Computerized implementation method

Abstract

SOLUTION: A computer-implemented method includes: receiving ECG data describing an electrocardiogram (ECG) of a subject in a period by a system formed of one or more computers; processing the ECG data to generate a predictive input; providing the predictive input to a survival estimation model which has been generated by processing multiple pieces of training data for predicting a survival estimation for the subject; and outputting the prediction of the survival estimation for the subject. The survival estimation is a future survival rate of the subject and estimated ejection rate characteristics are mapped to the future survival ratio of the subject.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] This specification describes computer-based techniques for analyzing physiological electrical data (e.g., electrocardiogram data) for purposes such as estimating the ejection fraction characteristics of a patient.

[0002] [Cross-Reference to Related Applications] This application claims priority to U.S. Patent Application No. 62 / 599,163, filed Dec. 15, 2017, and U.S. Patent Application No. 62 / 569,268, filed Oct. 6, 2017. The disclosures of these prior applications are considered part of the disclosure of this application and are hereby incorporated by reference in their entirety.

[0003] [Background] Ejection fraction is a major measure of the health of the hearts of humans and other mammals. Ejection fraction typically indicates the amount of blood ejected from the heart with each pumping. With each pumping cycle (i.e., "heartbeat" or "cardiac cycle"), the myocardium contracts and relaxes to pump blood into the subject's arterial system. When the heart relaxes, the ventricles fill with blood. Then, during the contraction phase of the pumping cycle, a portion of the blood is sent from the ventricles through the aorta into the arterial system. However, not all of the blood that filled the ventricles during relaxation can be pumped out during contraction. The percentage of the blood actually ejected from the ventricles during the pumping cycle is called the ejection fraction. Ejection fraction often relates only to the measurement of the portion of the blood ejected from the left ventricle, but for the purposes of this specification, ejection fraction can alternatively refer to the portion of the blood ejected from only the left ventricle, only the right ventricle, or both ventricles, and can indicate the strength or health of the myocardium.

[0004] Generally, an ejection fraction greater than approximately 50–55 percent is considered normal for humans. However, some people (often unnoticed) have low ejection fractions, for example, below the 50–55 percent range, or very low ejection fractions, for example, below 35 percent. This is called asymptomatic ventricular dysfunction (ASVD). Low or very low ejection fractions are often markers of serious cardiac complications such as cardiac arrest, sudden death, and various stages of heart failure. When ASVD is diagnosed, there are effective treatments for ASVD or for preventing these complications.

[0005] Ejection fraction has traditionally been measured using cardiac ultrasound. During cardiac ultrasound, an ultrasound technician uses an ultrasound transducer and specialized techniques to generate images of the heart. While cardiac ultrasound is non-invasive, it often requires a specialist to perform and interpret the procedure, and typically requires expensive equipment and infrastructure within the patient's healthcare provider's laboratory. Ejection fraction has also been measured using magnetic resonance imaging (MRI), computed tomography (CT), and nuclear medicine scans. All of these require specially trained personnel and expensive equipment.

[0006] [Summary] This specification discloses systems, methods, devices, and other techniques for estimating the cardiac ejection fraction of mammals from an electrocardiogram (ECG). An electrocardiogram is a measurement of the electrical activity of the mammalian heart. The pumping action of the heart is driven by a continuous cycle of electrical polarization and depolarization of the myocardium. This electrical activity can be captured by an electrocardiogram, by which electrodes are placed on the body surface of a subject (e.g., the chest and limbs of the subject) and the potential between each electrode pair is measured over time. The electrical signals captured by this process form an electrocardiogram. When the electrocardiogram is graphed to show the change in potential between electrodes over time, waveforms or ECG traces showing the polarization and depolarization of the heart in each of one or more cardiac cycles can be observed. For the purposes of this specification, the electrocardiogram may include any number of leads, from the conventional 12 leads, additional leads, or as few as a single lead. In addition, ECG can be obtained from adhesive electrodes, conductive electrodes, capacitive electrodes, handheld electrodes, wearable / clothing-type electrodes, subcutaneous electrodes, electrodes attached to implantable devices, or any combination thereof.

[0007] The shape of the ECG waveform is influenced by several factors. The techniques disclosed herein assume, in part, the recognition that the subject's ECG may be affected by the ejection fraction. For example, in subjects with a low or very low ejection fraction compared to another subject with a normal ejection fraction, one or more parts of the ECG waveform may be subtly deformed. Underlying conditions affecting the heart, whether due to atherosclerosis, muscle damage, inflammation, or valvular dysfunction, can impair the myocardium's pumping ability. Underlying conditions also affect the metabolism of individual muscle cells or their interconnections, potentially leading to fibrotic deposition or inflammatory cell infiltration, all of which result in subtle electrical changes. These localized electrical changes in the heart contribute to deformations recorded in surface ECGs. Such deformations may not be visible to the naked eye, but they may nevertheless be detectable using computer-based models developed by the techniques disclosed herein. Therefore, this specification describes how the subject's ECG can be used as a screening tool to predict the subject's ejection fraction characteristics. In many scenarios, ECG-based screening is preferred over measurement of ejection fraction by ultrasound or other means. ECG devices are often more widely available than ultrasound devices and can be performed more quickly without the need for a trained ECG technician. Furthermore, consumer and home-use ECG devices (e.g., single-lead ECG patches) are becoming increasingly common. Using the techniques disclosed herein, it is possible to increase the number of people screened for low ejection fraction and the number of screenings performed by using ECG devices in various settings, whether clinical or home-use. In some cases, if ECG screening of a subject indicates a high probability of low or very low ejection fraction, further evaluation of the condition can be performed to verify the condition by using other measurement means, such as ultrasound, MRI, or CT scans.

[0008] The subject matter disclosed herein includes computer implementation methods. These methods can be implemented by a system comprising one or more computers located in one or more locations. In some embodiments, the system comprises one or more processors and one or more computer-readable media encoded with instructions that, when executed by the one or more processors, cause the processors to implement the method. In some embodiments, the system comprises only computer-readable media encoded with instructions that, when executed, cause the processors to implement the method.

[0009] This method may involve the system receiving electrocardiogram (ECG) data describing the ECGs of mammals during a given period. The system provides a predictive input derived from the ECG data to an ejection fraction prediction model. The ejection fraction prediction model can be used to process the predictive input and generate an estimated ejection fraction characteristic of the mammal. The system then provides the estimated ejection fraction characteristic of the mammal as an output.

[0010] These and other implementations may optionally further include one or more of the following features:

[0011] ECG data can include one or more channels, each channel containing a subset of ECG data describing each lead of a mammalian ECG during a given period. The prediction input can characterize each lead of an ECG for each of one or more channels of the ECG data.

[0012] ECG data can include multiple channels, each channel containing a subset of ECG data describing a specific derivation of a mammalian ECG during a given period. The prediction input can characterize multiple ECG derivations for each of the multiple channels of ECG data.

[0013] The period over which ECG data describes a mammalian ECG can span multiple cardiac cycles.

[0014] Mammals can be considered equivalent to humans.

[0015] The ejection fraction prediction model can be a regression model, such as a logistic regression model.

[0016] The ejection fraction prediction model can be a machine learning model such as a neural network (e.g., a feedforward neural network, a convolutional neural network, or a recurrent neural network).

[0017] Ejection fraction models can be further refined by using clinical characteristics (including age, sex, weight, and / or the presence or absence of measurable medical conditions such as hypertension and diabetes) as input.

[0018] Ejection fraction models can also generate additional outputs, such as the risk of death over a given period (for example, the risk of death over the next year).

[0019] The estimated ejection fraction characteristic of a mammal can be a value that represents the absolute (e.g., intrinsic) estimate of the mammal's ejection fraction.

[0020] The estimated elimination rate characteristics of mammals can indicate the estimated range of the elimination rate of mammals.

[0021] Providing estimated ejection fraction characteristics of a mammal as output may include providing estimated ejection fraction characteristics for presentation to the mammal or to healthcare providers associated with that mammal.

[0022] This method may further include generating a predictive input by determining the values ​​for one or more morphological features of a mammalian ECG. The predictive input may represent the values ​​for one or more morphological features of a mammalian ECG.

[0023] The morphological features of the ECG of a mammal can include at least one of T wave amplitude, P wave amplitude, P wave area, T wave area, left inclination of the T wave, right inclination of the T wave, left inclination of the P wave, right inclination of the P wave, T wave duration, P wave duration, PR interval, QRS duration, QRS amplitude, QRS area, QRS energy, QRS peak-to-peak ratio, or QT segment length.

[0024] The ejection fraction prediction model can be personalized according to the mammal.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this invention pertains. Although methods and materials similar or equivalent to those described herein can be used to implement the present invention, suitable methods and materials are described below. All publications, patent applications, patents, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, this specification, including definitions, will control. In addition, the materials, methods, and examples are for illustrative purposes only and not intended to be limiting.

