KR102190682B1 - Method and system for screening diabetes using multiple biological signals - Google Patents

Abstract

A diabetes screening method and a diabetes screening system using multiple biological signals are disclosed. A diabetes screening method using multiple bio-signals according to an embodiment of the present invention includes the steps of measuring multiple bio-signals for a patient, and using a plurality of bio-signals selected from the multiple bio-signals. Constructing, and determining an order for the plurality of individual diagnostic models in consideration of performance related to diabetes prediction, and forming an integrated diagnostic model by integrating the plurality of individual diagnostic models according to the predetermined order, and And determining whether the patient has diabetes based on the final prediction probability value calculated using the integrated diagnostic model.

Description

Diabetes screening method and diabetes screening system using multiple biological signals {METHOD AND SYSTEM FOR SCREENING DIABETES USING MULTIPLE BIOLOGICAL SIGNALS}

The present invention relates to a non-invasive diabetes screening technology using multiple biological signals, and more particularly, to improvement of accuracy and reliability of diabetes screening through the integration of individual predictive models for diabetes determination.

Various prior literatures are known that present a technique for diagnosing diabetes using a bioelectrical method or a non-invasive biosignal using a spectral method.

1) US publication number: 2012-0065514 (2012.03.15), "Cardiohealth Methods and Apparatus"

2) US Publication No.: 10-2010-0081941 (2010.04.01), "CARDIOVASCULAR HEALTH STATION METHODS AND APPARATUS"

3) Korean registration number: 10-1440735 (2014.09.04), "Blood glucose measurement system and method that minimizes the number of blood collection"

Prior literatures have applied current to the human body to measure impedance and conductivity, and measure diabetes and body resistance based on biometric information measured by emitting infrared rays to measure electrocardiogram, or to measure pulse rate by applying infrared rays, based on this. It discloses about detecting.

As such, prior literature suggests a non-invasive diabetes screening technique that directly predicts blood sugar based on measured biometric information, but fails to present a predictive probability value for diabetes using information obtained from multiple bio signals.

In addition, in the existing prediction model, since the discrimination result may vary depending on the type of bio-signal, an integrated prediction model capable of predicting diabetes by analyzing a number of bio-signals is required, but by simply merging individual prediction models. However, there may be limitations in improving the accuracy of diabetes prediction.

Accordingly, in this specification, by sequentially combining individual prediction models constructed using multiple bio-signals according to the performance level, an integrated prediction model capable of more accurately predicting the risk of diabetes suspicious group can be constructed. I would like to propose a technique.

An embodiment of the present invention aims to improve the accuracy of diabetic discrimination through an integrated prediction model using multiple biological signals.

In an embodiment of the present invention, when forming an integrated prediction model for diabetic discrimination by combining individual prediction models constructed using multiple bio signals, the accuracy of diabetic discrimination and the accuracy of diabetic discrimination and It aims to improve reliability.

In an embodiment of the present invention, in the case of a diabetes suspected group whose predicted probability value falls within the range of, for example, '0.45 to 0.55%', after sequentially combining a certain number of individual diagnostic models with high diagnostic performance for each biological signal, an arbitrary biological signal The purpose of this study is to propose an integrated model capable of more accurate diabetes diagnosis by forming an integrated diagnosis model by additionally applying an individual diagnosis model according to.

An embodiment of the present invention proposes a method that can effectively integrate an individual diagnostic model according to the arbitrary biosignal into an existing diagnostic model when an arbitrary biosignal is added to improve the accuracy of diabetes screening. The purpose of this study is to improve accuracy through expansion of diagnostic models.

A diabetes screening method using multiple bio-signals according to an embodiment of the present invention includes the steps of measuring multiple bio-signals for a patient, and using a plurality of bio-signals selected from the multiple bio-signals. Constructing, and determining an order for the plurality of individual diagnostic models in consideration of performance related to diabetes prediction, and forming an integrated diagnostic model by integrating the plurality of individual diagnostic models according to the predetermined order, and Based on the final prediction probability value calculated using the integrated diagnostic model, it may include determining whether the patient has diabetes.

