Waveform-based hemodynamic instability warning

Abstract

A controller (150) for waveform-based hemodynamic instability warning includes a memory (151) storing instructions and a processor (152) executing the instructions. The controller (150) implements a process including: receiving an electrocardiogram (ECG) waveform via a first interface (153) connected to at least one ECG monitor (155) monitoring a patient; identifying (880) multiple heartbeats from the ECG waveform; separating the multiple heartbeats into first time windows; extracting features of the heartbeats in each of the first time windows as first extracted features for each first time window; generating generated features across a second time window including the multiple first time windows based on the first extracted features; applying trained artificial intelligence to the generated features; predicting hemodynamic instability for the patient based on the application of the trained artificial intelligence to the generated features; and outputting an alarm warning of the hemodynamic instability based on the predicted hemodynamic instability.

Description

[0001] Cross-reference to related applications

[0002] This application claims priority to jointly owned U.S. Provisional Application US 62 / 987602, filed March 10, 2020, pursuant to 35 USC §119(e). In particular, the entire disclosure of U.S. Provisional Application US 62 / 987602 is incorporated herein by reference. Background Technology

[0003] Hemodynamic instability (HI) is a condition that can be defined as the inability of blood to pass through organs and tissues, resulting in an inability to meet metabolic demands. Hemodynamic instability reflects problems in the circulatory system and can lead to cell dysfunction and death. Critically ill patients with impaired cardiac function are at high risk of circulatory failure and hemodynamic instability. In intensive care settings (e.g., intensive care unit (ICU)), intervention for hemodynamically unstable patients often involves fluid resuscitation to increase preload, administration of vasopressors to increase peripheral resistance (afterload) to maintain systemic blood pressure, and / or administration of inotropic agents to increase cardiac contractility.

[0004] Previous work on predicting hemodynamic instability has typically relied on nurse-drawn vital signs (e.g., temperature, pulse rate, respiratory rate, and / or blood pressure) and clinical measurements used to classify ICU stay durations as stable or unstable. In one study predicting hemodynamic instability, applying nurse-drawn vital signs and clinical measurements to a model yielded an area under the receiver operating characteristic (AUROC) of 0.82. In another study predicting hemodynamic instability, an AUROC of 0.87 was achieved by excluding missing data from the considerations. Another study predicting surgical ICU bleeding added bedside monitor vital signs, yielding an AUROC of 0.92. Bedside monitor vital signs were considered to be obtained more frequently than nurse-drawn vital signs. Yet another study predicting hemodynamic instability used ECG waveforms, but was conducted only in a simulated environment with healthy volunteers, resulting in AUROCs ranging from 0.86 to 0.88.

[0005] Continuous arterial blood pressure (ABP) waveforms carry relevant hemodynamic information. Arterial blood pressure has been used in ICU settings to monitor cardiovascular dysfunction and response to various interventions. The two main components of an individual ABP pulse cycle are systole and diastole, separated by the dicrotic notch. Aortic valve closure is graphically represented by the dicrotic notch in the ABP pulse cycle. Informational features, such as cardiac output (CO), as a measure of cardiac performance, can be estimated from the ABP waveform. Cardiac output is a function of heart rate, preload, contractility, and afterload. Other measures, such as the maximum value of the first derivative of the ABP pulse cycle, serve as indirect measures of cardiac contractility. ABP waveforms have been used to predict hypotensive episodes in ICU settings. For example, in a study predicting hypotensive events within the next 15 minutes, features extracted from the ABP waveform yielded AUROC values ​​ranging from 0.91 to 0.97. Summary of the Invention

[0006] According to one aspect of this disclosure, an apparatus includes a first interface, a memory, and a processor. The first interface is connected to at least one electrocardiogram (ECG) monitor for monitoring a patient. The memory stores instructions. The processor executes the instructions. When executed by the processor, the instructions cause the apparatus to identify a plurality of heartbeats from ECG waveforms received from the at least one ECG monitor via the first interface; separate the plurality of heartbeats into first time windows; and extract features of the heartbeats in each of the first time windows as a first extracted feature for each first time window. The instructions also cause the apparatus to: generate generated features across a second time window including the plurality of first time windows based on the first extracted features; apply trained artificial intelligence to the generated features; predict hemodynamic instability for the patient based on the application of the trained artificial intelligence to the generated features; and output an alarm warning of the hemodynamic instability based on the predicted hemodynamic instability.

[0007] According to another aspect of this disclosure, a method includes: receiving a plurality of electrocardiogram (ECG) waveforms via a first interface connected to at least one ECG monitor interface for monitoring a patient; identifying a plurality of heartbeats from the ECG waveforms; separating the plurality of heartbeats into first time windows; and extracting features of the heartbeats in each of the first time windows as a first extracted feature for each first time window. The method further includes: generating generated features across a second time window including the plurality of first time windows based on the first extracted features; applying trained artificial intelligence to the generated features; predicting hemodynamic instability for the patient based on the application of the trained artificial intelligence to the generated features; and outputting an alarm warning of the hemodynamic instability based on the predicted hemodynamic instability.

[0008] According to another aspect of this disclosure, a tangible non-transient computer-readable storage medium stores a computer program. When run by a processor, the computer program causes a system including the tangible non-transient computer-readable storage medium to: identify a plurality of heartbeats from electrocardiogram waveforms received from at least one electrocardiogram monitor via a first interface; separate the plurality of heartbeats into first time windows; and extract features of the heartbeats in each of the first time windows as a first extracted feature for each first time window. The computer program further causes the system to: generate generated features across a second time window including a plurality of the first time windows based on the first extracted features; apply trained artificial intelligence to the generated features; predict hemodynamic instability for a patient based on the application of the trained artificial intelligence to the generated features; and output an alarm warning of the hemodynamic instability based on the predicted hemodynamic instability. Attached Figure Description

[0009] When with attachment Figure 1 The exemplary embodiments will be best understood by reading the following description. It should be emphasized that the various features are not necessarily drawn to scale. In fact, dimensions may be increased or decreased arbitrarily for clarity of discussion. Wherever applicable and useful, the same reference numerals refer to the same elements.

[0010] Figure 1 The illustration shows a system for waveform-based hemodynamic instability warning according to a representative embodiment.

[0011] Figure 2 The illustration shows a processing pipeline and extracted features for waveform-based hemodynamic instability warnings according to a representative embodiment.

[0012] Figure 3The illustration depicts a method for waveform-based hemodynamic instability warning according to a representative embodiment.

[0013] Figure 4 The illustration shows a computer system for waveform-based hemodynamic instability warning according to a representative embodiment. Detailed Implementation

[0014] In the following detailed description, representative embodiments with specific details disclosed are set forth for purposes of explanation and not limitation, in order to provide a thorough understanding of embodiments according to this teaching. Descriptions of known systems, apparatuses, materials, methods of operation, and methods of manufacture may be omitted to avoid obscuring the description of representative embodiments. Nevertheless, systems, apparatuses, materials, and methods within the scope of this teaching are all within the capabilities of those skilled in the art and can be used according to representative embodiments. It should be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. The defined terms have meanings other than their scientific and technical meanings as commonly understood and accepted in the art of this teaching.