[0026] Details of one or more embodiments of the present invention are described in the accompanying drawings and the following description. Other features, objects, and advantages of the present invention will become apparent from the description, drawings, and claims.

Brief Description of the Drawings

[0027] [Figure 1] It is a conceptual diagram of an exemplary system that records and processes ECG data and uses the ECG data to estimate the ejection fraction characteristics of a subject. [Figure 2] It is a flowchart of an exemplary process for estimating the ejection fraction characteristics of a subject using ECG data. [Figure 3] It is a diagram of an exemplary ECG tracing for one heartbeat of a patient. The ECG tracing shows the constituent segments of the heartbeat and various morphological waveform features. [Figure 4]A flowchart of an exemplary process for selecting and using a suitable ejection fraction prediction model corresponding to the specified characteristics of the subject. [Figure 5] A flowchart of an exemplary process for using an ECG-based ejection fraction estimation as a screening tool to determine whether further evaluation is justified. [Figure 6] A flowchart of an exemplary process for training an ejection fraction prediction model such as a neural network-based model. [Figure 7] A flowchart of an exemplary process for predicting the survival rate of a subject from the subject's ECG data based on the heart condition such as low ejection fraction or ultra-low ejection fraction. [Figure 8] A diagram of an exemplary neural network model for estimating the ejection fraction characteristics of a subject from ECG data. [Figure 9] A diagram showing the results of the first neural network implementation form of an ejection fraction prediction model for classifying patients into groups according to whether the ejection fraction of the patient is predicted to be above 50 percent or below. [Figure 10] A diagram showing the results of the second neural network implementation form of an ejection fraction prediction model for classifying patients into groups according to whether the ejection fraction of the patient is predicted to be above 35 percent or below. [Figure 11] A graph showing the correlation between the model output and the true ejection fraction values in the study. [Figure 12] A block diagram of an exemplary computing device that can be used to implement the systems, methods, devices, and other techniques described herein. [Figure 13] A chart representing a data mining schema for training, validating, and testing a convolutional neural network ejection fraction prediction model. To avoid cross-contamination between datasets and network training based on the ECG of a specific patient, the ECG for each patient was used only once (in only one of the groups). For the follow-up analysis, patients with two or more data points were used. [Figure 14] This figure shows the receiver operating characteristic curves used to identify patients with an EF ≤ 35% using a convolutional neural network. ROC and area under the curve (AUC) were calculated using validation and test (holdout) datasets. Identical AUCs demonstrate the robustness of the algorithm to different datasets. [Figure 15] This pie chart shows the distribution of ejection fractions based on network classification. Among patients classified as having a low ejection fraction, 63.5% had an ejection fraction below 50%. Among those classified as normal, 1.3% had an ejection fraction ≤35%, and 90.1% had an EF ≥50%. [Figure 16] This table shows various patient characteristics and comorbidities for patients included in exemplary studies and evaluations of ejection fraction prediction models. [Figure 17] This graph shows the long-term outcomes for patients with an EF ≥ 50% at the time of initial classification. Patients who initially had a normal EF but were classified as having a low EF by the network had a significantly increased risk of future LV dysfunction compared to patients who had a normal EF and were classified as normal. [Modes for carrying out the invention]

[0028] Similar references and symbols in the figure indicate similar elements.

[0029] This specification discloses systems, methods, devices, and other techniques for estimating the ejection fraction of a mammal based on an electrocardiogram (ECG) of the mammal. In some implementations, a machine learning model, such as a neural network, is configured to process a predictive input characterizing ECG data and output a representation of the estimated ejection fraction of the mammal. This model can be trained to describe complex combinations of features that are not normally recognizable to humans but have been determined (e.g., through an iterative training process) to be correlated with specific ejection fraction characteristics. Additional details regarding these and other techniques are provided in the following description of Figures 1-12.

[0030] Figure 1 is a conceptual diagram of an exemplary system 100 that records and processes ECG data and uses the ECG data to estimate the ejection fraction characteristics of subject 102. For the purposes of this example, subject 102 is considered to be a human, more specifically a patient of a healthcare provider. However, it should be understood that the explanation is not limited to this example. In other implementations, subject 102 could be a human not particularly related to a healthcare provider, or any other mammal, and an appropriate model would be constructed to map the ECG data and estimate the ejection fraction for that mammal.

[0031] A pair of electrodes 104 are placed on the body surface of the patient 102, enabling the recording of signals indicating the electrical activity of the patient's heart during the ECG procedure. In some implementations, 10 leads can be attached to the patient to perform a standard 12-lead ECG recording (for example, some electrodes placed near the cardiac region in the chest and others placed on the patient's limbs). A 12-lead ECG is useful for obtaining multiple channels of ECG data, with the data for each channel representing a respective lead. Each lead is formed by the potential between a pair of electrodes. As a result of changes in electrode positioning, different pairs of electrodes for different leads form different angles, and each lead provides a different picture of the electrical activity of the patient's heart. For example, signals from each lead can be recorded simultaneously over a period (e.g., 5, 10, or 15 seconds) to capture information about the timing and location of electrical activity along different radial directions.

[0032] In some implementations, ECG data can be recorded using a sensor platform or electrode configuration other than a 12-lead ECG configuration, or with added features to a 12-lead ECG configuration. For example, a removable ECG patch can be attached to the patient's body surface, and the patch includes two or more electrodes that form one, two, or more leads capable of recording an ECG. In some implementations, the patient 102 can manually touch a fixed pair of external electrodes with their fingers, or the patient 102 can wear a watch, wristband, chest band, or other device that fixes two or more electrodes in place on the patient 102 to sense the electrical activity of the patient's heart. The patient 102 or a healthcare provider can use a mobile computing device (e.g., a smartphone 106) to configure an embodiment of the ECG procedure in a clinical or non-clinical setting.

[0033] System 100 further includes a data recorder and control unit 108, an ECG processing system 110, and one or more input / output devices 112a-b. In some implementations, wires extending from each of the electrodes 104 are coupled to the data recorder and control unit 108, enabling the recording of electrical signals sensed by the electrodes. Unit 108 may include, for example, an analog-to-digital (A / D) converter and other analog or digital signal conditioning circuits such as amplifiers and filters.

[0034] The data recorder and controller unit 108 is communicatively coupled to the ECG processing system 110. The ECG processing system 110 is a system consisting of one or more computers, which can be distributed in one or more physical locations. The system 110 may or may not be in the same physical location as the patient 102 and the data recorder and control unit 108. The system 110 can be coupled to a display screen 112a for presenting information to the user, and peripheral devices such as a mouse and keyboard 112 for receiving user input. Furthermore, the system 110 may include various components 114-126 that facilitate the processing of ECG data to determine estimated ejection fraction characteristics, estimated survival rates, and to present information regarding such estimates.

[0035] For example, system 110 may include a preprocessor 114 that digitally modifies the ECG signals received from the data recorder and control unit 108. The preprocessor 114 can perform noise reduction, anti-aliasing, or other digital techniques to prepare ECG data that describes patient 102's ECG for further processing.

[0036] System 110 further includes one or more ejection fraction prediction models 118. These models 118 are generally configured to process one or more predictive inputs characterizing the patient's ECG data and, based on the predictive inputs, generate an estimated ejection fraction characteristic for the patient 102. The ejection fraction characteristic may represent, for example, an absolute estimate of the patient's ejection fraction (e.g., a specific value such as 50 percent, 45 percent, 40 percent, 35 percent, 30 percent, or another value), or a classification of the patient's estimated ejection fraction (e.g., a normal ejection fraction greater than 50 percent, a low ejection fraction between 35 and 50 percent, or a very low ejection fraction below 35 percent).

[0037] As will be described in more detail below, the ejection fraction prediction model 118 can be a regression model, a machine learning model, or both. In some implementations, model 118 is a feedforward, recurrent, or convolutional neural network, or a capsule network. Neural network models can have fully connected layers, and autoencoder networks can be used. In some implementations, system 110 stores and maintains multiple ejection fraction prediction models 118. Each model 118 can correspond to a different set of patient characteristics (e.g., age, weight, sex, or other characteristics). When evaluating the ejection fraction for a new patient 102, system 110 can select an appropriate model from among the models 118 that corresponds to characteristics consistent with those of patient 102. Each model 118 can be trained, for example, on data points from patients having the corresponding characteristics for model 118. Other types of machine learning or regression models can also be applied, such as support vector machines (SVMs), hidden Markov models (HMMs), and other linear and nonlinear systems.