In addition, the diabetes screening system using multiple bio-signals according to an embodiment of the present invention uses a measuring unit for measuring multiple bio-signals related to a patient, and a plurality of bio-signals selected from among the multiple bio-signals. A model construction unit for constructing a diagnostic model, a performance evaluation unit for determining an order for the plurality of individual diagnostic models in consideration of performance related to diabetes prediction, and an integrated diagnostic model by integrating the plurality of individual diagnostic models according to the predetermined order. And, based on the final predicted probability value calculated using the integrated diagnostic model, a processing unit to determine whether the patient has diabetes.

According to an embodiment of the present invention, when diabetic prediction is performed based on an individual diabetes prediction model constructed by analyzing multiple biological signals of two or more of bioimpedance, electrochemical skin conductivity, and near-infrared, the prediction is unclear cases (0.45~ 0.55%), it is possible to perform accurate diabetes prediction by sequentially synthesizing prediction models according to the performance level of each prediction model.

According to an embodiment of the present invention, when developing a non-invasive diabetes screening device using multiple bio-signals, reliability and accuracy of screening may be improved through model integration.

According to an embodiment of the present invention, since the probability of diabetes is predicted through repetitive derivation of probability values using information obtained through multiple biological signals, a more accurate diagnosis can be performed.

According to an embodiment of the present invention, a diabetes diagnosis model is formed using information obtained through multiple bio-signals, but by identifying the high and low performance of each model, for a case designated as a diabetes suspected group, a diagnosis with excellent performance By sequentially applying the model and re-diagnosing it, the accuracy of diabetes prediction can be improved.

According to an embodiment of the present invention, when multiple biosignal measuring equipment such as'BIA' and'NIRS' is used as a diabetes screening tool, its accuracy can be improved.

According to an embodiment of the present invention, a prediction model for diabetes is generated by using variable information obtained by measuring biological impedance, and diabetes is predicted using variable information obtained by measuring electrochemical skin conductivity. Diabetes is predicted by creating a model, predicting diabetes by generating a prediction model for diabetes using variable information acquired by shooting near-infrared rays into the human body, and using each prediction model sequentially according to the level of diabetes prediction performance to suspect diabetes. You can accurately predict whether the group is at risk for diabetes.

1 is a block diagram showing the configuration of a diabetes screening system using multiple biological signals according to an embodiment of the present invention.
2A to 2C are graphs showing results of diabetic determination performance evaluation according to individual diagnostic models in the diabetes screening system according to an embodiment of the present invention.
3 is a graph showing a performance evaluation result according to an integrated diagnostic model in the diabetes screening system according to an embodiment of the present invention.
4 is a flow chart showing a procedure of a diabetes screening method using multiple biological signals according to an embodiment of the present invention.

Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings. However, since various changes may be made to the embodiments, the scope of the rights of the patent application is not limited or limited by these embodiments. It should be understood that all changes, equivalents, or substitutes to the embodiments are included in the scope of the rights.

The terms used in the examples are used for illustrative purposes only and should not be interpreted as limiting. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present specification, terms such as "comprise" or "have" are intended to designate the presence of features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, but one or more other features. It is to be understood that the presence or addition of elements or numbers, steps, actions, components, parts, or combinations thereof, does not preclude in advance.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiment belongs. Terms as defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in this application. Does not.

In addition, in the description with reference to the accompanying drawings, the same reference numerals are assigned to the same components regardless of the reference numerals, and redundant descriptions thereof will be omitted. In describing the embodiments, when it is determined that a detailed description of related known technologies may unnecessarily obscure the subject matter of the embodiments, the detailed description thereof will be omitted.

1 is a block diagram showing the configuration of a diabetes screening system using multiple biological signals according to an embodiment of the present invention.

Referring to FIG. 1, a diabetes screening system 100 using multiple biological signals according to an embodiment of the present invention includes a measurement unit 110, a model construction unit 120, a processing unit 130, and a performance evaluation unit 140. ) And the memory unit 150 may be included.

The measurement unit 110 measures multiple biological signals from the patient.

That is, the measuring unit 110 may simultaneously measure multiple biological signals related to the patient's diabetes level, such as a bioimpedance signal, an electrochemical skin conductivity signal, and a near-infrared signal, from the patient.

For example, the measurement unit 110 may measure a bioimpedance signal that responds to an AC current applied to a patient through an impedance measuring device (BIA) based on multiple frequencies.

In addition, the measurement unit 110 may measure a conductivity signal in response to a direct current applied to a patient through a conductivity meter (ESC).

In addition, the measurement unit 110 may measure a near-infrared signal irradiated to the patient's skin through a near-infrared measuring instrument (NIRS).