[0015] It should be understood that although the terms first, second, third, etc., may be used herein to describe various elements or components, these elements or components should not be limited by these terms. These terms are only used to distinguish one element or component from another. Therefore, the first element or component discussed below may also be referred to as the second element or component without departing from the teachings of the inventive concept.

[0016] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. The singular forms of the terms “a,” “an,” and “the” as used in the specification and claims are intended to include both the singular and plural forms unless the context clearly indicates otherwise. Additionally, when used herein, the terms “comprising” and / or “including” and / or similar terms specify the presence of the stated features, elements, and / or components, but do not exclude the presence or addition of one or more other features, elements, components, and / or groups thereof. The term “and / or” as used herein includes any and all combinations of one or more of the associated items listed.

[0017] Unless otherwise stated, when an element or component is said to be "connected to," "coupled to," or "adjacent to" another element or component, it should be understood that the element or component can be directly connected to or coupled to the other element or component, and that intermediate elements or components may be present. That is, these and similar terms cover situations where one or more intermediate elements or components may be used to connect two elements or components. However, when an element or component is said to be "directly connected to" another element or component, this only covers situations where two elements or components are connected to each other without any intermediary or intermediate elements or components.

[0018] This disclosure is therefore intended to provide one or more advantages specifically indicated below through its various aspects, embodiments, and / or particular features or sub-components. For purposes of explanation and not limitation, exemplary embodiments with specific details disclosed are set forth in order to provide a thorough understanding of embodiments pursuant to these teachings. However, other embodiments consistent with this disclosure but departing from the specific details disclosed herein remain within the scope of the claims. Furthermore, descriptions of well-known apparatuses and methods may be omitted to avoid obscuring the description of exemplary embodiments. Such methods and apparatuses are within the scope of this disclosure.

[0019] As described herein, ECG and / or ABP waveforms can be used to predict hemodynamic instability, whether in the absence of nurse-drawn vital signs and clinical measurements or in combination with nurse-drawn vital signs and clinical measurements. According to the embodiments described herein, when a baseline model for predicting hemodynamic instability is built using only clinical measurements and nurse-drawn vital signs, the AUROC is improved by 5% when features extracted from ECG and ABP waveforms are added, while the area under the precision-recall curve (AUPRC) is improved by 9%. When features extracted from ECG and ABP waveforms are used alone without any clinical measurements or nurse-drawn vital signs, a 4% improvement in AUROC and AUPRC is still observed compared to baseline results.

[0020] Figure 1 The illustration shows a system 100 for waveform-based hemodynamic instability warning according to a representative embodiment.

[0021] Figure 1 System 100 is a system for waveform-based hemodynamic instability warning and includes components that can be provided together or distributed. System 100 includes one or more ECG monitors 101, one or more ABP monitors 102, a workstation 140, a monitor 155, and an artificial intelligence controller 180. An artificial intelligence training system 190 provides trained artificial intelligence to the artificial intelligence controller 180.

[0022] One or more ECG monitors 101 are electrocardiogram (ECG) monitors. An ECG monitor monitors the electrical activity of the heart by measuring the current flowing through it. Each ECG monitor can record the intensity and timing of the signal in a graph and may include one or more electrodes attached to the patient's body via patches and wires that transmit the ECG trace signal to a receiver on workstation 140. The signals recorded by the one or more ECG monitors 101 may include P waves, QRS complexes, and T waves.

[0023] One or more ABP monitors 102 are arterial blood pressure monitors. One or more ABP monitors 102 monitor a patient's arterial blood pressure via a pressure transducer. The pressure transducer can be an external pressure sensor connected to an arterial vessel. An external pressure sensor is typically connected to the arterial vessel in a radial position via a fluid-filled catheter that enters the arterial vessel. Alternatively, the pressure transducer can be an internal pressure sensor mounted at the tip of the catheter that enters the arterial vessel. Continuous ABP signal waveforms can be recorded, and arterial blood pressure measurements can be measured using one or more ABP monitors 102. Arterial blood pressure measurements that can be measured using one or more ABP monitors 102 include systolic blood pressure (SPB), diastolic blood pressure (SBP), and mean arterial pressure (MAP). One or more ABP monitors 102 are connected to a receiver on workstation 140.

[0024] Workstation 140 includes a controller 150, a first interface 153, a second interface 154, and a touch panel 156. The controller 150 includes a memory 151 for storing instructions and a processor 152 for executing instructions. The controller 150 controls and implements some or all aspects of the methods attributable to workstation 140 as described herein. The first interface 153 connects one or more ECG monitors 101 to workstation 140. The second interface 154 connects one or more ABP monitors 102 to workstation 140. That is, the second interface 154 interfaces with one or more arterial blood pressure monitors. The first interface 153 and the second interface 154 may be ports, adapters, or other types of wired or wireless interfaces for sending and receiving data. The touch panel 156 may include a touch interface that accepts input via touch (e.g., by direct touch or via a mouse, keyboard, or other hand input mechanism). The touch panel 156 may also include a visual display for displaying the touch input, allowing the user to confirm the input. Workstation 140 may also include one or more other input interfaces. Workstation 140 may include one or more other input interfaces (not shown) that may include ports, disk drives, wireless antennas, or other types of receiver circuitry. These other input interfaces may also connect other user interfaces (e.g., mouse, keyboard, microphone, video camera, touchscreen display, or other components or parts) to workstation 140.

[0025] Monitor 155 is a visual electronic monitor that displays images and data. Monitor 155 may be a computer monitor, a display on a mobile device, a television, an electronic whiteboard, or another screen configured to display electronic images. Monitor 155 may also include one or more input interfaces, such as those mentioned above that allow connection of other components or parts to workstation 140, and a touchscreen that enables direct input via touch.

[0026] The AI ​​controller 180 includes a memory 181 for storing instructions and a processor 182 for executing instructions. The AI ​​controller 180 can dynamically apply trained artificial intelligence to data received from the controller 150 based on ECG signals from one or more ECG monitors 101 and ABP signals from one or more ABP monitors 102. In some embodiments, the AI ​​controller 180 is implemented as a component of workstation 140. In alternative embodiments, the controller 150 and the AI ​​controller 180 are implemented using the same memory and processor, rather than using different memories and processors.

[0027] The artificial intelligence training system 190 includes a memory 191 for storing instructions and a processor 192 for executing instructions. The AI ​​training system 190 can train the AI ​​based on datasets from many different instances of patient monitoring to learn the optimal correlation between input data and the resulting hemodynamic instability. Datasets can be provided from databases such as the Intensive Care Medical Information Marketplace (MIMIC) III database, described elsewhere in this document. As a result, the trained AI provided by the AI ​​training system 190 can be used to predict hemodynamic instability in the manner described herein and ultimately provide alerts to elicit interventions, thereby avoiding, managing, or otherwise resolving the predicted hemodynamic instability. Training can be based on analyses of thousands of patients, including hundreds of patients with hemodynamic instability. Training can involve binary classification and can improve AUROC by 5% or more when considering features from both ECG and ABP waveforms, as well as vital signs from laboratory results and nurse plots. When considering only features from ECG and ABP waveforms, training can also improve AUROC by 4% or more.

[0028] Figure 2 The illustration shows a processing pipeline and extracted features for waveform-based hemodynamic instability warnings according to a representative embodiment.

