TWI896995B - Method and system for artificial intelligence-enabled electrocardiogram screens low left ventricular ejection fraction with degree of confidence - Google Patents

Abstract

The present invention relates to a method and system for assisted diagnosis of left ventricular insufficiency using an artificial intelligence confidence model using an electrocardiogram. In addition to predicting the left ventricular ejection fraction by extracting high-level features from the electrocardiogram, it also provides variance prediction as a confidence index. The confidence level is used to measure the quality and prediction effect of the ECG. Therefore, the artificial intelligence ECG combined with the confidence level prediction index can be used as a more accurate auxiliary diagnostic tool, which is helpful to provide clinical prediction performance and decision-making reference.

Description

Method and system for diagnosing left ventricular dysfunction using electrocardiogram and artificial intelligence confidence model

This invention belongs to the field of assisted diagnosis technology. Specifically, it relates to a method and system for using electrocardiograms (ECGs) to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence (AI) confidence model. This method not only predicts left ventricular dysfunction using an ECG-based deep learning model, but also uses the standard deviation of the model output as a confidence indicator to improve the accuracy of predicting left ventricular dysfunction.

Approximately tens of millions of people worldwide suffer from left ventricular dysfunction (LVD), which can lead to heart failure in severe cases, impacting quality of life and prognosis. Clinically, left ventricular ejection fraction (LVF), cerebral palindromic acid (CPAP), and N-terminal pro-cerebral palindromic acid (NPAP) are important markers for evaluating heart failure, and treatment strategies vary depending on the patient's clinical parameters. Currently, left ventricular ejection fraction (EF) is the primary classification standard, and multiple EF tests are used for patients who respond to clinical treatment. LVF values are obtained and recorded using transthoracic echocardiograms, using either dynamic mode (M-Mode) or biplane Simpson's method, depending on clinical needs. BNP (brain-forming peptide) and N-terminal pro-brain-forming peptide (NT-pro-BNP) are currently the most commonly used diagnostic tools to complement cardiac ultrasound, for example, in monitoring treatment in adults with heart disease. However, previous studies have shown that the accuracy (AUC) of BNP and NT-pro-BNP is approximately 0.6-0.8, and is easily affected by gender, age, and medical history. Therefore, the development of more effective biomarkers for early cardiac detection is clinically needed.

ECG is currently the main tool for detecting heart function and electrical activity because it is fast, cheap, and suitable for community screening by the public. With the improvement of deep learning models (DLMs), ECG assisted by artificial intelligence (AI) has become a mature technology and is widely used for asymptomatic LVD screening. In addition, compared with true-negative patients, false-positive patients with normal EF values identified by AI ECG have a higher probability of developing severe LVD, demonstrating the potential of AI ECG to identify potential LVD patients. These features have increased the acceptance of AI-ECG in clinical practice. A randomized controlled trial showed that AI-ECG can identify an additional 43% of potential LVD patients. However, most AI-ECGs only provide a possibility test for LVD but do not assess its severity.

Furthermore, there are several electrocardiographic (ECG) features considered to be associated with typical LVD, including heart rate, left fascicular block (LBBB), and prolonged QRS duration. Furthermore, atrial fibrillation may cause acute cardiac dysfunction in some patients with low EF. Although patients with low EF may exhibit atrial fibrillation on the ECG, most patients with atrial fibrillation have normal EF, leading to greater uncertainty in predicting EF in patients with atrial fibrillation. This uncertainty in medical data complicates the diagnostic decision-making process. For these reasons, while artificial intelligence (AI) ECG models can be used as a tool for assessing left ventricular ejection fraction (EF) using echocardiography, they are unable to assess left ventricular function due to insufficient accuracy.

Furthermore, wearable physiological monitoring devices are transforming the healthcare industry, enabling people to monitor their physiological status and activities anytime, anywhere. They also offer advantages such as remote medical care and continuous monitoring, leading to their widespread use in various fields. Currently, wearable physiological monitoring devices can also acquire electrocardiograms (ECGs). If combined with ECG-assisted diagnostic tools, they can track patients' ECG signals over long periods of time. This can, for example, accurately detect left ventricular dysfunction in advance, effectively monitoring and preventing heart failure.

In other words, to better quantify the uncertainty of model predictions, developing a system and method using deep learning models to help physicians identify ECG changes associated with left ventricular dysfunction for early, objective, and accurate diagnosis will effectively serve as evidence for rapid treatment. Furthermore, developing an auxiliary diagnostic tool that can be applied to wearable physiological monitoring devices is a key issue in the industry.

