KR20180110310A - System and method for predicting and analysis stroke - Google Patents

Abstract

The present invention relates to a system and a method for predicting and analyzing a stroke, capable of finding out the stroke in advance in real time. The apparatus includes: a wearable data collecting device which collects real-time daily-life data through a wearable device; a history information data collecting device which collects information provided by Health Insurance Management Corporation and electronic medical record data of each person; and a stroke predicting device which finds out and predicts the stroke by using the data collected by the wearable data collecting device and the history information data collecting device. The stroke predicting device, which finds out and predicts the stroke, includes: a data pre-processing unit, which pre-processes and integrates the wearable data and the history information data together; and a stroke real-time finding and predicting unit which selects data having an ensemble structure from data input from the data pre-processing unit, generates multi-model learning & models by using the selected data, and predicts the stroke based on the multi-model learning & models.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system and method for predicting stroke,

The present invention relates to a stroke prediction system and method, and more particularly, to a stroke prediction system and method capable of early detection and prediction of stroke in real time using multi-machine learning or hierarchical multiple SVM.

Internet of Things (IoT) The Internet of Everything (IoE) is one of the technologies to connect the tangible and intangible things around us and provide more convenient and safe services to people. Recently, there has been great interest in various fields. The range of things in the IoT / IoE environment is the physical things that make up the natural environment such as people, cars, clocks, plants, roads, buildings, bridges, In addition, a virtual object such as a web, a social information, a database stored in a computer, and the like can be widely included.

Recently, IoT and IoE technologies, unlike existing Internet and communication methods, autonomously judge current situation or appropriate services, exchange information with each other, and provide new services. Services offered through IoT / IoE include public IoT / IoE services to help solve societal problems, individual IoT / IoE services for people's convenience and quality of life, Industrial IoT / IoE service, etc. In addition, various services such as health care, agriculture / fisheries, energy management, and distribution are being provided.

According to the Global Disease Burden Study (GBD), in the analysis of mortality from 20 major diseases from 1990 to 2013, the first and second were analyzed as ischemic heart disease and stroke, respectively. It is estimated that domestic socio-economic costs due to the occurrence of stroke disease vary depending on the method of calculation, but it is estimated to reach 3,700 billion won in 2005 and 4,150 billion won in 2010. In particular, it is very likely that a stroke is misdiagnosed as a malady in the golden time, or that a hemorrhagic / ischemic disease is diagnosed as a misdiagnosis of a stroke, which leads to a patient's disability or death in a short time. Early diagnosis of posterior cerebral infarction with dizziness related to this rate of misdiagnosis is about 35%. In 2009, the rate of misdiagnosis of stroke disease was ranked in the top 7 according to the rate of misdiagnosis in the Cook County Hospital in Chicago, USA. Therefore, it is urgent to develop a technique to lower the false rate and to predict and manage the possibility of stroke patients.

According to domestic and overseas research literature, studies are being actively conducted on models for predicting the long-term occurrence of stroke diseases, models for assessing the severity of stroke in hospitals, and prediction models of stroke prognosis. In particular, the long-term predictive model for stroke is provided by the Cox model (Wolf et al., 1991) and the Weibull model (Carroll, 2003) . In addition, research and development of the emergency room stroke parameterizing tool and screening tool instruction are being continued to reduce the time required for diagnosis and treatment of stroke in hospitals or medical institutions.

In recent years, FAST campaigns have been conducted to determine patients' self-awareness symptoms and to find hospitals. However, there is still no method or method for early detection or prediction of acute stroke using IoT / IoE technology and wearable devices or smart medical devices It's hard to see.

One embodiment of the present invention provides a method and system for diagnosing acute stroke in real time based on behavior of people in the IoT / IoE environment and surrounding living environment, individual medical care (e.g., EMR) history data and wearable data collected from real- The present invention provides an early detection and prediction system and method for early detection of stroke and prediction of stroke at the time of occurrence to cope with stroke.

It should be understood, however, that the technical scope of the present invention is not limited to the above-described technical problems, and other technical problems may exist.