The measurement unit 110 may record multiple bio-signals measured for a predetermined period in the memory unit 150 in order to collect multiple bio-signals in an amount required for learning modeling.

As an example, the measurement unit 110 reacts to a biological impedance signal that responds to the AC current applied to the patient measured through an impedance meter (BIA), and responds to the DC current applied to the patient measured through a conductivity meter (ESC). At least one biosignal of the conductivity signal measured through the near-infrared measuring instrument (NIRS) and the near-infrared signal irradiated to the patient's skin may be recorded in the memory unit 150 as the multiple biosignal.

The model building unit 120 constructs a plurality of individual diagnostic models by using a plurality of bio signals selected from the multiple bio signals.

The model building unit 120 analyzes two or more types of bio-signals selected from the memory unit 150 according to an ensemble technique among a logistic regression model to which a penalty point function is applied or a machine learning algorithm, and analyzes each of the plurality of individual diagnostic models. Can build.

For example, the model construction unit 120 analyzes the bio-impedance signal measured through an impedance meter (BIA) according to a logistic regression model to which a penalty point function is applied, and calculates a predictive probability value for diabetes prediction. You can build A.

In addition, the model building unit 120 may construct a diagnostic model B that analyzes the conductivity signal measured through a conductivity meter (ESC) according to an ensemble technique to calculate a prediction probability value for predicting diabetes.

In addition, the model construction unit 120 analyzes the near-infrared signal measured through a near-infrared measuring instrument (NIRS) according to a combination of a logistic regression model to which a penalty point function is applied and the ensemble technique, and calculates a prediction probability value for diabetes prediction. Diagnostic model C can be constructed.

The performance evaluation unit 140 determines an order for the plurality of individual diagnostic models in consideration of performance related to diabetes prediction.

That is, the performance evaluation unit 140 analyzes the bioimpedance signal, the conductivity signal, and the near-infrared signal measured from the patient, respectively, evaluates the performance of the plurality of individual diagnostic models, and based on the evaluation result of the performance, The order for sequentially integrating each of the plurality of individual diagnostic models may be determined.

The above sequence is a sequence for integrating each of the plurality of individual diagnostic models sequentially and step by step, and the performance evaluation unit 140 evaluates the diabetes prediction performance of the plurality of individual diagnostic models, and provides an individual diagnostic model with excellent diabetes prediction performance. You can give the order of relative dominance.

Specifically, the performance evaluation unit 140 evaluates whether the diabetes prediction result according to the prediction probability value calculated using the individual diagnosis model is close to the actual diabetes diagnosis result of the patient, and based on the evaluation result, the The order for sequentially integrating each of the plurality of individual diagnostic models may be determined.

For example, the performance evaluation unit 140 includes a diagnostic model A constructed using the patient's bio-impedance signal, a diagnostic model B constructed using the patient's conductivity signal, and a diagnostic model C constructed using the near-infrared signal. Diabetes prediction is performed by calculating a predicted probability value, and the diabetes prediction performance of an individual diagnostic model can be evaluated according to the degree to which the diabetes prediction result matches (closes) the diabetes diagnosis result of an actual patient.

The performance evaluation unit 140 sets the order of diagnostic models A with the best performance to '1', and then sets the order of diagnostic models C with excellent performance to '2', and the order of diagnostic models B with the lowest performance. Can be set to '3'.

As another example, the performance evaluation unit 140 evaluates the optimum coefficient and tuning parameter through cross-validation for each of a plurality of individual diagnostic models, and based on the evaluation result of the optimum coefficient and the tuning parameter, the plurality of The above order for sequentially integrating each of the individual diagnostic models may be determined.

That is, the performance evaluation unit 140 may evaluate the superiority (high and low) of the performance of each individual diagnostic model by comparing the optimal coefficient and the tuning parameter of the individual diagnostic model selected through cross-validity verification.

For example, referring to the graphs of FIGS. 2A to 2C, the performance evaluation unit 140 performs a '10-fold' cross-validation in order to select two tuning parameters representing optimal performance from a plurality of individual diagnostic models. Through this, it is possible to select the tuning parameter with the maximum AUC value, compare the selected tuning parameters, and give a relatively advanced order to the individual diagnostic model with a high tuning parameter.