[0029] exist Figure 2 In the upper left corner, the ABP processing pipeline is end-to-end and includes the following three steps: (1) heartbeat segmentation, (2) ABP feature extraction, and (3) signal anomaly detection. The end-to-end ABP processing pipeline is used to extract features from the ABP waveform. The upper left corner ECG pipeline is also end-to-end and also includes the following three steps: (1) heartbeat segmentation and labeling, (2) signal quality metrics, and (3) ECG feature extraction. Heartbeat segmentation in the ECG pipeline may include: detecting QRS complexes and then classifying the heartbeats. Signal quality metrics in the ECG pipeline may include signal quality assessment. ECG waveforms in the ECG pipeline may be derived from the MIMIC III database and may be recorded at frequencies of 125 Hz or 250 Hz. ECG waveforms may be upsampled from 125 Hz to 250 Hz to improve the temporal resolution and detectability of QRS complexes. Figure 2 The end-to-end ECG processing pipeline and ABP processing pipeline in the ICU use objective metrics of waveform signals commonly available in the ICU environment without requiring any additional special equipment.

[0030] exist Figure 2The right panel (A) shows the vital signs plotted by the nurse, and the right panel (B) shows the vital signs from the bedside monitor. Data from the ECG pipeline is shown in the right panel (C) ECG waveform HR (5 minutes), which is the 5-minute mean heart rate (HR) extracted from the ECG waveform. Data from the ABP pipeline is shown in the right panel (D) ABP waveform SBP (*), DBP (*), MAP (*) (5 minutes), which are the 5-minute mean systolic blood pressure (SBP), 5-minute mean diastolic blood pressure (DBP), and 5-minute mean arterial pressure (MAP), respectively, extracted from the ABP waveform. Figure 2 middle, Figure 4 Features in (A), (B), (C), and (D) were used for the sample patients. The feature for the ECG waveform shown in (C) was plotted based on the heartbeats divided into a first time window of 5 minutes. The feature shown in (D) was plotted based on the arterial blood pressure wave divided into the pulse cycle (which was then also divided into a first time window of 5 minutes).

[0031] Figure 2 The 6-hour window in the diagram is a second time window superimposed on (A), (B), (C), and (D). This 6-hour window ends at least one hour before the intervention time, as shown by the vertical lines superimposed on (A), (B), (C), and (D). This 6-hour window comprises 72 first time windows within the first time window, each first time window being 5 minutes for the ECG waveform in (C) and for the ABP waveform in (D). Features derived from the 6-hour windows in (A), (B), (C), and (D) are used in the hemodynamic instability prediction model in the lower left to classify stable and unstable patients. The derived features are derived from the time series data in (A), (B), (C), and (D), and therefore include features derived from the vital signs from the nurse plot from (A), features derived from the vital signs from the bedside monitor from (B), features derived from the segmented ECG waveform in (C), and features derived from the ABP waveform in (D).

[0032] Figure 2The vital signs shown in the nurse's drawing in (A) and the bedside monitor vital signs shown in (B) include heart rate (HR), systolic blood pressure (SBP), and diastolic blood pressure (DBP). Clinical measurements include many features extracted from laboratory test results and ventilator readings, including arterial alkalosis, aspartate aminotransferase (AST), bands, basophils, calcium, CO2, creatinine, eosinophils (EOS), fractional oxygen inhaled (FIO2), glucose, hematocrit, hemoglobin, ionized calcium, lactate, magnesium, mean airway pressure, partial pressure of carbon dioxide (PaCO2), peak inspiratory pressure, potassium, partial thromboplastin time (PTT), oxygen saturation (SaO2), sodium, bilirubin, white blood cell count, blood urea nitrogen (BUN), and central venous pressure. Clinical measurements may also include the patient's temperature and age at admission. Clinical measurements may also include those calculated as... Mean arterial pressure (MAP) and calculated as The shock index (SI).

[0033] Figure 3 The illustration depicts a method for waveform-based hemodynamic instability warning according to a representative embodiment.

[0034] Figure 3 The diagram illustrates a waveform-based hemodynamic instability warning process according to a representative embodiment. Figure 3 The method can be performed by a single device (e.g., controller 150 or workstation 140), a single system (e.g., system 100), by a single entity or on behalf of a single entity, or by a distributed device, a distributed system, or by multiple entities or on behalf of multiple entities. In some embodiments, when... Figure 1 When workstation 140 is integrated with or includes AI controller 180, workstation 140 can perform... Figure 3 The method; or, if workstation 140 and Figure 1 If the AI ​​controller 180 is provided separately, then the combination of workstation 140 and AI controller 180 can perform... Figure 3 The method.

[0035] At S305, Figure 3 The method includes receiving electrocardiogram (ECG) waveforms. This can be achieved from workstation 140 via... Figure 1 The first interface 153 receives electrocardiogram waves from one or more ECG monitors 101.

[0036] At S310, Figure 3 The method includes receiving arterial blood pressure waves. This can be achieved from workstation 140 via... Figure 1The second interface 154 receives arterial blood pressure waves from one or more ABP monitors 102.

[0037] At S315, Figure 3 The method includes identifying the heartbeat from an electrocardiogram (ECG) waveform. The heartbeat can be identified from the ECG waveform received at S305, and can be determined based on the ECG waveform received at S305. Figure 1 The controller 150 performs analysis to identify the heartbeat.

[0038] At S325, Figure 3 The method includes separating heartbeats into first-time windows. This can be achieved by... Figure 1 The controller 150 divides the heartbeats into first time windows. For example, each first time window may include a 5-minute interval, and as described herein, each 6-hour second time window may include 72 first time windows within the first time windows.

[0039] At S330, Figure 3 The method involves segmenting the arterial blood pressure wave into the pulse cycle. For example, Figure 1 The controller 150 can segment arterial blood pressure waves into individual pulse cycles. A pulse detection algorithm that detects the start of the arterial blood pressure pulse can be used to segment continuous arterial blood pressure waveforms into individual pulse cycles.

[0040] Additionally, although not shown, the pulse cycle can be separated into a first interval. For example, the heart rate and pulse cycle of the arterial blood pressure wave can be separated into the same first time window.

[0041] At S335, Figure 3 The methods include: if the heartbeat is noisy, excluding it from further processing. The exclusion at S335 can be achieved by... Figure 1The controller 150 in the system performs the operation and can involve one or more heartbeats. Correct extraction of ECG features depends on the quality of the ECG signal, and many factors affect this quality, including patient movement, poor electrode contact, incorrect electrode placement, and electrical interference. ECG quality assessment measures can be based on a combination of existing criteria and take into account various types of noise present in the data. Eight types of noise can be considered and filtered by a noise filter, including low-frequency noise, high-frequency noise, flat-line ratio, peak-to-peak amplitude, spike indicators with slope, amplitude saturation, electric field noise, and the ratio of outliers detected per-beat (QRS to QRS, or RR) intervals. For each type of noise, an empirically determined threshold can be tuned using ECG data from different patient groups. The total signal quality index (SQI) can be calculated using eight sub-indices. Signal quality can be assessed second by second. For a one-second ECG segment, the total SQI receives a value of 1 (acceptable) only when all sub-indices are below the tuned threshold.

