KR20210157795A - Method to train model - Google Patents

Abstract

According to one embodiment of the present disclosure, a computer program stored in a computer-readable storage medium is disclosed. When the computer program is executed on one or more processors, the computer program performs the following operations for training the model. The operations may include: identifying another medical image set which includes features similar to a target medical image set; and training a model for detecting medical data related to the target medical image set by using at least a portion of the other medical image set.

Description

How to train a model {METHOD TO TRAIN MODEL}

The present invention relates to a model learning method, and more particularly, to a model learning method using data including similar features.

In order to detect or classify data through machine learning, it is necessary to secure learning data for machine learning in advance. For machine learning, at least 1,000 to 10,000 training data must be secured.

However, there are cases in which it is impossible to secure sufficient training data in constructing the training data. For example, in the case of rare cancer, since the probability of occurrence of the rare cancer itself is low, it is impossible to secure a large amount of learning data on the rare cancer. When learning is performed using such a small amount of training data, the accuracy of the trained model may be reduced.

Accordingly, in training the model, there is a need in the art to perform learning using insufficient data.

Korean Patent Laid-Open No. 2016-0012537 discloses a method and apparatus for learning a neural network, and a data processing apparatus.

The present disclosure has been made in response to the above-mentioned background art, and an object of the present disclosure is to provide a model learning method.

As a computer program stored in a computer-readable storage medium according to an embodiment of the present disclosure for realizing the above-described task, the computer program performs the following operations for learning a model when the computer program is executed in one or more processors wherein the operations include: identifying another set of medical images, the other set of medical images including features similar to the set of target medical images; and training a model for detecting medical data related to the target medical image set by using at least a portion of the other medical image set.

In an alternative embodiment, the medical data may relate to at least one of an organ or a pathology included in the medical image.

In an alternative embodiment, the operation of identifying another medical image set, which includes features similar to the target medical image set, includes: a result of calculating a medical image using a medical data discrimination model to distinguish the medical data, When an image is classified into two or more classes, determining that two or more medical images corresponding to the two or more classes include similar features.

In an alternative embodiment, the identifying the other medical image set, which includes features similar to the target medical image set, comprises: identifying the target medical image using the other medical data detection model, pre-trained using the other medical image. outputting a detection result for the target medical image by calculating; and determining that the target medical image set and the other medical image set include similar features when the accuracy of the detection result is equal to or greater than a threshold value.

In an alternative embodiment, the accuracy of the detection result may include: a degree to which a medical image including the target medical data is distinguished as being included in the target medical data; or a degree to which a medical image that does not include the target medical data is discriminated as that does not include the target medical data; It may be determined by at least one of

In an alternative embodiment, the model may calculate at least one of target medical data or other medical data by calculating a medical image as an input.

In an alternative embodiment, the model may be trained on a training data set comprising at least a portion of the target medical image set and at least a portion of the other medical image set. In an alternative embodiment, the training data set includes at least a portion of the target medical image set, the first third-party medical image set, or the second third-party medical image set, and according to the feature similarity with the target medical image, the A ratio of the first other medical image set and the second other medical image set included in the training data set may be determined.

In an alternative embodiment, the model may be trained to detect the target medical image using at least a part of the structure of the other medical data detection model that has been pre-trained using the other medical image.

In an alternative embodiment, the model comprises: at least a portion of a first third-party medical data detection model, pre-trained using a first third-party medical image, and a second third-party medical image, pre-trained using a second third-party medical image Target medical data included in the medical image may be output by combining features of the output medical image using at least a portion of the data detection model, respectively.

In an alternative embodiment, determining the performance of the model by computing a medical image using the learned model; and adjusting the two or more medical image sets included in a training data set for training the model based on the performance of the model.

A method of training a model according to an embodiment of the present disclosure for realizing the above-described task, the method comprising: identifying another medical image set including features similar to a target medical image set; and training a model for detecting medical data related to the target medical image set by using at least a portion of the other medical image set.

As a server for learning a model according to an embodiment of the present disclosure for realizing the task as described above, the server comprising: a processor including one or more cores; and a memory, wherein the processor is configured to: identify another medical image set that includes features similar to a target medical image set, and use at least a portion of the other medical image set to be associated with the target medical image set. You can train a model to detect data.

As a computer-readable recording medium storing a data structure corresponding to a parameter of a neural network that is at least partially updated in the learning process according to an embodiment of the present disclosure for realizing the above-described task, the operation of the neural network is Based at least in part on the parameter, the learning process may include: identifying another set of medical images comprising features similar to a target medical image set; and training a model for detecting medical data related to the target medical image set by using at least a portion of the other medical image set.

The present disclosure may provide a method for learning a model.

1 is a diagram illustrating a block diagram of a computing device that performs an operation for providing a model learning method according to an embodiment of the present disclosure.
2 is a diagram illustrating a medical image according to an embodiment of the present disclosure.
3 is a diagram exemplarily illustrating a method of identifying two or more sets of medical images including similar features according to an embodiment of the present disclosure.
4 is a diagram exemplarily illustrating a method for identifying two or more sets of medical images including similar features according to an embodiment of the present disclosure.
5 is a diagram exemplarily illustrating a method for identifying two or more sets of medical images including similar features according to an embodiment of the present disclosure.
6 is a diagram exemplarily illustrating a model learning method according to an embodiment of the present disclosure.
7 is a flowchart of a model learning method according to an embodiment of the present disclosure.
8 is a block diagram of a computing device according to an embodiment of the present disclosure.

Various embodiments are now described with reference to the drawings. In this specification, various descriptions are presented to provide an understanding of the present disclosure. However, it is apparent that these embodiments may be practiced without these specific descriptions.

The terms “component,” “module,” “system,” and the like, as used herein, refer to a computer-related entity, hardware, firmware, software, a combination of software and hardware, or execution of software. For example, a component can be, but is not limited to being, a process running on a processor, a processor, an object, a thread of execution, a program, and/or a computer. For example, both an application running on a computing device and the computing device may be a component. One or more components may reside within a processor and/or thread of execution. A component may be localized within one computer. A component may be distributed between two or more computers. In addition, these components can execute from various computer readable media having various data structures stored therein. Components may communicate via a network such as the Internet with another system, for example, via a signal having one or more data packets (eg, data and/or signals from one component interacting with another component in a local system, distributed system, etc.) may communicate via local and/or remote processes depending on the data being transmitted).

In addition, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or.” That is, unless otherwise specified or clear from context, "X employs A or B" is intended to mean one of the natural implicit substitutions. That is, X employs A; X employs B; or when X employs both A and B, "X employs A or B" may apply to either of these cases. It should also be understood that the term “and/or” as used herein refers to and includes all possible combinations of one or more of the listed related items.

Also, the terms "comprises" and/or "comprising" should be understood to mean that the feature and/or element in question is present. However, it should be understood that the terms "comprises" and/or "comprising" do not exclude the presence or addition of one or more other features, elements and/or groups thereof. Also, unless otherwise specified or unless it is clear from context to refer to a singular form, the singular in the specification and claims should generally be construed to mean “one or more”.

Those skilled in the art will further appreciate that the various illustrative logical blocks, configurations, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be implemented in electronic hardware, computer software, or combinations of both. It should be recognized that they can be implemented with To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. However, such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

Descriptions of the presented embodiments are provided to enable those skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art of the present disclosure. The generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present invention is not limited to the embodiments presented herein. This invention is to be interpreted in its widest scope consistent with the principles and novel features presented herein.

In an embodiment of the present disclosure, the server may include other components for performing a server environment of the server. The server may include any type of device. The server is a digital device, and may be a digital device equipped with a processor, such as a laptop computer, a notebook computer, a desktop computer, a web pad, and a mobile phone, and having a computing power with a memory. The server may be a web server that processes a service. The above-described types of servers are merely examples, and the present disclosure is not limited thereto.

In this specification, neural network, artificial neural network, and network function may often be used interchangeably.

