KR102098561B1 - Method and program for stroke volume acquisition using arterial pressure waveform - Google Patents

Abstract

The present invention provides a method and a program for obtaining a cardiac output using an arterial pressure waveform. The method for acquiring a cardiac output using the arterial pressure waveform includes: a computer acquiring patient's cardiac output data for a predetermined period of time; the computer acquiring a patient's arterial pressure data including one or more heartbeat cycles; and the computer Generating a data set by mapping the cardiac output measurement data and the arterial pressure wave data, the computer performing learning by a learning model using the data set as input data, and the computer using the result of the learning And obtaining a cardiac output value by regression analysis, wherein the learning model includes a plurality of calculation steps, sequentially performs each calculation step, and performs subsampling of previous step data values ( subsampling) value is sent to the next step without going through each operation step. Characterized in that, the subsampling value is at least one of an average value, a maximum value, a minimum value, and a derivative value of the previous step data value.

Description

METHOD AND PROGRAM FOR STROKE VOLUME ACQUISITION USING ARTERIAL PRESSURE WAVEFORM

The present invention relates to a method for acquiring cardiac output using an arterial pressure waveform and a program thereof.

Cardiac output refers to the amount of blood your heart beats and squeezes. It is commonly called stroke volume (SV) or cardio output (CO). Stroke volume is the amount of blood squeezed out of a heartbeat, and is known to be about 70 ml in adult men. The cardio output is the amount of blood squeezed for 1 minute, which is the amount of strobulium multiplied by the heart rate per minute (Hear Rate). It is known as 2L.

Cardiac output is the most important and basic indicator for determining if a patient's circulatory function is operating normally, especially during surgery, but is difficult to measure directly.

Thermal dilution method or indicator dilution method for measuring cardiac output is not easy to measure, and it is not easy to measure, such as intubation into the left ventricle through the carotid artery, right atrium, right ventricle, or pulmonary vein using a catheter to measure cardiac output. Not provided.

In order to measure continuous cardiac output by thermal dilution, a method has been developed to apply a constant heat energy by using a hot wire connected to blood and measure a change in temperature, but side effects such as abnormally increasing the patient's body temperature appear.

The blood flow measuring device using an ultrasonic sensor must be directly attached to a blood vessel and attached to the sensor through surgery, so it is difficult to be frequently used to measure the patient's cardiac output, and the ultrasound imaging device can measure the temporary cardiac output by an expert. It is difficult to measure the patient's disease continuously for a long time.

Korean Registered Patent Publication No. 10-0821409, 2008.04.10.

In order to directly measure cardiac output, a pulmonary artery catheter must be mounted on the heart, but unless the patient has a large burden, it is monitored by measuring blood pressure in real time.

However, it is difficult to accurately measure cardiac output using blood pressure using a method known to date.

Therefore, the problem to be solved by the present invention is to provide a method for accurately obtaining a cardiac output through learning using an arterial pressure wave.

In addition, the problem to be solved by the present invention is to provide a method for accurately obtaining the cardiac output even by different arterial pressure wave data for each patient.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the following description.

In order to solve the above-described problem, a method for acquiring cardiac output using an arterial pressure waveform according to an embodiment of the present invention comprises: a computer acquiring data on a patient's cardiac output for a predetermined period of time; Obtaining a patient's arterial pressure wave data including a cycle, the computer mapping the cardiac output measurement data and the arterial pressure wave data to generate a data set, and the computer using the data set as input data by a learning model A step of performing learning and the computer using the result of the learning to obtain a cardiac output value by regression analysis (regression analysis), the learning model includes a plurality of calculation steps, each operation The steps are sequentially performed, and the subsampling value of the previous step data value is It is characterized in that it is sent to the next step without going through each operation step, and the subsampling value is at least one of an average value, a maximum value, a minimum value, and a derivative value of the previous step data value.

The learning model is a convolutional neural network model, and the plurality of computational steps include a convolution layer and a maxpooling layer.

When the subsampling value is a differential value, the differential value includes a differential value at all points of the arterial pressure wave.

