Intra-aortic pressure prediction

Abstract

To provide systems and methods for predicting an intra-aortic pressure of a patient receiving hemodynamic support from a transvalvular micro-axial heart pump.SOLUTION: In some implementations, an intra-aortic pressure time series is derived from measurements with a pressure sensor of the transvalvular micro-axial heart pump, and a motor speed time series is derived from measurements of the back electromotive force of a motor of the transvalvular micro-axial heart pump. Furthermore, in some implementations, machine learning algorithms such as deep learning are applied to the intra-aortic pressure and motor speed time series to accurately predict an intra-aortic pressure of a patient. In some implementations, the prediction is short-term (e.g., approximately 5 minutes in advance).SELECTED DRAWING: Figure 1b

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. Provisional Application No. 62 / 855,389, filed May 31, 2019, and incorporated herein by reference.

[0002] Technical Field The present technology relates to systems and methods for predicting intra-aortic pressure in a patient receiving hemodynamic support from a transvalve micro-axial heart pump. [Background technology]

[0003] background Machine learning has been successfully applied to provide predictions in a variety of different technology areas, including computer vision, natural language processing, speech recognition, and clinical healthcare. Examples of machine learning algorithms include Bayesian algorithms, clustering algorithms, decision tree algorithms, dimensionality reduction algorithms, instance-based algorithms, deep learning algorithms, regression algorithms, regularization algorithms, and rule-based machine learning algorithms. In clinical healthcare, machine learning algorithms have been used to model mortality risk, predict hospital length of stay, detect physiological decline, and classify phenotypes. See, for example, Harutyunyan et al., Multitask learning and benchmarking with clinical time series data, Scientific Data, doi: 10.1038 / s41597-019-0103-9, 2017; Purushothama et al., Benchmarking deep learning models on large healthcare datasets, Journal of Biomedical Informatics 83, 112-134, 2018. However, a need remains for systems and methods for predicting physiological responses that can aid physicians in early real-time detection of disease and patient response to treatment. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Harutyunyan et al., Multitask learning and benchmarking with clinical time series data, Scientific Data, doi: 10.1038 / s41597-019-0103-9, 2017 [Non-Patent Document 2] Purushothama et al., Benchmarking deep learning models on large healthcare datasets, Journal of Biomedical Informatics 83, 112-134, 2018 Summary of the Invention

[0005] overview To date, machine learning algorithms have not been used to predict intra-aortic pressure (e.g., latest intra-aortic pressure, mean intra-aortic pressure, median intra-aortic pressure, maximum intra-aortic pressure, minimum intra-aortic pressure, range of intra-aortic pressure, systolic intra-aortic pressure, diastolic intra-aortic pressure, etc.) in patients receiving hemodynamic support. Forecasting a patient's intra-aortic pressure is difficult, in part because high-frequency time series of intra-aortic pressure are not currently publicly available. Moreover, intra-aortic pressure time series can be noisy and highly non-stationary. Furthermore, forecast error and uncertainty increase dramatically for long-term predictions.

[0006] Being able to predict a patient's intra-aortic pressure would greatly enhance clinicians' ability to anticipate the patient's condition. For example, acute decompensated heart failure (ADHF) is a complex clinical event associated with excessive morbidity and mortality, and is typically indicated by a rapid drop in blood pressure and associated with an increased heart rate. A challenging aspect of ADHF is the lack of effective treatments that both reduce symptoms and improve clinical outcomes. Existing guideline recommendations are largely based on expert opinion. See, for example, Givertz et al., Acute Decompensated Heart Failure: Update on New and Emerging Evidence and Directions for Future Research, Journal of Cardiac Failure, Vol. 19, No. 6, 2013. Thus, being able to predict a patient's intra-aortic pressure trajectory would make it easier for health care providers to assess a patient's risk of ADHF and intervene before collapse. In addition, predicting intra-aortic pressure would provide guidance to help wean patients off support as their health improves.

[0007] In accordance with aspects of the present disclosure, systems and methods are described for predicting intra-aortic pressure in a patient receiving hemodynamic support from a transvalvular micro-axial heart pump. In some embodiments, a time series of intra-aortic pressure is derived from measurements of a pressure sensor in the transvalvular micro-axial heart pump, and a time series of motor speed is derived from measurements of a back electromotive force (EMF) in the motor of the transvalvular micro-axial heart pump. Furthermore, in some embodiments, machine learning algorithms, such as deep learning, are applied to the time series of intra-aortic pressure and motor speed to accurately predict the intra-aortic pressure in the patient. In some embodiments, the prediction is short-term (e.g., about 5 minutes ahead).

[0008] One aspect of the present disclosure relates to a system including a transvalvular micro axial heart pump including a motor and a pressure sensor, and one or more processors configured to: obtaining a set of intra-aortic pressure measurements corresponding to pressure values ​​measured by the pressure sensor during a period during which the transvalve micro axial pump is at least partially located within the patient's heart; obtaining a set of motor speed measurements corresponding to the rotational speed of the motor during said period; predicting an intra-aortic pressure for a patient based on a set of intra-aortic pressure measurements and a set of motor speed measurements using the trained machine learning model; and Automatically adjust motor speed settings based on the patient's predicted intra-aortic pressure.

[0009] In some embodiments, the one or more processors are further configured to obtain a set of current measurements corresponding to energy intake of the motor during the period, and the prediction is further based on the set of current measurements.

[0010] In some embodiments, the transvalve micro axial heart pump further includes a tube; an inlet area having one or more openings through which blood may be drawn into the tube by a motor; and an outlet area having one or more openings through which blood may be expelled from the tube by a motor, and a pressure sensor is coupled to the outlet area. In some embodiments, the transvalve micro axial heart pump further includes an additional pressure sensor coupled to the inlet area; the one or more processors are further configured to obtain a set of left ventricular pressure measurements corresponding to pressure values ​​measured by the additional pressure sensor during the time period; and the prediction is further based on the set of left ventricular pressure measurements.

[0011] In some embodiments, the machine learning model is a deep learning model. In some embodiments, the deep learning model is an Autoregressive Integrated Moving Average (ARIMA) model, a Deep Neural Network (DNN) model, a Recurrent Sequence to Sequence model, a Recurrent Sequence to Sequence model with Attention, a Transformer model, a Temporal Convolutional Neural Network (TCN) model, or a Convolutional Neural Pyramid model. In some embodiments, the deep learning model is a Recurrent Sequence to Sequence model with Legendre Memory Unit (LMU).

[0012] In some embodiments, the machine learning model is trained on a dataset having an increasing sequence, a decreasing sequence, and a stationary sequence, each sequence including an intra-aortic pressure measurement and a motor speed measurement. In some embodiments, a sequence is increasing if an intra-aortic pressure measurement in the sequence is increasing above a predefined threshold, a sequence is decreasing if an intra-aortic pressure measurement in the sequence is decreasing above a predefined threshold, and a sequence is stationary if an intra-aortic pressure measurement in the sequence is neither increasing nor decreasing above a predefined threshold. In some embodiments, the predefined threshold is 10 mmHg. In some embodiments, each sequence includes a predefined number of aortic pressure measurements and motor speed measurements. In some embodiments, each sequence includes real-time (RT) intra-aortic pressure measurements and motor speed measurements. In some embodiments, each sequence includes average time (AT) intra-aortic pressure measurements and motor speed measurements.

[0013] In some embodiments, the machine learning model is trained on a dataset having only increasing and decreasing sequences, each sequence including an intra-aortic pressure measurement and a motor speed measurement. In some embodiments, a sequence is increasing if the intra-aortic pressure measurement in the sequence is increasing above a predefined threshold, and is decreasing if the intra-aortic pressure measurement in the sequence is decreasing above a predefined threshold. In some embodiments, the predefined threshold is 10 mmHg. In some embodiments, each sequence includes a predefined number of aortic pressure measurements and motor speed measurements. In some embodiments, each sequence includes real-time (RT) intra-aortic pressure measurements and motor speed measurements. In some embodiments, each sequence includes average time (AT) intra-aortic pressure measurements and motor speed measurements.

[0014] In some embodiments, automatically adjusting a speed setting of the motor based on the patient's predicted intra-aortic pressure includes temporarily increasing the speed setting of the motor when the patient's predicted intra-aortic pressure is less than the patient's current intra-aortic pressure by more than a predetermined amount.

