WO2019235827A1 - 듀얼 클래스를 지원하는 질병 진단 시스템 및 그 방법 - Google Patents
듀얼 클래스를 지원하는 질병 진단 시스템 및 그 방법 Download PDFInfo
- Publication number
- WO2019235827A1 WO2019235827A1 PCT/KR2019/006758 KR2019006758W WO2019235827A1 WO 2019235827 A1 WO2019235827 A1 WO 2019235827A1 KR 2019006758 W KR2019006758 W KR 2019006758W WO 2019235827 A1 WO2019235827 A1 WO 2019235827A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- unit
- state
- states
- neural network
- disease
- Prior art date
Links
- 201000010099 disease Diseases 0.000 title claims abstract description 68
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 68
- 238000003745 diagnosis Methods 0.000 title claims abstract description 48
- 230000009977 dual effect Effects 0.000 title claims abstract description 25
- 238000000034 method Methods 0.000 title claims description 28
- 238000013528 artificial neural network Methods 0.000 claims abstract description 95
- 230000006870 function Effects 0.000 claims abstract description 32
- 238000002372 labelling Methods 0.000 claims abstract description 23
- 206010060862 Prostate cancer Diseases 0.000 claims description 10
- 208000000236 Prostatic Neoplasms Diseases 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 description 10
- 230000011218 segmentation Effects 0.000 description 9
- 210000001519 tissue Anatomy 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 206010028980 Neoplasm Diseases 0.000 description 2
- 201000011510 cancer Diseases 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007170 pathology Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000008827 biological function Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Images
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- 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
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Definitions
- the present invention relates to a system and method for diagnosing diseases using neural networks. More specifically, the neural network can be trained to diagnose diseases while allowing dual class labeling for specific unit units (e.g. pixels of images corresponding to biological tissue), which can be used to achieve higher accuracy.
- a system and method for constructing a network are provided.
- One of the main tasks of the pathology or pathology department is to read the biometric image of the patient and perform a diagnosis to determine the condition or indication for a particular disease. This diagnosis is dependent on the experience and knowledge of long-term skilled practitioners.
- diagnosis through deep learning using a neural network does not merely automate the experience and knowledge of a conventionally trained medical practitioner, but rather finds characteristic elements through self-learning and derives the desired solution. On the contrary, sometimes the characteristics of disease factors that are not known by skilled medical personnel are found in the image.
- a neural network eg, CNN
- the diagnosis of a disease through a neural network using a bioimage uses a piece of the bioimage, that is, a patch (or also called a tile). That is, a medical practitioner skilled in the patch will annotate a specific disease state (eg, whether cancer has been expressed), and train the neural network using these annotated multiple patches as training data.
- the neural network may use a convolutional neural network.
- the neural network is trained to determine whether the disease is manifested for each patch (for example, 512 by 512 pixels), so that there is a problem in which a part of the patch is not known.
- the neural network is also treated as a unit unit that has a plurality of states clearly like a unit unit having a specific state.
- the technical problem to be solved by the present invention is not the input data (unit unit) that can be clearly classified into any one of classes that the neural network outputs in the first state or the second state, but the determination of the class to be classified is ambiguous.
- the characteristic feature of the unit unit in which there is a possibility of a plurality of states is to provide a system and method for improving the performance by making it possible to use for learning. .
- the present invention provides a system and method for supporting a deep learning model that enables detailed segmentation of the disease-expressing part and the non-patch part by enabling pixel-specific diagnosis rather than patch-specific diagnosis.
- a system for diagnosing a disease using a biological image and the neural network is implemented in a system including a processor and a storage device for storing a neural network. At least one state of the plurality of states may be determined using learning data labeled that the unit unit has at least one of a plurality of states for a loss function and a predetermined unit unit included in the bioimage.
- the new network includes a specific layer outputting a plurality of feature values corresponding to a probability that the unit unit is determined as each of a plurality of states, and the loss function
- the plurality of states of the plurality of unit units-the plurality of states Includes a first state that is determined to have a higher probability state and a second state that is determined to have a lower state than the first state-for a dual labeling unit unit labeled as the second of the feature values of the particular layer.
- the loss is calculated by reflecting both the first feature value corresponding to the first state and the second feature value corresponding to the second state.
- the loss function may be defined such that the second feature value has more losses than the first feature value.
- the loss function may be defined by Equation 1 below.
- the feature value corresponding to the first state Is the first feature value, Is the second feature value.
- the unit unit may be a pixel unit of the biological image.
- the processor determines that the pixel is a disease pixel corresponding to a disease, and based on the determination result, the disease pixel in the biological image is another pixel. It can be displayed separately from.
- the disease may be prostate cancer.
- the plurality of conditions may include normal, Gleason score 3, Gleason score 4, and Gleason score 5.
- a method for diagnosing a disease using a biological image and a neural network which is implemented in a system including a processor and a storage device for solving the technical problem, may be performed by performing a predetermined loss function and the biological function on the neural network stored in the storage device.
- the loss may be defined by reflecting both a first feature value corresponding to a first state and a second feature value corresponding to the second state among the feature values of the specific layer.
- the method may include determining that the corresponding pixel is a disease pixel corresponding to a disease when each pixel has a predetermined determination value based on the determination value of the pixel unit, and different from the disease pixel in the biological image based on the determination result.
- the method may further include distinguishing and displaying the pixel.
- the above method may be implemented through a computer program installed in the data processing apparatus and hardware of the data processing apparatus capable of executing the computer program.