The method for acquiring cardiac output using an arterial pressure waveform according to an embodiment of the present invention for solving the above-described problem is obtained before the computer acquires the cardiac output value after the step of performing the learning, the computer is the result of the learning Comprising the step of calculating a constant value on the relationship formula using the cardiac output measurement data and the arterial pressure wave data as a variable, and the computer further comprises the step of performing learning by the learning model including the constant value , The result of the learning in the step of obtaining the cardiac output value is a result of learning by the learning model including the constant value.

In the method for acquiring cardiac output using an arterial pressure waveform according to an embodiment of the present invention for solving the above-described problem, the cardiac output measurement data and the arterial pressure wave data are respectively obtained for a plurality of patients, and the learning model comprises a plurality of It is generated for each patient, generated as a separate model, and before the step of calculating the constant value, the computer further comprises the step of evaluating the quality of the generated learning model for each of the plurality of patients.

The step of acquiring the cardiac output value is characterized by acquiring a cardiac output value by regression analysis using a result of learning by the patient's learning model that has obtained a quality equal to or greater than a predetermined value by the evaluating step. do.

The evaluating step is characterized in that the quality is calculated by using a result of applying a learning model of one patient to a patient other than the one of the plurality of patients.

A cardiac output acquisition program using arterial pressure waveforms according to another embodiment of the present invention for solving the above-described problems is combined with a computer that is hardware, and is stored in a medium to execute the above method.

A cardiac output acquisition server device using an arterial pressure waveform according to another embodiment of the present invention for solving the above-described problem, a cardiac output measurement data acquisition unit for acquiring cardiac output measurement data for a predetermined period of time, at least one An arterial pressure wave data acquisition unit for acquiring arterial pressure wave data of a patient including a heartbeat cycle, a data set generation unit mapping the cardiac output measurement data and the arterial pressure wave data to generate a data set, and inputting the data set And a cardiac output value acquiring unit for performing a learning by a low learning model and obtaining a cardiac output value by regression analysis using the results of the learning, wherein the learning model comprises a plurality of calculation steps. Including, sequentially performing each operation step, and subsampling the previous step data values ng) is characterized by sending the value to the next step without going through each operation step, and the subsampling value is at least one of an average value, a maximum value, a minimum value, and a derivative value of the previous step data value.

Other specific matters of the present invention are included in the detailed description and drawings.

According to the present invention, by using a learning model generated using the patient's cardiac output measurement data and arterial pressure wave data, the cardiac output can be obtained through the arterial pressure waveform.

According to the present invention, by applying not only the average value but also the maximum, minimum, and differential values to the learning model, it is possible to obtain cardiac output values of various patients by the learning model of the present invention even though the arterial pressure wave data varies from patient to patient. have.

According to the present invention, by using the learning model generated by learning the constant value on the relational expression using the cardiac output measurement data and arterial pressure wave data as variables, a more accurate cardiac output can be obtained.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a view for explaining a method of obtaining a cardiac output using an arterial pressure waveform according to an embodiment of the present invention.
2 is a view for explaining a learning model using cardiac output measurement data and arterial pressure wave data.
FIG. 3 is a diagram for explaining that learning is performed by calculating a constant value on a relational expression using cardiac output measurement data and arterial pressure wave data as variables.
4 is a diagram for explaining a method of generating a final learning model by evaluating the quality of the learning model for each of a plurality of patients.
5 is a graph showing a quality index of training data used in a learning model.
6 and 7 are diagrams showing the results of applying the analysis with values predicted through the final prediction model.

Advantages and features of the present invention, and methods for achieving them will be clarified with reference to embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms, only the present embodiments allow the disclosure of the present invention to be complete, and are common in the technical field to which the present invention pertains. It is provided to fully inform the skilled person of the scope of the present invention, and the present invention is only defined by the scope of the claims.

