KR20230002204A - Apparatus for quality management of medical image interpretation using machine learning, and method thereof - Google Patents

Abstract

The present disclosure relates to a device for managing a result of medical image reading by comparing a result of a medical image reading by a medical person and a reading result generated based on a machine learning model, and to a method thereof. The method comprises the steps of: receiving a medical image; receiving report information which is a result of a determination by a medical practitioner on the medical image; generating result information indicating the identity of first lesion information and second lesion information by applying, to a third analysis model, the first lesion information related to a lesion of the medical image obtained based on the medical image and the second lesion information related to the lesion of the medical image obtained based on the report information; and outputting the result information.

Description

Method and apparatus for quality management of medical image reading using machine learning

The present disclosure relates to a medical image analysis apparatus and method for managing a result of reading a medical image by comparing a result of reading a medical image by a medical person with a reading result generated based on a machine learning model.

Imaging tests are used to discover new diseases or to observe changes in existing lesions. Imaging is important because it is the starting point for patient diagnosis.

However, due to the aging population, the increase in private insurance coverage, and the expansion of medical care benefits, the scale of imaging tests is rapidly increasing, increasing the burden of interpretation by radiologists. In addition, since the accuracy of medical image interpretation can vary depending on various factors such as the reading environment, medical personnel's specialization, and medical personnel's concentration, there is a possibility that medical personnel may misinterpret medical images, and that medical personnel for the same medical images Readings can be different.

In practice, treatment may be delayed because the medical personnel cannot read the lesion in the medical image, or the wrong surgery or drug prescription may be received because the lesion is read as a different lesion from the one shown in the medical image.

Therefore, the need for quality control of medical image readings is emerging, and quality control methods of medical image readings can be largely divided into two types. The first method is to educate medical personnel. Since the medical image is read by a medical person, the quality of the medical image can be improved by regularly educating the medical person and conducting tests on the medical person.

However, this method has a problem in that it is difficult to evaluate reading accuracy in an actual clinical environment. In addition, there is a problem that measures cannot be taken for images that have been erroneously read in the past.

Another method is double reading. This is a method in which another medical person reads a medical image already read by a medical person. In this method, since medical images are read by different medical personnel, reading mistakes by one medical person can be prevented. However, in a situation where the number of radiology medical personnel is limited, there is a problem in that the possibility of realization is low in reality.

Accordingly, there is a demand for an apparatus and method for managing the quality of medical image reading.

The background technology of the present invention is disclosed in Korean Patent Registration No. 10-1811028 (2017.12.20.) and Korean Patent Publication No. 10-2018-0040287 (2018.04.20.).

An image reading method according to an embodiment of the present disclosure includes steps of receiving a medical image, receiving report information that is a result of a medical practitioner's judgment on the medical image, and a method related to a lesion of a medical image acquired based on the medical image. Generating result information indicating the identity of the first lesion information and the second lesion information by applying the second lesion information related to the lesion of the medical image obtained based on the first lesion information and the report information to a third analysis model, and the result It is characterized in that it includes the step of outputting information.

The first lesion information of the image reading method according to an embodiment of the present disclosure includes a medical image, first lesion information indicating whether or not a lesion exists in the medical image, and the location and size of the lesion in the medical image. 1 characterized in that it includes at least one of lesion area information or first report information automatically generated by the first analysis model using medical images.

The first lesion information of the image reading method according to an embodiment of the present disclosure includes first feature information for at least one feature related to the hidden layer of the first analysis model for generating the first lesion presence information from the medical image. It is characterized by including.

The second lesion information of the image reading method according to an embodiment of the present disclosure includes report information, second lesion presence or absence information representing the presence or absence of a lesion acquired based on the report information, or the position of the lesion acquired based on the report information and It is characterized in that it includes at least one of the second lesion area information including the size of the lesion.

The second lesion information of the image reading method according to an embodiment of the present disclosure includes second feature information for at least one feature related to the hidden layer of the second analysis model for generating the second lesion presence information from the report information. It is characterized by including.

An image reading method according to an embodiment of the present disclosure includes obtaining first lesion information by applying a medical image to a first analysis model, and obtaining second lesion information by applying report information to a second analysis model. The first analysis model is a model obtained by machine learning the correlation between a plurality of past medical images and a plurality of first past lesion information on the plurality of past medical images, and the second analysis model is a model obtained by machine learning a plurality of past reports. and a model obtained by machine learning the correlation between a plurality of second historical lesion information and a plurality of past report information, and the third analysis model includes a plurality of first past lesion information, a plurality of second past lesion information, and a plurality of past lesion information. It is characterized by being a model obtained by machine learning past result information indicating the identity of the first past lesion information of and the plurality of second past lesion information.

The second analysis model of the image reading method according to an embodiment of the present disclosure is characterized in that it is a rule-based model for extracting the second lesion information from report information.

The generating of result information of the image reading method according to an embodiment of the present disclosure may include determining whether the first lesion information included in the first lesion information and the second lesion information included in the second lesion information match each other. It is characterized in that it comprises the step of determining the result information based on.

The generating of result information of the image reading method according to an embodiment of the present disclosure may include: first lesion area information included in the first lesion information and second lesion area information included in the second lesion information, respectively. and generating result information based on the degree of matching between regions of the liver.

In the step of generating result information of the medical image analysis method according to an embodiment of the present disclosure, the first feature information included in the first lesion information and the second feature information included in the second lesion information are converted into a machine learning model. 3 It is characterized in that it includes the step of generating result information by applying to the analysis model.

The step of generating result information of the image reading method according to an embodiment of the present disclosure includes applying the first report information obtained by generating the first lesion information as a report and the report information to a third analysis model, which is a machine learning model, to obtain the result information. Characterized in that it comprises the step of generating.

In the step of generating result information of the image reading method according to an embodiment of the present disclosure, the result information is obtained by applying second lesion area information included in the medical image and second lesion information to a third analysis model, which is a machine learning model. It is characterized in that it includes the step of generating.

An apparatus for analyzing a medical image according to an embodiment of the present disclosure includes a processor and a memory, and the processor receives a medical image based on a command stored in the memory, and receives report information that is a result of a medical practitioner's determination of the medical image. a step of applying first lesion information related to the lesion of the medical image obtained based on the medical image and second lesion information related to the lesion of the medical image obtained based on the report information to a third analysis model to obtain a first It is characterized by performing the step of generating result information indicating the identity of the lesion information and the second lesion information and the step of outputting the result information.

The first lesion information of the medical image analysis apparatus according to an embodiment of the present disclosure includes a medical image, first lesion information indicating the presence or absence of a lesion in the medical image, and the location and size of the lesion in the medical image. 1 characterized in that it includes at least one of lesion area information or first report information automatically generated by the first analysis model using medical images.

The first lesion information of the medical image analysis apparatus according to an embodiment of the present disclosure includes first feature information for at least one feature related to a hidden layer of a first analysis model for generating first lesion information from a medical image. It is characterized in that it includes.

The second lesion information of the medical image analysis apparatus according to an embodiment of the present disclosure includes report information, second lesion presence or absence information representing the presence or absence of a lesion acquired based on the report information, or the location of the lesion acquired based on the report information. and at least one of second lesion area information including the size of the lesion.

The second lesion information of the medical image analysis apparatus according to an embodiment of the present disclosure includes second feature information for at least one feature related to a hidden layer of a second analysis model for generating second lesion presence/absence information from report information. It is characterized in that it includes.

A method for evaluating a report of a medical image analysis apparatus according to an embodiment of the present disclosure includes obtaining first lesion information by applying a medical image to a first analysis model and applying the report information to a second analysis model to obtain first lesion information. The method further includes obtaining 2 lesion information, wherein the first analysis model is a model obtained by machine learning a plurality of past medical images and a correlation between a plurality of first past lesion information for the plurality of past medical images, and The model includes a model obtained by machine learning the correlation between a plurality of past report information and a plurality of second past lesion information for the plurality of past report information, and the third analysis model includes a plurality of first past lesion information and a plurality of second past lesion information. 2 It is characterized by being a machine learning model of past lesion information and past result information indicating the identity of the plurality of first past lesion information and the plurality of second past lesion information.

The second analysis model of the medical image analysis apparatus according to an embodiment of the present disclosure is characterized in that it is a rule-based model for extracting second lesion information from report information.

The processor of the medical image analysis apparatus according to an embodiment of the present disclosure matches first lesion information included in the first lesion information with second lesion information included in the second lesion information, based on a command stored in a memory. It is characterized in that performing the step of determining the result information based on whether or not.

The processor of the medical image analysis apparatus according to an embodiment of the present disclosure generates first lesion area information included in the first lesion information and second lesion area information included in the second lesion information, respectively, based on a command stored in a memory. and generating result information based on the degree of region matching between lesion regions according to the method.

The processor of the medical image analysis apparatus according to an embodiment of the present disclosure converts first feature information included in the first lesion information and second feature information included in the second lesion information to a machine learning model based on a command stored in a memory. It is characterized in that performing the step of generating result information by applying to the third analysis model.

The processor of the medical image analysis apparatus according to an embodiment of the present disclosure applies first report information generated by generating a report of first lesion information and the report information to a third analysis model, which is a machine learning model, based on a command stored in a memory. It is characterized in that performing the step of generating result information by doing.

The processor of the medical image analysis apparatus according to an embodiment of the present disclosure applies second lesion area information included in the medical image and second lesion information to a third analysis model, which is a machine learning model, based on a command stored in a memory. It is characterized by performing the step of generating result information.

Also, a program for implementing the above-described image reading method may be recorded in a computer-readable recording medium.