[0038] In some implementations, system 110 maintains one or more survival estimation models 120. The survival estimation models 120 process ECG data, estimated ejection fraction characteristics, or both, and are configured to generate predictions of the patient's estimated future survival rates (e.g., 1, 2, 5, and / or 10-year survival rates) that indicate the patient's likelihood of surviving a cardiac condition, based on the patient's ejection fraction characteristics. For example, an ejection fraction characteristic model 118 can be used to generate estimated ejection fraction characteristics for a patient. The survival estimation model 120 can then map these estimated ejection fraction characteristics to future survival rates for the patient, as determined from empirical data. For example, the estimated 5-year survival rate for a patient with a very low ejection fraction may be lower than that for a patient with a normal ejection fraction. In other implementations, the survival estimation model 120 can process predictive inputs characterizing the patient's ECG without first starting with the patient's estimated ejection fraction, and derive estimated survival rates for patient 102 directly from these predictive inputs. In this way, when determining the estimated survival rate for patient 102, the patient's ECG characteristics other than or in addition to the ejection fraction can be considered. The survival estimation model 120 can be a regression model, a machine learning model, or both. In some implementations, model 120 is a feedforward, recurrent, or convolutional neural network, or a capsule network. The neural network model can have fully connected layers, and an autoencoder network can be used. In some implementations, system 110 stores and maintains multiple survival estimation models 120. Each model 120 can correspond to a different set of patient characteristics (e.g., age, weight, sex, or other characteristics). When evaluating the ejection fraction for a new patient 102, system 110 can select an appropriate model from among the models 120 that corresponds to characteristics consistent with patient 102's characteristics. Each model 120 can be trained, for example, on data points from patients having the corresponding characteristics for model 120.

[0039] The predictive input generator 116 processes ECG data from the data recorder and control unit 108 or the ECG signal preprocessor 114 to generate a predictive input suitable for processing by the ejection fraction prediction model 118 or the survival estimation model 120. For example, the predictive input generator 116 can normalize and vectorize ECG data from one or more channels (corresponding to one or more leads) into a format expected by the ejection fraction prediction model 118. In some implementations, the predictive input includes a time series of values ​​representing the amplitude of the ECG for one or more leads at each point in time at a specified sampling frequency over a period spanning one or more cardiac cycles (e.g., 1, 2, 5, or 10 seconds). In some examples, the predictive input can represent an ECG signal over a relatively short time period (e.g., 1, 2, 5, or 10 seconds), while in other examples, the predictive input can represent an ECG signal over a relatively longer time period (e.g., 30 seconds, 1 minute, 2 minutes, 5 minutes, 10 minutes, or longer). In some implementations, the predictive input generator 116 averages ECGs from two or more cardiac cycles to generate an average ECG representing the average cardiac cycle of patient 102. In some implementations, the predictive input generator 116 selects a portion of the ECG corresponding to a single cardiac cycle and characterizes only the selected portion of the ECG with the predictive input. Therefore, the selected portion is representative of patient 102's cardiac cycle.

[0040] In some implementations, instead of processing actual time-series data representing the amplitude of the ECG waveform over time, the ejection fraction prediction model 118 and the survival estimation model 120 can instead process a predictive input that represents the value of one or more morphological features of the patient's ECG. Morphological features are parameters that characterize the shape of the ECG waveform or a portion of the ECG waveform, such as the P wave, QRS complex, or T wave. Exemplary morphological features that can be identified in the predictive input include T-wave amplitude, P-wave amplitude, P-wave area, T-wave area, T-wave leftward tilt, T-wave rightward tilt, P-wave leftward tilt, P-wave rightward tilt, T-wave duration, P-wave duration, PR interval, QRS duration, QRS amplitude, QRS area, QRS energy, QRS peak-to-peak ratio, QT segment length, or two or more combinations thereof. These multiple features are graphically shown in the exemplary ECG waveform 300 in Figure 3. In some implementations, system 110 includes a feature extractor 126, which analyzes the ECG data and determines the value of any applicable morphological feature, including a predictive input that is processed by an ejection fraction prediction model 118.

[0041] System 110 further includes a training subsystem 122 and a notification manager 124. The training subsystem 122 is configured to train an ejection fraction prediction model 118, a survival estimation model 120, or both. An exemplary process 600 performed by the training subsystem 122 to train the ejection fraction prediction model 118 is described with reference to Figure 6. The notification manager 124 is configured to provide estimated ejection fraction characteristics, estimated survival rates, or both as output to one or more users. In some implementations, the notification manager 124 provides estimated ejection fraction characteristics for display on screen 112a. In addition, one or more notification services can be registered with the computing system 110, so that the notification manager 124 puts the results of estimated ejection fraction characteristics, survival rate estimates, or both into each of the registered services for presentation by one or more means (e.g., to a smartphone in a healthcare provider's clinic for visual, audible, and / or tactile presentation).

[0042] Figure 2 is a flowchart of an exemplary process 200 for estimating the ejection fraction characteristics of a subject using ECG data. Process 200 can be performed by a computing system, for example, system 110 in Figure 1. In step 202, the system acquires ECG data for a patient. The ECG data may include one or more channels, each channel representing the ECG signal for each lead of the ECG procedure, and the data from each channel can be temporally aligned. The ECG data may span a single cardiac cycle, a portion of a cardiac cycle, or multiple cardiac cycles of the patient. In step 204, the system's predictive input generator generates one or more predictive inputs from the ECG data and feeds them to an ejection fraction prediction model, for example, ejection fraction prediction model 118. In some implementations, the predictive inputs represent a time series of values ​​for the ECG waveform. In other implementations, the predictive inputs represent values ​​for morphological features of the ECG waveform (step 206). In step 208, the system provides the predictive inputs to the ejection fraction prediction model for processing. In step 210, the ejection fraction prediction model processes a predictive input representing the patient's ECG results to generate an estimated (predicted) ejection fraction characteristic. In some implementations (212), the ejection fraction characteristic is an absolute estimate of the patient's ejection fraction, identifying a specific value for the ejection fraction. In other implementations (214), the ejection fraction characteristic is a category or range of ejection fraction values ​​(e.g., very low ejection fraction, low ejection fraction, or normal ejection fraction). For example, an ejection fraction model can be trained to classify a patient's ejection fraction into one of two, three, or more possible ejection fraction categories defined by a specified threshold ejection fraction value. A binary classification model can classify a patient's ejection fraction into two possible categories. The system then stores and / or outputs the estimated ejection fraction characteristic in step 216, for example, for presentation to the patient or the patient's healthcare provider.

[0043] In some implementations, the ejection fraction prediction model may be trained and configured to predict the risk of a patient developing a low or very low ejection fraction at some point in the future (e.g., in the next month, six months, one year, two years, five years, or more) by detecting subtle features or changes in the patient's ECG that indicate early disease onset, rather than estimating the patient's current ejection fraction characteristics (as described in other implementations).

[0044] Figure 3 shows an exemplary ECG trace of a patient's single heartbeat. The ECG trace reveals the constituent segments of the heartbeat and various morphological waveform features.

[0045] Figure 4 is a flowchart of an exemplary process 400 for selecting and using an appropriate ejection fraction prediction model corresponding to patient characteristics. Process 400 can be carried out by a computing system such as system 110 in Figure 1. In some examples, a global ejection fraction prediction model can be trained and used for a wide range of patients. However, in other examples, a more accurate or reliable estimate of a patient's ejection fraction can be determined by using a patient-specific personalized model, or by using a semi-personalized model trained on ECG data and measured ejection fraction characteristics from a population of patients with the same or similar characteristics as the new patient for whom an ejection fraction estimate is desired. For example, multiple ejection fraction prediction models can be generated, each corresponding to a different set of patient characteristics, such as age, height, weight, body volume index (BMI), sex, family history, diabetes, hypertension, hyperlipidemia, high-sensitivity C-reactive protein (CRP), smoking, history of coronary artery disease, history of inflammatory disease, or any other indicator of comorbidity such as a combination of these. In step 402, the system identifies the set of characteristics for the patient for whom an estimated ejection fraction characteristic should be determined. In step 404, the system selects one of the ejection fraction prediction models that corresponds to the identified set of characteristics for the patient. For example, for a male patient over 50 years of age, a model trained on data from subjects with similar characteristics to the patient can be selected, rather than another model trained on data from subjects with different characteristics. In step 406, the system generates an ejection fraction prediction using the selected ejection fraction prediction model that corresponds to the patient's characteristics. For example, the system can generate an estimated ejection fraction characteristic by the selected model according to process 200 described with respect to Figure 2.