In addition, the performance evaluation unit 140 evaluates a plurality of individual diagnostic models and estimates the optimum coefficient, so that the processing unit 130, which will be described later, integrates the individual diagnostic models step by step in the order of high performance according to the optimum coefficient. In addition, by providing a graph (see 321 in FIG. 3) comparing the performance of the stepwise integrated diagnostic model to the manager, it is possible to determine whether to repeatedly perform the integration of individual diagnostic models.

The processing unit 130 forms an integrated diagnostic model by integrating each of the plurality of individual diagnostic models according to the order determined in consideration of the diabetes prediction performance.

As an example, the processing unit 130 selects and integrates two individual diagnostic models in the order of high performance among a plurality of individual diagnostic models, and selects and integrates an individual diagnostic model with the next highest performance in the integrated two individual diagnostic models. In this way, a plurality of individual diagnostic models can be sequentially and stepwise integrated.

In particular, when the prediction of diabetes by each individual diagnostic model is ambiguous or in a diabetes risk group, the processing unit 130 uses an integrated diagnostic model obtained by stepwise synthesis of individual diagnostic models according to different biological signals, thereby predicting diabetes (diabetes Discrimination) can increase the accuracy.

For example, the processing unit 130 corresponds to a prediction probability value calculated by an individual diagnostic model with the best performance among a plurality of individual diagnostic models to be '0.45 to 0.55%', so that the prediction of diabetes is ambiguous or a case belonging to the diabetes risk group For each diagnostic model, a diagnostic model can be sequentially synthesized according to the performance level of each diagnostic model, and diabetes prediction can be performed through the synthesized integrated diagnostic model.

Hereinafter, a process of synthesizing a plurality of individual diagnostic models constructed from multiple bio-signals in stages and sequentially in consideration of diabetes prediction performance will be described.

Specifically, the processing unit 130 calculates a first prediction probability value from the first linear prediction value calculated by the first individual diagnostic model having the highest order, and the range in which the first prediction probability value is selected (e.g., '0.45~ 0.55%'), by summing the second linear predicted value calculated by the second individual diagnostic model having a higher order excluding the first individual diagnostic model to the first linear predicted value, and calculating an average linear predicted value. , The first and second individual diagnostic models may be integrated.

Here, the selected range is a range for the predicted probability value when it is difficult to determine whether or not diabetes is presumed to belong to a diabetes risk group based on the predicted probability value calculated using an individual diagnostic model, and is '0.45 to 0.55%'. It can be illustrated.

In this way, the processing unit 130 may improve the accuracy of diabetes determination through sequential and stepwise integration in consideration of the performance of a plurality of individual diagnostic models.

On the other hand, for cases in which diabetes prediction is not ambiguous, the processing unit 130 provides the first individual diagnostic model with the best diabetes prediction performance without the need to integrate the second individual diagnostic model into the first individual diagnostic model. Diabetes determination (diabetes prediction) can be performed as it is.

That is, if the first prediction probability value calculated by the first individual diagnostic model with the best performance does not fall within the selected range (eg, '0.45 to 0.55%'), the processor 130 A probability value may be determined as the final predicted probability value, and whether or not diabetes may be determined according to the final predicted probability value.

According to an embodiment, when there is a third individual diagnostic model different from the first and second individual diagnostic models, the processing unit 130 further adds a third linear prediction value calculated by the third individual diagnostic model, By calculating the average linear prediction value, the third individual diagnostic model may be further integrated into the integrated first and second individual diagnostic models.

For example, the processing unit 130 may select a third individual diagnostic model with a higher order after the second individual diagnostic model among a plurality of individual diagnostic models, and further integrate them into the first and second individual diagnostic models. have.

In addition, the processing unit 130 randomly selects a third individual diagnostic model from among a plurality of individual diagnostic models excluding the first and second individual diagnostic models, regardless of the order, and selects the first and second individual diagnostic models. It can be further integrated.

In addition, the processing unit 130 selects a third individual diagnostic model constructed using a biosignal different from the first and second individual diagnostic models, among a plurality of individual diagnostic models, and selects the first and second individual diagnostic models. Can be further integrated into.

In addition, the processing unit 130 may further integrate an arbitrary individual diagnostic model built before or after the plurality of individual diagnostic models, as the third individual diagnostic model, into the first and second individual diagnostic models. have.