The terminology used herein is for describing the embodiments and is not intended to limit the present invention. In this specification, the singular form also includes the plural form unless otherwise specified in the phrase. As used herein, “comprises” and / or “comprising” does not exclude the presence or addition of one or more other components other than the components mentioned. Throughout the specification, the same reference numerals refer to the same components, and “and / or” includes each and every combination of one or more of the mentioned components. Although "first", "second", etc. are used to describe various components, it goes without saying that these components are not limited by these terms. These terms are only used to distinguish one component from another component. Therefore, it goes without saying that the first component mentioned below may be the second component within the technical spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used in the present specification may be used as meanings commonly understood by those skilled in the art to which the present invention pertains. In addition, terms defined in the commonly used dictionary are not ideally or excessively interpreted unless explicitly defined.

In the present specification, 'computer' includes all of various devices capable of performing arithmetic processing and providing a result to a user. For example, the computer is not only a desktop PC, a notebook, but also a smart phone, a tablet PC, a cellular phone, and a personal communication service phone (PCS phone), synchronous / asynchronous. A mobile terminal of an International Mobile Telecommunication-2000 (IMT-2000), a Palm Personal Computer (PC), and a Personal Digital Assistant (PDA) may also be applicable. Further, the computer may also be a medical computer, a medical PC, a medical tablet, a medical device, and a server that receives a request from a client and performs information processing.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a view for explaining a method of obtaining a cardiac output using an arterial pressure waveform according to an embodiment of the present invention.

Referring to FIG. 1, a method for acquiring cardiac output using an arterial pressure waveform includes: a computer acquiring patient cardiac output measurement data for a predetermined period of time (S100); Acquiring step (S200), mapping the cardiac output measurement data and arterial pressure wave data to generate a data set (S300), performing a learning by a learning model using the data set as input data (S400) and the result of the learning And obtaining a cardiac output value by regression analysis using (S500).

The step of acquiring cardiac output measurement data (S100) and the step of acquiring arterial pressure wave data (S200) may be performed simultaneously, the cardiac output measurement data may be obtained first, and then the arterial pressure wave data may be obtained, and the arterial pressure wave data may be obtained. The heart rate measurement data may be obtained after data is first acquired.

In step S100 of the computer acquiring the patient's cardiac output data for a predetermined period of time, the cardiac output data may be obtained using a pulmonary artery catheter or the Doppler effect.

Acquiring cardiac output measurement data using a pulmonary artery catheter is data obtained by directly inserting a catheter into the heart, and is used for learning as reference data together with corresponding arterial pressure wave data. In addition, acquiring cardiac output measurement data using a pulmonary artery catheter is measured for a predetermined period of time, and for example, it can be measured for 10 minutes as a minimum unit in consideration of a time delay of vigilance.

Acquisition of cardiac output data using the Doppler effect can be obtained by real-time measuring the velocity of blood flowing through the aorta using the Doppler effect, then integrating over time and multiplying the aortic cross-sectional area.

In the step of obtaining arterial pressure wave data of a patient including one or more heartbeat cycles (S200), the arterial pressure wave data includes all arterial pressure data, for example, femoral arterial pressure data or radial arterial pressure data. In addition, the arterial pressure wave data is obtained by including one or more heartbeat cycles, for example, the arterial pressure wave data measured for 10.24 seconds, which may completely cover at least one breathing cycle, is briefly 2 seconds, which is one heart rate. Arterial pressure wave data measured for less than is included.

The step (S300) of mapping the cardiac output measurement data and the arterial pressure wave data to generate a data set is to generate learning training data for performing learning by mapping the cardiac output measurement data and the arterial pressure wave data corresponding to each other.

When the cardiac output measurement data is obtained using the pulmonary artery catheter, an accurate cardiac output value can be obtained, and by training and training by mapping the cardiac output measurement data and corresponding arterial pressure wave data, when using a trained learning model, the arterial pressure is later Accurate cardiac output values can be obtained even by obtaining only wave data.

In step S400 of performing learning by a learning model using a data set as input data, the learning model includes a plurality of calculation steps, sequentially performs each calculation step, and subsampling the previous step data value. It is characterized in that the value is sent to the next step without going through each of the above steps.

