KR20200015280A - Method for predicting pneumonia - Google Patents

Abstract

According to an embodiment of the present invention, provided is a method for determining a probability of developing pneumonia by using a device for determining a probability of developing pneumonia. The method comprises the steps of: obtaining information on a plurality of items of a patient to be diagnosed and information on an environment; determining a node value of at least one input node included in an input layer by using the information on the plurality of items of the patient to be diagnosed and the information on the environment; determining a node value of at least one output node included in an output layer from a learned pneumonia occurrence probability calculator based on the node value of the at least one input node; and determining a probability of developing pneumonia of the patient to be diagnosed based on a node value of the at least one output node.

Description

How to determine the probability of developing pneumonia {METHOD FOR PREDICTING PNEUMONIA}

Embodiments of the present invention relate to a method for determining the incidence of pneumonia, and to a method for determining the incidence of pneumonia using a learned artificial neural network based on a plurality of learning data.

Pneumonia is an inflammation of the lungs. The most common cause is infection by bacteria, and viruses, fungi, or other microorganisms may be the cause. Rarely, it can be caused by inhaling allergic reactions or irritating chemicals. Older children, younger patients, or those with poor overall cough reflexes are more likely to develop inhaled pneumonia.

Pneumonia is the most common cause of death in the world today. In particular, pneumonia is the last disease to cause death in patients with other serious chronic diseases. In the United States, two to three million people develop pneumonia each year, of which about 60,000 die. In the United States, together with influenza, pneumonia is the eighth leading cause of death and the first infectious cause of death. Pneumonia is the most common cause of death during hospitalization and the most common cause of death in developing countries.

As such, pneumonia is a fatal disease despite its cure, and therefore, there is a need to proactively cope with a more accurate prediction of the probability of developing pneumonia.

The present invention seeks to reasonably infer the incidence of pneumonia from patient information and environmental information.

In addition, the present invention is to infer from the patient information with a high reliability in addition to the above-described probability of the occurrence of pneumonia, the probability of death, the occurrence of related diseases.

According to an embodiment of the present invention, a method of determining the incidence of pneumonia by using an apparatus for determining the incidence of pneumonia may include obtaining a plurality of items of information about a diagnosis target patient and information about an environment; Determining a node value of at least one input node included in an input layer by using a plurality of items of information on the diagnosis target patient and information on an environment; Determining a node value of at least one output node included in an output layer from a learned pneumonia incidence probability calculator based on the node value of the at least one input node; And determining a probability of developing pneumonia of the patient to be diagnosed based on the node value of the at least one output node.

In this case, the pneumonia incidence calculator is based on the plurality of items of information on each of the plurality of patients, the information on the environment and the plurality of patients based on the learning data including the probability of the occurrence of pneumonia for each of the plurality of items A neural network learned about the correlation between information and information about a plurality of environments and the probability of occurrence of pneumonia, wherein the neural network is hidden between the input layer, the output layer, the input layer and the output layer. The layer may include a first function between the input layer and the hidden layer and a second function between the hidden layer and the output layer.

The plurality of items of information about the patient includes at least one of personal information of the patient, vital sign information of the patient, and medical information of the patient, wherein the medical information of the patient includes a past medical history of the patient, It may include at least one of the medical image of the patient, the history of taking the drug of the patient and the history of symptoms of the patient.

The input layer may include a first input node to which the patient's personal information is input, a second input node to which the vital sign information of the patient is input, and a third input node to which the medical information of the patient is input.

The information on the environment includes at least one of information on concentration, humidity, and UV value of atmospheric fine dust for a predetermined period, and the input layer further includes a fourth input node to which the concentration of fine dust is input. can do.

The method for determining the onset probability of pneumonia further includes the step of learning the neural network based on the learning data, wherein the learning of the neural network comprises a plurality of items of information about any patient received from an external device, pneumonia. Classifying information of a patient including an occurrence diagnosis code into first valid learning data; Classifying information of a patient including a medical image related to pneumonia into second valid learning data, into a plurality of items of information about any patient received from the external device; Classifying patient information including a keyword suggestive of pneumonia into third valid learning data to a plurality of items of information about any patient received from the external device; Classifying the patient's information including the medical history of a specific disease within a predetermined period of time before the onset of pneumonia into fourth valid learning data in a plurality of items of information about any patient received from the external device; And determining at least some of the first valid learning data, the second valid learning data, the third valid learning data, and the fourth valid learning data as the learning data.

The determining of the learning data may include learning data belonging to all of the first valid learning data, the second valid learning data, the third valid learning data, and the fourth valid learning data as the learning data.

The neural network may be learned by further including a probability of death over time of the patient and a probability of developing the first disease of the patient. In this case, the output layer may further include a second output node corresponding to the probability of death over the time of the patient and a third output node corresponding to the occurrence probability of the first disease of the patient.

The first disease may be a plurality, and the output layer may include a third output node for each of the plurality of first diseases.

Other aspects, features, and advantages other than those described above will become apparent from the following drawings, claims, and detailed description of the invention.

According to embodiments of the present invention, it is possible to reasonably infer the probability of developing pneumonia from the patient's information.

In addition to the above-described probability of developing pneumonia, the probability of death and the probability of the associated disease can be inferred with high reliability.