[0015] Another aspect of the present disclosure relates to a system including a transvalvular micro axial heart pump, one or more processors, and a display. The transvalvular micro axial heart pump includes a motor and a pressure sensor. The one or more processors are configured to: obtaining a set of intra-aortic pressure measurements corresponding to pressure values ​​measured by the pressure sensor during a period during which the transvalve micro axial pump is at least partially located within the patient's heart; obtaining a set of motor speed measurements corresponding to the rotational speed of the motor during said period; and Predicting a patient's intra-aortic pressure based on a set of intra-aortic pressure measurements and a set of motor speed measurements using the trained machine learning model. The display is configured to display the patient's predicted intra-aortic pressure.

[0016] In some embodiments, the display is configured to display the patient's predicted intra-aortic pressure simultaneously with the patient's current intra-aortic pressure and the current speed setting of the motor. In some embodiments, the display is further configured to display an alert when the patient's predicted intra-aortic pressure is less than the patient's current intra-aortic pressure by more than a predetermined amount. In some embodiments, the display is configured to display the patient's predicted intra-aortic pressure as part of a graph.

[0017] Yet another aspect of the present disclosure relates to a method for treating a patient with a transvalve micro axial heart pump received within the patient, the method comprising the steps of: Inserting the transvalve micro axial heart pump into the patient's body; obtaining a set of intra-aortic pressure measurements corresponding to pressure values ​​measured by a pressure sensor located on the transvalve micro axial heart pump during a period when the transvalve micro axial heart pump is at least partially located within the patient's heart; obtaining a set of motor speed measurements corresponding to the rotational speed of the motor during said period; predicting an intra-aortic pressure for the patient based on the set of intra-aortic pressure measurements and the set of motor speed measurements using the trained machine learning model; and Automatically adjusting the speed setting of the motor based on the patient's predicted intra-aortic pressure.

[0018] In some embodiments, the method further includes obtaining a set of current measurements corresponding to energy uptake of the motor during said time period, and said prediction is further based on said set of current measurements.

[0019] In some embodiments, the transvalve micro axial heart pump further comprises a tube; an inlet area having one or more openings through which blood may be drawn into the tube by the motor; and an outlet area having one or more openings through which blood may be expelled from the tube by the motor, and a pressure sensor is coupled to the outlet area. In some embodiments, the transvalve micro axial heart pump further comprises an additional pressure sensor coupled to the inlet area; the method further comprises obtaining a set of left ventricular pressure measurements corresponding to pressure values ​​measured by the additional pressure sensor during said time period; and said prediction is further based on the set of left ventricular pressure measurements.

[0020] In some embodiments, the method further comprises adjusting the amount of drug provided to the patient based on the predicted intra-aortic pressure. In some embodiments, the method further comprises decreasing the motor speed if the intra-aortic pressure is predicted to increase. In some embodiments, the method further comprises increasing the motor speed if the intra-aortic pressure is predicted to decrease. [Brief description of the drawings]

[0021] [Figure 1a] 1 illustrates a transvalve micro axial heart pump. [Figure 1b] 1(a) illustrates the transvalve micro axial heart pump of FIG. 1(a) positioned within a patient's heart. [Figure 1c] 1 illustrates a ventricular support system. [Figure 2a] 1 illustrates information that may be displayed on a home screen. [Figure 2b] 1 illustrates information that may be displayed on an indwelling screen. [Figure 2c] 1 illustrates information that may be displayed on a purge screen. [Figure 2d] 14 illustrates information that may be displayed on an infusion history screen. [Figure 2e] 1 illustrates information that may be displayed on a home screen. [Figure 2f] 1 illustrates information that may be displayed on an indwelling screen. [Figure 2g] 1 illustrates information that may be displayed on an indwelling screen. [Figure 2h] 1 illustrates information that may be displayed on an indwelling screen. [Diagram 3] 1 illustrates a system for monitoring and / or controlling multiple medical devices, such as a transvalve micro axial heart pump. [Figure 4] The isovolumic relaxation, ejection, isovolumic relaxation, and filling phases of the cardiac cycle are illustrated. [Diagram 5] The regular waveforms of intra-aortic pressure (AoP), left ventricular pressure (LVP), differential pressure (dP), pump flow, and motor current and their relationship to systole and diastole are illustrated. [Figure 6] 1 illustrates the overall structure of a recursive sequence-to-sequence model. [Figure 7] Illustrates the overall structure of a Transformer model. [Figure 8] Illustrates the overall structure of a temporal convolutional neural network. [Figure 9] Illustrates the overall structure of a temporal convolutional neural pyramid. [Figure 10] Time series of intra-aortic pressure, motor speed, and motor current at 25 Hz (RT) for 10 seconds are shown. [Figure 11] A time series of mean intra-aortic pressure at 0.1 Hz (AT) for 20 min is shown. [Figure 12] The root mean square error of the selected models is plotted. [Figure 13] We plot the MAP predictions of the two deep learning models against the ground truth for a single recording over the course of a 24-hour period. [Figure 14] Illustrating MAP forecasts of two deep learning models with increasing, decreasing and stationary sequences [Figure 15] The root mean square error of the selected models is plotted. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Detailed Description The embodiments of the present disclosure will be described in detail with reference to the drawings; in the drawings, similar reference numerals refer to similar or identical elements. It should be understood that the embodiments of the present disclosure are merely examples of the present disclosure, which may be embodied in various forms. Well-known functions or constructions are not described in detail to avoid obscuring the present disclosure with unnecessary details. Therefore, the specific structural and functional details disclosed herein should not be interpreted as limiting, but only as a basis for claims and as a representative basis for teaching those skilled in the art to variously utilize the present disclosure in virtually any structure with reasonable details.

[0023] Efforts have been made to predict the peripheral blood pressure of patients by various machine learning models and statistical methods, see, for example, Abbasi et al., Long-term Prediction of Blood Pressure Time Series Using Multiple Fuzzy Functions, 21st Iranian Conference on Biomedical Engineering, ICBME, 2014; Peng et al., Long-term Blood Pressure Prediction with Deep Recurrent Neural Networks, arXiv:1705.04524v3, 2018.

[0024] Efforts have been made to predict whether a patient is likely to experience an acute hypotensive episode (AHE) using various machine learning models and statistical methods. For example, Henriques & Rocha, Prediction of Acute Hypotensive Episodes Using Neural Network Multi-models, Computers in Cardiology 36:549552, 2009; Moody & Lehman, Predicting Acute Hypotensive Episodes: The 10th Annual PhysioNet / Computers in Cardiology Challenge, Comput. Cardiol., 36(5445351): 541-544, 2009; Johnson et al., MIMIC-III, a freely accessible critical care database, Scientific Data, DOI: 10.1038 / sdata.2016.35, 2016; Hatib et al., Machine-learning Algorithm to Predict Hypotension Based on High-Fidelity Arterial Pressure Waveform Analysis, Anesthesiology, 129(4):663-674, Please refer to 2018.

[0025] Efforts have been made to predict acute decompensated heart failure (ADHF) by various machine learning models and statistical methods. For example, Kenney et al., Early Detection of Heart Failure Using Electronic Health Records, Circ. Cardiovasc. Qual. Outcomes, 9:649-658, 2016; Deo & Nallamothu, Learning About Machine Learning: The Promise and Pitfalls of Big Data and the Electronic Health Record, Circ. Cardiovasc. Qual. Outcomes, 9:618-620, 2016; Passantino et al. See al., Predicting mortality in patients with acute heart failure: Role of risk scores, World J. Cardiol., 7(12): 902911, 2015; Thorvaldsen et al., Predicting Risk in Patients Hospitalized for Acute Decompensated Heart Failure and Preserved Ejection Fraction, Circ. Heart Fail., 10:e003992, 2017.

[0026] However, none of the above cited studies describes a system or method for predicting intra-aortic pressure in a patient receiving hemodynamic support. Some of the cited studies describe a system or method for predicting a patient's peripheral blood pressure. However, whereas peripheral blood pressure provides an indirect index of a patient's cardiac function, intra-aortic pressure provides a direct index of a patient's cardiac function. Peripheral blood pressure may be obtained, for example, using a blood pressure cuff wrapped around the patient's limb (e.g., an arm cuff or wrist cuff), whereas intra-aortic pressure may be obtained, for example, using a transvalve micro-axial heart pump. As a result, peripheral blood pressure provides less information about a patient's condition than intra-aortic pressure.

[0027] In addition, some of these approaches described in the above cited studies require a very large number of input variables, which is impractical, at least from a clinical point of view. Moreover, some of the variables used in the above cited studies are not easily measurable. Moreover, some of the models proposed in the above cited studies are only suitable for the assessment of long-term mortality. They cannot assist physicians in the real-time early detection of diseases such as ADHF.