According to one aspect of the present invention, there is provided an early detection and prediction system for stroke, comprising: a wearable data collection device for collecting real-time living data through a wearable device; A history information data collection device for collecting individual electronic medical record data, a wearable data collection device, and a stroke detection and prediction device for predicting a stroke disease using data collected from the history information data collection device.

The stroke detection and prediction device may include a data preprocessing unit for preprocessing and integrating the wearable data and the hysteresis information data, a data preprocessing unit for selecting data of an ensemble structure among the data input from the data preprocessing unit, And a stroke real time detection and prediction unit for detecting and predicting a stroke based on the multiple machine learning models.

The system for early detection and prediction of stroke according to another aspect of the present invention includes a wearable data collection device for collecting real-time living data through a wearable device, a history information database for collecting information provided by the health insurance management corporation and individual electronic medical record data And a stroke finding and predicting device for predicting a stroke disease using the collected data, the wearable data collecting device, and the data collected from the history information data collecting device.

Here, the stroke detection and prediction apparatus may include a data preprocessing unit for preprocessing and integrating the wearable data and the history information data, and a data preprocessing unit for automatically inputting the data input from the data preprocessing unit into the types of stroke diseases using hierarchical multi- And a stroke real time detection and prediction unit for classifying and predicting a stroke.

According to another aspect of the present invention, there is provided a method for early detection and prediction of stroke, comprising the steps of collecting daily life data of a user through a wearable device, collecting history information data of health insurance corporation and electronic medical record established in a cloud environment , Finding and predicting a stroke using the daily life data and the history information data, analyzing a stroke prediction result, and visualizing the analysis result.

According to the present invention, in the present invention, not only can the stroke be predicted based on the multiple machine learning using the hysterical health examination data and the wearable device data, but also the hierarchical multiple SVM By predicting the stroke, it is possible to detect the stroke early.

The Stroke cube, which is a data cube model in the present invention, is composed of four dimensions of time, place, disease type, and disease cause. Multidimensional analysis can be performed through various OLAP operations according to the degree of abstraction for each dimension.

Based on such a data cube, the present invention can find and provide a potential pattern inherent in stroke disease through multidimensional analysis of stroke disease and analysis of disease growth rate by period.

In addition, according to the present invention, the stroke can be detected more quickly and accurately using the stroke cube, and the data cube of the present invention can be utilized as a scientific basis for stroke decision making and policy making.

1 is a block diagram of a system for early detection and prediction of stroke according to an embodiment of the present invention.
FIG. 2 is a diagram for explaining a real-time stroke detection and prediction method using multi-machine learning of the stroke detection and prediction unit of FIG. 1;
FIG. 3 is a diagram for explaining a real-time stroke detection and prediction method using the hierarchical multiple SVM of the stroke detection and prediction unit of FIG. 1;
Figure 4 is a detailed block diagram of a Stroke cube, which is a multidimensional data cube for various diseases, including stroke.
5 is a star schema diagram of the data cube of FIG.
6 is a QB void diagram for a multivariate analysis of stroke disease.
FIG. 7 is a block diagram of a data cube-based stroke multiscale analysis of the present invention.
Figure 8 is a two-dimensional data cube query result diagram for time and disease type dimensions.
FIG. 9 is a diagram illustrating a result of performing a Drill-Down operation on the Disease dimension in FIG.
10 is a conceptual hierarchical illustration of the Time dimension.
11 is a conceptual hierarchical illustration of a Location dimension.
Figure 12 is an exemplary conceptual hierarchy for the Disease dimension.
13 is an exemplary conceptual hierarchy for the Cause dimension.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, which will be readily apparent to those skilled in the art. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly explain the present invention in the drawings, parts not related to the description are omitted.

Whenever a component is referred to as "including" an element throughout the specification, it is to be understood that the element may include other elements, not the exclusion of any other element, unless the context clearly dictates otherwise.

Hereinafter, a stroke prediction system 1 and a stroke prediction method according to an embodiment of the present invention will be described in detail with reference to the drawings.