1 schematically illustrates a system for determining the probability of developing pneumonia according to an embodiment of the present invention.
2 schematically illustrates the configuration of an apparatus for determining the probability of developing pneumonia according to an embodiment of the present invention.
3 is a view for explaining the structure of the neural network learned by the control unit according to an embodiment of the present invention.
4 is a flowchart illustrating a method of determining the incidence of pneumonia performed by an apparatus for determining the incidence of pneumonia according to an embodiment of the present invention.
5 is an example of a screen displayed on a user terminal according to an embodiment of the present invention.

As the invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. Effects and features of the present invention, and methods for achieving them will be apparent with reference to the embodiments described below in detail in conjunction with the drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various forms.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings, and the same or corresponding components will be denoted by the same reference numerals, and redundant description thereof will be omitted. .

In the following embodiments, the terms first, second, etc. are used for the purpose of distinguishing one component from other components rather than a restrictive meaning. In the following examples, the singular forms “a”, “an” and “the” include plural forms unless the context clearly indicates otherwise. In the following examples, the terms including or having have meant that there is a feature or component described in the specification and does not preclude the possibility of adding one or more other features or components. In the drawings, components may be exaggerated or reduced in size for convenience of description. For example, the size and shape of each component shown in the drawings are arbitrarily shown for convenience of description, and thus the present invention is not necessarily limited to the illustrated.

1 schematically illustrates a system for determining the probability of developing pneumonia according to an embodiment of the present invention.

Referring to FIG. 1, a system for determining the probability of developing pneumonia according to an embodiment of the present invention may include a server 100, a user terminal 200, an external device 300, and a communication network 400 connecting them. .

The server 100 of the system for determining the incidence of pneumonia according to an embodiment of the present invention receives a plurality of items of information about the patient and information about the environment from the user terminal 200 and / or the external device 300, The probability of developing pneumonia in the patient may be determined based on the received information.

In addition, the server 100 may include information on a plurality of items for a plurality of patients, information on a plurality of environments, and a probability of developing pneumonia for each of the plurality of patients from the user terminal 200 and / or the external device 300. Neural networks can be learned based on facts of occurrence). Detailed description thereof will be described later.

In the present invention, 'a plurality of items of information about a patient' (or 'patient information') may include various items of information about a corresponding patient as its name. For example, the plurality of items of information about the patient may include at least one of personal information of the patient, vital sign information of the patient, and medical information of the patient.

The personal information of the patient may include general information about the patient such as, for example, the patient's name, patient's sex, patient's age, patient's date of birth, patient's name, patient's residence address, patient's caregiver information.

The vital sign information of the patient may include, for example, blood pressure information, pulse information, oxygen saturation information, body temperature information, etc. over time of the patient.

The patient's medical information may include, for example, the patient's medical history (e.g. sickness code), the patient's medical image (MRI image, CT image, X-ray image, etc.), the patient's history of drug use, the patient's symptom history, and Diagnostic history, and the like.

However, the above items are exemplary and the spirit of the present invention is not limited thereto. Thus, in addition to the listed items, general information about the patient may be included in the personal information of the patient, information on the vital signs of the patient in the vital sign information of the patient, and information about the patient's medical care, respectively.

In the present invention, 'environmental information' (or 'environmental information') may include information on various items of the natural environment. For example, the information on the environment may include information on the concentration, humidity, and UV value of atmospheric particulate matter for a predetermined time period.

In an alternative embodiment, the 'information about the environment' may include only various items of information about the environment around the patient's residence. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

In the present invention, 'the probability of developing pneumonia' refers to the probability of developing pneumonia in a particular patient, and is a fixed value (for example, 70%, etc.) regardless of the passage of time, or a value that changes with the passage of time. (For example, 18% after 1 week, 25% after 2 weeks, 11% after 3 weeks, etc.).

The 'neural network' included in the 'pneumonia incidence calculator' described later in the present invention may mean an artificial neural network learned by machine learning or deep learning. In this case, the artificial neural network may refer to a model in which artificial neurons (hereinafter, referred to as 'nodes') that form a network by synapse coupling learn a problem solving ability by changing the strength of synapses through learning. Can be. The structure of the neural network will be described later with reference to FIG. 3.

The user terminal 200 according to an embodiment of the present invention may refer to various devices capable of transmitting and receiving patient information and environment information with the server 100. In this case, the terminal may be a portable terminal 201 or may be a personal computer 202.

According to an embodiment of the present invention, the user terminal 200 includes display means for displaying a screen displaying a probability of developing pneumonia received from the server 100, input means for obtaining a user input for such a screen It can be provided. In this case, the input means and the display means may be configured in various ways. For example, the input means may include a keyboard, a mouse, a trackball, a microphone, a button, a touch panel, and the like, but is not limited thereto.

In the present invention, the external device 300 may refer to various devices for transmitting and receiving data through the server 100 and / or the user terminal 200 and the communication network 400.

According to an embodiment of the present invention, the external device 300 may be a device that provides training data for learning the pneumonia onset probability calculator provided in the server 100. For example, the external device 300 may be a server of a medical institution that provides a plurality of items of information about a plurality of patients.

In addition, the external device 300 may be a weather information providing server that provides information on the environment of various regions in various time zones. As described above, the external device 300 may be singular or plural.

In the present invention, the communication network 400 serves to connect the server 100, the user terminal 200 and the external device 300. For example, the communication network 400 provides a connection path for transmitting and receiving packet data after the user terminal 200 accesses the server 100. The communication network 400 may be a wired network such as local area networks (LANs), wide area networks (WANs), metropolitan area networks (MANs), integrated service digital networks (ISDNs), wireless LANs, CDMA, Bluetooth, satellite communications, and the like. Although it may encompass a wireless network of, the scope of the present invention is not limited thereto.