[0028] Patients with severe multivessel coronary artery disease (CAD), unprotected left main coronary artery stenosis, one last remaining vessel, and / or severely reduced left ventricular (LV) ejection fraction (EF) are frequently declined cardiac surgery and are increasingly referred for high-risk percutaneous coronary intervention (HR-PCI). To prevent hemodynamic instability and improve clinical outcomes, transvalvular microaxial heart pumps, such as the Impella 5.0® by Abiomed, Inc., Danvers, MA, as shown in Figure 1(a), are increasingly being used during HR-PCI.See, for example, Russo et al., Hemodynamics and its predictors during transvalvular-micro-axial-heart-pump-protected PCI in high risk patients with reduced ejection fraction, Int. J. Cardiol. 274:221-225, 2019; Dixon et al., A prospective feasibility trial investigating the use of the transvalvular micro-axial heart pump system in patients undergoing high-risk percutaneous coronary intervention (The Transvalvular Micro-axial Heart Pump Trial): initial U.S. experience, JACC Cardiovasc. Interv. 2 (2) 91-96, 2009; O'Neill et al., A prospective, randomized clinical trial of hemodynamic support with transvalvular micro-axial heart pump versus intra-aortic balloon pump in patients undergoing high-risk percutaneous coronary intervention: the transvalvular micro-axial heart pump study, Circulation 126 (14) 1717-1727, 2012.

[0029] A transvalvular micro-axial heart pump is a percutaneous catheter-based device that provides hemodynamic support to a patient's heart. As shown in FIG. 1(a), the transvalvular micro-axial heart pump 110 may include a pigtail 111, an inlet area 112, a cannula 113, a pressure sensor 114, an outlet area 115, a motor housing 116, and / or a catheter tube 117. The pigtail 111 may help stabilize the transvalvular micro-axial heart pump 110 within the patient's heart. In operation, blood may be drawn into one or more openings in the inlet area 112, flowed through the cannula 113, and expelled through one or more openings in the outlet area 115 by a motor (not shown) disposed within the motor housing 116. In some embodiments, the pressure sensor 114 may include a flexible membrane integrated within the cannula 113. One side of the pressure sensor 114 may be exposed to blood pressure outside the cannula 113, and the other side may be exposed to blood pressure inside the cannula 113. In some such embodiments, the pressure sensor 114 may generate an electrical signal proportional to the difference between the pressure outside the cannula 113 and the pressure inside the cannula 113. In some embodiments, the pressure difference measured by the pressure sensor 114 may be used to position the transvalve micro axial heart pump 110 within the patient's heart. In some embodiments, the pressure sensor 114 is an optical pressure sensor. A catheter tube 117 may provide one or more fluid and / or electrical connections between the transvalve micro axial heart pump 110 and more or more other devices of the ventricular support system.

[0030] As shown in FIG. 1(b), the transvalve micro axial heart pump 110 may be positioned within the patient's heart 120. As shown, the transvalve micro axial heart pump 110 may be inserted percutaneously, for example, via the femoral artery 122 into the ascending aorta 124, spanning the aortic valve 126, and into the left ventricle 128. In other embodiments, the transvalve micro axial heart pump may be inserted percutaneously, for example, via the axillary artery 123 into the ascending aorta 124, spanning the aortic valve 126, and into the left ventricle 128. In other embodiments, the transvalve micro axial heart pump may be inserted directly, for example, into the ascending aorta 124, spanning the aortic valve 126, and into the left ventricle 128. In operation, the transvalve micro axial heart pump 110 entrains blood from the left ventricle 128 and expels blood into the ascending aorta 124. As a result, the transvalve micro axial heart pump 110 performs a portion of the work normally done by the patient's heart 120. The hemodynamic effects of the transvalve micro axial heart pump include increased cardiac output and improved coronary blood flow, which results in reduced LV end-diastolic pressure, pulmonary capillary wedge pressure, myocardial work, and oxygen consumption. See, e.g., Burkhoff & Naidu, The science behind percutaneous hemodynamic support: a review and comparison of support strategies, Catheter Cardiovasc. Interv. 80:816-29, 2012.

[0031] As shown in FIG. 1(c), the transvalvular micro-axial heart pump 110 may be incorporated into a ventricular support system 100. The ventricular support system 100 also includes a controller 130 (e.g., Automated Impella Controller® by Abiomed, Inc., Danvers, MA), a display 140, a purging subsystem 150, a connector cable 160, a plug 170, and a repositioning unit 180. As shown, the controller 130 includes the display 140. The controller 130 monitors and controls the transvalvular micro-axial heart pump 110. During operation, the purging subsystem 150 delivers a purging fluid through the catheter tube 117 to the transvalvular micro-axial heart pump 110 to prevent blood from entering the motor (not shown) in the motor housing 116. In some embodiments, the purging fluid is a dextrose solution (e.g., a 5% dextrose in water solution containing 25 or 50 IU / mL heparin). A connector cable 160 provides an electrical connection between the transvalve micro axial heart pump 110 and the controller 130. A plug 170 connects the catheter tube 117, the purging subsystem 150, and the connector cable 160. In some embodiments, the plug 170 includes a memory for storing operating parameters in case the patient needs to be transferred to another controller. A repositioning unit 180 may be used to reposition the transvalve micro axial heart pump 110.

[0032] As shown, the purging subsystem 150 includes a container 151, a supply line 152, a purge cassette 153, a purge disk 154, a purge tubing 155, a check valve 156, a pressure reservoir 157, an infusion filter 158, and a side arm 159. The container 151 may be, for example, a bag or a bottle. The purging fluid is stored in the container 151. The supply line 152 provides a fluid connection between the container 151 and the purge cassette 153. The purge cassette 153 may control how the purging fluid in the container 151 is delivered to the transvalve micro axial heart pump 110. For example, the purge cassette 153 may include one or more valves for controlling the pressure and / or flow rate of the purging fluid. The purge disk 154 includes one or more pressure sensors and / or flow sensors for measuring the pressure and / or flow rate of the purging fluid. As shown, the controller 130 includes a purge cassette 153 and a purge disk 154. Purge tubing 155 provides a fluid connection between the purge disk 154 and a check valve 156. A pressure reservoir 157 provides additional filling volume during changes of the purge fluid. In some embodiments, the pressure reservoir 157 includes a flexible rubber septum that provides additional filling volume by means of an expansion chamber. An infusion filter 158 helps prevent bacterial contamination and air from entering the catheter tube 117. A side arm 159 provides a fluid connection between the infusion filter 158 and a plug 170.

[0033] In operation, the controller 130 receives measurements from the pressure sensor 114 and the purge disk 154 and controls the motor (not shown) in the motor housing 116 and the purge cassette 153. As described above, the controller 130 controls and measures the pressure and / or flow rate of the purge fluid via the purge cassette 153 and the purge disk 154. In operation, the purge fluid is passed through a purge lumen (not shown) in the catheter tube 117 and in the plug 170 after exiting the purge subsystem 150 through the side arm 159. A sensor cable (not shown) in the catheter tube 117, in the connector cable 160, and in the plug 170 provides an electrical connection between the pressure sensor 114 and the controller 130. A motor cable (not shown) in the catheter tube 117, in the connector cable 160, and in the plug 170 provides an electrical connection between the motor in the motor housing 116 and the controller 130. In operation, the controller 130 receives measurements from the pressure sensor 114 through the sensor cable and controls the power delivered through the motor cable to the motor in the motor housing 116. By controlling the power delivered to the motor in the motor housing 116, the controller 130 can control the speed of the motor in the motor housing 116.

[0034] Various modifications may be made to the ventricular support system 100 and one or more of its components. For example, as detailed in Abiomed, Impella® Ventricular Support Systems for Use During Cardiogenic Shock and High-Risk PCI: Instructions for Use and Clinical Reference Manual, Document No. 0042-9028 rG (Apr. 2020), which is incorporated herein by reference, the ventricular support system 100 may be modified to accommodate other types of transvalvular micro-axial heart pumps, such as Impella 2.5® catheters, Impella LD® catheters, and Impella CP® catheters. As another example, one or more sensors may be added to the transvalvular micro-axial heart pump 100. For example, a signal generator may be added to the transvalvular micro-axial heart pump 100 to generate a signal indicative of the rotational speed of the motor in the motor housing 116, as described in U.S. Patent Application No. 16 / 353,132, filed March 14, 2019, entitled "Blood Flow Rate Measurement System," which is incorporated herein. As another example, a second pressure sensor configured to measure the blood pressure of the left ventricle may be added to the transvalvular micro-axial heart pump 100 near the inlet area 112. In such an embodiment, an additional sensor cable may be disposed in the catheter tube 117, in the connector cable 160, and in the plug 170 to provide an electrical connection between the one or more additional sensors and the controller 130. As yet another example, one or more components of the ventricular support system 100 may be separated. For example, the display 140 may be incorporated into another device in communication with the controller 130 (e.g., wirelessly or through one or more electrical cables).