1 is a block diagram of a stroke prediction system 1 according to an embodiment of the present invention.

1, the stroke prediction system 1 includes a wearable data collection device 100, a history information data collection device 200, a stroke detection and prediction device 300, and a stroke analysis device 400 .

The wearable data collecting apparatus 100 collects daily life of wearable device users as data. The wearable data collecting apparatus 100 collects data related to daily life such as sleep data, driving data, and walking data through a wearable device.

In the present invention, the daily life data is exemplified as sleeping, driving and walking, but the present invention is not limited to this, and all the data that can be collected through the wearable device can be collected as daily life data.

The history information data collection device 200 collects information provided by the health insurance corporation established in the cloud environment and individual electronic medical record (EMR) data.

The stroke detection and prediction apparatus 300 includes a real-time living data preprocessing unit 310, a historical care data preprocessing unit 320, a data storage / management unit 330, a stroke real-time discovery and prediction unit 340, A stroke prediction emergency notification unit 360, and a stroke prediction result storage unit 370.

The real-time daily data preprocessing unit 310 preprocesses the data collected in real time in the wearable data collecting apparatus 100, and then transmits the preprocessed data to the data storage / management unit 330 and the stroke real-time detection and prediction unit 340.

The history medical care data preprocessing unit 320 preprocesses the data input from the history information data collection apparatus 200 and transmits the preprocessed data to the data storage / management unit 300 and the stroke real time detection and prediction unit 340.

The pre-processing may be Missing Value Processing, Data Cleaning, Feature Selection, Attribute Subset Selection, and the like.

The data storage / management unit 330 stores the processed data in the real-time life data preprocessing unit 310 and the historical care data preprocessing unit 320, and transfers the stored data to the stroke data analysis apparatus 400. In addition, the analysis result can be received from the data analysis apparatus 400 and stored.

The stroke real-time detection and prediction unit 340 detects and predicts a stroke disease using the preprocessed real-time wearable data and the history care data. The prediction result of the disease is stored in a predicted model history storage unit 350, (360) and the stroke prediction result storage unit (370). The construction and operation of the stroke real-time detection and prediction unit 340 will be described in more detail with reference to FIG. 2 and FIG.

The prediction model history storage unit 350 delivers the plurality of prediction models that have been stored in advance to the stroke real time detection and prediction unit 340 and outputs the prediction results of the disease from the stroke real time detection and prediction unit 340 to the model specification information And hyper parameters to perform incremental / adaptive updating of the predictive model.

The stroke prediction emergency notification unit 360 provides hospital visit medical care or emergency alarm information according to the stroke prediction result.

The stroke prediction result storage unit 370 stores the prediction results, and stores the results of the prediction.

At this time, the knowledge can be implemented through one of the following methods: regularization through association rules of machine learning, regularization through decision tree, and semantic ruleization such as RDF or OWL. Unit 370 is not limited to this and can predict the prediction results through various methods.

The stroke analyzer 400 provides the user with the data of the predicted stroke / knowledge data and the data stored through the data preprocessor by performing a visualization function of the risk factors of the stroke disease, various analyzes thereof, and analysis results thereof.

1 may be implemented in hardware such as software or an FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit), and may perform predetermined roles can do.

However, 'components' are not meant to be limited to software or hardware, and each component may be configured to reside on an addressable storage medium and configured to play one or more processors.

Thus, by way of example, an element may comprise components such as software components, object-oriented software components, class components and task components, processes, functions, attributes, procedures, Routines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables.

The components and functions provided within those components may be combined into a smaller number of components or further separated into additional components.

FIG. 2 is a view for explaining a real-time stroke prediction method using multi-machine learning of the stroke detection and prediction unit 340 of FIG.

In FIG. 2, the real-time multi-modal health data and the big data are data preprocessed by the real-time life data preprocessing unit 310 and the historical care data preprocessing unit 320 (S210) .

