WO2022197044A1 - 뉴럴 네트워크를 이용한 방광병변 진단 방법 및 그 시스템 - Google Patents
뉴럴 네트워크를 이용한 방광병변 진단 방법 및 그 시스템 Download PDFInfo
- Publication number
- WO2022197044A1 WO2022197044A1 PCT/KR2022/003546 KR2022003546W WO2022197044A1 WO 2022197044 A1 WO2022197044 A1 WO 2022197044A1 KR 2022003546 W KR2022003546 W KR 2022003546W WO 2022197044 A1 WO2022197044 A1 WO 2022197044A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- bladder
- lesion
- neural network
- lesions
- unit
- Prior art date
Links
- 230000003902 lesion Effects 0.000 title claims abstract description 239
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 140
- 238000003745 diagnosis Methods 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000001575 pathological effect Effects 0.000 claims abstract description 29
- 230000007170 pathology Effects 0.000 claims description 54
- 206010005003 Bladder cancer Diseases 0.000 claims description 15
- 206010044412 transitional cell carcinoma Diseases 0.000 claims description 12
- 208000023747 urothelial carcinoma Diseases 0.000 claims description 12
- 208000009458 Carcinoma in Situ Diseases 0.000 claims description 10
- 201000004933 in situ carcinoma Diseases 0.000 claims description 10
- 201000004686 non-invasive bladder papillary urothelial neoplasm Diseases 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 7
- 201000011177 bladder transitional cell papilloma Diseases 0.000 claims description 6
- 208000012477 urothelial papilloma Diseases 0.000 claims description 6
- 238000011065 in-situ storage Methods 0.000 claims description 5
- 230000003211 malignant effect Effects 0.000 claims description 5
- 230000035755 proliferation Effects 0.000 claims description 5
- 208000019515 urothelial dysplasia Diseases 0.000 claims description 5
- 208000026572 inverted urothelial papilloma Diseases 0.000 claims description 4
- 238000002271 resection Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 2
- 238000003860 storage Methods 0.000 description 14
- 238000009799 cystectomy Methods 0.000 description 13
- 208000007097 Urinary Bladder Neoplasms Diseases 0.000 description 12
- 201000005112 urinary bladder cancer Diseases 0.000 description 12
- 238000013527 convolutional neural network Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 7
- 238000007781 pre-processing Methods 0.000 description 7
- 238000013461 design Methods 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 6
- 238000011176 pooling Methods 0.000 description 5
- 238000012549 training Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 206010028980 Neoplasm Diseases 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 201000011510 cancer Diseases 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002062 proliferating effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Images
Classifications
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- 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
- G16H50/20—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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06T2207/20081—Training; Learning
-
- 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]
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Definitions
- the present invention relates to a method and system for diagnosing bladder lesions using a neural network, and more particularly, to a diagnostic system capable of diagnosing a plurality of bladder lesions using pathological images based on biological tissues of the bladder, effectively learning and diagnosing them. It relates to methods and systems that can be used.
- Bladder cancer is one of the most common cancers and has a high chance of recurrence.
- Transurethral resection of bladder (TURB) is mainly used as a diagnostic and local treatment method for bladder cancer.
- the diagnosis of bladder cancer is made by pathologically reading the tissue sample obtained through transurethral cystectomy, and the diagnosis of bladder cancer is made.
- the tissue sample obtained through transurethral cystectomy can be used not only for the diagnosis of bladder cancer, but also for diagnosing various lesions related to the bladder.
- tissue samples obtained through transurethral cystectomy there are many types of lesions that can be identified in tissue samples obtained through transurethral cystectomy, and can be largely divided into invasive cancer lesions and noninvasive lesions.
- noninvasive lesions include cancer lesions such as low/high grade noninvasive papillary urothelial carcinoma and urothelial carcinoma in situ (CIS), papillary urothelial neoplasm of low malignant potential (PUNLMP), urothelial proliferation of unknown malignant potential (UPUMP), and urothelial papilloma.
- cancer lesions such as low/high grade noninvasive papillary urothelial carcinoma and urothelial carcinoma in situ (CIS), papillary urothelial neoplasm of low malignant potential (PUNLMP), urothelial proliferation of unknown malignant potential (UPUMP), and urothelial papilloma.
- PUNLMP papillary urothelial neoplasm of low malignant potential
- UPUMP urothelial proliferation of unknown malignant potential
- proliferative lesions such as inverted urothelial papilloma and urothelial dysplasia.
- tissue sample for transurethral cystectomy since the size of the tissue sample for transurethral cystectomy is relatively large, it is difficult to find and determine the type of bladder cancer by examining a sample in the form of a glass slide with an optical microscope or reading a sample in the form of a pathological image with a monitor. It can be a tiring and arduous task for a specialist, and this demand is growing even more.
- Non-Patent Document 1 Non-Patent Document: (Antoni2016) Antoni, S. et al., Bladder Cancer Incidence and Mortality: A Global Overview and Recent Trends.