[0035] 2(a)-(h) illustrate different screens that may be displayed by display 140. FIG. For example, FIG. 2(a) illustrates a home screen 202 that includes heart pump type 211 (e.g., “Impella 5.0”), heart pump serial number 212 (e.g., “171000”), date and time 214 (e.g., “2019-08-21 15:56”), software version number 216 (e.g., “IC4048 V8.1”), power icon 218 (e.g., battery indicator), button labels 221, 222, 224, 226, and 228 (e.g., “Alarm Silence”, “Flow Control”, “Display”, “Purge Menu”, and “Menu”), heart pump current speed (performance) setting 230 (e.g., “P-4”), heart pump flow measurements 242, purge system measurements 244, status indicators 251 (e.g., “Impella Position OK”), line graphs 261, and notification area 270. The heart pump current speed (performance) setting 230 corresponds to the speed at which the motor in the motor housing 116 is operating. For example, "P-4" may indicate that the motor in the motor housing 116 is operating at approximately 22,000 rpm. The heart pump flow measurements 242 include the average flow rate (e.g., "1.6 L / min"), minimum flow rate (e.g., "1.1 L / min"), and maximum flow rate (e.g., "2.1 L / min") of blood through the transvalve micro axial heart pump 100. The heart pump flow measurements 242 may be derived from measurements taken by the pressure sensor 114 and / or the energy intake of the motor in the motor housing 116. The purge system measurements 244 include the current flow rate (e.g., "10.2 ml / hr") and current pressure (e.g., "99 mmHg") of the purge fluid through the purge subsystem 150. Purge system measurements 244 may be derived from measurements taken by purge disk 154. Diagram 161 illustrates how transvalve micro axial heart pump 110 should be positioned within a patient's heart. In FIG. 2(a), notification area 270 includes notifications 271, 272, and 273.

[0036] Each of notifications 271, 272, and 273 includes a header and a set of instructions. For example, notification 271 includes a header of "Purge System Open" and the instructions "1. Check purge system tubing for open connections or leaks" and "2. Press the Purge Menu softkey, then select Change Cassette & Bag." Notification 272 includes a header of "Suction" and the instructions "1. Reduce P Level," "2. Check Fill and Volume Status," and "3. Check Impella Position." Notification 273 includes a header of "Flight Mode Enabled" and the instructions "1. Ground Controller During Air Transport," "2. Enable Flight Mode on Module if equipped with Impella Connect," and "3. Disable Flight Mode in Menu upon arrival at Receiving Hospital." In other embodiments, notifications displayed within notification area 270 may be structured differently. For example, the header and instructions may be contained within a single box rather than within two different boxes. As another example, the notification may not include a header. As yet another example, the instructions may be replaced with a different type of information, such as an explanatory text. For example, a notification may serve as an alert and include a description explaining the cause of the alert.

[0037] 2(b)-(d) illustrate placement screen 204, purge screen 206, and infusion history screen 208, respectively. A user may switch between these screens using buttons positioned next to button labels 221, 222, 224, 226, and 228. In other embodiments, different user input devices may be used. For example, in some embodiments, display 140 may be a touch screen, and a user may switch between screens by tapping button labels 221, 222, 224, 226, and 228. As another example, in some embodiments, a user may switch between screens using a separate input device, such as a mouse or keyboard.

[0038] All data fields present on the home screen 202 are also included on the indwell screen 204, the purge screen 206, and the infusion history screen 208, except for the status indicators 251, the line diagram 261, and the notifications 271, 272, and 273. In other embodiments, additional data fields may be added or removed from these screens. For example, in some embodiments, the heart pump type 211 and the heart pump serial number 212 may only appear on the main screen 202.

[0039] The indwell screen 204, purge screen 206, and injection history screen 208 also include additional information. For example, as shown in FIG. 2(b), the indwell screen 204 includes a indwell signal graph 252, a indwell signal measurement 262, a motor current graph 253, and a motor current measurement 263. The indwell signal graph 252 illustrates pressure values ​​derived from measurements taken by the pressure sensor 114 over a period of time (e.g., "10 sec"). The indwell signal measurement 262 includes an average pressure value (e.g., "9 mmHg"), a minimum pressure value (e.g., "-17 mmHg"), and a maximum pressure value (e.g., "76 mmHg") derived from measurements taken by the pressure sensor 114 over the period of time. The motor current graph 253 illustrates current values ​​provided to the motor in the motor housing 116 over a period of time (e.g., "10 sec"). The motor current measurements 263 include the average current (e.g., 535 mA), the minimum current (e.g., 525 mA), and the maximum current (e.g., 556 mA) provided to the motor within the motor housing 116 over the period of time. The placement signal graph 252, the placement signal measurements 262, the motor current graph 253, and the motor current measurements 263 are collectively useful for determining the position of the transvalve micro axial heart pump 110 within the patient's heart.

[0040] As shown in FIG. 2(c), the purge screen 206 additionally includes a purge flow graph 254, a purge flow measurement 264, a purge pressure graph 255, and a purge pressure measurement 265. The purge flow graph 254 illustrates the flow rate of purge fluid through the purging subsystem 150 over a period of time (e.g., “1 hr”). The purge flow measurement 264 includes the current flow rate of purge fluid through the purging subsystem 150 (e.g., “17.9 ml / hr”). The purge pressure graph 255 illustrates the pressure of the purge fluid in the purging subsystem 150 over a period of time (e.g., “1 hr”). The purge pressure measurement 265 includes the current pressure of the purge fluid in the purging subsystem 150 (e.g., “559 mmHg”). Collectively, the purge flow graph 254, the purge flow measurements 264, the purge pressure graph 255, and the purge pressure measurements 265 can aid in patient management.

[0041] As shown in FIG. 2(d), the infusion history screen 208 additionally includes an infusion history table 256, a dextrose infusion measurement 266, and a heparin infusion measurement 267. The infusion history table 256 provides a summary of the amount of purge fluid, heparin, and dextrose delivered to the patient over each of a number of time periods (e.g., "10:00-11:00", "11:00-12:00", "12:00-13:00", "13:00-14:00", "14:00-15:00", and "15:00-15:08"). The dextrose infusion measurement 266 includes the most recent rate at which dextrose is being delivered to the patient (e.g., "935 mg / hr"). The heparin infusion measurement 267 includes the most recent rate at which heparin is being delivered to the patient (e.g., "935 IU / hr"). The infusion history table 256, dextrose infusion measurements 266, and heparin infusion measurements 267, collectively, can also aid in patient management.

[0042] 2(e)-(h) illustrate how different types of alerts may be presented to the user via the display 140. For example, when a patient has low native ventricular function and the controller 130 is unable to determine the position of the transvalvular micro-axial heart pump 110 within the patient's heart, the home screen 202 may be updated to the manner shown in FIG. 2(e). More specifically, the status indicator 251 may be updated to indicate "Impella position unknown" and a notification 274 may be added to the notification area 270. As another example, when the transvalvular micro-axial heart pump 110 is fully within the patient's ventricle or aorta, the placement screen 204 may be updated to the manner shown in FIG. 2(f). More specifically, a notification 275 may be added to the notification area 270. As yet another example, when the outlet area 115 is positioned on or near the patient's aortic valve, the placement screen 204 may be updated to the manner shown in FIG. 2(g). More specifically, a notification 276 may be added to the notification area 270. As yet another example, when the pressure sensor 114 fails and the controller 130 is unable to calculate the heart pump flow measurement 242, the indwelling screen 204 may be updated to the format shown in Figure 2(h). More specifically, the heart pump flow measurement 242 may be replaced with a table of estimated flows and corresponding MAPs, and a notification 277 may be added to the notification area 270.

[0043] 3 illustrates a system 300 for monitoring and / or controlling multiple medical device controllers, such as controller 130. System 300 may include medical device controllers 312, 314, 316, and 318, computer network 322, local area network (LAN) 324, remote link module 332, router 334, wireless access point 336, base station 338, server 342, data store 344, OCR engine 346, and / or monitoring stations 352 and 354. Computer network 322 may include wired and / or wireless segments and / or networks. For example, computer network 322 may include a wireless network conforming to the IEEE 802.11x standard (e.g., a wireless local area network (WLAN), commonly referred to as "Wi-Fi"), represented by wireless access point 336, and / or a cellular network, represented by base station 338. As another example, the computer network 322 may include private and / or public networks, such as a LAN 324, a metropolitan area network (MAN), and / or a wide area network (WAN), such as the Internet (not shown).