1 is a block diagram of a medical image analysis apparatus according to an embodiment of the present disclosure.
2 is a diagram illustrating a medical image analysis apparatus according to an embodiment of the present disclosure.
3 is a flowchart illustrating an operation of a medical image analysis apparatus according to an embodiment of the present disclosure.
4 is a block diagram illustrating a medical image analysis apparatus according to an embodiment of the present disclosure.
5 is a diagram for explaining a process of generating a machine learning model according to an embodiment of the present disclosure.
6 is a diagram for explaining a first analysis model according to an embodiment of the present disclosure.
7 is a diagram for explaining a second analysis model according to an embodiment of the present disclosure.
8 is a diagram for explaining a third analysis model according to an embodiment of the present disclosure.
9 is a diagram for explaining a region matching degree according to an embodiment of the present disclosure.
10 is a diagram illustrating an operation of a medical image analysis apparatus according to an embodiment of the present disclosure.
11 is a diagram illustrating an operation of a medical image analysis apparatus according to an embodiment of the present disclosure.
12 is a diagram illustrating an operation of a medical image analysis system according to an embodiment of the present disclosure.
13 is a diagram illustrating an operation of a medical image analysis system according to an embodiment of the present disclosure.

Advantages and features of the disclosed embodiments, and methods of achieving them, will become apparent with reference to the following embodiments in conjunction with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below and may be implemented in various different forms, only the present embodiments make the present disclosure complete, and those of ordinary skill in the art to which the present disclosure belongs It is provided only to fully inform the person of the scope of the invention.

Terms used in this specification will be briefly described, and the disclosed embodiments will be described in detail.

The terms used in this specification have been selected from general terms that are currently widely used as much as possible while considering the functions in the present disclosure, but they may vary according to the intention of a person skilled in the related field, a precedent, or the emergence of new technology. In addition, in a specific case, there is also a term arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the invention. Therefore, terms used in the present disclosure should be defined based on the meaning of the term and the general content of the present disclosure, not simply the name of the term.

Expressions in the singular number in this specification include plural expressions unless the context clearly dictates that they are singular. Also, plural expressions include singular expressions unless the context clearly specifies that they are plural.

When it is said that a certain part "includes" a certain component throughout the specification, it means that it may further include other components without excluding other components unless otherwise stated.

Also, the term “unit” used in the specification means a software or hardware component, and “unit” performs certain roles. However, "unit" is not meant to be limited to software or hardware. A “unit” may be configured to reside in an addressable storage medium and may be configured to reproduce on one or more processors. Thus, as an example, “unit” can refer to components such as software components, object-oriented software components, class components and task components, processes, functions, properties, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays and variables. Functionality provided within components and "parts" may be combined into fewer components and "parts" or further separated into additional components and "parts".

According to an embodiment of the present disclosure, “unit” may be implemented as a processor and a memory. The term “processor” should be interpreted broadly to include general-purpose processors, central processing units (CPUs), microprocessors, digital signal processors (DSPs), controllers, microcontrollers, state machines, and the like. In some circumstances, “processor” may refer to an application specific integrated circuit (ASIC), programmable logic device (PLD), field programmable gate array (FPGA), or the like. The term "processor" refers to a combination of processing devices, such as, for example, a combination of a DSP and a microprocessor, a combination of a plurality of microprocessors, a combination of one or more microprocessors in conjunction with a DSP core, or any other such configuration. may also refer to

The term "memory" should be interpreted broadly to include any electronic component capable of storing electronic information. The term memory includes random access memory (RAM), read-only memory (ROM), non-volatile random access memory (NVRAM), programmable read-only memory (PROM), erasable-programmable read-only memory (EPROM), electrical may refer to various types of processor-readable media, such as erasable PROM (EEPROM), flash memory, magnetic or optical data storage, registers, and the like. A memory is said to be in electronic communication with the processor if the processor can read information from and/or write information to the memory. Memory integrated with the processor is in electronic communication with the processor.

Hereinafter, with reference to the accompanying drawings, embodiments will be described in detail so that those skilled in the art can easily carry out the embodiments. And in order to clearly describe the present disclosure in the drawings, parts irrelevant to the description are omitted.

1 is a block diagram of a medical image analysis apparatus 100 according to an embodiment of the present disclosure.

Referring to FIG. 1 , a medical image analysis apparatus 100 according to an exemplary embodiment may include a data learner 110 and a data recognizer 120 . As described above, the medical image analysis apparatus 100 may include a processor and a memory.

The data learning unit 110 may learn a machine learning model for performing a target task using a data set. The data learning unit 110 may receive label information related to a data set and a target task. The data learning unit 110 may obtain a machine learning model by performing machine learning on the relationship between the data set and the label information. As an embodiment, the machine learning model obtained by the data learning unit 110 may be a model for evaluating the accuracy of medical image reading by a medical person based on the medical image and report information prepared by the medical person.

The data recognition unit 120 may store the machine learning model of the data learning unit 110 . The data recognizer 120 may apply a machine learning model to the medical image and the report information prepared by the medical person to output information related to the accuracy of the report information prepared by the medical person. In addition, the data recognizer 120 may use the result output by the machine learning model to update the machine learning model.

At least one of the data learning unit 110 and the data recognizing unit 120 may be manufactured in the form of at least one hardware chip and mounted in an electronic device. For example, at least one of the data learning unit 110 and the data recognizing unit 120 may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or an existing general-purpose processor (eg, CPU). Alternatively, it may be manufactured as a part of an application processor) or a graphics-only processor (eg GPU) and mounted in various electronic devices.

Also, the data learning unit 110 and the data recognizing unit 120 may be mounted on separate electronic devices. For example, one of the data learner 110 and the data recognizer 120 may be included in the electronic device, and the other may be included in the server. In addition, the data learning unit 110 and the data recognizing unit 120 may provide machine learning model information built by the data learning unit 110 to the data recognizing unit 120 through wired or wireless communication, and data recognition Data input to the unit 120 may be provided to the data learning unit 110 as additional learning data.

Meanwhile, at least one of the data learning unit 110 and the data recognizing unit 120 may be implemented as a software module. When at least one of the data learning unit 110 and the data recognizing unit 120 is implemented as a software module (or a program module including instructions), the software module is a memory or a computer-readable non-transitory module. It may be stored in a temporary readable recording medium (non-transitory computer readable media). Also, in this case, at least one software module may be provided by an Operating System (OS) or a predetermined application. Alternatively, some of the at least one software module may be provided by an Operating System (OS), and the other part may be provided by a predetermined application.

The data learning unit 110 according to an embodiment of the present disclosure includes a data acquisition unit 111, a pre-processing unit 112, a training data selection unit 113, a model learning unit 114, and a model evaluation unit 115. can include

The data acquisition unit 111 may acquire data necessary for machine learning. Since a lot of data is required for learning, the data acquisition unit 111 may receive a data set including a plurality of data.

Label information may be assigned to each of a plurality of data. The label information may be information describing each of a plurality of pieces of data. Label information may be information to be derived by a target task. Label information may be obtained from a user input, from a memory, or from a result of a machine learning model. For example, if the target task is to determine information related to the accuracy of report information prepared by medical personnel from medical images and report information prepared by medical personnel based on the medical images, the plurality of data includes a plurality of medical images and reports of a plurality of medical personnel. information and the label information will be the accuracy of the report information prepared by medical personnel.

The pre-processing unit 112 may pre-process the acquired data so that the received data can be used for machine learning. The pre-processing unit 112 may process the acquired data set into a preset format so that the model learning unit 114 to be described later may use it.

The learning data selector 113 may select data necessary for learning from preprocessed data. The selected data may be provided to the model learning unit 114 . The learning data selector 113 may select data necessary for learning from preprocessed data according to a predetermined criterion. In addition, the learning data selection unit 113 may select data according to a predetermined criterion by learning by the model learning unit 114 to be described later.

The model learning unit 114 may learn a criterion for which label information to output based on the data set. In addition, the model learning unit 114 may perform machine learning using a data set and label information for the data set as training data. In addition, the model learning unit 114 may perform machine learning by additionally using a previously obtained machine learning model. In this case, the previously acquired machine learning model may be a pre-built model. For example, the machine learning model may be a model built in advance by receiving basic learning data.

The machine learning model may be built in consideration of the application field of the learning model, the purpose of learning, or the computer performance of the device. The machine learning model may be, for example, a model based on a neural network. For example, models such as Deep Neural Network (DNN), Recurrent Neural Network (RNN), Long Short-Term Memory models (LSTM), Bidirectional Recurrent Deep Neural Network (BRDNN), and Convolutional Neural Networks (CNN) can be used as machine learning models. may, but is not limited thereto.

According to various embodiments, when a plurality of pre-built machine learning models exist, the model learning unit 114 may determine a machine learning model having a high correlation between the input learning data and the basic learning data as the machine learning model to be learned. there is. In this case, the basic learning data may be pre-classified for each type of data, and the machine learning model may be pre-built for each type of data. For example, the basic learning data may be pre-classified according to various criteria such as a place where the learning data was created, a time when the learning data was created, a size of the learning data, a creator of the learning data, and a type of object in the learning data.

In addition, the model learning unit 114 may train the machine learning model using a learning algorithm including, for example, error back-propagation or gradient descent.

In addition, the model learning unit 114 may learn a machine learning model through, for example, supervised learning using learning data as an input value. In addition, the model learning unit 114 performs, for example, unsupervised learning in which a criterion for a target task is discovered by self-learning the type of data necessary for the target task without any guidance. Through this, a machine learning model can be obtained. In addition, the model learning unit 114 may obtain a machine learning model through semi-supervised learning or active learning. In addition, the model learning unit 114 may learn the machine learning model through, for example, reinforcement learning using feedback about whether the result of the target task according to learning is correct.

In addition, when the machine learning model is learned, the model learning unit 114 may store the learned machine learning model. In this case, the model learning unit 114 may store the learned machine learning model in the memory of the electronic device including the data recognizing unit 120 . Alternatively, the model learning unit 114 may store the learned machine learning model in a memory of a server connected to the electronic device through a wired or wireless network.

The memory in which the learned machine learning model is stored may also store, for example, commands or data related to at least one other component of the electronic device. Also, the memory may store software and/or programs. Programs may include, for example, kernels, middleware, application programming interfaces (APIs) and/or application programs (or “applications”).

The model evaluation unit 115 may input evaluation data to the machine learning model and, if a result output from the evaluation data does not satisfy a predetermined criterion, may cause the model learning unit 114 to learn again. In this case, the evaluation data may be preset data for evaluating the machine learning model.

