WO2019075035A1 - Système et procédé pour distinguer un événement cardiaque d'un bruit dans un signal d'électrocardiogramme (ecg) - Google Patents

Système et procédé pour distinguer un événement cardiaque d'un bruit dans un signal d'électrocardiogramme (ecg) Download PDF

Info

Publication number
WO2019075035A1
WO2019075035A1 PCT/US2018/055177 US2018055177W WO2019075035A1 WO 2019075035 A1 WO2019075035 A1 WO 2019075035A1 US 2018055177 W US2018055177 W US 2018055177W WO 2019075035 A1 WO2019075035 A1 WO 2019075035A1
Authority
WO
WIPO (PCT)
Prior art keywords
ecg
preconfirmation
period
event
noise
Prior art date
Application number
PCT/US2018/055177
Other languages
English (en)
Inventor
Adam Sullivan
Thomas E. Kaib
Francesco Nicolo
Steven J. Szymkiewicz
Jason T. Whiting
Original Assignee
Zoll Medical Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US15/782,568 external-priority patent/US20180116537A1/en
Application filed by Zoll Medical Corporation filed Critical Zoll Medical Corporation
Publication of WO2019075035A1 publication Critical patent/WO2019075035A1/fr

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/63—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for local operation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/35—Detecting specific parameters of the electrocardiograph cycle by template matching
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/361—Detecting fibrillation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/363—Detecting tachycardia or bradycardia
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/364—Detecting abnormal ECG interval, e.g. extrasystoles, ectopic heartbeats
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/366—Detecting abnormal QRS complex, e.g. widening
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
    • A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
    • A61B5/6802—Sensor mounted on worn items
    • A61B5/6804—Garments; Clothes
    • A61B5/6805—Vests
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/74—Details of notification to user or communication with user or patient ; user input means
    • A61B5/746—Alarms related to a physiological condition, e.g. details of setting alarm thresholds or avoiding false alarms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/25—Bioelectric electrodes therefor
    • A61B5/279—Bioelectric electrodes therefor specially adapted for particular uses
    • A61B5/28—Bioelectric electrodes therefor specially adapted for particular uses for electrocardiography [ECG]
    • A61B5/282—Holders for multiple electrodes
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/30—Input circuits therefor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G06F2218/16—Classification; Matching by matching signal segments
    • G06F2218/18—Classification; Matching by matching signal segments by plotting the signal segments against each other, e.g. analysing scattergrams
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Definitions