[0044] The system 300 illustrates several different ways in which the medical device controllers may be connected to the computer network 322. For example, the medical device controller 312 is directly connected to the computer network 322. As another example, the medical device controller 314 is optionally connected to the computer network 322 through a remote link module 332. As yet another example, the medical device controller 316 is connected to the computer network 322 through a LAN 324 and a router 334. As yet another example, the medical device controller 318 is connected to the computer network 322 through a LAN 324, a router 334, and a wireless access point 336. The medical device controller 318 is also connected to the computer network 322 through a base station 338. In other embodiments, medical device controllers may be added and / or removed from the system 300. Additionally, multiple medical device controllers may be connected to the computer network 322 in a similar manner. For example, multiple medical device controllers may be directly connected to the computer network 322 in much the same manner as the medical device controller 312.

[0045] The server 342 may be configured to request status information from the medical device controllers 312, 314, 316, and 318 over the computer network 322. In some embodiments, the server 342 requests the status information automatically and / or repeatedly. In some embodiments, the status information includes an image of the contents of a screen displayed by a display associated with the medical device controllers 312, 314, 316, and / or 318. For example, the status information may resemble an image of any one of the screens illustrated in FIGS. 2(a)-(h). The image may be sent as one or more messages encoded as a video frame or as a sequence of video frames. Additionally, the video frames may contain, for example, a pixelated copy of the image. In some embodiments, the status information includes information derived from one or more of the data fields displayed by a display associated with the medical device controllers 312, 314, 316, and / or 318. For example, the status information may include information from one or more of the following data fields: heart pump type 211, heart pump serial number 212, date and time 214, heart pump current speed (performance) setting 230, heart pump flow measurements 242, purge system measurements 244, status indicators 251, and / or notification area 270.

[0046] The server 342 may also be configured to process the received status information. For example, when the server 342 receives an image of the contents of a screen displayed by a display associated with the medical device controller 312, 314, 316, and / or 318, the server 342 may parse the image and extract text information by optical character recognition (OCR) of portions of the image. In some embodiments, the extracted text information includes information from one or more of the data fields displayed by the display associated with the medical device controller 312, 314, 316, and / or 318. In some embodiments, the server 342 includes an OCR engine to parse the image and extract the text information. In some embodiments, the server 342 communicates with an external OCR engine, such as the OCR engine 346, to parse the image and extract the text information.

[0047] The data store 344 may be configured to store the raw and / or processed status information. For example, the data store 344 may store images of screen content displayed by displays associated with the medical device controllers 312, 314, 316, and / or 318, and / or text information extracted from the images by the server 342 and / or the OCR engine 346. The data store 344 may also be configured to provide at least a portion of the raw and / or processed status information to the monitoring stations 352 and 354 upon request. The monitoring stations 352 and 354 may be, for example, phones, tablets, and / or computers. In some embodiments, the monitoring stations 352 and 354 may use cloud-based technology to securely and remotely display at least a portion of the raw and / or processed status information on associated displays. For example, monitoring stations 352 and 354 may use an online device management system, such as Impella Connect® by Abiomed, Inc., Danvers, MA, to securely and remotely display at least a portion of the raw and / or processed status information.

[0048] In some embodiments, the server 342 and / or the monitoring stations 352 and / or 354 may also be configured to remotely send commands to one or more medical device controllers (e.g., medical device controllers 312, 314, 316, and / or 318) in the system 300. For example, if a controller 130 is added to the system 300, the server 342 and / or the monitoring stations 352 and / or 354 may be configured to remotely adjust the power delivered to the motor in the motor housing 116, the flow rate of the purging fluid through the purging subsystem 150, and / or the pressure of the purging fluid in the purging subsystem 150 by remotely sending commands to the controller 130. In some embodiments, one or more medical device controllers (e.g., medical device controllers 312, 314, 316, and / or 318) in the system 300 may offload one or more computational processes to the server 342 and / or the monitoring stations 352 and / or 354. For example, if controller 130 is added to system 300, controller 130 may offload complex calculations (e.g., machine learning algorithms) to server 342 and / or monitoring stations 352 and / or 354. To reduce latency, controller 130 may also offload such calculations to another computing device (not shown) on the same LAN.

[0049] As shown in FIG. 4, the cardiac cycle contains four phases: isovolumic contraction 410, ejection 420, isovolumic relaxation 430, and filling 440. During the cardiac cycle, contraction and relaxation of the heart muscle in the heart chambers causes two valves, the mitral valve 452 and the aortic valve 454, to open and close due to pressure differences. During isovolumic contraction 410, the mitral valve 452 and the aortic valve 454 close, and the pressure in chamber 456 increases until it is high enough for the aortic valve 454 to open. During ejection 420, the mitral valve 452 closes, the aortic valve 454 opens, and blood leaves chamber 456 and flows into the aorta. During isovolumic relaxation 430, the mitral valve 452 and the aortic valve 454 close, and the pressure in chamber 456 decreases until it is low enough for the mitral valve 452 to open. During filling phase 440, mitral valve 452 opens and aortic valve 454 closes, allowing blood to flow into chamber 456. The first two phases are known as systole and the last two phases are known as diastole.

[0050] FIG. 5 illustrates the regular waveforms of intra-aortic pressure (AoP), left ventricular pressure (LVP), differential pressure (dP), pump flow, and motor current and their relationship to contraction and diastole. The AoP waveform corresponds to the pressure in the patient's ascending aorta (e.g., ascending aorta 124). The LVP waveform corresponds to the pressure in the patient's left ventricle (e.g., left ventricle 128). The dP waveform corresponds to the differential pressure between the patient's ascending aorta and left ventricle. The pump flow waveform corresponds to the rate at which blood is drawn from the left ventricle into the ascending aorta by the transvalve micro-axial heart pump (e.g., transvalve micro-axial heart pump 110). The motor current waveform corresponds to the current provided to the motor of the transvalve micro-axial heart pump (e.g., the motor in motor housing 116).

[0051] Maintaining a consistent mean aortic pressure (MAP) is crucial to ensure adequate organ perfusion. See, for example, Chemla et al., Mean aortic pressure is the geometric mean of systolic and diastolic aortic pressure in resting humans, Journal of Applied Physiology 99:6, 2278-2284, 2005. Studies have shown that increased time spent below the MAP threshold of 65 mmHg is associated with poorer patient outcomes, including risk of death or organ dysfunction. See, for example, Varpula et al., Hemodynamic variables related to outcome in septic shock, Intensive Care Med. 31:1066-1071, 2005; Dunser et al., Arterial blood pressure during early sepsis and outcome, Intensive Care Med. 35:1225-1233, 2009; Dunser et al., Association of arterial blood pressure and vasopressor load with septic shock mortality: a post hoc analysis of a multicenter trial, Crit. Care Lond. Engl. 13:R181, 2009. As shown in Figure 5, physiological waveforms acquired using catheter-based hemodynamic support devices such as transvalvular microaxial flow heart pumps can be a rich source of hemodynamic information. However, advance warning of patient status based on predicted time series of MAP using such devices is rare.

[0052] Aspects of the present disclosure describe systems and methods for predicting intra-aortic pressure in a patient receiving hemodynamic support from a transvalve micro-axial heart pump. Advance warning of impending changes in intra-aortic pressure (e.g., MAP), even if the warning occurs only 5-15 minutes in advance, may aid in rapid management of the patient before total hemodynamic collapse. For example, if the patient's intra-aortic pressure is predicted to increase or remain stable, a clinician may initiate or continue a percutaneous coronary intervention (PCI) procedure. Similarly, if the patient's intra-aortic pressure is predicted to decrease, a clinician may delay or terminate the PCI procedure. Generally, a significant decrease in a patient's predicted intra-aortic pressure (e.g., a decrease of at least 10 mmHg) is indicative of a worsening patient condition. However, a sustained increase may also be indicative of a deteriorating patient condition.

[0053] The predicted stable trend in intra-aortic pressure also serves as a signal to wean the patient from the transvalve micro-axial heart pump. Similarly, the estimated intra-aortic pressure may be used to assign a level of support provided to the patient during the weaning process. For example, a clinical practitioner may adjust the pharmacological support provided to the patient based on the predicted intra-aortic pressure (e.g., by adjusting the amount of medication, such as a vasopressor or inotrope, provided to the patient). As another example, the motor speed setting (e.g., the current speed (performance) setting 230 of the heart pump) may be adjusted manually by a clinical practitioner and / or automatically by a connected medical device controller (e.g., the controller 130) based on the estimated intra-aortic pressure. For example, in some embodiments, the medical device controller may be configured to wean the patient from support by automatically and gradually reducing the motor speed setting over time. In such an embodiment, the medical device may, for example, temporarily increase the motor speed setting if the patient's condition is predicted to deteriorate (e.g., if the patient's intra-aortic pressure is predicted to decrease significantly).