For example, the model evaluator 115 determines that a predetermined criterion is not satisfied when the number or ratio of evaluation data for which the recognition result is not accurate among the results of the machine learning model learned for the evaluation data exceeds a preset threshold. can be evaluated as For example, when a predetermined criterion is defined as a ratio of 2%, and a learned machine learning model outputs an erroneous recognition result for evaluation data exceeding 20 out of a total of 1000 evaluation data, the model evaluation unit 115 performs learning It can be evaluated that the applied machine learning model is not suitable.

On the other hand, when there are a plurality of learned machine learning models, the model evaluation unit 115 evaluates whether each learned video learning model satisfies a predetermined criterion, and a model that satisfies the predetermined criterion is used as the final machine learning model. can decide In this case, when there are a plurality of models that satisfy a predetermined criterion, the model evaluation unit 115 may determine one or a predetermined number of models preset in order of high evaluation scores as the final machine learning model.

Meanwhile, at least one of the data acquisition unit 111, the preprocessing unit 112, the learning data selection unit 113, the model learning unit 114, and the model evaluation unit 115 in the data learning unit 110 is at least one It can be manufactured in the form of a hardware chip and mounted in an electronic device. For example, at least one of the data acquisition unit 111, the pre-processing unit 112, the training data selection unit 113, the model learning unit 114, and the model evaluation unit 115 is artificial intelligence (AI). It may be manufactured in the form of a dedicated hardware chip for the computer, or may be manufactured as a part of an existing general-purpose processor (eg, CPU or application processor) or graphics-only processor (eg, GPU) and mounted in various electronic devices described above.

In addition, the data acquisition unit 111, the pre-processing unit 112, the training data selection unit 113, the model learning unit 114, and the model evaluation unit 115 may be mounted on one electronic device, or may be installed in separate electronic devices. It may also be mounted on each electronic device. For example, some of the data acquisition unit 111, the pre-processing unit 112, the training data selection unit 113, the model learning unit 114, and the model evaluation unit 115 are included in the electronic device, and the others are included in the electronic device. can be included in the server.

In addition, at least one of the data acquisition unit 111, the preprocessing unit 112, the learning data selection unit 113, the model learning unit 114, and the model evaluation unit 115 may be implemented as a software module. At least one of the data acquisition unit 111, the pre-processing unit 112, the training data selection unit 113, the model learning unit 114, and the model evaluation unit 115 is a software module (or a program including instructions) module), the software module may be stored in a computer-readable, non-transitory computer readable recording medium (non-transitory computer readable media). Also, in this case, at least one software module may be provided by an Operating System (OS) or a predetermined application. Alternatively, some of the at least one software module may be provided by an Operating System (OS), and the other part may be provided by a predetermined application.

The data recognition unit 120 according to an embodiment of the present disclosure includes a data acquisition unit 121, a preprocessing unit 122, a recognition data selection unit 123, a recognition result providing unit 124, and a model update unit 125. can include

The data acquisition unit 121 may receive input data. The preprocessing unit 122 may preprocess the acquired input data so that the acquired input data can be used in the recognition data selection unit 123 or the recognition result providing unit 124 .

The recognition data selection unit 123 may select necessary data from preprocessed data. The selected data may be provided to the recognition result provider 124 . The recognition data selection unit 123 may select some or all of the pre-processed data according to a predetermined criterion. In addition, the recognition data selection unit 123 may select data according to a standard set by learning by the model learning unit 114 .

The recognition result providing unit 124 may obtain result data by applying the selected data to the machine learning model. The machine learning model may be a machine learning model generated by the model learning unit 114 . The recognition result provider 124 may output result data.

The model updating unit 125 may update the machine learning model based on the evaluation of the recognition result provided by the recognition result providing unit 124 . For example, the model updater 125 may provide the recognition result provided by the recognition result provider 124 to the model learner 114 so that the model learner 114 updates the machine learning model. there is.

Meanwhile, at least one of the data acquisition unit 121, the pre-processing unit 122, the recognition data selection unit 123, the recognition result providing unit 124, and the model update unit 125 in the data recognizing unit 120, at least It is manufactured in the form of a single hardware chip and can be mounted in an electronic device. For example, at least one of the data acquisition unit 121, the pre-processing unit 122, the recognition data selection unit 123, the recognition result providing unit 124, and the model update unit 125 is artificial intelligence (AI). ), or may be manufactured as a part of an existing general-purpose processor (eg, CPU or application processor) or graphics-only processor (eg, GPU) and mounted in various electronic devices described above.

In addition, the data acquisition unit 121, the pre-processing unit 122, the recognition data selection unit 123, the recognition result providing unit 124, and the model update unit 125 may be mounted in one electronic device or separately. may be mounted on each of the electronic devices. For example, some of the data acquisition unit 121, the pre-processing unit 122, the recognition data selection unit 123, the recognition result providing unit 124, and the model update unit 125 are included in the electronic device, and the other part is included in the electronic device. may be included in the server.

In addition, at least one of the data acquisition unit 121, the preprocessing unit 122, the recognition data selection unit 123, the recognition result providing unit 124, and the model update unit 125 may be implemented as a software module. At least one of the data acquisition unit 121, the pre-processing unit 122, the recognition data selection unit 123, the recognition result providing unit 124, and the model update unit 125 includes a software module (or an instruction) When implemented as a program module), the software module may be stored in a computer-readable, non-transitory computer readable recording medium (non-transitory computer readable media). Also, in this case, at least one software module may be provided by an Operating System (OS) or a predetermined application. Alternatively, some of the at least one software module may be provided by an Operating System (OS), and the other part may be provided by a predetermined application.

Hereinafter, a method and apparatus for sequentially performing machine learning on data sets by the data learning unit 110 will be described in more detail.

2 is a diagram illustrating a medical image analysis apparatus according to an embodiment of the present disclosure.

The medical image analysis apparatus 100 may include a processor 210 and a memory 220 . The processor 210 may execute instructions stored in the memory 220 .

As described above, the medical image analysis apparatus 100 may include at least one of the data learning unit 110 and the data recognizing unit 120 . At least one of the data learning unit 110 and the data recognizing unit 120 may be implemented by the processor 210 and the memory 220 .

Hereinafter, the operation of the medical image analysis apparatus 100 will be described in more detail below.

3 is a flowchart illustrating an operation of a medical image analysis apparatus according to an embodiment of the present disclosure.

The medical image analysis apparatus 100 may perform step 310 of receiving a medical image from the medical imaging apparatus. The medical image may be at least one of a Computed Tomography (CT), X-RAY, Mammography, or Magnetic Resonance Imaging (MRI) image. However, it is not limited thereto, and the medical image may include a video or an image expressed as a graph. The image expressed as a graph may include an electrocardiogram (EKG) graph.

The medical image analysis apparatus 100 may receive a medical image from the medical imaging device wired or wirelessly. Here, the medical imaging device may include a medical imaging device and may include a picture archiving and communication system. Also, the medical image analysis apparatus 100 may acquire medical images stored in the memory 220 .

The medical image analysis apparatus 100 may perform step 320 of receiving report information, which is a result of a medical practitioner's determination of a medical image. The medical image analysis apparatus 100 may receive report information wired or wireless. The medical image analysis apparatus 100 may receive report information from an input unit (not shown).

The medical image analysis apparatus 100 may receive medical images and report information in real time. That is, the medical image analysis apparatus 100 may receive medical image or report information as soon as it is generated. In addition, the medical imaging apparatus, gateway, or server may collect a plurality of report information or a plurality of medical images and transmit them to the medical image analysis apparatus 100 in non-real time, so that the medical image analysis apparatus 100 collects a plurality of report information or a plurality of medical images. It is possible to receive medical images in non-real time. Also, the medical image analysis apparatus 100 may automatically or manually receive medical images and report information.

The report information may be an analysis result prepared by a medical person by analyzing a medical image. Report information may include text information. Also, the report information may include window information of a medical image. Window information may include window center information and window width information. The window center information and the window width information may be information for adjusting brightness and contrast of a medical image. Also, the report information may include various numerical information related to medical images. Also, the report information may include at least one of lesion information and patient information. Also, the report information may include marker information displayed on an image. The marker information may include at least one pixel value or marker location information. The marker information may indicate the position of the region of interest or the length of the region of interest marked on the image by a medical practitioner.

The information about the lesion may include at least one of the name, type, size, location, and severity of the lesion shown in the medical image. Also, the patient information may include at least one of the patient's medical history, age, gender, and family history.

The medical image analysis apparatus 100 applies first lesion information based on the medical image and second lesion information based on the report information to a third analysis model to generate result information representing the sameness of the first lesion information and the second lesion information. Step 330 may be performed. The first lesion information and the second lesion information may be information related to the lesion of the medical image. The medical image analysis apparatus 100 applies the first lesion information obtained based on the medical image and the second lesion information obtained based on the report information to a third analysis model to determine the identity of the first lesion information and the second lesion information. It is possible to generate result information indicating. The first lesion information, the second lesion information, and the third analysis model will be described in detail together with FIGS. 4 to 9 .

The result information may indicate the degree of identity or similarity between the first lesion information and the second lesion information. According to an embodiment of the present disclosure, the medical image analysis apparatus 100 may digitally indicate the identity of the first lesion information and the second lesion information. For example, the medical image analysis apparatus 100 may indicate result information as '1' when the first lesion information and the second lesion information are identical, and may indicate result information as '0' when they are not identical. However, it is not limited thereto, and the medical image analysis apparatus 100 may indicate the result information as '0' when the first lesion information and the second lesion information are the same, and may indicate the result information as '1' when they are not identical. .

According to another embodiment of the present disclosure, the medical image analysis apparatus 100 may indicate a similarity between the first lesion information and the second lesion information as a real number or a natural number. For example, the medical image analysis apparatus 100 may display a higher number as the first lesion information and the second lesion information are similar, and may display a lower number as the number of dissimilarities. However, the present invention is not limited thereto, and the medical image analysis apparatus 100 may display a lower number as the first lesion information and the second lesion information are similar, and may display a higher number as the number is dissimilar.