  • the present invention generally relates to systems and methods for detecting arrhythmia and, more particularly, to systems and methods for use with a cardiac monitoring device to reduce the frequency of false detections of arrhythmia.
  • a normal heart beat wave starts at the sinoatrial node (SA node) and progresses toward the far lower corner of the left ventricle.
  • SA node sinoatrial node
  • a wave starting in the ventricles and resulting in a rate over 100 beats per minutes (or in uncoordinated ventricular movement) is called a ventricular tachyarrhythmia.
  • ECG electrocardiography
  • VF ventricular fibrillation
  • VT ventricular tachycardia
  • VT is a cardiac tachyarrhythmia originating from a ventricular ectopic focus, characterized by a rate typically greater than 100 beats per minute and typically by wide QRS complexes.
  • VT may be monomorphic (identical QRS complexes) or polymorphic (varying QRS complexes).
  • a heart in the VT state may or may not produce a pulse (i.e., pulsatile movement of blood through the circulatory system). If there is no pulse or a weak pulse, then the VT is considered to be unstable and a life threatening condition.
  • An unstable VT may be treated with an electrical shock or defibrillation.
  • VF is a pulseless arrhythmia with chaotic electrical activity and uncoordinated ventricular contraction in which the heart immediately loses its ability to function as a pump.
  • VF and unstable VT are the primary arrhythmias that cause sudden cardiac arrest (SCA) and lead to sudden cardiac death (SCD).
  • An electrical shock to the heart can correct VF and unstable VT rhythms.
  • a simplified understanding of the process is that an electrical shock of sufficient size can force all of the cardiac cells in the heart to depolarize at the same time. Subsequently, all of the cardiac cells experience a short resting period with the hope that the sinoatrial node (SA node) will recover from this shock before any of the other cells, and that the resulting rhythm will be a pulse-producing rhythm, if not a normal sinus rhythm.
  • SA node sinoatrial node
  • Various devices are currently available for providing an electrical shock to the heart.
  • some implantable devices commonly referred to as pacemakers, deliver microjoule electrical shocks to a slowly beating heart in order to speed the heart rate up to an acceptable level.
  • Other implantable devices commonly referred to as implantable cardioverter defibrillators, deliver electrical shocks in the range of 10 to 40 joules to correct VT or VF.
  • high energy shocks e.g., 180 to 360 joules
  • a defibrillator via external paddles applied to the chest wall in order to correct VT or VF, and prevent the possible fatal outcome of these arrhythmias.
  • implantable pacemakers and defibrillators have significantly improved the ability to treat these otherwise life-threatening conditions. Being implanted within the patient, such devices continuously or substantially continuously monitor the patient's heart for treatable arrhythmias and, when such is detected, the device applies corrective electrical shocks directly to the heart. External pacemakers and defibrillators that apply corrective electrical shocks to the patient's chest wall also are used to correct such life-threatening arrhythmias but suffer from a drawback insofar as it may not be possible to use the device to apply treatment in time during an acute arrhythmic emergency to save the patient's life. Such treatment is needed within a few minutes to be effective.
  • wearable defibrillators have been developed for patients that are susceptible to ventricular tachyarrhythmias and are at temporary or uncertain permanent risk of sudden death, or are awaiting an implantable device. Such wearable defibrillators are typically configured to provide external treatment if a life-threatening arrhythmia is detected.
  • a wearable defibrillator is available from ZOLL Lifecor Corporation of Pittsburgh, PA.
  • the sensitivity of the process used to detect such life threatening arrhythmia is very high and is designed to deliver treatment to every person who requires a treatment. The tradeoff to this high sensitivity is a higher level of false-positive detection of arrhythmias due to signal noise.
  • wearable defibrillators use an alarm sequence to alert conscious patients of an impending treatment, who by virtue of being conscious are not experiencing a lethal arrhythmia, and allows them to stop the treatment.
  • the user may ignore the alarm, may not hear the alarm, may not be able to respond to the alarm, and/or may forget to depress the button to stop treatment.
  • defibrillators such as automated external defibrillators (AED) and implantable defibrillators, also monitor the ECG signal of a patient to determine whether a cardiac event has occurred.
  • AED automated external defibrillators
  • CPR cardio-pulmonary resuscitation
  • the first ECG analysis is usually initiated within a few seconds following attachment of the defibrillation electrodes to the patient. Subsequent ECG analyses may or may not be initiated based upon the results of the first analysis.
  • the rescuer is advised to deliver a defibrillation shock.
  • a signal processing routine that determines whether the detected arrhythmia is an actual arrhythmia or simply caused by noise so that a shock is only delivered to the patient if he/she is experiencing a ventricular tachyarrhythmia.
  • a typical implantable defibrillator detects physiological changes in patient conditions through the retrieval and analysis of signals stored in an on-board, volatile memory.
  • these devices can store more than thirty minutes of per heartbeat data recorded on a per heartbeat, binned average basis, or on a derived basis from which can be measured or derived various measures of physiologic activity; for example, atrial or ventricular electrical activity, minute ventilation, patient activity score, and the like. From this information, a cardiac event can be detected and the system can determine whether to deliver a therapeutic shock.
  • the system can include a signal processing routine to distinguish a cardiac event from noise.
  • Defibrillators may be provided with systems and processes designed using the power spectral density and trained using machine learning to help reduce the frequency of false detections, which if persistent enough, leads to alarm fatigue and potentially unnecessary treatments.
  • a cardiac monitoring device comprises: at least one sensing electrode for obtaining an electrocardiogram (ECG) signal from a patient; a processing unit comprising at least one processor operatively coupled to the at least one sensing electrode; and at least one non-transitory computer-readable medium comprising program instructions that, when executed by the at least one processor, causes the cardiac monitoring device to: obtain the ECG signal from the at least one sensing electrode; determine a transformed ECG signal based on the ECG signal; extract at least one value representing at least one feature of the transformed ECG signal; provide the at least one value to determine a score associated with the ECG signal, thereby providing an ECG-derived score; compare the ECG-derived score to a predetermined threshold score determined by machine learning; and provide an indication of a cardiac event if the ECG-derived score is one of above or below the predetermined threshold score determined by the machine learning.
  • ECG electrocardiogram
  • the transformed ECG signal may comprise a frequency-domain representation of the ECG signal or a representation of a power distribution of the ECG signal over a range of frequencies of the ECG signal.
  • the transformed ECG signal comprises a power spectral density (PSD) of the ECG signal with the PSD being determined by calculating a fast Fourier transform (FFT) of the ECG signal.
  • PSD power spectral density
  • FFT fast Fourier transform
  • At least four features of the PSD may be extracted and provided to the machine learning. Such features may include, but are not limited to, at least one value representing a dominant frequency of the PSD, at least one value representing in-band entropy of the PSD between frequencies of 2Hz and 6Hz, at least one value representing first-band entropy of the PSD between frequencies of OHz and 2Hz, and at least one value representing a variance of the PSD.
  • the machine learning may be a multivariate adaptive regression splines classifier, a neural network classifier, or any suitable classifier.
  • the cardiac monitoring device may further include providing an instruction signal for taking an action based on the indication.
  • the action may be at least one of applying a therapy to a patient and providing a warning signal to the patient.
  • the cardiac monitoring device may also further include providing the indication and the ECG-derived score to the machine learning to refine the predetermined threshold score.
  • the program instructions that are executed by the at least one processor may be initiated for a portion of the ECG signal that is stored in a memory device when the at least one processor detects a triggering event.
  • the portion of the ECG signal is a predetermined time period of the ECG signal that precedes the triggering event.
  • the predetermined time period may be 20 seconds, for example.
  • a wearable defibrillator comprises: at least one therapy pad for rendering treatment to a patient wearing the wearable defibrillator; at least one sensing electrode for obtaining an electrocardiogram (ECG) signal from a patient; a processing unit comprising at least one processor operatively coupled to the at least one therapy pad and the at least one sensing electrode; and at least one non-transitory computer-readable medium comprising program instructions that, when executed by the at least one processor, causes the cardiac monitoring device to: obtain the ECG signal; determine a transformed ECG signal based on the ECG signal; extract at least one value representing at least one feature of the transformed ECG signal; provide the at least one value to determine a score associated with the ECG signal, thereby providing an ECG-derived score; compare the ECG-derived score to a predetermined threshold score determined by machine learning; and provide an indication of a cardiac event if the ECG-derived score is one of above or
  • the wearable defibrillator may include at least one alert system operatively connected to the at least one processor for conveying an alert signal to the patient.
  • the alert system comprises a display and a patient notification device.
  • the wearable defibrillator may also include at least one response mechanism operatively connected to the at least one processor.
  • the wearable defibrillator may be configured to prevent rendering treatment to the patient wearing the wearable defibrillator in response to a patient actuation of the at least one response mechanism.
  • a method for distinguishing a cardiac event from noise in an electrocardiogram (ECG) signal comprises: detecting an event in a portion of the ECG signal at a first module with a first signal processing routine; sending a signal that the event has been detected and the portion of the ECG signal to the second module; and evaluating the portion of the ECG signal at the second module with a second signal processing routine to determine whether the event detected by the first module is an actual cardiac event or noise based on output from a classifier.
  • ECG electrocardiogram
  • the second module may be dormant until a cardiac event is detected by the first module. If the second module determines that the cardiac event detected by the first module is an actual cardiac event, an alarm may be initiated providing a user with an indication that the actual cardiac event has occurred. If the second module determines that a cardiac event detected by the first module is noise, a silent delay period may be initiated in which no indication is provided to a user that the actual cardiac event has occurred. The second module may continue to analyze the portion of the ECG signal during the silent delay period to confirm that the cardiac event detected by the first module is noise. The second module may initiate an alarm providing a user with an indication that the actual cardiac event has occurred if the second module determines that the cardiac event detected by the first module is the actual cardiac event.
  • the second signal processing routine performed by the second module may comprise: obtaining the portion of the ECG signal; determining a PSD of the portion of the ECG signal; extracting at least one value representing at least one feature of the PSD; providing the at least one value to determine a score associated with the ECG signal, thereby providing an ECG-derived score; comparing the ECG-derived score to a predetermined threshold score determined by machine learning; and providing an indication of a cardiac event if the ECG- derived score is one of above or below the predetermined threshold score determined by the machine learning.
  • the machine learning may be one of a multivariate adaptive regression splines classifier and a neural network classifier.
  • the ECG signal may be obtained from a cardiac monitoring device.
  • the cardiac monitoring device may be one of a wearable defibrillator, an implantable defibrillator, a subcutaneous cardioverter defibrillator, an automated external defibrillator (AED), a mobile cardiac telemetry device, an ECG rhythm classifier, a ventricular arrhythmia detector, a Holter monitor, cardiac event monitor, and an implantable loop recorder.
  • a method is provided for distinguishing a cardiac event, such as VT or VF, from noise in an electrocardiogram (ECG) signal.
  • ECG electrocardiogram
  • the method comprises: obtaining the ECG signal; determining a transformed ECG signal based on the ECG signal; extracting at least one value representing at least one feature of the transformed ECG signal; providing the at least one value to determine a score associated with the ECG signal, thereby providing an ECG-derived score; comparing the ECG-derived score to a predetermined threshold score determined by machine learning; and providing an indication of a cardiac event if the ECG-derived score is one of above or below the predetermined threshold score determined by the machine learning.
  • a system includes an ambulatory medical device and a server.
  • the ambulatory medical device comprises: a digital signal processing module configured to: detect an abnormal rhythm from an electrocardiogram (ECG) signal of a patient using a first signal processing routine; and generate a first flag indicating an abnormal rhythm is detected; and a noise detector module configured to: receive the ECG signal from the digital signal processing module; execute a second signal processing routine to classify the abnormal rhythm as one of an arrhythmia event and a noise event; and, if the abnormal rhythm is classified as a noise event, initiate a preconfirmation period during which the noise detector module continues to evaluate the abnormal rhythm and classify the abnormal rhythm as one of an arrhythmia event and a noise event using the second signal processing routine; and generate a second flag indicating the start of the preconfirmation period; and a server configured to: receive the ECG signal, the first flag indicating the abnormal rhythm, and the second flag indicating the start of the preconfirmation period; and provide a
  • the ambulatory medical device is configured to monitor and treat the patient for arrhythmia.
  • the ambulatory medical device may be a wearable defibrillator.
  • the digital signal processing module may be configured to receive the ECG signal of the patient from an electrode assembly removably attached to the patient.
  • the second signal processing routine may comprise: obtaining the ECG signal; determining a transformed ECG signal based on the ECG signal; extracting at least one value representing at least one feature of the transformed ECG signal; providing the at least one value to determine a score associated with the ECG signal, thereby providing an ECG-derived score; comparing the ECG-derived score to a predetermined threshold score determined by machine learning; and classifying the abnormal rhythm as an arrhythmia event or a noise event based on the comparison of the ECG-derived score with the predetermined threshold score.
  • the machine learning may be one of a multivariate adaptive regression splines classifier and a neural network classifier.
  • the second signal processing routine may comprise: obtaining the ECG signal; determining a transformed ECG signal based on the ECG signal; extracting at least one value representing at least one feature of the transformed ECG signal; providing the at least one value to determine a score associated with the ECG signal, thereby providing an ECG-derived score; comparing the ECG-derived score to a predetermined threshold score determined by machine learning; and classifying the abnormal rhythm as an arrhythmia event or a noise event based on the comparison of the ECG-derived score with the predetermined threshold score.
  • the machine learning may be one of a multivariate adaptive regression splines classifier and a neural network classifier.
  • the visual indication of the preconfirmation period may comprise a visual indication of a start of the preconfirmation period that is displayed on a display device operatively connected to the server.
  • the visual indication of the preconfirmation period may comprise an annotation of a visual representation of the ECG signal for a duration of the preconfirmation period.
  • the preconfirmation period may be annotated by highlighting a portion of the visual representation of the ECG signal corresponding to the preconfirmation period.
  • the preconfirmation period may be highlighted by providing a portion of the visual representation of the ECG signal corresponding to the preconfirmation period a color that is different than a remainder of the visual representation of the ECG signal.
  • the preconfirmation period may be highlighted by providing a portion of the visual representation of the ECG signal corresponding to the preconfirmation period with at least one of icons and text to differentiate the portion from a remainder of the visual representation of the ECG signal.