In addition, various machine learning algorithms can be applied to the learning model, and for example, as a deep learning model is applied, it can be built into a deep neural network.

A deep neural network (DNN) according to embodiments of the present invention refers to a system or a network in which one or more layers are built in one or more computers to perform judgment based on a plurality of data. For example, the deep neural network may be implemented as a set of layers including a convolutional pooling layer, a locally-connected layer, and a fully-connected layer. The convolutional pooling layer or the local access layer can be configured to extract features in the image. The fully connected layer can determine a correlation between features of an image. In some embodiments, the overall structure of the deep neural network may be formed in such a way that a local connection layer continues to the convolutional pooling layer and a full connection layer goes to the local connection layer. The deep neural network may include various criteria (ie, parameters), and may add new criteria (ie, parameters) through input image analysis.

The deep neural network according to embodiments of the present invention has a structure called a convolutional neural network, and extracts a feature extraction layer for self-learning features having the greatest discriminative power from given image data. Based on the features, a prediction layer that learns a prediction model to produce the highest prediction performance may be configured as an integrated structure.

The feature extraction layer spatially integrates a convolution layer that creates a feature map by applying a plurality of filters for each region of the image and a feature map spatially, so that the feature that does not change in position or rotation changes. It can be formed into a structure that repeats the multiple layers alternately several times to allow extraction. Through this, various levels of features can be extracted from low-level features such as points, lines, and faces to complex and meaningful high-level features.

The convolution layer obtains a feature map by taking a non-linear activation function in the filter and the dot product of the local receptive field for each patch of the input image, compared to other network structures. Therefore, CNN has a feature of using a filter having sparse connectivity and shared weights. Such a connection structure reduces the number of parameters to be learned and efficiently improves prediction performance by efficiently learning through a backpropagation algorithm.

The integrated layer (Pooling Layer or Sub-sampling Layer) creates a new feature map by utilizing the region information of the feature map obtained from the previous convolution layer. In general, the feature map newly created by the integrated layer is reduced to a smaller size than the original feature map. As a typical method of integration, the maximum pooling that selects the maximum value of the area in the feature map and the corresponding feature within the feature map And Average Pooling, which calculates the average value of a region. In general, the feature map of the integrated layer may be less affected by the position of any structure or pattern existing in the input image than the feature map of the previous layer. That is, the integrated layer can extract features that are more robust to regional changes such as noise or distortion in an input image or a previous feature map, and these features can play an important role in classification performance. Another integrated layer's role is to enable the deeper structure to reflect the characteristics of a wider area as it goes up to the upper learning layer. As the feature extraction layer is accumulated, the lower layer reflects regional features and goes up to the upper layer. More and more features can be generated that reflect the characteristics of a more abstract overall image.

As such, the final feature extracted through the iteration of the convolutional layer and the integration layer is that a classification model such as a multi-layer perception (MLP) or a support vector machine (SVM) is a fully connected layer. -connected layer) to be used for classification model learning and prediction.

However, the structure of the deep neural network according to the embodiments of the present invention is not limited to this, and may be formed of neural networks having various structures.

In one embodiment, the learning model is a convolutional neural network model, and the plurality of computational steps include a convolution layer and a maxpooling layer, and a description of the learning model using the convolutional neural network model is illustrated. It will be described later in the description of 2.

The shape of the arterial pressure wave may be the same for each patient, but since the data values may be different, accuracy may be deteriorated by learning data sets of two or more patients as input data. Therefore, the input data for learning uses one patient's data set.

In the step (S500) of acquiring a cardiac output value by regression analysis using a result of learning, regression analysis is an analysis method for estimating a relationship between a dependent variable and two or more independent variables by a least square method, the present invention Essay uses regression analysis to estimate a single number rather than picking one of the answers through a learning model.

2 is a view for explaining a learning model using cardiac output measurement data and arterial pressure wave data.

The input data 10 of the learning model is a data set of mapped cardiac output data and arterial pressure wave data described above in the description of FIG. 1. As shown in FIG. 2, input data 10 of the learning model is input, and is output to the fully connected rater 50 through the convolution layers 31 to 36 and the maxpooling layers 41 to 46.