The medical image analysis apparatus 100 may determine result information indicating whether the first lesion information and the second lesion information are identical by comparing the degree of similarity with a threshold value. For example, when the degree of similarity is higher than a threshold value, the medical image analysis apparatus 100 may indicate that the first lesion information and the second lesion information are identical as result information. Also, when the degree of similarity is lower than the threshold value, the medical image analysis apparatus 100 may display result information indicating that the first lesion information and the second lesion information are not identical. However, the medical image analysis apparatus 100 is not limited thereto, and when the similarity is higher than the threshold value, the result information may indicate that the first lesion information and the second lesion information are not the same, and the similarity is the threshold value. If it is lower than that, the result information may indicate that the first lesion information and the second lesion information are identical.

Also, the medical image analysis apparatus 100 may perform step 340 of outputting result information. Result information may be transmitted wired or wirelessly to a hospital server, a medical practitioner's terminal, a patient's terminal, or a payer server. Result information may be displayed on the display of the medical image analysis apparatus 100 or output through a speaker.

When the report information of the medical person indicates that there is no lesion in the medical image, the medical image analysis apparatus 100 may obtain result information about the medical image based on the steps of FIG. 3 . The medical image analysis apparatus 100 may determine whether a lesion is present in the medical image based on result information. That is, it may be determined that the medical personnel's report information 421 may be incorrect. For example, the medical personnel's report information may indicate that the medical image is normal. However, the medical image analysis apparatus 100 may determine that there is microcalcification in the medical image. The medical image analysis apparatus 100 may display lesions discovered by the medical image analysis apparatus 100 but not found by a medical person. Therefore, the medical image analysis apparatus 100 can prevent omission of lesions by repeatedly reviewing medical images read by medical personnel.

Also, when the report information of the medical person indicates that there is a lesion in the medical image, the medical image analysis apparatus 100 may obtain result information about the medical image based on the steps of FIG. 3 . Based on the result information, the medical image analysis apparatus 100 may determine that a lesion exists in the medical image, but a lesion is missing from the report information or that the lesion described in the report information is incorrect. For example, in the report information of a medical person, only a pneumothorax was found in a chest x-ray, but the medical image analysis apparatus 100 may find a pneumothorax and a lung nodule based on a machine learning model. there is.

The medical image analysis apparatus 100 may output result information so that medical personnel take necessary measures. For example, a medical person may review a medical image based on result information. Alternatively, the medical image analysis apparatus 100 may change a work list of medical personnel so that a medical person other than the medical person who read the medical image reviews the medical image. Alternatively, the apparatus 100 for analyzing medical images may change a work list of medical personnel to allow a plurality of medical personnel to review medical images. Also, the health care provider may suggest additional tests to the patient. Accordingly, the medical image analysis apparatus 100 may improve the analysis result of the medical image.

4 is a block diagram illustrating a medical image analysis apparatus according to an embodiment of the present disclosure.

The medical image analysis apparatus 100 may include a first analysis model, a second analysis model, or a third analysis model. The first analysis model or the second analysis model may be omitted.

The medical image analysis apparatus 100 may receive the medical image 411 . The medical image analysis apparatus 100 may perform a step of obtaining first lesion information by applying the medical image 411 to the first analysis model 410 . The first analysis model 410 may be a model obtained by machine learning a correlation between a plurality of past medical images and a plurality of first past lesion information corresponding to the plurality of past medical images.

Here, the 'past' may mean learning data for performing machine learning. For example, the plurality of past medical images and the plurality of first past lesion information may refer to learning data used for machine learning of the first analysis model 410 .

The medical image analysis apparatus 100 may omit the first analysis model 410 . In this case, the first lesion information 412 may include a medical image 411 .

Also, the medical image analysis apparatus 100 may apply the report information 421 to the second analysis model 420 to obtain second lesion information 422 . The second analysis model 420 may include a model obtained by machine learning a plurality of past report information and a correlation between a plurality of second past lesion information corresponding to the plurality of past report information. Here, the 'past' may mean learning data for performing machine learning. For example, the plurality of past report information and the plurality of second past lesion information may refer to learning data used for machine learning of the second analysis model 420 .

Also, the second analysis model 420 may not be a machine learning model but may be a rule based model that extracts information from report information. For example, the second analysis model 420 may be a model for determining whether lesion information is included in the report information. Here, the lesion information may include various types of information related to the lesion. For example, the lesion information may include at least one of the name, type, location, and size of the lesion.

The medical image analysis apparatus 100 may omit the second analysis model 420 . In this case, the second lesion information may include report information 421 .

The medical image analysis apparatus 100 may perform a step of obtaining result information 430 by applying the first lesion information 412 and the second lesion information 422 to the third analysis model 400 . The third analysis model 400 converts first plurality of past lesion information, second plurality of past lesion information, and past result information representing past similarities between the first plurality of past lesion information and the second plurality of past lesion information into a machine. It can be a trained model. Here, the 'past' may mean learning data for performing machine learning. For example, the plurality of first past lesion information, the plurality of second past lesion information, and the past result information may refer to learning data used for machine learning of the third analysis model 400 .

The first lesion information may include a medical image, first lesion information indicating the presence or absence of a lesion in the medical image, information on the type of the first lesion, first lesion area information including the location and size of the lesion in the medical image, or It may include at least one of first report information automatically generated by the first analysis model using the medical image.

As already described, the medical image analysis apparatus 100 may not include the first analysis model 410 . When the medical image analysis apparatus 100 does not include the first analysis model 410 , the first lesion information 412 may be the medical image 411 . The third analysis model 400 may receive the medical image 411 .

The medical image analysis apparatus 100 may apply the medical image 411 to the first analysis model 410 to generate first lesion presence/absence information, which is one-dimensional information. The first analysis model 410 may be a machine-learned model of a relationship between past medical images and past label information indicating whether a lesion exists in the past medical images.

As an embodiment, the first lesion presence information may have a value of '0' or '1'. For example, when the first analysis model 410 finds a lesion in the medical image 411, the first lesion presence information may be '1'. In addition, when the first analysis model 410 does not find a lesion in the medical image 411, the first lesion presence information may be '0'. However, it is not limited thereto, and when the first analysis model 410 finds a lesion in the medical image 411, the first lesion presence information may be '0', and the first analysis model 410 determines the medical image ( If no lesion is found in 411), the first lesion presence information may be '1'. The third analysis model 400 may receive first lesion information.

Also, the medical image analysis apparatus 100 may apply the medical image 411 to the first analysis model 410 to acquire information about the type of the first lesion. The first analysis model 410 may be a machine-learned model of a relationship between past medical images and information on types of past lesions present in the past medical images.

The information on the type of the first lesion may include a code representing the lesion or a name of the lesion. If the first analysis model 410 does not find a lesion from the medical image, information about the type of the first lesion may have a value of '0' or 'null'.

The first lesion information 412 may include first lesion area information. The medical image analysis apparatus 100 may obtain first lesion area information by applying the medical image 411 to the first analysis model 410 . The first analysis model 410 may be a machine-learned model of a relationship between past medical images and location information of past lesions corresponding to the past medical images.

According to an embodiment of the present disclosure, the first lesion area information may be 2-dimensional information. The first lesion area information may be displayed as a marker on the medical image 411 . The marker indicates the location and size of a lesion in the medical image 411 and may be expressed using at least one pixel included in the medical image 411 . A marker may have a shape such as a circle, an oval, a square or a triangle.

According to an embodiment of the present disclosure, the first lesion area information may indicate the location and size of the lesion. For example, the first lesion area information may include coordinate values representing the location of the lesion. Also, the first lesion area information may include a value representing the size of the lesion. The size may include at least one of a radius, a diameter, an area, a volume, or a side length. The third analysis model 400 may receive first lesion area information.

The first lesion information 412 may include first report information automatically generated by the first analysis model using the medical image. The first analysis model 410 may be a model obtained by machine learning a relationship between past medical images and past report information related to lesions included in the past medical images.

The medical image analysis apparatus 100 may generate first report information on the medical image by applying the first analysis model 410 to the medical image. A medical person may modify first report information automatically generated by the medical image analysis apparatus 100 without having to newly prepare a report. In addition, the medical personnel may determine the modified report information as a final diagnosis result and input the modified report information to the medical image analysis apparatus 100 . The final diagnosis result may be report information 421. Also, the third analysis model 400 may receive first report information.

The second lesion information 422 includes report information, second lesion presence/absence information indicating the presence or absence of a lesion based on the report information, information on the second lesion type, or the location of the lesion or the size of the lesion obtained based on the report information. and at least one of the second lesion area information to be included.

As already described, the medical image analysis apparatus 100 may not include the second analysis model 420 . When the medical image analysis apparatus 100 does not include the second analysis model 420 , the second lesion information 422 may be report information 421 . The third analysis model 400 may receive report information 421 .

The medical image analysis apparatus 100 may apply the report information 421 to the second analysis model 420 to generate second lesion presence/absence information that is one-dimensional information. The second analysis model 420 may be a machine-learned model of a relationship between past report information and label information indicating whether information related to a lesion exists in the past report information.

As an embodiment, the second lesion presence information may have a value of '0' or '1'. For example, when the second analysis model 420 finds information indicating that there is a lesion in the medical image 411 from the report information 421, the second lesion presence/absence information may be '1'. Also, when the second analysis model 420 does not find information indicating that there is a lesion in the medical image 411 from the report information 421, the second lesion presence/absence information may be '0'. However, it is not limited thereto, and when the second analysis model 420 finds information indicating that there is a lesion in the medical image 411 from the report information 421, the second lesion presence/absence information may be '0', If the second analysis model 420 does not find information indicating that there is a lesion in the medical image 411 from the report information 421, the second lesion presence information may be '1'. The third analysis model 400 may receive the second lesion presence/absence information.

Also, the medical image analysis apparatus 100 may apply the report information 421 to the second analysis model 420 to acquire information about the type of the second lesion. The second analysis model 420 may be a machine-learned model of a relationship between past report information and information on the type of past lesion present in the past report information.

The information on the type of the second lesion may include a code representing the lesion or a name of the lesion. When the second analysis model 420 does not find a lesion from the report information, information on the type of the second lesion may have a value of '0' or 'null'.

The second lesion information 422 may include second lesion area information. The medical image analysis apparatus 100 may acquire second lesion area information by applying the report information 421 to the second analysis model 420 . The second analysis model 420 may be a machine-learned model of a relationship between past report information and location information of a lesion described in the past report information.