The stroke real time detection and prediction unit 340 selects data suitable for the multiple stroke early detection and prediction model of the ensemble structure among the preprocessed multimodal data. When learning and predictable data are determined for each model, the stroke real-time finding and predicting unit 340 loads the hyperparameter and specification information of the predictive model that has been built or stored, learns it by fine tuning, (Multi-model Learning & Model Generator, S220).

The stroke real-time detection and prediction unit 340 collects the early detection and prediction models of the stroke disease generated and selects one or more best models suitable for the service or the user's needs, (Result Combine & Majority Vote, S230).

Finally, the real-time stroke detection and prediction unit 340 performs real-time stroke detection and decision (S240) by selecting an optimal model among the stroke prediction models.

The selected model may be a black box (eg, SVM, ANN, RF, et al.) Model, which has excellent prediction performance and can not be described in the prediction process. May be a model (White Box, Example, Association Rule Mining, Decision Tree, et al.).

As shown in FIG. 2, the stroke real-time detection and prediction unit 340 can detect and predict a real-time stroke using multi-machine learning.

FIG. 3 is a view for explaining a real-time stroke prediction method using the hierarchical multiple SVM of the stroke detection and prediction unit 340 of FIG.

의 관계가 되고, 이 하한 값의 거리에 있는 데이터는 최적 경계 초평면과 가장 가까운 거리에 위치하게 된다. 이 데이터들을 Support Vector라고 부른다. 따라서, 최적 경계 초평면에 의해 분류되는 두 클래스간의 거리는 가 되고, 이때 ρ를 분류 한도(Margin of Separation)라 정의한다.The SVM based on the Statistical Learning Theory (SVM) has shown excellent performance in the field of pattern recognition because it solves the problem by converting the given problem to the Convex Quadratic Problem which is always guaranteed the global optimal solution. The basic principle of SVM starts from the problem of linear separation. Given a d-dimensional input data x _i , consider the problem that the output of the training data is divided into binary values such as {-1, +1}. When the optimal boundary hyperplane is given as a classification function of learning data, the distance r from the learning data to the hyperplane is

And the data at the distance of the lower limit value is located at a distance closest to the optimal boundary hyperplane. These data are called Support Vector. Therefore, the distance between the two classes classified by the optimal boundary hyperplane is , Where ρ is defined as the Margin of Separation.

Due to the functional limitations of SVMs, called binary classifiers, SVMs can not be directly applied to multiple classification problems, such as the traffic classification that a given problem is currently dealing with. Therefore, it is more advantageous to design the decision boundary as a single-class SVM that independently represents the corresponding class than when designing a multi-class SVM. Accordingly, the present invention proposes a new device capable of automatically classifying and predicting types of strokes by designing a multi-class SVM based on Support Vector Data Description (SVDD), which is a typical algorithm of a single class SVM.

가 주어졌을 경우, 각각의 클래스를 분류하기 위한 분류기는 각 클래스의 학습 데이터를 포함하면서 체적을 최소화하는 구체(Sphere)를 구하는 문제로 정의되며, 수학식 1의 최적화 문제를 통하여 수식화 된다.A set of K-data residing on the d-dimensional input space

A classifier for classifying each class is defined as a problem of obtaining a sphere that minimizes the volume while including learning data of each class and is formulated through the optimization problem of Equation 1.

[Equation 1]

는 k-번째 클래스를 표현하는 구체의 중심이며,

은 구체의 반경의 제곱,

는 k-번째의 클래스에 속한 i-번째 학습 데이터가 구체에서 벗어나는 정도를 나타내는 벌점 항이며, C는 상대적 중요성을 조정하는 상수(Trade-off Constant)이다.From here,

Is the center of the sphere representing the k-th class,

Is the square of the radius of the sphere,

Is a penalty term indicating the extent to which the i-th learning data belonging to the k-th class deviates from the sphere, and C is a constant that adjusts the relative importance.

In Equation (2), a Lagrange Function L is introduced to obtain a dual problem (Equation 3).

&Quot; (2) "

변수에 대해서는 최소값을 변수

에 대해서는 최대값을 가져야하므로, 아래의 조건식인 수학식 3을 만족해야 한다.Equation (2)

For the variables,

, It is necessary to satisfy the following expression (3).