[0054] In some embodiments, a display associated with the transvalve micro axial heart pump (e.g., display 140) may be configured to display the predicted intra-aortic pressure so that a clinician can react accordingly. For example, with respect to the screens illustrated in Figures 2(a)-(h), the predicted intra-aortic pressure may be displayed alongside the heart pump flow measurements 242 and / or the purge system measurements 244. As another example, a significant change in the intra-aortic pressure (e.g., + / - 10 mmHg) may trigger the display of a notification in the notification area 270 or an update of the status indicator 251. As yet another example, an intra-aortic pressure forecast screen may be displayed that includes a graph of the predicted intra-aortic pressure over time, much like the placement signal graph 252. As yet another example, graphs of predicted intra-aortic pressure over time may be added to the home screen 202, the placement screen 204, the purge screen 206, and / or the infusion history screen 208 and / or may replace a data field on one of these screens (e.g., the placement signal graph 252, the motor current graph 253, the purge flow graph 254, and / or the purge pressure graph 255).

[0055] As described above, the transvalvular micro-axial heart pump not only provides hemodynamic support and thus aids in native heart recovery, but is also equipped with one or more sensors (e.g., pressure sensor 114) for capturing measurements at the source, rather than peripherally, for example. Collectively, measurements obtained from the one or more sensors of the transvalvular micro-axial heart pump and the operating characteristics of the motor (e.g., the motor in the motor housing 116) of the transvalvular micro-axial heart pump can provide a rich set of data to which machine learning algorithms can be applied to predict the intra-aortic pressure of the patient. For example, machine learning algorithms may be applied to a set of features including intra-aortic pressure, motor current, motor speed, and / or motor speed settings (e.g., P-0, P-1, P-2, P-3, P-4, P-5, etc., for the Impella Catheter by Abiomed, Inc., Danvers, MA). The intra-aortic pressure may be derived from measurements obtained by the pressure sensor of the transvalvular micro-axial heart pump. The motor current may be derived from the energy intake of the transvalvular micro-axial heart pump motor. The motor speed may be derived from measurements taken by a signal generator of the transvalvular micro-axial heart pump. The motor speed may also be derived from the back electromotive force (EMF) of the transvalvular micro-axial heart pump motor. In some embodiments, the transvalvular micro-axial heart pump motor includes three or more motor windings. In such embodiments, the back EMF may be derived, for example, from a voltage measured across a motor winding that is disconnected from a power source. In some embodiments, the power source may be within a connected medical device controller (e.g., controller 130).

[0056] A variety of different machine learning algorithms may be applied to the measurements from the transvalvular micro-axial heart pump to predict the patient's intra-aortic pressure, such as Bayesian algorithms, clustering algorithms, decision tree algorithms, dimensionality reduction algorithms, instance-based algorithms, deep learning algorithms, regression algorithms, regularization algorithms, and rule-based machine learning algorithms. Some examples of deep learning algorithms include autoregressive integrated moving average (ARIMA) models, deep neural network (DNN) models, recurrent sequence-to-sequence models, recurrent sequence-to-sequence models with attention, transformer models, time convolutional neural network (TCN) models, and convolutional neural pyramid models. In some embodiments, these machine learning algorithms may be implemented by a medical device controller (e.g., controller 130) connected to the transvalvular micro-axial heart pump. In other embodiments, some or all of this processing may be offloaded to another device (e.g., server 342) on a computer network.

[0057] ARIMA models are popular statistical methods for time series forecasting. The components of the model are Autoregression (AR), Integrated, and Moving Average (MA). As a result, the model uses (a) a dependency between an observation and several lagged observations, (b) differencing of raw observations (subtracting an observation from its observation at the previous time step) to make the time series stationary, and (c) a dependency between the observation and the residual from a moving average model applied to the lagged observations. Additional information on ARIMA models can be found in Hyndman & Athanasopoulos, Forecasting: principles and practice, 2nd edition, Chapter 8 ARIMA models, OTexts: Melbourne, Australia, OTexts.com / fpp2, 2018, which is incorporated herein by reference.

[0058] A feedforward deep neural network (DNN) may be formed with an input layer, multiple hidden layers, and an output layer. The DNN may be used in an autoregressive fashion. In such an embodiment, the DNN may be built with a single unit in the output layer to make one-step-ahead predictions and continue to inductively feed back predictions for multiple-step-ahead predictions. Additional information regarding DNN models can be found in Schmidhuber, Deep Learning in Neural Networks: An Overview, arXiv:1404.7828v4, 2014, which is incorporated herein by reference.

[0059] A recurrent sequence-to-sequence model uses one encoder to map an input sequence to a fixed-size vector, and then maps the vector to a target sequence in a decoder. Additional information regarding recurrent sequence-to-sequence models can be found in Sutskever et al., Sequence to Sequence Learning with Neural Networks, NeurIPS 2014, which is incorporated herein by reference. A recurrent neural network (RNN) model may be used to preserve temporal information in a time series, as its hidden layer can store information processed through shared weights. For the encoder, a bidirectional RNN model may be used to process data in both forward and reverse directions. In some embodiments, two separate hidden layers may be used and then merged into the same output layer. For the decoder, an RNN model may be used to decode the target sequence from the hidden state. However, RNN models have difficulty learning long-term dependencies due to vanishing gradients. Long-short-term memory (LSTM) units can alleviate the vanishing gradient problem by using memory cell states. The overall structure 600 of a recurrent sequence-to-sequence model with LSTM units is illustrated in Figure 6. Additional information regarding LSTMs can be found in Hochreiter & Schmidhuber, Long Short-Term Memory, Neural Computation, Volume 9 Issue 8, 1997, which is incorporated herein by reference. In the remainder of this disclosure, the recurrent sequence-to-sequence model with LSTM units will be referred to simply as "LSTM."

[0060] Since the recurrent sequence-to-sequence model decodes each time step of the output from one fixed-length vector, it is necessary to compress all the necessary information of the input into one fixed-length vector. As a result, it may be difficult for the encoder-decoder network to learn all the useful information. To alleviate this problem, an attention mechanism may be applied. The attention mechanism can learn local information by utilizing intermediate states of the encoder for the context vector used by the decoder. Therefore, an attention mechanism may be used instead of a function to overcome the shortcomings of the fixed-length context vector by creating a shortcut between the context vector and the entire source input. Additional information on attention mechanisms can be found in Luong et al., Effective Approaches to Attention-based Neural Machine Translation, arXiv:1508.04025, 2015, which is incorporated herein by reference.

[0061] The Legendre Memory Unit (LMU) further addresses the vanishing and exploding gradient problems commonly associated with training RNNs by using a cell structure derived from first principles to project continuous-time signals into an orthogonal d-dimension. LMUs provide theoretical guarantees for learning long-range dependencies even as the discrete time steps Δt approach zero. This allows gradients to flow over the continuous history of the internal feature representation. LMUs are a recent innovation that achieves state-of-the-art memory capacity while ensuring energy efficiency, making them particularly well suited for chaotic time series forecasting tasks in the medical domain. Additional information on LMUs can be found in Voelker et al., Legendre Memory Units: Continuous-Time Representation in Recurrent Neural Networks, NeurIPS 2019, incorporated herein by reference.

[0062] The Transformer model is a transduction model that relies entirely on self-attention (note that attention here is different from the previous one) to compute input and output representations without using sequence-aligned RNNs or convolutions. Both the encoding and decoding components are identical stacks of layers, each composed of two sublayers: one multi-head attention layer and one fully-connected layer. The decoder has both layers, but between them there is an attention layer that helps the decoder focus on the output of the encoder stack. Instead of using a single scaled dot-product attention, the Transformer model projects the query Q, key K, and value V to the output as follows:

number

[0063] The TCN model has a convolutional hidden layer that operates on one-dimensional sequences. Convolutional neural networks create a hierarchical representation of the input sequence, where close input elements interact at lower layers and distant elements interact at higher layers. This allows for shorter paths to capture long-range dependencies compared to the chain-like structures modeled by recurrent networks. In some embodiments, the overall structure of the TCN model includes multiple convolutional blocks, followed by a flattening layer and multiple fully connected layers. In some embodiments, the absolute positions of the input elements may be embedded to give the model a sense of order. In some embodiments, a leaky relu activation function may be applied to each layer of the TCN model to avoid the "dead relu" problem. In some embodiments, dropout may be used to avoid overfitting. In some embodiments, residual connections may be used to improve the performance of the TCN model. The overall structure 800 of the TCN model is illustrated in FIG. 8. As shown, the TCN model includes multiple convolutional layers, followed by a flattening layer and multiple fully connected layers with residual connections. Additional information regarding the TCN model can be found in Bai et al., An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling, arXiv:1803.01271v2, 2018, which is incorporated herein by reference.