- the case of the unit unit is not input data (unit unit) that can be clearly classified as any one of the plurality of states, but the unit unit in which there is a possibility that the state of the class at the time of labeling is ambiguous
- the characteristic features of the unit unit 1, which may exist in such a plurality of states may also be used for learning, thereby improving performance.
- FIG. 1 is a view showing a schematic system configuration of a system for diagnosing diseases using a neural network according to the technical spirit of the present invention.
- FIG. 2 is a view for explaining the logical configuration of a diagnosis system for diseases using a neural network according to an embodiment of the present invention.
- FIG. 3 is a diagram illustrating a hardware configuration of a system for diagnosing diseases using a neural network according to an exemplary embodiment of the present invention.
- FIG. 4 is a diagram conceptually illustrating a configuration of a neural network according to another embodiment of the present invention.
- FIG. 5 is a diagram exemplarily illustrating a segmentation result according to an exemplary embodiment of the present invention.
- first and second may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.
- the component when 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. Means that the data may be transmitted to the other component.
- the component 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. 1 is a view showing a schematic system configuration of a system for diagnosing diseases using a neural network according to the technical spirit of the present invention.
- a diagnosis system for a disease using a neural network (hereinafter, referred to as a diagnosis system) 100 according to the spirit of the present invention may be installed in a predetermined server 10 to implement the spirit of the present invention.
- the server 10 refers to a data processing apparatus having a computing capability for implementing the technical idea of the present invention, and generally provides a specific service such as a personal computer, a portable terminal, etc. as well as a data processing apparatus accessible by a client through a network. It will be readily apparent to the average person skilled in the art that any device that can perform may be defined as a server.
- the server 10 may include a processor 11 and a storage device 12 as shown in FIG. 3.
- the processor 11 may refer to an operation device capable of driving a program 12-1 for implementing the technical idea of the present invention, and the processor 11 includes the program 12-1 and the present invention. Diagnosis can be performed using a neural network (12-2) defined by the technical idea of.
- the program 12-1 may refer to software defined for implementing the technical idea of the present invention.
- the storage device 12 may mean data storage means capable of storing the program 12-1 and the neural network 12-2, and may be implemented as a plurality of storage means according to an embodiment.
- the storage device 12 may mean not only a main memory device included in the server 10, but also a temporary storage device or a memory that may be included in the processor 11.
- diagnostic system 100 is illustrated as being implemented as any one physical device in FIG. 1 or 3, a plurality of physical devices are organically coupled as necessary to provide a diagnostic system 100 according to the spirit of the present invention. It will be easy for the average person skilled in the art to infer that it can be implemented.
- that the diagnostic system 100 performs a predetermined function may mean that the processor 11 performs a predetermined function using the program 12-1.
- the diagnosis system 100 may mean a series of processes for receiving a bio-image in which a biological tissue is expressed for each predetermined unit unit and outputting output data defined in the present specification.
- the unit unit may mean a unit for performing diagnosis.
- the unit may be a pixel unit of the biometric image.
- the unit may be a patch unit of the biometric image or a slide unit.
- the unit unit is a pixel unit, therefore, a case in which it is determined whether the result diagnosed by the diagnosis system 100 for each pixel is a pixel in which a disease is expressed or a pixel that is not described as an example will be described. It will be readily apparent to the average person skilled in the art that the unit unit may vary.
- the output data may refer to information output from an output layer of a neural network used by the diagnostic system 100.
- the output data may include a biotissue corresponding to a unit unit included in the biometric image for a particular disease. It may include state information indicating whether the state.
- the status information may be information corresponding to a probability of whether a specific disease (eg, a certain kind of cancer) is expressed in the tissue corresponding to the unit unit.
- a specific disease eg, a certain kind of cancer
- information indicating the progress of a specific disease may be used.
- a Gleason Pattern or a Gleason Score which is an index indicating the progress of prostate cancer, may be included in the state information.
- the Gleason pattern has a value of 2 to 5, and the larger the number, the more severe the prostate cancer is expressed.
- the state information is information or normal corresponding to the probability that the biological tissue corresponding to the unit unit to be diagnosed corresponds to a specific value (eg, 3, 4, or 5) of the Gleason pattern (ie, disease is not expressed). Information corresponding to the probability).
- the state may be implemented by four channels.
- the first state may be a state in which the corresponding unit unit is normal
- the second state may be a state in which the Gleason pattern is 3
- the third state may be a state in which the Gleason pattern is 4
- the fourth state may be a state in which the Gleason pattern is 5.
- the feature value corresponding to each state channel may be a value corresponding to the probability of each state, and the neural network may be learned so that the greater the value, the greater the likelihood of that state.
- the layer for outputting feature values corresponding to the plurality of states may be a final layer of the neural network or may be a layer immediately before the final layer.
- the immediately preceding layer outputting feature values corresponding to the plurality of states outputs feature values corresponding to a plurality of states (for example, the four states described above), and the final layer is the most of the feature values of the previous layer.
- a state having a large value for example, one of normal and Gleason patterns 3, 4, and 5 may be output as an output value.
- the final layer may be a layer for outputting the plurality of states. This may be easily changed according to an embodiment of designing a neural network.
- the learning data of the neural network may be labeled as a plurality of states rather than being labeled with any one of the plurality of states.
- the unit unit is any one of a plurality of states. There may be cases where there is a possibility of being classified into a first state or a second state rather than being distinguishable.