According to an embodiment of the present disclosure, the second lesion area information may be 2-dimensional information. The second lesion area information may be displayed as a marker on the medical image 411 or the template image. The marker indicates the location and size of the lesion in the medical image 411 and may be expressed using at least one pixel included in the medical image 411 or the template image. The markers may have various shapes such as round, oval, square or triangular. The template image is an image corresponding to a patient's body part shown in the medical image 411 and may be stored in the memory of the medical image analysis apparatus 100 .

According to an embodiment of the present disclosure, the second lesion area information may indicate the location and size of the lesion. For example, the second lesion area information may include coordinate values representing the location of the lesion. Also, the second lesion area information may include a value representing the size of the lesion. The size may include at least one of a radius, a diameter, an area, a volume, or a side length. The third analysis model 400 may receive second lesion area information.

5 is a diagram for explaining a process of generating a machine learning model according to an embodiment of the present disclosure.

As already described, the first analysis model 410 and the second analysis model 420 may be machine learning models.

According to an embodiment of the present disclosure, the medical image analysis apparatus 100 may generate first lesion presence information by applying the medical image 411 to the first analysis model 410 . The first analysis model 410 may be a model obtained by machine learning a relationship between a plurality of past medical images and a plurality of past lesion presence/absence information corresponding to each of the plurality of past medical images.

5 is a block diagram illustrating a process in which the medical image analysis apparatus 100 generates a machine learning model based on a plurality of data. The medical image analysis apparatus 100 may receive input data 510 and output data 530 to acquire the first analysis model 410 . Input data 510 may be a plurality of past medical images. Also, the output data 530 may be information on whether or not a plurality of past lesions exist. The medical image analysis apparatus 100 uses the hidden layer 520 by using feed forward and back propagation in order to machine learn a model in which output data 530 is generated from input data 510. can be updated The medical image analysis apparatus 100 may generate the first analysis model 410 by updating the hidden layer 520 based on a plurality of past medical images and a plurality of past lesion information.

The medical image analysis apparatus 100 may generate first lesion presence information by applying the medical image 411 to the generated first analysis model 410 . That is, output data 530 may be generated through the hidden layer 520 by using a medical image as the input data 510 and forward propagation. The output data 530 may be first lesion information. The medical image analysis apparatus 100 may determine first feature information included in the hidden layer 520 immediately before the output data 530 as first lesion information. That is, the first lesion information may include first feature information related to the hidden layer 520 of the first analysis model for generating the first lesion information from the medical image. For example, the first feature information may include values included in an uppermost layer of the hidden layer 520 when the medical image is forward propagated.

Similarly, an embodiment included in the second lesion information will be described below. The medical image analysis apparatus 100 may receive input data 510 and output data 530 to acquire the second analysis model 420 . Input data 510 may be a plurality of past report information. In addition, the output data 530 may be a plurality of past lesion status information described in each of a plurality of past report information. The medical image analysis apparatus 100 uses the hidden layer 520 by using feed forward and back propagation in order to machine learn a model in which output data 530 is generated from input data 510. can be updated The medical image analysis apparatus 100 may generate the second analysis model 420 by updating the hidden layer 520 based on a plurality of past report information and a plurality of past lesion information.

The medical image analysis apparatus 100 may generate second lesion presence information by applying the report information 421 to the generated second analysis model 420 . That is, the report information 421 can be used as the input data 510, and the output data 530 can be generated through the hidden layer 520 by forward propagation. The medical image analysis apparatus 100 may determine second feature information included in the hidden layer 520 and immediately before the output data 530 as second lesion information. That is, the second lesion information may include second feature information for at least one feature related to the hidden layer of the second analysis model for generating the second lesion presence information from the report information. For example, the second feature information may include values included in an uppermost layer of the hidden layer 520 when the report information is forward propagated.

The medical image analysis apparatus 100 may generate a third analysis model obtained by machine learning a relationship between the first feature information, the second feature information, and label information indicating a similarity between the first feature information and the second feature information. . The medical image analysis apparatus 100 obtains the similarity between the lesion information included in the medical image and the lesion information described in the report information by using the first analysis model, the second analysis model, and the third analysis model generated as described above. can do.

Also, the medical image analysis apparatus 100 may receive input data 510 and output data 530 to acquire the third analysis model 400 . The input data 510 may be a plurality of first past lesion information and a plurality of second past lesion information. In addition, the output data 530 may be a plurality of past result information indicating the identity of the plurality of first past lesion information and the plurality of second past lesion information. The medical image analysis apparatus 100 uses the hidden layer 520 by using feed forward and back propagation in order to machine learn a model in which output data 530 is generated from input data 510. can be updated The medical image analysis apparatus 100 generates a third analysis model 400 by updating the hidden layer 520 based on the plurality of first past lesion information, the plurality of second past lesion information, and the plurality of past result information. can do.

6 is a diagram for explaining a first analysis model according to an embodiment of the present disclosure.

The first analysis model 410 may be a model obtained by machine learning a correlation between a plurality of past medical images 611 and a plurality of first past lesion information 612 corresponding to the plurality of past medical images. The plurality of past medical images 611 and the plurality of first past lesion information 612 corresponding to the plurality of past medical images may be ground truth information. Actual information may be information that has been found to be correct by a skilled expert.

The plurality of past medical images 611 and the plurality of first past lesion information 612 corresponding to the plurality of past medical images may be obtained from an external server. For example, the external server may include at least one of a hospital server, a health insurance server, and a server of a national institution.

The plurality of first past lesion information 612 may be a result of medical personnel diagnosing a plurality of past medical images 611 . However, it is not limited thereto, and the plurality of first past lesion information 612 may be information generated by a machine learning model based on a plurality of past medical images 611 . The plurality of first past lesion information 612 includes information on whether or not a lesion exists in the plurality of past medical images 611, information on the type of lesion, information on the location of the lesion, information on the size of the lesion, or information on the size of the lesion. It may include at least one of past report information generated by medical personnel based on a plurality of past medical images 611 .

The medical image analysis apparatus 100 uses the data learning unit 110 of FIG. 1 together with a plurality of past medical images 611 and a plurality of first past lesion information 612 to generate a first analysis model 410. can create

The medical image analysis apparatus 100 may store the first analysis model 410 in a memory or may transmit the first analysis model 410 to another medical image analysis apparatus wired or wirelessly.

The medical image analysis apparatus 100 may receive the medical image 411 . The medical image analysis apparatus 100 may derive first lesion information 412 by applying the first analysis model 410 to the medical image 411 . The medical image analysis apparatus 100 may automatically acquire first lesion information 412 based on the first analysis model 410 without the help of a medical person.

The first lesion information includes a medical image, first lesion information indicating the presence or absence of a lesion in the medical image, first lesion area information indicating the location and size of the lesion in the medical image, or the first lesion information using the medical image. It may include at least one of first report information automatically generated by the analysis model. Also, the first lesion information may include first feature information related to a hidden layer of the first analysis model for generating first lesion information from a medical image. Since the first feature information has been described above, a detailed description thereof will be omitted.

The medical image analysis apparatus 100 may provide first lesion information 412 to medical personnel. The medical personnel may generate final diagnosis information by referring to the first lesion information 412 .

7 is a diagram for explaining a second analysis model according to an embodiment of the present disclosure.

The second analysis model 420 may be a model obtained by machine learning the correlation between the plurality of past report information 711 and the plurality of second past lesion information 712 corresponding to the plurality of past report information. The plurality of past report information 711 and the plurality of second past lesion information 712 may be ground truth information.

The plurality of past report information 711 may be information received from medical personnel. A medical person may analyze a medical image, prepare at least one piece of past report information, and store it in an external server. At least one piece of past report information may include at least one of digital text information and image information. In addition, at least one piece of past report information may include a diagnosis result written in a person's handwriting or a comment written by a medical person on a medical image.

The external server may accumulate and store at least one past report information. The plurality of past report information 711 may be obtained from an external server. For example, the external server may include at least one of a hospital server, a health insurance server, and a server of a national institution.

The plurality of second past lesion information 712 corresponding to the plurality of past report information may be label information for the past report information. The plurality of second past lesion information 712 may be information prepared by a person or information automatically extracted from report information based on a machine learning model. The plurality of second past lesion information 712 includes information indicating whether information related to a lesion exists in past report information, information on the type of lesion derived from past reports, or a location where a lesion derived from past report information exists. Alternatively, at least one of information indicating the size of the lesion may be included.

The medical image analysis apparatus 100 uses the data learning unit 110 of FIG. 1 together with the plurality of past report information 711 and the plurality of second past lesion information 712 to generate the second analysis model 420. can create

The medical image analysis apparatus 100 may store the second analysis model 420 in a memory or may transmit the second analysis model 420 to another medical image analysis apparatus wired or wirelessly.

The medical image analysis apparatus 100 may receive report information 421 . The medical image analysis apparatus 100 may derive second lesion information 422 by applying the second analysis model 420 to the report information 421 . The medical image analysis apparatus 100 may automatically obtain computer-readable second lesion information 422 from the report information 421 based on the second analysis model 420 .

The second lesion information 422 includes report information, second lesion presence or absence information indicating the presence or absence of a lesion based on the report information, or second lesion area information indicating the location of a lesion or the size of a lesion based on the report information. may contain at least one. The second lesion information may include second characteristic information related to the hidden layer of the second analysis model for generating the second lesion presence information from the report information. Since the second characteristic information has been described above, a detailed description thereof will be omitted.

The medical image analysis apparatus 100 may acquire first lesion information 412 and second lesion information 422 through the processes described in FIGS. 6 and 7 . The medical image analysis apparatus 100 may generate, as result information, whether the first lesion information 412 and the second lesion information 422 are identical based on the third analysis model.

8 is a diagram for explaining a third analysis model according to an embodiment of the present disclosure.

The third analysis model 400 indicates the identity of the plurality of first past lesion information 811, the plurality of second past lesion information 812, the plurality of first past lesion information, and the plurality of second past lesion information. It may be a model obtained by machine learning the correlation of the past result information 813 . The plurality of first past lesion information 811 , the plurality of second past lesion information 812 , and the past result information 813 may be ground truth information.