The server 100 according to an exemplary embodiment of the present invention receives a plurality of items of information about the patient and information about the environment from the user terminal 200 and / or the external device 300 and based on the received information. The probability of developing pneumonia in a patient can be determined.

In addition, the server 100 may include information on a plurality of items for a plurality of patients, information on a plurality of environments, and a probability of developing pneumonia for each of the plurality of patients from the user terminal 200 and / or the external device 300. Neural networks can be learned based on facts of occurrence). To this end, the server 100 may include a device for determining the probability of developing pneumonia, as shown in FIG. 2.

2 schematically illustrates a configuration of an apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention.

Referring to FIG. 2, the apparatus 110 for determining the incidence of pneumonia according to an embodiment of the present invention may include a communication unit 111, a control unit 112, and a memory 113. In addition, although not shown, the apparatus 110 for determining the probability of developing pneumonia according to the present embodiment may further include an input / output unit, a program storage unit, and the like.

The communicator 111 may transmit / receive a signal such as a control signal or a data signal through the wired / wireless connection of the device 110 for determining the probability of occurrence of pneumonia with another network device such as the user terminal 200 and / or the external device 300. It may be a device including necessary hardware and software.

The controller 112 may include all kinds of devices capable of processing data, such as a processor. Here, the 'processor' may refer to a data processing apparatus embedded in hardware having, for example, a physically structured circuit for performing a function represented by code or instructions included in a program. As an example of the data processing device embedded in the hardware, a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated (ASIC) Although it may include a processing device such as a circuit, a field programmable gate array (FPGA), etc., the scope of the present invention is not limited thereto.

The memory 113 performs a function of temporarily or permanently storing data processed by the apparatus 110 for determining the probability of occurrence of pneumonia. The memory 113 may include a magnetic storage medium or a flash storage medium, but the scope of the present invention is not limited thereto.

Hereinafter, a description will be made on the premise that the apparatus 100 for determining the incidence of pneumonia is provided in the server 100. However, the apparatus 100 for determining the probability of developing pneumonia may be provided separately from the server 100 according to role allocation.

Apparatus 110 for determining the incidence of pneumonia in the server 100 according to an embodiment of the present invention for a plurality of items of information and environment for the patient from the user terminal 200 and / or external device 300 Information (hereinafter, referred to as 'patient and environmental information') may be received, and a probability of developing pneumonia in the patient may be determined based on the received information.

In addition, the apparatus 110 for determining the probability of developing pneumonia may include information on a plurality of items for a plurality of patients, information on a plurality of environments, and pneumonia for each of a plurality of patients from the user terminal 200 and / or the external device 300. The neural network can be trained based on the probability of occurrence (or fact of occurrence) (hereinafter referred to as 'learning data').

As described above, in the present invention, the pneumonia onset probability calculator may include a neural network learned by the control unit 112 of the onset probability determining unit 110. In this case, the neural network may refer to an artificial neural network that is learned by machine learning or deep learning techniques. Here, 'artificial neural network' is a model in which artificial neurons (hereinafter, referred to as 'nodes') that form a network by synapse coupling learn a problem solving ability by changing the strength of synapse through learning. Can mean.

3 is a view for explaining the structure of the neural network learned by the control unit 112 according to an embodiment of the present invention.

Referring to FIG. 3, a neural network according to an embodiment of the present invention includes an input layer L1 including at least one input node N1, a hidden layer L2 including a plurality of hidden nodes N2, and at least An output layer L3 including one output node N3 may be included. At this time

Values corresponding to the patient and environment information acquired by the controller 112 may be input to at least one input node N1 of the input layer L1. For example, a value corresponding to the personal information of the patient may be input to the first input node, and a value corresponding to the vital sign of the patient may be input to the second input node.

The hidden layer L2 may include one or more layers that are fully connected as illustrated. When the hidden layer L2 includes a plurality of layers, the neural network may include a function (not shown) that defines a relationship between each hidden layer.

At least one output node N3 of the output layer L3 may include an output value generated by the neural network from an input value of the input layer L1 under the control of the controller 112. For example, the first output node may include a value corresponding to the probability of developing pneumonia over time of the patient.

In this case, the value included in each node of each layer may be a vector. In addition, each node may include a weight corresponding to the importance of the node.

The neural network includes a first function F1 that defines a relationship between the input layer L1 and a hidden layer L2, and a second function F2 that defines a relationship between the hidden layer L2 and the output layer L3. can do.

The first function F1 may define a connection relationship between the input node N1 included in the input layer L1 and the hidden node N2 included in the hidden layer L2. Similarly, the second function F2 may define a connection relationship between the hidden node N2 included in the hidden layer L2 and the output node N2 included in the output layer L2.

The functions between the first function F1, the second function F2, and the hidden layer may include a Recurrent Neural Network (RNN) model that outputs a result based on an input of a previous node.

In the process of neural network learning by the controller 112, the first function F1 and the second function F2 may be learned based on the plurality of learning data. Of course, in the process of neural network learning, functions between a plurality of hidden layers may also be learned in addition to the first function F1 and the second function F2.