&Quot; (3) "

Substituting into the Lagrangian function L in Equation (3), the following problem of Equation (4) can be obtained.

&Quot; (4) "

Spheres defined on the input space can represent only very simple shapes. In order to overcome these limitations, it can be extended to use the sphere defined on the high dimensional feature space F defined through the kernel function k. Since the independent classes of objects or services can express their boundaries more accurately in their respective feature spaces, the learning of the system (1) can be expressed by Equation 5 below in consideration of the independence of the feature spaces to which the respective classes are mapped This is done by obtaining the corresponding Convex QP (Quadratic Problem) problem.

&Quot; (5) "

Particularly, when k (x, x) = 1 is established when the Gaussian kernel is used, the above Equation (5) is simplified as shown in Equation (6) below.

&Quot; (6) "

In the application process after completion of the learning, the determination function of each object or service class is defined as shown in Equation (7).

&Quot; (7) "

값은 각 클래스의 특징 공간상의 경계로부터 해당 테스트 데이터와의 절대 거리를 의미함으로, 서로 다른 특징 공간상의 절대거리를 비교하여 소속 클래스를 결정하는 것은 바람직하지 않다. 따라서 특징 공간상의 절대거리

를 특징 공간상에서 정의되는 구형체의 반경

로 나눔으로서 상대적 거리

을 계산하고, 상대거리가 가장 큰 클래스를 입력 데이터 x의 소속 클래스로 결정한다(아래의 수학식 8 참조).Output of a single SVM defined on different feature spaces

Value means the absolute distance to the test data from the boundary in the feature space of each class, so it is not preferable to determine the belonging class by comparing absolute distances on different feature spaces. Therefore, the absolute distance on the feature space

The radius of the spherical body defined in the feature space

As a result,

And determines the class having the largest relative distance as the belonging class of the input data x (see Equation 8 below).

&Quot; (8) "

Based on the health data P1 shown in FIG. 3, the SVM determines whether the patient is normal and stroke (S310). If the stroke is determined to be normal, the type of stroke is classified S320). The specific stroke can then be categorized for each type of stroke (S330).

As shown in FIGS. 2 and 3, in the present invention, not only the stroke can be predicted based on the multi-machine learning using the history health examination data and the wearable device data, but also the stroke is predicted using the hierarchical multiple SVM , And stroke can be detected early.

In addition, according to the present invention, risk factors and multidimensional analysis of stroke diseases can be performed by storing and knowledge of early detection and prediction results of stroke.

Risk factors and multidimensional analysis for the stroke disease are performed in the stroke analyzer of FIG. 1, and will be described in more detail with reference to FIGS. 4 to 13.

FIG. 4 is a diagram of a multidimensional data cube P2 for various diseases including stroke, and FIG. 5 is a star schema diagram of the data cube P2 of FIG.

The stroke analyzer 400 detects a potential pattern or knowledge inherent in real-time wearable device data and stroke-related disease data accumulated over a long period of time, and provides a data cube capable of performing various analyzes in order to increase the accuracy of prediction and early detection of stroke disease The model P2 can be constructed.

As shown in FIG. 4, the data cube P2 can be constructed for various diseases including brain disease, and can perform multidimensional analysis according to the level of abstraction of the concept hierarchy.

The data cube P2 is a structure in which data can be modeled into several dimensions, and is defined as a dimension and a fact. In the present invention, as an example, it is composed of four dimensions such as Time, Location, Disease, and Cause, and various On-Line Analytical Processing (OLAP) ) Operation can be performed. The fact table is connected to each of the four dimensions by a respective key. The data cube model (P2) named as StrokeCube newly proposed in the present invention is composed of four dimension tables and three fact tables. According to the combination of four dimensions, the number of strokes of the fact table, the risk of stroke, The value is provided by analyzing as a measure value. In the present invention, a star schema is used as a multidimensional analysis model, and a star schema for the data cube P2 can be expressed as shown in FIG.