  • the preconfirmation period may be between about 10 seconds and about 60 seconds.
  • an alert device may be operatively coupled to the at least one processor if the one or more ambulatory medical devices does not provide an alarm to the patient.
  • a system may be provided that comprises a wearable defibrillator and a server.
  • the wearable defibrillator comprises: therapy pads for rendering treatment to a patient wearing the wearable defibrillator; sensing electrodes for obtaining an electrocardiogram (ECG) signal from the patient; a digital signal processing module operatively connected to the therapy pads and the sensing electrodes; and a noise detector module.
  • ECG electrocardiogram
  • the digital signal processing module is configured to: detect an abnormal rhythm from an electrocardiogram (ECG) signal of a patient using a first signal processing routine; and generate a first flag indicating an abnormal rhythm is detected; and a noise detector module configured to: receive the ECG signal from the digital signal processing module; execute a second signal processing routine to classify the abnormal rhythm as one of an arrhythmia event and a noise event; and, if the abnormal rhythm is classified as a noise event, initiate a preconfirmation period during which the noise detector module continues to evaluate the abnormal rhythm and classify the abnormal rhythm as one of an arrhythmia event and a noise event using the second signal processing routine; and generate a second flag indicating the start of the preconfirmation period.
  • the server is configured to: receive the ECG signal, the first flag indicating the abnormal rhythm, and the second flag indicating the start of the preconfirmation period; and provide a visual indication of the preconfirmation period.
  • the wearable defibrillator may be configured to delay a treatment sequence culminating in a delivery of the treatment to the patient by a duration of the preconfirmation period.
  • the wearable defibrillator may be configured to initiate a treatment sequence culminating in a delivery of the treatment to the patient immediately if the abnormal rhythm is classified as an arrhythmia event.
  • the visual indication of the preconfirmation period may comprise visual indication of a start of the preconfirmation period that is displayed on a display device operatively connected to the server or the visual indication of the preconfirmation period may comprise an annotation of a visual representation of the ECG signal for the duration of the preconfirmation period.
  • FIG. 1 is a schematic view of a wearable defibrillator that utilizes a method to reduce the frequency of false detections of cardiac arrhythmia in accordance with the present invention
  • FIG. 2 is a block diagram of a controller unit of the wearable defibrillator of FIG.1 ;
  • FIG. 3 is a block diagram of a microcontroller of the controller unit of FIG. 2;
  • FIG. 4 is a flow diagram of one process performed by a noise detection module of the microcontroller of FIG. 3;
  • FIG. 5 is a graph illustrating the power spectral density of an ECG signal demonstrating a normal sinus rhythm and containing significant noise content
  • FIG. 6 is a graph illustrating the power spectral density of an ECG signal demonstrating VT arrhythmia
  • FIG. 7 is a flow diagram illustrating a methodology performed by the noise detector module
  • FIG. 8 is a state machine diagram illustrating various states available for the noise detection module of the microcontroller of FIG. 3;
  • FIG. 9 is a chart illustrating the timeline sequence of a methodology performed by the noise detection module of the microcontroller of FIG. 3;
  • FIG. 10 is a flow diagram illustrating an exemplary software methodology for implementing the invention.
  • FIG. 11 is a flow diagram illustrating an exemplary method for generating a visualization of an example preconfirmation or delay period in which the methodology of the noise detector module is performed;
  • FIG. 12 is an exemplary visualization of the ECG signal in which the example preconfirmation or delay period is annotated
  • FIG. 13 is another exemplary visualization of the ECG signal in which the example preconfirmation or delay period and the treatment period are annotated.
  • FIG. 14 is a flow diagram illustrating an exemplary method for generating time flags or stamps in which the methodology of the noise detector module is performed.
  • a wearable defibrillator such as the Ellipse ICD available from St. Jude Medical of St. Paul, MN
  • an automated external defibrillator such as an AED PlusTM Automated External Defibrillator device available from ZOLL Medical Corporation of Chelmsford, MA
  • a mobile cardiac telemetry device such as an MCOTTM device available from BioTelemetry, Inc.
  • an ECG rhythm classifier such as the ECG rhythm classifier, a ventricular arrhythmia detector, a Holter monitor (such as the M8 Holter Reporter monitoring system available from Applied Cardiac Systems of Website, CA), and any other suitable cardiac event recorder (such as the ER 920W cardiac event monitor available from Braemar Manufacturing, LLC, of Eagan, MN).
  • a Holter monitor such as the M8 Holter Reporter monitoring system available from Applied Cardiac Systems of Website, CA
  • any other suitable cardiac event recorder such as the ER 920W cardiac event monitor available from Braemar Manufacturing, LLC, of Eagan, MN).
  • arrhythmias such as premature atrial contractions (PACs), multifocal atrial tachycardia, atrial flutter, and atrial fibrillation, supraventricular tachycardia (SVT), junctional arrhythmias, tachycardia, junctional rhythm, junctional tachycardia, premature junctional contraction, and ventricular arrhythmias such as premature ventricular contractions (PVCs) and accelerated idioventricular rhythm, may also be detected.
  • atrial arrhythmias such as premature atrial contractions (PACs), multifocal atrial tachycardia, atrial flutter, and atrial fibrillation, supraventricular tachycardia (SVT), junctional arrhythmias, tachycardia, junctional rhythm, junctional tachycardia, premature junctional contraction, and ventricular arrhythmias such as premature ventricular contractions (PVCs) and accelerated idioventricular rhythm, may also be detected.
  • PVCs premature ventricular contractions
  • a wearable defibrillator 1 may be worn by a patient and may include a belt or harness or other apparel configured to permit the patient to wear the defibrillator 1.
  • a wearable defibrillator can be configured for long term or extended wear.
  • the wearable defibrillator may be typically worn nearly continuously for two to three months at a time.
  • the wearable defibrillator 1 may be configured to continuously or substantially continuously monitor the vital signs of the patient, to be user-friendly and accessible, to be as light-weight, comfortable, and portable as possible, and to be capable of delivering one or more life-saving therapeutic shocks when needed.
  • the wearable defibrillator 1 may comprise a controller unit 3 positioned within an external housing that is configured to be worn by a patient and connected to a therapeutic or treatment device, such as an upper body harness or vest that includes ECG electrodes 5a, 5b, 5c, and 5d and therapy pads 7.
  • a therapeutic or treatment device such as an upper body harness or vest that includes ECG electrodes 5a, 5b, 5c, and 5d and therapy pads 7.
  • the ECG electrodes 5a, 5b, 5c, and 5d and therapy pads 7 of the harness or vest are operatively connected to controller unit 3 via a trunk cable 9 or other suitable connection mechanism.
  • suitable wearable defibrillators are disclosed in United States Patent Nos.
  • the upper body harness or vest may also include other sensing electrodes (not shown) such as heart beat sensors, accelerometers, and sensors capable of measuring blood pressure, heart rate, thoracic impedance, respiration rate, heart sounds, acoustic sensors, audio transducers, and the activity level of the subject.
  • sensing electrodes such as heart beat sensors, accelerometers, and sensors capable of measuring blood pressure, heart rate, thoracic impedance, respiration rate, heart sounds, acoustic sensors, audio transducers, and the activity level of the subject.
  • Electrodes 5a, 5b, 5c, and 5d are removably attached to the patient when the wearable defibrillator 1 is worn by the patient.
  • the electrodes 5a, 5b, 5c, and 5d form part of electrode assembly 11.
  • the electrode assembly 11 receives ECG signals from a front-to-back channel and from a side-to-side channel.
  • the front-to-back (FB) channel includes an electrode 5a, 5b, 5c, or 5d positioned on the chest of the patient and another electrode 5a, 5b, 5c, or 5d positioned on the back of the patient.
  • the side-to-side (SS) channel includes an electrode 5a, 5b, 5c, or 5d positioned on the left side of the chest and another electrode 5a, 5b, 5c, or 5d positioned on the right side of the patient.
  • the controller unit 3 is operatively connected to the therapy pads 7, at least one tactile stimulator 12, electrode assembly 11, and one or more alert devices such as audio/visual alert devices 13.
  • the therapy pads 7 are removably connected to the patient when the defibrillator 1 is worn.
  • the controller unit 3 may be operatively connected to other electrodes/devices which provide data to the controller regarding other physiological conditions or parameters of the patient.
  • the alert device 13 in the present example is audio/visual in design, it may additionally or alternatively be tactile in design, to provide a tactile (such as buzzing) alert to a patient in addition to or instead of audible and/or visible alerts.
  • a trunk cable 9 may be used to connect the electrode assembly 11 to the controller unit 3
  • other types of cables or other connection devices to operatively connect the electrode assembly 11 to the controller unit 3 may also be used.
  • Wiring or other connection devices may be used to connect at least one portion of the electrode assembly 11 to the electrodes 5a, 5b, 5c, and 5d.
  • the controller unit 3 may alternatively be operatively connected to one or more of the electrodes 5a, 5b, 5c, and 5d, therapy pads 7, electrode assembly 11, and stimulator 12 by a wireless connection or a combination of wireless and wired connections.
  • the controller unit 3 may include, without limitation, one or more processors, one or more controllers and/or one or more programs or other software stored in memory operatively connected to one or more processors. More specifically, and with reference to FIG. 2, the controller unit 3 comprises a microcontroller 15, audio/visual alert device 13 operatively connected to the microcontroller 15, and a power supply 17.
  • the power supply 17 has sufficient capacity to administer one or more therapeutic shocks to the therapy pads 7 as well as provide power to all of the internal components of the wearable defibrillator 1.
  • the audio/visual alert device 13 may also include a microphone 19, a speaker 21 , and audio circuitry 23 operatively connected to the microcontroller 15 via an appropriate interface.
  • the microcontroller 15 may be configured to cause a voice responsiveness test to be run as part of determining that the patient is experiencing a condition requiring treatment.
  • the voice responsiveness test may include the speaker 21, to verbally ask the patient if the patient is conscious.
  • the microphone 19 or other audio device senses that the patient responds with a positive verbal comment, such as, for example, "yes”
  • the microcontroller 15 may be configured to delay treatment.
  • the microcontroller 15 may be configured to cause the speaker 21 to provide a verbal message asking the patient to press one or more response buttons 27 to verify that the patient is conscious.
  • the microcontroller 15 may be configured to cause the speaker 21 to ask the patient certain questions in the event a possible condition is identified that may require delivery of a treatment.
  • the speaker may be configured to ask the patient "Are you conscious?" or "If you are conscious, please state your name.”
  • the microcontroller 15 may be operatively connected to a memory device that contains the patient's voice signature to verify that the patient is answering the questions. Such verification prevents a passerby from preventing treatment of the patient by improperly responding to the questions.
  • the use of the microphone 19 and the speaker 21 permits the patient to have realtime input provided to the wearable defibrillator 1.
  • the microcontroller 15 may also be configured to record the audio input in the proximity of the wearable device for later review by emergency personnel or to connect for voice contact with a monitoring center. Such information may help care providers to determine a diagnosis for the patient or to treat the patient.
  • the audio/visual alert device 13 may further include a display screen 25 for providing information to a patient and for providing a user input device to the patient.
  • the display screen 25 may be configured to provide information such as, but not limited to, time, battery life, volume, signal strength, device status, and any other useful information to the patient.
  • the display screen 25 also allows the user to access various data regarding the wearable defibrillator 1 such as, but not limited to, the settings of the device, data stored by the device, and various other data accumulated by the wearable defibrillator 1.
  • the display screen 25 further acts as a communication interface to allow the patient to send and receive data.
  • At least one response button 27 may also be provided in operative connection with the microcontroller 15.
  • the response button 27 may be provided to prevent the microcontroller 15 from sending a signal to initiate the rendering of treatment to the patient wearing the wearable defibrillator 1 in response to a patient actuation of the response button 27 within a predetermined time period.
  • the microcontroller 15, which may be a single chip multiprocessor such as is available from ARM Ltd. of Cambridge, UK, may be configured to receive more than one channel (i.e., front-to-back and side-to-side as described above) of ECG information from the ECG electrodes 5a, 5b, 5c, and 5d, detect abnormal heart rhythms based on the information received from the ECG electrodes 5a, 5b, 5c, and 5d, and administer a therapeutic shock to the patient via the therapy pads 7 if an abnormal heart rhythm is detected, unless a user intervenes within a predetermined period of time via the response button 27.
  • the predetermined period of time may extend from a minimum of 30 seconds to a maximum of a few minutes, but it may differ based on the type of detected arrhythmia, device interaction, or on the presence of noise.
  • the microcontroller 15 is also configured to perform several other functions in addition to those described above. These other functions may leverage the robust computing platform provided by the microcontroller 15 without disrupting the functions described above. Some examples of these other functions include notifying emergency personnel of the location of a patient who just received a therapeutic shock via a communication module (not shown), providing users of the device with the historical physiological data of the wearer of the device via the display screen 25, and/or notifying the manufacturer at a central location 29 of the wearable defibrillator 1 of potential performance issues within the wearable defibrillator 1 that may require repair to or replacement of the wearable defibrillator 1 via the communication module.
  • these other functions may include maintaining a history of data and events by storing this information in the memory device, communicating with the user via the display screen 25, and/or reporting data and events via the communication module.
  • another function may perform additional operations on the history of critical data. For instance, in one example, a function analyzes the history of critical data to predict worsening heart failure or an increased risk of sudden cardiac death and/or monitors other patient physiological conditions and parameters.
  • the wearable defibrillator 1 is configured to stimulate the patient for a predetermined time period.
  • the stimulus may be any stimulus perceptible by the patient. Examples of stimuli that the defibrillator 1 may produce include visual (via the display screen 25), audio (via the speaker 21), tactile stimulation (via the tactile stimulator 12) or a mild stimulating alarm shock (via the therapy pads 7).
  • the response button 27 is provided to allow a user to turn off the stimulus by pressing the response button 27 within a predetermined time period, such as described above.
  • the wearable defibrillator 1 administers one or more therapeutic shocks to the patient via the therapy pads 7.
  • the microcontroller 15 includes a digital signal processing module 31, a shared memory module 33 operatively coupled to the digital signal processing module 31 , and an activation module 37 operatively coupled to the digital signal processing module 31 and the shared memory module 33.
  • the activation module 37 includes an I/O module 38 operatively coupled to the digital signal processing module 31 and the shared memory module 33, a noise detector module 35 operatively coupled to the digital signal processing module 31 and the shared memory module 33 via the I/O module 38, a shell 36 operatively connected to the noise detector module 35 via the I/O module 38 and configured to receive a signal from the noise detector module 35 and to provide a signal, via the I/O module to the therapy pads 7 to administer one or more therapeutic shocks if the signal from the noise detector module 35 indicates that an abnormal condition is present.
  • the noise detector module 35 may be included within the digital signal processing module 31.
  • the noise detector module 35 can be in communication with the shared memory module 33, and be able to access stored ECG data in the shared memory module 33.
  • the shell 36 is sized to store programming objects that are configured to provide functionality such as, but not limited to, a user interface, Bluetooth ® connectivity, Universal Asynchronous Receiver/Transmitter (UART), capacitor charging, and other functionality disclosed below.