- the technical problem to be achieved by the present invention is to implement a neural network-based diagnostic system that can additionally diagnose various bladder lesions through tissue samples obtained through transurethral cystectomy performed as a diagnostic and local treatment method for bladder cancer.
- a neural network is constructed using learning data annotating the lesion region only for bladder lesions that must be learned by annotating the lesion region, and learning data annotated only the type of lesion for the rest of the bladder lesions.
- the bladder lesion diagnosis system receives a unit pathology image, and the bladder lesion diagnosis system converts the unit pathology image into a first neural network. acquiring a diagnosis result of a first bladder lesion among a plurality of bladder lesions in the unit pathology image by inputting it into a network, and inputting the unit pathology image to a second neural network by the bladder lesion diagnosis system to obtain the unit pathology image obtaining a diagnosis result of a second bladder lesion excluding the first bladder lesion among the plurality of bladder lesions, wherein the first neural network includes a plurality of annotated lesion regions in which the first bladder lesion is expressed. characterized in that it is a neural network learned through the first learning data of 2 The training data is characterized in that it is a neural network learned through the data on which the annotation on the lesion region is not performed.
- the unit pathology image may be a patch image in which a pathology image corresponding to a tissue specimen obtained through transurethral resection of bladder (TURB) is divided into predetermined sizes.
- TURB transurethral resection of bladder
- the first bladder lesion may include a urothelial carcinoma in situ (CIS) lesion.
- CIS urothelial carcinoma in situ
- the second bladder lesion is invasive bladder cancer (invasive urothelial carcinoma), low/high grade noninvasive papillary urothelial carcinoma lesion, papillary urothelial neoplasm of low malignant potential (PUNLMP) lesion, urothelial proliferation of unknown malignant potential (UPUMP) lesion, urothelial papilloma lesion , at least one of an inverted urothelial papilloma lesion and a urothelial dysplasia lesion.
- invasive bladder cancer invasive urothelial carcinoma
- low/high grade noninvasive papillary urothelial carcinoma lesion papillary urothelial neoplasm of low malignant potential (PUNLMP) lesion
- UPUMP urothelial proliferation of unknown malignant potential
- the bladder lesion diagnosis system receives a unit pathology image, the bladder lesion diagnosis system inputs the unit pathology image to a second neural network, and the unit pathology obtaining a diagnosis result of a second bladder lesion excluding the first bladder lesion among a plurality of bladder lesions in an image, and the bladder lesion diagnosis system inputs the unit pathology image to the first neural network to obtain the unit pathology image and obtaining a diagnosis result of the first bladder lesion among the plurality of bladder lesions, wherein the first neural network uses a plurality of first learning data annotated with the lesion region in which the first bladder lesion is expressed.
- the second neural network is a plurality of second learning data annotated with an expression lesion type, which is whether at least one of the second bladder lesions is expressed - the second learning data is in the lesion region. It is characterized in that it is a neural network learned through unannotated data.
- the method may be implemented by a computer program stored in a computer-readable recording medium.
- a bladder lesion diagnosis system using a neural network includes a processor and a storage device in which a program executed by the processor is recorded, and the processor drives the program to produce a unit pathological image.
- a diagnosis result of a first bladder lesion among a plurality of bladder lesions is obtained in the unit pathology image by input to one neural network, and the unit pathology image is input to a second neural network to display the plurality of bladder lesions in the unit pathology image.
- a diagnosis result of a second bladder lesion excluding the first bladder lesion is obtained, and the first neural network is a neural network learned through a plurality of first learning data in which the lesion region in which the first bladder lesion is expressed is annotated.
- the second neural network is a plurality of second learning data annotated with an expression lesion type, which is whether at least one of the second bladder lesions is expressed; It is characterized in that it is a neural network learned through unperformed data.
- transurethral bladder through a neural network-based diagnostic system that can additionally diagnose various bladder lesions through tissue samples obtained through transurethral cystectomy as a local treatment method and diagnosis of bladder cancer. Resection is effective in diagnosing not only bladder cancer but also various bladder lesions.
- a neural network is constructed using learning data annotating the lesion region only for bladder lesions that must be learned by annotating the lesion region, and for the rest of the bladder lesions, only the type of lesion is annotated. Building a neural network using data has the effect of effectively building a diagnostic system.
- FIG. 1 is a diagram for explaining a schematic system for implementing a bladder lesion diagnosis method using a neural network according to the technical idea of the present invention.
- FIG. 2 is a diagram for explaining a logical configuration of a bladder lesion diagnosis system using a neural network according to the technical idea of the present invention.
- FIG. 3 is a diagram for explaining the physical configuration of a bladder lesion diagnosis system using a neural network according to the technical idea of the present invention.
- FIG. 4 is a diagram for explaining the concept of a bladder lesion diagnosis method using a neural network according to the technical idea of the present invention.
- FIG. 5 is a flowchart illustrating a method for diagnosing bladder lesions using a neural network according to the technical concept of the present invention.