In FIG. 5, the stroke count defined as the scale is a measure for calculating the number of stroke occurrences, and is provided by analyzing the frequency of occurrence by time, region, and risk factors. Level of stroke is an assessment of the level of risk for stroke predictions and is a measure of the degree of risk from low to high risk (eg, low-risk group 1 to high-risk group 5). Finally, the Disease Analysis scale is a measure that provides information on the causes and causes of stroke.

As described above, the data cube P2 proposed by the present invention can be used as a scientific basis for decision-making and stroke-related policy making for prediction and early detection of stroke disease.

 6 is a QB void diagram for a multivariate analysis of stroke disease.

As shown in FIG. 6, the stroke analyzer 400 of the present invention constructs a cubeoid lattice as a set of given dimensions (Time, Location, Disease, Cause) Provide the number of strokes, risk, and cause analysis at different stages.

The lowest-level summary is the highest level summary, and the 0-dimensional QB void is called the Apex Cuboid.

FIG. 7 is a block diagram of a data cube-based stroke multiscale analysis of the present invention.

As shown in FIG. 7, the OLAP of the data analysis apparatus 400 provides a tool for managing, summarizing, and aggregating a large amount of historical data. These OLAPs can be summarized and aggregated according to the selected dimension according to the purpose of prediction and multivariate analysis of stroke disease. In the present invention, a multi-dimensional analysis can be performed using various OLAP operations such as roll-up, drill-down, dice, and slice.

A roll-up operation is an operation that goes up a concept hierarchy for a selected dimension or performs an aggregation on a data cube by dimension reduction.

A drill-down operation is an operation that is an inverse operation of a roll-up, which means an operation that provides more detailed information of the selected dimension.

A die operation is an operation that selects two or more dimensions to create a partial cube. A slice operation is an operation to select a dimension from a given cube to create a partial cube.

Figure 8 is a two-dimensional data cube query result diagram for time and disease type dimensions.

The stroke analyzer 400 may provide the analysis data as shown in FIG. 8 for the query result as shown in the following query equation (1).

[Query expression 1]

Dice for Time = "January 2015 ~ December 2015" AND

Disease = "ALL" AND Measure = "stroke count"

FIG. 8 shows the two dimensions of time and disease, which are the queries of the query expression 1, and the fact shown in FIG. 8, that is, the scale is the stroke count value.

At this time, the present invention is not limited to the stroke count but may include the Disease Count.

FIG. 9 is a diagram illustrating a result of performing a Drill-Down operation on the Disease dimension in FIG.

The stroke analyzer 400 may query the query result of FIG. 8 as shown in the following query expression 2 and provide analysis data as shown in FIG.

[Query expression 2]

Dice for Time = "April and October and November in 2015" AND

Disease = "each disease types" AND Measure = "stroke count"

According to the query equation 2, the stroke analyzer 400 performs a Drill-Down operation on the Disease dimension, calculates Dice for the March, October, and November cases in which the disease occurrence frequency is large, to provide.

As a result of the analysis in FIG. 9, it can be confirmed that the disease condition occurring by the season is different, and it is also possible to provide a pre-alarm and a warning alarm for a specific disease history person and a highly likely elderly person.

The concept hierarchy of the data analysis apparatus 400 means that consecutive mappings of general concepts from a lower concept set to a higher concept set are mapped (A Sequence of Mapping). This concept hierarchy is illustrated in Figures 10-13.

FIG. 10 is a concept hierarchy diagram for a time dimension, FIG. 11 is an concept hierarchy diagram for a location dimension, FIG. 12 is an concept hierarchy diagram for a Disease dimension, FIG.

As shown in FIG. 10, the attributes in the time dimension table are lattice structures having a partial order, and the concept hierarchy for the time dimension is changed from the sub-concept 10 minutes to hour ) <Day (<month, quarter; Week (week)} <year (year).

The concept hierarchy for the location dimension shown in FIG. 11 is defined as a hierarchical tree structure ranging from the lowest hierarchical level (home) to the 'all' including the country.