[0042] At S340, Figure 3 The method includes excluding the ABP pulse cycle from further processing if the pulse cycle is abnormally noisy. That is, Figure 3 The method may include excluding at least one individual pulse cycle identified as abnormal from an application of trained artificial intelligence, based on identifying at least one individual pulse cycle as abnormal. The exclusion at S340 may be performed by… Figure 1 The controller 150 in the system is used for execution and may involve one pulse cycle or more pulse cycles. The Signal Abnormality Index (SAI) can be calculated on an individual arterial blood pressure pulse cycle to discard abnormal arterial blood pressure pulse cycles that are considered unusable. An ABP pulse cycle may be discarded when any of the following conditions are met: (1) SBP > 300 mmHg or DBP < 20 mmHg; (2) MAP < 30 mmHg or > 200 mmHg; (3) (pulse pressure) PP < 20 mmHg; (4) the sum of the negative slopes of the ABP pulse cycles is < -40 mmHg / 100 ms; (5) the difference between SBP and DBP between adjacent ABP cycles exceeds 20 mmHg; (6) the difference between the lengths of the ABP pulse cycles in seconds exceeds Seconds; (7) The order of SBP, diphtheria notch and DBP is inconsistent; (8) SBP <DBP。

[0043] At S345 Figure 3The method involves extracting features of heartbeats within a first time window. ECG waveforms are utilized by extracting heart rate at a higher temporal resolution and deriving heart rate variability (HRV) as an indirect measure of voluntary control and vascular tone. Examples of features extracted from heartbeats within the first time window can include averages calculated based on individual heartbeats, such as the average heart rate. Additionally, examples of features extracted from heartbeats within the first time window can include features across all heartbeats within the first time window, such as heart rate variability. ECG heartbeat detection and classification can be performed on each ECG lead in each patient using a commercially available (FDA-approved) ECG arrhythmia detection algorithm. For each detected heartbeat, the heartbeat location can be assigned to a QRS peak, and the heartbeat type can be assigned to one of the following predetermined labels based on the characteristics of each heartbeat: normal heartbeat (N), premature ventricular contractions (V), supraventricular premature contractions (S), paced heartbeat (P), problematic / unclassified heartbeat (Q), and learning heartbeat (L).

[0044] The following ECG features can be extracted / calculated for each ECG lead using only heartbeat locations and labels from segments with good signal quality. When ECG features are available from multiple leads, features can be selected from a single lead, for example, in the following order: II, V, MCL, aVF, III, aVR, aVL, V1, V3, V5, and V2. ECG leads can be ranked sequentially based on availability and clinical relevance.

[0045] Heart rate (HR) is the number of heartbeats per minute. Instantaneous heart rate can be calculated using the RR interval between two adjacent heartbeats (where the heartbeats are labeled N, V, S, P, Q, or L). The average of instantaneous HR values ​​can be obtained over a first time window of 5 minutes. HR values ​​can be averaged over a time series by adding time series from (A) nurse-drawn vital signs, (B) bedside monitor vital signs measured per minute, and (C) ECG waveforms.

[0046] Another extracted ECG feature can be heart rate asymmetry (HRA). HRA quantifies rapid accelerations and decelerations of heart rate over a period of time and concisely represents the imbalance as a single indicator.

[0047] Additional extracted ECG features can be heart rate variability (HRV). HRV is the variability within the time interval between adjacent normal heartbeats. Only heartbeats labeled "N" are used to calculate the HRV measure over a 5-minute time window. The HRV measure is grouped into the time domain, frequency domain, and nonlinear domain, as explained below. The time-domain HRV measure includes the standard deviation of the NN interval (SDNN), the root mean square of the difference between successive normal heartbeats (RMSSD), the standard deviation of the difference between successive NN intervals (SDSD), and the percentage of adjacent NN intervals that differ from each other by more than 20 milliseconds (pNN20) and 50 milliseconds (pNN50), respectively.

[0048] Additional extracted ECG features can be frequency-domain HRV measures, which utilize Fast Fourier Transform (FFT) over the NN interval time series to extract the contributions of different frequency components. The frequency components can be organized into four non-overlapping frequency bands: ultra-low frequency (0.0001 Hz to 0.003 Hz), extremely low frequency (0.004 to 0.04 Hz), low frequency (0.05 to 0.15 Hz), and high frequency (0.16 to 0.4 Hz). These frequency-domain measures may require computation over adjacent NN intervals without any gaps. Therefore, these measures can be computed only over the longest, continuous NN interval segment within a 5-minute window. For this segment, the NN intervals can be resampled at a sampling frequency of 4 Hz using cubic interpolation to obtain a regularly and frequently sampled time series. The average of the NN time series can be subtracted to eliminate the effect on baseline offset. The Welch periodogram can be programmed over this time series using a Hanning window. Because the Hanning window is set to a minimum length of 256 samples, NN time series with a length ≤255 samples can be discarded. Frequency domain metrics include power in four frequency bands, total power, normalized and relative power in the low-frequency (LF) band and high-frequency (HF) band, and... The ratio. The correlation between heart rate and HRV, calculated in a second time window of 6 hours, can also be added to the feature.

[0049] It can also include several nonlinear HRV measures. These measures can be computed on the Poincaré plot of the longest segment of the adjacent NN intervals, the same calculation as the frequency domain analysis above. An ellipse can be fitted to the points on the Poincaré plot, and the width (SD1) and length (SD2) of the ellipse can be computed. ratio.

[0050] The number of premature beats (PVCs) can be tracked by monitoring the count of ventricular premature contractions (PVCs) and supraventricular premature contractions within a 5-minute window. The frequency of PVCs is a predictor of heart failure and death derived from a decrease in left ventricular ejection fraction.

[0051] At S350, Figure 3 The method involves extracting features of arterial blood pressure within a first time window. ABP waveforms can be utilized by extracting blood pressure measurements at a relatively high temporal resolution and extracting ABP morphological features. These morphological features can be used to estimate CO, cardiac contractility, and other hemodynamic states.

[0052] Within each ABP pulse cycle, SBP can be calculated as the local maximum within the ABP pulse cycle, and DBP as the first point with zero slope when traversing the ABP pulse cycle from right to left. DBP can be calculated in this way, rather than using the minimum value. MAP and pulse pressure (PP) can be calculated based on SBP and DBP. Two methods can be used to locate the dicrotic notch: one is 0.3 × T, where T is the length of the ABP pulse cycle in seconds, and the other is the first zero-slope crossover point after SBP. Both methods locate the dicrotic notch at a point where there is no clear separation between systole and diastole. Therefore, the dicrotic notch is located at the first zero-slope crossover point after t, where t is chosen as a circle (0.3 × T). As demonstrated by visual evaluation, this combined method has fewer errors in locating the dicrotic notch. Additionally, if SBP, DBP, or the dicrotic notch is not found, the ABP pulse cycle can be excluded from the calculation at 340.

[0053] The derived morphological features include time to SBP, time to DBP, time from pulse onset to dicrotic notch, area under the systolic, diastolic, and complete ABP pulse cycle, and length of the ABP pulse cycle. Maximum slope (first derivative of the ABP pulse cycle). Indirect measures of cardiac contractility were calculated for use in titrating positive inotropic drugs. Two indirect measures of CO were added to these characteristics, as follows:

[0054]

[0055] Specifically, PP, SBP, DBP, and systolic area are extracted for each ABP pulse cycle. The heart rate that is temporally closest to the current ABP pulse cycle is used. Features are calculated for each ABP pulse cycle and averaged over a first time window of 5 minutes. Figure 2 In (D), the 5-minute average time series of SBP, DBP, and MAP were calculated based on the ABP waveform of the same patient.