In the present invention, "learning" means that the control unit 112 updates the aforementioned functions (F1, F2, functions between hidden layers, etc.) based on the plurality of learning data and stores them in the memory 113 or the like. can do. For example, the controller 112 may appropriately adjust the coefficients of the elements constituting the first function based on the plurality of learning data and store them in the memory 113. Similarly, of course, the controller 112 may appropriately adjust the coefficients of the elements constituting the second function and store them in the memory 113. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

The neural network according to an embodiment of the present invention may be learned in a supervised learning manner based on labeled learning data. In this case, the learning data may be data in which the probability of occurrence of pneumonia (or fact of occurrence) of the patient is labeled in a plurality of items of information about the patient and information about the environment. In this case, the probability of developing pneumonia is a fixed value (for example, 70%, etc.) regardless of the passage of time as described above, or a value that changes with the passage of time (for example, 18% after two weeks, two weeks). 25% after, 11% after 3 weeks, etc.). The fact of onset may be a value indicating whether pneumonia is present.

 The control unit 112 according to an embodiment of the present invention uses the plurality of learning data to input the learning data into a neural network so that the output value generated is close to the probability of occurrence of pneumonia labeled on the learning data. The neural network can be trained by repeatedly performing the process of updating the functions (F1, F2, functions between hidden layers, etc.).

At this time, the control unit 112 according to an embodiment of the present invention may update the above-described functions (F1, F2, functions between hidden layers, etc.) according to a back propagation algorithm. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

Of course, the neural network according to an embodiment of the present invention may be learned in an unsupervised learning manner based on unlabeled learning data. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

The controller 112 according to an embodiment of the present invention may obtain patient and environment information, and determine the node value of at least one input node N1 included in the input layer L1 described above.

For convenience of description, the neural network may include a first input node corresponding to the personal information of the patient, a second input node corresponding to the vital sign information of the patient, and a third input node corresponding to the medical information of the patient in the input layer L1. Suppose you include it.

In this case, the control unit 112 according to an embodiment of the present invention may determine the node value of each of the first to third input nodes from the obtained information. At this time, the control unit 112 according to an embodiment of the present invention may extract a feature value for each item by applying a convolutional neural network (CNN) to the information of each item, and the extracted feature value may be in the form of a vector. .

In other words, the control unit 112 according to an embodiment of the present invention may apply a convolutional neural network to the personal information of the patient to determine the first vector corresponding to the personal information of the patient as the node value of the first input node. . Similarly, the controller 112 may determine the node value of the second input node from the vital sign information of the patient and the node value of the third input node from the medical information of the patient.

The input layer L1 may include a fourth input node corresponding to the information about the environment, in addition to the above-described information about the patient. In this case, the fourth input node may be a node having a node value corresponding to the concentration of fine dust. The node value of the fourth input node may be generated by the controller 112 by applying a convolutional neural network to information about the environment.

Although the controller 112 exemplarily applies the convolutional neural network to information of each item as a method of determining the node values of the first to fourth input nodes of the above-described input layer, the convolutional neural network is not necessarily described. You do not have to use. Therefore, various algorithms may be used to derive the feature values of each item according to the data characteristics and / or the design of the system.

Meanwhile, the node values of the first to fourth input nodes may be vectors of n, m, i, and j dimensions, respectively.

The control unit 112 according to an embodiment of the present invention includes at least one included in the output layer from the pneumonia incidence probability calculator based on the input layer determined by the above-described process (ie, node values of the input nodes included in the input layer). The node value of the output node can be determined.

In other words, the controller 112 according to an embodiment of the present invention may determine the node value of the output node output by the neural network when the node values of the input nodes are input.

Looking at this process in detail, the controller 112 according to an embodiment of the present invention may determine the node values of the nodes of the first hidden layer by inputting the values of the input nodes to the first function F1 and outputting the same. Similarly, the controller 112 may determine the node values of the nodes of the second hidden layer by inputting the node values of the nodes of the first hidden layer into a function defining a relationship between the hidden layers. In this case, the above-described 'function defining a relationship between hidden layers' may be a function defining a relationship between a first hidden layer and a second hidden layer. In the same manner, the controller 112 may determine the node values of the output nodes by inputting and outputting node values of the nodes of the last hidden layer to the second function. As such, the controller 112 may determine the output of the neural network corresponding to the node values of the input nodes, that is, the node values of the output nodes.

The controller 112 according to an embodiment of the present invention may determine the probability of developing pneumonia in the patient based on the node value of the output node described above. In this case, the output layer L3 according to an embodiment of the present invention may include a first output node corresponding to the probability of developing pneumonia of the patient. The node value of the first output node may be in the form of a vector such as [0, 1, 0, 0, 0, 0, 0, 0, 0, 0].

Meanwhile, the neural network according to an exemplary embodiment of the present invention may be further learned by including a probability of death over time of the patient and a probability of occurrence of the first disease of the patient.

For example, the control unit 112 according to an embodiment of the present invention may learn the probability of death and the onset of the first disease of the patient over the time of the patient in a supervised learning manner. In this case, in addition to the probability of developing pneumonia, the learning data may be further labeled with a probability of death and a patient's first disease. Of course, the control unit 112 may learn the probability of death according to the time of the patient and the onset of the first disease of the patient in an unsupervised learning manner. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

The output layer L3 according to an embodiment of the present invention may include a second output node corresponding to the probability of death over time of the patient and a third corresponding to the probability of occurrence of the first disease of the patient together with the aforementioned first output node. It may further include an output node.

In this case, the first disease is a disease associated with pneumonia, and may be plural. When the first disease is plural, the output layer L3 may include a third output node for each of the plurality of first diseases.

As described above, the present invention can reasonably infer the probability of occurrence of a specific disease, especially pneumonia, from the information of the patient, and can also infer the probability of death of the patient.

