Method for generating prediction result for predicting occurrence of fatal symptoms of subject in advance and device using same

Abstract

The present invention relates to a method for generating a prediction result for predicting an occurrence of fatal symptoms of a subject in advance, a method for performing data classification by using data augmentation in mechanical learning for the same, and a computing device using the same. Particularly, the computing device according to the present invention acquires vital signs of the subject, converts the same into individuated data, generates analysis information from the individuated data on the basis of a machine learning model, generates a prediction result by referring to the analysis information, and provides the prediction result to an external entity.

What is claimed is: 
     
         1 . A method for a computing apparatus operating based on a machine learning model, the method comprising:
 acquiring first vital signs of a subject during a first time duration;   predicting normal vital information related to a normal state of the subject based on the first vital signs;   converting the acquired first vital signs into second vital signs that are individualized to fit a characteristic of the subject based on the normal vital information;   generating individuation data based on the individualized second vital signs;   generating analysis information about a prediction of fatal symptoms from the individuation data based on the machine learning model; and   generating the prediction result which is determined by a result of predicting occurrence of the fatal symptoms during a predetermined duration based on the analysis information.   
     
     
         2 . The method of  claim 1 , wherein the normal vital information includes a characteristic value for the normal state of the subject predicted based on the first vital signs. 
     
     
         3 . The method of  claim 2 , wherein the second vital signs are converted based on a difference between the characteristic value and the first vital signs. 
     
     
         4 . The method of  claim 3 , wherein the characteristic value is calculated based on an average value of at least one first vital signal corresponding to a second time duration among the first vital signs, and
 wherein the second time duration is a partial time duration predicted that the subject is in the normal state in the first time duration.   
     
     
         5 . The method of  claim 1 , further comprising:
 updating the machine learning model based on evaluation information about the prediction result.   
     
     
         6 . The method of  claim 1 , wherein the first vital signs are acquired from an electronic medical record (EMR) of the subject. 
     
     
         7 . The method of  claim 1 , wherein the individuation data is generated by;
 generating individuation data for the subject by calculating a standard score (z-score) for the second vital signs with reference to an average and variance of vital signs of other subjects.   
     
     
         8 . The method of  claim 1 , wherein the machine learning model comprises an analysis model comprising a recurrent neural network model as an analysis model,
 the recurrent neural network model follows s t =f(Ux t +Ws t-1 ) and y=g(Vs t ),   where the x t  denotes the individuation data that is an input vector at a point in time t or a value processed from the individuation data,   the s t  denotes a hidden state corresponding to a memory of the recurrent neural network model at the point in time t,   the s t-1  denotes the hidden state at the point in time t−1,   the U, V, and W denote neural network parameters shared equally across all the points in times of the recurrent neural network model,   the f denotes a predetermined first activation function that is selected to calculate the hidden state,   the y denotes an output layer that is a latent feature according to the recurrent neural network model at the point in time t as the analysis information, and   the g denotes a predetermined second activation function that is selected to calculate the output layer, and   the machine learning model further comprises a prediction model comprising at least one fully connected layer for calculating an occurrence probability of the fatal symptoms from the output layer.   
     
     
         9 . The method of  claim 1 , wherein the fatal symptoms comprise an occurrence of cardiac arrest or sepsis. 
     
     
         10 . The method of  claim 1 , further comprising:
 providing the generated prediction result to an external entity.   
     
     
         11 . A non-transitory computer-readable storage medium storing a program instructions that is executable by a computer to perform the method of  claim 1 . 
     
     
         12 . A computing apparatus operating based on a machine learning model, the computing apparatus comprising:
 a communicator; and   a processor configured to communicate through the communicator   wherein the processor is configured to:   acquire first vital signs of a subject during a first time duration,   predict normal vital information related to a normal state of the subject based on the first vital signs   convert the acquired first vital signs into second vital signs that are individualized to fit a characteristic of the subject based on the normal vital information,   generate individuation data based on the individualized second vital signs,   generate analysis information about a prediction of the fatal symptoms from the individuation data based on the machine learning model, and   generate the prediction result determined by a result of predicting occurrence of the fatal symptoms during a duration based on the analysis information.   
     
     
         13 . The apparatus of  claim 12 , wherein the processor is configured to update the machine learning model based on evaluation information about the prediction result. 
     
     
         14 . The apparatus of  claim 12 , wherein the machine learning model comprises:
 an analysis model comprising a recurrent neural network model; and   a prediction model,   wherein the recurrent neural network model follows s t =f(Ux t +Ws t-1 ) and y=g(Vs t ),   where the x t  denotes the individuation data that is an input vector at the point in time t or a value processed from the individuation data,   the s t  denotes a hidden state corresponding to a memory of the recurrent neural network model at the point in time t,   the s t-1  denotes the hidden state at the point in time t−1,   the U, V, and W denote neural network parameters shared equally across all the points in times of the recurrent neural network model,   the f denotes a predetermined first activation function that is selected to calculate the hidden state,   the y denotes an output layer that is a latent feature according to the recurrent neural network model at the point in time t as the analysis information, and   the g denotes a predetermined second activation function that is selected to calculate the output layer,   wherein the prediction model comprising at least one fully connected layer for calculating an occurrence probability of the fatal symptoms from the output layer.   
     
     
         15 . The apparatus of  claim 12 , wherein the fatal symptoms comprise an occurrence of cardiac arrest or sepsis.