesion due to a predetermined disease or an image not including the lesion
- the first state is a normal state in which the lesion does not exist
- the The second state may be an abnormal state in which the lesion exists.
- a computer program installed in a data processing apparatus and recorded in a medium for performing the above-described method.
- a computer-readable recording medium in which a computer program for performing the above-described method is recorded.
- a computing system comprising a processor and a memory, wherein the memory, when executed by the processor, causes the computing system to perform the above-described method.
- an autoencoder for determining whether an input data instance is in a first state or a second state and a possibility that the input data instance is in the first state or the second state.
- a storage module for storing the neural network; an extraction module for extracting, for each of a plurality of data bags labeled in either the first state or the second state, an instance for learning that is a part of the data instances included in the data bag; and a learning module for learning the neural network based on a learning instance corresponding to each of the plurality of data bags, wherein the extraction module inputs each data instance included in the data bag into the neural network being trained.
- a neural network learning system that determines a part of each data instance included in the data bag as an instance for learning based on the data bag.
- the extraction module is configured to determine whether the data instance input to the auto-encoder is a first state or a second state based on a difference between the data instance input to the auto-encoder and the output data output from the auto-encoder cognition can be judged.
- the autoencoder is trained in advance with only data instances in the first state
- the extraction module includes at least one of a possibility for each data instance included in the data bag, and each data instance included in the data bag.
- the second state is most likely Input to the autoencoder in the order of decreasing the probability of being in the second state from the data instance, and determining that the upper part of the data instance determined by the autoencoder to be the first state is the learning instance corresponding to the data bag, and the When the data bag is labeled as the second state, it is input to the autoencoder in the order from the data instance most likely to be in the second state to the less likely to be in the second state. It may be determined that some data instances are training instances corresponding to the data bag.
- an autoencoder for determining whether an input patch is in a first state or a second state -
- the patch is one of dividing an image into a predetermined size, and the autoencoder is pre-learned only with patches that are a plurality of first states
- a storage module for storing a neural network capable of outputting a possibility that the input patch is in a first state or a second state
- an extraction module for extracting a training patch that is a part of patches constituting the training image for each of the plurality of training images labeled in either the first state or the second state
- a learning module for learning the neural network based on a learning patch corresponding to each of the plurality of learning images, wherein the extraction module inputs each patch constituting the learning image to the neural network being trained, Calculating the probability of each patch constituting the training image, and based on the determination result of the autoencoder for at least a portion of the probability of each patch constituting the training image, and each patch constituting the
- the autoencoder is trained in advance with only the patch in the first state, and the extraction module includes at least some of the possibilities for each patch constituting the learning image, and each patch constituting the learning image.
- the extraction module includes at least some of the possibilities for each patch constituting the learning image, and each patch constituting the learning image.
- the patches of the upper part determined by the autoencoder to be in the 2nd state are input to the autoencoder in the order from the patch most likely to be in the 2nd state to the less likely to be in the 2nd state. It may be determined that the learning patch corresponds to the learning image.
- instances for learning multiple data instances can be extracted from one data bag.
- the method according to the technical idea of the present invention According to , by filtering the training instances using the pre-trained autoencoder, it is possible to significantly reduce the number of wrong training data instances being extracted. Therefore, it is possible to effectively train the neural network with less data It has the effect of being able to
- 1A is a diagram illustrating an example of learning data used for multi-instance learning.
- 1B is a diagram illustrating a pseudo code illustrating an example of a process of learning a neural network through a multi-instance learning technique.
- FIG. 2 is a diagram illustrating a schematic configuration of a computing system for performing a neural network learning method according to the technical idea of the present invention.
- FIG. 3 is a diagram schematically illustrating the structure of an autoencoder used in a neural network learning method according to the technical idea of the present invention.
- FIG. 4 is a diagram illustrating an example of a method of extracting a data instance for learning by a neural network learning system according to the technical idea of the present invention.
- step S120 of FIG. 4 is a diagram illustrating an example of a specific process of step S120 of FIG. 4 .
- FIG. 6 is a diagram illustrating an example of a specific process of step S121 of FIG. 5 .
- FIG. 7 is a diagram illustrating a schematic configuration of a decision system using a neural network according to an embodiment of the present invention.
- FIG. 8 shows an example in which the judgment system using a neural network according to an embodiment of the present invention outputs the lesion region in the pathological slide image in the form of a grid map.
- the component when any one component 'transmits' data to another component, the component may directly transmit the data to the other component or through at least one other component. This means that the data may be transmitted to the other component. Conversely, when one component 'directly transmits' data to another component, it means that the data is transmitted from the component to the other component without passing through the other component.
- FIG. 2 is a diagram illustrating a schematic configuration of a computing system for performing a neural network learning method according to the technical idea of the present invention.
- the neural network learning method according to the technical idea of the present invention may be performed by the neural network learning system 100 .
- the neural network learning system 100 may be a computing system that is a data processing device having computational capability for implementing the technical idea of the present invention, and in general, a personal computer as well as a server, which is a data processing device that a client can access through a network. or a computing device such as a portable terminal.
- the neural network learning system 100 may be implemented as any one physical device, but if necessary, a plurality of physical devices may be organically combined to implement the neural network learning system 100 according to the technical idea of the present invention. An average expert in the technical field of the present invention can easily infer.
- the neural network learning system 100 may include a storage module 110 , an extraction module 120 , and a learning module 130 .
- some of the above-described components may not necessarily correspond to the components essential for the implementation of the present invention, and according to the embodiment, the neural network learning system 100 Of course, it may include more components than this.
- the system 100 may include functions and/or resources of other components of the neural network learning system 100 (eg, the storage module 110, the extraction module 120, the learning module 130, etc.) It may further include a control module (not shown) for controlling the.
- the neural network learning system 100 may further include a database (DB) 140 for storing various types of information and/or data required to implement the technical idea of the present invention.