The term "image" or "image data" as used throughout the present description and claims refers to multidimensional data composed of discrete image elements (eg, pixels in a two-dimensional image), in other words, ( For example, a term that refers to a visible object (eg, displayed on a video screen) or a digital representation of the object (eg, a file corresponding to a pixel output of a CT, MRI detector, etc.).

For example, “image” or “image” can be defined as computed tomography (CT), magnetic resonance imaging (MRI), fundus imaging, ultrasound or any other medical imaging known in the art. It may be a medical image of a subject collected by the system. The image does not necessarily have to be provided in a medical context, but may be provided in a non-medical context, for example, there may be X-ray imaging for security screening.

Throughout the detailed description and claims of the present invention, the 'DICOM (Digital Imaging and Communications in Medicine)' standard is a generic term for various standards used for digital image expression and communication in medical devices. The DICOM standard is published by a joint committee formed by the American Society of Radiological Medicine (ACR) and the National Electrical Engineers Association (NEMA).

In addition, throughout the detailed description and claims of the present invention, 'Picture Archiving and Communication System (PACS)' is a term that refers to a system that stores, processes, and transmits in accordance with the DICOM standard, X-ray, CT , medical imaging images acquired using digital medical imaging equipment such as MRI are stored in DICOM format and transmitted to terminals inside and outside the hospital through a network, to which reading results and medical records can be added.

1 is a diagram illustrating a block diagram of a computing device that performs an operation for providing a model learning method according to an embodiment of the present disclosure.

In order to train a neural network model, the construction of a training data set for the neural network model must be preceded. In order to learn the neural network model, a training data set including at least 1,000 to 10,000 or more training data may be required. However, when training a neural network model for the medical field, training data is often insufficient. For example, in the case of a rare disease, since the probability of the occurrence of the rare disease itself is low, it may be difficult to collect learning data on the rare disease. There is a great need for detecting rare diseases as well as commonly known diseases. Therefore, in order to detect a rare disease, various methods for constructing a training data set for a rare disease are presented. For example, data aggregation for rare diseases may be performed using a general data aggregation method.

In general data aggregation, additional data is generated using real data, and the additional data may be generated by transforming at least a portion of the real data. For example, additional data may be generated by blurring or rotating at least a portion of the disease image. However, the data aggregation as described above generates data that is not real by transforming data that actually exists, and builds a training data set, and thus accuracy may be lowered. That is, since artificially transformed data is used instead of real data, when the model is trained using the training data generated according to the data aggregation method, the performance of the model may be partially degraded. Accordingly, the model learning method according to an embodiment of the present disclosure presents a learning data set construction method using a method different from the data aggregation described above.

The model learning method according to an embodiment of the present disclosure may not generate additional training data by transforming existing data, but may be used as additional training data using other types of data that actually exist. That is, the model learning method according to an embodiment of the present disclosure provides a method of using data on another disease as learning data when constructing learning data for one disease. For example, when training data for rare cancer is insufficient, data on other cancers may be used together as training data in addition to data on rare cancer. In this case, rare cancers and other cancers may share similar features. In this regard, it will be described later in detail.

Hereinafter, according to an embodiment of the present disclosure, a method of constructing data for learning a model will be described.

The processor 120 may identify two or more sets of medical images that contain similar features. The processor 120 may identify two or more sets of medical images for use as a training data set.

The medical image may be an image obtained by photographing an organ or an organ including a pathology. For example, the medical image may be an image obtained by photographing an organ using MRI, CT, or the like. The medical image may be an image of an organ. Alternatively, the medical image may be an image of an organ including a specific pathology. Referring to FIG. 2 , each image shown in FIGS. 2A and 2B may be a medical image. The detailed description of the medical image described above is only an example, and the present disclosure is not limited thereto.

The medical image set may be a set of medical images. A medical image set may include one or more medical images. Medical image sets may be generated on an organ or pathological basis. For example, a medical image set including a medical image of the lungs and a medical image set including a medical image of the pancreas may be generated. Alternatively, for example, lung cancer may also be classified into squamous cell carcinoma, adenocarcinoma, large cell carcinoma, etc. according to pathology, and a medical image set may also be separately generated for each of squamous cell carcinoma, adenocarcinoma, and large cell carcinoma. have. The detailed description of the medical image set described above is only an example, and the present disclosure is not limited thereto.

The medical data may relate to at least one of an organ or a pathology included in the medical image. The medical data may be related to an output obtained when a medical image is computed using the medical image as an input to the model. The medical data may be information about which organs are included in the medical image when the medical image is calculated using the medical image as an input to the model. Alternatively, the medical data may be information regarding a pathology included in the medical image when it is calculated using the medical image as an input to the model. For example, when the medical image does not include abnormal pathology, that is, an image obtained by photographing a normal lung, the medical data may be 'lung', which is a result of identification of an organ. Or, for example, if the medical image is an image of a lung with large cell cancer including abnormal pathology, the medical data is 'lung', which is the identification result of an organ, and 'large cell cancer', which is the identification result of the pathology. can be The detailed description of the medical data described above is only an example, and the present disclosure is not limited thereto.

The training data set may be generated using two or more sets of medical images. The two or more medical image sets may include a target medical image set and another medical image set. A set of target medical images may be associated with medical data for detection using the model. The target medical image set may be a medical image set for medical data lacking in training data. For example, the target medical image set may be a medical image set for a rare disease, such as a rare cancer. The other medical image set may include other medical images having features similar to the target medical image included in the target medical image set. The other medical image set may be a medical image set for supplementing the target medical image set lacking in training data. That is, the other medical image set may be a medical image set in which a lot of data exists. Other medical image sets may include medical images having features similar to the target medical image. That is, the processor 120 may use medical images for other diseases or organs having similar features to supplement the target medical image set. For example, if a medical image for a rare disease A and a medical image for a common disease B (i.e., a disease for which data is abundant) contain similar features, the processor 120 may set the medical image set for the rare disease A may be determined as a target medical image set, and a medical image set for a rare disease B may be determined as another medical image set. Referring to FIG. 2 , for example, FIG. 2(a) may be a medical image for rare disease A, and FIG. 2(b) may be a medical image for common disease B. FIG. The two medical images may include similar features, and to detect rare disease A for FIG. 2( a ), data about common disease B for FIG. 2( b ) may be used as additional training data . The detailed description of the above-described target medical image set and other medical image sets is merely an example, and the present disclosure is not limited thereto.

The training data set may include a medical image included in the medical image set as an input of the training data, and may include medical data for the medical image as a label. The training data set may include at least a portion of a target medical image set and other medical image sets. The training data set may include a medical image included in the target medical image set as an input of the training data, and medical data for the medical image as a label. Also, the training data set may include a medical image included in another medical image set as an input of the training data and medical data for the medical image as a label. For example, in order to detect adenocarcinoma, the processor 120 may generate a training data set. The processor 120 learns at least one of whether or not adenocarcinoma is included in the medical image, or the location of adenocarcinoma by using a medical image obtained by photographing an adenocarcinoma included in a target medical image set, an adenocarcinoma medical image set, as an input of learning data. It can be included as a label for the data. In order to assist in adenocarcinoma detection, the processor 120 uses, as an input of learning data, a medical image of large cell cancer included in another medical image set, which is a medical image set for large cell cancer having features similar to adenocarcinoma. , and at least one of whether large cell cancer is included in the medical image or the location of the large cell box may be included as a label of the training data. The detailed description of the above-described learning data set is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 120 may identify two or more medical image sets having a low degree of differentiation when performing an operation using a medical data classification model for distinguishing medical data. Hereinafter, a method of identifying two or more medical image sets will be described with reference to FIG. 3 .