[0064] Advantageously, the TCN model has low memory requirements for training. Table 1 shows the complexity per layer of the LMU model, LSTM model, DNN model, pyramid model, TCN model, and Transformer model. In Table 1, n is the input length, d is the hidden size of the model, and k is the kernel size. For long sequences, such as a 5-minute real-time (RT) input sequence (e.g., with 7500 samples), the LSTM model can easily exhaust all available memory and cause the vanishing gradient problem. Moreover, the Transformer is very inefficient when the input length is larger than the hidden size of the model. In contrast, the TCN model can efficiently encode high-frequency data.

[0065] (Table 1) [Table 1]

[0066] In the convolutional neural pyramid model, a cascade of features is learned in two streams. The first stream across different pyramid levels expands the receptive field. The second stream learns the information at each pyramid level, and finally merges it to give the final result. As shown in Figure 9, the structure 900 of the convolutional neural pyramid model contains levels from 1 to N, where N is the number of levels. We call these levels L i where i ∈ [1, ..., N]. Different-scale content is assigned to each level L i The feature extraction and reconstruction operations are applied to each level respectively. L i The input to is the downsampled L (i-1) The features extracted from level L i In L, 2i convolutional layers are used for feature extraction. Then, a reconstruction operation fuses the information from two adjacent levels. For example, L i and L i+1 In the case of L i+1The output of is upsampled and then L i In some embodiments, the downsampling operation is implemented as a max pooling layer, and the upsampling operation is implemented as a deconvolution layer. Additional information regarding convolutional neural pyramid models can be found in Shen et al., Convolutional Neural Pyramid for Image Processing, arXiv:1704.02071v1 [cs.CV], 2017, which is incorporated herein by reference.

[0067] To test the effectiveness of some of the above mentioned deep learning algorithms in predicting intra-aortic pressure, patient data from 67 cases of transvalvular micro-axial heart pumps were acquired. Of these cases, 57 were indicated for HR-PCI (41 elective and 16 emergency). The remaining 10 were indicated for cardiogenic shock (CGS) due to acute myocardial infarction (AMI). In addition, another batch of 17 cases of transvalvular micro-axial heart pumps was used to compare the performance with respect to the amount of data.

[0068] The data from these cases included 25HZ intra-aortic pressure, 25HZ motor current, 25HZ motor speed, and other waveforms derived from these three signals (e.g., motor speed setting, left ventricular pressure, and heart rate). The data was captured by a medical device controller (e.g., controller 130) connected to a transvalvular micro-axial heart pump (e.g., transvalvular micro-axial heart pump 110). The 25HZ time series is referred to herein as real-time (RT) data. Averaged time (AT) data was derived from the RT data by averaging every 250 RT data points. In other embodiments, different amounts of RT data points may be averaged together to obtain AT data. In some embodiments, the amount of RT data points may be selected based on the desired time scale of the prediction. FIG. 10 illustrates a 10 second sample of the 25HZ RT intra-aortic pressure and motor speed time series. FIG. 11 illustrates a 20 minute sample of the 0.1HZ AT intra-aortic pressure time series. As shown in the figure, the waveform of the mean intra-aortic pressure is non-stationary and can indicate the long-term trend of the intra-aortic pressure and the patient's physical condition.

[0069] Since features such as motor speed setting, left ventricular pressure, and heart rate can be derived from motor speed and intra-aortic pressure, only motor speed and intra-aortic pressure were used in testing the effectiveness of some of the above-mentioned deep learning algorithms. Motor current was also not included as a feature because the average sequence contains less variation in motor current than motor speed and intra-aortic pressure. However, in other embodiments, any of these data sets may be used in addition to or instead of motor speed and / or intra-aortic pressure.

[0070] A sliding window was used to generate sequences of 15,000 samples (10 min duration). Sequences with sensor artifacts not reflecting physiological MAP (i.e., <50 mmHg, >200 mmHg) were removed. Changes in intra-aortic pressure of >10 mmHg were considered significant. These time series were categorized into three types: increasing sequences (I), decreasing sequences (D), and steady sequences (S). The overall change in increasing and decreasing sequences was >10 mmHg, and the overall change in steady sequences was <10 mmHg. Finally, 50,705 increasing RT sequences, 50,577 decreasing RT sequences, and 419,559 steady RT sequences were collected. All of these sequences were also converted to 0.1 Hz AT sequences of length 60.

[0071] Ten deep learning algorithms (i.e., ARIMA with averaging time (AT) input, DNN with AT input, LMU with AT input, LSTM with AT input, LSTM with attention with AT input, TCN with real-time (RT) input, TCN with AT input, Transformer with AT input, Pyramid with AT input, and Pyramid with RT input) were trained to predict the 5-minute future mean aortic pressure (MAP). In other embodiments, the prediction window may be increased or decreased. For example, in other embodiments, the prediction window may be increased to 10 minutes or 15 minutes. The ten deep learning algorithms were also trained with an RMS propagation optimizer and a learning rate decay of 0.8. A training-validation-test split of 60%-20%-20% was used. Since there are many possible combinations of hyperparameters, a random grid search of the hyperparameters was performed on a 10% holdout dataset. See, for example, Bergstra & Bengio, Random Search for Hyper-Parameter Optimization, Journal of Machine Learning Research 13 281-305, 2012. The search ranges for hyper-parameters are listed in Table 2. The root mean square error (RMSE) was used as the evaluation metric. The running average of the RMSE calculated on the validation set was used as the early stopping criterion. The same batch size of 64 was used for all tests.

[0072] (Table 2) [Table 2]

[0073] Figure 12 presents a comparison of the average RMSE achieved by some of the deep learning algorithms tested. From left to right, each bar plot presents the average RMSE achieved by LMU with AT input, LSTM with AT input, LSTM with attention with AT input, DNN with AT input, TCN with AT input, Transformer with AT input, and Pyramid with AT input. As shown, we tested each model on the increasing (I) only dataset, the decreasing (D) only dataset, the stationary (S) only dataset, and the IDS dataset. The IDS dataset contained equal proportions of all three types of sequences. The I, D, and S datasets contained 50,000 sample sequences. The IDS dataset contained 150,000 sample sequences. All of the models were trained on the IDS dataset. Overall, the LMU model consistently achieved the best average RMSE scores, including an average RMSE of 1.837 mmHg on the IDS dataset.

[0074] Figures 13 and 14 illustrate the MAP predictions generated by the two top performing models, LMU with AT input and LSTM with attention with AT input. Figure 13 illustrates the MAP predictions against the ground truth (e.g., true intra-aortic pressure) for a single recording over the course of 24 hours. The black line is the ground truth and the colored lines are the model's predictions. Figure 14 illustrates the MAP predictions with increasing, decreasing and steady sequences. The dashed line is the ground truth and the solid line is the model's predictions. The preceding 5 minutes of intra-aortic pressure and motor speed are the inputs to generate the predicted intra-aortic pressure values. As can be seen, both models closely follow the ground truth.

[0075] Table 3 shows all RMSE values ​​(mmHg) per cohort for each model trained on permutations of the increasing-decreasing-stationary (I, D, S) data set. The top number in each entry is the RMSE result for the combined cohort. The three values ​​in brackets are the RMSE for each of the three test sets containing only increasing, decreasing, and stationary sequences, respectively. All results are averages of five runs. The IDS training set contained equal proportions of all three types of sequences. The ID-only training set contained equal proportions of increasing and decreasing sequences. The IS-only training set contained equal proportions of increasing and stationary sequences. The DS-only training set contained equal proportions of decreasing and stationary sequences.

[0076] Figure 15 presents a comparison of the average RMSE achieved by some of the deep learning algorithms tested. Each bar plot presents, from left to right, the average RMSE achieved by LMU with AT input, LSTM with AT input, LSTM with attention with AT input, DNN with AT input, TCN with AT input, Transformer with AT input, and Pyramid with AT input. Different datasets for training and testing were used with these models. The light grey portion of each bar represents the improvement in predictive performance between the original patient cohort (N=20) and the latest patient cohort (N=67). Each model was trained on a permutation of the increasing-decreasing-stationary (I, D, S) dataset, as described above with respect to Table 3. Additionally, each model was tested on the increasing (I) only dataset, the decreasing (D) only dataset, the stationary (S) only dataset, and the IDS dataset, as described above with respect to Figure 12. Even without stationary sequences in the training set, all models achieve comparable or even better performance for predicting stationary sequences. Moreover, the improvements shown above each bar illustrate the potential for even better model performance as more data is collected in the future.