In this way, the processing unit 130 may further improve the accuracy of diabetic determination by the integrated diagnostic model by additionally performing the integration of the individual diagnostic model constructed with different biological signals to the model in which the individual diagnostic model is integrated.

Depending on the embodiment, if the third individual diagnostic model that is not additionally integrated remains, the processing unit 130 may repeat the additional integration process as many as the number of the third individual diagnostic models.

At this time, the processing unit 130 evaluates the performance of diabetes prediction through the performance evaluation unit 140 whenever integration of the third individual diagnostic model is performed, and determines whether to additionally integrate another third individual diagnostic model. You can decide.

If the diabetes prediction performance is evaluated by the performance evaluation unit 140 as excellent, the processing unit 130 may stop additional integration of the third individual diagnostic model.

In this way, the processing unit 130 may continuously improve the accuracy of diabetic determination of the integrated diagnostic model by repeatedly integrating the integrated diagnostic model with the individual diagnostic model constructed with an arbitrary biological signal.

The processing unit 130 determines whether the patient has diabetes based on the final predicted probability value calculated using the integrated diagnostic model.

Specifically, the processing unit 130 calculates a prediction probability value from the average linear prediction value calculated by the integrated diagnostic model, determines the prediction probability value as the final prediction probability value, and whether the diabetes is diagnosed according to the final prediction probability value. Can be determined.

Here, the average linear predicted value may be calculated as an average of linear predicted values calculated by each of a plurality of individual diagnosis models combined when the integrated diagnosis model is formed.

For example, the performance evaluation unit 140 includes an individual diagnostic model ('diagnostic model A') constructed from the bio-impedance signal measured by an impedance measuring device (BIA), and the conductivity signal measured by a conductivity measuring device (ESC). Among the individual diagnostic models ('diagnostic model B') constructed with and the near-infrared signals measured by the NIRS ('diagnostic model C'), the performance varies according to diabetes prediction performance. The order of the most excellent diagnostic model A can be set to '1', the order of the diagnostic model C having excellent performance is set to '2', and the order of the diagnostic model B with the lowest performance can be set to '3'.

The processing unit 130 performs diabetes prediction with a prediction probability value (p ₁ ) calculated using the diagnostic model A having the best diabetes prediction performance, but the range in which the prediction probability value (p ₁ ) is selected ('0.45 ≤ p ₁ ≤ 0.55'), it can be determined that the diabetes prediction result is ambiguous or belongs to the diabetes risk group.

)을 연산하고, 상기 평균 선형예측값으로부터 예측확률값(p₁₊₃)을 다시 산출할 수 있다.In this case, the processing unit 130 adds the linear predicted value (η ₁ ) calculated by the diagnostic model A with the best performance and the linear predicted value (η ₃ ) calculated by the diagnostic model C with the next superior diabetes prediction performance. After averaging, the average linear predicted value (

) May be calculated, and a prediction probability value (p ₁₊₃ ) may be calculated again from the average linear prediction value.

)으로부터, 최종 예측확률값(p₁₊₂₊₃)을 산출할 수 있다.Thereafter, the processing unit 130 further sums the linear predicted value (η ₃ ) calculated by the diagnostic model B constructed using a biological signal ('skin conductivity signal') different from the diagnostic models A and C, and calculated the average linear predicted value (

), a final predicted probability value (p ₁₊₂₊₃ ) can be calculated.

As described above, according to the present invention, in the case of a diabetes suspected group whose prediction probability value falls within the range of, for example, '0.45 to 0.55%', after sequentially combining a certain number of individual diagnostic models with high diagnostic performance for each biological signal, an arbitrary By forming an integrated diagnosis model by additionally applying an individual diagnosis model according to the biological signal, the accuracy of diabetes determination can be further improved.

2A to 2C are graphs showing results of diabetic determination performance evaluation according to individual diagnostic models in the diabetes screening system according to an embodiment of the present invention.

FIG. 2A is a graph showing the performance of a diagnostic model A constructed from a bioimpedance signal measured by an impedance measuring device BIA.

In FIG. 2B, a graph showing the performance of a diagnostic model B constructed from a conductivity signal measured by a conductivity meter (ESC) is shown.

2C is a graph showing the performance of a diagnostic model C constructed from a near-infrared signal measured by a near-infrared measuring instrument (NIRS).