The processor 120 may output the operation result 320 by using the medical image 310 as an input of the medical data discrimination model 300 . The medical data discrimination model 300 may calculate the medical image 310 as an input and output medical data corresponding to the medical image 310 . For example, the processor 120 may calculate the input medical image 310 using the medical data discrimination model 300 , and output medical data related to an organ or disease included in the medical image 310 . can

The medical data discrimination model 300 may calculate the medical image 310 and output a score value for each of two or more classes. Each of the two or more classes may respectively correspond to two or more different medical data. For example, the first class may be lungs, the second class may be liver, and the third class may be organs such as small intestine. Or, for example, each of the two or more classes may be a disease of the liver, wherein the first class is hepatocarcinoma, the second class is cholangiocarcinoma, the third class is hepatoblastoma, the fourth class is angiosarcoma, and the like. Specific description of the above-described class is merely an example, and the present disclosure is not limited thereto.

The processor 120 may classify the medical image 310 into at least one class based on a score value for each of two or more classes, which is a result of an operation on the medical image 310 . According to an embodiment of the present disclosure, the processor 120 may classify the medical image 310 into a class corresponding to the largest score value. The processor 120 may classify the medical image 310 into the one class when the score value for one class among the score values for each of the two or more classes is greater than the score values for the other classes. . For example, when the score values for each of the two or more classes are 0.7, 0.3, 0.2, 0.2, 0.25, the processor 120 may classify the medical image 310 into a class corresponding to the score value of 0.7. According to an embodiment of the present disclosure, the processor 120 may classify the medical image 310 into a class corresponding to a score value greater than or equal to a threshold value. The processor 120 may classify the medical image 310 as belonging to the two classes, for example, when there are two or more classes corresponding to a score value greater than or equal to a threshold value. According to an embodiment of the present disclosure, if the difference between the largest score value and the second largest score value is less than a threshold ratio or a threshold value, the processor 120 assigns the medical image 310 to the highest score, not one class. It can be classified into classes corresponding to each value and the second largest score value. For example, if the score values for each of the two or more classes are 0.7, 0.6, 0.2, 0.1, 0.25, and the threshold value is 0.6, the processor 120 sets the score values 0.7 and 0.6 having a value greater than the threshold value 0.6. The medical image 310 may be classified into two classes corresponding to each. Alternatively, for example, the processor 120 may classify the medical image 310 into two classes respectively corresponding to the score values by determining that there is no significant difference between the score values of 0.7 and 0.6. The detailed description of the medical image classification described above is only an example, and the present disclosure is not limited thereto.

As a result of the operation, when the medical image 310 is classified into two or more classes, the processor 120 may determine that two or more medical image sets corresponding to the two or more classes include similar features. When the medical image 310 is calculated using the medical data discrimination model, it may not be possible to classify it as one medical data, and it may be possible to classify it into two or more medical data. When the processor 120 classifies the medical image 310 into two or more classes, the processor 120 may determine that medical data corresponding to each of the two or more classes have similar features. For example, when the medical image 310 can be classified into hepatocarcinoma and hemangiosarcoma, the processor 120 may determine that the medical image for each of hepatocarcinoma and hemangiosarcoma has similar features. The processor 120 may determine that the medical image included in the medical image set corresponding to each of hepatocarcinoma and hemangiosarcoma has similar features. The detailed description of the medical image having the above-described similar features is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 120 calculates a medical image as an input of the medical data discrimination model 300 and determines target medical data having similar features and other medical data using the output result. can The processor 120 may calculate the medical image as an input of the medical data discrimination model 300 . The medical image may be a target medical image, or it may be another medical image. The medical data discrimination model 300 may classify medical images into two or more classes. Medical data corresponding to two or more classes included in the classification result of the medical data discrimination model 300 may be determined to have similar features. The processor 120 may determine that target medical data corresponding to one of the two or more classes and medical data corresponding to the other classes (ie, other medical data) have similar features. The processor 120 may use other medical data having similar features when constructing a training data set for the target medical data.

According to an embodiment of the present disclosure, the processor 120 calculates a target medical image as an input of the medical data discrimination model 300 and uses the output result to generate a target medical image having features similar to the target medical image. can decide For example, angiosarcoma (ie, target medical data) is a rare cancer, so image data for angiosarcoma may be lacking. In order to find another carcinoma having features similar to hemangiosarcoma, the processor 120 may calculate a medical image of the hemangiosarcoma as an input of the medical data discrimination model 300 . The output of the medical data discrimination model 300 may include a result of classifying a medical image for angiosarcoma into two classes, angiosarcoma and cholangiocarcinoma. Since the two medical data of angiosarcoma and cholangiocarcinoma share similar features, the processor 120 may determine that the medical data discrimination model 300 classifies the medical image for angiosarcoma into the two classes. . The processor 120 may thus determine that the medical images corresponding to each of the hemangiosarcoma and the cholangiocarcinoma share similar features. The processor 120 may utilize at least a part of a medical image set for cholangiocarcinoma when constructing a training data set for detecting angiosarcoma. The detailed description of the medical image having the above-described similar features is merely an example, and the present disclosure is not limited thereto.

That is, when the processor 120 calculates a medical image using the medical data discrimination model, if there are two or more classes that are not well distinguished, the processor 120 may determine that medical data corresponding to the two or more classes have similar features. Accordingly, the processor 120 may generate a training data set using two or more medical image data sets having similar features. The processor 120 may use other medical data when constructing a training data set for the target medical data by using another medical image having features similar to the target medical image.

According to an embodiment of the present disclosure, the processor 120 may identify two or more medical image sets using other medical data detection models.

The medical data detection model may be a model for calculating a medical image as an input and detecting medical data included in the medical image. The other medical data detection model may be a model trained on medical data different from the target medical data. The other medical data detection model may be a model trained on medical data having sufficient training data. For example, since hepatocarcinoma is a liver cancer with the highest incidence among various types of liver cancer, there may be a considerable amount of learning data. The other medical data detection model may be, for example, a model trained for HCC. When a medical image is input, other medical data detection models may determine whether the medical image is an image relating to hepatocarcinoma or detect a location of hepatocarcinoma included in the medical image. The detailed description of the medical data detection model described above is only an example, and the present disclosure is not limited thereto.

A method in which the processor 120 determines another medical image having a feature similar to a target medical image will be described with reference to FIG. 4 , according to an embodiment of the present disclosure. The processor 120 may output a detection result 420 for the target medical image by calculating the target medical image 410 using the other medical data detection model 400 that has been pre-trained using the other medical image. can The processor 120 may calculate the target medical image 410 by using the other medical data detection model 400 learned on other medical data having sufficient training data. The detection result may include at least one of whether target medical data is included in the target medical image or a location 422 of the target medical data included in the target medical image. For example, the processor 120 may calculate a medical image of the hepatoblastoma by using the other medical data detection model 400 learned for the hepatocarcinoma. The processor 120 may output a detection result for an image (ie, a target medical image) of hepatoblastoma. The detailed description of the above-described calculation of the target medical image is merely an example, and the present disclosure is not limited thereto.

When the accuracy of the detection result is equal to or greater than the threshold, the processor 120 may determine that the target medical image set and the other medical image set include similar features. That is, when the accuracy is equal to or greater than the threshold, it may be determined that the target medical image is also detectable because other medical data and the target medical data share similar features.

The accuracy of the detection result may be determined by an Area Under the Receiver Operating Characteristics (AUROC) value. The description of AUROC in the present specification is incorporated by reference in its entirety by "The use of the area under the ROC curve in the evaluation of machine learning algorithms" Andrew P. Bradley, July 1997.

The AUROC value can be calculated based on the sensitivity and specificity of the test result. The ROC curve may be plotted according to 1-(specificity) of the x-axis and sensitivity of the y-axis. Specificity is a measure of what percentage of people with a disease can be determined to have the disease. Sensitivity is a measure of what percentage of disease-free people can be determined to be disease-free.

The accuracy of the detection result may be determined by a degree (specificity) of distinguishing a medical image including the target medical data as being included in the target medical data. The accuracy of the detection result may be determined by a degree (sensitivity) of distinguishing a medical image that does not include the target medical data as a medical image that does not include the target medical data.