[0077] (Table 3) [Table 3]

[0078] Overall, these test results demonstrate that the above-mentioned systems and methods can be used to accurately predict a patient's intra-aortic pressure. Providing advance warning of impending changes in a patient's intra-aortic pressure, even if the warning occurs only 5-15 minutes in advance, can greatly improve clinical outcomes. For example, the authors of Wijnberge et al., Effect of a Machine Learning-Derived Early Warning System for Intraoperative Hypotension vs Standard Care on Depth and Duration of Intraoperative Hypotension During Elective Noncardiac Surgery: The HYPE Randomized Clinical Trial, JAMA, Caring for the Critically Ill Patient, doi: 10.1001 / jama.2020.0592, 2020, observed that the time spent in hypoglycemic events during surgery was significantly shorter when a machine learning warning system was used to inform clinicians about possible hypotension. Being able to predict and notify caregivers of significant changes in intra-aortic pressure (e.g., + / - 10 mmHg) gives clinicians time to intervene appropriately before hemodynamic instability occurs. In addition, predicting intra-aortic pressure can assist in weaning patients from mechanical circulatory support after native heart recovery. Because the level of hemodynamic support can be varied by changing the motor speed of the transvalvular pump, predicting MAP in advance can also assist in maintaining / increasing hemodynamic support.

[0079] From the foregoing description and with reference to the various drawings, those skilled in the art will recognize that certain modifications may be made to the present disclosure without departing from the scope of the present disclosure. While the drawings show several embodiments of the present disclosure, the disclosure is not intended to be limited thereto; the disclosure is to be as broad as the art will permit, and the specification is intended to be so read. Thus, the foregoing description should not be regarded as limiting, but merely as exemplifications of certain embodiments. Those skilled in the art will envision other modifications within the scope and spirit of the appended claims.

Claims

1. a transvalve micro axial heart pump having a motor and a pressure sensor; obtaining a set of intra-aortic pressure measurements corresponding to pressure values ​​measured by the pressure sensor during a period during which the transvalve micro axial pump is at least partially located within the patient's heart; obtaining a set of motor speed measurements corresponding to the rotational speed of the motor during said period; predicting the patient's intra-aortic pressure based on the set of intra-aortic pressure measurements and the set of motor speed measurements using the trained machine learning model; and automatically adjusting a speed setting of the motor based on the predicted intra-aortic pressure of the patient. one or more processors configured to A system comprising:

2. 2. The system of claim 1, wherein the one or more processors are further configured to obtain a set of current measurements corresponding to an energy draw of the motor during the period, and wherein the prediction is further based on the set of current measurements.

3. A transvalve micro axial flow heart pump A tube; an inlet area having one or more openings through which blood may be drawn into the tube by a motor; an outlet area having one or more openings through which blood may be expelled from the tube by the motor; 2. The system of claim 1, further comprising: a pressure sensor coupled to the outlet area.

4. the transvalve micro axial heart pump further comprising an additional pressure sensor coupled to said inlet area; the one or more processors are further configured to obtain a set of left ventricular pressure measurements corresponding to pressure values ​​measured by the additional pressure sensor during the time period; and the prediction is further based on the set of left ventricular pressure measurements.

4. The system of claim 3.

5. The system of claim 1 , wherein the machine learning model is a deep learning model.

6. 6. The system of claim 5, wherein the deep learning model is an Autoregressive Integrated Moving Average (ARIMA) model, a Deep Neural Network (DNN) model, a Recurrent Sequence to Sequence model, a Recurrent Sequence to Sequence model with Attention, a Transformer model, a Temporal Convolutional Neural Network (TCN) model, or a Convolutional Neural Pyramid model.

7. 6. The system of claim 5, wherein the deep learning model is a recurrent sequence-to-sequence model with a Legendre memory unit (LMU).

8. The machine learning model is trained on a dataset including increasing sequences, decreasing sequences, and stationary sequences; and Each sequence includes an intra-aortic pressure measurement and a motor speed measurement.

2. The system of claim 1.

9. 9. The system of claim 8, wherein a sequence is increasing if the intra-aortic pressure measurements in the sequence are increasing above a predefined threshold, the sequence is decreasing if the intra-aortic pressure measurements in the sequence are decreasing above the predefined threshold, and the sequence is steady if the intra-aortic pressure measurements in the sequence are neither increasing nor decreasing above the predefined threshold.

10. 10. The system of claim 9, wherein the predetermined threshold is 10 mmHg.

11. 10. The system of claim 8, wherein each sequence includes a predetermined number of aortic pressure measurements and motor speed measurements.

12. 9. The system of claim 8, wherein each sequence includes real-time (RT) intra-aortic pressure and motor speed measurements.

13. 9. The system of claim 8, wherein each sequence includes an average time (AT) intra-aortic pressure measurement and a motor speed measurement.

14. the machine learning model is trained on a dataset containing only increasing and decreasing sequences; and Each sequence includes an intra-aortic pressure measurement and a motor speed measurement.

2. The system of claim 1.

15. 14. The system of claim 13, wherein a sequence is increasing if the intra-aortic pressure measurements in the sequence are increasing above a predetermined threshold, and wherein a sequence is decreasing if the intra-aortic pressure measurements in the sequence are decreasing above the predetermined threshold.

16. Automatically adjusting the motor speed setting based on the patient's predicted intra-aortic pressure. temporarily increasing the speed setting of the motor when the predicted intra-aortic pressure of the patient is less than the patient's current intra-aortic pressure by more than a predetermined amount.

2. The system of claim 1, comprising:

17. a transvalve micro axial heart pump having a motor and a pressure sensor; obtaining a set of intra-aortic pressure measurements corresponding to pressure values ​​measured by the pressure sensor during a period during which the transvalve micro axial pump is at least partially located within the patient's heart; obtaining a set of motor speed measurements corresponding to the rotational speed of the motor during said period; and predicting an intra-aortic pressure for the patient based on the set of intra-aortic pressure measurements and the set of motor speed measurements using a trained machine learning model. and one or more processors configured to: a display configured to display the predicted intra-aortic pressure for the patient; and A system comprising:

18. 20. The system of claim 17, wherein the display is configured to simultaneously display the patient's predicted intra-aortic pressure and the patient's current intra-aortic pressure and the current speed setting of the motor.

19. 20. The system of claim 17, wherein the display is further configured to display an alert when the patient's predicted intra-aortic pressure is less than the patient's current intra-aortic pressure by more than a predetermined amount.

20. 20. The system of claim 17, wherein the display is configured to display the patient's predicted intra-aortic pressure as part of a graph.

21. 1. A method for treating a patient with a transvalve micro axial heart pump received within the patient's body, the method comprising the steps of: inserting the transvalve micro axial heart pump into the patient's body; obtaining a set of intra-aortic pressure measurements corresponding to pressure values ​​measured by a pressure sensor located on the transvalve micro axial heart pump during a period when the transvalve micro axial heart pump is at least partially located within the patient's heart; obtaining a set of motor speed measurements corresponding to the rotational speed of the motor during said period; predicting, using a trained machine learning model, an intra-aortic pressure for the patient based on the set of intra-aortic pressure measurements and the set of motor speed measurements; and Automatically adjusting a speed setting of the motor based on the predicted intra-aortic pressure of the patient.

22. 22. The method of claim 21, further comprising obtaining a set of current measurements corresponding to an energy draw of the motor during said time period, and wherein said prediction is further based on said set of current measurements.

23. A transvalve micro axial flow heart pump A tube; an inlet area having one or more openings through which blood may be drawn into the tube by a motor; an outlet area having one or more openings through which blood may be expelled from the tube by the motor; and wherein the pressure sensor is coupled to the outlet area.

24. the transvalve micro axial heart pump further comprising an additional pressure sensor coupled to said inlet area; the method further comprising obtaining a set of left ventricular pressure measurements corresponding to pressure values ​​measured by the additional pressure sensor during the period; and the prediction is further based on the set of left ventricular pressure measurements.

24. The method of claim 23.

25. 22. The method of claim 21, further comprising adjusting the amount of drug provided to the patient based on the predicted intra-aortic pressure.

26. 22. The method of claim 21, further comprising the step of reducing motor speed if intra-aortic pressure is predicted to increase.

27. 22. The method of claim 21, further comprising increasing the motor speed if intra-aortic pressure is predicted to decrease.

Images