Hereinafter, a process of generating learning data from the unprocessed patient information and / or environment information by the control unit 112 according to an embodiment of the present invention will be described.

The control unit 112 according to an embodiment of the present invention may receive a plurality of patient information and / or environment information from the user terminal 200 and / or the external device 300. For example, the controller 112 may receive a plurality of patient information from a server (not shown) of the medical institution. In this case, the plurality of patient information received may or may not be related to pneumonia. In other words, the plurality of patient information may include both information about patients associated with pneumonia and information about patients associated with other diseases.

In addition, the controller 112 may receive environment information of a plurality of time zones of a plurality of regions from a weather information providing server (not shown). In this case, the environmental information may also include environmental information of a region and / or a time point related to the patient, or may include environmental information of a region and / or a time point unrelated to the patient.

Accordingly, in order to use the received information for learning in the neural network, it may be necessary to properly process.

The control unit 112 according to an embodiment of the present invention may generate training data by extracting patient information related to pneumonia from a plurality of patient information. In this case, the controller 112 may extract patient information related to pneumonia based on the presence or absence of a pneumonia occurrence diagnosis code (eg, ICD10 standard, J100 / J110 / J12 ~ 18 / J851, etc.) in the patient's medical information. In addition, the control unit 112 may extract patient information related to pneumonia based on whether or not a medical image related to pneumonia is included in the medical information of the patient. In addition, the control unit 112 is associated with pneumonia in the patient's medical information, or based on the presence or absence of pneumonia (Pneumonia, Pneumonic, Consolidation, Infiltration, Ground glass opacity (opacities) = GGO, Parapneumonic, etc.) Relevant patient information can be extracted. In addition, the controller 112 may extract patient information related to pneumonia based on the presence or absence of medical history of a specific disease within a predetermined period before the onset of pneumonia.

In an optional embodiment, the control unit 112 may extract the patient's information corresponding to any one of the four criteria described above as patient information related to pneumonia, and only the patient's information corresponding to all four criteria described above It can also be extracted as patient information associated with

The control unit 112 according to an embodiment of the present invention may generate learning data based on the plurality of patient information extracted according to the above criteria. At this time, the control unit 112 according to an embodiment of the present invention may generate learning data including appropriate environment information with reference to the personal information of the patient included in each patient information.

For example, the controller 112 may generate learning data by checking a living area of a patient and a time of occurrence of pneumonia of the patient, and mapping environmental information of a region and a period corresponding thereto with the patient's degree of the corresponding patient.

As such, the present invention can generate appropriate learning data for learning of neural networks, in particular for learning of neural networks that predict the onset probability of pneumonia.

4 is a flowchart illustrating a method of determining the incidence of pneumonia performed by the apparatus 110 for determining the incidence of pneumonia according to an embodiment of the present invention. Hereinafter, descriptions of contents overlapping with those described in FIGS. 1 to 3 will be omitted, but will be described with reference to FIGS. 1 to 3.

The apparatus 110 for determining the occurrence probability of pneumonia according to an embodiment of the present invention may train a pneumonia occurrence probability calculator based on a plurality of learning data.

In the present invention, the 'pneumonia incidence calculator' may include an artificial neural network learned by machine learning or deep learning. In this case, the artificial neural network may refer to a model in which artificial neurons that form a network by combining synapses learn a problem solving ability by changing the strength of synapses through learning.

Referring back to FIG. 3, the neural network according to an embodiment of the present invention includes an input layer L1 including at least one input node N1, a hidden layer L2 including a plurality of hidden nodes N2, and An output layer L3 including at least one output node N3 may be included. At this time

Values corresponding to patient and environmental information acquired by the apparatus 110 for determining the probability of occurrence of pneumonia may be input to at least one input node N1 of the input layer L1. For example, a value corresponding to the personal information of the patient may be input to the first input node, and a value input to the vital sign of the patient may be input to the second input node.

The hidden layer L2 may include one or more layers that are fully connected as shown. When the hidden layer L2 includes a plurality of layers, the neural network may include a function (not shown) that defines a relationship between each hidden layer.

At least one output node N3 of the output layer L3 may include an output value generated by the neural network from an input value of the input layer L1 under the control of the apparatus 110 for determining the occurrence probability of pneumonia. For example, the first output node may include a value corresponding to the probability of developing pneumonia over time of the patient.

In this case, the value included in each node of each layer may be a vector. In addition, each node may include a weight corresponding to the importance of the node.

Meanwhile, as shown in FIG. 3, the neural network defines a relationship between the first function F1 and the hidden layer L2 and the output layer L3, which define a relationship between the input layer L1 and the hidden layer L2. It may include a second function (F2). The first function F1 may define a connection relationship between the input node N1 included in the input layer L1 and the hidden node N2 included in the hidden layer L2. Similarly, the second function F2 may define a connection relationship between the hidden node N2 included in the hidden layer L2 and the output node N2 included in the output layer L2.

The functions between the first function F1, the second function F2, and the hidden layer may include a Recurrent Neural Network (RNN) model that outputs a result based on an input of a previous node.

 The first function F1 and the second function F2 may be learned based on the plurality of learning data in the process of neural network learning by the apparatus 110 for determining the incidence of pneumonia. Of course, in addition to the above-described first function F1 and second function F2, functions between a plurality of hidden layers may also be learned in the process of neural network learning, and further, each layer L1, L2, L3 may be configured. Weights of the plurality of nodes N1, N2, and N3 may also be learned.