  • the digital signal processing module 31 may be configured to receive signals from more than one channel (i.e., front-to-back and side-to-side as described above) of ECG signals from the electrode assembly 11.
  • the digital signal processing module 31 passes the "raw" ECG signals to the shared memory module 33 as indicated by arrow A.
  • the digital signal processing module 31 also processes the ECG signal using a process that detects abnormal heart rhythms in the ECG signal.
  • An example of the methods used to detect abnormal heart rhythms may be found in United States Patent No. 5,944,669, which is hereby incorporated by reference in its entirety.
  • the method used by the digital signal processing module 31 to detect abnormal heart rhythms may be by a first signal processing routine stored on memory of the digital signal processing module 31. If an abnormal heart rhythm, such as a VT/VF, is detected by the digital signal processing module 31, a signal is sent to the noise detector module 35, via the I/O module 38, as indicated by arrow B.
  • the noise detector module 35 may be configured to receive the signal from the digital signal processing module 31, via the I/O module 38, indicating that an abnormal heart rhythm has been detected and further to process this signal to determine whether the abnormal heart rhythm is a VT/VF condition or if it was caused by noise in the ECG signal using a methodology or a second signal processing routine that will be discussed in greater detail hereinafter.
  • the noise detector module 35 may also access the "raw" ECG signals from the shared memory module 33, via the I/O module 38, as indicated by arrow C to assist in the determination as to whether the abnormal heart rhythm detected by the digital signal processing module 31 is a VT/VF condition or if it was caused by noise in the ECG signal.
  • the shared memory module 33 may be sized to store months or years of sensor information, such as ECG data, that is gathered over several monitoring and treatment periods. These monitoring and treatment periods may include continuous monitoring periods of approximately 24 hours (and substantially continuous monitoring periods of approximately 1- 2 months) during which several treatments may have been delivered to the patient.
  • the microcontroller 15 is configured to analyze the stored sensor information and to determine adjustments to the treatment method, or alternative treatment methods, of benefit to the patient. For instance, in one example, the microcontroller 15 is configured to analyze ECG data collected substantially contemporaneously with each instance of patient initiated delay, or cancellation, of treatment.
  • the microcontroller 15 is configured to analyze the stored months of ECG data to recognize individualized, idiosyncratic rhythms that, while not normal, do not indicate a need for treatment. In some examples, the microcontroller 15 may automatically adjust the treatment method of the wearable defibrillator 1 to better suit the patient by not initiating treatment in response to the recognized, idiosyncratic rhythm. Such an adjustment may be performed in conjunction with review by appropriate medical personnel.
  • the noise detector module 35 is configured to send a signal to the shell 36, via the I/O module, indicating that an abnormal condition is present or that the abnormal condition detected by the digital signal processing module 31 was caused by noise. Based on this signal, the shell 36 instructs the activation module 37 to provide an instruction signal to the therapy pads 7, via the I/O module 38, to administer one or more therapeutic shocks or it provides an instruction signal to the audio/visual alert device 13 to provide a warning to the patient. If the patient does not provide a response to the warning provided by the alert device 13 via the response button 27, for instance, then the shell 36 provides an instruction signal to the therapy pads 7 to administer one or more therapeutic shocks.
  • a machine learning system may assist in determining whether the abnormal heart rhythm is a VT/VF condition or if it was caused by noise.
  • the machine learning system may be any conventional system of computer algorithms that improve automatically through experience. Suitable machine learning systems may employ unsupervised learning (in which patterns are identified in a stream of input without target labels) or supervised learning (where patterns are identified to match a designated target label).
  • unsupervised learning in which patterns are identified in a stream of input without target labels
  • supervised learning where patterns are identified to match a designated target label.
  • One form of supervised learning is classification, through which an item is categorized based on observation of examples of items from several categories.
  • Another form is numerical or statistical regression, in which a function is developed to describe the relationship between inputs and outputs and to predict how outputs should change as inputs change.
  • Suitable methods of supervised learning may also include rules-based decision trees, ensemble methods (bagging, boosting, random forest), the ⁇ -Nearest Neighbors algorithm (k- NN), linear or logistic regression, naive Bayes classifiers, neural networks, the perceptron algorithm, and the support vector machine (SVM) model.
  • Suitable machine learning classifiers include the MARSTM classifier available from Salford Systems of San Diego, CA, and the ARESLabTM Adaptive Regression Splines toolbox for Matlab/Octave available from Gints Jekabsons, of the Institute of Applied Computer Systems, Riga Technical University of Riga, Lithuania.
  • a suitable neural network classifier is Neural Network ToolboxTM from The Math Works, Inc. Classifiers are described in greater detail below.
  • Machine learning may be developed using a high-level technical computing language such as MATLABTM, which is available from The Math Works, Inc. of Natick, MA, and performed a priori to build a function, model, or classifier.
  • the output may be migrated into the noise detector module 35 for use in determining a score or probability associated with the quality of a given ECG signal (described in detail later).
  • a machine learning system including algorithms and software instructions, maybe located in any convenient location (such as shared memory module 33) and invoked when needed, or, as in the example shown in the FIG. 3, a machine learning system 39 may be included in noise detector module 35. Further, while any suitable machine learning system may be used, in the example shown in the Figures, the machine learning system constitutes a machine learning classifier.
  • the method for distinguishing a cardiac event from noise in an electrocardiogram (ECG) signal includes obtaining an indication from the digital signal processing module 31 that a cardiac event has occurred; obtaining the ECG signal corresponding to the cardiac event from the shared memory module 33 (block 100); determining a PSD of the ECG signal (block 101); extracting at least one feature of the PSD and determining an ECG-derived score (block 102); determining a predetermined threshold score with a machine learning classifier (block 103), that has been trained with a sample data set (block 403); comparing the ECG-derived score to the predetermined threshold score determined by the machine learning classifier (block 104); and providing an indication of a cardiac event (block 105) if the ECG-derived score is one of above or below the predetermined threshold score determined by the machine learning classifier or providing an indication that the signal is noise if the ECG-derived score is the other one of above or below the predetermined threshold score determined by the machine
  • the digital signal processing module 31 receives the more than one channel of ECG signals from the electrode assembly 11 and determines whether VT/VF is present in the signal using the abnormal heart rhythms heart rhythms detection methodology described above. If VT/VF is present, a signal is sent to the noise detector module 35 and the "raw" ECG signal is obtained by the noise detector module 35 from the shared memory module 33.
  • the noise detector module 35 is configured to distinguish an actual VT/VF condition from inappropriate sensing of a VT/VF condition due to noise caused by lead malfunction, electromagnetic interference, patient movement, etc.
  • the noise detector module 35 processes the time domain ECG signal to transform the ECG signal into a transformed ECG signal.
  • the module 35 may transform the time domain ECG signal into a frequency domain ECG signal.
  • the module 35 can transform the ECG signal into a representation of a power distribution of the ECG signal over a range of frequencies of the ECG signal.
  • the power distribution over the range of frequencies can be computed from the frequency domain ECG signal.
  • the transformed ECG signal can be a PSD, which describes how the power of the ECG signal is distributed over the different frequencies of the signal.
  • noise detector module 35 may generate the PSD by performing fast Fourier transform (FFT) operations on the time domain ECG signal, or it may employ other discrete Fourier transform (DFT) techniques to generate the PSD.
  • FFT fast Fourier transform
  • DFT discrete Fourier transform
  • the PSD may be calculated using a modification of Welch's method.
  • a 4096 sample (at 400Hz sampling rate) may be windowed into 512 sample windows with a 256 sample overlap.
  • a Hamming window is then applied to each segment.
  • the FFT of each segment is taken and all of the windows are averaged into a single mean periodogram.
  • Welch's method is described in Peter D. Welch, "The Use of Fast Fourier Transform for the Estimation of Power Spectra: A Method Based on Time Averaging Over Short Modified Periodograms", IEEE Trans. Audio and Electroacoust, Vol. AU-15, pp. 70-73, (June 1967).
  • the Welch method evaluates FFT, which is a complex number, and then produces the square of the modulus of the FFT to transform FFT into a real number as a final result.
  • FIG. 5 illustrates a PSD of an ECG signal demonstrating normal sinus rhythm and containing noise content
  • FIG. 6 illustrates a PSD of an ECG signal demonstrating VT arrhythmia.
  • the PSD has very distinct features when a signal that is in VT/VF is compared with a signal from an ECG demonstrating normal sinus rhythm, even in the presence of a significant amount of noise contamination.
  • the PSD of an ECG signal that is in VT/VF typically has several distinct dominant spectral bands while a normal sinus rhythm has a dominant spectral band at less than 2.5 Hz.
  • the dominant spectral band is the band of frequencies that corresponds to the maximum value of the PSD.
  • a PSD with multiple dominant spectral bands has more than one band of frequencies in which the power of the ECG signal is significant. Accordingly, as can also be seen in FIG. 5, normal sinus rhythm may have other dominant spectral bands due to noise content, but there is a high information content in the very low frequency bands (i.e., less than 2.5 Hz) of a PSD of normal sinus rhythm, as compared with the PSD of a VT/VF signal at the same very low frequency bands.
  • the information content in the PSD of the ECG signal that is in VT/VF is spread over more frequencies, and the frequency content is most dense around the frequency of the VT, which is typically greater than 2.5 Hz.
  • a PSD of normal sinus rhythm differs from a PSD of VT/VF arrhythmia.
  • Noise within the ECG signal may be characterized as entropy (i.e., randomness).
  • various entropy calculations may be performed on a PSD to differentiate between a normal sinus rhythm signal with noise and a VT/VF signal. For instance, an in-band entropy may be calculated for a PSD of an ECG signal. Entropy in the Electrical Engineering field was defined by Shannon in 1948.
  • a first-band entropy 63 may also be calculated for a PSD of an ECG signal.
  • the first-band entropy may be calculated by converting the PSD to a probability distribution function (PDF) and calculating the entropy of the signal between OHz and 2Hz.
  • PDF probability distribution function
  • the variance of the PSD may be calculated for a PSD of an ECG signal by treating the PSD as a PDF and calculating the second moment as will be discussed in greater detail hereinafter.
  • the four features of the PSD were selected as the features that would be extracted from the PSD at block 102 and submitted to the machine learning classifier at block 103 based on a combination of feature selection experimentation and physiological reasoning.
  • NSR normal sinus rhythm
  • some characteristics of the PSD remain the same.
  • entropy is a measure of randomness
  • the entropy in the 0-2Hz range of the PSD is similar for an NSR with and without noise.
  • the PSD for an NSR without the presence of noise would have much less information content in the 2-6Hz range than a PSD for an NSR with a noisy signal.
  • FIG. 6 shows a PSD of the ECG signal that is contaminated with motion artifact or machine noise. It can be seen that the PSD of FIG. 6 shows some information content in the 0-2Hz band. However, the information content is relatively minor, especially when compared to the information content of the PSD in the 2-6 HZ band. Even with noise contamination of the VT/VF signal, most of the frequency content is still greater than 2Hz, leaving the entropy for the 0-2Hz largely untouched.
  • the in-band entropy of the PSD between frequencies of 2Hz and 6Hz and first-band entropy of the PSD between frequencies of OHz and 2Hz were selected as features that would be extracted from the PSD at block 102 and submitted to the machine learning classifier at block 103.
  • the dominant frequency of the PSD is also investigated. For normal sinus rhythm, the dominant frequency is typically less than 2Hz. However, extreme noise content (either low or high frequency content) may skew this observation. For example, FIG.
  • the dominant frequency of the PSD was selected as a feature that would be extracted from the PSD at block 102 and submitted to the machine learning classifier at block 103.
  • variance was selected as a feature that would be extracted from the PSD at block 102 and submitted to the machine learning classifier at block 103 because the variance of a distribution provides a feel for the relative spread of the distribution. If a PSD has most of the energy in the 0-2Hz band and very little in the 2-6Hz band, the variance is relatively small. However, a PSD with much energy in the 2-6Hz band would provide a much wider variance of the PSD. As noted before, a PSD for an NSR has most of the energy in the 0-2Hz band, and a PSD for a VT/VF arrhythmia has more energy in the 2-6Hz band. In order to calculate variance, it is assumed that the PSD is a normal distribution. As clearly shown in FIGS. 5 and 6, neither the PSD of the normal sinus rhythm nor the PSD of the signal in VT/VF is a normal distribution, the variance of the PSD is calculated by treating the PSD as a PDF and calculating the second moment.
  • the PSD may be calculated using a modification of Welch's method.
  • a 4096 sample (at 400Hz sampling rate) is windowed into 512 sample windows with a 256 sample overlap. Accordingly, for each 4096 sample, the dominant frequency, in-band entropy, first-band entropy, and variance are calculated as discussed hereinabove. These values may then be mapped using a function derived from a training set that maps each value to a range of -1 to 1. If a new value falls outside of the original range of values, the new mapped value could be outside of the range of -1 to 1.
  • / ⁇ is the PDF derived from the PSD.
  • / ⁇ may also be calculated using a discrete formula, such as a summation, rather than a continuous (integral) formula presented above.
  • the variance is found by continuing to treat f(s) as a probability distribution function. Assuming that the mean of f(s) is ⁇ , the variance ( ) may be calculated using the following equation:
  • (y var ) may also be calculated using a discrete formula, such as a summation, rather than a continuous (integral) formula presented above.
  • f(s) represents a PDF
  • the first-band entropy (ym-i) may be calculated using the following equation:
  • in-band entropy (ym-i) may be calculated using the following equation:
  • the dominant frequency of the PSD the in-band entropy of the PSD between frequencies of 2Hz and 6Hz; the first-band entropy of the PSD between frequencies of OHz and 2Hz; and the variance of the PSD were described hereinabove as the features extracted from the PSD to be provided to the machine learning classifier, this is not to be construed as limiting the present invention as other features may be utilized.
  • Such features include, but are not limited to, mean, median, whole band entropy (0-200Hz), out-of-band entropy (6-18Hz), and negative exponential parameters.
  • a signal amplification e.g., signal gain
  • an ECG acquisition circuit can be used as a feature to be provided to the machine learning classifier.
  • a signal gain (e.g., positive or negative gain in accordance with a predetermined threshold) may be applied to, e.g., enhance or suppress an incoming ECG signal. For example, if an incoming ECG signal is suppressed (e.g., negative gain is applied to the signal) such information can indicate a higher probability that the ECG signal is a noisy ECG signal. In this manner, information regarding an amount of applied gain can be weighed in determining whether the sampled ECG signal is noise or a cardiac event.
  • a signal gain e.g., positive or negative gain in accordance with a predetermined threshold
  • any suitable machine learning classifier may be utilized such as, but not limited to, a multivariate adaptive regression splines classifier. In one illustrative example that is not necessarily preferred, such a classifier is utilized due to its portability to C++, low computational costs, and high success rate in dealing with multivariate data with nonlinear degrees of correlation.
  • MARS classifier in its generic sense to describe a multivariate adaptive regression splines classifier.
