Blood pressure prediction system

Abstract

The invention relates to a non-invasive blood pressure prediction system (10) from ECG signals with machine learning. As a novelty, it comprises an input unit (14) which ensures receipt of instant ECG signals from the user, and a processor unit (12) which transfers the results received from a memory unit (11) storing a machine learning model that calculates the predicted blood pressure value from the received ECG signal to a user interface, and the said machine learning model is configured to extract the attributes of the signal through complexity analysis from the real ECG signals entered into the memory unit (11) and instant ECG signals, to process the extracted attributes with a stacking regression model, and to use the real blood pressure values recorded in response to the said real ECG signals when calculating the estimated blood pressure value corresponding to the instant ECG signal.

Description

[0001] BLOOD PRESSURE PREDICTION SYSTEM

[0002] TECHNICAL FIELD

[0003] The invention relates to a non-invasive blood pressure prediction system from ECG signals with machine learning.

[0004] PRIOR ART

[0005] Almost one in three adults has high blood pressure (hypertension), but most patients are unaware of this condition. High blood pressure is a very serious disease that often causes no warning signs but can lead to life-threatening conditions such as heart attack or stroke. Although it is not considered as dangerous as high blood pressure, low blood pressure (hypotension) is a condition that requires attention as it causes problems such as dizziness, blackout, weakness and fainting. Early diagnosis of blood pressure diseases and changes in lifestyle as a result of diagnosis can prevent blood pressure from harming human health. For the early diagnosis of important and preventable blood pressure diseases, it is useful to constantly monitor people and evaluate the detected blood pressure results together with other biosignals such as electrocardiogram (ECG) to minimize blood pressure-related damages. Biosignals such as pulse, blood pressure, blood oxygen saturation are vital signals and their constant monitoring can ensure elimination of many life-threatening problems. The vast majority of preventable deaths are related to heart attacks and strokes. Increased blood pressure (BP) is a leading cause of cardiovascular disease and premature death worldwide. However, most of methods available for monitoring blood pressure require specialized equipment and lack convenience of continuous monitoring.

[0006] The publication numbered EP2992820A2 discloses a blood pressure measurement method. The said method is essentially based on the use of PPG (plurality of photoplethysmogram). In conclusion, all the problems mentioned above have entailed an innovation in the respective technical field.

[0007] SUMMARY OF THE INVENTION

[0008] The present invention relates to a blood pressure prediction system to eliminate the abovementioned disadvantages and bring new advantages to the respective technical field.

[0009] The object of the invention is to present a non-invasive blood pressure prediction system from ECG signals with machine learning.

[0010] In order to realize all the objects which are mentioned above and which will be described in detail hereinafter, the present invention relates to a non-invasive blood pressure prediction system from ECG signals with machine learning. Accordingly; it comprises an input unit which ensures receipt of instant ECG signals from the user, and a processor unit which transfers the results received from a memory unit storing a machine learning model that calculates the predicted blood pressure value from the received ECG signal to a user interface, and the said machine learning model is configured to extract the attributes of the signal through complexity analysis from the real ECG signals entered into the memory unit and instant ECG signals, to process the extracted attributes with a stacking regression model, and to use the real blood pressure values recorded in response to the said real ECG signals when calculating the estimated blood pressure value corresponding to the instant ECG signal.

[0011] In a possible embodiment of the invention, the machine learning model is configured to include at least two of the K-Nearest Neighbor, Support Vector, Ridge, Decision Tree Regression, AdaBoost and XGoost regressions in the first stage of the stacking regression model. In a possible embodiment of the invention, the machine learning model is configured to include a final regression model that takes as input the predicted values obtained in the first stage of the stacking regression model.

[0012] In a possible embodiment of the invention, in the machine learning model, a final regression model that takes as input the predicted values obtained in the first stage of the stacking regression model is the lasso regression model.

[0013] In a possible embodiment of the invention, the machine learning model is configured to extract at least one of the signal mobility, signal complexity, signal entropy, autocorrelation coefficient and fractal dimension attributes through complexity analysis.

[0014] BRIEF DESCRIPTION OF THE DRAWINGS

[0015] A representative schematic view of the blood pressure prediction system of the invention is given in Figure 1 .

[0016] DETAILED DESCRIPTION OF THE INVENTION

[0017] This detailed description the blood pressure prediction system of the invention by examples that will not constitute any limiting effect and are merely for better understanding of the subject.

[0018] Referring to Figure 1 , the blood pressure estimation system of the invention essentially comprises an input unit which ensures receipt of the instant ECG (electrocardiogram) signals from the user, a memory unit where various data and a machine learning model are stored, a processor unit which processes the data, and a user interface to which the predicted blood pressure value is transferred.

[0019] The said memory unit includes real ECG signals previously taken with professional medical devices and real blood pressure values corresponding to these data. Attributes of both real ECG signals and instant ECG signals are extracted through complexity analysis in the said machine learning model. The identified attributes aim to obtain information about heart functions by focusing on the abstract features of the signal such as regularity, irregularity and chaoticness. Signal mobility, signal complexity, signal entropy, autocorrelation coefficient and fractal dimension attributes are created through complexity analysis.

[0020] The identified features are used as input to the stacking regression model. In the stacking regression model, the data are first processed in different regression models to obtain intermediate results. Afterwards, the said intermediate results are processed in a final regression model to obtain the final prediction At least two of the K-Nearest Neighbor, Support Vector, Ridge, Decision Tree Regression, AdaBoost and XGoost regressions are the prediction models used in the first stage of the stacking model. Lasso Regression model is used as the regression model in which intermediate results are processed.

[0021] Thanks to using the attributes of ECG signals as the main input in the machine learning model of the blood pressure prediction system of the invention and making the prediction with a stacking regression model, it is ensured that much more accurate predictions are made compared to the previous technique.

[0022] The protection scope of the invention is specified in the appended claims, and it may in no way be limited to the disclosures made for illustrative purposes in this detailed description. Because, it is evident that a person skilled in the art can present similar embodiments in the light of the foregoing without deviating from the main theme of the invention.

Claims

CLAIMS1. A non-invasive blood pressure prediction system (10) from ECG signals with5 machine learning, characterized in that it comprises an input unit (14) which ensures receipt of instant ECG signals from the user, and a processor unit (12) which transfers the results received from a memory unit (11 ) storing a machine learning model that calculates the predicted blood pressure value from the received ECG signal to a user10 interface, and the mentioned machine learning model is configured to extract the attributes of the signal through complexity analysis from the real ECG signals entered into the memory unit (11 ) and instant ECG signals, to process the extracted attributes with a stacking regression model, and to15 use the real blood pressure values recorded in response to the said real ECG signals when calculating the estimated blood pressure value corresponding to the instant ECG signal.

2. A blood pressure prediction system according to Claim 1 , characterized in20 that the machine learning model is configured to include at least two of the K-Nearest Neighbor, Support Vector, Ridge, Decision Tree Regression, AdaBoost and XGoost regressions in the first stage of the stacking regression model.25 3. A blood pressure prediction system according to Claim 2, characterized in that the machine learning model is configured to include a final regression model that takes as input the predicted values obtained in the first stage of the stacking regression model.

4. A blood pressure prediction system according to Claim 3, characterized in that the30 machine learning model, a final regression model that takes as input the predicted values obtained in the first stage of the stacking regression model, is the lasso regression model.

5. A blood pressure prediction system according to Claim 1 , the machine learning model is configured to extract at least one of the signal mobility, signal complexity, signal entropy,35 autocorrelation coefficient and fractal dimension attributes through complexity analysis.6

Images