Method, system and storage medium for calculating cardiac output based on impedance cardiomyogram

Abstract

The present invention relates to the field of medical technology, and in particular to a method, system and storage medium for calculating cardiac output based on a cardiac impedance blood flow diagram, wherein the method comprises selecting a neural network model to construct a corresponding relationship between cardiac output and ICG characteristic parameters, wherein the structure of the neural network model is selected as a one-dimensional convolutional neural network, specifically comprising an input layer, three hidden layers and an output layer, wherein the input layer is an ICG characteristic parameter and a deviation, each hidden layer has 64 nodes, and the activation function between the hidden layers uses a linear rectifier function, and the output layer is a predicted cardiac output value; the present invention has higher accuracy than the existing electric velocity method model, and shows the superiority of the neural network model in predicting cardiac output, which is of great significance for monitoring heart disease. At the same time, the portable measurement method, which is simple, easy to use and inexpensive, will be widely used in clinical, exercise physiology and rehabilitation research and other fields.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a method, system and storage medium for calculating cardiac output based on a cardiac impedance blood flow diagram. Background Art

[0002] The blood of the human body circulates periodically under the action of the heart, forming systemic circulation and pulmonary circulation. In the process of human blood circulation, the systemic circulation starts from the left ventricle to the aorta, then flows from the aorta into arteries at all levels, and finally to the capillaries throughout the body, exchanging gases and substances, turning arterial blood into venous blood. The pulmonary circulation starts from the right ventricle, and venous blood enters the capillary network through the pulmonary artery, and turns into arterial blood after inhaling oxygen in the alveoli.

[0003] Cardiac output refers to the total amount of blood pumped out of the ventricles per minute. The average person's cardiac output ranges from 4-6L / min. The main factors that affect cardiac output are: contractility, heart rate, and the heart's preload and afterload. Heart rate refers to the heart rate. The faster the heart rate, the more blood is pumped in a specific period of time, and the greater the cardiac output. Contractility is equivalent to the heart muscle's ability to contract. As contractility increases, cardiac output will also increase. Preload is the degree of expansion of the myocardium before contraction. Preload depends largely on ventricular filling. When ventricular pressure is high, the work done to eject the same volume of blood is relatively less. Therefore, the greater the preload, the greater the cardiac output. Afterload is the resistance that the ventricles overcome in order to eject blood. It depends largely on arterial blood pressure and vascular tension.

[0004] In the prior art, the cardiac output measurement model based on ICG signals has evolved from the initial chest cylinder model to the chest truncated cone model and the later electric velocity method model. As the complexity of the model increases, the accuracy of the model also increases. However, so far, the correlation between the cardiac output measured by the ICG method and the measurement results of standard instruments is not high. This is mainly because the relationship between the ICG signal and the actual cardiac output is complex and is affected by many factors. It is difficult for a simple formula to accurately reflect the relationship between the two. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method, system and storage medium for calculating cardiac output based on a cardiac impedance blood flow diagram, so as to achieve accurate calculation of cardiac output.

[0006] In order to achieve the above-mentioned purpose, the first aspect of the present invention provides a method for calculating cardiac output based on cardiac impedance blood flow diagram, the method comprising selecting a neural network model to construct a corresponding relationship between cardiac output and ICG characteristic parameters, the structure of the neural network model is selected as a one-dimensional convolutional neural network, specifically comprising an input layer, three hidden layers and an output layer, wherein the input layer is the ICG characteristic parameters and deviations, each hidden layer has 64 nodes, and the activation function between the hidden layers uses a linear rectification function, and the output layer is the predicted cardiac output value.

[0007] Optionally, the ICG characteristic parameters include impedance differential maximum, left ventricular ejection time, heart rate, basal impedance, body surface area, mean pressure, systolic pressure and diastolic pressure.

[0008] Optionally, the ICG characteristic parameters also include (dz / dt) max ×LVET,

[0009] Where, (dz / dt) max is the maximum value of impedance differential, and LVET is the left ventricular ejection time.

[0010] Optionally, in the construction of the neural network model, ICG characteristic parameters are obtained as a data set through a monitoring device;

[0011] Then, 85% of the data was randomly selected from the dataset as the training set, and the remaining 15% was used as the test set. The algorithm was implemented using Keras, with TensorFlow as the backend. The model cost function was set to root mean square, the learning rate was 0.01, and the stochastic gradient descent was iterated with 2000 iterations.

[0012] On the other hand, the present invention also provides a system for calculating cardiac output based on an impedance cardiomyogram, wherein the system comprises a processor, and the processor is used to execute the above method.

