Predictive modeling of respiration signal within an implantable device

Abstract

A medical device configured to sense a first instance of a physiological signal indicative of respiration of a patient during a first period of time; deliver cardiac pacing pulses to the patient via a plurality of electrodes; predict, by a machine learning model for a second period of time, a second instance of the physiological signal based on the first instance of the physiological signal; adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal; and adjust the rate of cardiac pacing based on a third instance of the physiological signal in response to one of a duration of time since a most recent update of the machine learning model meeting a time threshold, a change in baseline heart rate of the patient meeting a change threshold, a change in patient activity meeting a change threshold.

Description

Attorney Docket No.: A0012462W001PREDICTIVE MODELING OF RESPIRATION SIGNAL WITHIN AN IMPLANTABLE DEVICECROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to U.S. Provisional Application No. 63 / 722,364 filed November 19, 2024, which is incorporated herein by reference in its entirety.FIELD

[0002] This disclosure generally relates to medical devices and, more particularly, to medical devices that deliver cardiac therapy.BACKGROUND

[0003] In healthy humans, heart rate naturally increases during inspiration and decreases during expiration. This phenomenon, known as respiratory sinus arrhythmia (RSA), supports ventilation / perfusion matching as blood enters the lungs, i.e., increases pulmonary blood flow when the lungs are inflated.SUMMARY

[0004] In general, this disclosure describes techniques for delivering cardiac pacing to mimic RSA, e.g., by increasing the cardiac pacing pulse rate during inspiration, and, in some examples, decreasing the pulse rate during expiration. Many current pacemakers pace monotonically in view of pulmonary inspiration / expiration, and the monotonic pacing rate increases with increased activity level (e.g., many current pacemakers provide pacing that does not vary within the respiratory cycle). Cardiac pacing to mimic RSA, i.e., respirophasic pacing, can provide therapeutic benefit, e.g., by increasing cardiac output and helping to reverse remodel the heart of heart failure (HF) patients. However, current methods of restoring RSA are based on several averaged prior-detected inspiration and expiration phases based on identified respiration cycles, which may mimic RSA when the patient is breathing at a stable rate, e.g., while the patient is sleeping, but may not accurately mimic RSA when the patient is breathing at a more variable rate, e.g., when the patient is active.

[0005] The techniques of this disclosure may be implemented by a cardiac therapy device, e.g., a pacemaker, that can continuously sense physiological signals, e.g., bioimpedance, accelerometer, and / or electrogram (EGM) signals, indicative of respiration ofAttorney Docket No.: A0012462W001 a patient. In some examples, the pacemaker may control therapy delivery circuitry to adjust cardiac pacing pulses based on the respiration signal, e.g., adjust cardiac pacing pulses to mimic RSA. The pacemaker may in some examples be configured to predict a subsequent physiological signal indicative of the respiration of the patient based on the sensed physiological signal and control therapy delivery circuitry to adjust cardiac pacing pulses. In some examples, the pacemaker may implement a machine learning model to predict the subsequent physiological signal. In some examples, the pacemaker may alternate between adjusting cardiac pacing pulses based on the sensed physiological signal and the predicted physiological signal. In some examples, the pacemaker may alternate between adjusting based on the sensed physiological signal and the predicted physiological signal based on one or more of a periodic schedule, in response to a change in patient heart rate, or in response to a change in patient activity.

[0006] In some examples, RSA pacing may be unnecessary, ineffective, or counterproductive under certain conditions, e.g., when the patient is already achieving RSA or when the patient’s heart rate exceeds a threshold. The techniques of this disclosure may de-prioritize delivering RSA pacing under such conditions by determining whether one or more criterion for adjusting cardiac pacing are satisfied based on one or more sensed patient parameters.

[0007] In some examples, the techniques of this disclosure may include selecting a type of physiological signal indicative of respiration of the patient to use for adjusting cardiac pacing pulses. As an example, the pacemaker may be configured to sense a bioimpedance signal, an accelerometer signal, and a cardiac EGM signal. The pacemaker may implement a machine learning algorithm to select the type of physiological signal, e.g., the bioimpedance signal, the accelerometer signal, or the cardiac EGM signal, for use in adjusting cardiac pacing pulses. The machine learning algorithm may select the physiological signal based on one or more of signal quality or battery drain associated with the physiological signal. The pacemaker may periodically update the selection and / or may update the selection in response to a change in signal quality or patient activity level, as examples.

[0008] In some example, a medical device system includes an implantable medical device (HMD) including: sensing circuitry configured to sense a first instance of a physiological signal corresponding to a first period of time, the first instance of the physiological signal being indicative of respiration of a patient over the first period of time; and therapy delivery circuitry configured to deliver cardiac pacing pulses to a heart of the patient via a plurality of electrodes; and processing circuitry configured to: predict, by a machine learning model, aAttorney Docket No.: A0012462W001 second instance of the physiological signal of the patient corresponding to a second period of time based on the first instance of the physiological signal; control the therapy delivery circuitry to adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal over the second period of time; and control the therapy delivery circuitry to adjust the rate of cardiac pacing based on a third instance of the physiological signal sensed via the sensing circuitry of the IMD and corresponding to a third period of time in response to one of: a duration of time since a most recent update of the machine learning model meeting a time threshold; a change in baseline heart rate of the patient meeting a change threshold; or a change in patient activity meeting a change threshold.

[0009] In another example, a method includes sensing, via sensing circuitry of an IMD of a medical device system, a first instance of a physiological signal corresponding to a first period of time, the first instance of the physiological signal being indicative of respiration of a patient over the first period of time; predicting, by processing circuitry of the medical device system and using a machine learning model, a second instance of the physiological signal of the patient corresponding to a second period of time based on the first instance of the physiological signal; controlling, by the processing circuitry, therapy delivery circuitry of the IMD to adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal over the second period of time; and controlling, by the processing circuitry, the therapy delivery circuitry to adjust the rate of cardiac pacing based on a third instance of the physiological signal sensed via the sensing circuitry of the IMD and corresponding to a third period of time in response to one of: a duration of time since a most recent update of the machine learning model meeting a time threshold; a change in baseline heart rate of the patient meeting a change threshold; or a change in patient activity meeting a change threshold.

[0010] In another example, a non-transitory computer-readable medium stores instructions that when executed cause processing circuitry to: predict, using a machine learning model, a second instance of a physiological signal of a patient corresponding to a second period of time based on a first instance of the physiological signal, the first instance of a physiological signal being sensed via sensing circuitry of an implantable medical device of the medical device system and being indicative of respiration of the patient over a first period of time; control therapy delivery circuitry of the implantable medical device to adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal over the second period of time; and control the therapy delivery circuitry to adjust the rate of cardiac pacing based on a third instance of the physiological signal sensed via the sensing circuitry ofAttorney Docket No.: A0012462W001 the implantable medical device and corresponding to a third period of time in response to one of: a duration of time since a most recent update of the machine learning model meeting a time threshold; a change in baseline heart rate of the patient meeting a change threshold; or a change in patient activity meeting a change threshold.

[0011] This summary is intended to provide an overview of the subject matter described in this disclosure. It is not intended to provide an exclusive or exhaustive explanation of the apparatus and methods described in detail within the accompanying drawings and description below. Further details of one or more examples are set forth in the accompanying drawings and the description below.BRIEF DESCRIPTION OF DRAWINGS

[0012] FIG. l is a conceptual diagram illustrating an example system configured to deliver cardiac pacing to mimic respiratory sinus arrhythmia (RSA), the system including an implantable medical device (IMD) coupled to implantable medical leads, in accordance with one or more techniques of this disclosure.

[0013] FIG. 2 is a conceptual drawing illustrating the example IMD and leads on FIG. 1 in conjunction with a heart, in accordance with one or more techniques of this disclosure.

[0014] FIG. 3 is a functional block diagram illustrating an example configuration of the IMD of FIG. 1, in accordance with one or more techniques of this disclosure.

[0015] FIG. 4 is a functional block diagram illustrating an example configuration of the computing system of FIG. 1, in accordance with one or more techniques of this disclosure.

[0016] FIG. 5 is a functional block diagram illustrating an example configuration of the external device of FIG. 1, in accordance with one or more techniques of this disclosure.

[0017] FIG. 6 is a flow diagram illustrating an example operation of a device to adjust a rate of cardiac pacing, in accordance with one or more techniques of this disclosure.

[0018] FIG. 7 is a flow diagram illustrating an example operation of a device to determine whether to de-prioritize adjusting cardiac pacing according to a respiratory sinus arrhythmia (RSA) pacing program, in accordance with one or more techniques of this disclosure.

[0019] FIG. 8 is a flow diagram illustrating an example operation of a device to select a physiological signal of a plurality of physiological signals, in accordance with one or more techniques of this disclosure.Attorney Docket No.: A0012462W001

[0020] FIG. 9 is a conceptual diagram illustrating an example machine learning model configured to predict a physiological signal, in accordance with one or more techniques of this disclosure.

[0021] FIG. 10 is a conceptual diagram illustrating an example training process for a machine learning model in accordance with one or more techniques of this disclosure.

[0022] FIGS. 11 A, 1 IB, and 11C are graphical representations of example cardiac pacing waveforms, in accordance with one or more techniques of this disclosure.

[0023] Like reference characters refer to like elements throughout the figures and description.DETAILED DESCRIPTION

[0024] A variety of types of implantable and external devices are configured to monitor health based on sensed physiological signals. External devices that may be used to non- invasively sense and monitor physiological signals include wearable devices with electrodes configured to contact the skin of the patient, such as patches, watches, rings, necklaces, hearing aids, a wearable cardiac monitor or automated external defibrillator (AED), clothing, car seats, or bed linens. Such external devices may facilitate relatively longer-term monitoring of patient health during normal daily activities.

[0025] Implantable medical devices (IMDs) also sense and monitor physiological signals and detect health events such as episodes of arrhythmia, cardiac arrest, myocardial infarction, stroke, and seizure. Example IMDs include pacemakers and implantable cardioverterdefibrillators, which may be coupled to intravascular or extravascular leads, as well as pacemakers with housings configured for implantation within the heart, which may be leadless, such as the Mi era™ leadless pacing device of Medtronic, Inc. Pacemakers provide cardiac pacing pulses to patients based on monitored physiological signals.

[0026] Many current pacemakers pace monotonically in view of pulmonary inspiration / expiration, and the monotonic pacing rate increases with increased activity (e.g., increases with increased activity level but does not vary within the respiratory cycle). Cardiac pacing to mimic RSA, i.e., respirophasic pacing, can provide therapeutic benefit, e.g., by increasing cardiac output and helping to reverse remodel the heart of heart failure (HF) patients. However, current methods of restoring RSA are based on several averaged prior- detected inspiration and expiration phases based on identified respiration cycles, which may mimic RSA when the patient is breathing at a stable rate, e.g., while the patient is sleeping,Attorney Docket No.: A0012462W001 but may not accurately mimic RSA when the patient is breathing at a more variable rate, e.g., when the patient is active.

[0027] Current pacemakers that continuously sense physiological signals may improve RSA pacing accuracy. In some examples, continuously sensing physiological signals may lead to increased power consumption, which may reduce battery life of pacemakers. The techniques of this disclosure may be implemented by a cardiac therapy device, e.g., a pacemaker, that can sense physiological signals, e.g., bioimpedance and / or electrogram (EGM) signals, indicative of inspiration phases and expiration phases and detect inspiration phases and expiration phases. In some examples, the pacemaker may control therapy delivery circuitry to adjust cardiac pacing pulses based on the current inspiration and / or expiration phases. The techniques may additionally include predicting future physiological signals, e.g., indications of inspiration and expiration in physiological signals. In some examples, the techniques include using a machine learning model to predict the physiological signals. Based on the predicted physiological signals, a pacemaker may control therapy delivery circuitry. The pacemaker may alternate between controlling therapy delivery circuitry based on the current, sensed physiological signal and controlling therapy delivery circuitry based on the predicted physiological signals. The machine learning model may be updated periodically and / or in response to changes in patient state, such as changes in patient activity level, changes in patient heart rate, etc., meeting a threshold change.