The diabetes screening system compares the performance of individual diagnostic models constructed through multiple bio-signals through the graphs of FIGS. 2A to 2C, and integrates a plurality of individual diagnostic models in the order of superior performance (diagnostic model A> Diagnostic model C> Diagnostic model B) can be set.

For example, the diabetes screening system selects the tuning parameter with the maximum AUC value through '10-fold' cross-validation in order to select two tuning parameters representing optimal performance from a plurality of individual diagnostic models. By comparing the tuning parameters, it is possible to give a relatively advanced order to individual diagnostic models with a high tuning parameter.

3 is a graph showing a performance evaluation result according to an integrated diagnostic model in the diabetes screening system according to an embodiment of the present invention.

3, the diabetes screening system according to an embodiment of the present invention, among a plurality of individual diagnostic models A, B, C (311, 312, 313), the most excellent diagnostic model A (311), Next, a diagnostic model C 313 having excellent performance is synthesized first, and a diagnostic model B 312 having excellent performance is further synthesized to form an integrated diagnostic model 322.

That is, the diabetes screening system includes a diagnostic model A 311 constructed with a biological impedance signal measured by an impedance measuring instrument (BIA), and a diagnostic model C 313 constructed with a near-infrared ray signal measured by a near-infrared ray measuring instrument (NIRS). And, to the synthesized diagnostic model (A+C) 321, a diagnostic model B 312 constructed from a conductivity signal measured by a conductivity meter (ESC) was additionally synthesized, and an integrated diagnostic model 322 Can be formed.

In this way, the diabetes screening system can improve the accuracy and reliability of diabetic determination through step-by-step integration in consideration of the performance of an individual prediction model built using multiple biological signals.

Hereinafter, in FIG. 4, the work flow of the diabetes screening system 100 according to embodiments of the present invention will be described in detail.

4 is a flow chart showing a procedure of a diabetes screening method using multiple biological signals according to an embodiment of the present invention.

The diabetes screening method according to the present embodiment may be performed by the diabetes screening system 100 described above.

Referring to FIG. 4, in step 410, the diabetes screening system 100 measures multiple bio-signals for a patient for a predetermined period of time.

As an example, the diabetes screening system 100 measures a bioimpedance signal in response to an alternating current applied to a patient through an impedance meter (BIA) based on multiple frequencies, and through a conductivity meter (ESC), to the patient. The conductivity signal responding to the applied direct current is measured, and the near-infrared signal irradiated to the patient's skin can be measured through a near-infrared meter (NIRS).

In step 420, the diabetes screening system 100 constructs a plurality of individual diagnostic models by using two or more types of bio-signals selected from the multiple bio-signals.

The diabetes screening system 100 may construct each of the plurality of individual diagnostic models by analyzing two or more types of bio-signals according to an ensemble technique among a logistic regression model to which a penalty point function is applied or a machine learning algorithm.

In step 430, the diabetes screening system 100 integrates a plurality of individual diagnostic models step by step according to an order determined in consideration of the performance related to diabetes prediction.

As an example, the diabetes screening system 100 evaluates the optimal coefficient and tuning parameter through cross-validation for each of a plurality of individual diagnostic models, and based on the evaluation result of the optimal coefficient and the tuning parameter, the plurality of The above order for sequentially integrating each of the individual diagnostic models may be determined.

For example, referring to the graphs of FIGS. 2A to 2C, the diabetes screening system 100, in order to select two tuning parameters representing optimal performance from a plurality of individual diagnostic models, '10-fold' cross-validation By selecting the tuning parameter with the maximum AUC value and comparing the selected tuning parameters, it is possible to give a relatively advanced order to an individual diagnostic model with a high tuning parameter.

In step 440, the diabetes screening system 100 determines whether the patient has diabetes based on the final predicted probability value calculated according to the integrated diagnosis model.

In this step 440, the diabetes screening system 100, the prediction probability value calculated by the individual diagnostic model with the best performance among the plurality of individual diagnostic models corresponds to '0.45 ~ 0.55%', the diabetes prediction is ambiguous or , For cases belonging to the diabetes risk group, diagnostic models can be sequentially synthesized according to the performance of each diagnostic model, and diabetes prediction can be performed through the synthesized integrated diagnostic model.