- 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. 1 is a diagram for explaining a schematic system for implementing a bladder lesion diagnosis method using a neural network according to the technical idea of the present invention.
- a diagnosis system 100 may be installed in a predetermined server 10 to implement the technical idea of the present invention.
- the server 10 refers to a data processing device having arithmetic capability for implementing the technical idea of the present invention, and generally provides a specific service such as a personal computer, a mobile terminal, etc. as well as a data processing device accessible by a client through a network.
- An average expert in the art of the present invention can easily infer that any device capable of performing can 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 mean an arithmetic device capable of driving the 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 may be performed using a plurality of neural networks (Nerual Networks, 12-2, 12-3) defined by the technical idea of .
- the processor 110 may mean an arithmetic device capable of executing a predetermined program (software code), and may include an implementation example of the data processing device or a vendor mobile processor, microprocessor, CPU, single processor, multiprocessor, It may be named by various names such as GPU, and may be implemented by one or more processors.
- An average expert in the technical field of the present invention can easily infer that the processor 110 can perform data processing necessary for the technical idea of the present invention by driving the program.
- the storage device 120 may mean a device in which a program for implementing the technical idea of the present invention is stored/installed. According to an embodiment, the storage device 120 may be divided into a plurality of different physical devices, and according to an embodiment, a part of the storage device 120 may exist inside the processor 110 .
- the storage device 120 may be implemented as a hard disk, a GPU, a solid state disk (SSD), an optical disk, a random access memory (RAM), and/or other various types of storage media, depending on the embodiment. Accordingly, it may be implemented in a detachable manner in the storage device 120 .
- the processor 11 drives the program 12-1 to perform a plurality of processes through the neural networks 12-2 and 12-3. It may mean outputting whether or not at least one of bladder lesions is expressed and/or an expression region.
- the expression region of a given bladder lesion may be determined in units of pixels, and for this purpose, it is known that a neural network for determining whether each pixel is included in the expression region of a given disease can be learned and utilized. The description will be omitted.
- the storage device 12 may mean a data storage means capable of storing the program 12-1 and the neural networks 12-2 and 12-3, and is implemented as a plurality of storage means according to an embodiment. it might be In addition, the storage device 12 may be meant to include not only the main storage device included in the server 10 , but also a temporary storage device or memory that may be included in the processor 11 .
- diagnosis system 100 is illustrated as being implemented as any one physical device in FIG. 1 or FIG. 3 , a plurality of physical devices are organically combined as necessary to provide the diagnosis system 100 according to the technical spirit of the present invention.
- An average expert in the technical field of the present invention can easily infer that it can be implemented.
- the diagnosis system 100 when it performs diagnosis, it may mean a series of processes of receiving a unit pathology image and outputting the output data defined in this specification.
- the output data may be data indicating the type of the expressed lesion when one or a plurality of bladder lesions are expressed in the unit pathology image as described above.
- the output data may further include not only the type of lesion but also information indicating the region in which the lesion is expressed (eg, whether each pixel is included in the lesion region) as described above.
- the unit pathology image may be a patch image in which a slide image of a tissue sample obtained through transurethral cystectomy is divided into predetermined sizes.
- the size of the patch image may be appropriately determined according to need.
- various bladder lesions can be diagnosed through tissue samples obtained through transurethral cystectomy.
- the method to obtain the most accurate results in developing such a machine learning model is learning by annotating the types and regions of lesions expressed for each pathological image (a slide image or a patch image in which the slide image is divided) of a tissue sample. It may be to secure a large amount of data and learn the secured large amount of learning data through a neural network. That is, by training a machine learning model in a supervised-learning method, a neural network that outputs (diagnosing) whether a bladder lesion is expressed on a pathological image and, if so, what type of bladder lesion is expressed, can be learned. . In addition, as described above, if necessary, the neural network may be trained to output even the region of the expressed bladder lesion.
- tissue sample and its pathological images obtained through transurethral cystectomy are relatively large and there are various types of bladder lesions, it is difficult for a pathologist to annotate the types and areas of lesions for each type of lesion one by one. It is a very difficult task, and accordingly, building a large amount of training data is also very costly and time consuming.
- annotating the expressed lesion region on the pathology image requires relatively large resources.
- a method of annotating only the type of lesion indicating which lesion is expressed in the pathological image without annotating the lesion region in the image may be considered.
- a machine learning model can be developed that determines the lesion region and/or type using predetermined supervised learning methods only with the lesion type information present in the pathological image, and in this case, each pathological image has A number of lesion types that can be identified are annotated, and supervised learning methods using them include semi/unsupervised learning (Xie2019) using consistency loss, etc., noisy label framework (Li2020), reject option (Geifman2019), (min/max/ attention) multiple-instance learning (Ilse2018) may be utilized.
- bladder lesions e.g., low/high grade noninvasive papillary urothelial carcinoma, urothelial carcinoma in situ (CIS), papillary urothelial neoplasm of low malignant potential (PUNLMP), urothelial proliferation of unknown malignant potential (UPUMP), urothelial papilloma, inverted Among the urothelial papilloma, urothelial dysplasia, etc.