[0046] ECG-based estimation of a patient's ejection fraction can be a useful screening measure, but the results of an ECG-based screening measure may justify further evaluation of the patient. Figure 5 is a flowchart of an exemplary process 500 that uses an ECG-based method to screen for potentially problematic ejection fraction levels. Process 500 can be automated and performed, for example, by a computing system 110, or by a healthcare provider or other appropriate individual. In step 502, the patient's estimated ejection fraction characteristics are obtained based on process 200, for example, as described with respect to Figure 2. In step 504, the system determines whether the estimated ejection fraction characteristic and, optionally, additional factors meet one or more screening criteria to guide a decision on whether further evaluation of the patient's condition is justified. For example, the estimated ejection fraction characteristic may be an absolute value representing the patient's predicted ejection fraction, and the screening criteria may include a threshold ejection fraction (e.g., 35 percent or 50 percent). If the patient's estimated ejection fraction falls below the threshold, subsequent actions for further evaluation may be taken for the patient. These subsequent actions may include, for example, cardiac ultrasound, MRI, CT scan, nuclear medicine diagnosis, or a combination thereof. In some implementations, if the patient's ejection fraction characteristic falls into a range of ECG values ​​considered dangerous or unsafe, further medical evaluation may be taken for the patient.

[0047] In some implementations, a patient may be administered medication or treated if their ejection fraction falls below a threshold (e.g., 35 percent or 50 percent), or if their ejection fraction characteristics fall within a range of ECG values ​​considered dangerous or unsafe. Such medications or treatments may include beta-blockers, angiotensin receptor blockers, statins (for coronary artery disease), implantable cardioverter-defibrillators, cardiac resynchronization devices, and other clearly defined therapies (see 2013 ACCF / AHA Guideline for the Management of Heart Failure. Circulation. 2013;128:e240-e327 and 2016 ACC / AHA / HFSA Focused Update on New Pharmacological Therapy for Heart Failure, Circulation. 2016;134:e282-e293). In some implementations, if a patient's ejection fraction falls below a threshold (e.g., 35 percent or 50 percent), or if the patient's ejection fraction characteristics fall into a range of ECG values ​​considered dangerous or unsafe, additional diagnostic testing with therapeutic interventions may include screening for inflammation and other systemic conditions, and therapeutic interventions may include specific treatments or coronary angiography or other imaging to detect coronary artery disease, for which such diseases have clearly defined treatments.

[0048] Figure 6 is a flowchart of an exemplary process 600 for training an ejection fraction prediction model, such as a neural network-based model. In some implementations, machine learning techniques, such as gradient descent (including batch gradient descent or stochastic gradient descent), can be used to train the ejection fraction model. Process 600 can be carried out by a computing system such as the training subsystem 122 described with respect to Figure 1. In step 602, the system acquires multiple training data pair sets. Each pair includes an ECG prediction input characterizing the ECG of a particular patient and a target ejection fraction characteristic for that patient. The target ejection fraction characteristic can be a “true” or measured ejection fraction characteristic determined by ultrasound or other available procedures. In some cases, the model is trained on data from many different patients. If the model is semi-personalized, the patients represented in the training set may share a common set of characteristics with each other. In step 604, the training system selects a first training data pair for processing. In step 606, the prediction input from the training data pair is provided to the ejection fraction prediction model. The ejection fraction prediction model can be, for example, a neural network machine learning model having one or more layers of perceptrons, each with associated weights / parameters for an activation function associated with that perceptron. These weights can be initialized randomly when training begins and can be gradually refined over time as additional training iterations are performed. In step 608, the system processes the predictive input according to the model's current weights / parameters to generate an estimated ejection fraction characteristic. In step 610, the estimated ejection fraction characteristic is compared to the target ejection fraction characteristic to determine the output error. This output error is then backpropagated through the network using gradient descent to update the neural network's current weights / parameters. The process then returns to step 604, where another pair of training data is selected and steps 604-612 are repeated until the training termination condition is met.

[0049] Ejection fraction prediction models can be trained on large datasets, but the models themselves can be relatively small. After training, the models can be relatively small and can predict ejection fraction characteristics with less computational demand than might have been required during training.

[0050] Process 600 generally belongs to supervised learning processes, but in some implementations, the ejection fraction prediction model can also be trained using unsupervised learning techniques.

[0051] Figure 7 is a flowchart of an exemplary process 700 for predicting a subject's survival rate from cardiac conditions such as low or very low ejection fraction from ECG data. In some implementations, process 700 is performed by a computing system that includes a survival estimation model, such as model 120 (Figure 1) of the ECG processing system 110. In step 702, the system acquires ECG data for the patient. The ECG data may include one or more channels, each channel representing the ECG signal for each lead of the ECG procedure, and the data from each channel can be temporally aligned. The ECG data may span a single cardiac cycle, a portion of a cardiac cycle, or multiple cardiac cycles of the patient. In step 704, the system's predictive input generator generates one or more predictive inputs from the ECG data and feeds them into the survival estimation model. In some implementations, the predictive inputs represent a time series of values ​​for the ECG waveform. In other implementations, the predictive inputs represent values ​​for morphological features of the ECG waveform (706). In step 708, the predictive input is provided to the survival estimation model, and in step 710, the model processes the predictive input to determine the patient's estimated survival rate based on the patient's ECG. The estimated survival rate can be, for example, a prediction of 1 month, 1 year, 2 years, 3 years, 4 years, 5 years, or 10 years. The system can then provide the survival rate estimate as an output for presentation to the patient, the patient's healthcare provider, or both (step 712).

[0052] Figure 8 shows an exemplary neural network model 800 for estimating the ejection fraction characteristics of a subject from ECG data. Model 800 processes an ECG prediction input that characterizes one or more channels of the patient's ECG. As can be seen in Figure 8, Model 800 may include an input layer of perceptrons and / or sigmoid neurons, an output layer, and one or more hidden layers between the input and output layers. For example, data from multiple channels may be processed and pooled until the data is provided to the fully connected layer of the network. Model 800 may output an estimate or classification of the ejection fraction characteristics for the patient.

[0053] In some implementations, the neural network model 800 includes multiple convolutional layers for feature extraction. Each convolutional layer may include a constant or variable-length filter that focuses on the current lead of the ECG or multiple leads simultaneously (e.g., k×1 or k×m leads). Following the convolutional layers, the network may include one or more fully connected layers, each having the same or different number of neurons. Zero, one, or more recurrent neural network layers may be added before, after, or in parallel with the convolutional layers for temporal feature extraction. Skip layers and neurons may or may not be included.

[0054] Figure 9 shows the results of a first neural network implementation of an ejection fraction prediction model that categorizes patients into groups according to whether their ejection fraction is predicted to be above or below 50 percent.

[0055] Figure 10 shows the results of a second neural network implementation of the ejection fraction prediction model, which categorizes patients into groups according to whether their ejection fraction is predicted to be above or below 35 percent.

[0056] Figure 11 is a graph showing the correlation between model output and true ejection fraction values ​​in the study.

[0057] While several examples of detecting (low) ejection fraction (ECU) impairment in patients using data from a standard 12-lead ECG are described herein, other implementations may effectively detect cases of low ECU using fewer leads, including single-lead ECGs, or identify individuals at higher risk of developing low ECU within a future period (e.g., 5 years). Ejection fraction prediction models can be trained and implemented based on data from single-lead, arbitrary multi-lead configurations, and any number of ECG leads, including a standard 12-lead ECG. For example, a single-lead ECG based on lead 1 has been shown to be particularly effective in providing reliable data from which the ECU prediction model can generate an ejection fraction prediction. Advantageously, ECU prediction based on single-lead ECGs or other non-conventional electrodes (e.g., fewer than a standard 12-lead ECG) may expand access to screening devices that implement the ECU prediction models described herein. For example, a pair of electrodes can be integrated with or communicatively coupled to a smartphone or other portable electronic device, allowing an individual to conveniently capture their own ECG and quickly obtain predictive results by, for example, touching the electrodes with their fingertips. In other implementations, electrodes can also be placed within a wearable patch for continuous monitoring over a period of time.

[0058] In certain cases, yet another advantage that can be achieved by at least some of the implementations described herein is the ability to detect low ejection fraction based on a short, single-time profile of ECG data for a patient (e.g., from a single lead or multiple leads). For example, the system may only need to process relatively short samples of ECG data to generate a highly reliable prediction of a patient's ejection fraction status, such as whether the patient has low EF. In some implementations, the ejection fraction prediction model is configured to predict whether a patient has low EF (e.g., EF less than 35, 40, or 50 percent) based on ECG samples of 60 seconds or less. In some implementations, the ejection fraction prediction model is configured to predict whether a patient has low EF (e.g., EF less than 35, 40, or 50 percent) based on ECG samples of 30 seconds or less. In some implementations, the ejection fraction prediction model is configured to predict whether a patient has low EF (e.g., EF less than 35, 40, or 50 percent) based on ECG samples of 20 seconds or less. In some implementations, the ejection fraction prediction model is configured to predict whether a patient has a low EF (e.g., less than 35, 40, or 50 percent EF) based on an ECG sample taken at 15 seconds or less. In some implementations, the ejection fraction prediction model is configured to predict whether a patient has a low EF (e.g., less than 35, 40, or 50 percent EF) based on an ECG sample taken at 10 seconds or less. In some implementations, the ejection fraction prediction model is configured to predict whether a patient has a low EF (e.g., less than 35, 40, or 50 percent EF) based on an ECG sample taken at 5 seconds or less. In some implementations, the ejection fraction prediction model is configured to predict whether a patient has a low EF (e.g., less than 35, 40, or 50 percent EF) based on an ECG sample taken at 2 seconds or less. Therefore, the acquisition time for the ECG can be relatively short, which is advantageous for the patient.In addition, using machine learning models such as convolutional neural networks trained on large datasets enables high-speed processing, allowing computing systems running these models to return ejection fraction prediction results based on ECG input in a relatively short amount of time.