The AUROC value may be a value obtained by calculating the base area of the ROC curve. It may be determined that the higher the AUROC value is, the higher the accuracy of the detection result is, and the smaller the AUROC value is, the lower the accuracy of the detection result is. For example, when the AUROC value is 1, the accuracy of the detection result is 100%, when the AUROC value is 0.5, the accuracy of the detection result is normal, and when the AUROC value is 0, it is determined that the accuracy of the detection result is bad. can

The processor 120 may determine that the higher the AUROC value, the higher the accuracy when the target medical image is calculated using the other medical data detection model 400 . The processor 120 may determine that the lower the AUROC value, the lower the accuracy when the target medical image is calculated using the other medical data detection model 400 . The processor 120 may compare the AUROC value with a predetermined threshold to determine whether other medical data and the target medical data have similar features. When the AUROC value is equal to or greater than a predetermined threshold, the processor 120 may determine that other medical data and the target medical data have similar features. When constructing a training data set for learning target medical data, the processor 120 may use at least a part of a medical image set for other medical data. When the AUROC value is less than a predetermined threshold value, the processor 120 may determine that other medical data and target medical data do not have similar features.

Hereinafter, a method of determining the accuracy of a detection result will be described with reference to FIG. 5 . In the table shown in FIG. 5 , the x-axis represents target medical data, and the y-axis represents other medical data. 5 is a table showing the accuracy of a detection result when a target medical image is calculated using a model learned on other medical data. A higher AUROC value may be displayed in a darker color in the table, and a lower AUROC value may be displayed in a lighter color in the table. The processor 120 may determine that other medical data and target medical data corresponding to the section marked with a dark color have similar features. The processor 120 may determine that LIHC (Liver hepatocellular carcinoma), THCA (Thyroid carcinoma), and BLCA (Bladder Urothelial Carcinoma) corresponding to the portion 510 indicated in dark color in FIG. 5 have similar features. When the LHIC is the target medical data, the processor 120 may determine medical data for each of THCA and BLCA as other medical data. The processor 120 may utilize at least a part of medical image sets for each of THCA and BLCA when generating training data of a model for LHIC detection. The detailed description of the medical image having the above-described similar features is merely an example, and the present disclosure is not limited thereto.

According to another embodiment of the present disclosure, the processor 120 may perform the above operation for a purpose other than the purpose of replenishing the target medical data lacking the training data. The processor 120 may identify two or more pieces of medical data having similar features in order to improve detection performance of the medical data. The processor 120 may generate a training data set by using a medical image set for medical data having similar features among medical data in which the training data is not insufficient.

The processor 120 may calculate the second medical image by using the first medical data detection model learned using the first medical data. The processor 120 may determine that the first medical data and the second medical data have similar features when the accuracy of the detection result for the second medical image is equal to or greater than the threshold value. The processor 120 may generate a training data set by using at least a part of a medical image set for each of the first medical data and the second medical data. The processor 120 may utilize the training data set as described above in order to improve detection performance for each of the first medical data and the second medical data.

Referring to FIG. 5 , two pieces of medical data corresponding to a section marked with a dark color in the table may have similar features. On the other hand, two pieces of medical data corresponding to a section indicated by a light color in the table may have relatively less similar features. The processor 120 may generate a training data set by using two or more medical data that share similar features. Accordingly, the processor 120 may determine to generate one training data set by grouping the sections indicated in dark colors in the table of FIG. 5 . For example, the processor 120 generates a first training data set using a medical image set corresponding to each of OV, LUSC, and LUAD, and a second training data set corresponding to each of LIHC, THCA, BLCA, and BRCA. It can be created using a set of medical images. That is, the processor 120 may generate a training data set by grouping medical data sharing similar features.

In this case, since the model is trained using training data sharing similar features, the ability to extract similar features shared between both data can be improved. In addition, it is possible to learn more quickly how to detect similar features in the learning process. Specific description of a medical image having similar features is merely an example, and the present disclosure is not limited thereto.

Each of the medical data discrimination model and the medical data detection model described above may be a deep neural network. Throughout this specification, the terms neural network, network function, and neural network may be used interchangeably. A deep neural network (DNN) may refer to a neural network including a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks can be used to identify the latent structures of data. In other words, it can identify the potential structure of photos, texts, videos, voices, and music (e.g., what objects are in the photos, what the text and emotions are, what the texts and emotions are, etc.) . Deep neural networks include a convolutional neural network (CNN), a recurrent neural network (RNN), a restricted boltzmann machine (RBM), and a deep belief network (DBN). , Q network, U network, Siamese network, and the like.

A convolutional neural network is a kind of deep neural network, and includes a neural network including a convolutional layer. Convolutional neural networks are a type of multilayer perceptorns designed to use minimal preprocessing. A CNN may consist of one or several convolutional layers and artificial neural network layers combined with them. CNNs can additionally utilize weights and pooling layers. Thanks to this structure, CNN can fully utilize the input data of the two-dimensional structure. A convolutional neural network can be used to recognize objects in an image. A convolutional neural network can process image data by representing it as a matrix with dimensions. For example, in the case of RGB (red-green-blue) encoded image data, each R, G, and B color may be represented as a two-dimensional (eg, two-dimensional image) matrix. That is, the color value of each pixel of the image data may be a component of the matrix, and the size of the matrix may be the same as the size of the image. Accordingly, image data can be represented by three two-dimensional matrices (three-dimensional data array).

In a convolutional neural network, a convolutional process (input/output of a convolutional layer) can be performed by moving the convolutional filter and multiplying the convolutional filter and matrix components at each position of the image. The convolutional filter may be composed of an n*n matrix. A convolutional filter may be composed of a fixed type filter that is generally smaller than the total number of pixels in an image. That is, when an m*m image is input as a convolutional layer (eg, a convolutional layer having a size of n*n convolutional filters), a matrix representing n*n pixels including each pixel of the image This convolutional filter can be multiplied with a component (ie, a product of each component of a matrix). A component matching the convolutional filter may be extracted from the image by multiplication with the convolutional filter. For example, a 3*3 convolutional filter for extracting upper and lower linear components from an image can be configured as [[0,1,0], [0,1,0], [0,1,0]] can When the 3*3 convolutional filter for extracting the upper and lower linear components from the image is applied to the input image, the upper and lower linear components matching the convolutional filter may be extracted and output from the image. The convolutional layer may apply a convolutional filter to each matrix (ie, R, G, B colors in the case of R, G, and B coded images) for each channel representing the image. The convolutional layer may extract a feature matching the convolutional filter from the input image by applying the convolutional filter to the input image. The filter value of the convolutional filter (ie, the value of each component of the matrix) may be updated by backpropagation in the learning process of the convolutional neural network.

A subsampling layer is connected to the output of the convolutional layer to simplify the output of the convolutional layer, thereby reducing memory usage and computational amount. For example, if you input the output of the convolutional layer to a pooling layer with a 2*2 max pooling filter, you can compress the image by outputting the maximum value included in each patch for every 2*2 patch in each pixel of the image. can The above-described pooling may be a method of outputting a minimum value from a patch or an average value of a patch, and any pooling method may be included in the present disclosure.

A convolutional neural network may include one or more convolutional layers and subsampling layers. The convolutional neural network may extract features from an image by repeatedly performing a convolutional process and a subsampling process (eg, the aforementioned max pooling, etc.). Through iterative convolutional and subsampling processes, neural networks can extract global features from images.

An output of the convolutional layer or the subsampling layer may be input to a fully connected layer. A fully connected layer is a layer in which all neurons in one layer and all neurons in neighboring layers are connected. The fully connected layer may refer to a structure in which all nodes of each layer are connected to all nodes of other layers in a neural network.

In an embodiment of the present disclosure, in order to perform segmentation on medical data included in a medical image, the neural network may include a deconvolutional neural network (DCNN). A deconvolutional neural network performs an operation similar to that calculated in the reverse direction of a convolutional neural network. The deconvolutional neural network may output features extracted from the convolutional neural network as a feature map related to original data. A description of a specific configuration for a convolutional neural network is discussed more specifically in US Patent No. US9870768B2, which is incorporated herein by reference in its entirety.

Hereinafter, a method for training a model according to an embodiment of the present disclosure will be described.