In light of this, the present invention is based on an in-depth exploration of the aforementioned need for diagnosing left ventricular dysfunction. Drawing on the inventors' years of experience in related development, they have actively sought a solution. Through continuous research and development, they have ultimately successfully developed a method and system for using electrocardiograms (ECGs) to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model. This method addresses the inconvenience and difficulties caused by the inability of existing diagnostic aids to effectively assess left ventricular function due to insufficient accuracy.

Therefore, the primary objective of the present invention is to provide a method and system for using electrocardiograms (ECGs) to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model. This system and method utilizes a patient's ECG signals and, through an artificial intelligence deep learning model, helps physicians identify ECG changes associated with left ventricular dysfunction, enabling early, objective, and accurate diagnosis of heart failure.

Secondly, another major objective of the present invention is to provide a method and system for using electrocardiograms to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model. This method can rapidly assess left ventricular function to predict the probability of heart failure, allowing medical personnel to monitor and intervene in real time.

Furthermore, another primary objective of the present invention is to provide a method and system for using electrocardiograms to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model. This method can provide remote, long-term tracking to predict the probability of heart failure, thereby further protecting the patient's life safety.

To this end, the present invention primarily achieves the above-mentioned objectives and effects through the following technical means, which include:

a step of establishing at least one ejection fraction model data: establishing at least one left ventricular ejection fraction model, wherein the model is trained using at least one electrocardiogram signal and at least one left ventricular ejection fraction value obtained from cardiac ultrasound corresponding to the electrocardiogram signal;

1. Obtaining an electrocardiogram to be measured: Obtaining an electrocardiogram to be measured of a monitored person;

A step of determining a confidence ejection fraction: comprising an input prediction module, a weighted average module, and a sum output module using a deep neural network architecture, wherein the input prediction module inputs the electrocardiogram to be tested as a sequence vector and generates an input prediction value, and then outputs a weighted value through the weighted average module, and the output of the weighted average module of at least one lead will pass through a fully connected layer together, and set the output to two hidden layers, and then perform a weighted average of the weight and the individual prediction results of the original input prediction module, wherein one hidden layer is Sigmoid. The probability vector is output and limited to point estimates between 10 and 90. Another hidden layer directly outputs the variance and then performs an exponential transformation to form a confidence index. This is then projected onto the probability density function of the normal distribution for model optimization. The transformation is performed using u as the actual left ventricular ejection fraction, y as the point estimate, and σ as the square root of the variance, as shown in Formula (1).

……………………Formula (1)

A step of converting the probability of heart failure: using the cumulative distribution function of the normal distribution to convert the point estimate and the variance into the probability of heart failure. The cumulative distribution function mainly describes the probability of the variable X being less than or equal to x under a given distribution. In the present invention, u is the point estimate, and x is set to a left ventricular ejection fraction less than or equal to 40 as the diagnostic standard for heart failure. If f(x) is a continuous function, then its derivative is used as the cumulative probability density function to obtain a predicted value of the probability of heart failure, as shown in formulas (2) and (3);

……………………Formula (2)

………………………………………….Formula (3)

A step of displaying the left ventricular function test results: after obtaining the electrocardiogram, the measured left ventricular ejection fraction, and the heart failure prediction probability value of the monitored person, transmitting and displaying these data to at least one monitor.

The method is performed using the following system, which comprises at least one left ventricular function measuring device;

The left ventricular function measurement device includes a processing unit and at least one memory unit and at least one storage unit respectively connected to the processing unit, wherein the storage units have at least one model data module and a measurement operation module. The model data module contains at least one ejection fraction model data, the ejection fraction model data including at least one electrocardiogram signal and at least one left ventricular ejection fraction value of a cardiac ultrasound wave corresponding to the electrocardiogram signal. The measurement operation module uses the method of using an electrocardiogram to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model as described in any one of claims 1 to 4 to convert a test electrocardiogram into a predicted heart failure probability value.

The left ventricular function measuring device can be connected to at least one electrocardiogram generating device and at least one monitoring and displaying device, so that the electrocardiogram to be measured obtained by the electrocardiogram generating device can be converted into a corresponding predicted heart failure probability value and displayed on the monitoring and displaying device.

Through the specific implementation of the aforementioned technical means, the present invention can significantly enhance its practicality, increase its added value, and improve its economic benefits.

To help you better understand the structure, features, and other purposes of this invention, several preferred embodiments of this invention are listed below, along with detailed descriptions of the drawings, so that those skilled in the art can implement them.