In the present invention, 'learning' means that the apparatus 110 for determining the incidence of pneumonia is based on a plurality of learning data, and the aforementioned functions (F1, F2, functions between hidden layers, etc.) and nodes (N1, N2). , N3) may be updated and stored in the memory 113 or the like. For example, the apparatus 110 for determining the probability of occurrence of pneumonia may appropriately adjust coefficients of elements constituting the first function based on the plurality of learning data and store them in the memory 113. Similarly, similarly, the apparatus 110 for determining the probability of developing pneumonia may appropriately adjust coefficients of elements constituting the second function and / or weights of the nodes, and store them in the memory 113. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

The neural network according to an embodiment of the present invention may be learned in a supervised learning manner based on the labeled learning data. In this case, the learning data may be data in which the probability of occurrence of pneumonia (or fact of occurrence) of the patient is labeled in a plurality of items of information about the patient and information about the environment. In this case, the probability of developing pneumonia is a fixed value (for example, 70%, etc.) regardless of the passage of time as described above, or a value that changes with the passage of time (for example, 18% after two weeks, two weeks). 25% after, 11% after 3 weeks, etc.). The fact of onset may be a value indicating whether pneumonia is present.

 The apparatus 110 for determining the incidence of pneumonia incidence according to an embodiment of the present invention uses a plurality of pieces of learning data, and an output value generated by inputting one piece of learning data into a neural network is labeled with the learning data. The neural network can be trained by repeatedly performing the above-described functions (F1, F2, functions between hidden layers, etc.) and / or updating the weights of the nodes N1, N2, N3 so as to be close to. In this case, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention may include the above-described functions (F1, F2, functions between hidden layers, etc.) and / or nodes according to a back propagation algorithm. The weights of (N1, N2, N3) can be updated. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

Of course, the neural network according to an embodiment of the present invention may be learned in an unsupervised learning manner based on unlabeled learning data. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

Meanwhile, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention may generate learning data from raw patient information and / or environment information for the aforementioned learning.

First, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention may receive a plurality of patient information and / or environment information from the user terminal 200 and / or the external device 300. For example, the apparatus 110 for determining the probability of developing pneumonia may receive a plurality of patient information from a server (not shown) of a medical institution. In this case, the plurality of patient information received may or may not be related to pneumonia. In other words, the plurality of patients may be patients associated with pneumonia or patients associated with other diseases.

In addition, the apparatus 110 for determining the probability of developing pneumonia may receive environment information of a plurality of time zones of a plurality of regions from a weather information providing server (not shown). In this case, the environmental information may also include environmental information of a region and / or a time point related to the patient, or may include environmental information of a region and / or a time point unrelated to the patient. Accordingly, in order to use the received information for learning in the neural network, it may be necessary to properly process.

The apparatus 110 for determining the incidence of pneumonia according to an embodiment of the present invention may generate learning data by extracting patient information related to pneumonia from a plurality of patient information. At this time, the apparatus 110 for determining the probability of developing pneumonia may extract patient information related to pneumonia on the basis of whether pneumonia occurrence diagnosis code (eg ICD10 standard, J100 / J110 / J12 ~ 18 / J851, etc.) is included in the patient's medical information. Can be. In addition, the apparatus 110 for determining the probability of developing pneumonia may extract patient information related to pneumonia based on whether a medical image of a chest is included in the medical information of the patient. In addition, the apparatus 110 for determining the incidence of pneumonia is associated with pneumonia in a patient's medical information, or includes or suggests pneumonia (Pneumonia, Pneumonic, Consolidation, Infiltration, Ground glass opacity (opacities) = GGO, Parapneumonic, etc.) Based on the patient information associated with pneumonia can be extracted. In addition, the apparatus 110 for determining the incidence of pneumonia may extract patient information related to pneumonia based on the presence or absence of treatment within a predetermined period before the onset of pneumonia.

In an exemplary embodiment, the apparatus 110 for determining the probability of developing pneumonia may extract patient information corresponding to any one of the four criteria described above as patient information related to pneumonia, and may correspond to all four criteria described above. Only patient information may be extracted as patient information related to pneumonia.

The apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention may generate learning data based on a plurality of patient information extracted according to the above criteria. In this case, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention may generate learning data including appropriate environment information with reference to personal information of a patient included in each patient information.

For example, the apparatus 110 for determining the incidence of pneumonia may determine learning area of a patient and the timing of occurrence of pneumonia of the patient, and map learning data by mapping the environmental information of the corresponding region and period to the patient's degree. Can be generated.

As such, the present invention can generate appropriate learning data for learning of neural networks, in particular for learning of neural networks that predict the onset probability of pneumonia.

The apparatus 110 for determining the incidence of pneumonia according to the present invention can determine the incidence of pneumonia with higher accuracy by learning a neural network based on the learning data generated by the above-described process.

Subsequently, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention obtains patient and environmental information (S42), and uses the same to include at least one input node N1 included in the input layer L1 described above. Node value may be determined (S43).

For convenience of description, the neural network may include a first input node corresponding to the personal information of the patient, a second input node corresponding to the vital sign information of the patient, and a third input node corresponding to the medical information of the patient in the input layer L1. Suppose you include it.

In this case, the apparatus 110 for determining the occurrence probability of pneumonia according to an embodiment of the present invention may determine the node value of each of the first to third input nodes from the obtained information. In this case, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention may extract a feature value for each item by applying a convolutional neural network (CNN) to information of each item, and extract the extracted feature value as a vector. It may be in the form of.