- the CIS type of lesion has a characteristic that it is difficult to diagnose with a neural network trained with learning data annotating only the type of lesion on the unit pathology image as described above. This may be because the CIS type lesion is often classified according to the cell type rather than the tissue type, and thus has a characteristic that must be distinguished in a different way from other types of lesion.
- the present invention can provide a technical idea for solving these problems.
- a plurality of first learning data in which various bladder lesions are classified into a first bladder lesion and a second bladder lesion, and both the lesion type and the lesion area are annotated for the lesion classified as the first bladder lesion.
- a neural network that is trained with a plurality of second learning data annotated only by the lesion type, that is, a second neural network, is constructed and the second neural network is used through the second neural network. Diagnosis can be performed.
- the second learning data may be learning data that is not annotated with respect to the aforementioned lesion region.
- the diagnostic performance for bladder lesions as a whole can be guaranteed by annotating the lesion type and lesion area only for limited lesions (eg, CIS lesions) where diagnostic performance is exhibited above a certain level only when the lesion area is additionally annotated due to the characteristics of the lesion. can have an effect.
- limited lesions eg, CIS lesions
- the diagnosis system 100 When the diagnosis system 100 is implemented by being included in a predetermined server 10 , the diagnosis system 100 communicates with at least one client (eg, 20 , 20 - 1 ) connectable to the server 10 . can also be performed.
- the client eg, 20, 20-1
- the diagnosis system 100 views the transmitted pathology image or unit pathology image. Diagnosis according to the technical spirit of the invention can be performed.
- the diagnosis result may be transmitted to the client (Yekerdae, 20, 20-1).
- the server 10 itself may be provided with an interface for receiving the pathological image or unit pathological image.
- the diagnosis system 100 may diagnose various bladder lesions using a neural network according to the technical concept of the present invention. Of course, in order to perform such a diagnosis, the process of training the neural network may be performed first.
- the diagnosis system 100 may be a system that performs diagnosis by receiving a neural network learned according to the technical idea of the present invention and a program for performing diagnosis using the neural network from the outside, or It may be a system that even performs learning.
- the diagnosis system 100 may be implemented as a dedicated device manufactured to implement the technical idea of the present invention rather than a general-purpose data processing device. In this case, means for scanning pathological images may be further provided. may be
- the diagnosis system 100 for implementing this technical idea may logically have the configuration shown in FIG. 2 .
- FIG. 2 is a diagram for explaining a logical configuration of a disease diagnosis system using a neural network according to an embodiment of the present invention.
- the diagnosis system 100 includes a control module 110 and a neural network module 120 in which a neural network is stored.
- the diagnosis system 100 may further include a pre-processing module 130 .
- the diagnostic system 100 may mean a logical configuration including hardware resources and/or software necessary for implementing the technical idea of the present invention, and necessarily means one physical component or one It doesn't mean the device. That is, the diagnosis 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 respective functions By doing so, it may be implemented as a set of logical configurations for implementing the technical idea of the present invention. Also, the diagnosis 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 may be located in the same physical device.
- the combination of software and/or hardware constituting each of the control module 110, the neural network module 120, and/or the preprocessing module 130 is also located in different physical devices, Components located in different physical devices may be organically coupled to each other to implement the respective 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, and does not necessarily mean physically connected code or one type of hardware. can be easily inferred to an average expert in the technical field of the present invention.
- the control module 110 controls other components (eg, the neural network module 120 and/or the preprocessing module 130 ) included in the diagnosis system 100 to implement the technical idea of the present invention. can do.
- control module 110 may perform the diagnosis according to the technical idea of the present invention by using the neural network stored in the neural network module 120 .
- the neural network module 120 may store a plurality of neural networks as described above.
- the neural network may refer to a set of information representing a series of design items defining the neural network.
- the neural network may be a convolutional neural network, but various types of neural networks capable of performing diagnosis by well extracting features based on the lesion type and/or region annotated on the pathological image may be used. An average expert in the technical field of the present invention can easily infer.
- the 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 a sub-sampling layer).
- a convolutional neural network may be designed to include a normalization layer such as a batch normalization (BN) layer.
- BN batch normalization
- a convolutional neural network may be defined by a function, a filter, a stride, a 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 to implement the technical idea of the present invention Separately defined functions may be used.
- 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), Swish, an exponential linear unit (ELU), and 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 trained, 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 control module 110 may sequentially input a unit pathology image, that is, a pathology image to be diagnosed, to a plurality of neural networks stored in the neural network module 120 .
- the unit pathology image may be a patch image in which a slide image is divided into a predetermined size.
- the plurality of neural networks may include a first neural network and a second neural network.
- the first neural network and the second neural network are respectively learned and constructed separately, and information annotated on the learning data used at this time may also be different.
- the types of deep learning models used for the first neural network and the second neural network may also be different from each other.
- the first neural network may be a neural network for diagnosing the first bladder lesion as described above.