- the unit of the biometric image used as the training data may not be clearly classified into any one state (class), but there may be a possibility of having a plurality of states, that is, an ambiguous case.
- the unit unit is selected as a selected state by selecting any one of a plurality of states, and the neural network is learned according to the labeling result.
- the ambiguity of the image characteristic itself may have useful information, and the image characteristic itself has ambiguity rather than being predicted and labeled in any one likely class (state).
- the neural network can be trained to have higher accuracy than conventional methods.
- the expert who performs the labeling judges that a particular unit unit is likely to be classified as a first state (eg, normal), but also slightly to be judged as a second state (eg, Gleason score 3). Can be.
- a first state eg, normal
- a second state eg, Gleason score 3
- the unit unit which can be clearly classified into this case and the first state (eg, normal), may be a case where there is a difference in any slight difference in the image characteristic. However, if the labeling is simply performed in one state (eg, normal) that has a high probability in the conventional manner, this may result in the above minute difference being ignored.
- a unit unit having such ambiguity is itself (that is, the first state and the first state determined to have a higher probability, but lower probability than the first state, but may also be the second state). State) and may be labeled so as to be distinguished as compared to being labeled as either the first state or the second state.
- this method that is, a method of labeling a plurality of states (in this specification, a case of labeling in two states is described as an example, and thus referred to as 'dual class labeling'). Labeling itself can be reflected in learning to improve the performance of neural networks.
- a particular unit unit may be labeled with a primary state, that is, a state determined to have a higher probability and a secondary state. That is, unit units with ambiguous (confusing) image characteristics may be allowed to be labeled as such dual classes.
- the loss function defining the neural network may be defined to reflect both the primary state and the secondary state. For example, if the neural network outputs a primary state with a high probability, it is defined to have a high compensation (low loss), and if the neural network outputs a secondary state with a high probability, it is defined to have a low compensation (high loss). Can be.
- the loss function can be defined so that the loss of the secondary state is reduced only when the feature value is output with a higher probability than the state other than the primary state or the secondary state.
- the neural network is trained to have the lowest loss, so it is trained to output the primary state with a higher probability for dual-class labeled unit units, but the secondary state may be in different states even if it has a lower probability than other primary states. In comparison, it can be learned to have feature values corresponding to a high probability.
- the confused state itself is reflected in the learning so that the confused image characteristic is not ignored but the confused image characteristic is reflected in the neural network diagnosis result.
- the diagnostic system 100 may include at least one client (eg, 20 and 20-1) that may be connected to the server 10. It can also communicate with.
- the client eg, 20, 20-1
- the diagnostic system 100 may transmit a biometric image to the diagnostic system 100, and the diagnostic system 100 may perform a diagnosis according to the technical spirit of the present invention with respect to the transmitted bioimage. Can be done.
- the diagnosis result may be transmitted to the client (eg, 20, 20-1).
- the diagnosis system 100 may perform diagnosis using a neural network according to the spirit of the present invention.
- a process of learning the neural network may be first performed through learning data labeled with any one of a plurality of state channels or dual classes.
- the diagnosis system 100 may be a system for performing diagnosis by receiving a neural network learned according to the technical spirit of the present invention and a program for performing diagnosis using the neural network from an external source. It may be a system that even performs learning.
- the diagnostic system 100 may be implemented as a dedicated device manufactured to implement the technical idea of the present invention, rather than a general data processing device. In this case, a means for scanning a biological image may be further provided. It may be.
- the neural network receives information (eg, RGB 3 channel information) of a predetermined unit unit (eg, pixel or patch), and each of a plurality of pieces of state information about a predetermined disease (eg, prostate cancer) that the unit unit has. May be a feature value or a network outputting one of the most probable states among the plurality of state information.
- information eg, RGB 3 channel information
- a predetermined unit unit eg, pixel or patch
- a predetermined disease eg, prostate cancer
- the neural network may use a well-known Resnet neural network and a deeplab model.
- the neural network is disclosed in the Korean patent application (Application No. 10-2016-0168176, the diagnosis system of the disease using the neural network and a method therefor, filed below) It may be a neural network using both a micro network and a macro network at the same time.
- the micro network may receive a unit unit, and the macro network may receive a peripheral unit unit including the unit unit to output status information related to the unit unit.
- the micro network may be defined to receive a specific pixel, and the macro network may receive predetermined input data including up to neighboring pixels of the specific pixel to output a state of the specific pixel. have.
- not only one unit unit but also neighboring unit units may affect whether or not the disease unit is judged to be more accurate learning may be possible.
- the accuracy can be improved to a very significant level for the diagnosis of a disease that should consider not only the living tissue but also the surrounding tissue of the living tissue for the diagnosis of the living tissue corresponding to a specific unit unit.
- the biological image when dividing the biological image into a plurality of unit units, it may have a strong effect on the effect of the diagnostic results that may occur depending on the division method of the patch or the location of the divided region.
- the diagnostic system 100 for implementing such technical idea may logically have a configuration as shown in FIG. 2.
- FIG. 2 is a view for explaining the logical configuration of a diagnosis system for diseases using a neural network according to an embodiment of the present invention.
- the diagnostic system 100 includes a control module 110 and a neural network module 120 in which a neural network is stored.
- the diagnostic system 100 may further include a preprocessing module 130.