This invention relates to a method and system for using an electrocardiogram (ECG) to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model. The accompanying figures illustrate specific embodiments of the invention and its components. References to anterior and posterior, left and right, top and bottom, upper and lower, and horizontal and vertical are for descriptive purposes only and are not intended to limit the invention or its components to any particular position or spatial orientation. Dimensions specified in the drawings and description may vary according to the design and requirements of the specific embodiments of the invention without departing from the scope of the claims.

The present invention is a method for using an electrocardiogram to assist in the diagnosis of left ventricular dysfunction with an artificial intelligence confidence model, which can be used to quickly determine the state of left ventricular dysfunction to predict the probability of heart failure. As shown in the first and second figures, the method is used in a system for using an electrocardiogram to assist in the diagnosis of left ventricular dysfunction with an artificial intelligence confidence model, the system comprising at least one electrocardiogram generating device (10), a left ventricular function measuring device (20) and At least one monitoring and display device (30), wherein the electrocardiogram generating devices (10), the left ventricular function measuring device (20) and the monitoring and display devices (30) can be an integrated structure, a combined structure or a separate structure, and if they are separate structures, they can be connected to each other using wired technology (such as Ethernet) or wireless technology (such as wireless medical system, Wi-Fi or mobile communication above 3G) for mutual data transmission.

As shown in the second figure, the implementation steps of the method include:

A step of establishing at least one ejection fraction model data (S01): first, training at least one left ventricular ejection fraction model (Left Ventricular Ejection Fraction Model, also known as EFMD), wherein the model training is performed using at least one electrocardiogram signal (Reference ECG, also known as R.ECG) (such as 12-lead electrocardiogram) obtained from the electrocardiogram generating device (10) or the electrocardiogram database of the medical institution and at least one left ventricular ejection fraction value (Reference Left Ventricular Ejection Fraction Model) of the cardiac ultrasound wave corresponding to the electrocardiogram signal. Fraction, also known as R.EF), in which the 12-lead electrocardiogram is converted into "time-series" data and recorded in the form of an electric potential signal in each small period of time, such as recording one signal every 2 milliseconds and continuously recording for 10 seconds. The left ventricular ejection fraction value is obtained and recorded by transthoracic echocardiogram, using dynamic mode (Motion Mode, M-Mode) or two-dimensional Simpson's method according to clinical needs, and stored in the aforementioned left ventricular function measurement device (20);

A step of obtaining at least one electrocardiogram to be measured (S02): obtaining the electrocardiogram to be measured of a monitored person using an electrocardiogram generating device (10);

A step of determining the ejection fraction with confidence (S03): In terms of the model, the present invention uses a deep neural network measurement operation prediction model (also known as ECG12Net) that can be set in the left ventricular function measurement device (20) to establish a prediction model. The ECG12Net is disclosed in the invention patent application No. 109108808 of the Republic of China, as shown in the third figure, wherein the ECG12Net includes three important structures, namely, an introduction prediction module (ECG lead block), a weighted average module (Attention block) and a sum output module (Sum Output block). After passing at least one lead (e.g., 12 leads, Lead I, Lead II, ..., Lead V6) through the ECG lead block, each of the 12 leads will have a feature vector of length N and a prediction result of length 1. This feature vector is then passed through the attention block for weighted prediction. The attention block structure is: fully connected layer → batch normalization layer → linear rectifier layer → fully connected layer → batch normalization layer. Ultimately, each attention block outputs a single value. The outputs of the 12-lead "Attention block" are passed through a fully connected layer and set as the output of two hidden layers. The weights are then weighted averaged with the individual prediction results of the original "ECG lead block". One hidden layer outputs a probability vector with a Sigmoid function and limits it to a point estimate between 10 and 90. The other hidden layer directly outputs the variance and then performs an exponential conversion to form a confidence index. The confidence index is then projected onto the probability density function of the normal distribution for model optimization, with u as the actual left ventricular ejection fraction, y as the point estimate, and σ as the square root of the variance, as shown in Formula (1).