The input data 10 passes through the convolution layer 31 and the max spooling layer 41, and the sub-sampling value 21 does not pass through the convolution layer 31 and the max spooling layer 41, and the input data 10 Send the value as it is to the next step. In addition, among the values that have passed through the convolution layer 31 and the maxpooling layer 41, this value is sent as the sub-sampling value 22 as it is to the next step. Repeat this step.

The sub-sampling values 21 to 26 are at least one of an average value, a maximum value, a minimum value, and a derivative value of the previous step data value. In addition to the average value, the performance of the learning model can be further improved by using the maximum, minimum, and differential values.

Preferably, the sub-sampling values 21-26 are derivatives of previous step data values. The differential value includes the differential value at all points of the arterial pressure wave. Since data of the arterial pressure wave is different for each patient, by learning using the shape of the arterial pressure wave, it can be reflected in the learning model as a common point for each patient. Therefore, the differential value is used as a sub-sampling value to learn using the shape of the arterial pressure wave.

FIG. 3 is a diagram for explaining that learning is performed by calculating a constant value on a relational expression using cardiac output measurement data and arterial pressure wave data as variables.

Referring to FIG. 3, a method for acquiring a cardiac output using an arterial pressure waveform of the present invention includes a cardiac output measurement data and a computer using a result of the learning before the computer acquires a cardiac output after the step of performing learning. The method further includes calculating a constant value (S430) on a relational expression using arterial pressure wave data as a variable (S460), and the computer performing learning by a learning model including the constant value. In addition, in the step of acquiring the cardiac output value, the cardiac output value is obtained by regression analysis using the result of learning by the learning model including the constant value (S502).

Calculating a constant value on a relational expression in which the cardiac output measurement data and arterial pressure wave data are variables, and performing learning including the constant value is a characteristic of each patient, that is, a relationship between the cardiac output measurement data and the arterial pressure wave data. Since the characteristics are different, the scale is also different for each patient, so in order to generate a learning model that fits a different scale for each patient, a constant value is calculated on the relational expression, and the calculated constant value is included. By performing learning, different scales can be adjusted for each patient.

If only one patient's data is used, in the following Equation 1, w _i = 1, b _i = 0, or the step of calculating a constant value may be omitted.

The relational expressions using the cardiac output measurement data and the arterial pressure wave data as variables are as shown in Equation 1 below.

(Equation 1)

SV _scaled = SV _CNN * w _i + b _i

SV _CNN is the SV value calculated by the CNN model, w _i and b _i are the individual scale coefficients, i is the individual patient, and SV _scaled is the individual scale coefficient before applying the individual scale coefficient. Corrected value.

4 is a diagram for explaining a method of generating a final learning model by evaluating the quality of the learning model for each of a plurality of patients.

Referring to FIG. 4, a method for acquiring cardiac output using an arterial pressure waveform of the present invention comprises: a step in which a computer acquires cardiac output measurement data of a plurality of patients for a predetermined time (S102), including one or more heartbeat cycles Acquiring the arterial pressure wave data of a plurality of patients, respectively (S202), mapping the cardiac output measurement data and the arterial pressure wave data for each patient and generating a data set (302), each data set for each patient as input data Performing learning by the learning model (S402), evaluating the quality of each generated learning model for each of the plurality of patients (S410), using the results of the training, cardiac output measurement data and arterial pressure wave data as variables Calculating a constant value on the relational expression (S430), performing a learning by a learning model including the constant value (S460) and includes a constant value Using a result of learning by the learning model includes the step (S502) for obtaining cardiac output value by regression analysis.

The step of acquiring cardiac output measurement data (S102) and the step of acquiring arterial pressure wave data (S202) may be performed simultaneously, the cardiac output measurement data may be obtained first, and then the arterial pressure wave data may be obtained, and the arterial pressure wave data may be obtained. The heart rate measurement data may be obtained after data is first acquired.