The concept hierarchy for the Disease dimension shown in FIG. 12 is defined as a concept hierarchy for disease types, and a subtype based on classification information for disease types and a Specific Types hierarchy for each disease.

The concept hierarchy for the cause cause (Cause) dimension shown in FIG. 13 is a conceptual hierarchy for the types of disease causes, and a 'Control Types' in the subordinate concept 'Each Cause Types' Layer concept is defined.

Thus, in the present invention, a data cube named StrokeCube is composed of four dimensions of time, place, disease type, and disease cause, and multidimensional analysis can be performed through various OLAP operations according to the degree of abstraction for each dimension .

Based on such a data cube, the present invention can find and provide a potential pattern inherent in stroke disease through multidimensional analysis of stroke disease and analysis of disease growth rate by period.

In addition, according to the present invention, a stroke can be detected more accurately and quickly using a data cube named StrokeCube, and the data cube of the present invention can be used as a scientific basis for stroke decision making and policy making.

The stroke prediction method according to an embodiment of the present invention may be implemented in the form of a computer program stored in a medium executed by a computer or a recording medium including instructions executable by the computer. Computer readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable medium may include both computer storage media and communication media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Communication media typically includes any information delivery media, including computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, or other transport mechanism.

While the methods and apparatus of the present invention have been described with reference to particular embodiments, some or all of those elements or operations may be implemented using a computer system having a general purpose hardware architecture.

It will be understood by those skilled in the art that the foregoing description of the present invention is for illustrative purposes only and that those of ordinary skill in the art can readily understand that various changes and modifications may be made without departing from the spirit or essential characteristics of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. For example, each component described as a single entity may be distributed and implemented, and components described as being distributed may also be implemented in a combined form.

The scope of the present invention is defined by the appended claims rather than the detailed description and all changes or modifications derived from the meaning and scope of the claims and their equivalents are to be construed as being included within the scope of the present invention do.

100: Wearable data collecting device
200: History information data collection device
300: stroke detection and prediction device
310: Real-time daily data preprocessing unit
320: Historical care data preprocessing section
330: Data storage / management unit
340: stroke real time detection and prediction unit
350: prediction model history storage unit
360: Stroke prediction emergency notification unit
370: Stroke prediction result storage unit
400: Stroke analyzer

Claims (19)