The plurality of first past lesion information 811 may be a result of medical personnel diagnosing a plurality of past medical images 611 . However, it is not limited thereto, and the first past lesion information 811 may be information generated by a machine learning model based on a past medical image 611 . For example, the first past lesion information 811 may be information generated by the first analysis model 410 of FIG. 6 . The first past lesion information 811 is information on whether a lesion exists in the past medical image 611, information on the type of lesion, information on the location of the lesion, information on the size of the lesion, or At least one of past report information generated based on the image 611 may be included.

Also, the first past lesion information 811 may include first past feature information related to the hidden layer of the first analysis model 410 . As described above, the first analysis model 410 may be a model for obtaining first past lesion presence/absence information from past medical images. The first past feature information may include values included in a hidden layer of the first analysis model 410 when past medical images are applied to the first analysis model 410 . Since the first past feature information is similar to the previously described first feature information, a duplicate description is omitted.

The plurality of second past lesion information 812 may be label information for past report information. The plurality of second past lesion information 812 may be information prepared by a person or information automatically extracted from report information based on a machine learning model. For example, the plurality of second past lesion information 812 may be information generated by the second analysis model 420 of FIG. 7 . The plurality of second past lesion information 812 includes information indicating whether information related to a lesion exists in past report information, information on the type of lesion derived from past reports, or a location where a lesion derived from past report information exists. Alternatively, at least one of information indicating the size of the lesion may be included.

Also, the second past lesion information may include second past feature information related to the hidden layer of the second analysis model 420 . As described above, the second analysis model 420 may be a model for obtaining second past lesion information from past report information. The second past feature information may include values included in the hidden layer of the second analysis model 420 when past report information is applied to the second analysis model 420. The second past feature information has already been described. Since it is similar to the second feature information, duplicate descriptions are omitted.

The past result information 813 indicating the identity of the first plurality of past lesion information and the second plurality of past lesion information includes one of the plurality of first past lesion information 911 and the corresponding second plurality of past lesion information. It may be information indicating the identity or similarity of (912).

The past result information 813 may indicate the degree of identity or similarity between the first past lesion information and the second past lesion information. The past result information 813 may be generated by a user's input or automatically generated by a machine learning model. According to an embodiment of the present disclosure, the medical image analysis apparatus 100 may digitally display past result information 813 indicating the identity of first past lesion information and second past lesion information. For example, the medical image analysis apparatus 100 displays the past result information 813 as '1' when the first past lesion information and the second past lesion information are identical, and displays the past result information 813 when they are not identical. It can be represented as '0'. However, it is not limited thereto, and the medical image analysis apparatus 100 displays the past result information 813 as '0' when the first past lesion information and the second past lesion information are the same, and when they are not the same, the past result information ( 813) can be represented as '1'.

According to another embodiment of the present disclosure, the medical image analysis apparatus 100 may indicate the past result information 813 indicating the similarity between the first past lesion information and the second past lesion information as a real number or a natural number. For example, the medical image analysis apparatus 100 may indicate the past result information 813 as a higher number as the first past lesion information and the second past lesion information are similar, and as a dissimilar number, as a lower number. However, it is not limited thereto, and the medical image analysis apparatus 100 displays the past result information 813 as a lower number as the first past lesion information and the second past lesion information are similar, and as a dissimilar number, as a higher number. can

The medical image analysis apparatus 100 indicates the identity of the plurality of first past lesion information 811, the plurality of second past lesion information 812, and the plurality of first past lesion information and the plurality of second past lesion information. The third analysis model 400 may be generated by machine learning the past result information 813 .

However, it is not limited thereto, and the third analysis model 400 may not be a machine learning model. That is, the third analysis model 400 may be a rule-based model. The third analysis model may be a model for determining whether the first lesion information 412 and the second lesion information 422 are simply the same.

In the third analysis model 400 using a machine learning model, even if the first lesion information 412 and the second lesion information 422 have different formats, the first lesion information 412 and the second lesion information 422 are It can be determined whether or not they are substantially the same. In the present disclosure, when the first lesion information and the second lesion information have different formats, it should be understood that the third analysis model uses a machine learning model. In addition, in the case of the third analysis model 400 that does not use a machine learning model, the third analysis model 400 can be implemented simply, and the processing speed can be relatively fast.

The medical image analysis apparatus 100 may store the third analysis model 400 in a memory or may transmit the third analysis model 400 to another medical image analysis apparatus wired or wirelessly.

The medical image analysis apparatus 100 may perform a step of obtaining result information 430 by applying the first lesion information 412 and the second lesion information 422 to the third analysis model 400 . Since the first lesion information 412 and the second lesion information 422, which are inputs of the third analysis model 400, and the result information 430, which are outputs of the third analysis model 400, have been described together with FIG. 4, Redundant descriptions are omitted.

Hereinafter, the result information 430 that is the output of the third analysis model 400 will be described in more detail.

According to an embodiment of the present disclosure, in the medical image analysis apparatus 100, the first lesion information included in the first lesion information 412 and the second lesion information included in the second lesion information 422 match. The step of determining the result information based on whether or not it is performed may be performed. In this case, the third analysis model may not be a model generated by machine learning.

For example, the first lesion information included in the first lesion information 412 may have a value of '0' or '1'. Also, the second lesion information included in the second lesion information 422 may have a value of '0' or '1'.

The medical image analysis apparatus 100 may simply determine whether the first lesion presence information and the second lesion presence information are the same. For example, when the first lesion presence information is '0' and the second lesion presence information is '0', the medical image analysis apparatus 100 may set the result information 430 to indicate 'same'. Also, when the first lesion presence information is '1' and the second lesion presence information is '1', the medical image analysis apparatus 100 may make the result information 430 indicate 'same'. Also, when the first lesion presence information is '0' and the second lesion presence information is '1', the medical image analysis apparatus 100 may make the result information 430 indicate 'not the same'. In addition, when the first lesion presence information is '1' and the second lesion presence information is '0', the medical image analysis apparatus 100 may make the result information 430 indicate 'not the same'.

According to another embodiment of the present disclosure, the medical image analysis apparatus 100 applies a degree of region matching to first lesion area information included in the first lesion information and second lesion area information included in the second lesion information, and results are obtained. Steps to generate information can be performed. This will be described with reference to FIG. 9 .

9 is a diagram for explaining a region matching degree according to an embodiment of the present disclosure.

The medical image analysis apparatus 100 may include first lesion information 412 based on the medical image 411 . The first lesion information 412 may include first lesion area information. The first lesion area information may include information related to the size of the lesion, the location of the lesion, or the shape of the lesion. Referring to FIG. 9 , the medical image analysis apparatus 100 may determine a first lesion area 910 based on first lesion area information.

The medical image analysis apparatus 100 may include second lesion information 422 based on the report information 421 . The second lesion information 422 may include second lesion area information. The second lesion area information may include information related to the size of the lesion, the location of the lesion, or the shape of the lesion. Referring to FIG. 9 , the medical image analysis apparatus 100 may determine a second lesion area 920 based on second lesion area information.

According to an embodiment of the present disclosure, the region matching degree may include an intersection over union (IOU) method. Also, the medical image analysis apparatus 100 may obtain result information 430 without a third analysis model based on the degree of region matching. The region matching degree is a method of measuring how much two different regions overlap each other. The medical image analysis apparatus 100 determines result information closer to 1 as the first lesion area 910 and the second lesion area 920 are the same based on the degree of matching of the regions, and And when the second lesion area 920 is not the same, result information may be determined closer to 0.

The medical image analysis apparatus 100 may generate result information based on the union region 930 and the intersection region 940 of the first lesion region 910 and the second lesion region 920 . For example, the medical image analysis apparatus 100 may generate result information according to Equation 1.

Result information = (area of intersection area 940 of first lesion area information and second lesion area information)/(area of intersection area 930 of first lesion area information and second lesion area information) ... ( Equation 1)

However, it is not limited to generating result information according to Equation 1 as above. According to another embodiment of the present disclosure, the medical image analysis apparatus 100 may obtain result information by applying the first lesion area information and the second lesion area information to the third analysis model 400 that is a machine learning model. . Result information may represent information about identity.

The third analysis model 400 may be a machine-learned model of first past lesion area information, second past lesion area information, and past identity information. The information on the past identity may be a value input by the user to the medical image analysis apparatus 100 based on the first past lesion area information and the second past lesion area information. Alternatively, information on past identity may be a value obtained by Equation 1.

The third analysis model 400 may receive the first lesion area information and the second lesion area information and output result information indicating information on identity.

Referring back to FIG. 8 , the medical image analysis apparatus 100 converts the first feature information included in the first lesion information 412 and the second feature information included in the second lesion information 422 into a machine learning model. 3 A step of generating result information 430 by applying the analysis model may be performed.

The first characteristic information has been described together with FIGS. 5 and 6 . The medical image analysis apparatus 100 may acquire first lesion presence information by applying the medical image 411 to the first analysis model 410 . When the medical image 411 is applied to the first analysis model 410, the medical image analysis apparatus 100 may determine immediately preceding values of the first lesion presence information as first feature information. The first feature information may be a value included in the hidden layer of the first analysis model 410 .

The second characteristic information has been described together with FIGS. 5 and 7 . The medical image analysis apparatus 100 may apply the report information 421 to the second analysis model 420 to obtain second lesion presence information. When applying the report information 421 to the second analysis model 420, the medical image analysis apparatus 100 may determine immediately preceding values of the second lesion presence/absence information as the second feature information. The second feature information may be a value included in the hidden layer of the second analysis model 420 .

Referring to FIG. 8 , the third analysis model 400 includes first past feature information included in the plurality of first past lesion information 811 and second past feature information included in the plurality of second past lesion information 812 . It may be a model obtained by machine learning information related to the identity of the first past feature information and the second past feature information included in the information and past result information 813 .

The third analysis model 400 may receive first feature information included in the first lesion information 412 and second feature information included in the second lesion information 422 as inputs. The third analysis model 400 may output result information 430 indicating the identity of the first characteristic information and the second characteristic information.