- DB database
- the neural network learning system 100 may mean a logical configuration having hardware resources and/or software necessary to implement the technical idea of the present invention, and necessarily means one physical component or It is not meant to be a single device. That is, the system 100 may mean a logical combination of hardware and/or software provided to implement the technical idea of the present invention, and if necessary, installed in devices spaced apart from each other to perform each function. It may be implemented as a set of logical configurations for implementing the technical idea of the present invention. In addition, the system 100 may mean a set of components separately implemented for each function or role for implementing the technical idea of the present invention. For example, each of the storage module 110 , the extraction module 120 , and the learning module 130 may be located in different physical devices or may be located in the same physical device.
- each of the storage module 110 , the extraction module 120 , and the learning module 130 is also located in different physical devices and located in different physical devices. Components may be organically combined with each other to implement each of the above modules.
- a module may mean a functional and structural combination of hardware for carrying out the technical idea of the present invention and software for driving the hardware.
- the module may mean a logical unit of a predetermined code and a hardware resource for executing the predetermined code, which necessarily means physically connected code or means one type of hardware. It can be easily deduced to an average expert in the art of the present invention.
- the storage module 110 may store the neural network 111 and the autoencoder 112 .
- a neural network includes a multilayer perceptron model, and may refer to a set of information representing a series of design items defining an artificial neural network.
- the neural network 112 may be a convolutional neural network.
- a convolutional neural network may include an input layer, a plurality of hidden layers, and an output layer.
- Each of the plurality of hidden layers may include a convolution layer and a pooling layer (or sub-sampling layer).
- the convolutional neural network may be defined by a function, filter, stride, weight factor, etc. for defining each of these layers.
- the output layer may be defined as a fully connected FeedForward layer.
- each layer constituting the convolutional neural network The design details for each layer constituting the convolutional neural network are widely known. For example, well-known functions may be used for each of the number of layers to be included in a plurality of layers, a convolution function for defining the plurality of layers, a pooling function, and an activation function, and implement the technical spirit of the present invention Separately defined functions may be used to do this.
- An example of the convolution function is a discrete convolution sum and the like.
- max pooling, average pooling, etc. may be used.
- An example of the activation function may be a sigmoid, a tangent hyperbolic (tanh), a rectified linear unit (ReLU), or the like.
- the convolutional neural network in which design matters are defined may be stored in a storage device. And when the convolutional neural network is learned, a weight factor corresponding to each layer may be specified.
- learning of the convolutional neural network may refer to a process in which weight factors of respective layers are determined. And when the convolutional neural network is learned, the learned convolutional neural network may receive input data to an input layer and output output data through a predefined output layer.
- a neural network according to an embodiment of the present invention may be defined by selecting one or a plurality of well-known design items as described above, or an independent design item may be defined for the neural network.
- the neural network 111 may be a classification neural network that can be used for classification of input data.
- the neural network 111 may be a neural network used for binary classification of the input data by outputting a possibility of whether the input data is in a predetermined first or second state.
- the neural network 111 may be a neural network for receiving a biological image and determining the possibility that a lesion caused by a predetermined disease (eg, cancer) exists in the corresponding image.
- a predetermined disease eg, cancer
- the neural network 111 may output a possibility that a value input to the neural network 111 is a predetermined first state or a predetermined second state.
- the first state may be either positive or negative, and the second state may be the other one of positive or negative.
- the first state may be a negative state in which no lesion is present, and the second state may be a positive state in which a lesion is present.
- the probability output by the neural network 111 is a loss function (eg, mean squared error (MSE), cross entropy error (CEE)) or two vectors in the neural network 111 . It may be a value calculated by a distance (eg, a function representing a Euclidean distance, an n-norm distance, a Manhattan distance, etc.).
- MSE mean squared error
- CEE cross entropy error
- the autoencoder 112 is a neural network structure mainly used in an unsupervised learning methodology.
- the autoencoder 112 is an unsupervised machine learning model in the form of reducing the dimension of an input value and then restoring it again, and has a function of learning characteristics of values used for learning.
- the autoencoder 112 learns a function to approximate the output value to the input value, extracts features for the input value through the encoder, and reconstructs the input value through the decoder.
- the autoencoder 112 may include an encoder part 112-1 including a convolutional layer and a decoder part 112-2 including a deconvolutional layer.
- the original data (x) is input to the encoder 111
- the autoencoder is also a type of neural network, learning by a plurality of training data is preceded. In the learning stage of the autoencoder, the following processes 1) to 3) are performed for each training data d.
- the training data d is input to the autoencoder 112, and through encoding and decoding processes, restored data d' corresponding to the training data d is generated.
- the autoencoder 112 may be used to determine whether the input value is in the first state or the second state.
- the autoencoder 112 may be pre-learned only with the learning data in the first state, and the autoencoder 112 is restored by inputting a predetermined prediction target value into the learned autoencoder 112 .
- the prediction target value is in the second state.
- the autoencoder 112 may be pre-learned only with data in the second state, and the autoencoder 112 is restored by inputting a predetermined prediction target value into the learned autoencoder 112.
- the prediction target value is in the first state.
- the autoencoder 112 may include a Variational AutoEncoder (VAE).
- VAE Variational AutoEncoder
- the DB 140 may store training data to be used for training the neural network 111 .
- the training data may be data for multi-instance training as described with reference to FIG. 1A. That is, each of the training data stored in the DB 140 may be a data bag including a plurality of data instances.
- each of the plurality of training data may be a hole-image, and a data instance constituting each training data may be each image patch obtained by dividing the corresponding hole-image into predetermined sizes.
- each of the training data may be an intact pathological slide image.
- the data bag becomes one complete pathology slide image, and the data instance included in the data bag may be an individual patch obtained by dividing the pathological slide image into predetermined sizes.
- the learning data stored in the DB 140 may be labeled as a first state or a second state, respectively.
- each of the learning data may be labeled with a diagnosis result (eg, presence or absence of a lesion) for the pathological slide image.
- the learning data is not stored in the DB 140, but may be input from a user through an external input means, or may be stored in the form of a file in a storage device such as HDD or SDD. .