[0059] We have described models for estimating or predicting an individual's ejection fraction in specific implementation forms. However, it should be understood that these techniques can be more generally extended to models that facilitate the detection of a variety of current or impending structural heart conditions, including ejection fraction abnormalities, left ventricular myocardial mass abnormalities (e.g., elevated left ventricular myocardial mass, low left ventricular myocardial mass), valvular heart disease, ischemic heart disease, atrial appendage abnormalities, the presence of shunts, or patent foramen ovale (PFO), dilation of cardiac chambers (e.g., left atrium, right atrium, left ventricle, right ventricle), or combinations thereof and / or other structural heart conditions.

[0060] Figure 12 is a block diagram of computing devices 1200, 1250 that can be used as clients or one or more servers to implement the systems and methods described herein. Computing device 1200 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Computing device 1250 is intended to represent various forms of mobile devices, such as personal digital assistants, mobile phones, smartphones, and other similar computing devices. In addition, computing device 1200 or 1250 may include a Universal Serial Bus (USB) flash drive. The USB flash drive may store an operating system and other applications. The USB flash drive may include input / output components such as a wireless transmitter or USB connector that can be inserted into a USB port of another computing device. The components shown herein, their connections and relationships, and functions are illustrative only and are not intended to limit the implementations described and / or claimed herein.

[0061] The computing device 1200 includes a processor 1202, memory 1204, storage device 1206, a high-speed interface 1208 connected to memory 1204 and a high-speed expansion port 1210, and a low-speed bus 1214 and a low-speed interface 1212 connected to storage device 1206. Components 1202, 1204, 1206, 1208, 1210, and 1212 are interconnected using various buses and can be mounted on a common motherboard or by other appropriate methods. The processor 1202 can process instructions for execution within the computing device 1200, including instructions stored in memory 1204 or on storage device 1206, to display graphical information as a GUI on an external input / output device such as a display 1216 coupled to the high-speed interface 1208. In other implementations, multiple processors and / or multiple buses can be used as appropriate, along with multiple memories and multiple types of memory. Furthermore, multiple computing devices 1200 can be connected, with each device providing several parts of the required operation (for example, as a server bank, a group of blade servers, or a multiprocessor system).

[0062] Memory 1204 stores information within the computing device 1200. In one implementation, memory 1204 is one or more volatile memory units. In another implementation, memory 1204 is one or more non-volatile memory units. Memory 1204 can also be another form of computer-readable medium, such as a magnetic or optical disk.

[0063] The storage device 1206 can provide a large-capacity storage device to the computing device 1200. In one implementation, the storage device 1206 may be an array of devices including computer-readable media such as floppy disk devices, hard disk devices, optical disk devices or tape devices, flash memory, or other similar solid-state storage devices, or devices in a storage area network or other configuration. A computer program product can be tangibly implemented within the information carrier. The computer program product may also include instructions that, when executed, perform one or more of the methods described above. The information carrier is a computer or machine-readable medium such as memory 1204, storage device 1206, or memory-on-processor 1202.

[0064] The high-speed controller 1208 manages bandwidth-intensive operations for the computing device 1200, while the low-speed controller 1212 manages less bandwidth-intensive operations. Such function assignments are merely illustrative. In one implementation, the high-speed controller 1208 is coupled to memory 1204, display 1216 (e.g., via a graphics processor or accelerator), and high-speed expansion port 1210, which can accept various expansion cards (not shown). In this implementation, the low-speed controller 1212 is coupled to the storage device 1206 and the low-speed expansion port 1214. The low-speed expansion port may include various communication ports (e.g., USB, Bluetooth®, Ethernet®, wireless Ethernet) and can be coupled to one or more input / output devices such as a keyboard, pointing device, scanner, or networking devices such as a switch or router, for example, via a network adapter.

[0065] The computing device 1200 can be implemented in several different forms, as shown in the figure. For example, the computing device 1200 can be implemented as a single standard server 1220, or multiple times within a group of such servers. The computing device 1200 can also be implemented as part of a rack server system 1224. In addition, the computing device 1200 can be implemented within a personal computer, such as a laptop computer 1222. Alternatively, components from the computing device 1200 can be combined with other components within a mobile device (not shown), such as device 1250. Each such device may contain one or more computing devices 1200, 1250, and the entire system can consist of multiple computing devices 1200, 1250 communicating with each other.

[0066] Computing device 1250 includes, among other components, input / output devices such as a processor 1252, memory 1264, and display 1254, a communication interface 1266, and a transceiver 1268. Device 1250 may also include storage devices such as a microdrive or other devices to provide additional storage. Components 1250, 1252, 1264, 1254, 1266, and 1268 are interconnected using various buses, and some of these components can be mounted on a common motherboard or by other suitable methods.

[0067] The processor 1252 can execute instructions in the computing device 1250, including instructions stored in memory 1264. The processor can be implemented as a chipset consisting of multiple chips containing separate analog and digital processors. In addition, the processor can be implemented using one of several architectures. For example, the processor 1252 can be a CISC (Complex Instruction Set Computer) processor, a RISC (Reduced Instruction Set Computer) processor, or a MISC (Minimum Instruction Set Computer) processor. The processor can provide coordination of other components of the device 1250, such as user interface control, applications run by the device 1250, and wireless communication by the device 1250.

[0068] The processor 1252 can communicate with the user via a control interface 1258 and a display interface 1256 coupled to the display 1254. The display 1254 can be, for example, a TFT (thin-film transistor liquid crystal display) display, an OLED (organic light-emitting diode) display, or other suitable display technology. The display interface 1256 may include suitable circuitry for driving the display 1254 to present graphical and other information to the user. The control interface 1258 can receive commands from the user and convert them for transmission to the processor 1252. In addition, an external interface 1262 may be provided to communicate with the processor 1252 to enable near-area communication between device 1250 and other devices. The external interface 1262 may provide, for example, wired communication in some implementations or wireless communication in other implementations, and multiple interfaces may be used.

[0069] Memory 1264 stores information within the computing device 1250. Memory 1264 can be implemented as one or more computer-readable media, one or more volatile memory units, or one or more non-volatile memory units. Furthermore, an extended memory 1274 can be provided and connected to the device 1250 via an expansion interface 1272, which may include, for example, a SIMM (Single In-Line Memory Module) card interface. Such an extended memory 1274 can provide extra storage space for the device 1250, or it can store applications or other information for the device 1250. Specifically, the extended memory 1274 may contain instructions that perform or supplement the processes described above, and may also contain secure information. Therefore, for example, the extended memory 1274 can be provided as a security module for the device 1250 and programmed with instructions that enable the secure use of the device 1250. In addition, a secure application can be provided via a SIMM card, along with additional information, such as placing identification information on the SIMM card to prevent hacking.

[0070] The memory may include, for example, flash memory and / or NVRAM memory, as discussed below. In one implementation, a computer program product is tangibly implemented within an information carrier. The computer program product includes instructions that, when executed, perform one or more of the methods described above. The information carrier is a computer or machine-readable medium such as memory 1264, extended memory 1274, or memory-on-processor 1252, which can be received, for example, via transceiver 1268 or external interface 1262.

[0071] Device 1250 can communicate wirelessly via a communication interface 1266, which may include digital signal processing circuitry if necessary. The communication interface 1266 can provide communication in various modes or protocols, including, among others, GSM® voice call, SMS, EMS, or MMS messaging, CDMA, TDMA, PDC, WCDMA®, CDMA2000, or GPRS. Such communication can be performed, for example, via a radio frequency transceiver 1268. In addition, short-range communication can be performed using Bluetooth®, WiFi, or other such transceivers (not shown). Furthermore, a GPS (Global Positioning System) receiver module 1270 can provide additional navigation and location-related radio data to device 1250, which can be used as appropriate by applications running on device 1250.