  • a MARS classifier is a non-parametric regression classifier that can automatically model non-linearities and interactions between variables. It has the ability to put "kinks” into the regression function.
  • a MARS classifier builds a model that is a weighted sum of basis functions.
  • a basis function is a constant, a hinge function, or a product of two or more hinge functions.
  • a hinge function partitions data into disjoint independent regions, with a constant (known as a knot) at the hinge of the regions.
  • the product of two or more hinge functions models interaction between two or more variables.
  • An exemplary description of a MARS classifier is discussed in Jerome H. Friedman, "Multivariate Adaptive Regression Splines", The Annals of Statistics, Vol. 19, No. 1, (1991).
  • the MARS classifier builds a model in two phases: a forward pass, followed by a backward pass.
  • the model starts with an intercept function, adding an intercept term (which is the mean of response values), and then finds basis functions (optimal knot functions) to add to the model.
  • the classifier continues to add basis functions until a residual error becomes too small to overcome.
  • a backward pass is then performed in which the classifier attempts to remove basis functions one-by-one to delete the least effective terms.
  • the dominant frequency, in-band entropy, first-band entropy, and variance are mapped using a function derived from a training set that maps each value to a range of -1 to 1. Once these features are mapped, the values are inputs to the MARS classifier function.
  • this classifier function is a machine-learning methodology that is trained, tested, and validated using standard machine learning techniques. For example, and without limiting the present invention, a collection of 120 signals derived from treatments performed by a wearable defibrillator 1 are stored as a training data set.
  • the training data set may include 60 noisy normal sinus rhythm signals (i.e., false positive detections) and 60 tachyarrhythmia signals.
  • the classifier was trained using the training data set described above.
  • the MARS classifier that is utilized in the current example includes two classifiers: one for the side-to-side channel and one for the front-to-back channel. Each classifier outputs a numeric value intended to be between 0 and 1.
  • any exemplary methodology utilized by the MARS classifier is as follows. Once all of the features are extracted from the PSD (denoted by ydf, yvar,ym-2, ym-6 as discussed above), they are fed into the MARS classifier function G.
  • This process is repeated independently for each channel.
  • a continuous 10-second buffer is maintained and the entire process is repeated in a sliding 10-second window with a 1- second overlap.
  • a methodology performed by the noise detector module 35 operates with the logic of the abnormal heart rhythms detection methodology of the digital signal processing module 31.
  • the methodology of the noise detector module 35 is started whenever a treatable VT or VF event (i.e., a triggering event) is detected by the abnormal heart rhythms detection methodology of the digital signal processing module 31 and evaluates the signal at intervals of a second (block 150). For every 10-second window, a score is produced for each channel. A final Boolean score is calculated by requiring both channel scores from the MARS classifier to be above a threshold. This methodology is discussed in greater detail hereinafter.
  • a 20-second buffer of the ECG signal is passed from the shared memory module 33 to be analyzed (block 152).
  • a master score is initialized to 20 "points" at the beginning of an event (block 154). If the result produced by the MARS classifier of the noise detector module 35 at block 156 for the initial 10 seconds of the ECG signal is below the pre-determined threshold (arrhythmia), the master score is incremented by 1 (blocks 158 and 160). If the result produced by the MARS classifier of the noise detector module 35 is above the threshold (noise), the master score is decremented by 2 (blocks 158 and 162).
  • Each subsequent 10 seconds of the ECG signal is evaluated in a sliding 10 second window with a 1 second overlap and integrated into the master score.
  • the maximum master score is 20 and the minimum is 0 (in the current example, the master score does not exceed this range).
  • the scoring methodology is run 20 times (10 scores of sliding 10 second windows, for each channel) (block 164)
  • the resulting master score determines the state of the event.
  • the threshold for noise classification is 10. If the score is above 10, the methodology performed by the noise detector module 35 ends and allows the treatment sequence to proceed (blocks 166 and 168). However, if an event score is less than 10, the event is classified as noise (blocks 166 and 170).
  • the treatment sequence is held off and a new score is created once per second using ensuing ECG data. If the score ever goes above 10, the event sequence is released and the scoring methodology stops.
  • the methodology of the noise detector module 35 will be discussed hereinafter in greater detail with reference to FIG. 8.
  • the methodology provides five main states available to the noise detector module 35: idle 200, working 201, arrhythmia 202, silent noise 203, and noise alarm 204.
  • the default state of the noise detector module 35 is idle 200.
  • a VT/VF event indication from the digital signal processing module 31 may trigger the state to switch to working 201.
  • the state may enter arrhythmia 202 or silent noise 203 state.
  • the silent noise 203 state progresses to the noise alarm 204 state. If the score ever goes above the threshold, the methodology will enter the arrhythmia 202 state.
  • Each of these states will be discussed in greater detail below.
  • the activation module 37 is typically dormant (in the idle 200 state) due to no events being detected. While idle, the digital signal processing module 31 monitors for arrhythmias and stores ECG signals in the shared memory module 33. Thus, the 0 stage may thus provide simultaneous power conservation and continual ECG monitoring.
  • the shared memory module 33 may have a memory structure to store the most recent 20 seconds of monitoring data so that such data is immediately available for transmission to the activation module 37 upon detection of a VT/VF event.
  • the noise detector module 35 is sent into the idle 200 state at startup and remains in the idle 200 state until a VT/VF event is detected, the algorithm of the noise detector module 35 exits the idle 200 state and enters the working 201 state.
  • the working 201 state is triggered if the state was idle 200 and a VT/VF event was detected.
  • the working 201 state there are two manners in which the system exits: a timeout or the initial evaluation of the score.
  • the incoming ECG signal from the shared memory module 33 is scanned for the VT/VF flag provided by the digital signal processing module 31.
  • the previous 20 seconds are evaluated using the MARS classifier and a score is produced. If the initial score is above 10, the state will switch to arrhythmia 202. If the initial score is less than 10, the state will switch to silent noise 203.
  • this searching routine for the flags in the ECG signal from the shared memory module 33 only continues for 7 seconds (which may be calculated in any conventional manner, in this example from the system clock, not the shared memory ticks).
  • 7 seconds which may be calculated in any conventional manner, in this example from the system clock, not the shared memory ticks.
  • the noise detector module 35 is sent into the arrhythmia 202 state. If an NSR event is detected, the state will switch to idle 200 after resetting all of the parameters.
  • the silent noise 203 state may be entered with an initial score greater than 10 produced while in the working 201 state.
  • the silent noise 203 state may continue up to a maximum time, for example, 10 seconds for a VT event or 30 seconds for a VF event (these times are pre-programmed and may be modified). Every second, the score and timer may be evaluated. In the current example, there are three exit possibilities for the silent noise 203 state: arrhythmia 202, noise alarm 204, or idle 200. If the score goes above 10, the state will switch to arrhythmia 202. If the score remains below 10 and the timer is greater than the timer threshold, the state will switch to the noise alarm 204. If an NSR event is detected, the state will switch to idle 200 after resetting all of the internal parameters. If the score remains below 10 and the timer is less than the threshold, the state will remain in silent noise 203.
  • the noise alarm 204 state may be entered from the silent noise 203 state.
  • the arrhythmia 202 state there are two exit possibilities: the arrhythmia 202 state or the idle 200 state. Every second the score is evaluated in the noise alarm 204 state. If the score goes above 10, the state switches to arrhythmia 202. If the score remains below 10 and the timer is greater than the timer threshold, the state also switches to the arrhythmia 202 as a failsafe method since, during the noise alarm 204 state, the patient has been asked to respond via the response button 27 or some other method by the audio/visual alert device 13 and no response has been provided. Therefore, the normal treatment sequence is resumed. If an NSR is detected, the state will switch to idle 200.
  • the arrhythmia 202 state is a "catch all" state in which there is only one way to exit: an NSR event.
  • the noise detector module 35 may remain in the arrhythmia 202 state until an NSR is detected.
  • the arrhythmia 202 state causes the noise detector module 35 to immediately release any events in the internal queue to the shell 36. All ECG signal searching, scoring, and evaluating may stop, and the noise detector module 35 directly transmits all information from the digital signal processing module 31 to the activation module 37.
  • the state Upon receipt of an NSR event from the digital signal processing module 31 to the noise detector module 35, the state will switch from arrhythmia 202 to idle 200.
  • the noise detector module 35 upon detection of a VT/VF signal flag from the digital signal processing module 31, the noise detector module 35 awakens from a dormant state and, in its working 201 state, evaluates the first 20 seconds of the ECG signal that the digital signal processing module 31 has indicated as being a VT/VF event. If the noise detector module 35 determines that the signal is a VT/VF event (by, as disclosed above, operating the learning classifier, scoring, and evaluating the score as being above the VT/VF event threshold), the only delay is the time it takes for the noise detector module 35 to process the signal and reach its determination. This delay is typically around one second.
  • the noise detector module 35 then enters the arrhythmia 202 state and releases the event to the activation module 37.
  • the activation module 37 initiates an alarm via the audio/visual alert device 13. If the patient does not respond to the alarm by pressing the response button 27 or through some other suitable method, the activation module 37 initiates treatment by sending a signal to deliver a therapeutic shock via the therapy pads 7. With reference to FIG. 9, the time to treatment is quicker for a VF event than for a VT event because VT events are more prone to being misclassified and VF events are more likely to be life-threatening events.
  • the noise detector module 35 (in its working 201 state) evaluates a 20 second buffer and determines that the ECG VT/VF event signal from the digital signal processing module 31 is most likely noise or contaminated NSR. It does this using the learning classifier system as described above.
  • a silent delay (shown by dotted cross hatching in FIG. 8) is initiated to allow the system time to kick out of the arrhythmia on its own and return to a dormant state. After this delay has elapsed (in the example shown, 30 seconds for VT and 10 seconds for VF), the noise detector module leaves the silent noise 203 state and enters the noise alarm 204 state.
  • the activation module 37 in the noise alarm 204 state initiates an alarm via the audio/visual alert device 13. If the patient does not respond to the alarm by pressing the response button 27 or through some other suitable method within a selected period (in the example shown, 30 seconds for VT and 25 seconds for VF), the activation module 37 initiates treatment by sending a signal to deliver a therapeutic shock via the therapy pads 7. If the response button 27 is pushed by the patient, the timer is reset. This extends the noise alarm 204 state by 60 seconds for a VT event or 35 seconds for a VF event (not shown). The response button 27 may be pushed indefinitely to hold off a treatment alarm, extending the noise alarm 204 state equally indefinitely, provided the score does not go above 10.
  • the pseudo-code may be utilized to perform the methodology of the noise detector module 35.
  • This pseudo-code is provided for exemplary purposes only and is not intended to limit the invention as any number of other coding techniques may be utilized to implement the methodology of the noise detector module 35.
  • a noise detector object is initialized at block 300. After initialization of the noise detector object, its main purpose is to receive "raw" ECG signals and events from the digital signal processing module 31 (block 302) to determine whether an event should be withheld or released to the shell module 36 of the activation module 37.
  • the noise detector module 35 may be in the idle 200 state and receive a normal heart rate event signal. In this case, the event signal will be immediately passed through to the shell 36.
  • a critical example is one where the noise detector module 35 is in the idle 200 state and receives a treatable VT event (block 312).
  • the I/O module 38 transfers the ECG signal provided from the shared memory module 33 to the noise detector module 35 which immediately processes the ECG signal until it finds the treatable VT flag. All VT/VF events are considered to be held if received when the noise detector module 35 is in the idle 200 state.
  • the noise detector module 35 evaluates the preceding 20 seconds of the ECG signal to determine whether an event is truly a VT/VF or is a noisy normal sinus rhythm (block 314). If the noise detector module 35 determines it is an arrhythmia 202 based on the first 20 seconds, it will release the event and go to the arrhythmia 202 state. However, if it determines that it is a noisy normal sinus rhythm, the noise detector module 35 continues to evaluate the incoming ECG signal. Any preceding events during this time of noise are evaluated for hold/return. For example, a SIDE SIDE NOISE event during the holding of an event because the noise will be placed into a queue. If the noise detector subsequently declares an arrhythmia, all of the events in the queue will be released. [00114] The following pseudo-code may be utilized by the noise detector module 35.
  • the noise detector object is first initialized using the following code:
  • the constructor is called and the pre-programmed parameters for the VT/VF silent period and the VT/VF noise alarm period are fetched.
  • these parameters may be VT(30, 30) and VF(10, 25), with the first parameter being the length of delay during the silent noise 203 state and the second parameter being the length of delay during the noise alarm 204 state.
  • the shell 36 of the activation module 37 sends the timer parameters to the noise detector module 35 using the following code:
  • the shell 36 of the activation module 37 is configured to send shared memory packets to the noise detector module 35 when it is available.
  • the noise detector module 35 has an internal buffer of 60 seconds of ECG signal from the shared memory module 33 that is held as well as a 20 second evaluation buffer for the most recent 20 seconds of the ECG signal from the shared memory module 33.
  • the following code may be utilized by the signal processing module 31 to send a shared memory packet to the noise detector module 35:
  • the digital signal processing module 31 sends a signal to the I/O module 38 of the activation module 37.
  • the signal could be for an event of any type, such as low battery, misaligned electrode, or a VT/VF event.
  • the digital signal processing module 31 may send a signal identifying that an event has occurred, and identifying the type of event, (such as "low battery” or "VT").
  • it could send a measure of the confidence in the identification of the event, or other measures such as voltage or capacitance.
  • the I/O module 38 may pass the signal to the noise detector module 35.
  • the I/O module 38 may use the following code to pass the event signal: int EventAlert(Event event), with
  • Event event being an identification of the event (“Event”) and a type of event (“event”).
  • the int EventAlert code may have associated with it a number of logic statements that allow the digital signal processing module 31 to alert the noise detector module 35 to information about the event, such as the criticality of the event, which may facilitate the decision of the noise detector module 35 as to whether to hold the event (meaning to take further processing action) or return it (meaning to ignore the event and return to idle 200 state). If the event is critical and needs to be held, the noise detector module 35 may set the return variable ("int") in the EventAlert code to 1. If the event is not critical (for example, garment maintenance is required soon), the noise detection module may set the return variable ("int") to 0. This allows for the noise detector module 35 to make quick decisions for lower level events that require no action.
  • the noise detector module 35 may confirm the criticality of the event. It may also send the int EventAlert(Event event) code back to the I/O module 38. For events deemed non-critical, the noise detector module 35 may then return to its idle 200 state, and the I/O module 38 may pass the code to the shell module in the activation module 37 to act as required. For events deemed critical, the I/O module 38 may take no further action because its receipt of the code indicates that the noise detector module 35 is taking the necessary actions with regard to the event.