…………………………Formula (1)

A step of converting the probability of heart failure (S04): Then, the electrocardiogram to be measured of the monitored person is introduced into the left ventricular function measurement device (20) for conversion, and the prediction operation mode of the "ECG lead block", "Attention block" and "Sum Output block" of ECG12Net in the left ventricular function measurement device (20) is used. The electrocardiogram to be measured is input as a feature vector by the method of importing prediction. After the determination of left ventricular ejection fraction (DEF) with a confidence model is obtained by formula (1), the present invention uses the cumulative distribution function of the normal distribution to convert the point estimate and the variance into the probability of heart failure. The cumulative distribution function mainly describes the probability that the variable X is less than or equal to x under a given distribution. In the present invention, u is a point estimate, and x is set to a left ventricular ejection fraction less than or equal to 40 as the diagnostic standard for heart failure. If f(x) is a continuous function, then its derivative is used as the cumulative probability density function to obtain a predicted heart failure probability value (PHF), as shown in Formula (2) and Formula (3).

……………………Formula (2)

…………………………………………Formula (3)

A step of displaying the left ventricular function measurement results (S05): after obtaining the electrocardiogram, the measured left ventricular ejection fraction and the heart failure prediction probability value of the monitored person, the data are transmitted and displayed on the monitoring display device (30) of at least one monitor.

The method for using an electrocardiogram to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model further includes a model learning step (S06), which utilizes a convolutional neural network to identify eigenvalues in a specified electrocardiogram in an unsupervised manner for learning. The algorithm used by the convolutional neural network of the model learning module during the learning process can be a known method, and the network parameters are adjusted to generate new ejection fraction model data for diagnosis.

The detailed structure of the system of the present invention is as disclosed in the first and fourth figures, wherein the electrocardiogram generating devices (10) have at least one electrode unit (11) for measuring the electrocardiogram signal of at least one monitored person. The electrocardiogram generating devices (10) can select electrode units (11) of 1 to 12 leads to generate an electrocardiogram (D1) to be measured with a corresponding number of leads. For example, when a standard 12-lead electrocardiogram is used as an example, the electrocardiogram generating device (10) needs to place 10 electrode units (11) such as RA, LA, RL, LL, V1, V2, V3, V4, V5 and V6, which can generate a standard 12-lead [Lead I, Lead II, ..., Lead V6], and the electrocardiogram generating device (10) has a transmission unit (12) electrically connected to the electrode unit (11), and the transmission unit (12) can transmit the electrocardiogram (D1) to be measured generated by the electrode unit (11) to the above-mentioned left ventricular function measuring device (20) using wired technology or wireless technology. According to some embodiments, the electrocardiogram generating device (10) can be a wearable physiological monitoring device (50) or a part thereof, which is directly worn by the patient to obtain at least one electrocardiogram (D1) to be measured, and is used for long-term random monitoring of the monitored person (such as patients in ambulances, patients with chronic diseases, etc.);

As shown in FIG5, the left ventricular function measuring device (20) can execute the method of using electrocardiogram to assist in diagnosing left ventricular dysfunction with an artificial intelligence confidence model, and the left ventricular function measuring device (20) can include a processing unit (21), a transmission unit (22), at least one memory unit (23) and at least one storage unit (24), wherein the processing unit (21) is used to execute various programs, instructions and functions of the system, and the transmission unit (22) can be wired technology. Or wireless technology allows the left ventricular function measuring device (20) and the aforementioned electrocardiogram generating device (10) and/or monitoring display device (30) to be interconnected to transmit various data, images or instructions, and the memory units (23) are electrically connected to the processing unit (21), and the memory units (23) are used to store system programs or instructions, and as temporary data storage media for operating systems or other programs being executed. The memory unit (23) can be a read-only memory unit [Read Only Memory, ROM] and/or Random Access Memory, RAM, and each of the storage units (24) can be connected to the processing unit (21) by wire or wirelessly, and each of the storage units (24) can be an internal storage device or an external storage device, such as a hard disk drive (HDD), a solid state disk (SSD) or a cloud hard disk (Online Hard Disk). Drive], and each of the storage units (24) has at least a built-in model data module (25) and a measurement operation module (26), so that after the left ventricular function measurement device (20) obtains the electrocardiogram to be measured, a heart failure prediction probability value corresponding to the electrocardiogram to be measured (D1) can be obtained through the calculation and measurement of the model data module (25) and the measurement operation module (26). According to some embodiments, the left ventricular function measurement device (20) and the electrocardiogram generation devices (10) can be combined to form an integrated wearable physiological monitoring device (50). Furthermore, according to some embodiments, the left ventricular function measurement device (20) can be a cloud server device, an information workstation, a personal computer or a portable mobile device, etc. According to some embodiments, the model determination device (20) may further include a graphics processing unit (27) connected to the processing unit (21) for performing graphics operations through analysis, deep learning, and machine learning algorithms to improve the operation speed and the accuracy of electrocardiogram interpretation. According to some embodiments, the model determination device (20) further includes a model learning module (28) with a convolutional neural network architecture for generating new ventricular ejection fraction model data for diagnosis. The model learning module (28) uses a convolutional neural network to learn by identifying the eigenvalues in a specified electrocardiogram in an unsupervised manner, and the algorithm used by the convolutional neural network of the model learning module (28) during learning processing can be a well-known method, and the network parameters (weight coefficients, bias, etc.) are adjusted. Moreover, the ventricular ejection fraction model data (structural data and learned weight parameters, etc.) formed by the model learning module (28) are stored in the storage unit (24) together with the model data module (25) or the measurement operation module (26), and the model learning module (28) can use a well-known backpropagation method (Backpropagation) to implement learning processing in a method of driving deep learning model training.