The processor 120 may train a model for detecting medical data related to at least one medical image set among the two or more medical image sets by using at least a portion of the two or more medical image sets. The processor 120 may generate a training data set by using at least a part of the target medical image set and other medical image sets. The model trained using the corresponding training data set may be a model for detecting at least one of target medical data or other medical data. That is, the model may detect at least one of target medical data or other medical data by calculating a medical image as an input.

Hereinafter, a method of constructing a training data set for training a model according to an embodiment of the present disclosure will be described.

A model may be trained on a training data set, including at least a portion of the two or more sets of medical images. The training data set may include at least a portion of a target medical image set and one or more other medical image sets.

The training data set may include at least a part of the target medical image set, the first other medical image set, or the second other medical image set. A ratio of the first other medical image set to the second other medical image set included in the training data set may be determined according to the feature similarity with the target medical image. That is, when generating the training data set by using two or more other medical image sets, it is possible to determine how much to use the other medical image sets according to the feature similarity with the target medical image.

According to an embodiment of the present disclosure, the processor 120 may determine how much to use other medical image sets by using the degree of discrimination between medical data using the medical data discrimination model. The processor 120 may determine to include more similar features as the degree of discrimination between medical data using the medical data discrimination model is lower. The processor 120 may further include other medical image data sets containing more similar features in the training data set.

The processor 120 may calculate a degree of differentiation between the target medical data and other medical data. When the medical image is calculated using the medical data discrimination model, the processor 120 may classify the medical image into two or more classes. The processor 120 may determine that the degree of differentiation between medical data is lower as the difference between the score values of the class corresponding to the target medical data and the class corresponding to other medical data is smaller. For example, the calculation result for the A medical image indicates that a score value of a class corresponding to the target medical data is 0.7, a score value of a class corresponding to the first other medical data is 0.65, and the second medical data is a score value of 0.65. The class may have a score value of 0.68. In this case, since the difference in score (0.03) between the classes corresponding to the target medical data and the second other medical data is smaller than the difference in the score (0.05) between the classes corresponding to the target medical data and the second other medical data, It may be determined that the second other medical data has a lower degree of differentiation from the target medical data than the first other medical data. The processor 120 may determine that a feature similarity between the second other medical data and the target medical data is higher than a feature similarity between the first other medical data and the target medical data. The processor 120 may include more of the second other medical image set than the first other medical image set in the training data set. The detailed description of the above-described generation of the training data set is only an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 120 may determine how much to use the other medical image set by using the accuracy of the detection result using the other medical data detection model. The processor 120 may determine that the feature similarity between the other medical data and the target medical data is higher as the accuracy of the detection result using the other medical data detection model is higher. The processor 120 may include more other medical image sets in the training data set with respect to other medical data having a higher feature similarity. The processor 120 may calculate detection accuracy when the target medical image is calculated using other medical data detection models. The processor 120 may determine that the higher the AUROC value, the higher the detection accuracy. For example, when an OV medical image is calculated using the LIHC medical data detection model, the AUROC value may be 0.78. Alternatively, when the OV medical image is calculated using the THCA medical data detection model, the AUROC value may be 0.83. The processor 120 may determine to include more THCA medical image sets with higher AUROC values in the training data set than LIHC medical image sets. The detailed description of the above-described generation of the training data set is only an example, and the present disclosure is not limited thereto.

Hereinafter, a method for training a model according to an embodiment of the present disclosure will be described.

The model according to an embodiment of the present disclosure may be trained to detect the target medical image by using at least a part of the structure of the other medical data detection model that has been previously learned using the other medical image.

The model according to an embodiment of the present disclosure may be trained using at least some of the weights of other pre-trained medical data detection models. The processor 120 may set at least some of the weights of other pre-trained medical data detection models as initial weights of the target medical data detection model. That is, by using the weight of the model for detecting other medical data having similar features as the initial weight of the model, it is possible to shorten the learning time required for extracting the similar features. For example, the processor 120 may use weights for all layers of other pre-trained medical data detection models as initial weights of the model. Alternatively, the processor 120 may use weights for the remaining previous layers except for the FCL layer of the last stage of the other pre-trained medical data detection model as the initial weights of the model. The processor 120 may update the initial weight while setting the initial weight to at least some of the weights of other pre-trained medical data detection models and training the model using the training data set. That is, the processor 120 may learn the target medical data detection model by using at least a part of the structure of the other medical data detection model. In addition, the processor 120 may include at least a part of other medical image sets in the training data set to train the target medical data detection model. The detailed description of the above-described method for learning the model is merely an example, and the present disclosure is not limited thereto.

The model (ie, the target medical data detection model) according to an embodiment of the present disclosure may be trained using at least a part of the structures of two or more pre-trained other medical data detection models. Hereinafter, a model learning method will be described with reference to FIG. 6 . The target medical data detection model may be trained using at least a part of the first other medical data detection model and at least a part of the second other medical data detection model, respectively. The first other medical data detection model may be a pre-trained model using the first other medical data. The second other medical data detection model may be a pre-trained model using the second other medical data. The first other medical data, the second other medical data, and the target medical data may each have similar features.

The target medical data detection model may be divided into a sub-model of a front stage and a sub-model of a rear stage. The sub-model of the previous stage may be a model using at least a part of other pre-trained medical data detection models. That is, the sub-model of the previous stage may be a model using the weights of other pre-trained medical data detection models as initial weights. By setting the initial weight as a weight of another pre-trained medical data detection model, and training the model using the training data set, the weight may be updated. The sub-model of the preceding stage may be a sub-model for extracting features from a medical image. The sub-model of the previous stage may be a model in which at least a portion of two or more other medical data detection models are connected in parallel. For example, the medical data having a feature similar to the target medical data may be the first other medical data and the second other medical data. The front end of the target medical data detection model may have a structure in which the first other medical data detection sub-model 622 and the second other medical data detection sub-model 624 are connected in parallel. The first other medical data detection sub-model 622 may be at least a part of the pre-trained first other medical data detection model. The second other medical data detection sub-model 624 may be at least a part of the pre-trained second other medical data detection model. The first other medical data detection sub-model 622 and the second other medical data detection sub-model 624 may be sub-models for extracting features from the medical image, which are part of the other medical data detection models, respectively. The target medical data detection model may extract the first feature of the medical image by calculating the input medical image 610 using the first other medical data detection sub-model 622 . The target medical data detection model may extract the second feature of the medical image by calculating the input medical image 610 using the second other medical data detection sub-model 624 . The processor 120 may combine features of the medical image extracted from other medical data detection sub-models. The processor 120 may combine the first feature and the second feature of the medical image. For example, the processor 120 may concatenate a first feature and a second feature of the medical image. The processor 120 may combine features of the medical image output from other medical data detection sub-models, and may perform an operation as an input of the target medical data detection sub-model 630 . The processor 120 may be trained to detect target medical data included in the medical image 610 using the features of the combined medical images. For example, the target medical data detection sub-model 630 may be a fully connected layer (FCL). That is, the front end of the target medical data detection model may partially use the structure of the other medical data detection sub-model that has been pre-trained using other medical data having similar features. In addition, the rear end of the target medical data detection sub-model may be trained to extract target medical data from the combined features. The detailed description of the above-described method for learning the model is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the learned target medical data detection model may detect target medical data. Since the target medical data detection model is a model trained through a training data set generated using at least a part of another medical image set, other medical data may also be detected. One model trained using two or more sets of medical images having similar features may detect two or more medical data having similar features. That is, to detect N organs or diseases, organs or diseases sharing similar features can be trained together, and therefore fewer than N models can be trained.

According to an embodiment of the present disclosure, a training data set for training a model may be adjusted using reinforcement learning.

The processor 120 may determine the performance of the learned target medical data detection model by calculating a medical image using the learned target medical data detection model. The processor 120 may adjust the training data set for training the target medical data detection model based on the performance of the target medical data detection model. The processor 120 may adjust two or more medical image sets included in the training data set.