- the plurality of first learning data used for learning of the first neural network may be learning data, that is, information in which a lesion region is annotated for each pathological image.
- learning data that is, information in which a lesion region is annotated for each pathological image.
- the types of lesions need to be separately annotated.
- the first bladder lesion may include a carcinoma in situ (CIS) lesion, and if necessary, other lesions may be further included in the first bladder lesion.
- CIS carcinoma in situ
- the second neural network may be a neural network for diagnosing the second bladder lesion as described above.
- the region of the bladder lesion is not annotated for each training data, that is, each pathological image, but only the type of the lesion may be annotated.
- These secondary bladder lesions are invasive bladder cancer (invasive urothelial carcinoma), low/high grade noninvasive papillary urothelial carcinoma, papillary urothelial neoplasm of low malignant potential (PUNLMP), urothelial proliferation of unknown malignant potential (UPUMP), urothelial papilloma, inverted urothelial papilloma , urothelial dysplasia, and if necessary, other lesions may be further included in the second bladder lesion.
- invasive bladder cancer invasive urothelial carcinoma
- PUNLMP papillary urothelial neoplasm of low malignant potential
- UPUMP urothelial proliferation of unknown malignant potential
- urothelial papilloma inverted urothelial papilloma
- urothelial dysplasia if necessary, other lesions may be further included in the second bladder lesion.
- the control module 110 may input the unit pathology image to the first neural network to first diagnose the first bladder lesion.
- the diagnosis of the second bladder lesion may be performed.
- the unit pathology image may be input to the second neural network first, and then the unit pathology image may be input to the first neural network, or the unit pathology image may be input simultaneously to perform diagnosis in parallel.
- the pre-processing module 130 may perform pre-processing of a pathological image necessary before performing a diagnosis using a neural network.
- the pre-processing of the pathological image may include dividing the pathological image into patches of a predefined size, and the present invention may also perform appropriate image processing in a manner suitable for each of the neural networks, if necessary. An average expert in the field of technology can easily infer.
- FIG. 4 is a diagram for explaining the concept of a bladder lesion diagnosis method using a neural network according to the technical idea of the present invention.
- FIG. 5 is a flowchart illustrating a method for diagnosing bladder lesions using a neural network according to the technical idea of the present invention.
- the diagnosis system 100 may receive a plurality of unit pathological images 31 , 31-1 , 31-2 , 31-3 , etc. to be diagnosed ( S100 ). ).
- the unit pathology images (31, 31-1, 31-2, 31-3, etc.) include at least one pathological image (eg, slide image, 30) generated from a tissue sample obtained by transurethral cystectomy. It may be a patch image divided by size.
- Each of the unit pathological images 31, 31-1, 31-2, 31-3, etc. may be input to both the first neural network and the second neural network (S110 and S120). At this time, each of the unit pathology images (31, 31-1, 31-2, 31-3, etc.) is input to the first neural network first, the first bladder lesion is diagnosed, and then the second neural network is input to the second bladder. It goes without saying that the lesions may be diagnosed, the order of which may be changed, or the diagnosis may be performed at the same time.
- the lesion type and/or area for the first bladder lesion through the first neural network and the second neural network When the diagnosis of the lesion type and/or area of the second bladder lesion is performed, the diagnosis result of each of the unit pathology images (31, 31-1, 31-2, 31-3, etc.) is mapped to a slide image. can be (S130).
- a diagnosis result of the slide image that is, information indicating the lesion area and/or type in the slide image may be obtained.
- a neural network-based diagnostic system that can additionally diagnose various bladder lesions through tissue samples obtained through transurethral cystectomy performed as a diagnostic and local treatment method for bladder cancer can be implemented.
- a neural network is constructed using learning data annotating the lesion region only for bladder lesions that must be learned by annotating the lesion region, and learning data annotated only the type of lesion for the rest of the bladder lesions. It has the effect of effectively constructing a diagnostic system by constructing a neural network using
- the method for diagnosing bladder lesions using a neural network can be implemented as computer-readable codes on a computer-readable recording medium.
- the computer-readable recording medium includes all types of recording devices in which data readable by a computer system is stored. Examples of the computer-readable recording medium include ROM, RAM, CD-ROM, magnetic tape, hard disk, floppy disk, and optical data storage device.
- the computer-readable recording medium is distributed in a computer system connected through a network, so that the computer-readable code can be stored and executed in a distributed manner. 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 pertains.
- the present invention can be applied to a method and system for diagnosing bladder lesions using a neural network.