[0013] In yet another aspect, the present invention further provides a storage medium storing instructions for being read by a machine so as to enable the machine to execute the above method.

[0014] Compared with the prior art, the method for calculating cardiac output based on the impedance cardiomyogram provided by the present invention has higher accuracy than the existing electric velocity method model, showing the superiority of the neural network model in predicting cardiac output, which is of great significance for the monitoring of heart disease. At the same time, the simple, easy-to-use and low-cost portable measurement method will be widely used in the fields of clinical, exercise physiology and rehabilitation research. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1A neural network model structure diagram provided according to a specific embodiment of the present invention is shown;

[0016] Figure 2 The neural network model measurement result in the embodiment of the present invention;

[0017] Figure 3 A Bland-Altman diagram of the predicted results and actual results of the neural network model in an embodiment of the present invention;

[0018] Figure 4 is the measurement result of the traditional electrical speed measurement model;

[0019] Figure 5 This is the Bland-Altman diagram of the measurement results of the traditional electrical speed measurement model and the actual measurement values. DETAILED DESCRIPTION

[0020] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific embodiments.

[0021] Based on the existing ICG signal measurement model for cardiac output, the main characteristics that affect cardiac output are: (dz / dt) max (maximum value of impedance differential), LVET (left ventricular ejection time), HR (heart rate), Z0 (basal impedance), BSA (body surface area), MABP (mean blood pressure), SABP (systolic blood pressure), DABP (diastolic blood pressure), etc.

[0022] In the above characteristics, except (dz / dt) max , and other characteristic parameters have specific physiological significance. Therefore, the inventors of this application have studied (dz / dt) max The physiological significance of

[0023] According to the formula for calculating stroke volume by Kubieck, Sramek, etc., (dz / dt) is usually max The unit of Ω·s -1 , indirectly used to represent blood flow rate, through strict biophysical inspection and waveform analysis, it was found that there are also some problems. According to the relationship formula between resistance and resistivity:

[0024]

[0025] Where A is the cross-sectional area, L is the length, ρ b is the resistivity. When measuring the stroke volume of a specific subject, L remains unchanged. Considering that the thoracic impedance changes during the cardiac cycle due to the resistivity ρ b and blood volume V b The changes can be obtained,

[0026]

[0027] In equation (2) is the average ohm acceleration (unit: Ω·s -2 ), is the ohmic mean flow rate (unit: Ω·s -1 ). b (t) / dtThe rate of change of blood flow impedance, dV b (t) / dt is the rate of change of volume, i.e. the flow rate.

[0028] Although the above formula describes the composition of dz / dt, it does not explain (dz / dt) max The source of, Bernstein and Lemmens et al. found (dz / dt) max The peak value is exactly in the time domain of the peak acceleration of aortic blood flow, not at the peak velocity or flow rate. In addition, it is clear that (dz / dt) max Maximum change of aortic blood pressure (dP / dt) max are highly consistent in time, which in turn accelerates aortic blood flow to peak, so Bernstein et al. suggested that (dz / dt) max is an estimate of Ohm's acceleration, which gives the following formula:

[0029]

[0030] From the maximum value of (dz / dt), equation (2) is indeed The maximum value is obtained when the ohm mean velocity is 0. In order to make the units consistent, the ohm mean acceleration is reduced to the ohm mean velocity. When calculating the stroke volume SV, the ohm mean velocity is defined as: That is, the square root of the ratio of the maximum value of the impedance differential to the basic impedance (in S -1 ),

[0031]

[0032] Discussion completed (dz / dt) max The present invention uses a neural network to predict the complex relationship between ICG signals and actual cardiac output. Feature selection mainly depends on commonly used feature parameters. (dz / dt) is selected. max (impedance differential maximum), LVET (left ventricular ejection time), HR (heart rate), Z0 (basal impedance), BSA (body surface area), MABP (mean blood pressure), SABP (systolic blood pressure), DABP (diastolic blood pressure), etc., taking into account the above (dz / dt) maxPhysiological significance of parameter analysis, increase (dz / dt) max ×LVET, And other features.

[0033] Model design, considering that cardiac output is related to many ICG characteristic parameters, and the specific relationship is difficult to express with a simple formula, for this reason, the present invention adopts a neural network model to construct the corresponding relationship between cardiac output and ICG characteristic parameters. The specific neural network model structure selects a one-dimensional convolutional neural network,

[0034] like Figure 1 As shown, CO_true is the true cardiac output, Output Layer is the output layer, HiddenLayer is the hidden layer, and Input Layer is the input layer.