  • the noise detector module 35 may be polled periodically by the shell 36 of the activation module 37. For example, the noise detector module 35 in its idle 200 state may be polled every five (5) minutes. When the noise detector module 35 is in its working 201 state while a "1" event has been held, the noise detector module 35 may also be polled by the shell 36 of the activation module 37 until its state returns back to idle 200.
  • One polling code may be:
  • the poll structure of the VALIDATEIDLE polling code may have multiple components: an ID component, an event component, a state component, a score component, and a flag with a recordFlag option to indicate whether or when to record an event flag in the ECG signal.
  • the following options may be utilized for the ID component: 0 may be used to indicate that no action is needed and 1 may be used to indicate that an event has been identified and action may be required.
  • the event component may have the same structure as the event alert that was passed into the noise detector module 35 using the int EventAlert(Event event) code.
  • the noise detector module 35 may populate the event component using information about these events retrieved from an internal event queue.
  • the state component may be used to identify the state of the noise detector module 35, and the score component may be used to identify the current score of the event being monitored.
  • the state component and the score component may be optional arguments that may be inputted into fields in the log.
  • the poll structure of the VALID ATEIDLE polling code may also have a timing flag to record a delay.
  • the noise detector module 35 populates the polling structure as appropriate. For example, if the state of the noise detector module 35 or the score of the event being monitored changes, the recordFlag option will be true. If the recordFlag option is true, the noise detector module 35 may instruct the shell 36 of the activation module 37 to record the current state and score components contained within the event poll structure using the following code:
  • debugging/logging/reset functions are available.
  • One is a noise predictor module status function.
  • the shell 36 of the activation module 37 is configured to send to the noise detector module 35 a request for status in the form of the following code:
  • the noise detector module 35 may populate the int GetDetectorStatus() code with its state.
  • the "int" portion of the GetDetectorStatus() code may be an integer defining one of the possible states of the noise detector module 35.
  • the noise detector module 35 may send the populated code to the shell 36 of the activation module 37.
  • the activation module 37 may readily have the current status of the states of the noise detector module 35.
  • a reset function may be used to reset the noise detector module 35 as required. For example, if the defibrillator system 1 locks or for some other reason requires resetting (such as during a disconnection of an electrode 5a, 5b, 5c, or 5d from the controller unit 3), the noise detector module 35 is configured to manually reset.
  • the reset function may take the same form as the function that resets the noise detector module 35 to the idle 200 state when the NO TREATABLE RHYTHM event is received from the digital signal processing module 31 via the I/O module 38.
  • the noise detector module 35 may invoke the reset by issuing the following exemplary code:
  • Another function may be used to provide the activation module with information about how long it takes for the noise detector module 35 to make a first decision in response to an event flag.
  • the function is configured to determine the elapsed time between the receipt of the first VT/VF event flag and the first decision by the noise detector module 35 in response to the event flag.
  • the shell 36 of the activation module 37 is configured to send to the noise detector module 35 a request for the elapsed time using the following code:
  • the noise detector module 35 may populate the int GetBufferTimingO code with the elapsed timing.
  • the noise detector module 35 may send the populated code to the shell 36 of the activation module 37.
  • visualization of the above described preconfirmation or delay period provided by the noise detector module 35 for later review by a physician may be beneficial.
  • FIG. 11 illustrates the manner in which a system provides a visualization of this preconfirmation or delay period.
  • the first signal processing routine of the digital signal processing module 31 of the wearable defibrillator 1 receives an ECG signal from the electrode assembly 11 and uses a combination of heart rate and morphology analysis, based on the baseline template of the patient's vectorcardiogram, to determine an abnormal rhythm.
  • the combination of these criteria provides a high degree of sensitivity and specificity in detection of ventricular tachyarrhythmias and a low rate of inappropriate shocks.
  • the first signal processing routine is running at all times to analyze the received ECG signals. If the first signal processing routine detects an abnormal rhythm at step 1100, a time flag or time stamp providing an indication of such an abnormal event is generated and associated with the received ECG signal by the digital signal processing module 31.
  • the first signal processing routine may also generate an indication indicating a level of confidence associated with the indication of the abnormal event.
  • a confidence indication may indicate a high level of confidence that the detected abnormal event is an arrhythmia event (such as ventricular tachycardia or ventricular fibrillation).
  • the confidence indication may indicate a low level of confidence that the detected abnormal event is an arrhythmia event, or may indicate that it is unable to assess either a high or low confidence level that the detected abnormal event is an arrhythmia event.
  • the noise detector module 35 receives the ECG signal, along with the time flag, from the digital signal processing module 31. Thereafter, using the second signal processing routine (also called a "preconfirmation" routine as described herein), the noise detector module 35 classifies the abnormal rhythm as an arrhythmia event or a noise event at step 1102. This process, in some implementations, can take about 1 second or less (e.g., in some examples, less than 100 ms after the first signal processing routine indicates the abnormal rhythm). At step 1103, if the second signal processing routine classifies the abnormal rhythm as an arrhythmia event (VT or VF event), then the treatment sequence is immediately initiated.
  • the second signal processing routine classifies the abnormal rhythm as an arrhythmia event (VT or VF event)
  • the controller initiates a preconfirmation or delay period in which the entire treatment sequence (which, in one example, begins with a vibration alarm and culminates in shock delivery) is delayed for a predetermined time in which the second signal processing routine confirms that the abnormal rhythm is noise.
  • the predetermined time of the delay period may be, in one example, 10 seconds for a VF event and up to 30 seconds for a VT event. In other examples, the additional delay period may be up to 35 seconds for a VF event and up to 60 seconds for a VT event.
  • the delay is configurable, e.g., during initial set up of the device prior to fitting on the patient.
  • the second signal processing routine of the noise detector module 35 also generates a time flag or stamp indicating the start and the end of the delay period. After 30 seconds, if the first signal processing, which is running in tandem with the second signal processing routine continues to detect arrhythmia, then the treatment sequence is launched. In implementation, the second signal processing routine may be continuously running throughout the delay period. As such, as the noise detector module 35 acquires additional ECG data, it re-assesses whether the abnormal rhythm is still noise or arrhythmia. In some examples, the second processing routine may be executed periodically during the delay period (e.g., every 2-5 seconds). In some examples, the process can be configured to wait for additional ECG data before declaring a result, and as such may take longer than 1 second (e.g., between 1-20 seconds).
  • the additional delay period may be, in one example, an additional up to 10 seconds for a VF event and up to 30 seconds for a VT event.
  • the additional delay period may be up to 35 seconds for a VF event and up to 60 seconds for a VT event.
  • the additional delay period may also be configurable, e.g., during initial set up of the device prior to fitting on the patient.
  • the wearable defibrillator 1 is configured to transmit the ECG signal obtained from the electrode assembly 11, along with all of the time indications, flags or stamps generated during a patient's use of the wearable defibrillator to a central server for review by a physician or other qualified technician (see step 1106).
  • the ECG signal and time flags or stamps (ECG information) is received by the server at step 1108.
  • the server is configured to provide a visualization of the ECG signal and annotate this visualization of the ECG signal at step 1110 in order to provide additional insight to the physician regarding the manner in which the wearable defibrillator provided treatment to a patient.
  • the annotated visualization of the ECG signal is then displayed on a display device operationally connected to the server for review by the physician (see step 1112).
  • FIG. 12 illustrates when the delay period in which the second signal processing routine of the noise detector module 35 is running is initiated. As shown, the delay period begins at around the 30 second mark.
  • the point at which the second signal processing routine causes the treatment sequence to be delayed can be indicated by a visual indication or annotation, e.g., "advanced arrhythmia discrimination" or "AAD". It is understood that other annotations may be used to indicate the start of the delay period.
  • the entire delay period may be annotated by showing the portion 1202 of the visualization of the ECG signal corresponding to the delay period as a dashed line, for example.
  • FIG. 13 illustrates that the portion 1202 of the visualization of the ECG signal corresponds to the delay period ends at around the 60 second mark.
  • the portion 1202 is again shown as a dashed line to visually show the delay.
  • the visualization of the ECG changes from a dashed line, indicating the delay period, to a bolded line 1204 to illustrate that the alarm period has begun.
  • the visualization of the ECG as shown in FIG. 13 may include icons and text 1206 providing further information to the physician.
  • FIG. 13 illustrates that the patient's heart rate is 164 bpm, which is well over an example VT threshold of 120 bpm.
  • the delay period may be annotated by highlighting a portion of the visual representation of the ECG signal corresponding to the delay period.
  • the delay period may be annotated by providing a portion of the visual representation of the ECG signal corresponding to the delay period a color (such as blue) that is different than a remainder of the visual representation of the ECG signal.
  • the delay period may be annotated by providing a portion of the visual representation of the ECG signal corresponding to the delay period with at least one of icons and text to differentiate the portion from a remainder of the visual representation of the ECG signal.
  • the above described visualization of the ECG signal beneficially provides an indication to the physician when the second signal processing routine of the noise detector module 35 was running, thereby providing insight into the manner in which the wearable defibrillator was working.
  • the first signal processing routine i.e., the arrhythmia detection algorithm or ADA
  • the second signal processing routine also called the "pre-confirmation" routine as described herein
  • classifies the abnormal rhythm as an arrhythmia event or a noise event at step 1402. If the second signal processing routine classifies the abnormal rhythm as an arrhythmia event (VT or VF event), then a flag 192 is generated and the treatment sequence is immediately initiated at step 1404 and as shown in Scenario 1.
  • the second signal processing routine classifies the abnormal rhythm as a noise event, a preconfirmation or delay period in which the entire treatment sequence is delayed for a predetermined time in which the second signal processing routine confirms that the abnormal rhythm is noise. In addition, a flag 191 indicating this preconfirmation is generated. If the first signal processing routine, which is running in tandem with the second signal processing routine does not detect arrhythmia at step 1406, a flag 215 is generated indicating that the second signal processing routine did not agree with the first signal processing routine that an arrhythmia was detected and normal monitoring is continued at step 1408 and as shown in Scenario 3.
  • the treatment sequence is launched at block 1412 and as shown in Scenario 4.
  • a flag 214 is generated indicating that the second signal processing routine never agreed with the first signal processing routine and timed out. If the preconfirmation period has not yet elapsed at block 1410, the second signal processing routine will continue to run to confirm whether the abnormal rhythm detected by the first signal processing routine is an arrhythmia at block 1414. If the second signal processing routine classifies the abnormal rhythm as an arrhythmia event (VT or VF event), then the treatment sequence is immediately initiated at block 1416 and as shown in Scenario 2. A flag 192 is generated indicating that the second signal processing routine agreed with the first signal processing routine that the detected signal is a treatable arrhythmia.
  • the visual indications and/or annotations shown above can be displayed on any display device or computer systems operatively connected to the server.
  • display devices can be desktop computers, personal digital assistants, smart phones, tablet devices, and the like.
  • a technician reviewing the ECG information along with such visual indications and annotations may generate reports for one or more caregivers to review.
  • reports may include the annotations along with ECG information.
  • Such a report may be accessible by the caregiver through an online portal that is protected through security credentials.
  • a technician may also be able to generate output files based on the information for review by interested persons.
  • a portable document format (PDF) or other electronic format document may be generated visually depicting the ECG information and annotations described above.
  • PDF portable document format
  • the second signal processing routine of the noise detector module 35 may instead suspend the shock delivery.
  • the alarm sequence may begin as soon as the first signal processing routine of the digital signal processing module 31 detects a possible arrhythmia event.
  • the alarm sequence may only be suspended if the patient responds by pressing and holding the response buttons 27.
  • the first signal processing routine of the digital signal processing module 31 may allow the patient around 60 seconds to respond before delivering the shock (e.g., in the event of a VT rhythm).
  • the wearable defibrillator 1 may allow the patient an additional 30 seconds to respond (i.e., a total of 90 seconds).
  • a visual indication similar to the one described above is employed when the ECG information is transmitted to a server for analysis by a technician.
  • a visual annotation or indicator can be displayed, and the ECG graph can be visually annotated as described above to depict the delay period.
  • the wearable defibrillator 1 may execute an alarm sequence (e.g., initiate vibration alarm(s), siren alarm(s), and/or audible warning messages).
  • the wearable defibrillator 1 determines if the patient is conscious through a series of vibration, voice, and siren alerts as described hereinabove. If the patient does not respond by pressing and holding the response buttons 27, the wearable defibrillator 1 automatically treats the patient. This patient interaction feature provides the advantage of potentially reducing or eliminating inappropriate treatments.
  • Captured ECG signal from users of wearable defibrillators have been analyzed and it has been found that, in certain instances ( ⁇ 5% of events), several factors may require additional processing in order for the digital signal processing module 31 to declare a treatable arrhythmia.
  • the second signal processing routine of the noise detector module 35 can intelligently decipher the validity of the arrhythmia with the goal of further reducing an already low rate of inappropriate therapies.
  • the second signal processing routine of the noise detector module 35 uses machine learning to process through a vast ECG repository to make data-driven decisions in order to determine if the arrhythmia is physiological and requires defibrillation.
  • the first signal processing routine of the digital signal processing module 31 can operate on conventional heart rate and morphology.
  • the second signal processing routine of the noise detector module 35 working in conjunction with first signal processing routine of the digital signal processing module31, can determine if an arrhythmia was life-threatening in an average of 12 seconds.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Cardiology (AREA)
  • Biomedical Technology (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Pathology (AREA)
  • Animal Behavior & Ethology (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Surgery (AREA)
  • Biophysics (AREA)
  • Veterinary Medicine (AREA)
  • Theoretical Computer Science (AREA)
  • Primary Health Care (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Epidemiology (AREA)
  • General Business, Economics & Management (AREA)
  • Business, Economics & Management (AREA)
  • Databases & Information Systems (AREA)
  • Physiology (AREA)
  • Electrotherapy Devices (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