- the extraction module 120 performs an extraction step of extracting some of the data instances included in the data bag as a learning instance for each of the plurality of data bags labeled in either the first state or the second state.
- the training instance extracted by the extraction module 120 may be used for training of the pepper neural network 111 .
- FIG. 4 is a diagram illustrating an example of a method in which the extraction module 120 extracts a data instance for learning.
- FIG. 4 exemplifies the case where the training data is as shown in FIG. 1A.
- the extraction module 120 may first perform steps S110 to S130 for each data bag B 1 to B M ( S100 ).
- P ij may be a possibility of the second state, and a cross-entropy loss for the data instance D ij may be calculated as a probability P ij .
- the extraction module 120 is configured to at least some of the possibilities (P i1 to P iN ) for each data instance included in the data bag B i , and each data instance (D i1 to D iN ) included in the data bag. to be on the basis of the determination result of the auto-encoder 112 to determine that the data back-B i for each instance of data (D i1 to D iN) portion of the training instance of included in and (S120), determine the a training instance to the label L i of B i can be labeled (S130).
- step S120 of FIG. 4 is a diagram illustrating an example of a specific process of step S120 of FIG. 4 .
- the autoencoder 112 has been previously trained only on the data instance in the first state.
- the extraction module 120 when the data bag is labeled in the first state, the extraction module 120 performs the autoencoder 112 in the order from the data instance most likely to be in the second state to the less likely to be in the second state. ), it can be determined that the upper part of the data instance determined to be in the first state by the autoencoder is the instance for learning corresponding to the data bag (S121). At this time, when the difference between the data instance input to the auto-encoder 112 and the output data output by the auto-encoder 112 is equal to or greater than a predetermined limit value, the extraction module 120 determines that the data instance input to the auto-encoder 112 is It can be determined as the first state.
- the extraction module 120 inputs to the autoencoder 112 in the order from the data instance most likely to be in the second state to the less likely to be in the second state. , it may be determined that the upper part of the data instance determined to be in the second state by the autoencoder 112 is a learning instance corresponding to the data bag (S122).
- FIG. 6 is a diagram illustrating an example of a specific process of step S121 of FIG. 5 .
- the extraction module 120 may sort the data instances D i1 to D iN in the data bag B i in descending order in the order of possibility of the second state ( S1211 ).
- the state of the instance A k may be determined ( S1213 ), and when it is determined that the data instance A k is in the first state, it may be determined that the data instance A k is a learning instance ( S1215 ).
- the extraction module 120 may perform steps S1213 to S1215 until the loop ends or a predetermined number of learning instances Z corresponding to the data bag B i are found (see S1212 , S1216 , and S1217 ).
- FIGS. 5 and 6 are examples of implementing step S120 of FIG. 4 , and it goes without saying that there may be various methods of implementing step S120 of FIG. 4 .
- the extraction module 120 may extract multiple data instances from one data bag as instances for learning.
- the conventional multi-instance learning method when multiple instances for training are extracted from one data bag, there is a high probability that an incorrect training data instance is extracted, which may have a negative effect on learning of the neural network, but the method according to the technical idea of the present invention According to , by filtering using the pre-trained autoencoder as a training data instance having only one state, there is an effect that it is possible to significantly reduce the extraction of the wrong training data instance.
- the learning module 130 may learn the neural network 111 based on the learning instance extracted by the extraction module 120 .
- the learning module 130 can learn the neural network 111 by backpropagating a loss error between a learning instance input to the neural network 111 and an output value to the neural network 111 . .
- the neural network learning method treats the learning data instance extraction process performed by the extraction module and the learning process performed by the learning module 130 as one epoch, and repeats this for a plurality of epochs By doing so, the performance of the neural network 111 can be improved.
- the neural network learning method according to the technical idea of the present invention can be applied to learning a neural network for image-based disease diagnosis, which can be used for image-based disease diagnosis or diagnosis assistance to help a doctor diagnose.
- image-based disease diagnosis which can be used for image-based disease diagnosis or diagnosis assistance to help a doctor diagnose.
- the neural network 111 receives an image patch obtained by dividing a whole-slide-image into a predetermined size and determines the presence or absence of a lesion due to a predetermined disease in the header image patch. It may be a neural network for diagnosis or a diagnosis aid.
- the DB 140 may store a plurality of pathological image slides.
- the pathology slide image may be various biometric images including tissue images. Meanwhile, each pathological slide image may be labeled with either a first state in which a lesion does not exist (normal state) or a second state in which the lesion exists (abnormal state).
- the autoencoder 112 may be previously learned only from image patches in a normal state in which lesions do not exist. For example, the learner selects only those labeled as normal among the pathological slide images stored in the DB 140 for learning the neural network 111 and divides them into patches, and then learns the autoencoder 112 in advance. have. Alternatively, the learner may learn the autoencoder 112 in advance by collecting a separate patch in a normal state that is not used for learning the neural network 111 .
- the extraction module 120 extracts a patch for learning, which is a part of the patches constituting the pathological slide image for learning, for each of a plurality of pathological slide images for learning that are labeled in either a normal state or an abnormal state. step may be performed, and the learning module 130 may perform a learning step of learning the neural network 111 based on the extracted learning patch.
- the extraction module 120 inputs each patch constituting the pathological slide image for training into the neural network 111 under training, and calculates the probability of each patch constituting the pathological slide image for training Possibility of each patch constituting the pathological slide image for learning, and the pathology slide image for learning based on the determination result of the auto encoder 112 for at least a portion of each patch constituting the pathological slide image for learning It can be determined that a part of each patch constituting the patch is a patch for learning.
- the extraction module 120 is input to the autoencoder 112 in the order from the patch most likely to be abnormal to the less likely to be abnormal. Accordingly, it may be determined that the upper part of the patch determined to be in a normal state by the autoencoder 112 is a patch for learning corresponding to the image for learning.
- the extraction module 120 inputs it to the auto encoder 112 in the order from the patch most likely to be abnormal to the less likely to be abnormal, and the auto encoder It may be determined that the upper part of the patch determined to be in an abnormal state by 112 is a patch for learning corresponding to the image for learning.