[0072] Device 1250 can also communicate audibly using the audio codec 1260, which can receive speech information from the user and convert it into usable digital information. The audio codec 1260 can similarly generate audible sound for the user, for example, within the handset of device 1250, via a speaker. Such sound may include sounds from voice calls, recorded sounds (e.g., voice messages, music files, etc.), and sounds generated by applications running on device 1250.

[0073] The computing device 1250 can be implemented in several different forms, as shown in the figure. For example, the computing device 1250 can be implemented as a mobile phone 1280. The computing device 1250 can also be implemented as part of a smartphone 1282, a personal digital assistant, or other similar mobile device.

[0074] Various implementations of the systems and techniques described herein can be realized in digital electronic circuits, integrated circuits, specially designed ASICs (application-specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs executable and / or interpretable on a programmable system, the programmable system including at least one programmable processor which may be special-purpose or general-purpose and coupled to receive data and instructions from and transmit data and instructions to a storage system, at least one input device, and at least one output device.

[0075] These computer programs (also known as programs, software, software applications, or code) include machine instructions to a programmable processor and can be implemented in high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. In this specification, the terms “machine-readable medium” and “computer-readable medium” refer to any computer program product, apparatus, and / or device (e.g., magnetic disks, optical disks, memory, programmable logic devices (PLDs)) used to provide machine instructions and / or data to a programmable processor, including machine-readable medium that receives machine instructions as machine-readable signals. The term “machine-readable signal” refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0076] To provide user interaction, the systems and techniques described herein can be implemented on a computer having a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) and a keyboard and pointing device (e.g., a mouse or trackball) on which the user can provide input to the computer. Other types of devices can also be used to provide user interaction; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback), and the input from the user can be received in any form, including acoustic, verbal, or tactile input.

[0077] The systems and techniques described herein can be implemented within a computing system, which may include backend components (e.g., as a data server), middleware components (e.g., an application server), or frontend components (e.g., a client computer having a graphical user interface or web browser that allows a user to interact with the implementation of the systems and techniques described herein), or any combination of such backend, middleware, or frontend components. The components of the system may be interconnected by digital data communications (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks ("LANs"), wide area networks ("WANs"), peer-to-peer networks (with temporary or static members), grid computing infrastructure, and the Internet.

[0078] A computing system can include clients and servers. Clients and servers are generally remote from each other and typically interact via a communication network. The relationship between a client and a server arises from computer programs running on each computer and having a client-server relationship with each other.

[0079] [Example Implementation] <Overview> This example belongs to a study based on artificial intelligence ("AI") developed using the techniques described herein to detect patients with asymptomatic low left ventricular dysfunction (ALVD) with high fidelity using a non-invasive 10-second digital ECG.

[0080] background ALVD is present in 2–9% of the population, is associated with a decline in quality of life and lifespan, and is generally treatable when detected. The area under the curve (AUC) for BNP screening blood tests is 0.79–0.89. This study hypothesized that the use of artificial intelligence (AI) could enable the identification of left ventricular systolic dysfunction using an ECG, a ubiquitous and inexpensive test.

[0081] method This study involved training a convolutional neural network to identify patients with ventricular dysfunction, defined as an ejection fraction (EF) ≤ 35%, using digitally stored pairs of 12-lead ECG and echocardiography images from 44,959 patients in the Mayo Clinic DataVault. This network was then tested on 52,870 patients reserved for external validation.

[0082] result Of the 52,870 patients tested, 4,131 (7.8%) had an EF ≤ 35%. The AUC of the ROC was 0.93. Sensitivity, specificity, and precision were 86.3%, 85.7%, and 85.7%, respectively. Of the 1,335 patients with abnormal AI screening but normal EF (false positive), 147 (11%) developed at least one abnormal EF in the future (5-year incidence rate 9.5%). This fourfold increase in the risk of developing low LVEF in the future suggests that the network may be detecting early, asymptomatic, metabolic, or structural abnormalities that appear on the ECG.

[0083] conclusion By applying artificial intelligence to ECGs, which are ubiquitous and typically low-cost tests, ECGs can serve as a powerful tool for screening left ventricular dysfunction and further identifying individuals at high risk of developing it in the future.

[0084] <Method> Data sources and research population Following approval by the institutional review board, the study acquired data from the Mayo Clinic digital data vault. 163,892 adult patients (18 years or older) were identified by at least one digital standard 10-second 12-lead ECG acquired in the supine position between January 1994 and February 2017, and at least one transthoracic ultrasound (TTE) acquired within 14 days of the index ECG (Figure 13). For patients with multiple ECG and TTE datasets meeting these criteria, the earliest pair was used for networking, validation, or testing, and subsequent TTE data were used for follow-up analysis. A prior proof-of-concept assessment was conducted to obtain an internal survey using 2200 ECG-TTE data pairs excluded from this analysis, leaving a cohort of 97,829 patients, whose first ECG-TEE pair datasets were used for the primary analysis.

[0085] ECGs were acquired at a sampling rate of 500 Hz using a General Electric Market ECG machine (Market, Wisconsin) and stored using the MUSE data management system for later retrieval. Comprehensive 2D or 3D Doppler ultrasound cardiac examinations were available to all patients. Quantitative data at the time of acquisition were recorded in a custom database (Echocardiography Image Management System, EIMS) developed by the Mayo Clinic. Left ventricular ejection fraction (EF) was routinely measured or estimated using standardized methods, and most reports allowed for the recording of two or more values. For the purposes of this study, the ejection fraction values ​​used in the model were the first values ​​available from a standard hierarchical sequence, and EF was determined using three-dimensional echocardiography (Yamani H, Cai Q, Ahmad M, Three-dimensional echocardiography in evaluation of left ventricular indices. Echocardiography 2012;29:66-75), a two-plane method using the Simpson method, a two-dimensional method (Quinones MA, Waggoner AD, Reduto LA et al., A new, simplified and accurate method for determining ejection fraction with two-dimensional echocardiography. Circulation 1981;64:744-53), M-mode, or, if none of the above were available, a visually estimated EF report. If the estimate was a range, the midpoint was used as a single EF value. Left ventricular ejection fraction (EF) was classified as low (≤35%), slightly decreased (35-49%), or normal (≥50%).

[0086] First and second stage results The primary result was the AI ​​network's ability to identify patients with LVEF of 35% or less using only ECG signals. This value was selected based on its clear clinical and therapeutic significance (Russo AM, Stainback RF, Bailey SR et al., ACCF / HRS / AHA / ASE / HFSA / SCAI / SCCT / SCMR 2013 appropriate use criteria for implantable cardioverter-defibrillators and cardiac resynchronization therapy: a report of the American College of Cardiology Foundation appropriate use criteria task force, Heart Rhythm Society, American Heart Association, American Society of Echocardiography, Heart Failure Society of America, Society for Cardiovascular Angiography and Interventions, Society of Cardiovascular Computed Tomography, and Society for Cardiovascular Magnetic Resonance. J Am Coll Cardiol 2013;61:1318-68). A secondary result was the AI ​​network's ability to identify individuals who had normal EF at the time of screening but whose risk of later low EF increased during follow-up.

[0087] Overview of AI Model Development This research involved developing a convolutional neural network (CNN) using the Keras Framework along with the TensorFlow® backend (Google, Mountain View, California) and Python (van Rossum G. Python tutorial, Technical Report CS-R9526. Amsterdam 1995 May). The CNN applied to images (or videos) works to extract very small patterns within the dataset using convolution. Each 12-lead ECG was treated as a 12 × 5000 "image" (i.e., 12 leads sampled at 500 Hz with a duration of 10 seconds) (van Rossum G. Python tutorial, Technical Report CS-R9526. Amsterdam 1995 May). The network consisted of N single-lead convolutional layers, each followed by a nonlinear "Relu" activation function, a batch normalization layer (Ioffe S, Szegedy C, Batch normalization: accelerating deep network training by reducing internal covariate shift, International Conference on Machine Learning 2015), and a maximum pooling layer (Nagi J, Ducatelle F, Di Caro G et al., Max-pooling convolutional neural networks for vision-based hand gesture recognition. 2011 IEEE International Conference 2011:342-7). Features extracted from each raw digital ECG lead were fused into another convolutional layer that simultaneously had access to all leads.The data was fed into a fully connected network with two hidden layers, including a dropout layer to avoid overfitting and an output layer activated using the "Softmax" function, following the final convolutional layer (Cristianini N, Shawe-Taylor J, New York, NY: Cambridge University Press; 1999).

[0088] AI model training Due to the large size of our dataset, we used approximately 50% of it to train the network. This allowed us to obtain a very large dataset for testing the network and better evaluating its robustness (Figure 13). After the initial split into development and test (holdout) datasets, we further split the development dataset into training data (80% of the development set) and internal validation data (20%).