As described above in the description of FIG. 1, the shape of the arterial pressure wave may be the same for each patient, but since data values may be different, accuracy may be deteriorated when learning a data set of two or more patients as input data. . Therefore, the input data for learning uses one patient's data set.

Even if a plurality of patient data sets are trained as input data, each data set for each patient is input data and learning is performed.

In the step of evaluating the quality of the learning model generated for each of the plurality of patients (S410), the quality is calculated by using a result of applying a learning model of one patient to the remaining patients excluding the one of the plurality of patients will be.

In generating a learning model, since an accurate learning model can be generated only when a learning model is generated using high-quality data, learning is performed using only a data set having a quality higher than a certain standard using a step of evaluating the quality of the data. Create the model.

In one embodiment, evaluating the quality of the learning model measures how well it fits by applying the learning model of patient A to the remaining 24 patients excluding patient A using data of patient A among 25 patients. By doing so, the quality of patient A's learning model can be evaluated.

The evaluation criterion of the quality of the learning model is, in one embodiment, determining the prediction accuracy of the learning model, and evaluating using the Pearson correlation coefficient of the predicted value, for example, learning in which the Pearson correlation coefficient is 0.45 or more. Can be based on a model.

Evaluate the quality of the learning model, and calculate the constant value on the relational expression using the cardiac output measurement data and the arterial pressure wave data as variables using the results of learning using only the data set of the learning model that has obtained a quality above a certain standard as input data. (S430). Learning by the learning model including the calculated constant value is performed (S460), and a cardiac output value is obtained by regression analysis using the results of learning by the learning model including the constant value (S502).

In order to confirm the effect of the cardiac output using the arterial pressure waveform of the present invention, the cardiac output data value measured using a pulmonary artery catheter (PAC) was limited to 2 seconds in Edward vigilance. As measured data, an experiment was conducted with data from 58 patients who had undergone liver transplant surgery at Seoul Asan Hospital between February 1, 2018 and April 4, 2018.

Of the 58 data, 27 data from February 1, 2018 to March 10, 2018 were used to build the model through training, and the remaining 31 between March 11, 2018 and April 4, 2018 The dog was actually used to apply the model and compare the results.

5 is a graph showing a quality index of training data used in a learning model.

Referring to the graph of FIG. 5, as a model was created through training, the vertical axis value is a data quality index (DQI), and 27 CNN prediction models are generated using 27 data one by one, and then the model is generated. Is applied to the remaining 26 data, respectively.

The data quality index was defined as the Pearson Correlation (PCR) of the predicted and actual values of the model, and the data quality was defined as the average value of the PCR of 26 prediction results. As shown in Fig. 5, the average of the actual PCRs can be found to be distributed between 0.21 and 0.43.

In FIG. 5, a final prediction model is generated based on the top 7 data having excellent data quality.

In the analysis, high-resolution analysis and low-resolution analysis were divided into two types. In the high-resolution analysis, the original data measured every 2 seconds were used as it was, and in the low-resolution analysis, sampling was performed once every 10 minutes.

Table 1 below shows the results of analyzing the correlations estimated from cardiac output measurement data values and arterial pressure wave data values. Table 1 below is a result of comparing the present invention and other products (EV1000). Pearson correlation coefficients and intraclass correlation coefficients (ICCs) were used as indicators for correlation.

Method of analysis Measure The present invention Other products High resolution analysis Pearson's correlation coefficient 0.55 0.44 Intra-day correlation coefficient 0.56 0.44 Low resolution analysis Pearson's correlation coefficient 0.54 0.44 Intra-day correlation coefficient 0.44 0.56

Referring to Table 1 above, in the high resolution analysis, the Pearson correlation coefficient and the intra-interval correlation coefficient were superior to those of other companies' products. In the low-resolution analysis, the present invention was based on the Pearson correlation coefficient, and the intra-interval correlation coefficient was Third-party products showed excellent results.

6 and 7 are diagrams showing the results of applying the analysis with values predicted through the final prediction model.