[0028] By alternating between adjusting cardiac pacing pulses based on the sensed physiological signals and predicted physiological signals, the techniques of this disclosure may maintain accurate RSA pacing while preventing battery drain that may be associated with continuously adjusting cardiac pacing pulses based on sensed physiological signals. As such, patients may receive cardiac pacing that more closely mimics RSA, which may improve patient outcomes.

[0029] RSA pacing may be unnecessary, ineffective, or counterproductive under certain conditions, e.g., when the patient is already achieving RSA or when the patient’s heart rate exceeds a threshold, when the patient is experiencing a cardiac event and / or treatment for the cardiac event, such as anti -tachycardia pacing (ATP), or when patient activity meets a threshold. The techniques of this disclosure may avoid delivering RSA pacing under such conditions by determining to de-prioritize RSA pacing when one or more patient parameters satisfy one or more criterion. In this manner, the techniques described herein may advantageously improve the operation of a device that delivers cardiac pacing to mimic RSA. By determining to de-prioritize an RSA pacing program based on a determination that one orAttorney Docket No.: A0012462W001 more RSA de-prioritization criterion have been met based on a patient state and / or a signal quality, a device configured according to the techniques of this disclosure may deliver RSA pacing at times when RSA pacing is more likely to be effective for the patient and avoid delivery of RSA pacing when RSA pacing is less likely to be effective for the patient, which may improve patient outcomes.

[0030] In some examples, the techniques of this disclosure may additionally include further adjusting cardiac pacing based on patient allometry, e.g., the pacemaker may adjust cardiac pacing differently for different patients based on the allometry of the heart. To adjust pacing based on patient allometry, the techniques of this disclosure can include setting a patient specific rate that is lower than a nominal rate. In some examples, the techniques of this disclosure may prioritize adjusting cardiac pacing based on patient allometry before prioritizing adjusting cardiac pacing based on RSA.

[0031] In some examples, the techniques of this disclosure may include selecting a type of physiological signal to sense for determining respiration of the patient. For example, a pacemaker may be configured to sense a bioimpedance signal and a cardiac electrogram (EGM) signal, and, using a machine learning model, the pacemaker may select the bioimpedance signal or the cardiac EGM signal. For example, the pacemaker may select the physiological signal based on a quality of the physiological signal for the patient and / or a required battery power associated with the physiological signal. In some examples, the physiological signal may switch from one physiological signal type to another, e.g., switching from sensing a bioimpedance signal to an EGM signal, responsive to changes in signal quality. In some examples, some physiological signals may be more noise-resistant than others, e.g., particularly during times of relatively high patient activity, but may also require more battery power. By switching between physiological signals in response to changes in patient activity level, the techniques of this disclosure may increase accuracy and battery life, thereby improving RSA pacing, which may improve patient outcomes.

[0032] FIG. l is a conceptual diagram illustrating an example system 2 configured to deliver cardiac pacing to mimic respiratory sinus arrhythmia (RSA), i.e., respirophasic pacing, in a patient 4, in accordance with one or more techniques of this disclosure. In the example of FIG. 1, system 2 includes an implantable medical device (IMD) 10, which is coupled to leads 18, 20, and 22, and an external device 24. IMD 10 may be, for example, an implantable pacemaker, cardioverter, and / or defibrillator that provides electrical signals to heart 12 via electrodes coupled to one or more of leads 18, 20, and 22. Patient 4 is ordinarily, but not necessarily, a human patient.Attorney Docket No.: A0012462W001

[0033] External device 24 is configured for wireless communication with IMD 10. External device 24 may be configured to communicate with computing system 8 via network 16. In some examples, external device 24 may provide a user interface and allow a user to interact with IMD 10. Computing system 8 may comprise external devices configured to allow a user to interact with IMD 10, or data collected from IMD 10, via network 16.

[0034] IMD 10 and external device 24 may communicate via wireless communication using any techniques known in the art. Examples of communication techniques may include, for example, radiofrequency (RF) telemetry or communication according to a Bluetooth® protocol, but other communication techniques such as magnetic coupling are also contemplated.

[0035] External device 24 may be used to retrieve data from IMD 10 and may transmit the data to computing system 8 via network 16. The retrieved data may include physiological signals recorded by IMD 10. In some examples, computing system 8 includes one or more handheld external devices, computer workstations, servers or other networked external devices. In some examples, computing system 8 may include one or more devices, including processing circuitry and storage devices, that implement a monitoring system 14. Computing system 8 may comprise a cloud computing system. Computing system 8, network 16, and monitoring system 14 may be implemented by the Medtronic CareLink™ Network or other patient monitoring system, in some examples.

[0036] Monitoring system 14 may analyze data received from medical devices, including IMD 10. Monitoring system 14 may implement machine learning models to predict physiological signals based on the data. The machine learning models may include neural networks, deep learning models, convolutional neural networks, or other types of predictive analytics systems.

[0037] Network 16 may include one or more external devices (not shown), such as one or more non-edge switches, routers, hubs, gateways, security devices such as firewalls, intrusion detection, and / or intrusion prevention devices, servers, computer terminals, laptops, printers, databases, wireless mobile devices such as cellular phones or personal digital assistants, wireless access points, bridges, cable modems, application accelerators, or other network devices. Network 16 may include one or more networks administered by service providers and may thus form part of a large-scale public network infrastructure, e.g., the Internet. Network 16 may provide external devices, such as computing system 8 and IMD 10, access to the Internet, and may provide a communication framework that allows the external devices to communicate with one another. In some examples, network 16 may be a private networkAttorney Docket No.: A0012462W001 that provides a communication framework that allows computing system 8, IMD 10, and / or external device 24 to communicate with one another but isolates one or more of computing system 8, IMD 10, or external device 24 from devices external to network 16 for security purposes. In some examples, the communications between computing system 8, IMD 10, and external device 24 are encrypted.

[0038] Computing system 8 is an example of a computing system configured to receive data stored by a medical device of a patient. Computing system 8 may be managed by a manufacturer of IMD 10 to, for example, provide cloud storage and analysis of collected data, maintenance and software services, or other networked functionality for their devices and users thereof. In the example illustrated by FIG. 1, computing system 8 implements a monitoring system 14. As will be described in greater detail below, monitoring system 14 may, in some examples, facilitate prediction of physiological signals of patient 4.

[0039] In the example of FIG. 1, leads 18, 20, 22 extend into the heart 12 of patient 4 to sense electrical activity of heart 12, e.g., one or more bioimpedance signals, one or more cardiac electrogram (EGM) signals, one or more accelerometer signals, and / or deliver electrical stimulation to heart 12. Leads 18, 20, and 22 may also be used to detect a physiological signal indicative of fluid volume in patient 4 and respiration of patient 4.

[0040] In the example shown in FIG. 1, right ventricular (RV) lead 18 extends through one or more veins (not shown), the superior vena cava (not shown), and right atrium 26, and into right ventricle 28. Left ventricular (LV) coronary sinus lead 20 extends through one or more veins, the vena cava, right atrium 26, and into the coronary sinus 30 to a region adjacent to the free wall of left ventricle 32 of heart 12. Right atrial (RA) lead 22 extends through one or more veins and the vena cava, and into the right atrium 26 of heart 12.

[0041] The illustrated number and positions of leads 18, 20, and 22 are examples. In other examples, IMD 10 may be coupled to one, two, or more than three leads that extend to a variety of positions. In some examples, system 2 may additionally or alternatively include one or more leads or lead segments (not shown in FIG. 1) that deploy one or more electrodes within the vena cava, or other veins. Furthermore, in some examples, system 2 may additionally or alternatively include extravascular leads with electrodes implanted outside of heart 12, instead of or in addition to transvenous, intracardiac leads 18, 20 and 22. Such leads may be used for one or more of cardiac sensing, pacing, or cardioversion / defibrillation. Additionally, in some examples, system 2 may include one or more leadless cardiac pacing devices, such as the Micra™ pacemakers commercially available from Medtronic, Inc., instead of or in addition to IMD 10. One or more leadless pacemakers may be configured toAttorney Docket No.: A0012462W001 deliver cardiac pacing according to an RSA mode in the manner described herein with respect to IMD 10. Furthermore, an external medical device may be configured to deliver cardiac pacing according to an RSA mode in the manner described herein with respect to IMD 10. In some examples, a system may additionally or alternatively include one or more implantable or external monitoring devices that monitor patient parameters but do not provide therapy, such as a Reveal LINQ™ insertable cardiac monitor, commercially available from Medtronic, Inc.

[0042] IMD 10 may sense electrical signals attendant to the depolarization and repolarization of heart 12 via electrodes (not shown in FIG. 1) coupled to at least one of the leads 18, 20, 22. In some examples, IMD 10 provides pacing pulses to heart 12 based on the electrical signals sensed within heart 12. The configurations of electrodes used by IMD 10 for sensing and pacing may be unipolar or bipolar. In some examples, IMD 10 may deliver cardiac pacing to provide cardiac resynchronization therapy (CRT). In some examples, IMD 10 may additionally or alternatively be configured to provide conduction system pacing, which may provide a more physiologic activation of heart 12 than conventional pacing. In such examples, leads 18, 20, 22 may be configured / positioned such that their electrode(s) access (are capable of stimulating) the heart’s conduction system, e.g., the His bundle, left bundle branch, or right bundle branch.

[0043] IMD 10 may detect arrhythmia of heart 12, such as tachycardia or fibrillation of the atria 26 and 36 and / or ventricles 28 and 32, and may also provide defibrillation therapy and / or cardioversion therapy via electrodes located on at least one of the leads 18, 20, 22. In some examples, IMD 10 may be programmed to deliver a progression of therapies, e.g., pulses with increasing energy levels, until a fibrillation of heart 12 is stopped. IMD 10 may detect fibrillation by employing one or more fibrillation detection techniques known in the art.

[0044] IMD 10 may utilize two of any electrodes carried on leads 18, 20, 22 to sense physiological signals. In some examples, IMD 10 may also use a housing electrode of IMD 10 (not shown) to sense physiological signals and monitor cardiac activity. Although these physiological signals may be used to monitor heart 12 for therapy, in some examples, IMD 10 may also use any two electrodes of leads 18, 20, and 22 or the housing electrode to sense a bioimpedance of patient 4 to monitor the condition of heart 12. For example, IMD 10 may monitor heart rate, heart rate variability, indicators of blood flow, or other indicators of the ability of heart 12 to pump blood or the progression of heart failure (HF) based on the bioimpedance signal or another sensed signal.Attorney Docket No.: A0012462W001

[0045] As another example, as the tissues within the thoracic cavity of patient 4 increase in fluid content, the impedance between two electrodes may also change. IMD 10 may use this bioimpedance to create a fluid index. As the fluid index increases, more fluid may be more likely to be retained within patient 4 and heart 12 may be stressed to keep up with moving the greater amount of fluid.

[0046] IMD 10 may additionally or alternatively utilize two of any electrodes carried on leads 18, 20, 22 to sense EGM signals. In some examples, IMD 10 may also use a housing electrode of IMD 10 (not shown) to sense EGM signals and monitor cardiac activity. Although these EGM signals may be used to monitor heart 12 for potential arrhythmias and other disorders for therapy, the EGM signals may also be used to monitor the condition of heart 12. For example, IMD 10 may monitor heart rate, heart rate variability, indicators of blood flow, or other indicators of the ability of heart 12 to pump blood or the progression of heart failure (HF) and / or another disease state, e.g., high blood pressure, based on the EGM signal or another sensed signal.