[0056] At S357 Figure 3The method involves extracting features across a second time window. The extracted ECG and ABP waveform features are obtained as follows: Figure 2 The time series shown in (C) and (D) are examples. A second time window may include, for example... Figure 2 The time series shown contains multiple first time windows, and may include, for example, 72 first time windows totaling 6 hours. Features extracted across second time windows can be extracted by calculating the time series features, such as the mean, standard deviation, and slope of features extracted from the first time windows included in the second time window.

[0057] Derived features (DFs) that can be extracted across a second time window can be extracted from the time series of the first time window to capture relevant information related to hemodynamic instability. (See above.) Figure 2 As shown, only time series available within a 6-hour window are summarized at S357, rather than data from the 1-hour window prior to intervention, as this provides physicians with sufficient time for intervention. The mean, standard deviation, median, minimum, and maximum values ​​can be calculated for each time series within the 6-hour window to capture the distributional properties of the time series. The slope and intercept for each time series are calculated to capture trend information and baseline offset, respectively. For meaningful comparisons across patients, trend characteristics can be calculated with respect to time series subtracted from the mean. Finally, the slope and approximate entropy can be calculated to capture regularity and volatility in the time series. For approximate entropy and sample entropy, the pattern length for finding similarities is set to a default value of 2.

[0058] In this embodiment, multiple second time windows can be used, and these second time windows can overlap. For example, when the second time window is 6 hours, a different second time window can start every 30 minutes, so that, for example, the feature generated at S337 can be generated every 30 minutes.

[0059] At S360, Figure 3 The method involves applying trained artificial intelligence to extracted features. An AI training system 190 can train the trained AI and provide it to an AI controller 180. The AI ​​controller 180 can apply the trained AI, or the workstation 140 can apply the trained AI when the controller 150 provides the AI ​​controller 180.

[0060] At S370 Figure 3 Methods include predicting hemodynamic instability. For example, hemodynamic instability can be predicted one hour or more in advance. Predictions can include confidence levels, such as 30%, 60%, or 80%. Predictions can also specify a particular time or time period at which hemodynamic instability is predicted to begin or become detectable.

[0061] At S380, Figure 3 The method involves determining whether the predicted hemodynamic instability is above a threshold. If the predicted hemodynamic instability is not above the threshold (S380 = No), then... Figure 3 The process will return to S305 and S310. For example, the threshold could be 50% or 75%. In some embodiments, multiple thresholds can be used. For example, a first threshold of 25% could be considered a warning threshold at which clinical staff should prepare to take action in response to the prediction, while a second threshold of 50% could be considered an action threshold at which clinical staff should take action in response to the prediction. The alert could include an inference based on an estimated probability exceeding a predetermined threshold that the patient will enter a hemodynamically unstable phase.

[0062] If the predicted value for hemodynamic instability is higher than the threshold (S380 = Yes), then Figure 3 Methods include issuing alerts warning of hemodynamic instability. Other actions to be taken when the predicted value exceeds a threshold may include intervention or initiating preparation for intervention.

[0063] in addition, Figure 3 The method can utilize trained artificial intelligence. The training of this AI can be based on a large patient population with relevant data, such as patients with information stored in the MIMIC III database. The MIMIC III database contains electronic health records (EHRs) and waveform recordings of a matched subset of patients. EHRs include entries from various laboratory test results, medication administration, fluid intake, and vital signs plotted by nurses. Waveform recordings contain multi-parameter physiological signals, including ECG and ABP. The MIMIC III database also includes vital signs generated by bedside monitors, such as heart rate and systolic blood pressure (SBP) available per minute.

[0064] The MIMIC III database includes EHR data from 46,520 unique ICU patients. The MIMIC III database was matched with physiological waveforms and bedside monitor vital signs for a subset of 10,282 patients. The process of training the trained artificial intelligence may involve first identifying hemodynamically unstable patients who received either strong or weak interventions associated with hemodynamic instability. Strong interventions may include the use of vasopressors such as dobutamine, dopamine, epinephrine, norepinephrine, phenylephrine, vasopressin, and isoproterenol. Weak interventions may include the use of fluid therapy, packed red blood cells, and agents such as lidocaine. In the MIMIC III database, 15,713 patients could be identified as having one or more strong interventions, while 9,949 patients were identified as having one or more weak interventions. Patients with only weak interventions could be excluded from subsequent treatment. The remaining 20,858 patients who did not receive any interventions could be identified as stable patients. Stable and unstable patients with waveforms can be filtered, resulting in 4460 stable patients with waveforms and 3037 unstable patients with waveforms. The selection can also be limited to unstable patients who have any one of the following vital signs within a time window prior to their first major intervention since admission: laboratory, nurse-drawn vital signs, bedside monitor vital signs, ECG, or ABP. In some embodiments described herein, the second time window can be selected from 6 hours to 1 hour prior to the intervention. For stable patients, a 6-hour window can be pre-selected randomly. Stable and unstable patients with similar levels of care (e.g., at least one nurse-drawn HR and SBP within the 6-hour window) can be used as a training basis, and in one set of experiments, this led to the identification of 880 unstable patients and 2501 stable patients.

[0065] The problem of predicting hemodynamic stability in unstable patients can be treated as a binary classification problem. Gradient Boosting Tree (XGBoost) models can be used as boosting models with an overall ensemble of decision trees. In each iteration, decision trees can be added to the existing tree ensemble to correct errors from previous iterations, with the goal of minimizing the overall loss function. Implementations of Gradient Boosting Trees available in Python can be used. Hyperparameter tuning can be performed on regularization λ from 10⁻⁴ to 1, learning rates from 10⁻⁴ to 0.3, and maximum depth of each decision tree from 1 to 5. All other settings can be set to default values. Hierarchical 5-layer nested cross-validation can be performed on all patients. Hyperparameter tuning can be performed on four training layers, and testing can be performed on the extended fifth layer. The average AUROC and average AUPRC curves for the five test layers can be reported. The standard errors for the five test layers can also be reported.

[0066] The results of two experiments are reported. In Experiment I, the performance of adding features with higher temporal resolution was compared to the baseline model. The baseline model could only utilize the last entries of vital signs from clinical measurements and nurse plots (at the end of the second time window). The last entries in a 6-hour window following forward filling of vital signs from clinical measurements and nurse plots could be used, thus minimizing the number of missing features for each patient. This experiment included data from all stable and unstable patients. The experiment simulated a real-world ICU environment where the hemodynamic instability prediction model utilized information available in the ICU to improve predictive performance.