- the diagnostic system 100 may refer to a logical configuration having hardware resources and / or software necessary for implementing the technical idea of the present invention, and necessarily means one physical component or one It does not mean a device. That is, the diagnostic system 100 may mean a logical combination of hardware and / or software provided to implement the technical idea of the present invention. If necessary, the diagnostic system 100 may be installed in devices spaced apart from each other to perform respective functions. As a result, it may be implemented as a set of logical configurations for implementing the technical idea of the present invention. In addition, the diagnostic system 100 may refer to a set of components separately implemented for each function or role for implementing the technical idea of the present invention.
- each of the control module 110, the neural network module 120, and / or the preprocessing module 130 may be located in different physical devices or in the same physical device.
- a combination of software and / or hardware configuring each of the control module 110, the neural network module 120, and / or the preprocessing module 130 may also be located on different physical devices, and Configurations located in different physical devices may be organically coupled to each other to implement each of the modules.
- module in the present specification may mean a functional and structural combination of hardware for performing 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 performing the predetermined code, and does not necessarily mean a physically connected code or a kind of hardware. Can be easily inferred by the average expert in the art.
- the control module 110 controls other components (eg, the neural network module 120 and / or the preprocessing module 130, etc.) included in the diagnostic system 100 to implement the technical idea of the present invention. can do.
- control module 110 may perform a diagnosis according to the spirit of the present invention by using a neural network stored in the neural network module 120.
- Performing the diagnosis may mean outputting a channel value (feature value) of at least one channel defined in the output layer as described above.
- Each feature value may correspond to a probability that a unit unit to be diagnosed corresponds to information defined by a corresponding channel.
- the neural network module 120 may store a neural network.
- the neural network may refer to a set of information representing a series of design items defining a neural network.
- the neural network may be a convolutional neural network.
- a well-known deeplab neural network may be used as described above, or the neural network disclosed in the previous application may be used.
- the neural network may define a loss function that enables learning using dual class labeled learning data.
- the neural network is trained so that the losses defined by the Loss function can be minimized.
- a unit unit when labeled as a dual class, that is, it may be labeled with a plurality of states, wherein the plurality of states may include a first state and a state of higher probability.
- the loss function When labeled to include a second state that is determined to have a lower state than the first state, the loss function may be applied to the first feature value corresponding to the first state and the second state of the feature values corresponding to the plurality of states. It can be defined that all of the corresponding second feature values are reflected in the loss.
- the neural network may be defined such that the first feature value corresponding to the unit unit has a greater influence than the second feature value.
- the feature value corresponding to the remaining state may be defined such that the higher the value, the greater the loss.
- the loss function may be defined to improve the conventional cross entropy, that is, the second feature value is also reflected in the loss.
- Loss function according to an embodiment of the present invention may be as shown in Equation 1 below.
- the feature value corresponding to the first state Is the first feature value, May be the second feature value.
- And p may have a value between 0 and 1.
- the neural network when a specific unit unit is input, the neural network includes a predetermined layer that outputs a feature value of each of a plurality of predefined state channels (eg, normal, Gleason score 3, Gleason score 4, and Gleason score 5). can do.
- the layer may be designed as a final layer or may be a layer immediately before the final layer.
- the specific unit unit may be a unit unit labeled with a primary state and a secondary state, that is, a dual class labeling.
- the larger the feature value of the primary state that is, the larger the feature value corresponding to the primary state among the feature values output from the neural network.
- the larger the feature value corresponding to the secondary state The value of becomes large, and as a result, the overall loss can be made small.
- the feature value of the primary state has a value smaller than 1, compared to the feature value of the secondary state, the loss has a greater effect on reducing the loss.
- the neural network has a large value of the primary state and a relatively large value of the secondary state (although it has a smaller value than the primary state) due to the loss function as described above, and the remaining state is a small value. Can be learned to have.
- the result may reflect the result of dual class labeling labeled by a real expert.
- Confusion matrices are matrices of the number of things whose predicted value is 'B' for the correct answer to 'A' at the position of (A, B). Performance measurements of segmentation, such as pixel-by-pixel accuracy and section over union (IoU), are based on this fusion matrix.
- IoU section over union
- p may be a value determined from 0 to 1.
- control module 110 may input input data into a neural network stored in the neural network module 120, that is, a learned neural network.
- the output data may be output by performing operations defined by the neural network.
- the preprocessing module 130 may perform preprocessing of the necessary bioimage before performing diagnosis using a neural network.
- the preprocessing of the biological image may include dividing the biological image into unit units having a predefined size. If necessary, the preprocessing module 130 may not be necessary if the unit is a pixel unit, and in addition to the division into the unit, the preprocessing module 130 may be performed by various preprocessing modules 130 for neural network learning as needed. It will be readily apparent to the average expert in the art.
- each of a predetermined state e.g., normal, Gleason score 3, Gleason score 4, and Gleason score 5
- a predetermined state e.g., normal, Gleason score 3, Gleason score 4, and Gleason score 5
- It may include a layer that outputs a feature value.
- FIG. 4 An example conceptually illustrating such a neural network may be as shown in FIG. 4.
- FIG. 4 is a diagram illustrating a concept of a neural network according to an embodiment of the present invention.
- the neural network 300 may receive a predetermined unit unit (eg, a pixel, a patch, etc.) as shown in FIG. 4.
- a predetermined unit unit e.g, a pixel, a patch, etc.
- FIG. 4 illustrates that a patch in which a biological image is divided into a predetermined size (for example, 512 by 512) is input, a pixel value (for example, an RGB three channel value) may be input as input data for each patch unit.
- a predetermined unit unit eg, a pixel, a patch, etc.
- a predetermined skilled person may mark a specific area from a bio-image using a predetermined tool and perform labeling on the marked area. Then, the pixels included in the labeled area may have a labeling value in the corresponding area.
- labeling may be performed for each patch.
- the neural network 300 may be a known deeplab model as described above, or may be a neural network disclosed in a previous application. Or another neural network.
- the neural network 300 may include a predetermined layer (eg, 310).
- the layer eg, 310) may be a layer that outputs feature values corresponding to a plurality of states for the unit unit.
- the layer may output feature values corresponding to four channels, respectively, each feature value being in a plurality of predetermined states (eg, normal, Gleason score 3, Gleason score 4, and Gleason score 5). It may be a value corresponding to a corresponding probability.
- the layer (eg, 310) may be a final layer, and there may be a final layer (eg, 320) immediately after the layer 310.
- the final layer (eg, 320) may be a layer that outputs a state having the largest value among feature values of the layer (eg, 310).
- the output value output by the layer may be a result of learning that the loss is minimized by the loss function defined in the neural network 300.
- the loss function is smaller for a specific unit unit labeled as a dual class, and smaller for a primary state feature, and smaller for a primary state but larger for a secondary state feature, and smaller for the remaining state.
- Equation 1 Such an example may be Equation 1 as described above, but is not limited thereto, and embodiments of other loss functions may be possible.
- the confused unit may reflect the state of confusion in the learning itself, resulting in an overall improvement in performance.
- the type of disease diagnosed by the diagnosis system 100 is diagnosed as an example of prostate cancer is described as an example, an average expert in the art of the present disclosure may not easily apply the technical idea of the present invention to the prostate cancer. I can reason about it.
- the unit unit is a pixel
- the corresponding pixel when each pixel has a predetermined determination value based on the determination value of the pixel unit (eg, Gleason score 3 to 5), the corresponding pixel is It may be determined by the disease pixel corresponding to the disease.
- the disease pixel may be distinguished from other pixels in the biological image based on the determination result.
- the diagnosis may be made in units of pixels, and in this case, only a part where the disease is expressed in the biological image may be segmented.
- One such example may be as shown in FIG. 5.
- FIG. 5 is a diagram exemplarily illustrating a segmentation result according to an exemplary embodiment of the present invention.
- the left portion of FIG. 5 may represent a bioimage labeled by a trained expert, and each labeled region may represent a region in which the disease is expressed.
- FIG. 5 represents a result of performing segmentation on a pixel-by-pixel basis by expressing a pixel, which is determined to have a disease on a pixel-by-pixel basis, differently from other regions (for example, in yellow) according to the spirit of the present invention.
- the disease diagnosis method supporting dual classes may be implemented as computer readable codes on a computer readable recording medium.
- Computer-readable recording media include all kinds of recording devices that store data that can be read by a computer system. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, hard disk, floppy disk, optical data storage, and the like.
- the computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion. And functional programs, codes and code segments for implementing the present invention can be easily inferred by programmers in the art to which the present invention belongs.
- the present invention can be used in "a disease diagnosis system and method supporting dual classes".
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
Description
Claims (12)
- 프로세서 및 뉴럴 네트워크를 저장하는 저장장치를 포함하는 시스템에 구현되며 생체이미지와 상기 뉴럴 네트워크를 이용한 질병의 진단 시스템에 있어서,상기 프로세서는,상기 저장장치에 저장된 상기 뉴럴 네트워크를 소정의 로스 펑션과 상기 생체이미지에 포함된 소정의 단위 유닛에 대해 상기 단위 유닛이 복수의 상태들 중 적어도 하나의 상태를 가질 것으로 라벨링된 학습 데이터를 이용하여, 상기 복수의 상태들 중 적어도 하나의 상태를 가질 확률에 대응하는 판단 값을 출력하도록 학습하며,상기 뉴렬 네트워크는,상기 단위 유닛이 복수의 상태들 각각으로 판단될 확률에 대응되는 복수의 피쳐 값들을 출력하는 특정 레이어를 포함하고,상기 로스 펑션은,다수의 단위 유닛들 중 상기 복수의 상태-상기 복수의 상태는 보다 높은 확률의 상태를 가질 것으로 판단되는 제1상태 및 제1상태보다는 낮은 상태 가질 것으로 판단되는 제2상태를 포함함-로 라벨링된 듀얼 라벨링 단위 유닛에 대해, 상기 특정 레이어의 피쳐 값들 중 제1상태에 상응하는 제1피쳐 값과 상기 제2상태에 상응하는 제2피쳐 값을 모두 반영하여 로스를 연산하도록 정의되는 것을 특징으로 하는 듀얼 클래스를 지원하는 질병 진단 시스템.
- 제1항에 있어서, 상기 로스 펑션은,상기 제1피쳐 값에 비해 상기 제2피쳐 값이 보다 많은 로스를 가지도록 정의되는 것을 특징으로 하는 듀얼 클래스를 지원하는 질병 진단 시스템.
- 제1항에 있어서, 상기 단위 유닛은,상기 생체이미지의 픽셀 단위인 것을 특징으로 하는 듀얼 클래스를 지원하는 질병 진단 시스템.
- 제4항에 있어서, 상기 프로세서는,상기 픽셀 단위의 판단 값에 기초하여 각각의 픽셀이 미리 정해진 판단 값을 가지는 경우 해당 픽셀은 질병에 해당하는 질병 픽셀로 판단하고, 판단결과에 기초하여 상기 생체이미지에서 질병 픽셀을 다른 픽셀과 구분하여 표시하는 것을 특징으로 하는 듀얼 클래스를 지원하는 질병 진단 시스템.
- 제1항에 있어서, 상기 질병은,전립선 암인 것을 특징으로 하는 뉴럴 네트워크를 이용한 질병의 진단 시스템.
- 제6항에 있어서, 상기 복수의 상태들은,정상, 글리슨 스코어 3, 글리슨 스코어 4, 및 글리슨 스코어 5를 포함하는 것을 특징으로 하는 듀얼 클래스를 지원하는 질병 진단 시스템.
- 프로세서 및 저장장치를 포함하는 시스템에 구현되며 생체이미지와 뉴럴 네트워크를 이용한 질병의 진단 시스템이 수행하는 방법에 있어서,상기 저장장치에 저장된 상기 뉴럴 네트워크를 소정의 로스 펑션과 상기 생체이미지에 포함된 소정의 단위 유닛에 대해 상기 단위 유닛이 복수의 상태들 중 적어도 하나의 상태를 가질 것으로 라벨링된 학습 데이터를 입력받는 단계;입력받은 학습 데이터를 이용하여 각각의 단위 유닛에 대해 상기 복수의 상태들 중 적어도 하나의 상태를 가질 확률에 대응하는 판단 값을 출력하도록 학습하는 단계를 포함하며,상기 뉴렬 네트워크는,상기 단위 유닛이 복수의 상태들 각각으로 판단될 확률에 대응되는 복수의 피쳐 값들을 출력하는 특정 레이어를 포함하고,상기 로스 펑션은,다수의 단위 유닛들 중 상기 복수의 상태-상기 복수의 상태는 보다 높은 확률의 상태를 가질 것으로 판단되는 제1상태 및 제1상태보다는 낮은 상태 가질 것으로 판단되는 제2상태를 포함함-로 라벨링된 듀얼 라벨링 단위 유닛에 대해, 상기 특정 레이어의 피쳐 값들 중 제1상태에 상응하는 제1피쳐 값과 상기 제2상태에 상응하는 제2피쳐 값을 모두 반영하여 로스를 연산하도록 정의되는 것을 특징으로 하는 방법.
- 제8항에 있어서, 상기 로스 펑션은,상기 제1피쳐 값에 비해 상기 제2피쳐 값이 보다 많은 로스를 가지도록 정의되는 것을 특징으로 하는 듀얼 클래스를 지원하는 질병 진단 시스템.
- 제8항에 있어서, 상기 방법은,상기 픽셀 단위의 판단 값에 기초하여 각각의 픽셀이 미리 정해진 판단 값을 가지는 경우 해당 픽셀은 질병에 해당하는 질병 픽셀로 판단하는 단계; 및판단결과에 기초하여 상기 생체이미지에서 질병 픽셀을 다른 픽셀과 구분하여 표시하는 단계를 더 포함하는 방법.
- 데이터 처리장치에 설치되며 제8항 내지 제11항 중 어느 한 항에 기재된 방법을 수행하기 위한 매체에 기록된 컴퓨터 프로그램.
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201980037858.6A CN112236829A (zh) | 2018-06-04 | 2019-06-04 | 一种支持双重类别的疾病诊断***及其方法 |
JP2020566818A JP7109815B2 (ja) | 2018-06-04 | 2019-06-04 | デュアルクラスに対応できる疾病診断システム及び該方法 |
US16/972,231 US12009099B2 (en) | 2018-06-04 | 2019-06-04 | Disease diagnosis system for supporting dual class, and method therefor |
EP19815441.1A EP3799070A4 (en) | 2018-06-04 | 2019-06-04 | DISEASE DIAGNOSTIC SYSTEM FOR SUPPORTING A DUAL CLASS AND ASSOCIATED METHOD |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR10-2018-0064331 | 2018-06-04 | ||
KR1020180064331A KR102162895B1 (ko) | 2018-06-04 | 2018-06-04 | 듀얼 클래스를 지원하는 질병 진단 시스템 및 그 방법 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2019235827A1 true WO2019235827A1 (ko) | 2019-12-12 |
Family
ID=68770904
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/KR2019/006758 WO2019235827A1 (ko) | 2018-06-04 | 2019-06-04 | 듀얼 클래스를 지원하는 질병 진단 시스템 및 그 방법 |
Country Status (6)
Country | Link |
---|---|
US (1) | US12009099B2 (ko) |
EP (1) | EP3799070A4 (ko) |
JP (1) | JP7109815B2 (ko) |
KR (1) | KR102162895B1 (ko) |
CN (1) | CN112236829A (ko) |
WO (1) | WO2019235827A1 (ko) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR102215269B1 (ko) * | 2018-08-07 | 2021-02-15 | 주식회사 딥바이오 | 진단 결과 생성 시스템 및 방법 |
EP3608701A1 (de) * | 2018-08-09 | 2020-02-12 | Olympus Soft Imaging Solutions GmbH | Verfahren zur bereitstellung wenigstens einer auswertemethode für proben |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7467119B2 (en) * | 2003-07-21 | 2008-12-16 | Aureon Laboratories, Inc. | Systems and methods for treating, diagnosing and predicting the occurrence of a medical condition |
KR20140042531A (ko) * | 2012-09-28 | 2014-04-07 | 삼성전자주식회사 | 카테고리별 진단 모델을 이용한 병변 진단 장치 및 방법 |
US20150213302A1 (en) * | 2014-01-30 | 2015-07-30 | Case Western Reserve University | Automatic Detection Of Mitosis Using Handcrafted And Convolutional Neural Network Features |
KR20160034814A (ko) | 2014-09-22 | 2016-03-30 | 삼성전자주식회사 | 뉴럴 네트워크를 수반한 클라이언트 장치 및 그것을 포함하는 시스템 |
KR20160037022A (ko) * | 2014-09-26 | 2016-04-05 | 삼성전자주식회사 | 부스트 풀링 뉴럴 네트워크 기반의 데이터 분류 장치 및 그 데이터 분류 장치를 위한 뉴럴 네트워크 학습 방법 |