[0089] During training, ECGs were fed into the network, and the Adam optimizer (Kingma DP, Ba J. Adam: A Method for Stochastic Optimization 2014) was used to update the network weights using binary cross-entropy as the loss function. After each epoch, the network was tested using an internal validation dataset, and training was stopped after optimization. Network hyperparameters were also tuned during this process, and the network with the lowest loss value was selected. All ECGs were passed through a low-pass filter (100 Hz) to remove high-frequency noise and quantization errors.

[0090] Primary result - AI extended ECG to identify low EF. After selecting the optimal network using validation data, a receiver operating curve (ROC) was created using the same validation set, and the area under the curve (AUC) was measured as a primary assessment of network strength. Using the ROC from the validation dataset, two thresholds were selected for the probability of having a low LVEF. The first threshold was selected by giving equal weights to sensitivity and specificity, and the second threshold was selected to give 90% sensitivity on the validation dataset. The CNN model was then used on the test data to test its ability to predict low LVEF. Sensitivity, specificity, and accuracy on the test data not used for model training or threshold selection were calculated using the two thresholds. The impact of age and sex on the network's predictive function was also evaluated by creating separate networks with those variables as input and by training the networks to determine whether age and sex could be determined from ECG alone.

[0091] Secondary results - AI extended ECG to predict future low EF We hypothesized that in the early stages of any disease affecting EF, ECG signals would show subtle abnormal patterns due to metabolic and structural disturbances that did not affect enough myocardial layer to cause a decrease in EF. Furthermore, we hypothesized that CNN would classify some of these cases as abnormal, which initially appear to be false-positive tests but may become true-positive tests over time. To test this hypothesis, we designed sub-studies to identify patients who met the following conditions: 1) the network predicted the patient would have low EF, 2) an echocardiogram was performed on the patient within 14 days demonstrating normal EF (≥50%) indicating false-positive detection by the algorithm, and 3) at least one additional echocardiogram (not used for training or testing) was available to the patient at a later date. A control group was created using true-negative cases (where both algorithmic and clinical judgments were consistent in not having low EF). Subjects were selected for the control group to have LVEF ≥50%. Kaplan-Meier analysis was used to show the incidence of low EF in terms of true-negative vs. false-positive over time. Next, we used Cox proportional hazards regression to estimate the hazard for low EF after adjusting for age and sex. In addition, we performed a sensitivity analysis by classifying the predicted probabilities for low EF to determine whether there is a monotonic pattern in the predicted probabilities for future occurrence of low EF.

[0092] statistical considerations For measuring diagnostic performance (AUC ROC, sensitivity), the sample size is sufficiently large, and therefore the normal confidence interval (CI) is expected to have a width of less than 0.5%. These CIs are therefore not reported along with estimates due to their high precision. Continuous data are presented as mean + / - SD. Two-tailed p-values ​​are presented for the Cox model. Survival analysis was performed using SAS version 9.4. CNNs were trained using Keras (version 2.0.3) and TensorFlow (version 1.0.1).

[0093] Research support The research was conceived, funded, and conducted entirely by the Mayo Clinic. There was no industrial support of any kind.

[0094] <Result> Study population A total of 625,326 patients who underwent ECG-TTE pairs were selected, and the study cohort chosen for analysis was identified (Figure 13). The first ECG-TTE data pairs from patients who underwent ECG and echocardiography at intervals of 2 weeks or less constituted the analysis dataset of 97,829 patients, including 35,970 in the training set, 8,989 in the validation set, and 52,870 in the holdout trial set. No patients belonged to more than one group (Figure 13). The mean age of the overall patient population was 61.8 ± 16.5 years, and 7.8% of the population had an EF ≤ 35%. Table 1 shows patient characteristics for the training, validation, and trial sets. In the study dataset, 4,131 patients (7.8%) had an EF of 35% or less, 6,740 patients (12.7%) had an LVEF between 35% and 50%, and 41,999 patients (79.5%) had an LVEF of 50% or more. Over 89% of the TTE was performed within one day of the indicative ECG.

[0095] Primary result In Figure 14, the AUC of the holdout trial dataset was 0.93, identical to the AUC of the internal validation dataset (0.93). When a threshold was selected without prioritizing sensitivity, the overall precision was 85.7%, specificity was 85.7%, sensitivity was 86.3%, and negative predictive value was 98.7%. When the algorithm was applied to the trial dataset using a threshold that gave 90% sensitivity in the validation set, the sensitivity was 89.1%, specificity was 83%, overall precision was 83.5%, and negative predictive value was 98.9%. When patients without known comorbidities (see Figure 16) were analyzed separately by the network, the AUC increased to 0.98, with sensitivity of 95.6%, specificity of 92.4%, negative predictive value of 99.8%, and precision of 92.5%.

[0096] Network performance did not improve with the addition of age and sex as inputs. To understand whether the network was "determining" age and sex based solely on the ECG, the network was retrained to predict age and sex from the ECG. It predicted age as 8.7 ± 6 years and accurately determined sex from the ECG with 87 percent accuracy, resulting in an AUC of 0.94. This indicates that the network was able to determine age and sex with high confidence from the ECG alone, and explains why its performance was not affected by age or sex, since the network should have been able to explain these variables.

[0097] Distribution of EF values ​​by AI algorithm classification When thresholds were selected without prioritizing sensitivity (i.e., thresholds that yielded equal sensitivity and specificity using validation data), 10,544 patients (19.9%) in the study set were identified by the network as having low EF. Of these 10,544 patients, 33.8% had an EF of ≤35%, 29.5% had an EF of 36–50%, and 36.6% had a normal EF. Within the group identified as normal by the network, 1.3% had an EF of ≤35%, 8.6% had a LVEF of 36–50%, and the remainder (90.1%) had a normal LVEF (Figure 17).

[0098] Secondary results - AI extended ECG to predict future low EF Of the patients identified by the network as having normal EF and also having conclusive, concurrent normal EF by echocardiography ("true negative"), 11,515 underwent follow-up echocardiography. Of these true negative patients, 302 developed low EF at a median follow-up of 3.8 (1.4–7.5) years (IQR) (Figure 4: 10-year incidence of 4.4%). In contrast, 1,335 patients were indicated by the network as having low EF, but concurrent echocardiography demonstrated normal EF ("false positive"). Of these 1,335 patients with initial "false positive" results, 147 (Figure 4: 10-year incidence of 20.8%) developed left ventricular dysfunction during a median follow-up of 3.4 (IQR 1.2–6.8) years. This indicates that when the AI ​​algorithm defines an ECG as abnormal, the risk of future low EF is four times higher (adjusted for age and sex, HR=4.1 [3.3~5.0], p<0.001), suggesting that the network identified the ECG abnormality before overt ventricular dysfunction appeared.

[0099] <Discussion> Left ventricular systolic dysfunction is associated with a reduced quality of life, increased morbidity, and increased mortality (McDonagh TA, McDonald K, Maisel AS, Screening for asymptomatic left ventricular dysfunction using B-type natriuretic peptide. Congest Heart Fail 2008;14:5-8). Major cardiovascular specialist organizations endorse evidence-based treatments that improve symptoms and survival rates once they are detected (Al-Khatib SM, Stevenson WG, Ackerman MJ et al., 2017 AHA / ACC / HRS Guideline for Management of Patients With Ventricular Arrhythmias and the Prevention of Sudden Cardiac Death: Executive Summary: A Report of the American College of Cardiology / American Heart Association Task Force on Clinical Practice Guidelines and the Heart Rhythm Society. Circulation 2017; Yancy CW, Jessup M, Bozkurt B et al., 2013 ACCF / AHA guideline for the management of heart failure: a report of the American College of Cardiology Foundation / American Heart Association Task Force on Practice Guidelines. J Am Coll Cardiol 2013;62:e147-239).However, there is a lack of effective screening for ventricular dysfunction (McDonagh TA, McDonald K, Maisel AS, Screening for asymptomatic left ventricular dysfunction using B-type natriuretic peptide. Congest Heart Fail 2008;14:5-8; Redfield MM, Rodeheffer RJ, Jacobsen SJ, Mahoney DW, Bailey KR, Burnett JC, Jr. Plasma brain natriuretic peptide to detect preclinical ventricular systolic or diastolic dysfunction: a community-based study. Circulation 2004;109:3176-81). This study found that applying artificial intelligence using a convolutional neural network to a standard 12-lead ECG, an inexpensive and widely available common clinical trial, enables the detection of left ventricular dysfunction with an AUC of 0.93. The performance of this trial is comparable to other common screening tests such as prostate-specific antigen for prostate cancer (0.92), mammography for breast cancer (0.85), and cervical cytology for cervical cancer (0.70). In contrast to BNP, accuracy was not affected by age or sex. In addition to effectively identifying individuals with ventricular systolic dysfunction, the network also predicted individuals who initially had normal LV function but later developed low EF. Patients with normal EF but abnormal network screening ("false positives") had a four-fold increased risk of developing ventricular dysfunction in the next five years (a 10% risk in five years). This suggests that the network detected early, asymptomatic, metabolic, or structural abnormalities that manifested on the ECG.