[0067] The baseline model (using clinical measurements and vital signs) can have an AUROC of 0.89 and an AUPRC of 0.79. The availability of these features is highly dependent on the caregiver and indirectly reflects clinician attention or other biases. For the baseline model, derived features from nurse-drawn vital signs (nurse DF) and bedside monitor vital signs (monitor DF) can be added. Adding derived features of vital signs to the baseline model improves the AUROC by 2%, while the change in AUPRC is even more significant. P-values ​​obtained from paired t-tests of AUROC performed on the baseline model from five levels can be reported. The performance improvement caused by including nurse-drawn vital signs shows a greater difference compared to bedside monitor vital signs. Next, higher temporal resolution features are added to the baseline model. Derived features from ABP waveforms and ECG waveforms are added, and features from ABP waveforms and ECG waveforms are combined. These high temporal resolution features can include basic vital signs as well as other ABP and ECG-specific features. When ECG and ABP features are included, AUROC achieves a 4% improvement and AUPRC achieves a 7% improvement compared to the baseline model.

[0068] In Experiment I, the relative contributions of the different feature groups were unknown because the availability of feature groups across patients was uneven. For example, only one-third of the patients had ABP waveforms, but almost all patients had ECG waveforms. To compare the relative performance of the different feature groups, Experiment II was conducted, in which only patients with data from all five feature groups within a 6-hour window were included, and each feature group was evaluated individually. This resulted in only 243 unstable patients and 230 stable patients. All other details of Experiment II were similar to those of Experiment I.

[0069] The final entry for vital signs using clinical measurements and nurse plots yielded an AUROC of 0.82, a 7% decrease compared to Experiment I, which used a larger number of patients. No significant improvement in AUROC was observed when using nurse plotted vital signs and monitored vital signs. Finally, when evaluating waveform features, an AUROC of 0.84 was obtained for ABP and 0.85 for ECG+ABP. Although performance decreased when using ECG features alone, this was not statistically significant (P=0.36). A slight improvement in performance was observed when using ABP alone compared to using ABP in combination with ECG. This suggests that the predictive model can utilize information available from ABP rather than ECG (e.g., cardiac output or systolic force). Furthermore, none of the performance metrics showed a significant difference from the baseline model. AUPRC followed a very similar trend but was better than AUROC because there were more positive examples (243 unstable patients) compared to negative examples. The results of this experiment indicate that even in the absence of clinical measurements and nurse records of vital signs, objective measurements (e.g., vital signs from bedside monitors and characteristics from waveforms) can carry relevant information related to hemodynamic instability.

[0070] As a final analysis, the clinical relevance of these features was explored, and the extent to which these waveform features represent the underlying systems (i.e., cardiac performance, voluntary tone, and vascular tone) was examined. Feature importance was examined based on the ranking of all five CV layers by the XGBoost model in Experiment II. A smaller cohort was chosen to examine feature importance because all patients had all five feature groups, and the proportion of stable to unstable patients was roughly equal (230 vs. 243). However, within each feature group, the availability of features influencing its importance could be unbalanced. To correct for this, the importance of each feature was subtracted from the importance of its missing patterns. First, a data matrix was created by setting features to 1 (1 when the feature is present, 0 otherwise)—this simply encodes the missing patterns. Second, this binary data matrix was used in the predictive model, and the feature importance of missing patterns was calculated. Following this, a corrected feature importance score is calculated within the range of "1.0" (where the feature is highly ranked and its missing pattern has little or no effect) to "0.0" (where the feature's importance is completely offset by its missing pattern) to "-1.0" (where the feature is not important, but its missing pattern carries information about stable patients compared to unstable patients).

[0071] Clinically measured characteristics include SBP, MAP, SI, hematocrit, hemoglobin, PaCO2, FiO2, alkali excess, and mean airway pressure, many of which are clinically relevant to hemodynamic monitoring. When using only the ABP waveform, systolic blood pressure, diastolic blood pressure, shock index, CO (Zander), dicrotic notch amplitude, ABP pulse cycle length, minimum MAP, and approximate entropy of the area under the systolic phase can also be used. When using only the ECG waveform, LF, ULF power, minimum LF, median VLF power, approximate entropy (its regularity in capturing the heart rate signal), and the Porta index can also be used. The LF component is generally accepted as an indirect measure of sympathetic and vagal nervous system activity, and heart rate has been observed to be influenced by both components. During hemodialysis-induced hypotension, there are significant differences in frequency domain HRV measurements (particularly VLF, LF, and HF) between stable and unstable patients.

[0072] Figure 4 The illustration shows a computer system for waveform-based hemodynamic instability warning according to a representative embodiment.

[0073] Figure 4 The illustration shows a computer system according to another representative embodiment, on which a method for waveform-based hemodynamic instability warning is implemented.

[0074] Figure 4 The computer system 400 illustrates a complete set of components for a communication device or computer device. However, the "controller" as described herein can be used in fewer than [number missing]. Figure 4 This set of components is used for implementation. For example, it can be implemented through a combination of memory and processor. Computer system 400 may include some or all of the components of one or more of the components of a system for waveform-based hemodynamic instability warning described herein, but any such device may not necessarily include one or more of the components described for computer system 400 and may include other components not described.

[0075] refer to Figure 4 The computer system 400 includes a set of software instructions that can be executed to cause the computer system 400 to perform some or all of the computer-based functions disclosed herein. The computer system 400 can operate as a standalone device or, for example, be connected to other computer systems or peripheral devices using a network 401. In embodiments, the computer system 400 performs logic processing based on digital signals received via an analog-to-digital converter.

[0076] In a networked deployment, computer system 400 operates as a server, or as a client computer in a server-client user network environment, or as a peer-to-peer (or distributed) computer system in a peer-to-peer (or distributed) network environment. Computer system 400 can also be implemented as various devices or incorporated into various devices, for example, Figure 1 The computer system 400 may include a workstation 140, an artificial intelligence controller 180, and / or an artificial intelligence training system 190, a fixed computer, a mobile computer, a personal computer (PC), a laptop computer, a tablet computer, or other machine capable of (sequentially or otherwise) running a set of software instructions specifying actions to be performed by that machine. The computer system 400 can be incorporated as a device or incorporated into a device, which in turn is included in an integrated system including additional devices. In embodiments, the computer system 400 can be implemented using electronic devices that provide voice, video, or data communication. Furthermore, although the computer system 400 is illustrated as a single system, the term "system" should also be considered to include any collection of systems or subsystems that individually or jointly run one or more sets of software instructions to perform one or more computer functions.

[0077] like Figure 4 As shown, the computer system 400 includes a processor 410. The processor 410 can be considered as... Figure 1 The processor 152 of the controller 150 in the middle Figure 1 The processor 182 and / or in the artificial intelligence controller 180 Figure 1A representative example of the processor 192 in the artificial intelligence training system 190 is shown. Processor 410 executes instructions to implement some or all aspects of the methods and processes described herein. Processor 410 is tangible and non-transient. As used herein, the term "non-transient" should not be interpreted as a perpetual state characteristic, but rather as a characteristic of a state that will persist for a period of time. The term "non-transient" specifically denies transient characteristics, such as carrier waves or signals or other forms of characteristics that exist only momentarily at any time and place. Processor 410 is an article of manufacture and / or a machine part. Processor 410 is configured to execute software instructions to perform the functions described in the various embodiments herein. Processor 410 may be a general-purpose processor or a part of an application-specific integrated circuit (ASIC). Processor 410 may also be a microprocessor, microcomputer, processor chip, controller, microcontroller, digital signal processor (DSP), state machine, or programmable logic device. Processor 410 may also be a logic circuit (including a programmable gate array (PGA) such as a field-programmable gate array (FPGA)) or another type of circuit comprising discrete gate and / or transistor logic units. Processor 410 may be a central processing unit (CPU), a graphics processing unit (GPU), or both. Additionally, any processor described herein may include multiple processors, parallel processors, or both. Multiple processors may be included in or coupled to a single device or multiple devices.