[0047] IMD 10 may communicate with external device 24. In some examples, external device 24 comprises a handheld computing device, computer workstation, or networked computing device. External device 24 may be configured to retrieve data from IMD 10, e.g., for presentation to a clinician or other user, such as the physiological signal of patient 4, and / or data regarding the operation of IMD 10. In some examples, external device 24 may provide the retrieved data to a cloud computing system, such as the CareLink™ system available from Medtronic, Inc., which may analyze the data and provide reports of the analysis and / or the data to clinicians or other users. In some examples, a clinician or other user may also interact with external device 24 to program IMD 10, e.g., select values for operational parameters of IMD 10. Although the user is typically a clinician, the user may be patient 4 in some examples.

[0048] IMD 10 is an example of a device configured to deliver cardiac pacing pulses to a heart of a patient via a plurality of electrodes, sense a physiological signals of patient 4 over a first period of time, predict the physiological signal for a second period of time based on the physiological signal over the first period of time, and adjust cardiac pacing pulses based on the predicted physiological signal during the second period of time. IMD 10 may adjust the cardiac pacing pulses based on the sensed physiological signal during the first period of time.

[0049] IMD 10 may alternate between adjusting the cardiac pacing pulses based on sensed physiological signals and predicted physiological signals based a period schedule, e.g., may switch every 5 minutes, every 10 minutes, or every 30 minutes. Additionally orAttorney Docket No.: A0012462W001 alternatively, IMD 10 may switch from adjusting the cardiac pacing pulses based on the predicted physiological signal to adjusting the cardiac pacing pulses based on the sensed signal in response to one or more of patient 4’s heart rate meeting a heart rate threshold or patient 4’s activity level meeting a threshold.

[0050] FIG. 2 is a conceptual drawing illustrating IMD 10 and leads 18, 20, and 22 of system 2 in greater detail, in accordance with one or more techniques of this disclosure. As shown in FIG. 2, IMD 10 is coupled to leads 18, 20, and 22. Leads 18, 20, 22 may be electrically coupled to therapy delivery circuitry and sensing circuitry of IMD 10 via connector block 34. In some examples, proximal ends of leads 18, 20, 22 may include electrical contacts that electrically couple to respective electrical contacts within connector block 34 of IMD 10. In addition, in some examples, leads 18, 20, 22 may be mechanically coupled to connector block 34 with the aid of set screws, connection pins, snap connectors, or another suitable mechanical coupling mechanism.

[0051] Each of the leads 18, 20, 22 includes an elongated insulative lead body, which may carry a number of concentric coiled conductors separated from one another by tubular insulative sheaths. Bipolar electrodes 40 and 42 are located adjacent to a distal end of lead 18 in right ventricle 28. In addition, bipolar electrodes 44 and 46 are located adjacent to a distal end of lead 20 in coronary sinus 30 and bipolar electrodes 48 and 50 are located adjacent to a distal end of lead 22 in right atrium 26. In the illustrated example, there are no electrodes located in left atrium 33. However, other examples may include electrodes in left atrium 33. Furthermore, in examples in which IMD 10 is configured to deliver conduction system pacing, lead 18 may configured / positioned differently than illustrated in FIG. 2 so that electrode 42 may stimulate the conduction system, e.g., His bundle, left bundle branch, or right bundle branch. For example, electrode 42 may be positioned on or in the ventricular septum.

[0052] Electrodes 40, 44, and 48 may take the form of ring electrodes, and electrodes 42, 46 and 50 may take the form of fixed or extendable helix tip electrodes mounted to insulative electrode heads 52, 54 and 56, respectively. In other examples, one or more of electrodes 42, 46 and 50 may take the form of small circular electrodes at the tip of a tined lead or other fixation element. Leads 18, 20, 22 also include elongated electrodes 62, 64, 66, respectively, which may take the form of a coil. Each of the electrodes 40, 42, 44, 46, 48, 50, 62, 64 and 66 may be electrically coupled to a respective one of the coiled conductors within the lead body of its associated lead 18, 20, 22, and thereby coupled to respective ones of the electrical contacts on the proximal end of leads 18, 20 and 22.Attorney Docket No.: A0012462W001

[0053] In some examples, as illustrated in FIG. 2, IMD 10 includes one or more housing electrodes, such as housing electrode 58, which may be formed integrally with an outer surface of hermetically-sealed housing 60 of IMD 10, or otherwise coupled to housing 60. In some examples, housing electrode 58 is defined by an uninsulated portion of an outward facing portion of housing 60 of IMD 10. Other division between insulated and uninsulated portions of housing 60 may be employed to define two or more housing electrodes. In some examples, housing electrode 58 comprises substantially all of housing 60. As described in further detail with reference to FIG. 3, housing 60 may enclose therapy delivery circuitry configured to generate therapeutic signals, such as cardiac pacing pulses and defibrillation shocks, as well as sensing circuitry for sensing the rhythm of heart 12 and other patient parameters.

[0054] IMD 10 may sense electrical signals attendant to the depolarization and repolarization of heart 12 via electrodes 40, 42, 44, 46, 48, 50, 62, 64 and 66. The electrical signals are conducted to IMD 10 from the electrodes via the respective leads 18, 20, 22. IMD 10 may sense such electrical signals via any bipolar combination of electrodes 40, 42, 44, 46, 48, 50, 62, 64 and 66. Furthermore, any of the electrodes 40, 42, 44, 46, 48, 50, 62, 64 and 66 may be used for unipolar sensing in combination with housing electrode 58. The combination of electrodes used for sensing may be referred to as a sensing configuration or electrode vector.

[0055] In some examples, IMD 10 delivers pacing pulses via bipolar combinations of electrodes 40, 42, 44, 46, 48 and 50 to produce depolarization of cardiac tissue of heart 12. In some examples, IMD 10 delivers pacing pulses via any of electrodes 40, 42, 44, 46, 48 and 50 in combination with housing electrode 58 in a unipolar configuration. Furthermore, IMD 10 may deliver defibrillation pulses to heart 12 via any combination of elongated electrodes 62, 64, 66, and housing electrode 58. Electrodes 58, 62, 64, 66 may also be used to deliver cardioversion pulses to heart 12. Electrodes 62, 64, 66 may be fabricated from any suitable electrically conductive material, such as, but not limited to, platinum, platinum alloy or other materials known to be usable in implantable defibrillation electrodes. The combination of electrodes used for delivery of therapy or sensing, their associated conductors and connectors, and any tissue or fluid between the electrodes, may define an electrical path.

[0056] In some examples, to stimulate and sense a bioimpedance signal, IMD 10 stimulates via a ring electrode located in RV 28, e.g., electrode 40, and a ring electrode located in LV 32, e.g., electrode 44, and senses the bioimpedance signal via a tip electrode located in RV 28, e.g., electrode 42, and a tip electrode located in LV 32, e.g., electrode 46.Attorney Docket No.: A0012462W001IMD 10 may also stimulate via a tip electrode located in RV 28, e.g., electrode 42, and a coil electrode located in RV 28, e.g., electrode 62, and may sense the bioimpedance signal via a ring electrode located in RV 28, e.g., electrode 40, and a coil electrode in RV 28, e.g., electrode 62. IMD 10 may also stimulate via a tip electrode located in right atrium 26, e.g., electrode 50, and a tip electrode in RV 28, e.g., electrode 42, and may sense the bioimpedance signal via a ring electrode located in right atrium 26, e.g., electrode 48, and a ring electrode located in RV 28, e.g., electrode 40. The stimulation and sensing configurations described herein serve merely as examples. Several other stimulation and sensing configurations are also possible.

[0057] In some examples, IMD 10 may additionally or alternatively sense a far-field EGM signal via coil electrode 62 positioned in RV 28 and housing electrode 58.Additionally, or alternatively, IMD 10 may sense the EGM signal via tip electrode 42 and housing electrode 58. Other EGM signal sensing configurations are also possible.

[0058] In addition to sensing bioimpedance to identify inspiration phases and / or expiration phases, any of electrodes 40, 42, 44, 46, 48, 50, 62, 64, 66, and 58 may be used to sense non-cardiac signals. For example, two or more electrodes may be used to measure a bioimpedance, e.g., within the thoracic cavity of patient 4. This bioimpedance may be used to generate a fluid index patient metric that indicates the amount of fluid building up within patient 4. Since a greater amount of fluid may indicate increased pumping loads on heart 12, the fluid index may be used as an indicator of HF risk level. IMD 10 may periodically measure the intrathoracic bioimpedance to identify a trend in the fluid index over days, weeks, months, and even years of patient monitoring.

[0059] In some examples, the two electrodes used to measure the intrathoracic bioimpedance may be located at two different positions within the chest of patient 4. For example, coil electrode 62 and housing electrode 58 may be used as the sensing vector for intrathoracic impedance because electrode 62 is located within RV 28 and housing electrode 58 is located at the IMD 10 implant site generally in the upper chest region. However, other electrodes spanning multiple organs or tissues of patient 4 may also be used, e.g., an additional implanted electrode used only for measuring thoracic bioimpedance.

[0060] FIG. 3 is a functional block diagram illustrating an example configuration of IMD 10, in accordance with one or more techniques of this disclosure. In the illustrated example, IMD 10 includes processing circuitry 380, sensing circuitry 382, one or more sensors 384, therapy delivery circuitry 386, communication circuitry 388, and memory 390. Memory 390 includes computer-readable instructions that, when executed by processing circuitry 380,Attorney Docket No.: A0012462W001 cause IMD 10 and processing circuitry 380 to perform various functions attributed to IMD 10 and processing circuitry 380 herein. Memory 390 may include any volatile, non-volatile, magnetic, optical, or electrical media, such as a random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), electrically-erasable programmable ROM (EEPROM), flash memory, or any other digital or analog media.

[0061] Processing circuitry 380 may include any one or more of a microprocessor, a controller, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or equivalent discrete or analog logic circuitry. In some examples, processing circuitry 380 may include multiple components, such as any combination of one or more microprocessors, one or more controllers, one or more DSPs, one or more ASICs, or one or more FPGAs, as well as other discrete or integrated logic circuitry. The functions attributed to processing circuitry 380 herein may be embodied as software, firmware, hardware or any combination thereof, e.g., may be embodied as software or firmware executed on processing circuitry.

[0062] Processing circuitry 380 controls therapy delivery circuitry 386 to deliver therapy to heart 12 according to a therapy parameters and programs which may be stored in memory 390. An example of therapy parameters stored in memory 390 are RSA pacing parameters 396 for delivery of cardiac pacing. RSA pacing parameters 396 may include timing and duration parameters, such as an amount of time to increase and / or decrease a rate of cardiac pacing pulses to mimic RSA or an extent to which to increase and / or decrease a rate of cardiac pacing pulses to mimic RSA. Therapy delivery circuitry 386 is electrically coupled to electrodes 40, 42, 44, 46, 48, 50, 58, 62, 64, and 66, e.g., via conductors of the respective lead 18, 20, 22, or, in the case of housing electrode 58, via an electrical conductor disposed within housing 60 of IMD 10. In the illustrated example, therapy delivery circuitry 386 is configured to generate and deliver electrical therapy to heart 12. For example, therapy delivery circuitry 386 may deliver defibrillation shocks to heart 12 via at least two electrodes 58, 62, 64, 66. Therapy delivery circuitry 386 may deliver pacing pulses via ring electrodes 40, 44, 48 coupled to leads 18, 20, and 22, respectively, and / or helical electrodes 42, 46, and 50 of leads 18, 20, and 22, respectively. In some examples, therapy delivery circuitry 386 delivers pacing, cardioversion, or defibrillation stimulation in the form of electrical pulses. In other examples, therapy delivery circuitry 386 may deliver one or more of these types of stimulation in the form of other signals, such as sine waves, square waves, or other substantially continuous time signals.Attorney Docket No.: A0012462W001

[0063] Therapy delivery circuitry 386 includes circuitry, such as charge pumps, capacitors, current mirrors, or other signal generation circuitry for generating a pulse or other signal. Therapy delivery circuitry 386 may include a switch module, and processing circuitry 380 may use the switch module to select, e.g., via a data / address bus, which of the available electrodes are used to deliver antitachyarrhythmia shocks or pacing pulses. The switch module may include a switch array, switch matrix, multiplexer, or any other type of switching device suitable to selectively couple stimulation energy to selected electrodes.