In a stroke early detection and prediction system,
A wearable data collecting device for collecting real-time living data through a wearable device,
A history information data collection device for collecting information provided by the health insurance management corporation and individual electronic medical record data, and
And a stroke detection and prediction device for detecting and predicting a stroke disease using data collected from the wearable data collection device and the history information data collection device,
The stroke finding and predicting device comprises:
A data preprocessing unit for preprocessing and integrating the wearable data and the history information data,
And a stroke real time detection and prediction unit for generating a plurality of machine learning models using the selected data and predicting a stroke based on the multiple machine learning models, Includes a stroke early detection and prediction system. The method according to claim 1,
The stroke finding and predicting device comprises:
Further comprising a prediction model history storage unit storing a plurality of prediction models and performing cumulative and adaptive updating of the prediction models by receiving model specification information and hyper parameters from the stroke real time detection and prediction unit,
Wherein the stroke real time detection and prediction unit loads the model specification information and the hyperparameter from the prediction model history storage unit and learns fine tuning to generate the multiple machine learning models. 3. The method of claim 2,
The stroke real-time detection and prediction unit integrates the results of the generation of the multiple machine learning models and the results thereof, and performs the detection and prediction of the stroke disease in real time by selecting a model having the highest prediction performance among the multiple machine learning models Early detection and prediction systems for stroke. 3. The method of claim 2,
Wherein the stroke real-time detection and prediction unit integrates the multiple machine learning models and the results thereof, selects a model capable of describing the prediction process and the procedure among the multiple machine learning models, and calculates a stroke Early detection and prediction systems. The method according to claim 1,
Further comprising a stroke analyzer for analyzing, storing, and visualizing data input from the stroke finding and predicting device. 6. The method of claim 5,
The stroke analyzer generates data cubes composed of the four dimensions of time, place, disease type, and cause of disease from the stroke finding and predicting device, and performs various OLAP operations according to the degree of abstraction for each dimension Stroke early detection and prediction system that performs multidimensional analysis. The method according to claim 6,
Wherein the stroke analyzer is capable of analyzing the number of strokes, the risk of stroke, and disease analysis values of a fact table according to a combination of four dimensions of the data cube as a measure value. The method according to claim 6,
The data analyzing apparatus includes an early detection means for performing a multidimensional analysis using at least one OLAP operation of a roll-up, a drill-down, a dice, and a slice, And prediction system. In a stroke early detection and prediction system,
A wearable data collecting device for collecting real-time living data through a wearable device,
A history information data collection device for collecting information provided by the health insurance management corporation and individual electronic medical record data, and
And a stroke detection and prediction device for detecting and predicting a stroke disease using data collected from the wearable data collection device and the history information data collection device,
The stroke finding and predicting device comprises:
A data preprocessing unit for preprocessing and integrating the wearable data and the history information data,
And a stroke real time detection and prediction unit for automatically classifying the data input from the data preprocessing unit into types of stroke disease using hierarchical multiple SVM to predict a stroke. 10. The method of claim 9,
Wherein the stroke real-time detection and prediction unit designs a multi-class SVM based on Support Vector Data Description (SVDD). 10. The method of claim 9,
Further comprising a stroke analyzer for analyzing, storing, and visualizing data input from the stroke finding and predicting device. 12. The method of claim 11,
The stroke analyzer generates a data cube composed of four dimensions of time, place, type of disease, disease cause, and the data inputted from the stroke finding and predicting device, An early detection and prediction system for stroke that performs analysis as an enemy. 12. The method of claim 11,
Wherein the stroke analyzer is capable of analyzing the number of strokes, the risk of stroke, and disease analysis values of a fact table according to a combination of four dimensions of the data cube as a measure value. 12. The method of claim 11,
The data analyzing apparatus includes an early detection means for performing a multidimensional analysis using at least one OLAP operation of a roll-up, a drill-down, a dice, and a slice, And prediction system. Early detection and prediction of stroke in early detection and prediction systems of stroke,
Collecting the daily life data of the user through the wearable device,
Collecting history information data of health insurance corporation and electronic medical record established in a cloud environment,
Detecting and predicting a stroke using the daily life data and the history information data, and
An early detection and prediction method of stroke including a step of analyzing the findings and prediction results of the stroke, and visualizing the analysis results. 16. The method of claim 15,
Wherein the step of detecting and predicting stroke comprises:
Pre-processing the daily life data and the history information data,
Selecting data of an ensemble structure among the preprocessed data and generating multiple machine learning models using the selected data, and
The method comprising the steps of: generating the multiple machine learning models and integrating the results; and selecting a model having the highest prediction performance among the multiple machine learning models to perform detection and prediction of the stroke disease in real time, And prediction method. 16. The method of claim 15,
Wherein the step of detecting and predicting stroke comprises:
Pre-processing the daily life data and the history information data,
Selecting data of an ensemble structure among the preprocessed data and generating multiple machine learning models using the selected data, and
A step of integrating the results of the multi-machine learning model and the results thereof, and performing a stroke stroke diagnosis and prediction in real time by selecting a model capable of predicting a procedure and a procedure among the multiple machine learning models, Discovery and Forecasting Methods. 16. The method of claim 15,
Wherein the step of detecting and predicting stroke comprises:
Wherein said daily life data and said history information data are automatically classified into types of stroke disease using hierarchical multiple SVM to detect and predict strokes. 16. The method of claim 15,
Analyzing the findings and prediction results of the stroke, and visualizing the analysis results,
Generating a data cube consisting of four dimensions of the stroke and the prediction result of time, place, disease type, disease cause; and
And performing an OLAP operation on the data cubes according to the level of abstraction for each dimension to analyze the data cubes in a multidimensional manner.

Images