According to another embodiment of the present disclosure, the first lesion information may include first report information. The medical image analysis apparatus 100 may perform a step of generating result information by applying the first report information obtained by generating the first lesion information as a report and the report information to a third analysis model that is a machine learning model.

Referring to FIG. 6 , the medical image analysis apparatus 100 may obtain first report information as first lesion information 412 by applying the medical image 411 to the first analysis model 410 . The first report information may be a report automatically obtained based on the medical image 411 .

Referring to FIG. 7 , the medical image analysis apparatus 100 may not include a second analysis model. The medical image analysis apparatus 100 may directly input the report information 421 to the third analysis model. That is, the second lesion information 422 may include report information 421 .

Referring to FIG. 8 , the third analysis model 400 includes first past report information included in a plurality of first past lesion information 811, past report information included in a plurality of second past lesion information 812, and It may be a model obtained by machine learning information related to the identity of the first past report information included in the past result information 813 and the past report information.

The third analysis model 400 may receive first report information included in the first lesion information 412 and report information 421 included in the second lesion information 422 as inputs. The third analysis model 400 may output result information 430 indicating the identity of the first report information and the report information 421 .

According to another embodiment of the present disclosure, the medical image analysis apparatus 100 generates result information by applying second lesion area information included in the medical image and second lesion information to a third analysis model that is a machine learning model. can be performed.

Referring to FIG. 6 , the medical image analysis apparatus 100 may not include the first analysis model 410 . The medical image analysis apparatus 100 may directly input the medical image 411 to the third analysis model 400 . That is, the first lesion information 412 may include a medical image 411 .

Referring to FIG. 7 , the medical image analysis apparatus 100 may obtain second lesion area information as second lesion information 422 by applying the report information 421 to the second analysis model 420 . The second lesion area information may be information automatically acquired based on the report information 421 .

Referring to FIG. 8 , the third analysis model 400 includes past medical images included in the plurality of first past lesion information 811 and second past lesion area information included in the plurality of second past lesion information 812 . and information related to the identity of the past medical image included in the past result information 813 and the second past lesion area information may be machine-learned models.

The third analysis model 400 may receive second lesion area information included in the medical image 411 and the second lesion information 422 as an input. The third analysis model 400 may output result information 430 indicating the identity of the first report information and the report information 421 .

10 is a diagram illustrating an operation of a medical image analysis apparatus according to an embodiment of the present disclosure.

Based on the result information 430 , the medical image analysis apparatus 100 may perform step 1010 of determining the medical image reading accuracy of the medical personnel. Referring to FIG. 4 , first lesion information 412 may represent a diagnosis result automatically generated for a medical image by the medical image analysis apparatus 100 . Also, the second lesion information 422 may indicate a diagnosis result of a medical image created by a medical person. The result information 430 may indicate the identity of the first lesion information 412 and the second lesion information 422 . The medical image analysis apparatus 100 may accumulate and store result information on medical personnel. When the reliability of the first lesion information 412 is high, the medical image analysis apparatus 100 may determine medical image reading accuracy of a medical person based on the result information 430 . That is, since the first lesion information 412 can be trusted, the accuracy of reading the medical image by the medical personnel is determined based on the first lesion information 412 .

Specifically, the accuracy of medical image reading by a medical person may be based on the number of times that the first lesion information 412 and the second lesion information 422 coincide. In addition, the medical person's image reading accuracy may be based on a ratio of the number of coincidences between the first lesion information 412 and the second lesion information 422 to the total number of medical image readings by the medical person.

However, the present invention is not limited thereto, and a skilled medical person may evaluate whether the report information 421 of another medical person is correct, and the medical image analysis apparatus 100 may receive the evaluation result of the skilled medical person. The medical image analysis apparatus 100 may determine the evaluation result of a skilled medical person as the accuracy of reading the medical image of the medical person.

The medical image analysis apparatus 100 may perform step 1020 of acquiring lesion information and accuracy of the lesion information from the medical image based on the machine learning model. The machine learning model may include the first analysis model described above. The machine learning model may obtain accuracy of lesion information together with at least one of first lesion presence information, first lesion type information, first lesion area information, or first report information from a medical image. The accuracy of lesion information may vary depending on the medical image.

For example, the machine learning model may obtain a probability of existence of a lesion in a medical image and a probability of non-existence of a lesion, respectively. The machine learning model may determine the first lesion information based on a high probability. For example, when the probability that a lesion exists in a medical image is higher than the probability that a lesion does not exist, the machine learning model may make the first lesion information indicate that a lesion exists in the medical image.

The machine learning model may determine that the accuracy of the lesion information is low when the difference between the probability of existence and the probability of non-existence is lower than a predetermined threshold value. Conversely, the machine learning model may determine that the accuracy of the lesion information is high when the difference between the probability of existence and the probability of non-existence is higher than a predetermined threshold value.

Although the case where the machine learning model acquires the first lesion information has been described above, the same process may be performed for the first lesion type information, the first lesion area information, or the first report information.

The medical image analysis apparatus 100 may perform step 1030 of automatically allocating a medical person to read a medical image based on the medical person's medical image reading accuracy and the accuracy of the lesion information of the machine learning model. The medical image analysis apparatus 100 may perform step 1010 to determine the medical image reading accuracy of the medical personnel. The medical image analysis apparatus 100 may obtain accuracy of lesion information of the machine learning model for the medical image by performing step 1020 . The accuracy of the lesion information of the machine learning model may differ for each medical image.

The medical image analysis apparatus 100 may assign to read a medical image in which the accuracy of the lesion information of the machine learning model is low as the medical person's medical image reading accuracy is high. High accuracy of medical image reading by a medical person may mean that the medical person is skilled. In addition, low accuracy of lesion information of the machine learning model may mean that it is difficult to read medical images by the machine learning model. Also, this may mean that it is a medical image of a rare case. The medical image analysis apparatus 100 may prevent medical accidents from occurring by assigning medical images that are difficult to read to skilled medical personnel.

Conversely, the medical image analysis apparatus 100 may assign to read a medical image in which the accuracy of the lesion information of the machine learning model is high as the medical person's medical image reading accuracy is low. Low accuracy of medical image reading by medical personnel may mean that the medical personnel are not skilled. In addition, high accuracy of lesion information of the machine learning model may mean that the medical image is easy to read by the machine learning model. Also, this may mean that it is a medical image of a frequently appearing case. The apparatus 100 for analyzing medical images may prevent medical accidents from occurring by allocating medical images that are easy to read to unskilled medical personnel, while enabling medical personnel to deal with various cases.

So far, the process of allocating medical images to medical personnel according to the skill level of the medical personnel has been described. However, the same process can be performed for hospitals that are not medical personnel. In detail, the medical image analysis apparatus 100 may perform step 1010 of determining the medical image reading accuracy of the hospital based on the result information 430 . In addition, the medical image analysis apparatus 100 may perform step 1020 of obtaining lesion information and accuracy of the lesion information from the medical image based on the machine learning model. The medical image analysis apparatus 100 may perform step 1030 of automatically allocating a hospital to read a medical image based on the accuracy of reading the medical image of the hospital and the accuracy of the lesion information of the machine learning model. Since the operation performed by the medical image analysis apparatus 100 is the same except that the medical personnel is changed to the hospital, duplicate descriptions are omitted.

11 is a diagram illustrating an operation of a medical image analysis apparatus according to an embodiment of the present disclosure.

The medical image analysis apparatus 100 may perform step 1110 of receiving a user's evaluation result for the first lesion information 412 . The user may determine whether the first lesion information 412 obtained by the first analysis model 410 is accurate or incorrect and input the information to the medical image analysis apparatus 100 . The medical image analysis apparatus 100 may determine whether the first lesion information 412 is accurate or incorrect based on the user's input. This process may be performed by the model evaluation unit 115 of the medical image analysis apparatus 100 .

When the evaluation result is inaccurate, the medical image analysis apparatus 100 may perform step 1120 of acquiring past medical image and past lesion information. The past medical image and past lesion information may include updated past medical images and updated past lesion information in order to update the first analysis model 410 . The medical image analysis apparatus 100 may perform step 1120 using the model updater 125 .

The medical image analysis apparatus 100 may perform step 1130 of re-learning the first analysis model based on past medical images and past lesion information. The medical image analysis apparatus 100 may store the retrained first analysis model in a memory or may transmit the retrained first analysis model to another medical image analysis apparatus. The retrained first analysis model can derive lesions from medical images more accurately and more quickly than the existing first analysis model. Here, re-learning may mean additional learning or continuous learning. The apparatus 100 for analyzing medical images of the present disclosure has an effect of supporting a medical person's interpretation of medical images by continuously improving the first analysis model.

12 is a diagram illustrating an operation of a medical image analysis system according to an embodiment of the present disclosure.

The medical image analysis system may include at least one of the medical image analysis device 100 , hospital server 1200 , patient terminal 1201 , medical personnel terminal, and payer server. The payer server may include at least one of an insurance company server, a medical insurance corporation server, and a Center for Medicare & Medicaid Service (CMS) server.

A medical person may generate report information 421 by diagnosing a patient. The medical image analysis apparatus 100 may determine the accuracy of the medical personnel's report information 421 as follows.

The medical image analysis apparatus 100 may accumulate result information 430 and store it in a memory. The medical image analysis apparatus 100 may perform step 1211 of transmitting the result information 430 to the hospital server 1200 . The result information 430 may include information about the identity of the first lesion information 412 and the second lesion information 422 . When the hospital server 1200 determines that the first lesion information 412 is different from the second lesion information 422 based on the result information 430, the first lesion information 412 or the second lesion information 422 You can determine which information is inaccurate.

The medical image analysis apparatus 100 may obtain accuracy of the first lesion information 412 from the first analysis model. Accuracy is how high the probability of the first lesion information derived by applying the first analysis model 410 to the medical image 411 is compared to other lesion information that can be derived by the first analysis model 410. It can be determined based on whether Since this has already been described with reference to FIG. 10 , duplicate descriptions will be omitted.