[0064] Sensing circuitry 382 monitors signals from at least one of electrodes 40, 42, 44, 46, 48, 50, 58, 62, 64 or 66 to monitor one or more physiological signal(s) of the heart and / or electrical activity of heart 12, respiration of patient 4, or other patient parameters, values of which may be stored as patient parameter data 392 in memory 390. Sensing circuitry 382 may detect intrinsic cardiac depolarizations, determine heart rates or heart rate variability, or may detect arrhythmias or other electrical signals. Sensing circuitry 382 may include one or more filters, amplifiers, analog-to-digital converters, or other sensing circuitry.

[0065] Sensing circuitry 382 may also include a switch module to select which of the available electrodes are used to sense the heart activity, depending upon which electrode combination, or electrode vector, is used in the current sensing configuration. In some examples, processing circuitry 380 may select the electrodes that function as sense electrodes, i.e., select the sensing configuration, via the switch module within sensing circuitry 382. Sensing circuitry 382 may include one or more detection channels, each of which may be coupled to a selected electrode configuration for detection of cardiac signals via that electrode configuration. Some detection channels may be configured to detect cardiac events, such as P-waves or R-waves, and provide indications of the occurrences of such events to processing circuitry 380.

[0066] Processing circuitry 380 may be configured to determine respiration of patient 4 based on one or more of a bioimpedance signal, EGM signal, accelerometer signal, or another physiological signal. In some examples, processing circuitry 380 may additionally or alternatively sense a photoplethysmography (PPG) signal, a heart sounds signal, and / or a hemodynamic signal. The plurality of physiological signals may measure intracardiac and / or arterial pressures directly or may serve as surrogates indicative of intracardiac and / or arterial pressures. In some examples, sensing circuitry 382 may sense other physiological signals containing a respiratory signature. To determine respiration, processing circuitry 380 may, for example, detect peaks and troughs in the physiological signal, e.g., the bioimpedance signal, by identifying zero slope points (zero crossings in a derivative or differential of the signal), orAttorney Docket No.: A0012462W001 using any other peak / trough detection techniques. Processing circuitry 380 may determine an expiration phase as an interval or window from an identified peak to a subsequent trough, and an inspiration phase as an interval or window from an identified trough to a subsequent peak. Processing circuitry 380 may determine respiration effort based on one or more of a peak-to-trough amplitude or a slope of the signal within the inspiration phase. Processing circuitry 380 may determine tidal volume based on an area under the curve during the respiration cycle. In some examples, processing circuitry 380 may determine tidal volume based on a peak-to-trough amplitude, which may vary with tidal volume. Based on the sensed physiological signal, and, in some examples, the determined inspiration and / or expiration phases, processing circuitry 380 may control therapy delivery circuitry 386 to adjust cardiac pacing pulses over a period of time corresponding to the sensed physiological signal. In some examples, processing circuitry 380 may implement an application 370 to apply machine learning model(s) 374 of monitoring system 372 to predict the physiological signal over a subsequent period of time based on the sensed physiological signal over the first period of time.

[0067] Memory 390 may store applications 370 executable by processing circuitry 380. Applications 370 may include monitoring system 372. Processing circuitry 380 may execute monitoring system 372 to predict a physiological signal of patient 4.

[0068] As an example, monitoring system 372 may include machine learning model(s) 374, which may be configured to predict a physiological signal of patient 4. In some examples, monitoring system 372 may additionally detect cardiac events of patient 4, as well as potential device function issues, such as sensing issues, e.g., oversensing, under-sensing, lead fracture, or loss of capture, and / or battery depletion. Processing circuitry 380 may determine to update machine learning model(s) 374 on a periodic schedule, e.g., daily or weekly, based on sensed physiological signal data.

[0069] Machine learning model(s) 374 may additionally or alternatively be configured to select a physiological signal type for subsequent use in adjusting cardiac pacing pulses. As an example, IMD 10 may be configured to sense a plurality of physiological signals, e.g., a bioimpedance signal, a cardiac EGM, and an accelerometer signal. Processing circuitry 380 may implement machine learning model(s) 374 to select a physiological signal, e.g., the bioimpedance signal, for use in adjusting the cardiac pacing pulses delivered by therapy delivery circuitry 386. Machine learning model(s) 374 may be configured to select the physiological signal based on one or more of signal quality or associated battery usage for each of the plurality of physiological signals.Attorney Docket No.: A0012462W001

[0070] One or more sensor(s) 384 may include, as examples, one or more accelerometers, microphones, temperature sensors, or optical sensors that are configured to provide signals or data representing one or more patient parameters to processing circuitry 380 via sensing circuitry 382. In some examples, based on a signal from one or more accelerometers, processing circuitry 380 may determine postures and / or activity levels of patient 4. In some examples, processing circuitry 380 may determine the physiological signal using one or more of sensor(s) 384. For example, processing circuitry 380 may select the physiological signal to be the accelerometer signal. In some examples, processing circuitry 380 may select multiple physiological signals, such as the accelerometer signal in addition to a bioimpedance and / or EGM signal. In some examples, local minima and local maxima in the accelerometer signal may be indicative of beginnings of inspiration phases and expiration phases, respectively.

[0071] Processing circuitry 380 may implement programmable counters that control the basic time intervals associated with DDD, VVI, DVI, VDD, AAI, DDI, DDDR, VVIR, DVIR, VDDR, AAIR, DDIR, CRT, and other modes of pacing. Intervals defined by processing circuitry 80 may include atrial and ventricular pacing escape intervals, A-V intervals, V-V intervals, and refractory periods during which sensed P-waves and R-waves are ineffective to restart timing of the intervals. The durations of these intervals may be determined by processing circuitry 380 in response to stored data in memory 390.

[0072] In some examples, processing circuitry 380 may modify escape intervals based on a rate responsive pacing mode. Processing circuitry 380 may determine a sensor indicated pacing rate based on sensed parameters of patient 4, such as one or more of activity level or respiration rate, and thereby modify the escape interval and pacing rate to provide cardiac pacing that supports the activity of patient 4. In some examples, processing circuitry 380 may additionally modify the pacing mode based on a patient disease state or other patient data. In some examples, processing circuitry 380 may control IMD 10 to provide CRT by controlling delivery of pacing pulses to one or both of RV 28 and LV 32 based on atrioventricular timing and interventricular timing specified by one or more A-V intervals and V-V intervals.

[0073] Interval counters implemented by processing circuitry 380 may be reset upon sensing of R-waves and P-waves with detection channels of sensing circuitry 382. In examples in which IMD 10 provides pacing, therapy delivery circuitry 86 may include pacer output circuits that are coupled, e.g., selectively by a switching module, to any combination of electrodes 40, 42, 44, 46, 48, 50, 58, 62, or 66 appropriate for delivery of a bipolar or unipolar pacing pulse to one of the chambers of heart 12. In such examples, processing circuitry 380 may reset the interval counters upon the generation of pacing pulses by therapyAttorney Docket No.: A0012462W001 delivery circuitry 386, and thereby control the basic timing of cardiac pacing functions, including anti-tachyarrhythmia pacing.

[0074] The value of the count present in the interval counters when reset by sensed R- waves and P-waves may be used by processing circuitry 380 to measure the durations of R-R intervals, P-P intervals, P-R intervals and R-P intervals, which are measurements that may be stored in memory 390. Processing circuitry 380 may use the count in the interval counters to detect a tachyarrhythmia event, such as atrial fibrillation (AF), atrial tachycardia (AT), ventricular fibrillation (VF), or ventricular tachycardia (VT). These intervals may also be used to detect the overall heart rate, ventricular contraction rate, and heart rate variability. A portion of memory 390 may be configured as a plurality of recirculating buffers, capable of holding series of measured intervals, which may be analyzed by processing circuitry 380 in response to the occurrence of a pace or sense interrupt to determine whether the patient’s heart 12 is presently exhibiting atrial or ventricular tachyarrhythmia.

[0075] In some examples, processing circuitry 380 may determine that tachyarrhythmia has occurred by identification of shortened R-R (or P-P) interval lengths. Generally, processing circuitry 380 detects tachycardia when the interval length falls below 220 milliseconds (ms) and fibrillation when the interval length falls below 180 ms. These interval lengths are merely examples, and a user may define the interval lengths as desired, which may then be stored within memory 390. This interval length may need to be detected for a certain number of consecutive cycles, for a certain percentage of cycles within a running window, or a running average for a certain number of cardiac cycles, as examples.

[0076] In the event that processing circuitry 380 detects an atrial or ventricular tachyarrhythmia based on signals from sensing circuitry 382, and an anti -tachyarrhythmia pacing regimen is desired, timing intervals for controlling the generation of anti -tachyarrhythmia pacing therapies by therapy delivery circuitry 386 may be loaded by processing circuitry 380 to control the operation of the escape interval counters therein and to define refractory periods during which detection of R-waves and P-waves is ineffective to restart the escape interval counters for the an anti-tachyarrhythmia pacing. In the event that processing circuitry 380 detects an atrial or ventricular tachyarrhythmia based on signals from sensing circuitry 382, and a cardioversion or defibrillation shock is desired, processing circuitry 380 may control the amplitude, form and timing of the shock delivered by therapy delivery circuitry 386.

[0077] Memory 390 may be configured to store a variety of operational parameters, therapy parameters, sensed and detected data, and any other information related to theAttorney Docket No.: A0012462W001 therapy and treatment of patient 4. In the example of FIG. 3, memory 390 includes patient parameter data 392, pacing programs 394, and RSA pacing parameters 396. Patient parameter data 392 may store all of the data generated from the sensing and detecting of patient parameters described herein, such as heart rates, inspiration phases and / or expiration phases, activity, posture, fluid index, an atrial tachycardia or fibrillation burden, a ventricular contraction rate during atrial fibrillation, a nighttime heart rate, a difference between night and day heart rate, a heart rate variability, a cardiac resynchronization therapy percentage, a bradyarrhythmia pacing therapy percentage (in a ventricle and / or atrium), and number or frequency of electrical shock events, blood pressure, right ventricular pressure, pulmonary artery pressure, patient temperature, or biomarkers such as a brain natriuretic peptide (BNP), troponin, or related surrogates.

[0078] Pacing programs 394 may include a plurality of pacing programs, including an RSA pacing program. In some examples, pacing programs 394 may include a patient allometry program, an ATP program, and an intrinsic RSA program. In some examples, pacing programs 394 may be patient specific. In some examples, the patient allometry program may include patient specific rate that is lower than and different from a nominal rate of adjusting cardiac pacing. Processing circuitry 380 may be configured to select and / or prioritize a pacing program based on patient parameter data 392 and / or the physiological signal of patient 4. Processing circuitry 380 may prioritize one or more of pacing programs 394 over the RSA pacing programming depending on patient parameter data 392 and the physiological signal. As an example, if processing circuitry 380 determines patient 4 is achieving intrinsic RSA, processing circuitry 380 may prioritize the intrinsic RSA program.

[0079] RSA pacing parameters 396 includes one or more criteria that processing circuitry 380 may apply to patient parameter data 392 to determine whether to adjust cardiac pacing to mimic RSA. Processing circuitry 80 may adjust pacing to mimic RSA if patient parameter data 392 satisfies RSA pacing parameters 396. RSA pacing parameters 396 may be fixed, programmable by a user, or variable based on conditions determined by processing circuitry 380. To adjust cardiac pacing to mimic RSA, processing circuitry 380 control therapy delivery circuitry 386 to deliver pacing pulses, according to RSA pacing parameters 396, based on the physiological signal. In some examples, processing circuitry 380 controls therapy delivery circuitry 386 to deliver pacing pulses with increasing rates during an inspiration phase of a respiratory cycle, and, in some examples, decreasing rates during an expiration phase of the cardiac cycle, as described herein. The increasing and decreasing of pacing rates may be sequential, on a beat-to-beat or other basis.Attorney Docket No.: A0012462W001

[0080] Communication circuitry 388 includes any suitable hardware, firmware, software or any combination thereof for communicating with another device, such as external device 24 (FIG. 1). Under the control of processing circuitry 380, communication circuitry 388 may communicate with external device 24 with the aid of an antenna, which may be internal and / or external.

[0081] FIG. 4 is a functional block diagram illustrating an example configuration of the computing system of FIG. 1, in accordance with one or more techniques of this disclosure. FIG. 4 is a block diagram illustrating an example configuration of computing system 8, in accordance with one or more techniques of this disclosure. In the illustrated example, computing system 8 includes processing circuitry 402 for executing applications 424 that include monitoring system 450 or any other applications described herein. Computing system 8 may be any component or system that includes processing circuitry or other suitable computing environment for executing software instructions and, for example, need not necessarily include one or more elements shown in FIG. 4 (e.g., input devices 404, communication circuitry 406, user interface devices 410, or output devices 412; and in some examples components such as storage device(s) 408 may not be co-located or in the same chassis as other components). In some examples, computing system 8 may be a cloud computing system distributed across a plurality of devices.

[0082] In the example of FIG. 4, computing system 8 includes processing circuitry 402, one or more input devices 404, communication circuitry 406, one or more storage device(s) 408, user interface (UI) device(s) 410, and one or more output devices 412. Computing system 8, in some examples, further includes one or more application(s) 424 such as monitoring system 450, and operating system 416 that are executable by computing system 8. Each of components 402, 404, 406, 408, 410, and 412 are coupled (physically, communicatively, and / or operatively) for inter-component communications. In some examples, communication channels 414 may include a system bus, a network connection, an inter-process communication data structure, or any other method for communicating data. As one example, components 402, 404, 406, 408, 410, and 412 may be coupled by one or more communication channels 414.

[0083] Processing circuitry 402, in one example, is configured to implement functionality and / or process instructions for execution within computing system 8. For example, processing circuitry 402 may be capable of processing instructions stored in storage device(s) 408. Examples of processing circuitry 402 may include any one or more of a microprocessor, a controller, a digital signal processor (DSP), an application specific integrated circuitAttorney Docket No.: A0012462W001(ASIC), a field-programmable gate array (FPGA), or equivalent discrete or integrated logic circuitry.

[0084] One or more storage device(s) 408 may be configured to store information within computing system 8 during operation. Storage device(s) 408, in some examples, is described as a computer-readable storage medium. In some examples, storage device(s) 408 is a temporary memory, meaning that a primary purpose of storage device(s) 408 is not long-term storage. Storage device(s) 408, in some examples, is described as a volatile memory, meaning that storage device(s) 408 does not maintain stored contents when the computer is turned off. Examples of volatile memories include random access memories (RAM), dynamic random access memories (DRAM), static random access memories (SRAM), and other forms of volatile memories known in the art. In some examples, storage device(s) 408 is used to store program instructions for execution by processing circuitry 402. Storage device(s) 408, in one example, is used by software or applications 424 running on computing system 8 to temporarily store information during program execution.

[0085] Storage device(s) 408, in some examples, also include one or more computer- readable storage media. Storage device(s) 408 may be configured to store larger amounts of information than volatile memory. Storage device(s) 408 may further be configured for longterm storage of information. In some examples, storage device(s) 408 include non-volatile storage elements. Examples of such non-volatile storage elements include magnetic hard discs, optical discs, floppy discs, flash memories, or forms of electrically programmable memories (EPROM) or electrically erasable and programmable memories (EEPROM).

[0086] Computing system 8, in some examples, also includes communication circuitry 406 to communicate with other devices and systems, such as HMD 10 and external device 24 of FIG. 1, as well as other networked client external devices of various users. Communication circuitry 406 may include a network interface card, such as an Ethernet card, an optical transceiver, a radio frequency transceiver, or any other type of device that can send and receive information. Other examples of such network interfaces may include 3G and Wi-Fi radios.

[0087] Computing system 8, in one example, also includes one or more user interface devices 410. User interface devices 410, in some examples, are configured to receive input from a user through tactile, audio, or video feedback. Examples of user interface devices(s) 410 include a presence-sensitive display, a mouse, a keyboard, a voice responsive system, video camera, microphone or any other type of device for detecting a command from a user. In some examples, a presence-sensitive display includes a touch-sensitive screen.Attorney Docket No.: A0012462W001

[0088] One or more output device(s) 412 may also be included in computing system 8. Output device(s) 412, in some examples, is configured to provide output to a user using tactile, audio, or video stimuli. Output device(s) 412, in one example, includes a presencesensitive display, a sound card, a video graphics adapter card, or any other type of device for converting a signal into an appropriate form understandable to humans or machines. Additional examples of output device(s) 412 include a speaker, a cathode ray tube (CRT) monitor, a liquid crystal display (LCD), or any other type of device that can generate intelligible output to a user.

[0089] Computing system 8 may include operating system 416. Operating system 416, in some examples, controls the operation of components of computing system 8. For example, operating system 416, in one example, facilitates the communication of one or more applications 424 and monitoring system 450 with processing circuitry 402, communication circuitry 406, storage device(s) 408, input devices 404, user interface devices 410, and output devices 412.

[0090] Applications 424 may also include program instructions and / or data that are executable by computing system 8. Example application(s) 424 executable by computing system 8 may include monitoring system 450. Other additional applications not shown may alternatively or additionally be included to provide other functionality described herein and are not depicted for the sake of simplicity.

[0091] In accordance with the techniques of the disclosure, computing system 8 receives physiological signal data sensed by IMD 10 via communication circuitry 406, e.g., to predict a subsequent physiological signal and / or to select a physiological signal type. Monitoring system 450, as implemented by computing system 8 including processing circuitry 402 and storage device(s) 408, controls the prediction of physiological signals, and, in some examples, determines whether to prioritize RSA pacing and / or determines a type of physiological signal for use in adjusting cardiac pacing pulses.

[0092] Processing circuitry 402 may apply machine learning models 452 and apply the physiological data, e.g., a first instance of the physiological data corresponding to a first period of time, as an input to the one or more machine learning model(s) 452. Machine learning model(s) 452 may be configured to output a predicted physiological signal, e.g., a second instance of the physiological signal corresponding to a second period of time, based on the input physiological signal. Additionally, or alternatively, machine learning model(s) 452 may be configured to select a type of physiological signal of a plurality of types of physiological signals for use in adjusting cardiac pacing pulses. Machine learning model(s)Attorney Docket No.: A0012462W001452 may additionally or alternatively be configured to determine whether to prioritize RSA pacing based on the physiological signal and / or one or more additional signals, e.g., a signal indicative of patient heart rate, such as a cardiac EGM signal.

[0093] Machine learning model(s) 452 may include, as examples, neural networks, such as deep neural networks, which may include convolutional neural networks, multi-layer perceptrons, transformers, recurrent neural networks, and / or echo state networks, as examples. For example, machine learning models 452 may include a long short-term memory (LSTM) recurrent neural network. As another example, machine learning model(s) 452 may include a seasonal and trend decomposition using Loess (STL decomposition).

[0094] FIG. 5 is a functional block diagram illustrating an example configuration of external device 24 of FIG. 1, in accordance with one or more techniques of this disclosure. External device 24 is configured to communicate with IMD 10, in accordance with one or more techniques of this disclosure. In the example of FIG. 5, external device 24 includes processing circuitry 502, communication circuitry 506, user interface 504, power source 510, and memory 508.

[0095] Processing circuitry 502, in one example, may include one or more processors that are configured to implement functionality and / or process instructions for execution within external device 24. For example, processing circuitry 502 may be capable of processing instructions stored in memory 508. Processing circuitry 502 may include, for example, microprocessors, DSPs, ASICs, FPGAs, GPUs, TPUs, or equivalent discrete or integrated logic circuitry, or a combination of any of the foregoing devices or circuitry. Accordingly, processing circuitry 502 may include any suitable structure, whether in hardware, software, firmware, or any combination thereof, to perform the functions ascribed herein to processing circuitry 502.

[0096] Communication circuitry 506 may include any suitable hardware, firmware, software or any combination thereof for communicating with another device, such as IMD 10. Under the control of processing circuitry 502, communication circuitry 506 may receive downlink telemetry from, as well as send uplink telemetry to, IMD 10, or another device, such as computing system 8.

[0097] A user, such as a clinician or patient 4, may interact with external device 24 through user interface 504. User interface 504 includes a display (not shown), such as an LCD or LED display or other type of screen, with which processing circuitry 502 may present information related to IMD 10. In addition, user interface 504 may include an input mechanism to receive input from the user. The input mechanisms may include, for example,Attorney Docket No.: A0012462W001 any one or more of buttons, a keypad (e.g., an alphanumeric keypad), a peripheral pointing device, a touch screen, or another input mechanism that allows the user to navigate through user interfaces presented by processing circuitry 502 of external device 24 and provide input. In other examples, user interface 504 also includes audio circuitry for providing audible notifications, instructions or other sounds to patient 4, receiving voice commands from patient 4, or both. Memory 508 may include instructions for operating user interface 504 and for managing power source 510.

[0098] Power source 510 is configured to deliver operating power to the components of external device 24. Power source 510 may include a battery and a power generation circuit to produce the operating power. In some examples, the battery is rechargeable to allow extended operation. Recharging may be accomplished by electrically coupling power source 510 to a cradle or plug that is connected to an alternating current (AC) outlet. In addition, recharging may be accomplished through proximal inductive interaction between an external charger and an inductive charging coil within external device 24. In other examples, traditional batteries (e.g., nickel cadmium or lithium ion batteries) may be used. In addition, external device 24 may be directly coupled to an alternating current outlet to operate.

[0099] Memory 508 may be configured to store information within external device 24 during operation. In some examples, memory 508 may be referred to as a storage device and include computer-readable instructions that, when executed by processing circuitry 502, cause external device 24 and processing circuitry 502 to perform various functions attributed to external device 24 and processing circuitry 502 herein. Memory 508 may include any volatile, non-volatile, magnetic, optical, or electrical media, such as RAM, DRAM, SRAM, magnetic discs, optical discs, flash memories, ROM, NVRAM, EPROM, EEPROM, flash memory, or any other digital media. Memory 508 may also store data generated by sensing circuitry 382 of IMD 10, such as signals corresponding to indications of detections.

[0100] Memory 508 may be configured to store information within external device 24 during operation. Memory 508, in some examples, is described as a computer-readable storage medium. In some examples, memory 508 is a temporary memory, meaning that a primary purpose of memory 508 is not long-term storage. Memory 508, in some examples, is described as a volatile memory, meaning that memory 508 does not maintain stored contents when the computer is turned off. Examples of volatile memories include random access memories (RAM), dynamic random access memories (DRAM), static random access memories (SRAM), and other forms of volatile memories known in the art. In some examples, memory 508 is used to store program instructions for execution by processingAttorney Docket No.: A0012462W001 circuitry 502. Memory 508, in one example, is used by software or applications 512 running on external device 24 to temporarily store information during program execution.

[0100] Memory 508 may, in some examples, also include one or more computer-readable storage media. Memory 508 may be configured to store larger amounts of information than volatile memory. Memory 508 may further be configured for long-term storage of information. In some examples, Memory 508 include non-volatile storage elements. Examples of such non-volatile storage elements include magnetic hard discs, optical discs, floppy discs, flash memories, or forms of electrically programmable memories (EPROM) or electrically erasable and programmable memories (EEPROM).

[0101] Applications 512 may also include program instructions and / or data that are executable by external device 24. Example application(s) 512 executable by external device 24 may include monitoring system 514. Other additional applications not shown may alternatively or additionally be included to provide other functionality described herein and are not depicted for the sake of simplicity.

[0102] In accordance with the techniques of the disclosure, external device 24 receives physiological signal data sensed by IMD 10 via communication circuitry 506, e.g., to predict a subsequent physiological signal and / or to select a physiological signal type. Monitoring system 514, as implemented by external device 24 including processing circuitry 502 and memory 508, controls the prediction of physiological signals, and, in some examples, determines whether to prioritize RS A pacing and / or determines a type of physiological signal for use in adjusting cardiac pacing pulses.

[0103] Processing circuitry 502 may apply machine learning models 516 and apply the physiological data, e.g., a first instance of the physiological data corresponding to a first period of time, as an input to the one or more machine learning models 516. Machine learning models 516 may be configured to output a predicted physiological signal, e.g., a second instance of the physiological signal corresponding to a second period of time, based on the input physiological signal. Additionally, or alternatively, machine learning models 516 may be configured to select a type of physiological signal of a plurality of types of physiological signals for use in adjusting cardiac pacing pulses. Machine learning models 516 may additionally or alternatively be configured to determine whether to prioritize RSA pacing based on the physiological signal and / or one or more additional signals, e.g., a signal indicative of patient heart rate, such as a cardiac EGM signal.

[0104] Machine learning models 516 may include, as examples, neural networks, such as deep neural networks, which may include convolutional neural networks, multi-layerAttorney Docket No.: A0012462W001 perceptrons, transformers, recurrent neural networks, and / or echo state networks, as examples. For example, machine learning models 516 may include an LSTM recurrent neural network. As another example, machine learning models 452 may include an STL decomposition neural network.

[0105] FIG. 6 is a flow diagram illustrating an example operation of a device to adjust a rate of cardiac pacing, in accordance with one or more techniques of this disclosure. Sensing circuitry 382 of IMD 10 senses a first instance of a physiological signal, e.g., a bioimpedance signal, an accelerometer signal, and / or a cardiac EGM signal, corresponding to a first period of time. In some examples, the first period of time comprises a monitoring period. Processing circuitry of system 2, e.g., processing circuitry 380 of IMD 10, predicts, by a machine learning model, a second instance of a physiological signal of a patient, e.g., patient 4, corresponding to a second period of time based on the sensed first instance of the physiological signal corresponding to the first period of time (602). In some examples, during the first period of time, processing circuitry 380 controls therapy delivery circuitry 386 to adjust a rate of cardiac pacing during the first period of time based on the sensed first instance of the physiological signal.

[0106] In some examples, the machine learning model may be a time series forecasting model, such as an LSTM neural network or an STL neural network. In examples in which the machine learning model comprises an LSTM neural network and the physiological signal comprises a bioimpedance signal, processing circuitry 380 may implement the machine learning model by normalizing the bioimpedance signal and reshaping the bioimpedance signal into a 3D array or otherwise preparing the bioimpedance signal to be input into an artificial intelligence (Al) routine / library, such as TensorFlow. The model defines a sequential model including a linear stack of layers, e.g., an LSTM layer with 100 network units. A return sequence value is set to “True” to cause the output of the LSTM layer to be another sequence of the same length, and another LSTM layer with 100 network units can be added to the sequential model with a return sequence value set to “False” so that only the last output in the resulting output sequence is returned. A dense neural network layer with 25 network units can be added to the sequential model, followed by another dense layer specifying the output of 1 network unit. The machine learning model then trains the sequential model using the prepared sensed bioimpedance signal and / or the bioimpedance signal features.

[0107] Processing circuitry 380 controls therapy delivery circuitry 386 to adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal over theAttorney Docket No.: A0012462W001 second period of time (604). In some examples, processing circuitry 380 controls therapy delivery circuitry 386 to adjust the rate of cardiac pacing according to an RS A therapy program. Adjusting the rate of cardiac pacing according to the RSA therapy program may include pacing at a relatively higher rate during inspiration of patient 4 than during expiration of patient 4. For example, therapy delivery circuitry 386 may increase the rate of cardiac pacing relative to a baseline during inspiration and may decrease the rate of cardiac pacing relative to the baseline during expiration.

[0108] During the second period of time, processing circuitry 380 may monitor for changes in patient heart rate and patient activity level. As an example, processing circuitry 380 may determine a median cycle length of a plurality of previous heartbeats of patient 4. As another example, processing circuitry 380 may determine a median activity count averaged over an interval of time, e.g., an interval of time that is shorter than the second period of time, of patient 4.

[0109] Processing circuitry 380 may additionally determine whether the second period of time has ended. The second period of time may be a predetermined amount of time, e.g., 5 minutes, 30 minutes, 1 hour, or several hours. If processing circuitry 380 determines none of a change in heart rate criterion, a change in patient activity level criterion, or an expiration of the second period criterion has been met (“NO” of 606), processing circuitry 380 continues to control therapy delivery circuitry 386 to adjust the rate of cardiac pacing based on the predicted second instance of the physiological signal.

[0110] If the second period of time ends, thereby meeting the expiration of the second period criterion, and / or a change in heart rate and / or a change in patient activity level meets the corresponding change criterion, e.g., if the median cycle length of the plurality of previous heartbeats of patient 4 and / or the median activity count averaged over the interval of time changes by a threshold amount and / or the median cycle length and / or median activity count meets a respective threshold value (“YES” of 606), processing circuitry 380 may determine to control therapy delivery circuitry 386 to adjust the rate of cardiac pacing based on a sensed third instance of the physiological signal corresponding to a third period of time (608). In examples in which the second period of time ends before any other criteria are met, the third period of time and the second period of time may not overlap. In examples in which the patient activity level criterion or the patient heart rate criterion is met, the second period of time and the third period of time may overlap. Said another way, the anticipated duration of the second period of time may be shortened if the patient activity level criterion or the patient heart rate criterion is met before the expiration of the second period of time.Attorney Docket No.: A0012462W001Processing circuitry 380 may determine to control therapy delivery circuitry 386 based on the sensed third instance of the signal instead of the predicted second instance of the signal over a period of time including some portion of the anticipated duration of the second period of time.[oni] In some examples, processing circuitry 380 updates the machine learning model based on the third instance of the physiological signal, e.g., processing circuitry 380 may provide the third instance of the physiological signal as input to the machine learning model (610). In some examples, processing circuitry 380 may additionally update the machine learning model based on a physiological signal sensed during the second period of time. In some examples, sensing circuitry 382 may continue to sense the physiological signal during the second period of time, and processing circuitry 380 may store the sensed signal without using the sensed signal to control therapy delivery circuitry 386. In some examples, processing circuitry 380 updates the machine learning model periodically, e.g., daily.

[0112] Although the example operation of FIG. 6 is described with respect to processing circuitry 380, the techniques of this disclosure are not so limited. In some examples, the machine learning model may be edge-based or cloud-based. As an example, the machine learning model may be machine learning model(s) 516 of external device 24 if the machine learning model is edge-based or may be machine learning model(s) 452 of computing system 8 if the machine learning model is cloud-based. Communication circuitry 388 may transmit the physiological signal, features of the physiological signal, and / or one or more additional signals and / or patient information, to one or more of external device 24 or computing system 8 to provide input to machine learning model(s) 516 or machine learning model(s) 452.

[0113] FIG. 7 is a flow diagram illustrating an example operation of a device to determine whether to de-prioritize adjusting cardiac pacing according to an RSA pacing program, in accordance with one or more techniques of this disclosure. Processing circuitry of system 2, e.g., processing circuitry 380 of IMD 10, determines one or more of a patient state or a signal quality of one or more of the physiological signal, e.g., the third instance of the physiological signal, or one or more additional signals, e.g., one or more additional signals corresponding to the third period of time (702). In some examples, the one or more additional signals may include a cardiac EGM signal. Based on one or more of the patient state or the signal quality, processing circuitry 380 determines whether one or more RSA deprioritization criterion are met (704). The one or more RSA de-prioritization criterion may include one or more of a change in patient state criterion, e.g., a patient disease state criterion, a cardiac episode criterion, an intrinsic RSA criterion, a patient heart rate criterion, or aAttorney Docket No.: A0012462W001 patient activity level criterion, or a signal quality criterion, such as a signal to noise ratio (SNR) criterion.

[0114] In some examples, processing circuitry of system 2 may be configured to determine a patient state of patient 4 based on one or more of historical patient data, the physiological signal, or one or more additional physiological signals. The patient state may be indicative of a patient disease state, a patient activity level, a patient heart rate, a cardiac episode, and / or intrinsic RSA.

[0115] The patient disease state may, in some examples, worsen or improve over time. If the patient disease state changes by a threshold amount, e.g., if the patient disease state worsens by a threshold amount, processing circuitry 380 may determine the patient state criterion has been met. In some examples, if patient 4’s heart rate and / or activity level is above a corresponding threshold and / or has changed by a threshold amount, processing circuitry 380 may determine the patient heart rate and / or activity level criterion has been met. In some examples, processing circuitry 380 may determine patient 4 is experiencing a cardiac episode, such as a tachyarrhythmia episode.

[0116] Based on a classification of the cardiac episode, processing circuitry 380 may determine the patient state criterion has been met. In some examples, processing circuitry 380 may determine patient 4 has intrinsic RSA. If patient 4 has intrinsic RSA, processing circuitry 380 may determine the patient state criterion has been met. If patient 4 is achieving intrinsic RSA, it may, in some examples, be relatively inefficient or in some cases counterproductive to adjust the rate of cardiac pacing according to the RSA program.

[0117] Processing circuitry 380 may additionally or alternatively determine a signal quality of the physiological signal and / or one or more additional signals. If the signal quality passes a threshold, e.g., if the SNR is below a threshold value, processing circuitry 380 may determine the signal quality criterion has been met. The SNR may fall below a threshold due to one or more of patient posture, patient environment, or device-based sensing issues. In some examples, if the signal quality is relatively low, e.g., if the SNR is below the threshold value, processing circuitry 380 may be unable to effectively and / or accurately control therapy delivery circuitry 386 to adjust the rate of cardiac pacing based on the physiological signal.

[0118] If processing circuitry 380 determines that the patient state and / or the signal quality meet the corresponding one or more RSA de-prioritization criterion (“YES” of 704), processing circuitry 380 may determine to de-prioritize the RSA program and may adjust the rate of the cardiac pacing according to another program, such as a patient allometry program, a monotonic program, a cardiac episode program, etc., over a subsequent period of time, e.g.,Attorney Docket No.: A0012462W001 a fourth period of time (708). Processing circuitry may continue to adjust the rate of the cardiac pacing according to the other program until the one or more RSA de-prioritization criterion are no longer met.

[0119] If processing circuitry 380 determines that the patient state and / or the signal quality do not meet the corresponding one or more RSA de-prioritization criterion (“NO” of 704), processing circuitry 380 may determine to adjust the rate of cardiac pacing according to the RSA program (706). In some examples, the RSA program may be patient specific. For example, the RSA program may include adjusting the pacing rate in a stepwise manner, in a ramping matter, or in a sinusoidal manner, as will be discussed with respect to FIGS. 11 A, 1 IB, and 11C, respectively. In some examples, processing circuitry 380 may determine the extent to which to determine to adjust the rate of the cardiac pacing during inspiration and expiration based on the patient state.

[0120] FIG. 8 is a flow diagram illustrating an example operation of a device, e.g., IMD 10, external device 24, and / or computing system 8, to select a physiological signal of a plurality of physiological signals, in accordance with one or more techniques of this disclosure. Sensing circuitry 382 senses a plurality of physiological signals indicative of respiration of patient 4, such as a bioimpedance signal, a cardiac EGM signal, and an accelerometer signal (802). In some examples, sensing circuitry 382 may additionally or alternatively sense a photoplethysmography (PPG) signal, a heart sounds signal, and / or a hemodynamic signal. The plurality of physiological signals may measure intracardiac and / or arterial pressures directly or may serve as surrogates indicative of intracardiac and / or arterial pressures. In some examples, sensing circuitry 382 may sense other physiological signals containing a respiratory signature. Processing circuitry of system 2, e.g., processing circuitry 380, determines which physiological signal of a plurality of physiological signals to use in controlling therapy delivery circuitry 386 to adjust cardiac pacing pulses, e.g., according to an RSA program (804). In some examples, processing circuitry 380 determines the physiological signal by implementing a machine learning model, e.g., machine learning model(s) 374. In some examples, processing circuitry 380 may perform the example operation of FIG. 8 prior to performing the example operation of FIG. 6.

[0121] To determine which physiological signal of the plurality of physiological signals to use in controlling therapy delivery circuitry, processing circuitry 380 may implement machine learning model(s) 374 and may provide the bioimpedance signal, the cardiac EGM signal, and the accelerometer signal as input. Additionally, or alternatively, processing circuitry 380 may determine one or more features of each of the signals, such as a signalAttorney Docket No.: A0012462W001 quality feature and may provide the one or more features as input to machine learning model(s) 374. Processing circuitry 380 may additionally or alternatively provide power consumption information associated with sensing each of the physiological signals as input to machine learning model(s) 374. Using the input, processing circuitry 380, by implementing machine learning model(s) 374, selects the physiological signal.

[0122] To select the physiological signal, processing circuitry 380 may assign weights to the signal quality feature and the power consumption information. In some examples, processing circuitry 380 may select a first physiological signal at a first instance and may select a second physiological signal at a second instance. As an example, processing circuitry 380 may perform the example operation of FIG. 8 a first time and select the cardiac EGM signal. Processing circuitry 380 may perform the example operation of FIG. 8 a second time and select the bioimpedance signal. In some examples, processing circuitry 380 may determine, based on the weighted input information and, optionally, using machine learning model(s) 374, to switch to the bioimpedance signal. In some examples, to determine to switch to the bioimpedance signal, processing circuitry 380 may determine a benefit associated with signal quality outweighs an impact on power consumption associated with selecting the bioimpedance signal.

[0123] Although the example operation of FIG. 8 is described with respect to processing circuitry 380, the techniques of this disclosure are not so limited. In some examples, the machine learning model may be edge-based or cloud-based. As an example, the machine learning model may be machine learning model(s) 516 of external device 24 if the machine learning model is edge-based or may be machine learning model(s) 452 of computing system 8 if the machine learning model is cloud-based. Communication circuitry 388 may transmit the plurality of physiological signals and / or features of the plurality of physiological signals to one or more of external device 24 or computing system 8 to provide input to machine learning model(s) 516 or machine learning model(s) 452.

[0124] FIG. 9 is a conceptual diagram illustrating an example machine learning model configured to predict a physiological signal, in accordance with one or more techniques of this disclosure. Machine learning model 900 is an example of a set of rules, e.g., a set of rules implemented by machine learning model(s) 374 of IMD 10, by machine learning model(s) 452 of computing system 8 in wireless communication with IMD 10, or by machine learning model(s) 516 of external device 24 in wireless communication with IMD 10, as discussed above. Machine learning model 900 may be an example of a deep learning model, or deep learning algorithm, trained to predict a physiological signal over a period of time. One orAttorney Docket No.: A0012462W001 more of IMD 10, external device 24, or a computing system 8 may train, store, and / or utilize machine learning model 900, but other devices may apply inputs associated with a particular patient to machine learning model 900 in other examples. As discussed above, other types of machine learning and deep learning models or algorithms may be utilized in other examples. For examples, a convolutional neural network model of ResNet-18 may be used. Some nonlimiting examples of models that may be used for transfer learning include AlexNet, VGGNet, GoogleNet, ResNet50, DenseNet, transformer models such as bidirectional encoder representations from transformers (BERT) and generative pre-trained transformers (GPT), etc. Some non-limiting examples of machine learning techniques include Support Vector Machines, K-Nearest Neighbor algorithm, and Multi-layer Perceptron.

[0125] As shown in the example of FIG. 9, machine learning model 900 may include three layers. These three layers include input layer 902, hidden layers 904, and output layer 906. Output layer 906 comprises the output from the transfer function 905 of output layer 906. Input layer 902 represents each of the input values XI through X4 provided to machine learning model 900. The number of inputs may be less than or greater than 4, including much greater than 4, e.g., hundreds or thousands. In some examples, the input values may any of the of values input into a machine learning model, as described above. In some examples, input values may include samples of the physiological signal, e.g., a bioimpedance signal, a cardiac EGM signal, and / or an accelerometer signal. In addition, in some examples, input values of machine learning model 900 may include additional data, such as data relating to one or more additional parameters of patient 4. In some examples, processing circuitry 380 normalizes the physiological signal and reshapes the physiological signal into a 3D array or otherwise prepares the physiological signal to be input into an artificial intelligence (Al) routine / library, such as TensorFlow.

[0126] Each of the input values for each node in the input layer 902 is provided to each node of hidden layer 904. In the example of FIG. 9, hidden layers 904 include two layers, one layer having four nodes and the other layer having three nodes, but fewer or greater number of nodes may be used in other examples. Each input from input layer 902 is multiplied by a weight and then summed at each node of hidden layers 904. During training of machine learning model 900, the weights for each input are adjusted to establish the relationship between the input, e.g., the physiological signal over a first period of time, to predict the physiological signal over a second period of time. In some examples, one hidden layer may be incorporated into machine learning model 900, or three or more hidden layers may beAttorney Docket No.: A0012462W001 incorporated into machine learning model 900, where each layer includes the same or different number of nodes.

[0127] The result of each node within hidden layers 904 is applied to the transfer function of output layer 906. The transfer function may be linear or non-linear, and in some examples may depend on the number of layers within machine learning model 900. Example non-linear transfer functions may be a sigmoid function or a tanh function. In examples in which machine learning model 900 is an LSTM, hidden layers 904 may include LSTM cells, including, input, forget, and output gates. The output 907 of the transfer function may be a prediction of the physiological signal over the second period of time.

[0128] In an example, machine learning model 900 defines a sequential model including a linear stack of layers, e.g., an LSTM layer with 100 network units. A return sequence value is set to “True” to cause the output of the LSTM layer to be another sequence of the same length, and another LSTM layer with 100 network units can be added to the sequential model with a return sequence value set to “False” so that only the last output in the resulting output sequence is returned. A dense neural network layer with 25 network units can be added to the sequential model, followed by another dense layer specifying the output of 1 network unit.

[0129] By applying the physiological signal to a machine learning model, such as machine learning model 900, processing circuitry, such as processing circuitry 380 of IMD 10, is able to predict the physiological signal over the second period of time with great accuracy.

[0130] In some examples, machine learning model 900 is additionally or alternatively configured to select a type of physiological signal for use in controlling therapy delivery circuitry 386 to adjust cardiac pacing pulses, e.g., adjusts cardiac pacing pulses according to an RSA program. In some examples, machine learning model 900 is additionally or alternatively configured to determine whether to prioritize the RSA program.

[0131] The architecture of model 900 is but one example, and other examples may be utilized in implementing the techniques of this disclosure. For example, a machine learning model may include different types of layers, such as pooling layers. As another example, a machine learning model may include features not illustrated in the example of FIG. 9, such as skip connections and weight sharing.

[0132] FIG. 10 is a conceptual diagram illustrating an example training process for a machine learning model 1000 in accordance with one or more techniques of this disclosure.

[0133] Machine learning model 1000 may be substantially similar to or the same as machine learning model 900 (FIG. 9). Machine learning model 1000 may be implementedAttorney Docket No.: A0012462W001 using any number of models for supervised and / or reinforcement learning, such as but not limited to, an artificial neural network, a decision tree, a naive Bayes network, a support vector machine, or a k-nearest neighbor model. In other examples, machine learning model 1000 may be implemented using any number of models for unsupervised learning. In some examples, processing circuitry of one or more of IMD 10, external device 24, and / or computing system 8 initially trains the machine learning model 1000 based on training set data 1002, e.g., the first instance of the physiological signal corresponding to the first period of time. A prediction of a second instance of the physiological signal by the machine learning model 1000 may be compared 1004 to the target output 1003. Based on an error signal representing the comparison, the processing circuitry implementing a leaming / training function 1005 may send or apply a modification to weights of machine learning model 1000 or otherwise modify / update the machine learning model 1000. For example, one or more of IMD 10, external device 24, and / or computing system 8 may, for each training instance in the training set 1002, modify machine learning model 1000 in response to data applied to the machine learning model 1000.

[0134] FIGS. 11 A, 1 IB, and 11C are graphical representations of example cardiac pacing waveforms, in accordance with one or more techniques of this disclosure. Each of the graphical representations of FIGS. 11 A, 1 IB, and 11C may correspond to an RSA pacing program. The example of FIG. 11 A includes a stepwise pacing function 1106. Portion 1108 corresponds to an interval of inspiration of a patient, e.g., patient 4, and portion 1110 corresponds to an interval of expiration of patient 4. Processing circuitry 380 may determine a minimum pacing rate 1104 and a maximum pacing rate 1102. In some examples, the difference between minimum pacing rate 1104 and maximum pacing rate 1102 may range from about 1 beat per minute to 20 beats per minute, such as 3 beats per minute. When patient 4 begins the inspiration process, processing circuitry 380 controls therapy delivery circuitry 386 to deliver pacing at maximum pacing rate 1102. When patient begins the expiration process, processing circuitry 380 controls therapy delivery circuitry 386 to deliver pacing at minimum pacing rate 1104.

[0135] The example of FIG. 1 IB includes a ramp pacing function 1114. During inspiration portion 1108, therapy delivery circuitry 386 linearly increases the pacing rate to maximum pacing rate 1102 and begins to decrease the pacing rate halfway through inspiration portion 1108. When inspiration portion 1108 ends and expiration portion 1110 beings, the pacing rate is at baseline rate 1112. Baseline rate 1112 may be based on patient 4’s intrinsic heart rate and / or may be an average heart rate of patient 4, e.g., a movingAttorney Docket No.: A0012462W001 average heart rate of patient 4. Therapy delivery circuitry 386 continues to decrease the pacing rate until reaching minimum pacing rate 1104 halfway through expiration phase 1110, at which point therapy delivery circuitry 386 increases the pacing rate.

[0136] The example of FIG. 11C includes a non-linear, e.g., sinusoidal, pacing function 1116. During inspiration portion 1108, therapy delivery circuitry 386 non-linearly increases the pacing rate to maximum pacing rate 1102 and begins to decrease the pacing rate halfway through inspiration portion 1108. When inspiration portion 1108 ends and expiration portion 1110 beings, the pacing rate is at baseline rate 1112. Therapy delivery circuitry 386 continues to non-linearly decrease the pacing rate until reaching minimum pacing rate 1104 halfway through expiration phase 1110, at which point therapy delivery circuitry 386 increases the pacing rate.

[0137] In some examples, processing circuitry 380 determines which pacing function of pacing functions 1106, 1114, and 1116 to select for patient 4. In some cases, the selection is patient specific. In some examples, processing circuitry 380 determines which pacing function to select based on a power consumption associated with each of the pacing functions.

[0138] Example 1. A medical device system comprising: an implantable medical device (IMD) comprising: sensing circuitry configured to sense a first instance of a physiological signal corresponding to a first period of time, the first instance of the physiological signal being indicative of respiration of a patient over the first period of time; and therapy delivery circuitry configured to deliver cardiac pacing pulses to a heart of the patient via a plurality of electrodes; and processing circuitry configured to: predict, by a machine learning model, a second instance of the physiological signal of the patient corresponding to a second period of time based on the first instance of the physiological signal; control the therapy delivery circuitry to adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal over the second period of time; and control the therapy delivery circuitry to adjust the rate of cardiac pacing based on a third instance of the physiological signal sensed via the sensing circuitry of the IMD and corresponding to a third period of time in response to one of: a duration of time since a most recent update of the machine learning model meeting a time threshold; a change in baseline heart rate of the patient meeting a change threshold; or a change in patient activity meeting a change threshold.Attorney Docket No.: A0012462W001

[0139] Example 2. The medical device system of example 1, wherein the processing circuitry is further configured to: update the machine learning model based on the third instance of the physiological signal.