In other words, the apparatus 110 for determining the incidence of pneumonia according to an embodiment of the present invention applies a convolutional neural network to personal information of a patient, thereby applying a first vector corresponding to the personal information of the patient to a node of the first input node. Can be determined by value. Similarly, the apparatus 110 for determining the probability of developing pneumonia may determine the node value of the second input node from the vital sign information of the patient and the node value of the third input node from the medical information of the patient.

The input layer L1 may include a fourth input node corresponding to the information about the environment, in addition to the above-described information about the patient. In this case, the fourth input node may be a node having a node value corresponding to the concentration of fine dust. Of course, the node value of the fourth input node may be generated by the apparatus 110 for determining the probability of occurrence of pneumonia by applying a convolutional neural network to information about the environment.

Although the apparatus for determining the probability of occurrence of pneumonia 110 applies a convolutional neural network to information of each item as a method of determining the node values of the first to fourth input nodes of the above-described input layer, it is necessary. You do not have to use a convolutional neural network like this. Therefore, various algorithms may be used to derive the feature values of each item according to the data characteristics and / or the design of the system.

Meanwhile, the node values of the first to fourth input nodes may be vectors of n, m, i, and j dimensions, respectively.

Meanwhile, the apparatus 110 for determining the incidence of pneumonia may be obtained by receiving the above-described patient and environment information from the user terminal 200 and / or the external device 300 described above, and read from the memory 113. May obtain patient and environmental information in a manner.

Apparatus 110 for determining the incidence of pneumonia according to an embodiment of the present invention is a node of at least one output node included in the output layer from the pneumonia onset probability calculator based on the node value of the input nodes determined by the above-described process The value can be determined. (S44)

In other words, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention may determine a node value of an output node output by a neural network when node values of input nodes are input.

Looking at this process in detail, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention inputs the values of the input nodes into the first function F1 and outputs the node values of the nodes of the first hidden layer. You can decide. Similarly, the apparatus 110 for determining the occurrence probability of pneumonia may determine the node values of the nodes of the second hidden layer by inputting the node values of the nodes of the first hidden layer into a function defining a relationship between the hidden layers. In this case, the above-described 'function defining a relationship between hidden layers' may be a function defining a relationship between a first hidden layer and a second hidden layer. In a similar manner, the apparatus 110 for determining the probability of occurrence of pneumonia may determine node values of output nodes by inputting and outputting node values of nodes of the last hidden layer to a second function.

As described above, the apparatus 110 for determining the probability of occurrence of pneumonia may determine an output of a neural network corresponding to node values of input nodes, that is, a node value of an output node.

The apparatus 110 for determining the incidence of pneumonia according to an embodiment of the present invention may determine a pneumonia incidence of a patient based on the node value of the output node described above (S45). In this case, according to an embodiment of the present invention The output layer L3 may include a first output node corresponding to the probability of developing pneumonia in the patient. The node value of the first output node may be in the form of a vector such as [0, 1, 0, 0, 0, 0, 0, 0, 0, 0], where each component of the vector corresponds to a probability Can be. For example, each of the ten components described above may correspond to the probability of developing pneumonia of 100%, 90%, 80% 70%, 60%, 50%, 40%, 30%, 20%, 10%.

Meanwhile, the neural network according to an exemplary embodiment of the present invention may be further learned by including a probability of death over time of the patient and a probability of occurrence of the first disease of the patient.

For example, the apparatus 110 for determining the probability of developing pneumonia according to an embodiment of the present invention may learn a probability of death according to a patient's time and a probability of occurrence of a first disease of a patient in a supervised learning manner. In this case, in addition to the probability of developing pneumonia, the learning data may be further labeled with a probability of death and a patient's first disease. Of course, the apparatus 110 for determining the incidence of pneumonia can learn the probability of death according to the time of the patient and the incidence of the first disease of the patient in an unsupervised learning manner. However, this is merely exemplary and the spirit of the present invention is not limited thereto.

The output layer L3 according to an embodiment of the present invention, together with the aforementioned first output node, includes a second output node corresponding to the probability of death over time of the patient and a second corresponding to the probability of occurrence of the first disease of the patient. It may further include three output nodes.

Meanwhile, such a first disease may be a plurality of diseases associated with pneumonia. When the first disease is plural, the output layer L3 may include a third output node for each of the plurality of first diseases.

As described above, the present invention can reasonably infer the probability of occurrence of a specific disease, especially pneumonia, from the information of the patient, and can also infer the probability of death of the patient.

5 is an example of a screen 500 displayed on the user terminal 200 according to an embodiment of the present invention.

Hereinafter, for convenience of description, the user of the user terminal 200 transmits a plurality of items of information about a patient (diagnosis target patient) who wants to know the incidence of pneumonia to the apparatus 110 for determining the incidence of pneumonia, and It is assumed that the result is received from the apparatus for determining the probability of occurrence of 110 and displayed as shown on the screen 500.

Referring to FIG. 5, the screen 500 includes an area 510 in which patient information is displayed, an area 520 in which environmental information is displayed, an area 530 in which a probability of developing pneumonia is displayed, and an area in which associated probability is displayed. 540 may include.

In the area 510 in which the patient information is displayed, vital sign information and medical information, as well as personal information of the patient, may be displayed. In other words, the apparatus 110 for determining the probability of occurrence of pneumonia according to an embodiment of the present invention may provide the user terminal 200 with main information about a patient.

In the area 520 in which the environmental information is displayed, fine dust numerical information of an area and a time point that is expected to be related to the pneumonia development of the patient may be displayed. In addition, in the area 520 where the environmental information is displayed, the threshold fine dust value 521 is expected to have a significant correlation with the invention of pneumonia, and the threshold fine dust value 521 is expected to contribute to the development of pneumonia. Time period 522 may be displayed. Accordingly, the apparatus 110 for determining the incidence of pneumonia according to an embodiment of the present invention extracts fine dust numerical information of an area and a time point that is expected to be related to pneumonia symptoms of a patient from environmental information, and then the user terminal 200. ) Can be provided. For example, the apparatus 110 for determining the incidence of pneumonia according to an embodiment of the present invention may provide fine dust numerical information at a predetermined time point in a predetermined area in consideration of a living area and a hospitalization time point of a patient.

An area 530 in which the probability of developing pneumonia is displayed may display a degree of occurrence of pneumonia over time. In this case, the risk section 531 may be further displayed in the region 530 in which the probability of developing pneumonia is displayed. The incidence of pneumonia displayed in the area 530 in which the incidence of pneumonia is displayed may be based on the node value of the first output node.

In the region 540 in which the associated probability is displayed, for example, the probability of death of the patient over time and the probability of occurrence of the first disease may be further displayed. In this case, the probability of death of the patient and the probability of occurrence of the first disease may also be based on the node value of the output node corresponding to each included in the output layer L3.

Embodiments according to the present invention described above may be implemented in the form of a computer program that can be executed through various components on a computer, such a computer program may be recorded in a computer-readable medium. In this case, the medium may be a program executable by a computer. Examples of media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, And ROM, RAM, flash memory, and the like, configured to store program instructions.

On the other hand, the computer program may be one specially designed and configured for the present invention or may be known and available to those skilled in the computer software field. Examples of computer programs may include high-level language code that can be executed by a computer using an interpreter as well as machine code such as produced by a compiler.

Particular implementations described in the present invention are embodiments and do not limit the scope of the present invention in any way. For brevity of description, descriptions of conventional electronic configurations, control systems, software, and other functional aspects of the systems may be omitted. In addition, the connection or connection members of the lines between the components shown in the drawings are illustrative of the functional connection and / or physical or circuit connections as an example, in the actual device replaceable or additional various functional connections, physical It can be represented as a connection, or circuit connections. In addition, unless specifically mentioned, such as "essential", "important" may not be a necessary component for the application of the present invention.

Therefore, the spirit of the present invention should not be limited to the above-described embodiments, and all the scope equivalent to or equivalent to the scope of the claims as well as the claims to be described below are within the scope of the spirit of the present invention. Will belong to.

100: server
110: device for determining the probability of developing pneumonia
111: communication unit
112: control unit
113: memory
200: user terminal
300: external device
400: network

Claims (6)

In the method of determining the incidence of pneumonia using a device for determining the incidence of pneumonia,
Obtaining a plurality of items of information on the diagnosis target patient and information on the environment;
Determining a node value of at least one input node included in an input layer by using a plurality of items of information about the diagnosis target patient and information about an environment;
Determining a node value of at least one output node included in an output layer from a learned pneumonia incidence probability calculator based on the node value of the at least one input node; And
Determining a probability of developing pneumonia of the patient to be diagnosed based on node values of the at least one output node;
The pneumonia incidence probability calculator is configured to provide information on the plurality of items based on learning data including information on a plurality of items for each of a plurality of patients, information on a plurality of environments, and onset probability of pneumonia for each of the plurality of patients. And a neural network learned about a correlation between information on a plurality of environments and a probability of occurrence of pneumonia.
The neural network may include the input layer, the output layer, a hidden layer defined between the input layer and the output layer, a first function between the input layer and the hidden layer, and a second function between the hidden layer and the output layer. A method for determining the probability of developing pneumonia. The method of claim 1
Multiple items of information about the patient
At least one of personal information of the patient, vital sign information of the patient, and medical information of the patient;
The medical information of the patient
And at least one of a past medical history of the patient, a medical image of the patient, a history of taking the drug of the patient, and a history of symptoms of the patient. The method of claim 2
The input layer is
And a third input node to which the patient's personal information is input, a second input node to which the vital sign information of the patient is input, and a third input node to which the medical information of the patient is input. The method of claim 3,
The information about the environment
At least one of information on the concentration of atmospheric particulate matter, humidity, and ultraviolet ray value for a predetermined period of time;
The input layer is
And a fourth input node to which the concentration of the fine dust is input. The method of claim 1
The probability of determining the incidence of pneumonia is
Training the neural network based on the training data;
Learning the neural network
Classifying information of a patient including a pneumonia incidence diagnosis code into first valid learning data in a plurality of items of information about any patient received from an external device;
Classifying information of a patient including medical images related to pneumonia into second valid learning data in a plurality of items of information about any patient received from the external device;
Classifying patient information including a keyword suggestive of pneumonia into third valid learning data in a plurality of items of information about any patient received from the external device;
Classifying patient information including a medical history of a specific disease within a predetermined period of time before the onset of pneumonia into fourth valid learning data, in a plurality of items of information about any patient received from the external device; And
And determining at least a portion of the first valid learning data, the second valid learning data, the third valid learning data, and the fourth valid learning data as the learning data. The method of claim 5
Determining the learning data is
And determining learning data belonging to all of the first valid learning data, the second valid learning data, the third valid learning data, and the fourth valid learning data as the learning data.

Images