L'invention concerne un système qui comprend un dispositif médical ambulatoire comprenant : un module de traitement de signal numérique configuré pour : détecter un rythme anormal à partir d'un signal d'électrocardiogramme (ECG) d'un patient et générer un premier indicateur indiquant qu'un rythme anormal est détecté ; et un module de détecteur de bruit configuré pour : recevoir le signal d'ECG et classer le rythme anormal comme l'un parmi un événement d'arythmie et un événement de bruit ; et, s'il est classé comme événement de bruit, initier une période de pré-confirmation pendant laquelle le module de détecteur de bruit continue à évaluer le rythme anormal et à classer le rythme anormal comme l'un parmi un événement d'arythmie et un événement de bruit ; et générer un second indicateur indiquant le début de la période de pré-confirmation ; et un serveur configuré pour : recevoir le signal d'ECG, le premier indicateur indiquant le rythme anormal, et le second indicateur indiquant le début de la période de pré-confirmation ; et fournir une indication visuelle de la période de pré-confirmation.
PCT/US2018/055177 2017-10-12 2018-10-10 Système et procédé pour distinguer un événement cardiaque d'un bruit dans un signal d'électrocardiogramme (ecg) WO2019075035A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US15/782,568 2017-10-12
US15/782,568 US20180116537A1 (en) 2014-07-07 2017-10-12 System and Method for Distinguishing a Cardiac Event From Noise in an Electrocardiogram (ECG) Signal