For example, referring to FIGS. 2A to 2C, the diabetes screening system 100 includes a performance graph of a diagnostic model A constructed from a bioimpedance signal measured by an impedance measuring device (BIA) (FIG. 2A ), and a near-infrared measuring instrument. By comparing the performance graph of the diagnostic model C constructed with the near-infrared signal measured by (NIRS) (Fig. 2b) and the diagnostic model B (Fig. 2c) constructed with the conductivity signal measured by the conductivity meter (ESC), performance In this excellent order, an order (diagnostic model A> diagnosis model C> diagnosis model B) to be integrated into a plurality of individual diagnosis models may be given, and an integrated diagnosis model may be formed according to the order.

Referring to FIG. 3, the diabetes screening system 100 is constructed with a near-infrared signal measured by a near-infrared measuring instrument (NIRS) in a diagnostic model A 311 constructed with a biological impedance signal measured by an impedance measuring instrument (BIA). One diagnostic model C (313) was first synthesized, and then to the synthesized diagnostic model (A+C) 321, a diagnostic model B (312) constructed from the conductivity signal measured by a conductivity meter (ESC) was additionally synthesized. Thus, an integrated diagnostic model 322 can be formed.

As described above, according to an embodiment of the present invention, when diabetic prediction is performed based on an individual diabetes prediction model constructed by analyzing two or more multiple biological signals among bioimpedance, electrochemical skin conductivity, and near-infrared rays, the prediction is unclear. For (0.45~0.55%), accurate prediction of diabetes can be performed by sequentially synthesizing prediction models according to the performance level of each prediction model.

According to an embodiment of the present invention, when developing a non-invasive diabetes screening device using multiple bio-signals, reliability and accuracy of screening may be improved through model integration.

According to an embodiment of the present invention, since the probability of diabetes is predicted through repetitive derivation of probability values using information obtained through multiple biological signals, a more accurate diagnosis can be performed.

According to an embodiment of the present invention, a diabetes diagnosis model is formed using information obtained through multiple bio-signals, but by identifying the high and low performance of each model, for a case designated as a diabetes suspected group, a diagnosis with excellent performance By sequentially applying the model and re-diagnosing it, the accuracy of diabetes prediction can be improved.

According to an embodiment of the present invention, when multiple bio-signal measuring equipment such as'BIA' and'NIRS' is used as a diabetes screening tool, its accuracy can be improved.

According to an embodiment of the present invention, a prediction model for diabetes is generated by using variable information obtained by measuring biological impedance, and diabetes is predicted using variable information obtained by measuring electrochemical skin conductivity. Diabetes is predicted by creating a model, predicting diabetes by generating a prediction model for diabetes using variable information acquired by shooting near-infrared rays into the human body, and using each prediction model sequentially according to the level of diabetes prediction performance to suspect diabetes. You can accurately predict whether the group is at risk for diabetes.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of the program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to behave as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodyed in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

As described above, although the embodiments have been described by the limited drawings, a person of ordinary skill in the art can apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments and claims and equivalents fall within the scope of the following claims.

100: Diabetes Screening System
110: measurement unit 120: model building unit
130: processing unit 140: performance evaluation unit
150: memory unit

Claims (15)