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Public Health (AREA)
- General Physics & Mathematics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Biomedical Technology (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
Description
Claims (8)
- 방광병변 진단 시스템이 단위 병리 이미지를 입력 받는 단계;상기 방광 병변 진단 시스템이 상기 단위 병리 이미지를 제1뉴럴 네트워크에 입력하여 상기 단위 병리 이미지에 복수의 방광병변들 중 제1방광 병변의 진단결과를 획득하는 단계; 및상기 방광 병변 진단 시스템이 상기 단위 병리 이미지를 제2뉴럴 네트워크에 입력하여 상기 단위 병리 이미지에 상기 복수의 방광병변들 중 상기 제1방광 병변을 제외한 제2방광 병변의 진단결과를 획득하는 단계를 포함하며,상기 제1뉴럴 네트워크는,상기 제1방광 병변이 발현된 병변 영역이 어노테이션된 복수의 제1학습 데이터를 통해 학습된 뉴럴 네트워크인 것을 특징으로 하고,상기 제2뉴럴 네트워크는,상기 제2방광 병변 중 적어도 하나가 발현되었는지 여부인 발현 병변 종류가 어노테이션된 복수의 제2학습데이터-제2학습데이터는 병변 영역에 대한 어노테이션이 미수행된 데이터임-를 통해 학습된 뉴럴 네트워크인 것을 특징으로 하는 뉴럴 네트워크를 이용한 방광병변 진단 방법.
- 제1항에 있어서, 상기 단위 병리 이미지는,경요도방광절제술(transurenthral resection of bladder, TURB)를 통해 확보된 조직 검체에 상응하는 병리 이미지가 소정의 크기로 분할된 패치이미지인 것을 특징으로 하는 뉴럴 네트워크를 이용한 방광병변 진단 방법.
- 제1항에 있어서, 상기 제1방광 병변은,urothelial carcinoma in situ(CIS) 병변을 포함하는 뉴럴 네트워크를 이용한 방광병변 진단 방법.
- 제1항에 있어서, 상기 제2방광 병변은,침윤성 방광암(invasive urothelial carcinoma), low/high grade noninvasive papillary urothelial carcinoma 병변, papillary urothelial neoplasm of low malignant potential (PUNLMP) 병변, urothelial proliferation of unknown malignant potential (UPUMP) 병변, urothelial papilloma 병변, inverted urothelial papilloma 병변, urothelial dysplasia 병변 중 적어도 하나를 포함하는 뉴럴 네트워크를 이용한 방광병변 진단 방법.
- 방광병변 진단 시스템이 단위 병리 이미지를 입력 받는 단계;상기 방광 병변 진단 시스템이 상기 단위 병리 이미지를 제2뉴럴 네트워크에 입력하여 상기 단위 병리 이미지에 복수의 방광병변들 중 제1방광 병변을 제외한 제2방광 병변의 진단결과를 획득하는 단계; 및상기 방광 병변 진단 시스템이 상기 단위 병리 이미지를 제1뉴럴 네트워크에 입력하여 상기 단위 병리 이미지에 상기 복수의 방광병변들 중 상기 제1방광 병변의 진단결과를 획득하는 단계를 포함하며,상기 제1뉴럴 네트워크는,상기 제1방광 병변이 발현된 병변 영역이 어노테이션된 복수의 제1학습 데이터를 통해 학습된 뉴럴 네트워크인 것을 특징으로 하고,상기 제2뉴럴 네트워크는,상기 제2방광 병변 중 적어도 하나가 발현되었는지 여부인 발현 병변 종류가 어노테이션된 복수의 제2학습데이터-제2학습데이터는 병변 영역에 대한 어노테이션이 미수행된 데이터임-를 통해 학습된 뉴럴 네트워크인 것을 특징으로 하는 뉴럴 네트워크를 이용한 방광병변 진단 방법.
- 데이터 처리장치에 설치되며 제1항 내지 제5항 중 어느 한 항에 기재된 방법을 수행하기 위한 컴퓨터 판독가능한 기록매체에 저장된 컴퓨터 프로그램.
- 프로세서; 및상기 프로세서에 의해 실행되는 프로그램이 기록된 메모리를 포함하며,상기 프로세서는 상기 프로그램을 구동하여,단위 병리 이미지를 제1뉴럴 네트워크에 입력하여 상기 단위 병리 이미지에 복수의 방광병변들 중 제1방광 병변의 진단결과를 획득하고, 상기 단위 병리 이미지를 제2뉴럴 네트워크에 입력하여 상기 단위 병리 이미지에 상기 복수의 방광병변들 중 상기 제1방광 병변을 제외한 제2방광 병변의 진단결과를 획득하며,상기 제1뉴럴 네트워크는,상기 제1방광 병변이 발현된 병변 영역이 어노테이션된 복수의 제1학습 데이터를 통해 학습된 뉴럴 네트워크인 것을 특징으로 하고,상기 제2뉴럴 네트워크는,상기 제2방광 병변 중 적어도 하나가 발현되었는지 여부인 발현 병변 종류가 어노테이션된 복수의 제2학습데이터-제2학습데이터는 병변 영역에 대한 어노테이션이 미수행된 데이터임-를 통해 학습된 뉴럴 네트워크인 것을 특징으로 하는 뉴럴 네트워크를 이용한 방광병변 진단 시스템.
- 제7항에 있어서, 상기 제1방광 병변은,urothelial carcinoma in situ(CIS) 병변을 포함하는 뉴럴 네트워크를 이용한 방광병변 진단 시스템.