When the medical image analysis apparatus 100 has high accuracy of the first lesion information 412 and the first lesion information 412 and the second lesion information 422 are different, the hospital server 1200 provides the second lesion information ( 422) can be determined to be inaccurate. Also, the medical image analysis apparatus 100 may determine that the second lesion information 422 is inaccurate based on the user's input. That the second lesion information 422 is inaccurate means that there is information that is described in the first lesion information 412 but is missing in the second lesion information 422, or that it is not described in the first lesion information 412 but the second lesion information 412 is incorrect. This may mean a case in which there is information described in the lesion information 422 .

The hospital server 1200 may generate visit request information when there is inaccurate information in the second lesion information 422 . The visit request information may include at least one of result information 430 and first lesion information 412 . The hospital server 1200 may perform step 1221 of transmitting visit request information to the patient terminal 1201 to request a reexamination to the patient. The patient can check the information displayed on the patient terminal 1201 and visit the hospital. The visit request information may include at least one of result information 430 and first lesion information 412 . However, it is not limited thereto, and the user may request a patient to visit the hospital by mail or e-mail according to the visit request information displayed by the hospital server 1200 .

Also, when there is inaccurate information in the second lesion information 422, the medical image analysis apparatus 100, not the hospital server 1200, may generate hospital visit request information. Also, the medical image analysis apparatus 100 may transmit visit request information to the patient terminal 1201 or the hospital server 1200 .

Also, the hospital server 1200 or the medical image analysis apparatus 100 may transmit visit request information to a doctor terminal or an insurance company server in addition to the patient terminal 1201 . The hospital server 1200 may determine at least one of schedule information for performing a reexamination on a patient, a processing flow within the hospital, and a patient worklist based on the visit request information.

The hospital server 1200 may accumulate result information 430 and store it in memory. The hospital server 1200 may perform step 1212 of generating evaluation information of medical personnel belonging to the hospital based on the accumulated result information. For example, the hospital server 1200 may obtain medical personnel report information 421 or second lesion information 422 . The hospital server 1200 may generate medical personnel evaluation information based on the number of times that the medical personnel report information 421 or the second lesion information 422 is the same as the first lesion information. Alternatively, a skilled medical person may input evaluation information into the hospital server 1200 based on the medical person's report information 421 or the second lesion information 422 . The hospital server 1200 may determine evaluation information of a medical person based on an input of a skilled medical person.

The hospital server 1200 may perform step 1213 of providing incentives to medical personnel based on the evaluation information. For example, the hospital server 1200 may provide an incentive to the medical personnel as the evaluation information increases in association with the medical staff pay server. In addition, when the evaluation information is low, the hospital server 1200 may impose a penalty on the medical personnel or suggest retraining of the medical personnel to the hospital manager.

The hospital server 1200 may perform step 1214 of determining the medical personnel's patient worklist based on the evaluation information. For example, the hospital server 1200 may determine a patient worklist so that medical personnel with high evaluation information take care of urgent and difficult patients. In addition, the hospital server 1200 may determine a patient worklist so that medical personnel with low evaluation information take on patients with low difficulty level.

13 is a diagram illustrating an operation of a medical image analysis system according to an embodiment of the present disclosure.

The medical image analysis apparatus 100 may accumulate result information 430 and store it in a memory. The medical image analysis apparatus 100 may perform step 1311 of generating hospital evaluation information based on the accumulated result information. The hospital evaluation information may represent the rate at which medical personnel belonging to the hospital correctly read the medical image. The hospital evaluation information may be obtained based on a ratio of the number of times indicating that the first lesion information 412 and the second lesion information 422 are identical to the total number of readings performed by medical personnel belonging to the hospital. The number of times indicating that the first lesion information 412 and the second lesion information 422 are the same may be obtained based on the third analysis model 400 . As already described, the first lesion information 412 is information on a lesion for which a medical image has been read based on the first analysis model 410, and the second lesion information 422 is the second analysis model in the medical personnel's report information. This is information about the lesion obtained by applying (420). When the first lesion information 412 and the second lesion information 422 match, the medical image analysis apparatus 100 may determine that the medical person has correctly read the medical image.

Also, the medical image analysis apparatus 100 may determine hospital evaluation information by receiving hospital evaluation information from a skilled medical person. When the first lesion information and the second lesion information match, the medical image analysis apparatus 100 may determine that the medical person has correctly read the medical image.

When the first lesion information and the second lesion information are different from each other, a skilled medical person may select incorrect information from among the first lesion information and the second lesion information. If the first lesion information is incorrect, the medical image analysis apparatus 100 may update the first analysis model 410 . The medical image analysis apparatus 100 may update the first analysis model 410 as described in FIG. 6 . The medical image analysis apparatus 100 may update the first analysis model 410 based on the plurality of past medical images 611 and the plurality of first past lesion information 612 . The medical image analysis apparatus 100 may use the plurality of past medical images 611 and the plurality of first past lesion information 612 as actual information. In addition, the first past lesion information 612 may further include second lesion information determined to be accurate information by a skilled medical practitioner.

If the second lesion information is inaccurate, the medical image analysis apparatus 100 may determine hospital evaluation information based on the number of times indicating that the first lesion information and the second lesion information are identical to the total number of readings by medical personnel. The number of times indicating that the first lesion information 412 and the second lesion information 422 are the same may be obtained based on the third analysis model 400 .

The medical image analysis apparatus 100 may provide hospital evaluation information according to hospitals. Also, the medical image analysis apparatus 100 may provide information on an overall average of hospital evaluation information. In addition, the medical image analysis apparatus 100 may provide information about the average of hospital evaluation information for each region. Patients may check statistical data of hospital evaluation information from the medical image analysis device 100 . Patients can choose a hospital based on objective information, so they can receive high-quality services. In addition, hospitals will make efforts to increase hospital evaluation information, so the overall service quality will be improved.

The medical image analysis apparatus 100 may provide various statistical data of hospitals. The medical image analysis apparatus 100 may provide information on the number of readings for each modality of the hospital. For example, the medical image analysis apparatus 100 may provide information on the number of times medical personnel belonging to a hospital read an ultrasound image, a CT image, or an MRI image. Patients can determine the proficiency of medical personnel belonging to the hospital based on the number of readings for each modality.

Also, the medical image analysis apparatus 100 may provide information related to the number of normal or abnormal readings of medical images by medical personnel belonging to the hospital. Normal reading means that there is no lesion in the medical image, and abnormal reading means that there is a lesion in the medical image.

In addition, the medical image analysis apparatus 100 may provide information about the accuracy of reading medical images by medical personnel belonging to the hospital. For example, the medical image analysis apparatus 100 determines the number of times that the lesion information read by the medical person and the lesion information read by the first analysis model 410 are different with respect to the total number of times the medical person belonging to the hospital reads the medical image. Information about accuracy can be determined based on the ratio of . The medical image analysis apparatus 100 may use the third analysis model 400 to determine whether the lesion information read by the medical personnel and the lesion information read by the first analysis model 410 are different.

The medical image analysis apparatus 100 may transmit hospital evaluation information to an external server. The external server may provide result information or hospital evaluation information in the form of a website. The user can check hospital evaluation information using a web browser. Also, the medical image analysis apparatus 100 may transmit hospital evaluation information to the user terminal. The user can easily check hospital evaluation information about the hospital he plans to use or the hospital he has used in the past using his terminal.

In addition, the medical image analysis apparatus 100 may perform step 1312 of transmitting hospital evaluation information to a payer server 1300 . The payer server 1300 may refer to a server corresponding to a subject paying medical expenses, and for example, at least one of an insurance company server, a medical insurance corporation server, or a Center for Medicare & Medicaid Service (CMS) server. may contain one.

The payer server 1300 may perform step 1313 of changing medical expense payment standard information based on hospital evaluation information. For example, the payer server 1300 may change the medical expense payment standard information so that the higher the hospital evaluation information, the higher the medical fee. In addition, the payer server 1300 may change medical expense payment standard information so that the lower the hospital evaluation information, the lower the medical expense. However, the payer server 1300 is not limited thereto, and the payer server 1300 may change the medical cost payment standard information so that the lower the hospital evaluation information is, the lower the medical cost is paid, and the lower the hospital evaluation information is, the higher the medical cost payment standard information is. can be changed The payer server 1300 may generate information guiding more patients to go to the hospital instead of paying lower medical expenses as the hospital evaluation information is higher.

The payer server 1300 may transmit changed medical expense payment standard information to the hospital server 1200, an external server, or a user terminal. In addition, the hospital manager may strive to improve the accuracy of medical image reading of the hospital based on the medical expense payment standard information displayed on the hospital server 1200 .

In addition, the external server may provide users with the received medical expense payment standard information in the form of a website. Users can check medical expense payment standard information for hospitals they plan to use or hospitals they have used in the past through a web browser.

In addition, the user terminal may display the received medical expense payment standard information. The user can easily check medical expense payment standard information using the terminal.

So far, the medical image analysis apparatus 100 according to the present disclosure has been described. The medical image analysis apparatus 100 according to the present disclosure may compare medical image reading using a machine learning model and medical image reading by a medical person. Therefore, the accuracy of medical image reading can be improved. In reality, it is difficult for a plurality of medical personnel to discriminate a single medical image in terms of time and cost. , it can reduce the omission of lesion information that can occur in medical image reading. In addition, according to the medical image analysis apparatus 100 according to the present disclosure, medical services can be improved and medical costs can be reduced through early detection of lesions. In addition, the medical image analysis apparatus 100 according to the present disclosure can improve the performance of the machine learning model.

So far, we have looked at various embodiments. Those skilled in the art to which the present invention pertains will be able to understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments should be considered from an illustrative rather than a limiting point of view. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the equivalent scope will be construed as being included in the present invention.

On the other hand, the above-described embodiments of the present invention can be written as a program that can be executed on a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable recording medium. The computer-readable recording medium includes storage media such as magnetic storage media (eg, ROM, floppy disk, hard disk, etc.) and optical reading media (eg, CD-ROM, DVD, etc.).

Claims (1)

As a medical image analysis device,
Receives a medical image, receives report information that is a result of a medical practitioner's judgment on the medical image, and compares lesion information obtained from the medical image with lesion information obtained from the report information using an analysis model. Based on the results, a processor that determines the medical personnel's worklist
A medical image analysis device comprising a.

Images