[0078] As used herein, the term "processor" encompasses any electronic component capable of running programs or machine-executable instructions. References to computing devices that include "processor" should be interpreted as including more than one processor or processing core, as is the case in a multi-core processor. A processor can also refer to a collection of processors within a single computer system or distributed across multiple computer systems. The term computing device should also be interpreted as including a collection or network of computing devices, each including one or more processors. A program has software instructions that are executed by one or more processors, which may be located within the same computing device or distributed across multiple computing devices.

[0079] Computer system 400 also includes main memory 420 and static memory 430, wherein the memories in computer system 400 communicate with each other and with processor 410 via bus 408. One or both of main memory 420 and static memory 430 can be considered as... Figure 1 The memory 151 of the controller 150 in the middle Figure 1 The memory 181 and / or of the artificial intelligence controller 180 in the middle Figure 1A representative example of memory 191 in an artificial intelligence training system 190 is shown. Main memory 420 and static memory 430 may store instructions for implementing some or all aspects of the methods and processes described herein. The memory described herein is a tangible, non-transient, computer-readable storage medium for storing data and executable software instructions, and is non-transient during the time the software instructions are stored therein. As used herein, the term "non-transient" should not be construed as a perpetual state characteristic, but rather as a characteristic of a state that will last for a period of time. The term "non-transient" specifically denies transient characteristics, such as carrier waves or signals or other forms of characteristics that exist only momentarily at any time and place. Main memory 420 and static memory 430 are articles of manufacture and / or machine parts. Main memory 420 and static memory 430 are computer-readable media from which a computer (e.g., processor 410) can read data and executable software instructions. Each of the main memory 420 and static memory 430 may be implemented as one or more of the following: random access memory (RAM), read-only memory (ROM), flash memory, electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, magnetic tapes, compact disc read-only memory (CD-ROM), digital versatile disks (DVDs), floppy disks, Blu-ray discs, or any other form of storage medium known in the art. The memory may be volatile or non-volatile, secure and / or encrypted, insecure and / or unencrypted.

[0080] “Memory” is an example of a computer-readable storage medium. Computer memory is any memory that a processor can directly access. Examples of computer memory include, but are not limited to, RAM, registers, and register files. The reference to “computer memory” or “memory” should be interpreted as potentially referring to multiple memories. Memory can be, for example, multiple memories within the same computer system. Memory can also be multiple memories distributed across multiple computer systems or computing devices.

[0081] As shown in the figure, the computer system 400 may also include, for example, a video display unit 450 (e.g., a liquid crystal display (LCD), an organic light-emitting diode (OLED), a flat panel display, a solid-state display, or a cathode ray tube (CRT)). Additionally, the computer system 400 includes an input device 460 (e.g., a keyboard / virtual keyboard, a touch-sensitive input screen, or a voice input unit with voice recognition) and a cursor control device 470 (e.g., a mouse, a touch-sensitive input screen, or a pad). The computer system 400 may also optionally include a disk drive unit 480, a signal generation device 490 (e.g., a speaker or a remote control), and a network interface device 440.

[0082] In an embodiment, such as Figure 4 As shown, the disk drive unit 480 includes a computer-readable medium 482 in which one or more sets of software instructions 484 (software) are embedded. These software instructions 484, to be executed by the processor 410, are read from the computer-readable medium 482. Additionally, the software instructions 484, when executed by the processor 410, perform one or more steps of the methods and processes described herein. In embodiments, the software instructions 484 reside wholly or partially in main memory 420, static memory 430, and / or reside wholly or partially in the processor 410 during operation by the computer system 400. Furthermore, the computer-readable medium 482 may include the software instructions 484 or receive and execute the software instructions 484 in response to a propagated signal, causing a device connected to the network 401 to transmit voice, video, or data on the network 401. The software instructions 484 may be sent or received on the network 401 via a network interface device 440.

[0083] In embodiments, dedicated hardware implementations (e.g., application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), programmable logic arrays, and other hardware components) are constructed to implement one or more of the methods described herein. One or more embodiments described herein may use two or more specific interconnected hardware modules or devices to implement functionality, having associated control and data signals capable of communicating between and through these modules. Therefore, this disclosure covers software, firmware, and hardware implementations. Nothing in this application should be construed as being implemented solely using software and not using hardware such as tangible non-transient processors and / or memory.

[0084] According to various embodiments of this disclosure, the methods described herein can be implemented using a hardware computer system running software programs. Additionally, in exemplary non-limiting embodiments, implementations can include distributed processing, component / object distributed processing, and parallel processing. Virtual computer system processing can implement one or more of the methods or functions described herein, and the processors described herein can be used to support virtual processing environments.

[0085] Based on the experiments described herein, adding ABP and ECG waveforms to clinical measurements and nurse plots of vital signs yielded an AUROC of 0.93, while using ABP and ECG waveforms alone yielded an AUROC of 0.85. Therefore, waveform-based hemodynamic instability warnings help ensure adequate tissue perfusion and early detection of inadequate perfusion and end-organ dysfunction. ECG and / or ABP waveforms can be used alone, together, or in conjunction with laboratory tests and nurse plots of vital signs to predict hemodynamically unstable patients. The end-to-end waveform processing pipeline described herein may include heartbeat segmentation, signal quality assessment, and feature extraction, resulting in a usable prediction of hemodynamic instability. Predicting patients who may have hemodynamic instability allows for early intervention and management, which generally leads to better patient outcomes. Nevertheless, waveform-based hemodynamic instability warnings are not limited to the application of the specific details described herein, but are applicable to other embodiments where alternative details may be feasible.

[0086] While waveform-based hemodynamic instability warnings have been described with reference to several exemplary embodiments, it should be understood that the language used is descriptive and illustrative, not limiting. Changes, such as those presently described and modified, may be made within the scope and spirit of the claims without departing from the scope and spirit of waveform-based hemodynamic instability warnings in all their aspects. Although waveform-based hemodynamic instability warnings have been described with reference to specific means, materials, and embodiments, they are not intended to be limited to the disclosed details; rather, they are extended to all functionally equivalent structures, methods, and uses as described within the scope of the claims.

[0087] The embodiments described herein are intended to provide a general understanding of the structure of various embodiments. These descriptions are not intended to constitute a complete description of all elements and features of the disclosure described herein. Many other embodiments will be apparent to those skilled in the art upon review of this disclosure. Other embodiments can be utilized and derived from this disclosure, allowing structural and logical substitutions and changes to be made without departing from the scope of this disclosure. Furthermore, these illustrations are representative only and may not be drawn to scale. Some scales in the illustrations may be enlarged, while others may be minimized. Therefore, this disclosure and the accompanying drawings should be considered illustrative rather than restrictive.

[0088] In this document, for convenience only, the term "invention" may be used individually and / or collectively to refer to one or more embodiments of this disclosure, without intending to limit the scope of this application to any particular invention or inventive concept. Furthermore, while specific embodiments have been illustrated and described herein, it should be understood that any subsequent arrangements designed to achieve the same or similar purpose with respect to the specific embodiments shown may be substituted. This disclosure is intended to cover any and all subsequent modifications or variations of the various embodiments. After reading the specification, combinations of the above embodiments and other embodiments not specifically described herein will be apparent to those skilled in the art.