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP22771710.5A EP4290527A1 (en) | 2021-03-15 | 2022-03-14 | Bladder lesion diagnosis method using neural network, and system thereof |
CN202280017419.0A CN116917999A (zh) | 2021-03-15 | 2022-03-14 | 利用神经网络的膀胱病变诊断方法及其系统 |
US18/282,213 US20240144476A1 (en) | 2021-03-15 | 2022-03-14 | Bladder lesion diagnosis method using neural network, and system thereof |
JP2023551759A JP2024512272A (ja) | 2021-03-15 | 2022-03-14 | ニューラルネットワークを用いた膀胱病変の診断方法及びそのシステム |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020210033189A KR102354476B1 (ko) | 2021-03-15 | 2021-03-15 | 뉴럴 네트워크를 이용한 방광병변 진단 시스템 제공방법 및 그 시스템 |
KR10-2021-0033189 | 2021-03-15 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022197044A1 true WO2022197044A1 (ko) | 2022-09-22 |
Family
ID=80050166
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/KR2022/003546 WO2022197044A1 (ko) | 2021-03-15 | 2022-03-14 | 뉴럴 네트워크를 이용한 방광병변 진단 방법 및 그 시스템 |
Country Status (6)
Country | Link |
---|---|
US (1) | US20240144476A1 (ko) |
EP (1) | EP4290527A1 (ko) |
JP (1) | JP2024512272A (ko) |
KR (1) | KR102354476B1 (ko) |
CN (1) | CN116917999A (ko) |
WO (1) | WO2022197044A1 (ko) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117694839A (zh) * | 2024-02-05 | 2024-03-15 | 四川省肿瘤医院 | 基于图像的非肌层浸润性膀胱癌复发率预测方法和系统 |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR102354476B1 (ko) * | 2021-03-15 | 2022-01-21 | 주식회사 딥바이오 | 뉴럴 네트워크를 이용한 방광병변 진단 시스템 제공방법 및 그 시스템 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180240551A1 (en) * | 2017-02-20 | 2018-08-23 | General Electric Company | Methods and systems for hierarchical machine learning models for medical imaging |
US20180263568A1 (en) * | 2017-03-09 | 2018-09-20 | The Board Of Trustees Of The Leland Stanford Junior University | Systems and Methods for Clinical Image Classification |
JP2019148473A (ja) * | 2018-02-27 | 2019-09-05 | シスメックス株式会社 | 画像解析方法、画像解析装置、プログラム、学習済み深層学習アルゴリズムの製造方法および学習済み深層学習アルゴリズム |
KR20200092805A (ko) * | 2019-01-25 | 2020-08-04 | 주식회사 딥바이오 | 커스터마이징 진단 시스템 제공방법 및 그 시스템 |
KR102354476B1 (ko) * | 2021-03-15 | 2022-01-21 | 주식회사 딥바이오 | 뉴럴 네트워크를 이용한 방광병변 진단 시스템 제공방법 및 그 시스템 |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2021529585A (ja) * | 2018-06-29 | 2021-11-04 | ミラキ イノベーション シンク タンク エルエルシー | 機械学習及び人工知能を使用する小型の体内で制御可能な医療機器 |
-
2021
- 2021-03-15 KR KR1020210033189A patent/KR102354476B1/ko active IP Right Grant
-
2022
- 2022-03-14 CN CN202280017419.0A patent/CN116917999A/zh active Pending
- 2022-03-14 JP JP2023551759A patent/JP2024512272A/ja active Pending
- 2022-03-14 US US18/282,213 patent/US20240144476A1/en active Pending
- 2022-03-14 WO PCT/KR2022/003546 patent/WO2022197044A1/ko active Application Filing
- 2022-03-14 EP EP22771710.5A patent/EP4290527A1/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180240551A1 (en) * | 2017-02-20 | 2018-08-23 | General Electric Company | Methods and systems for hierarchical machine learning models for medical imaging |
US20180263568A1 (en) * | 2017-03-09 | 2018-09-20 | The Board Of Trustees Of The Leland Stanford Junior University | Systems and Methods for Clinical Image Classification |
JP2019148473A (ja) * | 2018-02-27 | 2019-09-05 | シスメックス株式会社 | 画像解析方法、画像解析装置、プログラム、学習済み深層学習アルゴリズムの製造方法および学習済み深層学習アルゴリズム |
KR20200092805A (ko) * | 2019-01-25 | 2020-08-04 | 주식회사 딥바이오 | 커스터마이징 진단 시스템 제공방법 및 그 시스템 |