KR20180021635A (ko) * | 2016-08-22 | 2018-03-05 | 한국과학기술원 | 3차원 의료 영상에서 깊이 방향 재귀 학습을 이용하는 병변 특징 표현 분석 방법 및 시스템 |
Family Cites Families (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040199482A1 (en) * | 2002-04-15 | 2004-10-07 | Wilson Scott B. | Systems and methods for automatic and incremental learning of patient states from biomedical signals |
US7761240B2 (en) | 2004-08-11 | 2010-07-20 | Aureon Laboratories, Inc. | Systems and methods for automated diagnosis and grading of tissue images |
US7599893B2 (en) | 2005-10-13 | 2009-10-06 | Aureon Laboratories, Inc. | Methods and systems for feature selection in machine learning based on feature contribution and model fitness |
ES2795036T3 (es) * | 2008-07-25 | 2020-11-20 | Fund D Anna Sommer Champalimaud E Dr Carlos Montez Champalimaud | Sistemas y métodos para tratar, diagnosticar y predecir la aparición de una afección médica |
JP6654634B2 (ja) | 2014-12-03 | 2020-02-26 | ベンタナ メディカル システムズ, インコーポレイテッド | 不均一なバイオマーカー分布を定量的に分析するための方法、システム及び装置 |
WO2016094330A2 (en) * | 2014-12-08 | 2016-06-16 | 20/20 Genesystems, Inc | Methods and machine learning systems for predicting the liklihood or risk of having cancer |
JP6644075B2 (ja) | 2015-08-19 | 2020-02-12 | 興和株式会社 | 画像処理装置、画像処理方法、及び画像処理プログラム |
US10339650B2 (en) * | 2016-01-07 | 2019-07-02 | Koios Medical, Inc. | Method and means of CAD system personalization to reduce intraoperator and interoperator variation |
US10572996B2 (en) | 2016-06-28 | 2020-02-25 | Contextvision Ab | Method and system for detecting pathological anomalies in a digital pathology image and method for annotating a tissue slide |
US10991093B2 (en) * | 2016-09-21 | 2021-04-27 | The General Hospital Corporation | Systems, methods and media for automatically generating a bone age assessment from a radiograph |
JP6965343B2 (ja) | 2016-10-31 | 2021-11-10 | コニカ ミノルタ ラボラトリー ユー.エス.エー.,インコーポレイテッド | 制御フィードバックを用いる画像セグメンテーションの方法及びシステム |
KR101869438B1 (ko) * | 2016-11-22 | 2018-06-20 | 네이버 주식회사 | 딥 러닝을 이용하여 환자의 진단 이력으로부터 질병 예후를 예측하는 방법 및 시스템 |
KR101944536B1 (ko) | 2016-12-11 | 2019-02-01 | 주식회사 딥바이오 | 뉴럴 네트워크를 이용한 질병의 진단 시스템 및 그 방법 |
US20190205758A1 (en) * | 2016-12-30 | 2019-07-04 | Konica Minolta Laboratory U.S.A., Inc. | Gland segmentation with deeply-supervised multi-level deconvolution networks |
US10489908B2 (en) * | 2017-02-22 | 2019-11-26 | Siemens Healthcare Gmbh | Deep convolutional encoder-decoder for prostate cancer detection and classification |
US10977843B2 (en) * | 2017-06-28 | 2021-04-13 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for determining parameters for medical image processing |
CN107492099B (zh) * | 2017-08-28 | 2021-08-20 | 京东方科技集团股份有限公司 | 医学图像分析方法、医学图像分析***以及存储介质 |
US10650286B2 (en) * | 2017-09-07 | 2020-05-12 | International Business Machines Corporation | Classifying medical images using deep convolution neural network (CNN) architecture |
US10592779B2 (en) * | 2017-12-21 | 2020-03-17 | International Business Machines Corporation | Generative adversarial network medical image generation for training of a classifier |
US10540578B2 (en) * | 2017-12-21 | 2020-01-21 | International Business Machines Corporation | Adapting a generative adversarial network to new data sources for image classification |
US10937540B2 (en) * | 2017-12-21 | 2021-03-02 | International Business Machines Coporation | Medical image classification based on a generative adversarial network trained discriminator |
US11256973B2 (en) * | 2018-02-05 | 2022-02-22 | Nuance Communications, Inc. | Neural network with embedded filter layer |
-
2018