[0140] Example 3. The medical device system of any of examples 1 or 2, wherein to adjust the rate of the cardiac pacing, the processing circuitry is configured to: control the therapy delivery circuitry to deliver the cardiac pacing pulses to more closely mimic respiratory sinus arrhythmia (RSA).

[0141] Example 4. The medical device system of example 3, wherein controlling the therapy delivery circuitry to deliver the cardiac pacing pulses to more closely mimic RSA comprises executing an RSA program that further configures the processing circuitry to: determine one or more of a patient state of the patient or a signal quality based on one or more of the third instance of the physiological signal or one or more additional signals corresponding to the third period of time; and determine to de-prioritize the RSA program and deliver cardiac pacing pulses according to another program different from the RSA program during a fourth period of time based on the one or more of the patient state or the signal quality of the third instance physiological signal or the one or more additional signals meeting one or more RSA de-prioritization criteria.

[0142] Example 5. The medical device system of any of examples 1 through 4, wherein the baseline heart rate of the patient comprises a median cycle length of a plurality of previous heartbeats of the patient.

[0143] Example 6. The medical device system of any of examples 1 through 5, wherein the processing circuitry is configured to determine the change in patient activity level by comparing median activity counts averaged over corresponding time intervals.

[0144] Example 7. The medical device system of any of examples 1 through 6, wherein the processing circuitry is further configured to: sense a plurality of physiological signals; and determine, by another machine learning model, which physiological signal of the plurality of physiological signals to use as the physiological signal for use in controlling the therapy delivery circuitry.

[0145] Example 8. The medical device system of example 7, wherein the plurality of physiological signals comprises two or more of: a bioimpedance signal; a cardiac electrogram signal; or an accelerometer signal.

[0146] Example 9. The medical device system of any of examples 1 through 8, wherein the processing circuitry is configured to control the therapy delivery circuitry toAttorney Docket No.: A0012462W001 adjust the rate of the cardiac pacing based on the first instance of the physiological signal during the first period of time.

[0147] Example 10. The medical device system of any of examples 1 through 9, wherein the processing circuitry is further configured to control the therapy delivery circuitry to adjust the rate of cardiac pacing based on patient allometry.

[0148] Example 11. The medical device system of any of examples 1 through 10, wherein the second period of time and the third period of time overlap.

[0149] Example 12. The medical device system of any of examples 1 through 11, wherein the first instance of the physiological signal comprises a bioimpedance signal.

[0150] Example 13. The medical device system of any of examples 1 through 12, wherein the machine learning model comprises a long short term memory (LSTM) neural network.

[0151] Example 14. The medical device system of any of examples 1 through 13, wherein the machine learning model is one or more of: implantable medical device-based; edge-based; or cloud-based.

[0152] Example 15. The medical device system of any of examples 1 through 14, wherein the IMD comprises a pacemaker.

[0153] Example 16. The medical device system of any of examples 1 through 15, wherein the processing circuitry is a component of the IMD.

[0154] Example 17. The medical device system of any of examples 1 through 15, wherein the processing circuitry is a component of a non-implantable external device.

[0155] Example 18. A method comprising: sensing, via sensing circuitry of an implantable medical device (IMD) of a medical device system, a first instance of a physiological signal corresponding to a first period of time, the first instance of the physiological signal being indicative of respiration of a patient over the first period of time; predicting, by processing circuitry of the medical device system and using a machine learning model, a second instance of the physiological signal of the patient corresponding to a second period of time based on the first instance of the physiological signal; controlling, by the processing circuitry, therapy delivery circuitry of the IMD to adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal over the second period of time; and controlling, by the processing circuitry, the therapy delivery circuitry to adjust the rate of cardiac pacing based on a third instance of the physiological signal sensed via the sensing circuitry of the IMD and corresponding to a third period of time in response to one of: a duration of time since a most recent update of the machine learning model meeting aAttorney Docket No.: A0012462W001 time threshold; a change in baseline heart rate of the patient meeting a change threshold; or a change in patient activity meeting a change threshold.

[0156] Example 19. The method of example 18, further comprising: updating, by the processing circuitry, the machine learning model based on the third instance of the physiological signal.

[0157] Example 20. The method of any of examples 18 or 19, wherein adjusting the rate of the cardiac pacing comprises: controlling, by the processing circuitry, the therapy delivery circuitry to deliver the cardiac pacing pulses to more closely mimic respiratory sinus arrhythmia (RSA).

[0158] Example 21. The method of example 20, wherein controlling the therapy delivery circuitry to deliver the cardiac pacing pulses to more closely mimic RSA comprises: executing, by the processing circuitry, an RSA program, wherein executing the RSA program configures the processing circuitry to: determine one or more of a patient state of the patient or a signal quality based on one or more of the third instance of the physiological signal or one or more additional signals corresponding to the third period of time; and determine to deprioritize the RSA program and deliver cardiac pacing pulses according to another program different from the RSA program during a fourth period of time based on the one or more of the patient state or the signal quality of the third instance physiological signal or the one or more additional signals meeting one or more RSA de-prioritization criteria.

[0159] Example 22. The method of any of examples 18 through 21, wherein the baseline heart rate of the patient comprises a median cycle length of a plurality of previous heartbeats of the patient.

[0160] Example 23. The method of any of examples 18 through 22, wherein determining the change in patient activity level comprises: comparing, by the processing circuitry, median activity counts averaged over corresponding time intervals.

[0161] Example 24. The method of any of examples 18 through 23, further comprising: sensing, by the sensing circuitry, a plurality of physiological signals; and determining, by the processing circuitry and using another machine learning model, which physiological signal of the plurality of physiological signals to use as the physiological signal for use in controlling the therapy delivery circuitry.

[0162] Example 25. The method of example 24, wherein the plurality of physiological signals comprises two or more of: a bioimpedance signal; a cardiac electrogram signal; or an accelerometer signal.Attorney Docket No.: A0012462W001

[0163] Example 26. The method of any of examples 18 through 25, further comprising: controlling, by the processing circuitry, the therapy delivery circuitry to adjust the rate of the cardiac pacing based on the first instance of the physiological signal during the first period of time.

[0164] Example 27. The method of any of examples 18 through 26, further comprising: controlling, by the processing circuitry, the therapy delivery circuitry to adjust the rate of cardiac pacing based on patient allometry.

[0165] Example 28. The method of any of examples 18 through 27, wherein the second period of time and the third period of time overlap.

[0166] Example 29. The method of any of examples 18 through 28, wherein the first instance of the physiological signal comprises a bioimpedance signal.

[0167] Example 30. The method of any of examples 18 through 29, wherein the machine learning model comprises a long short term memory (LSTM) neural network.

[0168] Example 31. The method of any of examples 18 through 30, wherein the machine learning model is one or more of: implantable medical device-based; edge-based; or cloud-based.

[0169] Example 32. The method of any of examples 18 through 31, wherein theIMD comprises a pacemaker.

[0170] Example 33. The medical device system of any of examples 18 through 32, wherein the processing circuitry is a component of the IMD.

[0171] Example 34. The medical device system of any of examples 18 through 32, wherein the processing circuitry is a component of a non-implantable external device.

[0172] Example 35. A non-transitory computer-readable medium storing instructions that when executed cause processing circuitry to: predict, using a machine learning model, a second instance of a physiological signal of a patient corresponding to a second period of time based on a first instance of the physiological signal, the first instance of a physiological signal being sensed via sensing circuitry of an implantable medical device of the medical device system and being indicative of respiration of the patient over a first period of time; control therapy delivery circuitry of the implantable medical device to adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal over the second period of time; and control the therapy delivery circuitry to adjust the rate of cardiac pacing based on a third instance of the physiological signal sensed via the sensing circuitry of the implantable medical device and corresponding to a third period of time in response to one of: a duration of time since a most recent update of the machine learning model meeting aAttorney Docket No.: A0012462W001 time threshold; a change in baseline heart rate of the patient meeting a change threshold; or a change in patient activity meeting a change threshold.

[0173] Various examples have been described. These and other examples are within the scope of the following claims.

Claims

Attorney Docket No.: A0012462W001WHAT IS CLAIMED IS:

1. A medical device system comprising: an implantable medical device (IMD) comprising: sensing circuitry configured to sense a first instance of a physiological signal corresponding to a first period of time, the first instance of the physiological signal being indicative of respiration of a patient over the first period of time; and therapy delivery circuitry configured to deliver cardiac pacing pulses to a heart of the patient via a plurality of electrodes; and processing circuitry configured to: predict, by a machine learning model, a second instance of the physiological signal of the patient corresponding to a second period of time based on the first instance of the physiological signal; control the therapy delivery circuitry to adjust a rate of cardiac pacing based on the predicted second instance of the physiological signal over the second period of time; and control the therapy delivery circuitry to adjust the rate of cardiac pacing based on a third instance of the physiological signal sensed via the sensing circuitry of the IMD and corresponding to a third period of time in response to one of: a duration of time since a most recent update of the machine learning model meeting a time threshold; a change in baseline heart rate of the patient meeting a change threshold; or a change in patient activity meeting a change threshold.

2. The medical device system of claim 1, wherein the processing circuitry is further configured to: update the machine learning model based on the third instance of the physiological signal.

3. The medical device system of any of claims 1 or 2, wherein to adjust the rate of the cardiac pacing, the processing circuitry is configured to: control the therapy delivery circuitry to deliver the cardiac pacing pulses to more closely mimic respiratory sinus arrhythmia (RSA).Attorney Docket No.: A0012462W0014. The medical device system of claim 3, wherein controlling the therapy delivery circuitry to deliver the cardiac pacing pulses to more closely mimic RSA comprises executing an RSA program that further configures the processing circuitry to: determine one or more of a patient state of the patient or a signal quality based on one or more of the third instance of the physiological signal or one or more additional signals corresponding to the third period of time; and determine to de-prioritize the RSA program and deliver cardiac pacing pulses according to another program different from the RSA program during a fourth period of time based on the one or more of the patient state or the signal quality of the third instance physiological signal or the one or more additional signals meeting one or more RSA deprioritization criteria.

5. The medical device system of any of claims 1 through 4, wherein the baseline heart rate of the patient comprises a median cycle length of a plurality of previous heartbeats of the patient.

6. The medical device system of any of claims 1 through 5, wherein the processing circuitry is configured to determine the change in patient activity level by comparing median activity counts averaged over corresponding time intervals.

7. The medical device system of any of claims 1 through 6, wherein the processing circuitry is further configured to: sense a plurality of physiological signals; and determine, by another machine learning model, which physiological signal of the plurality of physiological signals to use as the physiological signal for use in controlling the therapy delivery circuitry.

8. The medical device system of claim 7, wherein the plurality of physiological signals comprises two or more of: a bioimpedance signal; a cardiac electrogram signal; or an accelerometer signal.Attorney Docket No.: A0012462W0019. The medical device system of any of claims 1 through 8, wherein the processing circuitry is configured to control the therapy delivery circuitry to adjust the rate of the cardiac pacing based on the first instance of the physiological signal during the first period of time.

10. The medical device system of any of claims 1 through 9, wherein the processing circuitry is further configured to control the therapy delivery circuitry to adjust the rate of cardiac pacing based on patient allometry.

11. The medical device system of any of claims 1 through 10, wherein the second period of time and the third period of time overlap.

12. The medical device system of any of claims 1 through 11, wherein the first instance of the physiological signal comprises a bioimpedance signal.

13. The medical device system of any of claims 1 through 12, wherein the machine learning model comprises a long short term memory (LSTM) neural network.

14. The medical device system of any of claims 1 through 13, wherein the machine learning model is one or more of:IMD-based; edge-based; or cloud-based.

15. The medical device system of any of claims 1 through 14, wherein the IMD comprises a pacemaker, and wherein the processing circuitry is a component of the pacemaker.

Images