The processor 120 may adjust the training data set of the target medical data detection model by using the reinforcement learning model. The reinforcement learning model according to an embodiment of the present disclosure may be a model using a policy network and the value network. The policy network may be a teacher learned policy network (SL policy network) and/or a reinforcement learned policy network (RL policy network). The reinforcement-learned policy network may be initialized based on the teacher-learned policy network, and may be a network that learns to maximize rewards.

A policy according to an embodiment of the present disclosure may be a configuration of a training data set. The policy may be a composition and/or proportion of a target medical image set and one or more other medical image sets included in the training data set.

The reward according to an embodiment of the present disclosure may be determined according to the performance of the target medical data detection model. When the medical image is calculated using the learned target medical data detection model, the processor 120 may calculate the accuracy of the output result. The processor 120 may determine the reward to be larger as the accuracy of the output result is higher, and conversely, the lower the accuracy, the lower the reward may be determined. The accuracy of the output result may be determined by, for example, the AUROC value, as described above.

A value network according to an embodiment of the present disclosure may be a network that predicts a value function value given a current state and/or a policy as input. The value function may derive an expected reward based on the input current state and policy. For example, the reward output from the value function may be an average of the sum of future rewards. A reward may be a reward obtained when a certain policy is taken in the current state.

That is, the processor 120 trains the target medical data detection model using the training data set, evaluates the performance of the target medical data detection model, and determines a reward according to the performance to adjust the training data set again. . The processor 120 may adjust the configuration of the training data set to maximize the performance of the target medical data detection model by using reinforcement learning. The detailed description of the reinforcement learning method described above is only an example, and the present disclosure is not limited thereto.

The computing device 100 for providing a model learning method according to an embodiment of the present disclosure may include a network unit 110 , a processor 120 , and a memory 130 .

The network unit 110 may transmit and receive medical images and medical data according to an embodiment of the present disclosure to other computing devices, servers, and the like. In addition, the network unit 110 may enable communication between a plurality of computing devices so that operations for learning a model are performed distributedly in each of the plurality of computing devices. The network unit 110 may enable communication between a plurality of computing devices to distribute calculations for model learning using a network function.

The network unit 110 according to an embodiment of the present disclosure may operate based on any type of wired/wireless communication technology currently used and implemented, such as short-distance (short-range), long-distance, wired and wireless, etc., and may operate in other networks as well. can be used

The processor 120 may include one or more cores, and a central processing unit (CPU) of a computing device, a general purpose graphics processing unit (GPGPU), and a tensor processing unit (TPU). unit) and the like, and may include a processor for learning the model. The processor 120 may provide a model learning method according to an embodiment of the present disclosure by reading a computer program stored in the memory 130 . According to an embodiment of the present disclosure, the processor 120 may perform calculation to provide a model learning method.

The memory 130 may store a computer program for providing a model learning method according to an embodiment of the present disclosure, and the stored computer program may be read and driven by the processor 120 .

The memory 130 according to embodiments of the present disclosure may store a program for the operation of the processor 120 , and may temporarily or permanently store input/output data or events. The memory 130 may store data related to a display and sound. The memory 130 may include a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (eg, SD or XD memory, etc.), RAM (Random Access Memory, RAM), SRAM (Static Random Access Memory), ROM (Read-Only Memory, ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, magnetic It may include at least one type of storage medium among a disk and an optical disk.

A computer-readable medium storing a data structure is disclosed according to an embodiment of the present disclosure.

The data structure may refer to the organization, management, and storage of data that enables efficient access and modification of data. A data structure may refer to an organization of data to solve a specific problem (eg, data retrieval, data storage, and data modification in the shortest time). A data structure may be defined as a physical or logical relationship between data elements designed to support a particular data processing function. The logical relationship between data elements may include a connection relationship between user-defined data elements. Physical relationships between data elements may include actual relationships between data elements physically stored on a computer-readable storage medium (eg, persistent storage). A data structure may specifically include a set of data, relationships between data, and functions or instructions applicable to data. Through an effectively designed data structure, the computing device can perform an operation while using the resource of the computing device to a minimum. Specifically, the computing device may increase the efficiency of operations, reads, insertions, deletions, comparisons, exchanges, and retrievals through effectively designed data structures.

A data structure may be classified into a linear data structure and a non-linear data structure according to the type of the data structure. The linear data structure may be a structure in which only one piece of data is connected after one piece of data. The linear data structure may include a list, a stack, a queue, and a deck. A list may mean a set of data in which an order exists internally. The list may include a linked list. The linked list may be a data structure in which data is linked in such a way that each data is linked in a line with a pointer. In a linked list, a pointer may contain information about a link with the next or previous data. A linked list may be expressed as a single linked list, a doubly linked list, or a circularly linked list according to a shape. A stack can be a data enumeration structure with limited access to data. A stack can be a linear data structure in which data can be manipulated (eg, inserted or deleted) at only one end of the data structure. The data stored in the stack may be a data structure (LIFO-Last in First Out). A queue is a data listing structure that allows limited access to data, and unlike a stack, it may be a data structure that is stored later (FIFO-First in First Out). A deck can be a data structure that can process data at either end of the data structure.

The non-linear data structure may be a structure in which a plurality of data is connected after one data. The nonlinear data structure may include a graph data structure. A graph data structure may be defined as a vertex and an edge, and the edge may include a line connecting two different vertices. A graph data structure may include a tree data structure. The tree data structure may be a data structure in which one path connects two different vertices among a plurality of vertices included in the tree. That is, it may be a data structure that does not form a loop in the graph data structure.

Throughout this specification, computational model, neural network, network function, and neural network may be used interchangeably. Hereinafter, the neural network is unified and described. The data structure may include a neural network. And the data structure including the neural network may be stored in a computer-readable medium. Data structures including neural networks also include preprocessed data for processing by the neural network, data input to the neural network, weights of the neural network, hyperparameters of the neural network, data obtained from the neural network, activation functions associated with each node or layer of the neural network, and neural networks. It may include a loss function for learning of . A data structure comprising a neural network may include any of the components disclosed above. That is, the data structure including the neural network includes preprocessed data for processing by the neural network, data input to the neural network, weights of the neural network, hyperparameters of the neural network, data obtained from the neural network, activation function associated with each node or layer of the neural network, and neural network It may be configured to include all or any combination thereof, such as a loss function for learning of . In addition to the above-described configurations, a data structure including a neural network may include any other information that determines a characteristic of a neural network. In addition, the data structure may include all types of data used or generated in the operation process of the neural network, and is not limited to the above. Computer-readable media may include computer-readable recording media and/or computer-readable transmission media. A neural network may be composed of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. A neural network is configured to include at least one or more nodes.

The data structure may include data input to the neural network. A data structure including data input to the neural network may be stored in a computer-readable medium. The data input to the neural network may include learning data input in a neural network learning process and/or input data input to the neural network in which learning is completed. Data input to the neural network may include pre-processing data and/or pre-processing target data. The preprocessing may include a data processing process for inputting data into the neural network. Accordingly, the data structure may include data to be pre-processed and data generated by pre-processing. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

The data structure may include the weights of the neural network. (In this specification, weight and parameter may be used interchangeably.) And the data structure including the weight of the neural network may be stored in a computer-readable medium. The neural network may include a plurality of weights. The weight may be variable, and may be changed by a user or an algorithm in order for the neural network to perform a desired function. For example, when one or more input nodes are interconnected to one output node by respective links, the output node sets values input to input nodes connected to the output node and links corresponding to the respective input nodes. A data value output from the output node may be determined based on the weight. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

By way of example and not limitation, the weight may include a weight variable in a neural network learning process and/or a weight in which neural network learning is completed. The variable weight in the neural network learning process may include a weight at a time point at which a learning cycle starts and/or a weight variable during the learning cycle. The weight for which neural network learning is completed may include a weight for which a learning cycle is completed. Accordingly, the data structure including the weights of the neural network may include a data structure including the weights that vary in the neural network learning process and/or the weights on which the neural network learning is completed. Therefore, it is assumed that the above-described weights and/or combinations of weights are included in the data structure including the weights of the neural network. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

The data structure including the weights of the neural network may be stored in a computer-readable storage medium (eg, memory, hard disk) after being serialized. Serialization can be the process of converting a data structure into a form that can be reconstructed and used later by storing it on the same or a different computing device. The computing device may serialize the data structure to send and receive data over the network. A data structure including weights of the serialized neural network may be reconstructed in the same computing device or in another computing device through deserialization. The data structure including the weight of the neural network is not limited to serialization. Furthermore, the data structure including the weights of the neural network is a data structure to increase computational efficiency while using the resources of the computing device to a minimum (e.g., B-Tree, Trie, m-way search tree, AVL tree, Red-Black Tree). The foregoing is merely an example, and the present disclosure is not limited thereto.