As for the monitoring and display device (30), it has a transmission unit (31), and the transmission unit (31) can use wired technology or wireless technology to receive the detection value and/or electrocardiogram signal transmitted by the above-mentioned vector conversion device (20), and the monitoring and display device (30) has a display unit (32), and the display unit (32) can be used to display the left ventricular ejection fraction, the heart failure prediction probability value and/or the electrocardiogram signal (D1). The electrocardiogram generated by the electrocardiogram is provided for medical personnel to interpret and predict the probability of heart failure. According to some embodiments, the monitoring and display device (30) further has an alarm sending unit (35). The alarm sending unit (35) can send the left ventricular ejection fraction, heart failure prediction probability value and/or electrocardiogram generated by the electrocardiogram signal (D1) that exceeds the preset value to emergency personnel, responsible doctors or remote monitoring devices for medical personnel to monitor and intervene in real time.

In summary, it can be understood that the present invention is a highly creative creation that not only effectively solves problems faced by practitioners but also significantly enhances efficacy. Furthermore, no identical or similar products have been created or publicly used in the same technical field, and this invention also exhibits enhanced efficacy. Therefore, the present invention meets the requirements of "novelty" and "advancedness" for invention patents, and therefore, an invention patent application is filed in accordance with the law.

10: ECG generation device 11: Electrode unit 12: Transmission unit 15: ECG database 20: Left ventricular function measurement device 21: Processing unit 22: Transmission unit 23: Memory unit 24: Storage unit 25: Model data module 26: Measurement calculation module 27: Graphics processing unit 28: Model learning module 30: Monitoring and display device 31: Transmission unit 32: Display unit 35: Alarm transmission unit 50: Wearable physiological monitoring device S01: Create at least one ejection fraction model data set S02: Obtain an ECG to be measured S03: Determine confidence in ejection fraction S04: Conversion probability to heart failure S05: Display left ventricular function assessment results S06: Model learning

Figure 1 is a schematic diagram illustrating the operation of the system of the present invention that uses an electrocardiogram (ECG) and an artificial intelligence confidence model to assist in the diagnosis of left ventricular dysfunction.

Figure 2 is a schematic diagram of the process architecture of the method of the present invention for using an electrocardiogram to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model.

Figure 3 is a schematic diagram of the model architecture for left ventricular function measurement in the method of the present invention.

Figure 4 is a schematic diagram of the system architecture of the present invention.

Figure 5 is a schematic diagram of the structure of the left ventricular function measurement device in the system of the present invention.

Figure 6 is a diagram illustrating data conversion during actual application of the system of the present invention.

S01: Create at least one ejection fraction model data set S02: Obtain a test electrocardiogram S03: Determine the ejection fraction with confidence S04: Conversion probability of heart failure S05: Display left ventricular function test results S06: Model learning

Claims (10)