In the diabetes screening method using multiple biological signals implemented by the diabetes screening system,
In the diabetes screening system, measuring multiple biological signals related to a patient;
In the diabetes screening system, constructing a plurality of individual diagnostic models by using a plurality of biological signals selected from among the multiple biological signals;
In the diabetes screening system, determining an order of the plurality of individual diagnostic models in consideration of performance related to diabetes prediction;
Calculating, in the diabetes screening system, a first prediction probability value from a first linear prediction value calculated by a first individual diagnostic model having the highest order;
When the first predicted probability value is within a selected range belonging to the diabetes risk group,
In the diabetes screening system, a second linear prediction value calculated by a second individual diagnosis model having the highest order excluding the first individual diagnosis model is added to the first linear predicted value and then averaged, and an average linear predicted value Forming an integrated diagnostic model by integrating the first and second individual diagnostic models by calculating. And
In the diabetes screening system, determining whether the patient has diabetes based on a final predicted probability value calculated using the integrated diagnostic model
Diabetes screening method using multiple biological signals comprising a. The method of claim 1,
The step of determining the order,
Evaluating whether a diabetes prediction result according to a prediction probability value calculated using an individual diagnosis model is close to an actual diabetes diagnosis result of the patient; And
Determining the sequence for sequentially integrating each of the plurality of individual diagnostic models based on the evaluation result
Diabetes screening method comprising a. The method of claim 1,
The step of determining the order,
Evaluating an optimum coefficient and a tuning parameter through cross-validation of each of the plurality of individual diagnostic models; And
Determining the order for sequentially integrating each of the plurality of individual diagnostic models based on the evaluation result of the optimum coefficient and the tuning parameter
Diabetes screening method comprising a. The method of claim 1,
The step of determining the order,
Analyzing a bioimpedance signal, a conductivity signal, and a near-infrared signal measured from the patient, respectively, and evaluating the performance of the plurality of individual diagnostic models; And
Determining the sequence for sequentially integrating each of the plurality of individual diagnostic models based on the evaluation result of the performance
Diabetes screening method comprising a. delete The method of claim 1,
When there is a third individual diagnostic model different from the first and second individual diagnostic models,
Forming the integrated diagnostic model,
The third linear prediction values calculated by the third individual diagnostic models are further summed and averaged, and the average linear predicted values are calculated, whereby the third individual diagnostic model is added to the integrated first and second individual diagnostic models. Further incorporating; And
If the third individual diagnostic model that has not been further integrated remains, repeating the additional integration step
Diabetes screening method further comprising. The method of claim 1,
The step of determining whether or not diabetes,
Determining a predicted probability value calculated from the average linear predicted value as the final predicted probability value, and determining whether or not the diabetes is based on the final predicted probability value
Diabetes screening method comprising a. The method of claim 1,
If the first prediction probability value does not fall within the selected range,
The step of determining whether or not diabetes,
Determining the first predicted probability value as the final predicted probability value, and determining whether or not the diabetes is diagnosed according to the final predicted probability value
Diabetes screening method comprising a. The method of claim 1,
In the diabetes screening system, a biological impedance signal in response to an AC current applied to the patient measured through an impedance meter (BIA), and a conductivity signal in response to a direct current applied to the patient measured through a conductivity meter (ESC) , And recording at least one of the near-infrared signals irradiated to the patient's skin measured through a near-infrared measuring instrument (NIRS) in a memory unit
Diabetes screening method further comprising. The method of claim 9,
The step of constructing the plurality of individual diagnostic models,
Constructing each of the plurality of individual diagnostic models by analyzing two or more types of biosignals selected from the memory unit according to an ensemble method among a logistic regression model to which a penalty point function is applied or machine learning algorithms
Diabetes screening method comprising a. A measuring unit for measuring multiple biological signals related to a patient;
A model construction unit for constructing a plurality of individual diagnostic models by using a plurality of bio signals selected from the multiple bio signals;
A performance evaluation unit for determining an order of the plurality of individual diagnostic models in consideration of performance related to diabetes prediction; And
The first predicted probability value is calculated from the first linear predicted value calculated by the first individual diagnostic model with the highest order, and if the first predicted probability value falls within a selected range belonging to the diabetes risk group, the first individual diagnosis Excluding the model, the second linear predicted value calculated by the second individual diagnostic model with the highest order is added to the first linear predicted value and then averaged, and the average linear predicted value is calculated. A processing unit that forms an integrated diagnostic model by integrating the diagnostic model, and determines whether the patient has diabetes based on the final predicted probability value calculated using the integrated diagnostic model
Diabetes screening system using multiple biological signals comprising a. The method of claim 11,
The performance evaluation unit,
Through cross-validation for each of the plurality of individual diagnostic models, the optimum coefficient and the tuning parameter are evaluated,
To determine the order for sequentially integrating each of the plurality of individual diagnostic models, based on the evaluation result of the optimal coefficient and the tuning parameter
Diabetes screening system. delete The method of claim 11,
When there is a third individual diagnostic model different from the first and second individual diagnostic models,
The processing unit,
The third linear prediction values calculated by the third individual diagnostic models are further summed and averaged, and the average linear predicted values are calculated, whereby the third individual diagnostic model is added to the integrated first and second individual diagnostic models. If the third individual diagnostic model that is additionally integrated and has not been further integrated remains, the process of further integration is repeated.
Diabetes screening system. The method of claim 11,
The model building unit,
Two or more types of bio-signals are selected from the memory unit recording the multiple bio-signals, and the selected two or more bio-signals are analyzed according to an ensemble technique among a logistic regression model or machine learning algorithm to which a penalty point function is applied, To build each diagnostic model
Diabetes screening system.

Images