- the neural network learned by the above-described neural network learning method may be mounted on a predetermined decision system to perform prediction on input data.
- a neural network mounted in the judgment system may be used to judge an image. That is, the neural network may be for determining whether a predetermined image is in the first state or the second state.
- the determination system may be a predetermined computing system.
- FIG. 7 is a diagram illustrating a schematic configuration of a decision system using a neural network according to an embodiment of the present invention.
- the determination system 300 using the neural network may include a storage module 310 , a patch unit determination module 320 , and an output module 330 .
- the storage module 310 may store the neural network learned by the above-described neural network learning method.
- the storage module 310 may be a storage means capable of storing various types of information and data including a neural network.
- the patch unit determination module 320 may obtain a determination result corresponding to each of the plurality of diagnostic patches by inputting each of a plurality of diagnostic patches obtained by dividing a predetermined determination target image into the neural network.
- the output module 330 may output a grid map of the determination target image based on determination results of each of the plurality of diagnostic patches obtained by the patch unit diagnostic module.
- the grid map may mean a map capable of visually distinguishing the patch area in the first state and the patch area in the second state.
- the determination system using the neural network ( ) may be used as a system for receiving a pathological slide image and outputting a lesion area.
- the determination system 300 may receive a pathological slide image including a lesion caused by a predetermined disease, and the patch unit determination module 320 divides the pathological slide image into patch units, The presence or absence of a lesion can be determined for each patch. Thereafter, the output module ( ) may output the lesion area of the pathological slide image in the form of a grid map divided in units of patches based on the determination result for each patch.
- FIG. 8 shows an example of outputting the lesion area in the pathological slide image in the form of a grid map.
- a region (lesion region) corresponding to the patch determined as the second state is displayed in a dark color.
- the computing device 100 may include a processor and a storage device.
- the processor may mean an arithmetic device capable of driving a program for implementing the technical idea of the present invention, and may perform a neural network learning method defined by the program and the technical idea of the present invention.
- the processor may include a single-core CPU or a multi-core CPU.
- the storage device may mean a data storage means capable of storing a program and various data necessary for implementing the technical idea of the present invention, and may be implemented as a plurality of storage means according to an embodiment.
- the storage device may be meant to include not only the main storage device included in the computing device 100 , but also a temporary storage device or memory that may be included in the processor.
- the memory may include high-speed random access memory and may include non-volatile memory such as one or more magnetic disk storage devices, flash memory devices, or other non-volatile solid-state memory devices. Access to memory by the processor and other components may be controlled by a memory controller.
- the method according to the embodiment of the present invention may be implemented in the form of a computer-readable program command and stored in a computer-readable recording medium, and the control program and the target program according to the embodiment of the present invention are also implemented in the computer. It may be stored in a readable recording medium.
- the computer-readable recording medium includes all types of recording devices in which data readable by a computer system is stored.
- the program instructions recorded on the recording medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the software field.
- Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and floppy disks, and hardware devices specially configured to store and execute program instructions, such as magneto-optical media and ROM, RAM, flash memory, and the like.
- the computer-readable recording medium is distributed in network-connected computer systems, and computer-readable codes can be stored and executed in a distributed manner.
- Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by an apparatus for electronically processing information using an interpreter or the like, for example, a computer.
- the hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.
- the present invention can be used in a method for learning a neural network through autoencoder and multi-instance learning, and a computing system performing the same.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Epidemiology (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Description
Claims (17)
- 입력된 데이터 인스턴스(instance)가 제1상태인지 제2상태인지를 판단하기 위한 오토인코더 및 입력된 데이터 인스턴스가 제1상태 또는 제2상태일 가능성을 출력할 수 있는 뉴럴 네트워크를 포함하는 컴퓨팅 시스템에서 수행되는 뉴럴 네트워크 학습 방법으로서,제1상태 또는 제2상태 중 어느 하나로 라벨링되어 있는 복수의 데이터 백(bag) 각각에 대하여, 상기 데이터 백에 포함된 데이터 인스턴스들 중 일부인 학습용 인스턴스를 추출하는 추출단계; 및상기 복수의 데이터 백 각각에 상응하는 학습용 인스턴스에 기초하여 상기 뉴럴 네트워크를 학습하는 학습단계를 포함하되,상기 추출단계는,상기 데이터 백에 포함된 각 데이터 인스턴스를 학습 중인 상기 뉴럴 네트워크에 입력하여, 상기 데이터 백에 포함된 각 데이터 인스턴스 별 가능성을 산출하는 단계; 및상기 데이터 백에 포함된 각 데이터 인스턴스 별 가능성, 및 상기 데이터 백에 포함된 각 데이터 인스턴스 중 적어도 일부에 대한 상기 오토인코더의 판단 결과에 기초하여 상기 데이터 백에 포함된 각 데이터 인스턴스 중 일부를 학습용 인스턴스라고 판단하는 단계를 포함하는 뉴럴 네트워크 학습 방법.
- 제1항에 있어서,상기 뉴럴 네트워크 학습 시스템은,상기 오토인코더에 입력된 데이터 인스턴스와 상기 오토인코더에서 출력된 출력 데이터와의 차이에 기초하여, 상기 오토인코더에 입력된 데이터 인스턴스가 제1상태인지 제2상태인지를 판단하는 뉴럴 네트워크 학습 방법.
- 제1항에 있어서,상기 오토인코더는, 제1상태인 데이터 인스턴스만으로 미리 학습되어 있으며,상기 데이터 백에 포함된 각 데이터 인스턴스 별 가능성, 및 상기 데이터 백에 포함된 각 데이터 인스턴스 중 적어도 일부에 대한 상기 오토인코더의 판단 결과에 기초하여 상기 데이터 백에 포함된 각 데이터 인스턴스 중 일부를 학습용 인스턴스라고 판단하는 단계는,상기 데이터 백이 제1상태로 라벨링된 경우,제2상태일 가능성이 가장 높은 데이터 인스턴스에서부터 제2상태일 가능성이 낮아지는 순서로 상기 오토인코더에 입력하여, 상기 오토인코더에 의해 제1상태라고 판단된 상위 일부의 데이터 인스턴스를 상기 데이터 백에 상응하는 학습용 인스턴스라고 판단하는 단계; 및상기 데이터 백이 제2상태로 라벨링된 경우,제2상태일 가능성이 가장 높은 데이터 인스턴스에서부터 제2상태일 가능성이 낮아지는 순서로 상기 오토인코더에 입력하여, 상기 오토인코더에 의해 제2상태라고 판단된 상위 일부의 데이터 인스턴스를 상기 데이터 백에 상응하는 학습용 인스턴스라고 판단하는 단계를 포함하는 뉴럴 네트워크 학습 방법.
- 제1항에 있어서,상기 뉴럴 네트워크 학습 방법은,상기 추출단계 및 상기 학습단계를 1 에폭(epoch) 이상 반복 수행하는 단계를 더 포함하는 뉴럴 네트워크 학습 방법.
- 제1항에 있어서,상기 복수의 데이터 백 각각은, 홀-이미지(whole-image)이며,상기 복수의 데이터 백 각각에 포함된 데이터 인스턴스(instance)는, 상기 데이터 백에 상응하는 홀-이미지를 소정의 크기로 분할한 각각의 이미지 패치인 것을 특징으로 하는 뉴럴 네트워크 학습 방법
- 입력된 패치가 제1상태인지 제2상태인지를 판단하기 위한 오토인코더-여기서, 상기 패치는, 이미지를 소정의 크기로 분할한 것 중 하나임-; 및 입력된 패치가 제1상태 또는 제2상태일 가능성을 출력할 수 있는 뉴럴 네트워크를 포함하는 컴퓨팅 시스템에서 수행되는 뉴럴 네트워크 학습 방법으로서,제1상태 또는 제2상태 중 어느 하나로 라벨링되어 있는 복수의 학습용 이미지 각각에 대하여, 상기 학습용 이미지를 구성하는 패치 중 일부인 학습용 패치를 추출하는 추출단계; 및상기 복수의 학습용 이미지 각각에 상응하는 학습용 패치에 기초하여 상기 뉴럴 네트워크를 학습하는 학습단계를 포함하되,상기 추출단계는,상기 학습용 이미지를 구성하는 각 패치를 훈련중인 상기 뉴럴 네트워크에 입력하여, 상기 학습용 이미지를 구성하는 각 패치 별 가능성을 산출하는 단계; 및상기 학습용 이미지를 구성하는 각 패치 별 가능성, 및 상기 학습용 이미지를 구성하는 각 패치 중 적어도 일부에 대한 상기 오토인코더의 판단 결과에 기초하여 상기 학습용 이미지를 구성하는 각 패치 중 일부를 학습용 패치라고 판단하는 단계를 포함하는 뉴럴 네트워크 학습 방법.
- 제6항에 있어서,상기 오토인코더는, 제1상태인 패치만으로 미리 학습되어 있으며,상기 학습용 이미지를 구성하는 각 패치 별 가능성, 및 상기 학습용 이미지를 구성하는 각 패치 중 적어도 일부에 대한 상기 오토인코더의 판단 결과에 기초하여 상기 학습용 이미지를 구성하는 각 패치 중 일부를 학습용 패치라고 판단하는 단계는,상기 학습용 이미지가 제1상태로 라벨링된 경우,제2상태일 가능성이 가장 높은 패치에서부터 제2상태일 가능성이 낮아지는 순서로 상기 오토인코더에 입력하여, 상기 오토인코더에 의해 제1상태라고 판단된 상위 일부의 패치를 상기 학습용 이미지에 상응하는 학습용 패치라고 판단하는 단계; 및상기 학습용 이미지가 제2상태로 라벨링된 경우,제2상태일 가능성이 가장 높은 패치에서부터 제2상태일 가능성이 낮아지는 순서로 상기 오토인코더에 입력하여, 상기 오토인코더에 의해 제2상태라고 판단된 상위 일부의 패치를 상기 학습용 이미지에 상응하는 학습용 패치라고 판단하는 단계를 포함하는 뉴럴 네트워크 학습 방법.
- 제6항에 있어서,상기 학습용 이미지 각각은, 소정의 질병으로 인한 병변이 포함된 이미지 혹은 상기 병변이 포함되지 않는 이미지 중 어느 하나이며,상기 제1상태는 상기 병변이 존재하지 않는 정상 상태이며,상기 제2상태는 상기 병변이 존재하는 비정상 상태인 뉴럴 네트워크 학습 방법.
- 제6항에 기재된 뉴럴 네트워크 학습 방법에 의하여 학습된 뉴럴 네트워크를 저장하는 저장모듈;소정의 판단 대상 이미지를 분할한 복수의 진단 패치 각각을 상기 뉴럴 네트워크에 입력하여 상기 복수의 진단 패치 각각에 상응하는 판단 결과를 획득하는 패치단위 판단모듈;상기 패치단위 진단모듈에 의해 획득된 상기 복수의 진단 패치 각각의 판단 결과에 기초하여, 상기 판단 대상 이미지의 히트맵을 출력하는 출력모듈을 포함하는 뉴럴 네트워크를 이용한 판단 시스템.
- 데이터 처리장치에 설치되며 제1항 내지 제8항 중 어느 한 항에 기재된 방법을 수행하기 위한 매체에 기록된 컴퓨터 프로그램.
- 제1항 내지 제8항 중 어느 한 항에 기재된 방법을 수행하기 위한 컴퓨터 프로그램이 기록된 컴퓨터 판독가능한 기록매체.
- 컴퓨팅 시스템으로서,프로세서 및 메모리를 포함하되,상기 메모리는, 상기 프로세서에 의해 수행될 경우, 상기 컴퓨팅 시스템이 제1항 내지 제8항 중 어느 한 항에 기재된 방법을 수행하도록 하는 컴퓨팅 시스템.
- 입력된 데이터 인스턴스(instance)가 제1상태인지 제2상태인지를 판단하기 위한 오토인코더 및 입력된 데이터 인스턴스가 제1상태 또는 제2상태일 가능성을 출력할 수 있는 뉴럴 네트워크를 저장하는 저장모듈;제1상태 또는 제2상태 중 어느 하나로 라벨링되어 있는 복수의 데이터 백(bag) 각각에 대하여, 상기 데이터 백에 포함된 데이터 인스턴스들 중 일부인 학습용 인스턴스를 추출하는 추출모듈; 및상기 복수의 데이터 백 각각에 상응하는 학습용 인스턴스에 기초하여 상기 뉴럴 네트워크를 학습하는 학습모듈을 포함하되,상기 추출모듈은,상기 데이터 백에 포함된 각 데이터 인스턴스를 학습 중인 상기 뉴럴 네트워크에 입력하여, 상기 데이터 백에 포함된 각 데이터 인스턴스 별 가능성을 산출하고,상기 데이터 백에 포함된 각 데이터 인스턴스 별 가능성, 및 상기 데이터 백에 포함된 각 데이터 인스턴스 중 적어도 일부에 대한 상기 오토인코더의 판단 결과에 기초하여 상기 데이터 백에 포함된 각 데이터 인스턴스 중 일부를 학습용 인스턴스라고 판단하는 뉴럴 네트워크 학습 시스템.
- 제13항에 있어서,상기 추출모듈은,상기 오토인코더에 입력된 데이터 인스턴스와 상기 오토인코더에서 출력된 출력 데이터와의 차이에 기초하여, 상기 오토인코더에 입력된 데이터 인스턴스가 제1상태인지 제2상태인지를 판단하는 뉴럴 네트워크 학습 시스템.
- 제12항에 있어서,상기 오토인코더는, 제1상태인 데이터 인스턴스만으로 미리 학습되어 있으며,추출모듈은, 상기 데이터 백에 포함된 각 데이터 인스턴스 별 가능성, 및 상기 데이터 백에 포함된 각 데이터 인스턴스 중 적어도 일부에 대한 상기 오토인코더의 판단 결과에 기초하여 상기 데이터 백에 포함된 각 데이터 인스턴스 중 일부를 학습용 인스턴스라고 판단하기 위하여,상기 데이터 백이 제1상태로 라벨링된 경우,제2상태일 가능성이 가장 높은 데이터 인스턴스에서부터 제2상태일 가능성이 낮아지는 순서로 상기 오토인코더에 입력하여, 상기 오토인코더에 의해 제1상태라고 판단된 상위 일부의 데이터 인스턴스를 상기 데이터 백에 상응하는 학습용 인스턴스라고 판단하고,상기 데이터 백이 제2상태로 라벨링된 경우,제2상태일 가능성이 가장 높은 데이터 인스턴스에서부터 제2상태일 가능성이 낮아지는 순서로 상기 오토인코더에 입력하여, 상기 오토인코더에 의해 제2상태라고 판단된 상위 일부의 데이터 인스턴스를 상기 데이터 백에 상응하는 학습용 인스턴스라고 판단하는 뉴럴 네트워크 학습 시스템.
- 입력된 패치가 제1상태인지 제2상태인지를 판단하기 위한 오토인코더-여기서, 상기 패치는, 이미지를 소정의 크기로 분할한 것 중 하나이며, 상기 오토인코더는, 복수의 제1상태인 패치만으로 미리 학습됨; 및입력된 패치가 제1상태 또는 제2상태일 가능성을 출력할 수 있는 뉴럴 네트워크를 저장하는 저장모듈;제1상태 또는 제2상태 중 어느 하나로 라벨링되어 있는 복수의 학습용 이미지 각각에 대하여, 상기 학습용 이미지를 구성하는 패치 중 일부인 학습용 패치를 추출하는 추출모듈; 및상기 복수의 학습용 이미지 각각에 상응하는 학습용 패치에 기초하여 상기 뉴럴 네트워크를 학습하는 학습모듈을 포함하되,상기 추출모듈은,상기 학습용 이미지를 구성하는 각 패치를 훈련중인 상기 뉴럴 네트워크에 입력하여, 상기 학습용 이미지를 구성하는 각 패치 별 가능성을 산출하고,상기 학습용 이미지를 구성하는 각 패치 별 가능성, 및 상기 학습용 이미지를 구성하는 각 패치 중 적어도 일부에 대한 상기 오토인코더의 판단 결과에 기초하여 상기 학습용 이미지를 구성하는 각 패치 중 일부를 학습용 패치라고 판단하는 뉴럴 네트워크 학습 시스템.
- 제16항에 있어서,상기 오토인코더는, 제1상태인 패치만으로 미리 학습되어 있으며,상기 추출모듈은, 상기 학습용 이미지를 구성하는 각 패치 별 가능성, 및 상기 학습용 이미지를 구성하는 각 패치 중 적어도 일부에 대한 상기 오토인코더의 판단 결과에 기초하여 상기 학습용 이미지를 구성하는 각 패치 중 일부를 학습용 패치라고 판단하기 위하여,상기 학습용 이미지가 제1상태로 라벨링된 경우,제2상태일 가능성이 가장 높은 패치에서부터 제2상태일 가능성이 낮아지는 순서로 상기 오토인코더에 입력하여, 상기 오토인코더에 의해 제1상태라고 판단된 상위 일부의 패치를 상기 학습용 이미지에 상응하는 학습용 패치라고 판단하고,상기 학습용 이미지가 제2상태로 라벨링된 경우,제2상태일 가능성이 가장 높은 패치에서부터 제2상태일 가능성이 낮아지는 순서로 상기 오토인코더에 입력하여, 상기 오토인코더에 의해 제2상태라고 판단된 상위 일부의 패치를 상기 학습용 이미지에 상응하는 학습용 패치라고 판단하는 뉴럴 네트워크 학습 시스템.
Priority Applications (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| EP21818809.2A EP4148629A4 (en) | 2020-06-05 | 2021-06-03 | NEURAL NETWORK TRAINING METHOD USING AUTOENCODER AND MULTIPLE INSTANCE LEARNING AND COMPUTER SYSTEM FOR IMPLEMENTING IT |
| US18/008,407 US20230244930A1 (en) | 2020-06-05 | 2021-06-03 | Neural network learning method using auto encoder and multiple instance learning and computing system performing the same |
| CN202180040650.7A CN115777107A (zh) | 2020-06-05 | 2021-06-03 | 通过自动编码器及多实例学习进行的神经网络学习方法以及执行其的计算系统 |
| JP2022572516A JP7628715B2 (ja) | 2020-06-05 | 2021-06-03 | オートエンコーダ及びマルチインスタンス学習を通じたニューラルネットワーク学習方法並びにこれを実行するコンピューティングシステム |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR1020200068640A KR102163519B1 (ko) | 2020-06-05 | 2020-06-05 | 오토인코더 및 다중 인스턴스 학습을 통한 뉴럴 네트워크 학습 방법 및 이를 수행하는 컴퓨팅 시스템 |
| KR10-2020-0068640 | 2020-06-05 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2021246810A1 true WO2021246810A1 (ko) | 2021-12-09 |
Family
ID=72883662
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/KR2021/006966 Ceased WO2021246810A1 (ko) | 2020-06-05 | 2021-06-03 | 오토인코더 및 다중 인스턴스 학습을 통한 뉴럴 네트워크 학습 방법 및 이를 수행하는 컴퓨팅 시스템 |
Country Status (6)
| Country | Link |
|---|---|
| US (1) | US20230244930A1 (ko) |
| EP (1) | EP4148629A4 (ko) |
| JP (1) | JP7628715B2 (ko) |
| KR (1) | KR102163519B1 (ko) |
| CN (1) | CN115777107A (ko) |
| WO (1) | WO2021246810A1 (ko) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2024080767A (ja) * | 2022-12-05 | 2024-06-17 | 株式会社東芝 | 学習装置、方法及びプログラム |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102163519B1 (ko) * | 2020-06-05 | 2020-10-07 | 주식회사 딥바이오 | 오토인코더 및 다중 인스턴스 학습을 통한 뉴럴 네트워크 학습 방법 및 이를 수행하는 컴퓨팅 시스템 |
| KR102556173B1 (ko) * | 2021-02-16 | 2023-07-14 | 한국기술교육대학교 산학협력단 | 기계학습 모델의 이상 데이터 추출 방법 및 장치 |
| KR102317992B1 (ko) * | 2021-03-26 | 2021-10-28 | 주식회사 트윔 | 뉴럴 네트워크를 이용한 제품 검사 방법, 장치 및 제품 검사 장치 학습 방법 |
| US20240203100A1 (en) | 2021-04-22 | 2024-06-20 | Nec Corporation | Training device, prediction device, training method, and recording medium |
| KR102821942B1 (ko) | 2022-12-28 | 2025-06-17 | 중앙대학교 산학협력단 | 멀티-변이형 오토 인코더를 활용한 사용자 예측 데이터 생성 장치 및 방법 |
| KR102809726B1 (ko) * | 2023-10-23 | 2025-05-20 | 주식회사 테렌즈 | Ai를 이용하는 유두갑상선암 검출 방법 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019186198A1 (en) * | 2018-03-29 | 2019-10-03 | Benevolentai Technology Limited | Attention filtering for multiple instance learning |
| KR20190143510A (ko) * | 2018-06-04 | 2019-12-31 | 주식회사 딥바이오 | 투 페이스 질병 진단 시스템 및 그 방법 |
| KR20200044183A (ko) * | 2018-10-05 | 2020-04-29 | 주식회사 딥바이오 | 병리 이미지 검색을 위한 시스템 및 방법 |
| KR20200057547A (ko) * | 2018-11-16 | 2020-05-26 | 주식회사 딥바이오 | 지도학습기반의 합의 진단방법 및 그 시스템 |
| KR102163519B1 (ko) * | 2020-06-05 | 2020-10-07 | 주식회사 딥바이오 | 오토인코더 및 다중 인스턴스 학습을 통한 뉴럴 네트워크 학습 방법 및 이를 수행하는 컴퓨팅 시스템 |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0457181A (ja) * | 1990-06-26 | 1992-02-24 | Toshiba Corp | 多重神経回路網の構築方法およびその装置 |
| KR101623431B1 (ko) * | 2015-08-06 | 2016-05-23 | 주식회사 루닛 | 의료 영상의 병리 진단 분류 장치 및 이를 이용한 병리 진단 시스템 |
| CN108053027B (zh) * | 2017-12-18 | 2021-04-30 | 中山大学 | 一种加速深度神经网络的方法及装置 |
| CN108596200A (zh) * | 2018-01-03 | 2018-09-28 | 深圳北航新兴产业技术研究院 | 医学图像分类的方法和装置 |
| US10699407B2 (en) | 2018-04-11 | 2020-06-30 | Pie Medical Imaging B.V. | Method and system for assessing vessel obstruction based on machine learning |
| EP3877948A4 (en) * | 2018-11-05 | 2022-11-23 | Huron Technologies International Inc. | SYSTEMS AND METHODS FOR MANAGEMENT OF MEDICAL IMAGES |
| US12159229B2 (en) * | 2019-05-29 | 2024-12-03 | Georgia Tech Research Corporation | Transfer learning for medical applications using limited data |
-
2020
- 2020-06-05 KR KR1020200068640A patent/KR102163519B1/ko active Active
-
2021
- 2021-06-03 US US18/008,407 patent/US20230244930A1/en active Pending
- 2021-06-03 EP EP21818809.2A patent/EP4148629A4/en not_active Withdrawn
- 2021-06-03 JP JP2022572516A patent/JP7628715B2/ja active Active
- 2021-06-03 WO PCT/KR2021/006966 patent/WO2021246810A1/ko not_active Ceased
- 2021-06-03 CN CN202180040650.7A patent/CN115777107A/zh active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019186198A1 (en) * | 2018-03-29 | 2019-10-03 | Benevolentai Technology Limited | Attention filtering for multiple instance learning |
| KR20190143510A (ko) * | 2018-06-04 | 2019-12-31 | 주식회사 딥바이오 | 투 페이스 질병 진단 시스템 및 그 방법 |
| KR20200044183A (ko) * | 2018-10-05 | 2020-04-29 | 주식회사 딥바이오 | 병리 이미지 검색을 위한 시스템 및 방법 |
| KR20200057547A (ko) * | 2018-11-16 | 2020-05-26 | 주식회사 딥바이오 | 지도학습기반의 합의 진단방법 및 그 시스템 |
| KR102163519B1 (ko) * | 2020-06-05 | 2020-10-07 | 주식회사 딥바이오 | 오토인코더 및 다중 인스턴스 학습을 통한 뉴럴 네트워크 학습 방법 및 이를 수행하는 컴퓨팅 시스템 |
Non-Patent Citations (3)
| Title |
|---|
| IWATA TOMOHARU; TOYODA MACHIKO; TORA SHOTARO; UEDA NAONORI: "Anomaly detection with inexact labels", MACHINE LEARNING, vol. 109, no. 8, 31 May 2020 (2020-05-31), New York, pages 1617 - 1633, XP037469118, ISSN: 0885-6125, DOI: 10.1007/s10994-020-05880-w * |
| See also references of EP4148629A4 * |
| TONG-TONG CHEN; CHAN-JUAN LIU; HAI-LIN ZOU; SHU-SEN ZHOU; YING LIU; XIN-MIAO DING: "A multi-instance multi-label scene classification method based on multi-kernel fusion", 2015 SAI INTELLIGENT SYSTEMS CONFERENCE (INTELLISYS), IEEE, 10 November 2015 (2015-11-10), pages 782 - 787, XP032834533, DOI: 10.1109/IntelliSys.2015.7361229 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2024080767A (ja) * | 2022-12-05 | 2024-06-17 | 株式会社東芝 | 学習装置、方法及びプログラム |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115777107A (zh) | 2023-03-10 |
| KR102163519B1 (ko) | 2020-10-07 |
| US20230244930A1 (en) | 2023-08-03 |
| EP4148629A4 (en) | 2024-05-15 |
| EP4148629A1 (en) | 2023-03-15 |
| JP7628715B2 (ja) | 2025-02-12 |
| JP2023529585A (ja) | 2023-07-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021246810A1 (ko) | 오토인코더 및 다중 인스턴스 학습을 통한 뉴럴 네트워크 학습 방법 및 이를 수행하는 컴퓨팅 시스템 | |
| WO2021246811A1 (ko) | 중증도 판단용 뉴럴 네트워크 학습 방법 및 시스템 | |
| Lyu et al. | Classification of alzheimer's disease via vision transformer: Classification of alzheimer's disease via vision transformer | |
| WO2020071877A1 (ko) | 병리 이미지 검색을 위한 시스템 및 방법 | |
| WO2018230832A1 (en) | Image processing apparatus and method using multi-channel feature map | |
| WO2022065798A1 (ko) | 병리 이미지 분석 결과 출력 방법 및 이를 수행하는 컴퓨팅 시스템 | |
| WO2017213398A1 (en) | Learning model for salient facial region detection | |
| US7711157B2 (en) | Artificial intelligence systems for identifying objects | |
| WO2018212494A1 (ko) | 객체를 식별하는 방법 및 디바이스 | |
| WO2021045507A2 (ko) | Ct 영상 기반 부위별 대뇌 피질 수축율 예측 방법 및 장치 | |
| WO2020196985A1 (ko) | 비디오 행동 인식 및 행동 구간 탐지 장치 및 방법 | |
| WO2021040287A1 (ko) | 사람 재식별 장치 및 방법 | |
| WO2021153861A1 (ko) | 다중 객체 검출 방법 및 그 장치 | |
| Li et al. | Knowledge-spreader: Learning semi-supervised facial action dynamics by consistifying knowledge granularity | |
| WO2023191564A1 (ko) | 심층신경망 기반의 관심질병 예측 장치, 방법 및 이를 위한 컴퓨터 판독가능 프로그램 | |
| WO2018212584A2 (ko) | 딥 뉴럴 네트워크를 이용하여 문장이 속하는 클래스를 분류하는 방법 및 장치 | |
| WO2021071288A1 (ko) | 골절 진단모델의 학습 방법 및 장치 | |
| WO2019035544A1 (ko) | 학습을 이용한 얼굴 인식 장치 및 방법 | |
| WO2023132428A1 (ko) | 재순위화를 통한 객체 검색 | |
| WO2022158843A1 (ko) | 조직 검체 이미지 정제 방법, 및 이를 수행하는 컴퓨팅 시스템 | |
| WO2022019356A1 (ko) | 준-지도학습을 이용하여 질병의 발병 영역에 대한 어노테이션을 수행하기 위한 방법 및 이를 수행하는 진단 시스템 | |
| WO2021101052A1 (ko) | 배경 프레임 억제를 통한 약한 지도 학습 기반의 행동 프레임 검출 방법 및 장치 | |
| WO2020209487A1 (ko) | 인공 신경망 기반의 장소 인식 장치 및 이의 학습 장치 | |
| WO2021015489A2 (ko) | 인코더를 이용한 이미지의 특이 영역 분석 방법 및 장치 | |
| WO2022019355A1 (ko) | 다중 페이즈 생체 이미지를 이용하여 학습된 뉴럴 네트워크를 이용한 질병 진단 방법 및 이를 수행하는 질병 진단 시스템 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21818809 Country of ref document: EP Kind code of ref document: A1 |
|
| ENP | Entry into the national phase |
Ref document number: 2022572516 Country of ref document: JP Kind code of ref document: A |
|
| ENP | Entry into the national phase |
Ref document number: 2021818809 Country of ref document: EP Effective date: 20221205 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| WWW | Wipo information: withdrawn in national office |
Ref document number: 2021818809 Country of ref document: EP |