6 is a result of applying the blend-altman plot analysis. Referring to FIG. 6, in high-resolution data, the third-party product (a in FIG. 6) is 14.77, and the present invention (b in FIG. 6) is 13.63 with a deviation of the third-party product. (A in FIG. 6) showed an error rate of 5.01%, and the present invention (b in FIG. 6) was 4.46%. In the low-resolution data, each of the other products (c in FIG. 6) was 16.08, and the present invention (d in FIG. 6). The deviation of 15.15 and the other products (c in FIG. 6) showed 4.55%, and the present invention (d in FIG. 6) showed an error rate of 4.03%.

Therefore, it can be confirmed that the present invention is superior to other companies' products because both the deviation and the error rate of the present invention are low.

7 is a result of applying the quadrant plot analysis. Referring to FIG. 7, in high-resolution data, the third-party product (a in FIG. 7) shows an error rate of 25.03% and the present invention (b in FIG. 7) is 24.0%. It can be seen that the invention is superior to other products.

Cardiac output acquisition server device using the arterial pressure waveform according to another embodiment of the present invention, a cardiac output measurement data acquisition unit for acquiring the cardiac output data of a patient for a predetermined period of time, the arterial pressure of a patient including one or more heart rate cycles Arterial pressure wave data acquisition unit that acquires wave data, data set generation unit that maps cardiac output measurement data and arterial pressure wave data to generate a data set, and performs training by a learning model using the data set as input data, and results of learning It includes a cardiac output value acquisition unit for acquiring the cardiac output value by regression analysis.

To obtain the cardiac output measurement data, the cardiac output measurement data may be obtained using a pulmonary artery catheter, or the cardiac output measurement data may be obtained using a Doppler effect.

Acquiring cardiac output measurement data using a pulmonary artery catheter is data obtained by directly inserting a catheter into the heart, and is used for learning as reference data together with corresponding arterial pressure wave data. In addition, acquiring cardiac output measurement data using a pulmonary artery catheter is measured for a predetermined period of time, and for example, it can be measured for 10 minutes as a minimum unit in consideration of a time delay of vigilance.

Acquisition of cardiac output data using the Doppler effect can be obtained by real-time measuring the velocity of blood flowing through the aorta using the Doppler effect, then integrating over time and multiplying the aortic cross-sectional area.

The arterial pressure wave data includes all arterial pressure data, for example, femoral arterial pressure data or radial arterial pressure data. In addition, the arterial pressure wave data is obtained by including one or more heartbeat cycles, for example, the arterial pressure wave data measured for 10.24 seconds, which may completely cover at least one breathing cycle, is briefly 2 seconds, which is one heart rate. Arterial pressure wave data measured for less than is included.

The data set generating unit is to generate learning training data for performing learning by mapping corresponding cardiac output measurement data and arterial pressure wave data.

When the cardiac output measurement data is obtained using the pulmonary artery catheter, an accurate cardiac output value can be obtained, and by training and training by mapping the cardiac output measurement data and the corresponding arterial pressure wave data, when using a trained learning model, the arterial pressure is later Accurate cardiac output values can be obtained even by obtaining only wave data.

The learning model is characterized by including a plurality of operation steps, sequentially performing each operation step, and sending a subsampling value of the previous step data value to the next step without going through each operation step.

The subsampling value is at least one of an average value, a maximum value, a minimum value, and a derivative value of the previous step data value, and is preferably a derivative value of the previous step data value. The differential value includes the differential value at all points of the arterial pressure wave. Since data of the arterial pressure wave is different for each patient, by learning using the shape of the arterial pressure wave, it can be reflected in the learning model as a common point for each patient. Therefore, the differential value is used as a sub-sampling value to learn using the shape of the arterial pressure wave.

1 to 4, the above description is equally applied to a cardiac output server device using an arterial pressure waveform according to another embodiment of the present invention.

The steps of a method or algorithm described in connection with an embodiment of the present invention may be implemented directly in hardware, a software module executed by hardware, or a combination thereof. The software modules may include random access memory (RAM), read only memory (ROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, hard disk, removable disk, CD-ROM, or It may reside on any type of computer readable recording medium well known in the art.