[0035] The model has 5 layers, input layer, hidden layer (3 layers) and output layer. The input layer is the above ICG feature parameters and deviations; the hidden layer has 3 layers, each with 64 nodes, and the activation function between the hidden layers uses a linear rectifier function. The final output layer is the predicted cardiac output CO value.

[0036] Dataset, the present invention uses the cardiac output value measured by the standard instrument Physioflow's Enduro device as the label value. A total of 467 subjects were measured in the experiment, aged between 18 and 65 years old, including 252 males and 215 females. Each subject measured 8-15 sets of data at rest, and the abnormal measurement results (CO less than 2L / min or greater than 10L / min) of subjects with loose electrodes or severe heart failure were removed, and a total of 5685 sets of data were obtained.

[0037] The test was conducted using TL-NiCON-100 equipment to measure (dz / dt) max (impedance differential maximum), LVET (left ventricular ejection time), HR (heart rate), Z0 (basal impedance) and other ICG characteristic parameters, and the height, weight, systolic blood pressure and diastolic blood pressure of the subjects were recorded before measurement. The experiment was approved by the Ethics Committee of Anhui Medical University (ethics approval number: PJ2021-006-01).

[0038] Inclusion criteria: (1) Aged 18 years or older, regardless of gender; (2) Subjects need to undergo cardiac output testing or voluntarily participate in cardiac output testing; (3) Subjects or their guardians can understand the purpose of the study, show sufficient compliance with the study protocol, and sign the informed consent. Exclusion criteria: Patients with chest trauma or chest surgery; patients with skin allergies to electrodes; patients with pacemakers; patients with extracorporeal circulation, etc.

[0039] For training, 85% (4832 groups) of data were randomly selected as the training set, and the remaining 15% (853 groups) were used as the test set. The algorithm was implemented using Keras, with TensorFlow as the backend. The model cost function was set to RMS, the learning rate was 0.01, and the stochastic gradient descent iteration was used with 2000 iterations.

[0040] The prediction results of the test set are obtained through training, such as Figure 2 As shown in the figure, in the 853 test sets, the predicted results were compared with the actual standard instrument measurement results, and the correlation reached 0.90, P<0.001, so the predicted results were highly correlated with the actual results. Figure 3 The Bland-Altman graph of the predicted results and the actual results is shown in the figure. The mean square error is 0.57, and only 44 sets of data are outside the mean±1.96δ range, accounting for about 5.16%. In addition, the neural network model evaluation parameter R2-score also reached 0.81.

[0041] In order to compare the effectiveness of the test results, this section also uses the electric speed method model to verify the test set, such as Figure 4 As shown in the figure, the electric velocity method model has a certain correlation with the actual measured value, and the correlation coefficient is 0.81. Figure 5 The Bland-Altman plot of the results measured by the electric velocity method model and the actual measured values ​​shows a mean square error of 0.79, and about 76 sets of data are outside the mean ± 1.96δ range, accounting for about 8.9%. Therefore, compared with the electric velocity method model, the neural network model has more accurate measurement results.

[0042] The above shows and describes the basic principles, main features and characteristics of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of the present invention to be protected is defined by the attached claims and their equivalents.

Claims

1. A method for calculating cardiac output based on cardiac impedance blood flow diagram, characterized in that: The method includes selecting a neural network model to construct a corresponding relationship between cardiac output and ICG characteristic parameters, wherein the structure of the neural network model is selected as a one-dimensional convolutional neural network, specifically including an input layer, three hidden layers and an output layer, wherein the input layer is the ICG characteristic parameters and deviations, each hidden layer has 64 nodes, and the activation function between the hidden layers uses a linear rectification function, and the output layer is the predicted cardiac output value; The ICG characteristic parameters include impedance differential maximum, left ventricular ejection time, heart rate, basal impedance, body surface area, mean pressure, systolic pressure and diastolic pressure; The ICG characteristic parameters also include (dz / dt) max ×LVET, Where, (dz / dt) max is the maximum value of impedance differential, LVET is the left ventricular ejection time; In the construction of the neural network model, ICG characteristic parameters are obtained as data sets through monitoring equipment; Then, 85% of the data was randomly selected from the dataset as the training set, and the remaining 15% was used as the test set. The algorithm was implemented using Keras, with TensorFlow as the backend. The model cost function was set to root mean square, the learning rate was 0.01, and the stochastic gradient descent was iterated with 2000 iterations.

2. A system for calculating cardiac output based on cardiac impedance blood flow diagram, characterized in that: The system comprises a processor configured to execute the method of claim 1 .

3. A storage medium, characterized in that: The storage medium stores instructions for being read by a machine to enable the machine to execute the method according to claim 1 .

Images