A method for using electrocardiogram to assist in the diagnosis of left ventricular dysfunction with an artificial intelligence confidence model comprises: a step of establishing at least one ejection fraction model data: establishing at least one left ventricular ejection fraction model, wherein the model training is performed using at least one electrocardiogram signal and at least one left ventricular ejection fraction value of a cardiac ultrasound wave corresponding to the electrocardiogram signal; a step of obtaining an electrocardiogram to be measured: obtaining an electrocardiogram to be measured of a monitored person; a step of determining an ejection fraction with confidence: comprising using a deep neural network to determine the ejection fraction with confidence. The neural network architecture is composed of an input prediction module, a weighted average module and a sum output module. The input prediction module inputs the electrocardiogram to be tested as a sequence vector and generates an input prediction value. The weighted average module then outputs a weighted value. The output of the weighted average module of at least one leader will pass through a fully connected layer together and set the output to 2 hidden layers. The weight is then weighted averaged with the individual prediction results of the original input prediction module, one of which is a Sigmoid layer. The probability vector is output and limited to point estimates between 10 and 90. Another hidden layer directly outputs the variance and then performs an exponential transformation to form a confidence index. This is then projected onto the probability density function of the normal distribution for model optimization. The transformation is performed using u as the actual left ventricular ejection fraction, y as the point estimate, and σ as the square root of the variance, as shown in Formula (1). Formula (1) - Step of converting the probability of heart failure: using the cumulative distribution function of the normal distribution to convert the point estimate and the variance into the probability of heart failure. The cumulative distribution function mainly describes the probability that the variable X is less than or equal to x under a given distribution. In the present invention, u is the point estimate, and x is set to the left ventricular ejection fraction less than or equal to 40 as the diagnostic standard for heart failure. If f(x) is a continuous function, then its derivative is used as the cumulative probability density function to obtain a predicted value of the probability of heart failure, as shown in Formula (2) and Formula (3); ……………………Formula (2) Formula (3) - Step of displaying the left ventricular function measurement results: After obtaining the electrocardiogram, the measured left ventricular ejection fraction and the predicted probability value of heart failure of the monitored person, such data are transmitted and displayed to at least one monitor. As described in item 1 of the patent application, a method for using an electrocardiogram to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model is provided, wherein the left ventricular ejection fraction value in the step of establishing at least one ejection fraction model data is obtained and recorded by chest ultrasound examination using a dynamic mode or a two-dimensional Simpson two-section scan according to clinical needs. As described in Item 1 of the patent application, a method for using an electrocardiogram to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model is provided, wherein the electrocardiogram in the step of establishing at least one ejection fraction model data can be generated by electrocardiogram signals of different lead numbers, and corresponding ejection fraction model data of various different lead numbers are established. As described in item 1 of the patent application, a method for using an electrocardiogram to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model, wherein the method further includes a model learning step, which utilizes a convolutional neural network to identify feature values in a specified electrocardiogram in an unsupervised manner for learning, and the algorithm used by the convolutional neural network of the model learning module during learning processing can be a publicly known method, and the network parameters are adjusted to generate new ejection fraction model data for diagnosis. A system for using electrocardiograms to assist in the diagnosis of left ventricular dysfunction using an artificial intelligence confidence model, the system comprising at least one left ventricular function measurement device; the left ventricular function measurement device comprising a processing unit and at least one memory unit and at least one storage unit respectively connected to the processing unit, wherein the storage units comprise at least one model data module and a measurement operation module, the model data module comprising at least one ejection fraction model data, the ejection fraction model data comprising at least one electrocardiogram signal and at least one corresponding to the electrocardiogram signal The left ventricular ejection fraction value of the cardiac ultrasound is measured, and the measurement operation module uses the method of applying electrocardiogram to assist in the diagnosis of left ventricular dysfunction with an artificial intelligence confidence model as described in any one of items 1 to 4 of the patent application to convert a test electrocardiogram into a predicted heart failure probability value; the left ventricular function measurement device can be connected to at least one electrocardiogram generation device and at least one monitoring and display device, so that it can convert the test electrocardiogram obtained by the electrocardiogram generation device into a corresponding predicted heart failure probability value and display it on the monitoring and display device. In the system described in claim 5, the left ventricular function measuring device may further include a graphics processing unit connected to the processing unit to increase the calculation speed. As described in item 5 of the patent application, the left ventricular function measurement device further includes a model learning module, and the model learning module uses a convolutional neural network to identify feature values in a specified electrocardiogram in an unsupervised manner for learning. The algorithm used by the convolutional neural network of the model learning module during learning processing can be a well-known method, and the network parameters are adjusted to generate new ejection fraction model data for diagnosis. As described in item 5 of the patent application scope, the electrocardiogram generating device, the left ventricular function measuring device and the monitoring and display device can be composed of a wearable physiological monitoring device. As described in item 5 of the patent application, the processing unit of the left ventricular function measurement device can be connected to a transmission unit, and the electrocardiogram generation device and the monitoring and display device each have a transmission unit that can be connected to each other for transmission, so that the left ventricular function measurement device can connect to the electrocardiogram generation device and the monitoring and display device located at a remote location to transmit data. As described in item 5 of the patent application scope, the monitoring and display device has a display unit and an alarm sending unit, the display unit can be used to display an electrocardiogram for determining myocardial infarction and/or electrocardiographic signal generation, and the alarm sending unit can be used to notify medical personnel for real-time monitoring and intervention.