Publications (1)

Publication Number Publication Date
WO2019075035A1 true WO2019075035A1 (fr) 2019-04-18

Family

ID=64270944

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2018/055177 WO2019075035A1 (fr) 2017-10-12 2018-10-10 Système et procédé pour distinguer un événement cardiaque d'un bruit dans un signal d'électrocardiogramme (ecg)

Country Status (1)

Country Link
WO (1) WO2019075035A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113796873A (zh) * 2021-10-26 2021-12-17 山东建筑大学 穿戴式动态心电信号分类方法及系统

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4928690A (en) 1988-04-25 1990-05-29 Lifecor, Inc. Portable device for sensing cardiac function and automatically delivering electrical therapy
US5078134A (en) 1988-04-25 1992-01-07 Lifecor, Inc. Portable device for sensing cardiac function and automatically delivering electrical therapy
US5741306A (en) 1996-05-23 1998-04-21 Lifecor, Inc. Patient-worn energy delivery apparatus
US5944669A (en) 1997-11-20 1999-08-31 Lifecor, Inc. Apparatus and method for sensing cardiac function
US6065154A (en) 1998-04-07 2000-05-23 Lifecor, Inc. Support garments for patient-worn energy delivery apparatus
US6253099B1 (en) 1999-08-19 2001-06-26 Lifecor, Inc. Cardiac monitoring electrode apparatus and method
US6280461B1 (en) 1996-05-23 2001-08-28 Lifecor, Inc. Patient-worn energy delivery apparatus
US6681003B2 (en) 1999-10-05 2004-01-20 Lifecor, Inc. Data collection and system management for patient-worn medical devices
US20120226178A1 (en) * 2004-05-12 2012-09-06 Freeman Gary A Automatic Cardiac Therapy Advisor With Hidden Markov Model Processing
US8271082B2 (en) 2007-06-07 2012-09-18 Zoll Medical Corporation Medical device configured to test for user responsiveness
US8369944B2 (en) 2007-06-06 2013-02-05 Zoll Medical Corporation Wearable defibrillator with audio input/output
US20160000349A1 (en) * 2014-07-07 2016-01-07 Zoll Medical Corporation System and Method for Distinguishing a Cardiac Event From Noise in an Electrocardiogram (ECG) Signal

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4928690A (en) 1988-04-25 1990-05-29 Lifecor, Inc. Portable device for sensing cardiac function and automatically delivering electrical therapy
US5078134A (en) 1988-04-25 1992-01-07 Lifecor, Inc. Portable device for sensing cardiac function and automatically delivering electrical therapy
US5741306A (en) 1996-05-23 1998-04-21 Lifecor, Inc. Patient-worn energy delivery apparatus
US6280461B1 (en) 1996-05-23 2001-08-28 Lifecor, Inc. Patient-worn energy delivery apparatus
US5944669A (en) 1997-11-20 1999-08-31 Lifecor, Inc. Apparatus and method for sensing cardiac function
US6065154A (en) 1998-04-07 2000-05-23 Lifecor, Inc. Support garments for patient-worn energy delivery apparatus
US6253099B1 (en) 1999-08-19 2001-06-26 Lifecor, Inc. Cardiac monitoring electrode apparatus and method
US6681003B2 (en) 1999-10-05 2004-01-20 Lifecor, Inc. Data collection and system management for patient-worn medical devices
US20120226178A1 (en) * 2004-05-12 2012-09-06 Freeman Gary A Automatic Cardiac Therapy Advisor With Hidden Markov Model Processing
US8369944B2 (en) 2007-06-06 2013-02-05 Zoll Medical Corporation Wearable defibrillator with audio input/output
US8271082B2 (en) 2007-06-07 2012-09-18 Zoll Medical Corporation Medical device configured to test for user responsiveness
US20160000349A1 (en) * 2014-07-07 2016-01-07 Zoll Medical Corporation System and Method for Distinguishing a Cardiac Event From Noise in an Electrocardiogram (ECG) Signal

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
C.E. SHANNON: "A Mathematical Theory of Communication", THE BELL SYSTEM TECHNICAL JOURNAL, vol. XXVII, no. 3, July 1948 (1948-07-01), XP001041986
JEROME H. FRIEDMAN: "Multivariate Adaptive Regression Splines", THE ANNALS OF STATISTICS, vol. 19, no. 1, 1991
PETER D. WELCH: "The Use of Fast Fourier Transform for the Estimation of Power Spectra: A Method Based on Time Averaging Over Short Modified Periodograms", IEEE TRANS. AUDIO AND ELECTROACOUST., vol. AU-15, June 1967 (1967-06-01), pages 70 - 73, XP002586039

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113796873A (zh) * 2021-10-26 2021-12-17 山东建筑大学 穿戴式动态心电信号分类方法及系统
CN113796873B (zh) * 2021-10-26 2024-05-03 山东建筑大学 穿戴式动态心电信号分类方法及系统

Similar Documents

Publication Publication Date Title
US11160484B2 (en) System and method for distinguishing a cardiac event from noise in an electrocardiogram (ECG) signal
US10136826B2 (en) System and method for distinguishing a cardiac event from noise in an electrocardiogram (ECG) signal
WO2017058524A1 (fr) Modes de fonctionnement de dispositif médical
US11529086B2 (en) Medical device for sensing cardiac function
JP6975141B2 (ja) Ｅｃｇデータにおけるショック適応リズムまたはショック非適応リズムの高速識別
US11324443B2 (en) Amplitude spectrum area considerations for an external medical monitoring and treatment device
CN117222358A (zh) 语音辅助急性健康事件监测
WO2023154864A1 (fr) Classification de tachyarythmie ventriculaire
WO2019075035A1 (fr) Système et procédé pour distinguer un événement cardiaque d'un bruit dans un signal d'électrocardiogramme (ecg)
US20220314011A1 (en) Heart rate calculator with reduced overcounting
US12023166B2 (en) Medical device for sensing cardiac function
Pai et al. Cardiac monitoring & resuscitation system for ventricular fibrillation detection
WO2023192123A1 (fr) Détection d'arythmie à distance et analyse de traitement dans des dispositifs cardiaques portables
WO2024059101A1 (fr) Vérification adaptative de problèmes de santé graves par l'utilisateur
WO2024059160A1 (fr) Détection d'un problème de santé aigu pendant une attaque médicamenteuse
EP4329597A1 (fr) Détection de paramètres de respiration en tant qu'indicateur d'un événement d'arrêt cardiaque soudain
CN117015336A (zh) 急性健康事件监测与指导
Au Yeung Prediction of Sudden Cardiac Arrest for Patients with Congestive Heart Failure
JP2019180798A (ja) 異なるチャネルからのｅｃｇ信号を乗算することによって患者の心拍数を計算する着用型自動除細動器（ｗｃｄ）システム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18800789

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18800789

Country of ref document: EP

Kind code of ref document: A1