The embodiments of the present invention have been described above with reference to the accompanying drawings, but a person skilled in the art to which the present invention pertains may implement the present invention in other specific forms without changing its technical spirit or essential features. You will understand. Therefore, it should be understood that the above-described embodiments are illustrative in all respects and not restrictive.

10: input data
21, 22, 23, 24, 25, 26: sub-sampling value
31, 32, 33, 34, 35, 36: Convolution Layer
41, 42, 43, 44, 45, 46: Maxpooling layer
50: fully connected layer

Claims (9)

The computer acquiring the patient's cardiac output measurement data for a predetermined period of time;
The computer obtaining arterial pressure wave data of a patient including one or more heartbeat cycles;
The computer mapping the cardiac output measurement data and the arterial pressure wave data to generate a data set;
The computer performing learning by a learning model using the data set as input data; And
The computer comprises the step of obtaining a cardiac output value by regression analysis using the results of the learning,
The learning model,
It includes a plurality of operation steps, each operation step sequentially, characterized in that the sub-sampling (subsampling) value of the previous step data value is sent to the next step without going through each operation step,
The subsampling value is,
At least one of an average value, a maximum value, a minimum value, and a derivative value of the previous step data value,
A method for acquiring cardiac output using arterial pressure waveforms. According to claim 1,
The learning model is a convolutional neural network model,
The plurality of calculation steps include a convolution layer and a maxpooling layer,
A method for acquiring cardiac output using arterial pressure waveforms. According to claim 1,
When the subsampling value is a derivative value,
The derivative value is,
Characterized in that it comprises a differential value at all points of the arterial pressure wave,
A method for acquiring cardiac output using arterial pressure waveforms. According to claim 1,
After the step of performing the learning, before the step of obtaining the cardiac output value,
Calculating, by the computer, a constant value on a relational expression using the cardiac output measurement data and the arterial pressure wave data as variables using the results of the learning; And
The computer further comprises the step of performing learning by the learning model including the constant value,
The result of the learning in the step of obtaining the cardiac output value,
The result of learning by the learning model including the constant value,
A method for acquiring cardiac output using arterial pressure waveforms. According to claim 4,
The cardiac output measurement data and the arterial pressure wave data are respectively obtained for a plurality of patients,
The learning model,
It is generated for each of a plurality of patients, and is generated as a separate model,
Before the step of calculating the constant value,
The computer further comprising the step of evaluating the quality of the learning model generated for each of the plurality of patients,
A method for acquiring cardiac output using arterial pressure waveforms. The method of claim 5,
The step of obtaining the cardiac output value,
Characterized in that the cardiac output value is obtained by regression analysis using a result of learning by a learning model of a patient who has obtained a quality higher than a predetermined value by the evaluation step,
A method for acquiring cardiac output using arterial pressure waveforms. The method of claim 5,
The evaluation step,
Characterized in that the quality is calculated by using a result of applying a learning model of one patient to a patient other than the one of the plurality of patients,
A method for acquiring cardiac output using arterial pressure waveforms. A program for acquiring cardiac output using arterial pressure waveforms, stored in a medium in combination with a computer, which is hardware, to execute the method of claim 1. A cardiac output measurement data acquisition unit for acquiring cardiac output data for a predetermined period of time;
An arterial pressure wave data acquisition unit for acquiring arterial pressure wave data of a patient including at least one heartbeat cycle;
A data set generator configured to generate a data set by mapping the cardiac output measurement data and the arterial pressure wave data;
And a cardiac output value acquisition unit configured to perform learning by a learning model using the data set as input data, and to obtain a cardiac output value by regression analysis using the results of the learning,
The learning model,
It includes a plurality of operation steps, each operation step sequentially, characterized in that the sub-sampling (subsampling) value of the previous step data value is sent to the next step without going through each operation step,
The subsampling value is,
At least one of an average value, a maximum value, a minimum value, and a derivative value of the previous step data value,
Server device for acquiring cardiac output using arterial pressure waveform.

Images