The data structure may include hyper-parameters of the neural network. In addition, the data structure including the hyperparameters of the neural network may be stored in a computer-readable medium. The hyper parameter may be a variable variable by a user. Hyperparameters are, for example, learning rate, cost function, number of iterations of the learning cycle, weight initialization (e.g., setting the range of weight values to be initialized for weights), Hidden Unit The number (eg, the number of hidden layers, the number of nodes of the hidden layer) may be included. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

7 is a flowchart of a model learning method according to an embodiment of the present disclosure.

The computing device 100 may identify 710 other medical image sets that include features similar to the target medical image set. The medical image set may be a set of medical images. A medical image set may include one or more medical images. Medical image sets may be generated on an organ or pathological basis. The medical data may relate to at least one of an organ or a pathology included in the medical image.

The computing device 100 may identify two or more medical image sets having a low degree of discrimination when performing an operation using a medical data discrimination model for distinguishing medical data.

The computing device 100 may output a calculation result by using the medical image as an input of the medical data discrimination model. When the medical image is classified into two or more classes as a result of the calculation, the computing device 100 may determine that two or more medical image sets corresponding to the two or more classes include similar features.

The computing device 100 may output a detection result for the target medical image by calculating the target medical image by using the other medical data detection model that has been previously learned using the other medical image. When the accuracy of the detection result is equal to or greater than the threshold, the computing device 100 may determine that the target medical image set and other medical image sets include similar features. The accuracy of the detection result may be determined by a degree to which a medical image including the target medical data is distinguished as being included in the target medical data. The accuracy of the detection result may be determined by a degree to which a medical image not including the target medical data is distinguished as not including the target medical data.

The computing device 100 may train ( 720 ) a model for detecting medical data related to a target medical image set by using at least a portion of another medical image set. The model may calculate at least one of target medical data or other medical data by calculating a medical image as an input.

A model may be trained on a training data set, including at least a portion of the two or more sets of medical images. The training data set may include at least a part of the target medical image set, the first other medical image set, or the second other medical image set. In the training data set, a ratio of the first other medical image set and the second other medical image set included in the training data set may be determined according to the feature similarity with the target medical image.

The model may be trained to detect the target medical image by using at least a part of the structure of the other medical data detection model that has been previously trained using other medical images.

The model includes at least a part of the first other medical data detection model pre-trained using the first other medical image and at least a part of the second other medical data detection model pre-trained using the second other medical image, respectively By using the combination of features of the output medical image, target medical data included in the medical image may be output.

The computing device 100 may determine the performance of the model by calculating a medical image using the learned model. The computing device 100 may adjust two or more medical image sets included in the training data set for training the model based on the performance of the model.

The model learning method according to an embodiment of the present disclosure may be implemented by modules, circuits, means and logic that perform the above operations.

8 is a block diagram of a computing device according to an embodiment of the present disclosure.

8 depicts a simplified, general schematic diagram of an example computing environment in which embodiments of the present disclosure may be implemented.

Although the present disclosure has been described above generally in the context of computer-executable instructions that may be executed on one or more computers, those skilled in the art will appreciate that the present disclosure may be implemented in combination with other program modules and/or as a combination of hardware and software. will be.

Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will appreciate that the methods of the present disclosure are suitable for single-processor or multiprocessor computer systems, minicomputers, mainframe computers as well as personal computers, hand held computing devices, microprocessor-based or programmable consumer electronics, etc. (each of which is It will be appreciated that other computer system configurations may be implemented, including those that may operate in connection with one or more associated devices.

The described embodiments of the present disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

Computers typically include a variety of computer-readable media. Any medium accessible by a computer may be a computer-readable medium. Computer-readable media includes volatile and nonvolatile media, transitory and non-transitory media, removable and non-removable media. By way of example, and not limitation, computer-readable media may include computer-readable storage media and computer-readable transmission media. Computer readable storage media includes volatile and nonvolatile media, temporary and non-transitory media, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. includes media. Computer storage media may include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital video disk (DVD) or other optical disk storage device, magnetic cassette, magnetic tape, magnetic disk storage device or other magnetic storage device; or any other medium that can be accessed by a computer and used to store the desired information.

Computer-readable transmission media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as other transport mechanisms, and includes all information delivery media. do. The term modulated data signal means a signal in which one or more of the characteristics of the signal is set or changed so as to encode information in the signal. By way of example, and not limitation, computer-readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above are also intended to be included within the scope of computer-readable transmission media.

An example environment 1100 implementing various aspects of the present disclosure is shown including a computer 1102 , the computer 1102 including a processing unit 1104 , a system memory 1106 , and a system bus 1108 . do. A system bus 1108 couples system components, including but not limited to system memory 1106 , to the processing device 1104 . The processing device 1104 may be any of a variety of commercially available processors. Dual processor and other multiprocessor architectures may also be used as processing unit 1104 .

The system bus 1108 may be any of several types of bus structures that may further interconnect a memory bus, a peripheral bus, and a local bus using any of a variety of commercial bus architectures. System memory 1106 includes read only memory (ROM) 1110 and random access memory (RAM) 1112 . A basic input/output system (BIOS) is stored in non-volatile memory 1110, such as ROM, EPROM, EEPROM, etc., the BIOS is the basic input/output system (BIOS) that helps transfer information between components within computer 1102, such as during startup. contains routines. RAM 1112 may also include high-speed RAM, such as static RAM, for caching data.

The computer 1102 may also include an internal hard disk drive (HDD) 1114 (eg, EIDE, SATA) - this internal hard disk drive 1114 may also be configured for external use within a suitable chassis (not shown). Yes-, magnetic floppy disk drive (FDD) 1116 (eg, for reading from or writing to removable diskette 1118), and optical disk drive 1120 (eg, CD-ROM) for reading from, or writing to, disk 1122, or other high capacity optical media, such as DVD. The hard disk drive 1114 , the magnetic disk drive 1116 , and the optical disk drive 1120 are connected to the system bus 1108 by the hard disk drive interface 1124 , the magnetic disk drive interface 1126 , and the optical drive interface 1128 , respectively. ) can be connected to The interface 1124 for implementing an external drive includes at least one or both of Universal Serial Bus (USB) and IEEE 1394 interface technologies.

These drives and their associated computer readable media provide non-volatile storage of data, data structures, computer executable instructions, and the like. In the case of computer 1102, drives and media correspond to storing any data in a suitable digital format. Although the description of computer readable media above refers to HDDs, removable magnetic disks, and removable optical media such as CDs or DVDs, those skilled in the art will use zip drives, magnetic cassettes, flash memory cards, cartridges, etc. It will be appreciated that other tangible computer-readable media, such as, etc., may also be used in the example operating environment and that any such media may include computer-executable instructions for performing the methods of the present disclosure.

A number of program modules may be stored in the drive and RAM 1112 , including an operating system 1130 , one or more application programs 1132 , other program modules 1134 , and program data 1136 . All or portions of the operating system, applications, modules, and/or data may also be cached in RAM 1112 . It will be appreciated that the present disclosure may be implemented in various commercially available operating systems or combinations of operating systems.