[0089] This abstract of disclosure is provided in accordance with 37 C. FR § 1.72(b) and is to be understood at the time of filing not to be used for interpreting or limiting the scope or meaning of the claims. Additionally, in the foregoing detailed description, various features may be grouped together or described in a single embodiment for the purpose of simplification. This disclosure should not be construed as reflecting an intention that the claimed embodiments require more features than expressly recited in each claim. Rather, as reflected in the claims, the inventive subject matter may refer to all features of fewer than any of the disclosed embodiments. Therefore, the claims are incorporated into the detailed description, with each claim independently defining a separately claimed subject matter.

[0090] The foregoing description of the disclosed embodiments is provided to enable any person skilled in the art to practice the concepts described in this disclosure. Therefore, the subject matter disclosed above should be considered illustrative rather than restrictive, and the claims are intended to cover all such modifications, enhancements, and other embodiments falling within the true spirit and scope of this disclosure. Accordingly, to the fullest extent permitted by law, the scope of this disclosure will be determined by the broadest permissible interpretation of the claims and their equivalents, and should not be limited by the foregoing detailed description.

Claims

1. A device for waveform-based hemodynamic instability warning, comprising: The first interface (153) is connected to at least one electrocardiogram monitor (155) interface for monitoring the patient; Memory (151), its storage instructions, and Processor (152), which executes the instructions, wherein, when executed by processor (152), the instructions cause the device to: Multiple heartbeats are identified from the electrocardiogram waves received from the at least one electrocardiogram monitor (155) via the first interface (153); The multiple heartbeats are separated into a first time window; The heartbeat features in each of the first time windows are extracted and used as the first extracted features for each first time window; Generate features for a second time window that includes multiple first time windows, based on the first extracted features; The trained artificial intelligence is applied to the generated features; Based on applying the trained artificial intelligence to the generated features to predict hemodynamic instability for the patient; and An alert warning of the hemodynamic instability is output based on the prediction of the hemodynamic instability.

2. The apparatus according to claim 1, further comprising: A second interface (154) is connected to an interface of a patient's arterial blood pressure monitor (155), wherein, when run by the processor (152), the instructions also cause the device to: The arterial blood pressure wave received from the arterial blood pressure monitor (155) via the second interface (154) is segmented into individual pulse cycles, wherein each of the first time windows includes multiple individual pulse cycles; and Features of the arterial blood pressure wave in each of the first time windows are extracted as second extracted features for each first time window, wherein the generated features across the second time window are additionally based on the second extracted features.

3. The apparatus according to claim 1, in, The alert includes an inference that the patient will enter a hemodynamically unstable phase based on an estimated probability that exceeds a predetermined threshold.

4. The apparatus according to claim 1, in, When executed by the processor (152), the instructions also cause the device to: The plurality of heartbeats are continuously identified from the electrocardiogram waveforms of a plurality of the first time windows within the second time window.

5. The apparatus according to claim 1, further comprising: The monitor displays the alarm. The device includes a monitor (155).

6. A method for waveform-based hemodynamic instability warning, comprising: Multiple electrocardiogram waves are received via a first interface (153) connected to at least one electrocardiogram monitor (155) monitoring the patient; Identify (880) multiple heartbeats from the electrocardiogram waveform; The multiple heartbeats are separated into a first time window; The heartbeat features in each of the first time windows are extracted and used as the first extracted features for each first time window; Generate features for a second time window that includes multiple first time windows, based on the first extracted features; The trained artificial intelligence is applied to the generated features; Based on applying the trained artificial intelligence to the generated features to predict hemodynamic instability for the patient; and An alert warning of the hemodynamic instability is output based on the prediction of the hemodynamic instability.

7. The method according to claim 6, further comprising: Arterial blood pressure waves are received via a second interface (154) connected to an interface of an arterial blood pressure monitor (155) monitoring the patient; The arterial blood pressure wave is segmented into individual pulse cycles, wherein each first time window includes multiple individual pulse cycles; and Features of the arterial blood pressure wave in each of the first time windows are extracted as second extracted features for each first time window, wherein the generated features across the second time window are additionally based on the second extracted features.

8. The method according to claim 6, further comprising: Based on the characteristics of each of the plurality of heartbeats, each of the plurality of heartbeats is identified by one of a plurality of predetermined labels; and By applying a noise filter to the plurality of heartbeats, at least one of the plurality of heartbeats is excluded from the application of the trained artificial intelligence.

9. The method according to claim 7, further comprising: At least one individual pulse cycle (880) is identified as abnormal; and Based on identifying the at least one individual pulse cycle as abnormal (880), the at least one individual pulse cycle identified as abnormal is excluded from the application of the trained artificial intelligence.

10. The method according to claim 6, wherein, Extracting features of the heartbeats in each time window includes: Heart rate variability was extracted as the first extracted feature for each time window.

11. A tangible, non-transient, computer-readable storage medium for storing a computer program, said computer program causing a system (100) including said tangible, non-transient, computer-readable storage medium to function when run by a processor (152): Multiple heartbeats are identified from electrocardiogram waves received from at least one electrocardiogram monitor (155) via a first interface (153); The multiple heartbeats are separated into a first time window; The heartbeat features in each of the first time windows are extracted and used as the first extracted features for each first time window; Generate features for a second time window that includes multiple first time windows, based on the first extracted features; The trained artificial intelligence is applied to the generated features; Based on applying the trained artificial intelligence to the generated features to predict hemodynamic instability for patients; and An alert warning of the hemodynamic instability is output based on the prediction of the hemodynamic instability.

12. The tangible non-transient computer-readable storage medium according to claim 11, wherein, When run by the processor (152), the computer program also enables the system (100) including the tangible, non-transient, computer-readable storage medium to: The arterial blood pressure wave received from the arterial blood pressure monitor (155) via the second interface (154) is segmented into individual pulse cycles, wherein each of the first time windows includes multiple individual pulse cycles; and Features of the arterial blood pressure wave in each of the first time windows are extracted as second extracted features for each first time window, wherein the generated features across the second time window are additionally based on the second extracted features.

13. The tangible non-transient computer-readable storage medium according to claim 11, wherein, When run by the processor (152), the computer program also enables the system (100) including the tangible, non-transient, computer-readable storage medium to: Based on the characteristics of each of the plurality of heartbeats, each of the plurality of heartbeats is identified by one of a plurality of predetermined labels; and By applying a noise filter to the plurality of heartbeats, at least one of the plurality of heartbeats is excluded from the application of the trained artificial intelligence.

14. The tangible non-transient computer-readable storage medium according to claim 11, wherein, When run by the processor (152), the computer program also enables the system (100) including the tangible, non-transient, computer-readable storage medium to: At least one individual pulse cycle is identified as abnormal; and Based on identifying the at least one individual pulse cycle as abnormal (880), the at least one individual pulse cycle identified as abnormal is excluded from the application of the trained artificial intelligence.

15. The tangible, non-transient, computer-readable storage medium according to claim 11, in, The tangible, non-transient, computer-readable storage medium is provided as a component of the monitor (155) that displays the alarm.

Images