[0100] Congestive heart failure affects more than 5 million people in the United States alone and costs more than $30 billion annually in healthcare (Heidenreich PA, Albert NM, Allen LA et al., Forecasting the impact of heart failure in the United States: a policy statement from the American Heart Association. Circ Heart Fail 2013;6:606-19; Mozaffarian D, Benjamin EJ, Go AS et al., Heart disease and stroke statistics--2015 update: a report from the American Heart Association. Circulation 2015;131:e29-322). Early detection and prevention are paramount in health management. Asymptomatic left ventricular dysfunction affects more than 7 million Americans and many more people worldwide. Six percent of patients in the inventors' population had low ejection fraction (EF), which was found to be consistent with previous studies (McDonagh TA, McDonald K, Maisel AS, Screening for asymptomatic left ventricular dysfunction using B-type natriuretic peptide. Congest Heart Fail 2008;14:5-8). BNP and NT-BNP have been proposed for the detection of left ventricular systolic dysfunction.Bhalla and collaborators evaluated BNP to screen for systolic and diastolic dysfunction, finding that the AUC was 0.60 for BnP and 0.70 for NTproBNP, and that the results improved to 0.70 and 0.73 with the addition of impedance cardiography (Bhalla V, Isakson S, Bhalla MA et al., Diagnostic ability of B-type natriuretic peptide and impedance cardiography: testing to identify left ventricular dysfunction in hypertensive patients. Am J Hypertens 2005;18:73S-81S). A Mayo Clinic study from Olmsted County evaluating individuals aged 45 and older found that the AUC was higher in individuals with more severe systolic dysfunction (0.82-0.92) than in individuals with other systolic dysfunction (0.51-0.74). Furthermore, the optimal level of discrimination for BNP varied with age and sex. In contrast, this study found that superior AI network performance was invariant with respect to age and sex. This was validated by creating a network capable of determining the age and sex of patients within the range of 8 ± 6.3 years with 87% accuracy based solely on ECG.Previous literature has described changes in ECG patterns with age and sex (Daly C, Clemens F, Lopez Sendon JL et al., Gender differences in the management and clinical outcome of stable angina. Circulation 2006;113:490-8; Khane RS, Surdi AD, Bhatkar RS, Changes in ECG pattern with advancing age. J Basic Clin Physiol Pharmacol 2011;22:97-101; Kuo TB, Lin T, Yang CC, Li CL, Chen CF, Chou P, Effect of aging on gender differences in neural control of heart rate. Am J Physiol 1999;277:H2233-9, Stramba-Badiale M, Locati EH, Martinelli A, Courville J, Schwartz PJ, Gender and the relationship between ventricular repolarization and cardiac cycle length during 24-h Holter recordings. Eur Heart J 1997;18:1000-6). During the training process, the network integrated and accounted for age and sex characteristics so as not to affect its discriminative ability to identify ventricular dysfunction. This ability is considered to be inherent to neural network selection.

[0101] The specific ECG characteristics used by unsupervised convolutional neural networks to classify whether an individual has low ejection fraction (EF) or not are unknown due to the nature of neural networks. However, by training the network in a large cohort of approximately 45,000 ECG and EF data pairs, as demonstrated by an AUC of 0.93 when tested in a population of 52,870 individuals, the network was exposed to a sufficient number of ECG variants to reliably classify individuals with low ejection fraction. In contrast to traditional applications of neural networks in medicine to mimic human skills, such as identifying mammogram lesions (Salazar-Licea LA, Pedraza-Ortega JC, Pastrana-Palma A, Aceves-Fernandez MA, Location of mammograms ROI's and reduction of false-positive. Comput Methods Programs Biomed 2017;143:97-111), this study expands the use of AI and extends its capabilities beyond human skills.

[0102] A key characteristic of the inventors' network is its use of the 12-lead ECG, an inexpensive and standard ubiquitous test, as its input. In many rural areas of the United States (Gruca TS, Pyo TH, Nelson GC, Providing Cardiology Care in Rural Areas Through Visiting Consultant Clinics. J Am Heart Assoc 2016;5) and in developing countries, access to cardiac care and imaging is limited.The availability of portable, inexpensive testing for ventricular systolic dysfunction enables the optimal use of limited imaging resources, allowing individuals to benefit from proven treatments such as beta-adrenergic blockers, angiotensin receptor blockers, and, where available, implantable devices (defibrillators and cardiac resynchronization systems). (Yancy CW, Jessup M, Bozkurt B et al., 2013 ACCF / AHA guideline for the management of heart failure: a report of the American College of Cardiology Foundation / American Heart Association Task Force on Practice Guidelines. J Am Coll Cardiol 2013;62:e147-239; Yancy CW, Jessup M, Bozkurt B et al., 2017 ACC / AHA / HFSA Focused Update of the 2013 ACCF / AHA Guideline for the Management of Heart Failure: A Report of the American College of Cardiology / American Heart Association Task Force on Clinical Practice Guidelines and the Heart Failure Society of America. J Am Coll Cardiol 2017;70:776-803).With the emergence of smartphone-compatible electrodes (Yasin OZ, Attia Z, Dillon JJ et al., Noninvasive blood potassium measurement using signal-processed, single-lead ECG acquired from a handheld smartphone. J Electrocardiol 2017;50:620-5), mobile applications enable use in resource-limited areas. Furthermore, the software-based nature of the network test "sample" conducted in this study allows for continuous feedback and improvement, enabling the rapid distribution of system improvements.

[0103] In this study, based on the well-established results and therapeutic impact of this value, an EF of 35% or less was selected as the detection threshold (Yancy CW, Jessup M, Bozkurt B, et al., 2017 ACC / AHA / HFSA Focused Update of the 2013 ACCF / AHA Guideline for the Management of Heart Failure: A Report of the American College of Cardiology / American Heart Association Task Force on Clinical Practice Guidelines and the Heart Failure Society of America. J Am Coll Cardiol 2017;70:776-803). However, the identification of EF below 50% is also clinically significant. Therefore, 45% of the “false positive” value is actually clinically meaningful (Figure 15). In addition, the network can be easily configured to use a different detection threshold instead, e.g., identification of EF < 40%. ECG-TTE pairs were not acquired simultaneously. However, this study includes over 100,000 ECG-TTE pairs datasets, and since over 89% of TTEs are performed within 24 hours of the ECG, the likelihood of errors related to time delays is small. Specifically, the convolutional neural network in this study was trained to detect low EF, i.e., diastolic dysfunction, and not heart failure. However, evidence-based therapies that reduce morbidity and mortality have been established for the treatment of low EF, which can sometimes be asymptomatic and therefore its detection is of paramount importance.

[0104] Although several implementation forms have been described in detail above, other modifications are also possible. Furthermore, the logical flow shown in the diagram does not require a specific order or sequence to achieve the desired result. Other steps can be provided, or steps can be omitted from the described flow; other components can be added to the described system, or removed from the described system. Therefore, other implementation forms are also within the scope of the following claims.

Claims

1. A system consisting of one or more computers receives ECG data describing the target electrocardiogram (ECG) during the period, The process of generating a prediction input by processing the aforementioned ECG data, The prediction input is provided to a survival estimation model generated by processing multiple training data to predict the survival estimation for the aforementioned target. Outputting a survival estimate for the aforementioned subject, Includes, The survival estimate is the future survival rate of the subject. The estimated ejection fraction characteristics generated based on the aforementioned prediction input are mapped to the future survival rate for the subject. Computerized implementation method.

2. The computer implementation method according to claim 1, wherein the survival estimation model is a regression model.

3. The computer implementation method according to claim 1, wherein the survival estimation model is a machine learning model.

4. The computer implementation method according to claim 2, wherein the survival estimation model is a neural network.

5. The computer implementation method according to claim 4, wherein the neural network is a feedforward neural network, a convolutional neural network, or a recurrent neural network.

6. The computer implementation method according to claim 1, wherein the predictive input is the value of one or more morphological features of the ECG.

7. The computer implementation method according to claim 6, wherein one or more morphological features include at least one of T-wave amplitude, P-wave amplitude, P-wave area, T-wave area, T-wave left tilt, T-wave right tilt, P-wave left tilt, P-wave right tilt, T-wave duration, P-wave duration, PR interval, QRS duration, QRS amplitude, QRS area, QRS energy, QRS peak-to-peak ratio, or QT segment length.

Images