A user may enter commands and information into the computer 1102 via one or more wired/wireless input devices, for example, a pointing device such as a keyboard 1138 and a mouse 1140 . Other input devices (not shown) may include a microphone, IR remote control, joystick, game pad, stylus pen, touch screen, and the like. Although these and other input devices are connected to the processing unit 1104 through an input device interface 1142 that is often connected to the system bus 1108, parallel ports, IEEE 1394 serial ports, game ports, USB ports, IR interfaces, It may be connected by other interfaces, etc.

A monitor 1144 or other type of display device is also coupled to the system bus 1108 via an interface, such as a video adapter 1146 . In addition to the monitor 1144, the computer typically includes other peripheral output devices (not shown), such as speakers, printers, and the like.

Computer 1102 may operate in a networked environment using logical connections to one or more remote computers, such as remote computer(s) 1148 via wired and/or wireless communications. Remote computer(s) 1148 may be workstations, computing device computers, routers, personal computers, portable computers, microprocessor-based entertainment devices, peer devices, or other common network nodes, and are typically connected to computer 1102 . Although it includes many or all of the components described for it, only memory storage device 1150 is shown for simplicity. The logical connections shown include wired/wireless connections to a local area network (LAN) 1152 and/or a larger network, eg, a wide area network (WAN) 1154 . Such LAN and WAN networking environments are common in offices and companies, and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to a worldwide computer network, for example, the Internet.

When used in a LAN networking environment, the computer 1102 is coupled to the local network 1152 through a wired and/or wireless communication network interface or adapter 1156 . Adapter 1156 may facilitate wired or wireless communication to LAN 1152 , which LAN 1152 also includes a wireless access point installed therein for communicating with wireless adapter 1156 . When used in a WAN networking environment, the computer 1102 may include a modem 1158, be connected to a communication computing device on the WAN 1154, or establish communications over the WAN 1154, such as over the Internet. have other means. A modem 1158 , which may be internal or external and a wired or wireless device, is coupled to the system bus 1108 via a serial port interface 1142 . In a networked environment, program modules depicted relative to computer 1102 , or portions thereof, may be stored in remote memory/storage device 1150 . It will be appreciated that the network connections shown are exemplary and other means of establishing a communication link between the computers may be used.

The computer 1102 may be associated with any wireless device or object that is deployed and operates in wireless communication, for example, a printer, scanner, desktop and/or portable computer, portable data assistant (PDA), communication satellite, wireless detectable tag. It operates to communicate with any device or place, and phone. This includes at least Wi-Fi and Bluetooth wireless technologies. Accordingly, the communication may be a predefined structure as in a conventional network or may simply be an ad hoc communication between at least two devices.

Wi-Fi (Wireless Fidelity) makes it possible to connect to the Internet, etc. without a wired connection. Wi-Fi is a wireless technology such as cell phones that allows these devices, eg, computers, to transmit and receive data indoors and outdoors, ie anywhere within range of a base station. Wi-Fi networks use a radio technology called IEEE 802.11 (a, b, g, etc.) to provide secure, reliable, and high-speed wireless connections. Wi-Fi can be used to connect computers to each other, to the Internet, and to wired networks (using IEEE 802.3 or Ethernet). Wi-Fi networks can operate in unlicensed 2.4 and 5 GHz radio bands, for example, at 11 Mbps (802.11a) or 54 Mbps (802.11b) data rates, or in products that include both bands (dual band). have.

One of ordinary skill in the art of this disclosure will understand that information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, instructions, information, signals, bits, symbols, and chips that may be referenced in the above description are voltages, currents, electromagnetic waves, magnetic fields or particles, optical field particles or particles, or any combination thereof.

Those of ordinary skill in the art of the present disclosure will recognize that the various illustrative logical blocks, modules, processors, means, circuits, and algorithm steps described in connection with the embodiments disclosed herein include electronic hardware, (convenience For this purpose, it will be understood that it may be implemented by various forms of program or design code (referred to herein as "software") or a combination of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. A person skilled in the art of the present disclosure may implement the described functionality in various ways for each specific application, but such implementation decisions should not be interpreted as a departure from the scope of the present disclosure.

The various embodiments presented herein may be implemented as methods, apparatus, or articles of manufacture using standard programming and/or engineering techniques. The term “article of manufacture” includes a computer program or media accessible from any computer-readable device. For example, computer-readable media include magnetic storage devices (eg, hard disks, floppy disks, magnetic strips, etc.), optical disks (eg, CDs, DVDs, etc.), smart cards, and flash memory. devices (eg, EEPROMs, cards, sticks, key drives, etc.). Also, various storage media presented herein include one or more devices and/or other machine-readable media for storing information.

It is understood that the specific order or hierarchy of steps in the presented processes is an example of exemplary approaches. Based on design priorities, it is to be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present disclosure. The appended method claims present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art of the present disclosure. The generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present disclosure is not intended to be limited to the embodiments presented herein, but is to be construed in the widest scope consistent with the principles and novel features presented herein.

Claims (14)

A computer program stored in a computer readable storage medium, wherein the computer program, when executed on one or more processors, performs the following operations for training a model, the operations comprising:
identifying another set of medical images comprising features similar to the set of target medical images; and
training a model for detecting medical data related to the target medical image set by using at least a part of the other medical image set;
containing,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
wherein the medical data relates to at least one of an organ or a pathology included in the medical image;
A computer program stored in a computer-readable storage medium.
The method of claim 1,
Identifying another set of medical images comprising features similar to the target medical image set includes:
As a result of calculating a medical image using a medical data discrimination model for distinguishing medical data, when the medical image is classified into two or more classes, the two or more medical images corresponding to the two or more classes include similar features decision to be taken;
containing,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
Identifying another set of medical images comprising features similar to the target medical image set includes:
outputting a detection result for the target medical image by calculating a target medical image using another medical data detection model that has been previously learned using another medical image; and
determining that the target medical image set and the other medical image set include similar features when the accuracy of the detection result is equal to or greater than a threshold value;
containing,
A computer program stored in a computer-readable storage medium.
5. The method of claim 4,
The accuracy of the detection result is:
a degree to which a medical image including the target medical data is discriminated as including the target medical data; or
a degree to which a medical image that does not include the target medical data is discriminated as that does not include the target medical data;
determined by at least one of
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The model calculates a medical image as an input to detect at least one of target medical data or other medical data,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
wherein the model is trained with a training data set comprising at least a portion of the target medical image set and at least a portion of the other medical image set;
A computer program stored in a computer-readable storage medium.
8. The method of claim 7,
The training data set includes at least a part of the target medical image set, the first other medical image set, or the second other medical image set, and
a ratio of the first other medical image set and the second other medical image set included in the training data set is determined according to a feature similarity with a target medical image;
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The model is trained to detect medical data related to a target medical image using at least a part of the structure of the other medical data detection model, which has been previously learned using other medical images.
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The model includes at least a part of the first other medical data detection model pre-trained using the first other medical image, and at least a part of the second other medical data detection model pre-trained using the second other medical image. outputting target medical data included in the medical image by combining features of the output medical image by using
A computer program stored in a computer-readable storage medium.
The method of claim 1,
determining performance of the model by calculating a medical image using the learned model; and
adjusting two or more medical image sets included in a training data set for training the model based on the performance of the model;
further comprising,
A computer program stored in a computer-readable storage medium.
A method of learning a model, comprising:
identifying another set of medical images comprising features similar to the set of target medical images; and
training a model for detecting medical data related to the target medical image set by using at least a part of the other medical image set;
containing,
How the model is trained.
As a server for model training,
a processor including one or more cores; and
Memory;
including,
The processor is
identify another set of medical images that contain features similar to the set of target medical images, and
training a model for detecting medical data related to the target medical image set using at least a part of the other medical image set;
Server for training the model.
A computer-readable recording medium storing a data structure corresponding to a parameter of a neural network that is at least partially updated in a learning process, wherein the operation of the neural network is based at least in part on the parameter, and the learning process includes:
identifying another set of medical images comprising features similar to the set of target medical images; and
training a model for detecting medical data related to the target medical image set by using at least a part of the other medical image set;
containing,
A computer-readable recording medium having a data structure stored thereon.

Images