- 2018-06-04 KR KR1020180064331A patent/KR102162895B1/ko active IP Right Grant
-
2019
- 2019-06-04 CN CN201980037858.6A patent/CN112236829A/zh active Pending
- 2019-06-04 EP EP19815441.1A patent/EP3799070A4/en active Pending
- 2019-06-04 US US16/972,231 patent/US12009099B2/en active Active
- 2019-06-04 JP JP2020566818A patent/JP7109815B2/ja active Active
- 2019-06-04 WO PCT/KR2019/006758 patent/WO2019235827A1/ko unknown
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7467119B2 (en) * | 2003-07-21 | 2008-12-16 | Aureon Laboratories, Inc. | Systems and methods for treating, diagnosing and predicting the occurrence of a medical condition |
KR20140042531A (ko) * | 2012-09-28 | 2014-04-07 | 삼성전자주식회사 | 카테고리별 진단 모델을 이용한 병변 진단 장치 및 방법 |
US20150213302A1 (en) * | 2014-01-30 | 2015-07-30 | Case Western Reserve University | Automatic Detection Of Mitosis Using Handcrafted And Convolutional Neural Network Features |
KR20160034814A (ko) | 2014-09-22 | 2016-03-30 | 삼성전자주식회사 | 뉴럴 네트워크를 수반한 클라이언트 장치 및 그것을 포함하는 시스템 |
KR20160037022A (ko) * | 2014-09-26 | 2016-04-05 | 삼성전자주식회사 | 부스트 풀링 뉴럴 네트워크 기반의 데이터 분류 장치 및 그 데이터 분류 장치를 위한 뉴럴 네트워크 학습 방법 |
KR20180021635A (ko) * | 2016-08-22 | 2018-03-05 | 한국과학기술원 | 3차원 의료 영상에서 깊이 방향 재귀 학습을 이용하는 병변 특징 표현 분석 방법 및 시스템 |
Also Published As
Publication number | Publication date |
---|---|
KR102162895B1 (ko) | 2020-10-07 |
JP2021526261A (ja) | 2021-09-30 |
US12009099B2 (en) | 2024-06-11 |
US20210142900A1 (en) | 2021-05-13 |
EP3799070A4 (en) | 2022-03-02 |
KR20190143509A (ko) | 2019-12-31 |
CN112236829A (zh) | 2021-01-15 |
JP7109815B2 (ja) | 2022-08-01 |
EP3799070A1 (en) | 2021-03-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2018106005A1 (ko) | 뉴럴 네트워크를 이용한 질병의 진단 시스템 및 그 방법 | |
WO2019235828A1 (ko) | 투 페이스 질병 진단 시스템 및 그 방법 | |
WO2020032559A2 (ko) | 뉴럴 네트워크를 이용한 질병의 진단 시스템 및 방법 | |
WO2020045848A1 (ko) | 세그멘테이션을 수행하는 뉴럴 네트워크를 이용한 질병 진단 시스템 및 방법 | |
WO2019132168A1 (ko) | 수술영상데이터 학습시스템 | |
WO2020111754A9 (ko) | 세미 슈퍼바이즈드 학습을 이용한 진단 시스템 제공방법 및 이를 이용하는 진단 시스템 | |
WO2021060899A1 (ko) | 인공지능 모델을 사용 기관에 특화시키는 학습 방법, 이를 수행하는 장치 | |
WO2021182889A2 (ko) | 영상 기반의 안질환 진단 장치 및 방법 | |
WO2020032562A2 (ko) | 생체 이미지 진단 시스템, 생체 이미지 진단 방법, 및 이를 수행하기 위한 단말 | |
WO2019235827A1 (ko) | 듀얼 클래스를 지원하는 질병 진단 시스템 및 그 방법 | |
WO2021025458A1 (ko) | 멀티미디어 정보를 이용한 휴대용 체외진단키트 해석장치 | |
WO2021194056A1 (en) | Method for training deep learning network based on artificial intelligence and learning device using the same | |
WO2021010671A2 (ko) | 뉴럴 네트워크 및 비국소적 블록을 이용하여 세그멘테이션을 수행하는 질병 진단 시스템 및 방법 | |
WO2020101457A2 (ko) | 지도학습기반의 합의 진단방법 및 그 시스템 | |
WO2022146050A1 (ko) | 우울증 진단을 위한 인공지능 연합학습 방법 및 시스템 | |
WO2020032561A2 (ko) | 다중 색 모델 및 뉴럴 네트워크를 이용한 질병 진단 시스템 및 방법 | |
WO2022131642A1 (ko) | 의료 영상 기반 질환 중증도 결정 장치 및 방법 | |
WO2022197044A1 (ko) | 뉴럴 네트워크를 이용한 방광병변 진단 방법 및 그 시스템 | |
WO2022092993A1 (ko) | 대상 이미지에 대한 추론 작업을 수행하는 방법 및 시스템 | |
WO2020032560A2 (ko) | 진단 결과 생성 시스템 및 방법 | |
WO2024014789A1 (ko) | 차트축감지방법 및 장치 | |
WO2015126058A1 (ko) | 암 예후 예측 방법 | |
WO2022149658A1 (ko) | 뉴럴 네트워크의 출력 해석 방법 및 이를 위한 시스템 | |
WO2022158843A1 (ko) | 조직 검체 이미지 정제 방법, 및 이를 수행하는 컴퓨팅 시스템 | |
WO2023167448A1 (ko) | 병리 슬라이드 이미지를 분석하는 방법 및 장치 |
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: 19815441 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2020566818 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2019815441 Country of ref document: EP Effective date: 20201130 |
|
ENP | Entry into the national phase |
Ref document number: 2019815441 Country of ref document: EP Effective date: 20201130 |