KR102354476B1 (ko) * | 2021-03-15 | 2022-01-21 | 주식회사 딥바이오 | 뉴럴 네트워크를 이용한 방광병변 진단 시스템 제공방법 및 그 시스템 |
Non-Patent Citations (2)
Title |
---|
ANTONI, S. ET AL., BLADDER CANCER INCIDENCE AND MORTALITY: A GLOBAL OVERVIEW AND RECENT TRENDS, 2016 |
NEGASSI MISGANA; SUAREZ-IBARROLA RODRIGO; HEIN SIMON; MIERNIK ARKADIUSZ; REITERER ALEXANDER: "Application of artificial neural networks for automated analysis of cystoscopic images: a review of the current status and future prospects", WORLD JOURNAL OF UROLOGY, SPRINGER BERLIN HEIDELBERG, BERLIN/HEIDELBERG, vol. 38, no. 10, 10 January 2020 (2020-01-10), Berlin/Heidelberg, pages 2349 - 2358, XP037251855, ISSN: 0724-4983, DOI: 10.1007/s00345-019-03059-0 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117694839A (zh) * | 2024-02-05 | 2024-03-15 | 四川省肿瘤医院 | 基于图像的非肌层浸润性膀胱癌复发率预测方法和系统 |
CN117694839B (zh) * | 2024-02-05 | 2024-04-16 | 四川省肿瘤医院 | 基于图像的非肌层浸润性膀胱癌复发率预测方法和系统 |
Also Published As
Publication number | Publication date |
---|---|
US20240144476A1 (en) | 2024-05-02 |
JP2024512272A (ja) | 2024-03-19 |
CN116917999A (zh) | 2023-10-20 |
KR102354476B1 (ko) | 2022-01-21 |
EP4290527A1 (en) | 2023-12-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022197044A1 (ko) | 뉴럴 네트워크를 이용한 방광병변 진단 방법 및 그 시스템 | |
WO2017095014A1 (ko) | Dnn 학습을 이용한 세포이상 여부 진단시스템 및 그 진단관리 방법 | |
WO2017213398A1 (en) | Learning model for salient facial region detection | |
WO2016171341A1 (ko) | 클라우드 기반 병리 분석 시스템 및 방법 | |
WO2022149894A1 (ko) | 병리 검체에 대한 판단 결과를 제공하는 인공 뉴럴 네트워크의 학습 방법, 및 이를 수행하는 컴퓨팅 시스템 | |
WO2014051246A1 (en) | Method and apparatus for inferring facial composite | |
WO2021137454A1 (ko) | 인공지능 기반의 사용자 의료정보 분석 방법 및 시스템 | |
WO2017069596A1 (ko) | Cad기반 디지털 엑스레이의 자동 결핵 진단 예측 시스템 | |
WO2019235828A1 (ko) | 투 페이스 질병 진단 시스템 및 그 방법 | |
WO2018155963A1 (en) | Method of accelerating execution of machine learning based application tasks in a computing device | |
WO2021071288A1 (ko) | 골절 진단모델의 학습 방법 및 장치 | |
WO2019143179A1 (ko) | 동일 피사체에 대하여 시간 간격을 두고 촬영된 영상 간에 동일 관심구역을 자동으로 검출하는 방법 및 이를 이용한 장치 | |
WO2020204635A1 (ko) | 병리 진단 지원 장치 및 방법 | |
WO2019054711A1 (ko) | 스마트 기기를 이용한 딥 러닝 기반 의료용 초음파 영상 분류 방법 및 장치 | |
WO2022103134A1 (ko) | 질병 통합 진단 시스템 및 동작 방법 | |
WO2016204535A1 (ko) | 의료 검사를 위한 이미지 분석 관리 방법 및 서버 | |
WO2016114574A2 (ko) | 패치 쉬프트를 이용한 텍스처 필터링 방법 및 장치 | |
WO2021002669A1 (ko) | 병변 통합 학습 모델을 구축하는 장치와 방법, 및 상기 병변 통합 학습 모델을 사용하여 병변을 진단하는 장치와 방법 | |
WO2017010612A1 (ko) | 의료 영상 분석 기반의 병리 진단 예측 시스템 및 방법 | |
WO2023191472A1 (ko) | 면역조직화학 염색 이미지를 분석하기 위한 기계학습모델을 학습하는 방법 및 이를 수행하는 컴퓨팅 시스템 | |
WO2022191539A1 (ko) | Turp 병리 이미지로부터 전립선암을 검출하기 위한 용도의 인공 뉴럴 네트워크를 학습하는 방법 및 이를 수행하는 컴퓨팅 시스템 | |
WO2022158843A1 (ko) | 조직 검체 이미지 정제 방법, 및 이를 수행하는 컴퓨팅 시스템 | |
WO2022231200A1 (ko) | 유방암 병변 영역을 판별하기 위한 인공 신경망을 학습하기 위한 학습 방법, 및 이를 수행하는 컴퓨팅 시스템 | |
WO2023013959A1 (ko) | 아밀로이드 베타 축적 예측 장치 및 방법 | |
WO2022245063A1 (ko) | 인공 지능 기반의 유전체와 의료 정보 분석 및 의약 물질 개발 방법 및 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: 22771710 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2023551759 Country of ref document: JP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202280017419.0 Country of ref document: CN |
|
ENP | Entry into the national phase |
Ref document number: 2022771710 Country of ref document: EP Effective date: 20230906 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18282213 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |