Systems and methods for determining flow rate and cardiac output

Abstract

A system for determining a patient's hemodynamic state includes a first hemodynamic sensor and a display. The first hemodynamic sensor generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which instructions, when executed by one or more processors, cause the system to receive the first hemodynamic sensor signal, convert the right ventricular pressure waveform into an estimate of the blood flow waveform, and extract features from the right ventricular pressure waveform. The instructions further cause the system to estimate the patient's cardiac output based on the right ventricular pressure waveform or features extracted from the patient's right ventricular pressure waveform, and to output the blood flow waveform and cardiac output to the display.

Description

[Technical Field]

[0001] Cross-reference of related applications This application claims the interests of U.S. Provisional Application No. 63 / 492,176, filed on March 24, 2023, entitled “SYSTEMS AND METHODS FOR DETERMINING FLOW AND CARDIAC OUTPUT,” the disclosure of which is incorporated herein by reference in its entirety. This application also claims the interests of U.S. Provisional Application No. 63 / 492,173, filed on March 24, 2023, entitled “SYSTEMS AND METHODS FOR DETERMINING FILTERED CARDIAC OUTPUT,” the disclosure of which is incorporated herein by reference in its entirety. This application also claims the interests of U.S. Provisional Application No. 63 / 492,159, filed on March 24, 2023, entitled “SYSTEMS AND METHODS FOR VALIDATING HEMODYNAMIC DATA,” the disclosure of which is incorporated herein by reference in its entirety. [Background technology]

[0002] This disclosure relates in general to hemodynamic monitoring, and more specifically to determining blood flow and cardiac output in a patient using monitored hemodynamic data.

[0003] Monitoring a patient's hemodynamic variables enables improved patient care. Cardiac output, the amount of blood pumped by the heart per minute, is a crucial hemodynamic variable. Cardiac output is monitored by clinicians to make diagnoses and provide interventions in critically ill patients. However, the accuracy of cardiac output monitoring often decreases at high cardiac output levels. Furthermore, delays in cardiac output measurement can affect the accuracy of cardiac output information. The equipment required to monitor cardiac output is often expensive, difficult to obtain, or too invasive. [Overview of the project] [Means for solving the problem]

[0004] In one example, a system for determining a patient's hemodynamic state includes a first hemodynamic sensor and a display. The first hemodynamic sensor continuously generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which instructions, when executed by one or more processors, cause the system to receive the first hemodynamic sensor signal representing the patient's right ventricular pressure waveform, convert the patient's right ventricular pressure waveform into an estimate of a blood flow waveform, and extract features from the patient's right ventricular pressure waveform. The instructions further cause the system to estimate the patient's cardiac output based on the patient's right ventricular pressure waveform or features extracted from the patient's right ventricular pressure waveform, and to output the patient's blood flow waveform and the patient's cardiac output to the display.

[0005] In another example, a system for determining a patient's hemodynamic status includes a first hemodynamic sensor, a catheter connected to the first hemodynamic sensor, and a display. The first hemodynamic sensor continuously generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which instructions, when executed by one or more processors, are encoded to cause the system to receive the first hemodynamic sensor signal representing the patient's right ventricular pressure waveform and to extract features from the patient's right ventricular pressure waveform. The instructions further cause the system to compare the features extracted from the patient's right ventricular pressure waveform with one or more specified ranges of features in order to determine that the catheter is correctly positioned if the values ​​of the features extracted from the patient's right ventricular pressure waveform fall within one or more specified ranges of features, or to determine that the catheter is incorrectly positioned if the values ​​of the features extracted from the patient's right ventricular pressure waveform fall within one or more specified ranges of features. The command further instructs the system to use a machine learning model to convert the patient's right ventricular pressure waveform into an estimate of the patient's blood flow waveform, and to integrate the patient's blood flow waveform to determine the patient's cardiac output. The command further instructs the system to use a regression model to estimate the change in the patient's cardiac output based on features extracted from the right ventricular pressure waveform, and to estimate the patient's filtered cardiac output by filtering the patient's cardiac output using a Kalman filter algorithm that uses the change in the patient's cardiac output. The command further instructs the system to output the patient's blood flow waveform and / or the patient's filtered cardiac output to the display.

[0006] In another example, a system for estimating a patient's blood flow includes a hemodynamic sensor and a display. The hemodynamic sensor continuously generates a hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which, when executed by one or more processors, instructions are encoded to cause the system to receive the hemodynamic sensor signal representing the patient's right ventricular pressure waveform and to use a machine learning model to convert the patient's right ventricular pressure waveform into an estimate of the patient's blood flow waveform. The instructions further cause the system to output the estimate of the patient's blood flow waveform to a display or a module for determining the patient's cardiac output.

[0007] In another example, a method for estimating a patient's blood flow includes the steps of: receiving sensed hemodynamic data representing the patient's right ventricular pressure waveform by a hemodynamic monitoring system; and using a machine learning model, converting the patient's right ventricular pressure waveform into an estimate of the patient's blood flow waveform by the hemodynamic monitoring system. The method further includes the step of outputting the estimate of the patient's blood flow waveform by the hemodynamic monitoring system to a display for monitoring the patient's blood flow based on the patient's right ventricular pressure waveform, or to a module for determining the patient's cardiac output by the hemodynamic monitoring system.

[0008] In another example, a method for training an autoencoder model to estimate a patient's blood flow includes the steps of training an autoencoder model using measured animal right ventricular pressure waveforms and measured animal blood flow to obtain a trained autoencoder model of an animal. The method further includes the steps of inputting the measured animal right ventricular pressure waveforms into the trained autoencoder model of the animal to generate a predicted animal blood flow, and comparing the measured animal blood flow to a predicted animal blood flow to validate the trained autoencoder model of the animal. The method further includes the step of determining whether the predicted animal blood flow from the trained autoencoder model of the animal is valid or invalid. The method further includes the steps of inputting a human right ventricular pressure waveform into the trained autoencoder model of the animal to generate a raw human blood flow, and scaling the raw human blood flow using intermittent cardiac output to generate a scaled raw human blood flow. The method further includes the step of retraining an animal-trained autoencoder model using human right ventricular pressure waveforms and human scaled raw blood flow to obtain a human-trained autoencoder model for estimating patient blood flow based on the patient's right ventricular pressure waveform.

[0009] In another example, a system for determining a patient's hemodynamic state includes a first hemodynamic sensor and a display. The first hemodynamic sensor continuously generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which instructions, when executed by one or more processors, cause the system to receive the first hemodynamic sensor signal representing the patient's right ventricular pressure waveform and to extract information from the patient's right ventricular pressure waveform. The instructions further cause the system to estimate changes in cardiac output based on features extracted from the patient's right ventricular pressure waveform using a regression model, and to output the changes in cardiac output to a display and / or a cardiac output estimator submodule for calculating cardiac output.

[0010] In another example, a system for determining a patient's hemodynamic state includes a first hemodynamic sensor, a second hemodynamic sensor, and a display. The first hemodynamic sensor continuously generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The second hemodynamic sensor continuously generates a second hemodynamic sensor signal representing the patient's pulmonary artery pressure waveform. The system further includes one or more processors and a computer-readable memory in which, when executed by one or more processors, instructions are encoded to cause the system to receive the first hemodynamic sensor signal representing the patient's right ventricular pressure waveform and to extract features from the patient's right ventricular pressure waveform. The instructions further cause the system to receive the second hemodynamic sensor signal representing the patient's pulmonary artery pressure waveform and to extract features from the patient's pulmonary artery pressure waveform. The command further instructs the system to use a regression model to estimate changes in cardiac output based on features extracted from the patient's right ventricular pressure waveform and features extracted from the patient's pulmonary artery pressure waveform, and to output the changes in cardiac output to the display and / or to the cardiac output estimator submodule for calculating cardiac output.

[0011] In another example, a method for determining a patient's hemodynamic state includes the steps of: receiving sensed hemodynamic data representing the patient's right ventricular pressure waveform by a hemodynamic monitoring system; and performing waveform analysis of the hemodynamic data by the hemodynamic monitoring system to extract features from the patient's right ventricular pressure waveform. The method further includes the steps of: estimating changes in cardiac output based on features extracted from the patient's right ventricular pressure waveform using a regression model by the hemodynamic monitoring system; and outputting the changes in cardiac output to a display and / or cardiac output estimator submodule for calculating cardiac output. [Brief explanation of the drawing]

[0012] [Figure 1] This schematic block diagram illustrates an exemplary hemodynamic monitoring system that determines a patient's cardiac output and blood flow based on hemodynamic data. [Figure 2] This is a perspective view of an exemplary hemodynamic monitor that analyzes right ventricular pressure waveforms and pulmonary artery pressure waveforms to provide patient cardiac output and blood flow. [Figure 3] This is a perspective view of an exemplary catheter that can be inserted into a patient and connected to one or more hemodynamic sensors to provide hemodynamic data to a hemodynamic monitor. [Figure 4] This is a perspective view of an exemplary minimally invasive pressure sensor that can be attached to a patient to sense hemodynamic data representing the patient's right ventricular pressure or pulmonary artery pressure. [Figure 5] This is a perspective view of an exemplary oxygen saturation measurement module for receiving oxygen saturation measurement data from a catheter inserted into a patient. [Figure 6] This is a schematic block diagram showing the inputs and outputs of a module for a hemodynamic monitoring system. [Figure 7] This graph shows an exemplary right ventricular pressure waveform trace, including exemplary indices showing blood flow and cardiac output. [Figure 8]This graph shows an exemplary pulmonary artery pressure waveform trace, including exemplary indices showing blood flow and cardiac output. [Figure 9] Figure 1 is a schematic block diagram showing the verification module. [Figure 10] This graph shows an exemplary right ventricular pressure waveform trace, including indicators of catheter placement and data quality. [Figure 11] This graph shows exemplary right ventricular pressure waveform traces and exemplary pulmonary artery pressure waveform traces, including exemplary indicators of catheter placement and data quality. [Figure 12] Figure 1 is a schematic block diagram showing the flow module. [Figure 13] This is a schematic block diagram showing how the right ventricular pressure waveform is converted to a processed flow waveform via a flow module. [Figure 14] This is a schematic block diagram showing the autoencoder model of the flow module. [Figure 15] This is a schematic diagram showing the filter of an autoencoder model. [Figure 16] This flowchart illustrates an exemplary process for training an autoencoder model to predict human blood flow based on right ventricular pressure waveforms. [Figure 17] Figure 1 is a schematic block diagram showing the COAE module. [Figure 18] Figure 1 is a schematic block diagram showing the COLR module. [Figure 19] This is a schematic block diagram showing the reference and current features used as input to the regression model of the COLR module. [Figure 20A] Figure 1 is a schematic block diagram showing the COfiltered module, including a first example of a filter submodule. [Figure 20B] This is a schematic block diagram showing the COfiltered module shown in Figure 1, including a second example of a filter submodule. [Figure 21A] This graph shows example prediction values ​​for the COfiltered module. [Figure 21B] This graph shows an example of time-series measurements for the COfiltered module. [Figure 21C] This is an example graph from the COfiltered module showing filtered values ​​over time. [Figure 22] This graph compares exemplary cardiac output values ​​from the COAE module, COLR module, and COfiltered module with exemplary intermittent cardiac output values ​​over time for patients in the pre-bypass, post-bypass, and ICU states. [Figure 23] Figure 1 is a schematic block diagram showing the inputs and outputs of each module in an exemplary implementation of a hemodynamic monitoring system. [Modes for carrying out the invention]

[0013] As described herein, a hemodynamic monitoring system implements a series of modules that generate continuous blood flow and cardiac output measurements of a patient. The hemodynamic monitoring system utilizes catheters and hemodynamic sensors to generate right ventricular pressure ("RVP") waveforms and optionally pulmonary artery pressure ("PAP") waveforms for input into the hemodynamic monitoring system modules to generate patient blood flow and cardiac output during procedures in, for example, an operating room (OR), intensive care unit (ICU), or other patient care environment. Healthcare professionals can use the blood flow and cardiac output information to improve patient care.

[0014] Hemodynamic monitoring system (Figures 1-8) Figure 1 is a schematic block diagram of a hemodynamic monitoring system 10 that determines a patient's cardiac output and blood flow rate based on hemodynamic data. Figure 1 shows the hemodynamic monitoring system 10, which includes a hemodynamic monitor 12 and hemodynamic sensors 14 (including hemodynamic sensors 14A, 14B, 14C, and 14D). The hemodynamic monitor 12 includes a system processor 20, a system memory 22, a display 24, an analog-to-digital converter (ADC) 26, and a digital-to-analog converter (DAC) 28. The system memory 22 includes an RVP feature module 32, a PAP feature module 34, a validation module 36, a flow module 38, and CO AE Module 40 and CO LR Module 42 and CO filtered The system includes a flow rate and cardiac output software code 30, which includes a module 44. The display 24 includes a user interface 46, which includes a control element 48 and a sensory alarm 50. Figure 1 also shows a patient 16 and a healthcare worker 18.

[0015] As shown in Figure 1, the hemodynamic monitoring system 10 includes a hemodynamic monitor 12 and hemodynamic sensors 14 (including hemodynamic sensors 14A, 14B, 14C, and 14D). The hemodynamic monitoring system 10 can be implemented in a patient care environment, such as an ICU, OR, or other patient care environment, to monitor the hemodynamic status of a patient. As shown in Figure 1, the patient care environment may include a patient 16 and healthcare professionals 18 who have been trained to utilize the hemodynamic monitoring system 10.

[0016] As will be discussed later with respect to Figure 2, the hemodynamic monitor 12 can be an integrated hardware unit including a system processor 20, system memory 22, display 24, ADC 26, and DAC 28. In other examples, one or more components and / or functions of the hemodynamic monitor 12 can be distributed across multiple hardware units. For example, in some examples, the display 24 can be a separate display device located away from the hemodynamic monitor 12 and operably coupled to the hemodynamic monitor 12. Similarly, at least some of the data processing within the hemodynamic monitoring system 10 can be performed via a smart cable connected between a catheter (e.g., catheter 54 shown in Figure 5) or sensor and the hemodynamic monitor 12. Generally, although illustrated and described as an integrated hardware unit in the example of Figure 1, it should be understood that the hemodynamic monitor 12 can include any combination of devices and components electrically, communicatively, or otherwise operably connected to perform the functions belonging to the hemodynamic monitor 12 herein.

[0017] As shown in Figure 1, the system memory 22 stores the flow rate and cardiac output software code 30. The flow rate and cardiac output code 30 includes the RVP feature module 32, the PAP feature module 34, the verification module 36, the flow rate module 38, and CO AE Module 40 and CO LR Module 42 and CO filtered The module 44 is included. The display 24 provides a user interface 46 which includes control elements 48 that enable user interaction with the hemodynamic monitor 12 and / or other components of the hemodynamic monitoring system 10. As shown in Figure 1, the user interface 46 also provides sensory alarms 50 that provide alerts to healthcare workers based on the patient 16's flow rate and / or cardiac output, as will be further described below.

[0018] The hemodynamic sensor 14 can be attached to patient 16 to sense hemodynamic data representing the patient's RVP waveform, PAP waveform, blood oxygen saturation ("SvO2"), cardiac output, or any combination of these hemodynamic data. The hemodynamic sensor 14 is operably connected to the hemodynamic monitor 12 to provide the sensed hemodynamic data to the hemodynamic monitor 12 (e.g., electrically and / or communicatively connected via wired, wireless, or both). In some examples, the hemodynamic sensor 14 provides the patient's hemodynamic data to the hemodynamic monitor 12 as an analog signal, which is then converted by the ADC 26 into digital hemodynamic data representing the RVP waveform and / or PAP waveform. In other examples, the hemodynamic sensor 14 can provide the sensed hemodynamic data to the hemodynamic monitor 12 in digital format, in which case the hemodynamic monitor 12 may not include or utilize the ADC 26. In yet another example, the hemodynamic sensor 14 can provide the hemodynamic data of the patient 16 as an analog signal to the hemodynamic monitor 12, which then analyzes this hemodynamic data in its analog form.

[0019] The hemodynamic sensor 14 may include one or more non-invasive, minimally invasive, or invasive sensors attached to the patient 16. For example, the hemodynamic sensor 14 may take the form of an invasive hemodynamic sensor 14A, such as a second pressure transducer 14A that provides RVP waveform data sensed at the right ventricular port 68C located in the right ventricle of the patient's heart (shown in Figure 4). The hemodynamic sensor 14 may take the form of an invasive hemodynamic sensor 14B, such as an oxygen saturation measurement module 14B that provides blood oxygen saturation data in the pulmonary artery based on optical pulses radiated from module 14B into the pulmonary artery, reflected and returned, and received by module 14B via the optical connector 62 of catheter 54 (shown in Figure 5). Furthermore, the hemodynamic sensor 14 may take the form of a non-invasive hemodynamic sensor. In some cases, the hemodynamic sensor 14 can be non-invasively attached to the limbs of patient 16, such as the forehead, wrists, arms, fingers, ankles, toes, or other limbs. The hemodynamic sensor 14 may also take the form of other invasive, minimally invasive, or non-invasive sensors.

[0020] In certain examples, the hemodynamic sensor 14 can be configured to sense the patient's RVP, PAP, or both right ventricular pressure and pulmonary artery pressure. In some examples, the hemodynamic sensor 14 may also be used to sense the patient's cardiac output, blood oxygen saturation in the pulmonary artery, or both cardiac output and blood oxygen saturation, in addition to the right ventricular pressure waveform and pulmonary artery pressure waveform. For example, one or more hemodynamic sensors 14 can be attached to the patient 16 via a radial artery catheter inserted in the patient's arm. In other examples, the hemodynamic sensor 14 can be attached to the patient 16 via a femoral artery catheter inserted in the patient's leg. Such techniques can also enable multiple hemodynamic sensors 14 to provide substantially continuous heartbeat-by-heart rate monitoring of RVP and PAP, as well as monitoring of the patient's cardiac output and blood oxygen saturation, or any combination of these hemodynamic data, over a long period of time, such as several minutes or several hours.

[0021] The system processor 20 executes flow rate and cardiac output software code 30 that implements an RVP feature module 32, a PAP feature module 34, a verification module 36, a flow rate module 38, a CO AE module 40, and a CO LR module 四十二, and a CO filtered module 44 to generate blood flow measurement values and / or cardiac output measurement values using the RVP waveform and optionally the PAP waveform.The example of the system processor 20 can 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 other equivalent discrete logic circuit or integrated logic circuit.

[0022] The system memory 22 can be configured to store information within the hemodynamic monitor 12 during operation.The system memory 22 is described as a computer-readable storage medium in some examples.In some examples, the computer-readable medium can include a non-transitory medium.The term "non-transitory" can indicate that the storage medium is not embodied in a carrier wave or a propagated signal.In a particular example, the non-transitory storage medium can store data that can change over time (e.g., in RAM or a cache).The system memory 22 can include volatile and non-volatile computer-readable memory.Examples of volatile memory can include random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), and other forms of volatile memory.Examples of non-volatile memory can include, for example, magnetic hard disks, optical disks, flash memory, or forms of electrically programmable memory (EPROM) or electrically erasable programmable (EEPROM) memory.

[0023] It should be noted that there seems to be an error in the original text where "CO " appears multiple times in an unclear context. The translation attempts to make sense of the overall structure and content as best as possible based on the available information. Also, "四十二" in the original was likely a misspelling or an incomplete expression, and the translation treats it as "module 42" for the sake of continuity.The display 24 can be a liquid crystal display (LCD), light-emitting diode (LED) display, organic light-emitting diode (OLED) display, or other display device suitable for providing information to the user in a graphical format. The user interface 46 can include graphical and / or physical control elements that enable user input to interact with the hemodynamic monitor 12 and / or other components of the hemodynamic monitoring system 10. In some examples, the user interface 46 can take the form of a graphical user interface (GUI) that presents graphical control elements presented on, for example, the touch-sensitive and / or presence-sensitive display screen of the display 24. In such examples, user input can be received in the form of gesture input, such as touch gestures, scroll gestures, zoom gestures, or other gesture inputs. In certain examples, the user interface 46 can take the form of physical control elements, such as physical buttons, keys, knobs, or other physical control elements configured to receive user input to interact with components of the hemodynamic monitoring system 10, and / or may include them.

[0024] An example of a hemodynamic monitor 12 is shown in Figure 2. An example of a catheter is shown in Figure 3. An example of a minimally invasive pressure sensor is shown in Figure 4. An example of an oxygen saturation measurement module is shown in Figure 5.

[0025] Figure 2 is a perspective view of a hemodynamic monitor 12 that analyzes RVP and PAP waveforms to provide cardiac output and blood flow of a patient 16. As shown in Figure 2, the hemodynamic monitor 12 includes a display 24 that presents a graphical user interface, including control elements 48 (e.g., graphical control elements) that enable user interaction with the hemodynamic monitor 12, in the example of Figure 1. The hemodynamic monitor 12 may also include a number of input and / or output (I / O) connectors 52 configured for wired connections (e.g., electrical and / or communicative connections) to one or more peripheral components, such as one or more hemodynamic sensors 14. While the example in Figure 2 shows five separate I / O connectors 52, it should be understood that in other examples, the hemodynamic monitor 12 may include fewer than five I / O connectors 52 or more than five I / O connectors 52. In yet another example, the hemodynamic monitor 12 may not include I / O connectors 52 and instead communicate wirelessly with various peripheral devices.

[0026] As illustrated with respect to Figure 1, the hemodynamic monitor 12 includes one or more system processors 20 and a computer-readable system memory 22 that stores executable flow rate and cardiac output software code 30 for generating continuous blood flow and cardiac output measurements. The hemodynamic monitor 12 can receive sensed hemodynamic data representing RVP and PAP waveforms, for example, via one or more hemodynamic sensors 14 connected to the hemodynamic monitor 12 via an I / O connector 52. The hemodynamic monitor 12 uses the received hemodynamic data and several profiling parameters (e.g., input features) to execute the flow rate and cardiac output software code 30 to obtain blood flow and cardiac output measurements, as further described below.

[0027] As shown in Figure 1, the hemodynamic monitor 12 can present a graphical user interface on the display 24. The display 24 can be an LCD, LED display, OLED display, or other display device suitable for providing information to the user in a graphical format. In some examples, such as the example in Figure 2, the display 24 can be a touch-sensitive and / or presence-sensitive display device configured to receive user input in the form of gesture input, such as touch gestures, scroll gestures, zoom gestures, swipe gestures, or other gesture inputs.

[0028] The hemodynamic monitor 12 receives hemodynamic data from the patient 16 via one or more hemodynamic sensors 14A, 14B, 14C, and 14D (collectively referred to as hemodynamic sensors 14). In response to receiving the hemodynamic data from the patient 16, the hemodynamic monitor 12 executes the flow rate and cardiac output software code 30 to determine the blood flow rate and / or cardiac output and to display the blood flow rate and / or cardiac output on the display 24. The hemodynamic monitor 12 can also activate sensory alarms (e.g., sensory alarm 50 shown in Figure 1) such as audible alarms, tactile alarms, or other sensory alarms in response to the blood flow rate and / or cardiac output measurements. Thus, the hemodynamic monitor 12 can provide alarms to warn healthcare professionals regarding the blood flow rate or cardiac output measurements.

[0029] Figure 3 is a perspective view of a catheter 54 that can be inserted into a patient 16 and connected to one or more hemodynamic sensors 14 to provide hemodynamic data to a hemodynamic monitor 12. For example, the catheter 54 may be connected to one or more pressure-sensing hemodynamic sensors 14A to detect the patient 16's RVP, PAP, or both right ventricular pressure and pulmonary artery pressure. The catheter 54 may also interface with an oxygen saturation measurement module 14B to sense the patient's mixed venous blood oxygen saturation. Protected by a sheath 56, the catheter 54 includes multiple lumens 58 to which a fluid connector 60, an optical connector 62, a thermistor connector 64, and a thermal filament connector 66 are connected to one of the ports 68, an implantable hemodynamic sensor 14C (e.g., a thermistor), or an implantable hemodynamic sensor 14D (e.g., a thermal filament). To facilitate insertion of the catheter 54 into the patient 16, or for specific hemodynamic measurements, the catheter 54 includes a balloon 70 located at the tip 72 of the catheter 54.

[0030] As shown in Figure 3, the catheter 54 includes a distal port connector 60A that communicates with port 68A at its tip 72. A proximal infusion connector 60B communicates with the proximal port 68B, located approximately 30 cm from the tip 72, and can be used to administer fluids and drugs into the patient's heart. A right ventricular pacing connector 60C communicates with the right ventricular port 68C, which may be located approximately 19 cm from the tip 72, or approximately 12-13 cm from the tip 72. Connector 60C can be used to sense RVP. A thermistor connector 64 electrically connects to a hemodynamic sensor 14C (e.g., a thermistor) located near the tip 72 of the catheter 54 to measure deep blood temperature in the pulmonary artery. In some examples of the catheter 54, a thermal filament connector 66 electrically connects to a hemodynamic sensor 14D (e.g., a thermal filament) embedded in the catheter 54 and located in the patient's right ventricle. In some examples, the catheter 54 does not include a thermal filament or a corresponding thermal filament connector 66. The balloon connector 60D communicates with the balloon 70 and can be used with a syringe 74 to inflate and deflate the balloon 70.

[0031] For example, after insertion into patient 16 via the introducer, the distal port connector 60A and the right ventricular paging connector 60C can be connected to separate pressure transducer sensors 14A. The first pressure transducer 14A provides PAP waveform data sensed at the distal port 68A located in the pulmonary artery of the patient's heart to the hemodynamic monitor 12, and the second pressure transducer 14A provides RVP waveform data sensed at the right ventricular port 68C located in the right ventricle of the patient's heart to the hemodynamic monitor 12. Pulmonary artery blood oxygen saturation data can be provided by the oxygen saturation measurement module 14B based on light pulses irradiated into the pulmonary artery from the oxygen saturation measurement module 14B and the reflected light received by the oxygen saturation measurement module 14B via the optical connector 62 of the catheter 54. Furthermore, using the thermal filament connector 66 and thermistor connector 64 and associated wiring, the hemodynamic monitor 12 can receive cardiac output data of patient 16, for example, using a thermodilution technique. Cardiac output measured via a thermal filament and the corresponding thermal filament connector 66 can be considered continuous cardiac output ("CCO"). If the catheter 54 does not contain a thermal filament, cardiac output can be determined using a thermistor after injecting a fluid bolus (or set of boluses) of known volume and temperature via the proximal injection port 68B using a thermodilution technique. After injection of the fluid bolus, cardiac output measured via the thermistor and the corresponding thermistor connector 64 can be considered intermittent cardiac output ("iCO"). Intermittent cardiac output iCO measurements can be taken at a frequency of, for example, several minutes, several hours, or even longer intervals, depending on the level of monitoring required by the patient. For example, a clinician may administer a bolus set of 3-4 fluid boluses, with one fluid bolus in the set administered approximately every minute so that the complete bolus set lasts for about 3 minutes. In one example, when a clinician is evaluating a patient's response to drug therapy or another medical intervention, fluid boluses may be administered very frequently, such as every minute or every few minutes.In another example, if the patient is relatively stable in the ICU, fluid boluses can be administered at a lower frequency, such as every hour or every six hours. Thus, catheter 54 can be used to deliver CCO and / or ICO, as described below with reference to Figures 6, 16, and 18-20A.

[0032] Catheter 54 is an example of a catheter that can be used to measure RVP, PAP, CCO, and / or iCO. In other examples, any catheter configured to measure RVP, PAP, CCO, and / or iCO can be used.

[0033] Figure 4 is a perspective view of a hemodynamic sensor 14A that can be attached to patient 16 to sense hemodynamic data representing patient RVP or PAP of patient 16. As shown in Figure 4, the hemodynamic sensor 14A includes a housing 76, a fluid input port 78, a catheter-side fluid port 80, and / or an I / O cable 82. The fluid input port 78 is configured to connect to a fluid source, such as a saline bag or other fluid input source, via piping or other hydraulic connections. The catheter-side fluid port 80 is configured to connect to a catheter inserted in the patient's arm (i.e., a radial artery catheter) or a catheter inserted in the patient's leg (i.e., a femoral artery catheter) (e.g., a radial artery catheter or a femoral artery catheter) via piping or other hydraulic connections. The I / O cable 82 is configured to connect to a hemodynamic monitor 12, for example, via one or more of the I / O connectors 52 (shown in Figure 2). The housing 76 of the hemodynamic sensor 14A encloses one or more pressure transducers, a communication circuit, a processing circuit, and corresponding electronic components to sense the fluid pressure corresponding to the patient's RVP or PAP, which is transmitted to the hemodynamic monitor 12 (shown in Figure 2) via the I / O cable 82.

[0034] During operation, a column of fluid (e.g., saline solution) is introduced from a fluid source (e.g., a saline bag) through a fluid input port 78 to a hemodynamic sensor 14A and then to a catheter-side fluid port 80 leading to a catheter inserted into the patient 16. The RVP or PAP is transmitted through the fluid column to a pressure sensor located in a housing 76 that senses the pressure in the fluid column. The hemodynamic sensor 14A converts the sensed fluid column pressure into an electrical signal via a pressure transducer and outputs the corresponding electrical signal to the hemodynamic monitor 12 (shown in Figure 1) via an I / O cable 82. Thus, the hemodynamic sensor 14 transmits analog sensor data (or a digital representation of analog sensor data) representing substantially continuous inter-cardiac monitoring of the patient 16's RVP or PAP to the hemodynamic monitor 12 (shown in Figure 1).

[0035] Figure 5 is a perspective view of the oxygen saturation measurement module 14B for receiving oxygen saturation measurement data from a catheter inserted into patient 16. As shown in Figure 5, the hemodynamic sensor 14B includes an optical transmitter and optical receiver, which are housed in a housing 86 and arranged to communicate with the catheter via an I / O connector 84 accessible through a protective door 88. Within the housing 86, as shown in Figure 5, the hemodynamic sensor 14B includes a communication circuit, a processing circuit, and corresponding electronic components to sense blood oxygen saturation data derived from light emission transmitted to the patient via the catheter and the corresponding return light received from patient 16 via the catheter. An electrical signal indicating the patient's blood oxygen saturation level is transmitted to the hemodynamic monitor 12 via a cable 90 and a connector 92 which interfaces with one of the I / O connectors 52 (shown in Figure 2).

[0036] Figure 6 is a schematic block diagram showing the inputs and outputs of each module of the hemodynamic monitoring system 10. Although the hemodynamic monitoring system 10 is described in relation to catheter 54, any suitable catheter can be used with the hemodynamic monitoring system 10. Figure 6 shows the RVP feature module 32, the PAP feature module 34, the validation module 36, the flow module 38, and CO AE Module 40 (a module that estimates cardiac output based on an autoencoder model), and CO LR Module 42 (a module that estimates cardiac output using a linear regression model), and CO filtered The image shows module 44 (a module for filtering one or more estimates of cardiac output) and output device 102. It also includes RVP feature module 32, PAP feature module 34, validation module 36, flow module 38, and CO2. AE Module 40, CO LR Module 42, and CO filtered Each of the modules 44 is a functional module of the flow rate and cardiac output software code 30, as shown in Figure 1. While the flow rate and cardiac output software code 30 is described herein as being divided into seven modules, in other examples the functionality of the flow rate and cardiac output software code 30 may be described as more or fewer modules, which may depend in some examples how the code is written or organized. Furthermore, any module and / or submodule can be a completely separate set of code. The modules of the flow rate and cardiac output software code 30 will be described in turn, but these modules may contain overlapping or mixed functionality.

[0037] The RVP feature module 32 is the first module of the flow rate and cardiac output software code 30 in the hemodynamic monitoring system 10. The RVP feature module 32 is the RVP waveform ("RVP waveform ) from RVP features ("RVP featuresThe code includes a method for extracting the following: RVP feature module 32 waveform It receives RVP as input. waveform This corresponds to hemodynamic data sensed by one of the hemodynamic sensors 14A and received by the hemodynamic monitor 12. waveform The sensed hemodynamic data corresponding to the RVP is passed to the RVP feature module 32. The RVP feature module 32 then processes the RVP waveform From RVP features Extract the following. The RVP feature module 32 is RVP features The RVP feature module 32 outputs the following to the verification module 36. features to CO LR Output is also sent to module 42.

[0038] The PAP feature module 34 is the second module of the flow rate and cardiac output software code 30 in the hemodynamic monitoring system 10. The PAP feature module 34 is the PAP waveform ("PAP waveform ) from PAP characteristics ("PAP features The code includes a method for extracting "). PAP Feature Module 34 is PAP waveform It receives as input. PAP waveform This corresponds to hemodynamic data sensed by one of the hemodynamic sensors 14A and received by the hemodynamic monitor 12. PAP waveform The corresponding hemodynamic data is passed to the PAP feature module 34. The PAP feature module 34 then processes the PAP waveform From PAP features Extract the following. PAP Feature Module 34 is PAP features The PAP feature module 34 outputs to the verification module 36. features to CO LR Output is also sent to module 42.

[0039] The validation module 36 is the third module of the flow rate and cardiac output software code 30 in the hemodynamic monitoring system 10. The validation module 36 cleans the data and performs RVPwaveform and PAP waveform The code includes methods to ensure that it is valid and reliable. Verification module 36 uses RVP waveform And, PAP waveform And, RVP features And, PAP features It receives and as input. RVP waveform and PAP waveform The corresponding sensed hemodynamic data is passed to the validation module 36. features Features are supplied from the RVP feature module 32 to the verification module 36. Similarly, PAP features This is supplied from the PAP feature module 34 to the validation module 36. The validation module 36 determines whether a valid RVP ("RVP_valid") and a valid PAP ("PAP_valid") are true or false (i.e., RVP waveform and PAP waveform The module containing the indication of whether it is valid or not (i.e., flow module 38, CO) is located within the dashed box shown in Figure 6. AE Module 40, CO LR Module 42, and CO filtered Output to module 44) and RVP waveform and PAP waveform A determination indicating whether it is valid can be output to an output device 102 such as a display 24 (shown in Figure 1). Instructions indicating that RVP_valid and PAP_valid are true (for example, "RVP_valid=true" and "PAP_valid=true") are used to output RVP waveform and PAP waveform The corresponding hemodynamic data is valid for further processing by the module contained within the dashed box. Instructions indicating that RVP_valid and PAP_valid are false (e.g., "RVP_valid=false" and "PAP_valid=false") indicate that RVP waveform and PAP waveform This means that the corresponding hemodynamic data is not valid for further processing. Validation module 36 also smooths the RVPwaveform The flow rate is output to the flow module 38, and RVP features And, PAP features and CO LR Output to module 42.

[0040] The flow module 38 is the fourth module of the flow rate and cardiac output software code 30 in the hemodynamic monitoring system 10. The flow module 38 is RVP waveform The code includes a method for estimating the blood flow waveform from the RVP. The flow module 38 uses RVP. waveform It receives RVP as input. waveform The corresponding sensed hemodynamic data is passed to the flow module 38. The flow module 38 receives the smoothed RVP from the validation module 36. waveform It can also receive the following. The flow module 38 determines whether the valid flow rate ("flow_valid") is true or false (i.e., the processed blood flow rate ("flow_valid") processed Instructions indicating whether ) is valid or invalid (CO) AE Output to module 40, flow processed A determination indicating whether it is valid can be output to an output device 102 such as a display 24 (shown in Figure 1). The instruction that flow_valid is true (for example, "flow_valid=true") indicates the estimated blood flow waveform flow processed However, CO AE This means that it is valid for further processing by module 40. The instruction that flow_valid is false (e.g., "flow_valid=false") means that the estimated blood flow waveform flow processed However, this means that it is not effective for further processing. The flow module 38 further, flow processed to CO AE Output to module 40. Flow module 38 outputs flow processed It can also be output to output device 102.

[0041] CO AEModule 40 is the fifth module of the flow and cardiac output software code 30 in the hemodynamic monitoring system 10. CO AE Module 40 includes a method in the code for deriving cardiac output from the estimated blood flow waveform. CO AE Module 40 receives flow processed as an input. flow processed is supplied from the flow module 38 to the CO AE Module 40. CO AE Module 40 also receives an indication from the flow module 38 indicating whether flow_valid is true or false, and whether CO AE Module 40 can continue or not. CO AE Module 40 outputs the autoencoder cardiac output (「CO AE 」) to the CO LR Module 42 and the CO filtered Module 44. CO AE Module 40 can also output CO AE to the output device 102.

[0042] CO LR Module 42 is the sixth module of the flow and cardiac output software code 30 in the hemodynamic monitoring system 10. CO LR Module 42 includes a method in the code for estimating the change in cardiac output and cardiac output based on RVP features and PAP features . CO LR Module 42 receives RVP features , PAP features , CO AE , CCO, and iCO as inputs. RVP features is supplied from the RVP feature module 32 or the verification module 36 to the CO LR Module 42. Similarly, PAP features is supplied from the PAP feature module 34 or the verification module 36 to the CO LR Module 42. CO AE is CO AECO from module 40 LR It is supplied to module 42. CCO and iCO correspond to patient 16 cardiac output data received by hemodynamic monitor 12 via catheter 54 using a thermal filament and / or thermistor and thermodilution technique, as described above with reference to Figure 3. CCO and iCO are CO LR Passed to module 42. CO LR Module 42 is linear regression cardiac output ("CO LR ) and linear regression changes in cardiac output ("ΔCO LR ) and CO filtered Output to module 44. CO LR Module 42 is CO LR and ΔCO LR It is also possible to output to output device 102.

[0043] CO filtered Module 44 is the seventh module of the flow rate and cardiac output software code 30 in the hemodynamic monitoring system 10. filtered Module 44 includes a method within the code for filtering estimates of cardiac output or changes in cardiac output via the Kalman filter algorithm. filtered Module 44 is CO AE CO LR , CCO, iCO, and ΔCO LR It receives CO as input. AE CO AE CO from module 40 filtered It is supplied to module 44. LR and ΔCO LR CO LR CO from module 42 filtered It is supplied to module 44. LR In the same manner as described above for Module 42, the CCO and iCO are CO filtered It is also passed to module 44. filtered Module 44 is filtered cardiac output ("CO filtered ") and filtered changes in cardiac output ("ΔCOfiltered The output is sent to output device 102.

[0044] The output device 102 is a device for receiving outputs from the modules of the flow rate and cardiac output software code 30. The output device 102 may include a display 24, as shown in Figure 1. For example, the output device 102 may receive final estimates of flow rate, cardiac output, and / or changes in cardiac output for display via the display 24. The output device 102 is connected to the flow module 38, CO AE Module 40, CO LR Module 42, and CO filtere It can receive the output from module 44. More specifically, output device 102 receives flow processed CO AE CO LR And, ΔCO LR CO filtered And, ΔCO filtered The system receives the following outputs. Each of these outputs can be displayed via the display 24 as a corresponding graph representing the value over time.

[0045] Although some modules are shown in Figure 6 as having multiple inputs and outputs, some of the inputs and outputs are arbitrary, allowing for many configurations of the hemodynamic monitoring system 10 (shown in Figure 1). In some examples, PAP features PAP waveform CCO and / or iCO may not be used. In some cases, the estimated cardiac output of patient 16 (shown in Figure 1) is CO AE Module 40, CO LR Module 42, and CO filtered Output can be obtained from any one or more of the modules 44. These configurations may depend, for example, on the combination of hemodynamic sensors 14 (shown in Figure 1) used or the desired output.

[0046] Each module of the hemodynamic monitoring system 10 will be described in more detail below. The RVP feature module 32 will be described with reference to Figure 7. The PAP feature module 34 will be described with reference to Figure 8. The verification module 36 will be described with reference to Figures 9 to 11. The flow rate module 38 will be described with reference to Figures 12 to 16. AE Module 40 will be explained with reference to Figure 17. LR Module 42 will be explained with reference to Figures 18 and 19. filtered Module 44 will be explained with reference to Figures 20A to 22.

[0047] Figure 7 is a graph showing the RVP waveform trace 104 including indices 106, 108, 110, 112, 114, 116, 118, and 120. In Figure 7, the RVP waveform trace 104 is RVP waveform This is an example waveform of RVP. waveform This corresponds to hemodynamic data sensed by one of the hemodynamic sensors 14A and received by the hemodynamic monitor 12. The RVP waveform trace 104 (represented by digital hemodynamic data) can include various indices showing the blood flow and cardiac output of the patient 16. features As explained above with reference to Figure 1, RVP is transmitted via the RVP feature module 32. waveform Extracted from RVP. waveform Before extracting metrics from the data, the heart rate detection algorithm identifies the start and end of individual heartbeats for each waveform. The RVP heart rate detection algorithm identifies the start of a heartbeat based on the maximum RVP, minimum RVP, the maximum or minimum rate of change in the RVP, and / or the second derivative with respect to time in the RVP. waveform After identifying the heart rate, various indicators of blood flow and cardiac output can be continuously extracted from the waveform for each heartbeat.

[0048] Index 106 of the RVP waveform trace 104 corresponds to minimum diastolic blood pressure. Index 108 of the RVP waveform trace 104 corresponds to end-diastolic blood pressure. Index 110 of the RVP waveform trace 104 corresponds to maximum systolic blood pressure. Index 112 of the RVP waveform trace 104 corresponds to end-systolic blood pressure. Slope S1 is the slope of the RVP waveform trace 104, which can also provide an index. Slope S1 is drawn at one location, but is representative of multiple slopes that can be determined at multiple locations along the RVP waveform trace 104. For example, index 114, which corresponds to the rate of change of maximum blood pressure over time during systolic rise (dP / dt), and index 116, which corresponds to the rate of change of minimum blood pressure over time during relaxation after end-systolic pressure (dP / dt), are further exemplary indices. Similarly, the second time derivative of the RVP waveform trace 104 can be determined at any location along the RVP waveform trace 104 and used as an index. Other exemplary indices include the RVP gradient, or the difference in blood pressure at different points in time during diastolic or systolic. For example, index 118 corresponds to the diastolic gradient, or the difference between minimum diastolic blood pressure (index 106) and end diastolic blood pressure (index 108). Index 120 corresponds to the right ventricular pulse pressure, or the blood pressure gradient equal to the difference between end diastolic blood pressure (index 108) and maximum systolic blood pressure (index 110).

[0049] For example, using the RVP feature module 32, the flow rate and cardiac output software code 30 can extract additional indices from the RVP waveform trace 104 based on the RVP waveform trace 104 over various periods. For instance, systolic rise (indicators 108-110), systolic fall (indicators 110-112), isovolumetric relaxation (indicators 112-106), diastolic (indicators 106-108), and heart rate interval (between indicators 106) can be determined by the flow rate and cardiac output software code 30. Such indices may include the average RVP over one of the intervals referred to above.

[0050] Figure 8 is a graph showing a PAP waveform trace 122 including indices 124, 126, 128, 130, and 132 indicating blood flow and cardiac output. In Figure 8, the PAP waveform trace 122 is PAP waveform This is an example waveform. PAP waveform This corresponds to hemodynamic data sensed by one of the hemodynamic sensors 14A and received by the hemodynamic monitor 12. The PAP waveform trace 122 (represented by digital hemodynamic data) may include various indices indicating the blood flow and cardiac output of the patient 16. features As explained above with reference to Figure 1, the PAP feature module 34 is used to access the PAP. waveform Extracted from PAP. waveform Before extracting metrics, the heart rate detection algorithm identifies the start and end of individual heartbeats for each waveform. The PAP heart rate detection algorithm identifies the start of a heartbeat based on the maximum PAP, minimum PAP, maximum or minimum rate of change in PAP, and / or the second derivative with respect to time in PAP. waveform After identifying the heart rate, various indicators of blood flow and cardiac output can be continuously extracted from the waveform for each heartbeat.

[0051] Indicator 124 of the PAP waveform trace 122 corresponds to the onset of a heartbeat. Indicator 126 of the PAP waveform trace 122 corresponds to the maximum systolic blood pressure, indicating the end of the systolic rise. Indicator 128 of the PAP waveform trace 122 corresponds to the presence of a overlapping notch and blood pressure, indicating the end of the systolic decay. Indicator 130 of the PAP waveform trace 122 corresponds to the minimum diastolic blood pressure of the patient's heartbeat. The mean PAP can also be used as an indicator. The PAP slope, or the blood pressure difference between points in the PAP waveform trace 122, can also be used as an indicator. For example, indicator 132 corresponds to the pulmonary artery pressure, or the difference between the minimum diastolic blood pressure (indicator 130) and the maximum systolic blood pressure (indicator 126). S2 is the slope of the PAP waveform trace 122, which can also provide an indicator. Slope S2 is drawn at one location, but is representative of multiple slopes that can be determined at multiple locations along the PAP waveform trace 122. For example, the indicator may include the maximum and / or minimum time derivatives of the PAP waveform trace 122.

[0052] For example, using the PAP feature module 34, the flow rate and cardiac output software code 30 can extract additional metrics from the PAP waveform trace 122 based on the PAP waveform trace 122 at various intervals. For example, the interval from maximum systolic blood pressure at metric 126 to diastolic blood pressure at metric 128, and the interval from the onset of heartbeat at metric 124 to diastolic blood pressure at metric 130 can be extracted from the PAP waveform trace 122. The flow rate and cardiac output software code 30 can identify additional metrics from the PAP waveform trace 122 at various intervals. For example, systolic rise (metrics 124-126), systolic fall (metrics 126-128), systole (metrics 124-128), diastolic period (metrics 128-130), and heart rate interval (between metrics 124) can be determined by the flow rate and cardiac output software code 30. Such metrics may include the average PAP during one of the intervals referred to above. The area under the curve of the PAP waveform trace 122, and the standard deviation of the PAP waveform trace 122 determined for the intervals referenced above, can also serve as additional indicators for patient 16.

[0053] Verification module (Figures 9-11) Figure 9 is a schematic block diagram showing a validation module 36 including a signal quality detector 134 and a determination block 136. Figure 10 is a graph showing an RVP waveform trace 137 including indices 108 and 110 indicating the placement of catheter 54 (shown in Figure 3) and data quality. Figure 11 is a graph showing RVP waveform traces 104 and PAP waveform traces 122 including indices 110, 126, 138, 140, and 142 indicating the placement of catheter 54 and data quality. Figures 9, 10, and 11 will be discussed together. The validation module 36 detects and continuously monitors catheter 54 placement problems and signal quality problems that affect data quality, using RVP waveform and PAP waveform Use data from here.

[0054] The verification module 36 filters or cleans the data and performs RVP waveform PAP waveform RVP features , and PAP features This ensures that it is valid and reliable. The RVP derived from the RVP feature module 32 and the PAP feature module 34, respectively. waveform RVP features and PAP waveform PAP features This is the input to verification module 36. RVP waveform and PAP waveform This is also input to the verification module 36. The verification module 36 performs RVP in the 10-second segment. waveform and PAP waveform Analyze the following: RVP waveform and PAP waveformThe segment may be approximately 10 seconds (e.g., 9.5 to 10.5 seconds), or any other suitable time segment. The validation module 36 can be implemented within a patient care environment, such as an ICU, OR, or another patient care environment. The validation module 36 refines the data collected from the catheter 54 and determines whether the catheter 54 is positioned correctly to yield high-quality data via the signal quality detector 134 and the determination block 136. Thus, the validation module 36 checks whether the data from the catheter 54 is correct for the flow module 38 and CO LR Determine whether it can be used for further analysis, such as in Module 42.

[0055] RVP features and PAP features This is input to the signal quality detector 134. The signal quality detector 134 is RVP features and PAP features The data generated is cleaned and the RVP signal quality index ("SQI") is used. RVP ) and PAP signal quality index ("SQI") PAP ") and, respectively, RVP waveform and PAP waveform Each of these is assigned to a 10-second segment. The signal quality detector 134 is RVP features and PAP features The standard deviation and RVP waveform and PAP waveform Artifacts in and RVP waveform and PAP waveform Check the standard deviation of the features that represent the difference between [the two points].

[0056] The signal quality detector 134 is RVP features and PAP features RVP features and PAP features To analyze the data by comparing it to a specified range of values, the signal quality algorithm RVP is used. features and PAP features It consumes and. RVP waveform and PAP waveformWhen both are present, features can be compared between waveforms, and features representing the differences between waveforms can be analyzed. The specified range of values ​​is the physiological limit plus the system tolerance, or RVP features PAP features , and RVP waveform and PAP waveform It corresponds to the standard deviation for each feature that represents the difference between and . The signal quality detector 134 is RVP features and PAP features By comparing it with the above, data quality is also checked. Therefore, the specified range of values ​​is RVP, which excludes heartbeats or noise. waveform and PAP waveform It can be used to identify artifacts and signal quality issues such as under-attenuated or over-attenuated signals. Data outside the specified range of values, such as negative pressure, abnormally high pressure, or maximum systolic blood pressure 126 in the PAP waveform trace 122 which is higher than the maximum systolic blood pressure 110 in the RVP waveform trace 104, represent physiologically impossible data.

[0057] Values ​​outside the specified range represent error data. The signal quality detector 134 cleans the data by removing the error data and performs RVP (Reverse Value Processing). waveform and PAP waveform The signal quality detector 134 smooths the signal based on the amount of error data detected. waveform and PAP waveform SQI in each 10-second segment RVP and SQI PAP Assigns to the signal quality detector 134. RVP and SQI PAP The results are output to the judgment block 136. The analyzed RVP waveform and PAP waveformError data in each 10-second segment (identified by features outside the specified value range) is removed from the 10-second segment at the heart rate level or any other appropriate level. If RVP_valid is true (e.g., "RVP_valid=true") and / or PAP_valid is true (e.g., "PAP_valid=true"), RVP waveform The remaining 10-second segment and / or PAP waveform The remaining 10-second segment, as well as the related RVP features and / or PAP features Each will be passed on for further analysis.

[0058] Figure 10 shows an exemplary RVP waveform trace 137 that can be analyzed by the signal quality detector 134. The signal quality detector 134 analyzes the RVP of the RVP waveform trace 137. features The RVP waveform trace 137 is similar to the RVP waveform trace 104 shown in Figure 7. waveform This is another exemplary waveform. In this example, the signal quality detector 134 compares the end-diastolic blood pressure 108 and maximal systolic blood pressure 110 of each heartbeat in the RVP waveform trace 137 with a specified range of values ​​for the RVP end-diastolic blood pressure 108 and RVP maximal systolic blood pressure 110, respectively. Data from each heartbeat, from one RVP end-diastolic blood pressure 108 to the next RVP end-diastolic blood pressure 108, or from one RVP maximal systolic blood pressure 110 to the next RVP maximal systolic blood pressure 110, are analyzed against the corresponding specified range of values. The RVP waveform trace 137 includes an end-diastolic blood pressure exclusion 108E and a maximal systolic blood pressure exclusion 110E.

[0059] Figure 10 shows heartbeats in an RVP waveform trace 137 extending from one RVP diastolic end-blood pressure 108 to the next RVP diastolic end-blood pressure 108, or from one RVP maximal systolic blood pressure 110 to the next maximal systolic blood pressure 110. The signal quality detector 134 detects, flags, and removes heartbeats containing data outside a specified value range via a signal quality algorithm. Thus, the diastolic end-blood pressure exclusion 108E and the maximal systolic blood pressure exclusion 110E are within heartbeats from segments shown in the RVP waveform trace 137 that contain erroneous data, and RVP waveform Since it is removed from the segment, it is removed and excluded from further analysis. As shown in Figure 10, the heartbeats from the maximum systolic blood pressure 110 to the next maximum systolic blood pressure exclusion 110E contain blood pressure values ​​outside the specified range because the RVP waveform trace 137 contains artifacts, noise, and the preceding end-diastolic blood pressure exclusion 108E. As a result, the signal quality detector 134 flags and excludes that heartbeat (from the maximum systolic blood pressure 110 to the maximum systolic blood pressure exclusion 110E in the RVP waveform trace 137), and all data from that heartbeat is excluded from the RVP waveform It is discarded. The end-diastolic blood pressure exclusion 108E, which is within the range of the aforementioned heart rate, is an end-diastolic blood pressure measurement that is excluded from further analysis. As can be seen further in Figure 10, the heart rate from the end-diastolic blood pressure exclusion 108E to the next end-diastolic blood pressure 108 in the RVP waveform trace 137 includes a sharp drop in blood pressure following the maximum systolic blood pressure exclusion 110E, which is outside the range of the specified value. The sharp drop in blood pressure in the RVP waveform trace 137 after the maximum systolic blood pressure exclusion 110E is outside the range of the specified value because such a sharp drop in blood pressure following the maximum systolic blood pressure 110 is not physiologically expected. As a result, the signal quality detector 134 flags and excludes that heart rate, and all data from that heart rate (from the end-diastolic blood pressure exclusion 108E to the next end-diastolic blood pressure 108 in the RVP waveform trace 137) is excluded from the RVP waveform It is discarded. The maximum systolic blood pressure excluded within the aforementioned heartbeat 110E is a maximum systolic blood pressure measurement that is excluded from further analysis. That heartbeat is associated with RVP features RVP waveformBefore transferring the segment to the subsequent module, the RVP corresponding to the RVP waveform trace 137 waveform Excluded from the segment, thereby RVP waveform Clean the signal quality detector 134. RVP In determining the PAP, excluded heartbeats are used from the maximum systolic blood pressure 110 to the maximum systolic blood pressure exclusion 110E, and from the end-diastolic blood pressure exclusion 108E to the next end-diastolic blood pressure 108, including noise and a sudden drop in blood pressure. The signal quality detector 134 is used for PAP. waveform PAP to clean and analyze features Using PAP in a similar way waveform It can be analyzed.

[0060] The specified range of values ​​is RVP waveform and PAP waveform If both are available, RVP features and PAP features It can also address the differences between them. Furthermore, RVP waveform and PAP waveform Various features can be derived from the differences between them, and the features being examined are RVP waveform and PAP waveform It depends on whether it is synchronous or asynchronous. Values ​​outside the specified range represent error data.

[0061] As can be seen in Figure 11, the signal quality detector 134 detects the RVP of the RVP waveform trace 104. features and PAP waveform trace 122 PAP features Features representing the differences between them can be analyzed. The RVP waveform trace 104 is synchronized with the PAP waveform trace 122. The RVP waveform trace 104 is RVP waveform This is an example waveform, and the PAP waveform trace 122 is PAP waveform This is an example waveform. In this example, the signal quality detector 134 analyzes the pulse travel time 138, the systolic gradient 140, and the mean blood pressure 142.

[0062] Indicator 110 is determined from the RVP waveform trace 104. Indicator 110 is the maximum systolic blood pressure or the end of the systolic rise in the RVP waveform trace 104. Indicator 126 is determined from the PAP waveform trace 122. Indicator 126 is the maximum systolic blood pressure or the end of the systolic rise in the PAP waveform trace 122. Indicators 138, 140, and 142 are determined from the difference between the RVP waveform trace 104 and the PAP waveform trace 122. Indicator 138 is the pulse duration, or the time elapsed between the maximum systolic blood pressure (indicator 110) in the RVP waveform trace 104 and the maximum systolic blood pressure (indicator 126) in the PAP waveform trace 122. Pulse duration 138 is RVP waveform and PAP waveform Delay between or between the heart rate and RVP waveform and PAP waveform This represents the time difference when detected in both. Indicator 140 is the systolic gradient, or the blood pressure gradient between the maximum systolic blood pressure of the RVP waveform trace 104 (indicator 110) and the maximum systolic blood pressure of the PAP waveform trace 122 (indicator 126). Indicator 142 is the mean blood pressure of the PAP waveform trace 122 and the RVP waveform trace 104.

[0063] The signal quality detector 134 compares the pulse travel time 138, systolic gradient 140, and mean blood pressure 142 with specified ranges of values ​​for the pulse travel time 138, systolic gradient 140, and mean blood pressure 142, respectively. The signal quality detector 134 also compares the difference between the maximum systolic blood pressure 110 of the RVP waveform trace 104 and the maximum systolic blood pressure 126 of the PAP waveform trace 122, or the RVP features Heart rate and PAP calculated from features It is also possible to check for other differences between the features of the synchronized RVP waveform trace 104 and PAP wavelength trace 122, such as the difference between the calculated heart rate and the RVP. waveform This indicates the presence of five heartbeats and synchronized PAP waveform If the presence of 10 heartbeats within the same time window indicates the presence of RVP, the signal quality detector 134 will determine the RVP waveform and / or PAP waveformThe system detects that the data from the segment is invalid. If the RVP waveform trace 104 and PAP waveform trace 122 are asynchronous, the signal quality detector 134 detects other additional RVP, such as changes in pulse travel time between the RVP waveform trace 104 and the PAP waveform trace 122. features and PAP features It can be analyzed.

[0064] The signal quality detector 134 detects any errors or values ​​outside the specified range for each indicator via the signal quality algorithm and flags them. The signal quality detector 134 uses RVP waveform And, PAP waveform And, RVP features And, PAP features Before transferring the data to the next module, the heart rate corresponding to the heart rate from the RVP waveform trace 104 and PAP waveform trace 122, which contain data flagged with errors, is transferred to the RVP waveform and PAP waveform Exclude from, thereby RVP waveform and PAP waveform Clean and

[0065] The signal quality detector 134 is SQI RVP and SQI PAP When determining this, RVP includes pulse running time 138, systolic gradient 140, and mean blood pressure 142. features and PAP features Any errors in the indicator identified based on the difference between the two are used. Among other parameters determined or derived from the RVP waveform trace 104 and PAP waveform trace 122, indicators 138, 140, and 142 are used when the signal quality detector 134 is RVP waveform and PAP waveform This provides additional data that can be analyzed. Therefore, RVP waveform and PAP waveform Analyzing both allows for a more comprehensive assessment of data quality, resulting in a more accurate analysis.

[0066] As discussed above, the signal quality detector 134 uses the SQI algorithm to perform SQI RVP And, SQI PAP This is combined with the RVP and PAP signal quality algorithm ("SQI"). COMB In order to determine "), each of them, RVP waveform and PAP waveform RVP from 10-second intervals features and PAP features In addition, such information is used. The signal quality detector 134 each uses RVP features And, PAP features And, RVP waveform and PAP waveform Based on the characteristics that represent the differences between them, SQI RVP And, SQI PAP And, SQI COMB Assign and .

[0067] SQI RVP and SQI PAP This refers to the quality of each segment of the waveform, or each RVP. waveform and PAP waveform RVP calculated from the 10-second segment features and PAP features This value indicates the degree of deviation in the signal quality. For example, the signal quality detector 134 can be assigned a signal quality index from zero to 5 (0 to 5), where zero (0) is the highest quality and 5 is the lowest quality. RVP and SQI PAP This can be any appropriate numerical range.

[0068] The signal quality detector 134, as discussed above, RVP waveform RVP features and PAP waveform PAP features RVP features The specified range of values ​​and PAP features It compares with the specified range of values. The signal quality detector 134 compares with various RVPs. features and PAP featuresHowever, based on how close each value is to the specified range, RVP waveform and PAP waveform SQI RVP and SQI PA Assign P. RVP waveform and PAP waveform Each of these includes RVP features and PAP features If the value is not within the specified range, the SQI values ​​from 2 to 5 will be used. RVP and SQI PAP It can be assigned. RVP waveform and PAP waveform Each of these includes RVP features and PAP features If the values ​​are within the specified range, each SQI has a value between zero and one (0 to 1). RVP and SQI PAP It can be assigned. The signal quality detector 134 is SQI RVP and SQI PAP The output is sent to the judgment block 136.

[0069] SQI RVP and SQI PAP This is supplied to the determination block 136 of the verification module 36, and the determination block 136 is RVP waveform and PAP waveform The determination block 136 determines whether the 10-second segment is valid. The determination block 136 can output the determination to the output device 102 (shown in Figure 6), which can be, for example, the display 24 (shown in Figure 1). The determination block 136 is a binary system, and each RVP outside the specified range of values features and PAP features RVP waveform and PAP waveform Assign zero (0) to each, and each RVP is within the specified range of values. features and PAP features RVP waveform and PAP waveform Assign 1 to it. For example, RVP features and PAPfeatures If the signal quality index has 2 to 5, the determination block 136 determines, respectively, RVP waveform and PAP waveform Assign zero (0) to it. RVP features and PAP features If the signal quality index is zero or 1 (0 or 1), the determination block 136 determines, respectively, RVP waveform and PAP waveform Assign 1 to it.

[0070] SQI RVP If zero (0) is assigned to it, the determination block 136 determines that RVP_valid is false (for example, "RVP_valid=false"). This is indicated by the arrow labeled "No" in Figure 9. The determination that RVP_valid is false can be sent from the determination block 136 to the display, which may display "INVALID RVP WAVEFORM". SQI RVP If 1 is assigned to it, the determination block 136 determines that RVP_valid is true (e.g., "RVP_valid=true"). This is indicated by the arrow labeled "Yes" in Figure 9. The determination that RVP_valid is true can be sent from the determination block 136 to the display, which may display "VALID RVP WAVEFORM". The verification module 36 is contained within the dashed box shown in Figure 6 (i.e., the flow module 38, CO AE Module 40, CO LR Module 42, and CO filtered Module 44) can also output an instruction indicating whether RVP_valid is true or false. Subsequent modules will, if RVP_valid is true, execute RVP waveform You can proceed using the data from the filtered 10-second segment, and if RVP_valid is false, RVP waveform You cannot continue using the data from here.

[0071] If SQIPAP is assigned zero (0), the determination block 136 determines that PAP_valid is false (e.g., "PAP_valid=false"). This is indicated by the arrow labeled "No" in Figure 9. The determination that PAP_valid is false can be sent from the determination block 136 to the display, which may display "INVALID PAP WAVEFORM". PAP signal quality index SQI PAP If 1 is assigned to it, the determination block 136 determines that PAP_valid is true (e.g., "PAP_valid=true"). This is indicated by the arrow labeled "Yes" in Figure 9. The determination that PAP_valid is true can be sent from the determination block 136 to the display, which may display "VALID PAP WAVEFORM". The verification module 36 is contained within the dashed box shown in Figure 6 (i.e., the flow module 38, CO AE Module 40, CO LR Module 42, and CO filtered Module 44) can also output an instruction indicating whether PAP_valid is true or false. Subsequent modules will, if PAP_valid is true, then PAP waveform You can proceed using the data from the filtered 10-second segment, and if PAP_valid is false, PAP waveform You cannot continue using the data from here.

[0072] SQI COMB RVP waveform and PAP waveform It is generated in the same way using features that represent the differences between the two. SQI COMB RVP waveform 10-second segments and / or PAP waveform The effectiveness of the 10-second segment can be further analyzed by the determination block 136.

[0073] If RVP_valid and / or PAP_valid are true, catheter 54 is properly positioned and RVP features and RVP waveform PAP features and PAP waveform It is of high quality (i.e., has few artifacts and is suitable for further processing and analysis). If RVP_valid is true and PAP_valid is false, then RVP features and RVP waveform It may have fewer artifacts, still be of high quality, and still be acceptable for further processing and analysis. For example, SQI COMB 1 is assigned to SQI RVP If 1 is assigned to it, RVP_valid is still true, and RVP features and RVP waveform It is of high quality and acceptable for further analysis. For example, validation module 36 smooths the RVP waveform This can be output to the flow module 38 for determining blood flow rate. If RVP_valid is false on its own, or false in combination with PAP_valid, the catheter 54 is not properly positioned or there is a signal quality problem, and the catheter 54 should be adjusted. RVP derived from a properly positioned catheter 54 features and RVP waveform and / or PAP features and PAP waveform It is excluded as it is of low quality and unacceptable for further analysis.

[0074] The validation module 36 determines whether the signal quality is sufficient and whether the catheter 54 is operating within physiological limits. In some cases, the validation module 36 continuously monitors the placement of the catheter 54 via signal quality. The validation module 36 determines that the catheter 54 is correctly positioned and therefore has a valid RVP that can derive blood flow and cardiac output measurements. waveform and PAP waveformTo determine whether accurate data is being collected that leads to this, RVP waveform RVP features and PAP waveform PAP features We will analyze this. For example, RVP waveform or PAP waveform If the line is flat, the catheter 54 may not be properly connected to the hemodynamic monitor 12, and blood flow and cardiac output cannot be derived. waveform or PAP waveform If the value exceeds or falls below the specified range, catheter 54 may be improperly positioned and delivering data indicating central venous pressure instead of RVP and / or PAP, respectively. If the data exceeds or falls below the specified range, the data is of poor quality and should not be used, and catheter 54 should be adjusted. If the data is within the specified range, the data is of high quality and can be used for further analysis in further modules, etc. Therefore, it is important to identify and monitor positioning problems and / or other data quality problems in real time to notify clinicians and disable the output of error information.

[0075] Flow module (Figures 12-16) Figure 12 is a schematic block diagram showing the flow module 38. Figure 13 is RVP waveform flow through flow module 38 processed This is a schematic block diagram showing the conversion to RVP. Figures 12 and 13 will be discussed together. waveform This is input to the flow module 38 to predict hemodynamics. The flow module 38 includes an autoencoder model (or "autoencoder") 144, a flow filter 146, and a decision block 148.

[0076] Flow module 38 is RVP waveform The input is received, and the estimated value of the processed blood flow waveform of patient 16 is flow processed And the processed blood flow signal quality index ("SQI").flow The flow module 38 outputs the smoothed or cleaned RVP from the verification module 36, as discussed above with respect to Figures 9 to 11. waveform It may receive RVP. waveform The absolute value of RVP is shown in the flow module 38 (and CO2 shown in Figure 17). AE Module 40) is not stored or analyzed, and relative RVP values ​​and RVP waveform For each patient (e.g., patient 16 shown in Figure 1), the shape can be normalized so that only the shape is used. As seen in Figure 13, RVP waveform The flow is transmitted via the flow module 38. processed Converted to RVP. waveform The raw blood flow rate is measured by the autoencoder model 144. raw It is converted into an estimated value of the waveform of ''). flow raw The flow then passes through the flow filter 146. processed It is converted to flow. raw This reflects the physiological prediction of blood flow, and SQI flow The autoencoder model 144 and the flow filter 146 are processed to generate the flow for each heartbeat. processed To generate each, RVP waveform and flow raw Analyze the 10-second interval. In some examples, the flow module 38 is flow processed It continuously outputs this.

[0077] Autoencoder Model 144 is RVP waveform Input and flow raw The autoencoder model 144 is a deep learning-based model that functions correctly when patient 16 has both low and high cardiac output. The autoencoder model 144 is discussed below with respect to Figure 16, RVP waveform to raw It is trained to convert to flow. The autoencoder model 144 is trained to convert to flow raw This is output to the flow filter 146.

[0078] The flow filter 146 receives flow from the autoencoder model 144. raw It receives as input, flow processed and SQI flow It outputs the following. The flow filter 146 is flow raw By comparing the characteristics of flow with a specified range of values ​​corresponding to the physiological limit, raw The flow filter 146 filters or cleans the flow, removing artifacts or physiological inaccuracies such as negative flow rates, flow rates in the thousands, square flow rates, and inconsistencies such as excessive or abrupt increases in flow rates, or excessively different flow rates per heartbeat. processed To remove or exclude from flow raw This filters out the flow. As a result, the flow filter 146 produces a smoother, more consistent, and more accurate blood flow waveform. processed This results in flow from the flow filter 146. processed is flow raw It is more accurate and physiologically valid.

[0079] Flow filter 146 is SQI flow Determine the flow processed Assign to SQI. flow is flow processed This value indicates the quality of the waveform. The flow filter 146 is flow processed In addition to SQI flow In order to output flow, as discussed above, raw The flow filter 146 compares this to a specified range of blood flow values. raw Based on the amount of physiological inaccuracies and / or inconsistencies, or error data detected and removed from SQI flow to processed Assign it to flow. processed The SQI is between zero and 5 (0 and 5). flow A value is assigned, with zero (0) being the highest quality and 5 being the lowest quality. SQIflow This can be any appropriate numerical range. The flow filter 146 is flow raw The flow filter 146 compares this to a specified range of blood flow values. raw Based on how close it is to the specified value range, SQI flow to processed Assign to flow raw If the flow is not within all or some of the specified ranges of values, or is not close to them, processed This includes SQI values ​​ranging from 2 to 5. flow It can be assigned. flow raw If it is within or close to all specified ranges of values, flow processed SQI has values ​​ranging from zero to one. flow It can be assigned. Flow filter 146 is SQI flow Output this to judgment block 148.

[0080] SQI flow is flow processed The flow is supplied to the determination block 148 of the flow module 38, which determines whether it is valid. processed The determination of whether it is enabled or disabled can be output to output device 102 (shown in Figure 6), and output device 102 can be, for example, display 24 (shown in Figure 1). The determination block 148 is a binary system and if the flow is not within the specified range of values processed Assign zero (0) to the flow within the specified range of values. processed Assign 1 to it. For example, SQI flow If the value is between 2 and 5, the determination block 148 is flow processed Assign zero (0) to SQI. flow If the value is zero or 1 (0 or 1), the determination block 148 determines flow processed Assign 1 to it.

[0081] SQI flowIf zero (0) is assigned to it, the determination block 148 determines that flow_valid is false (for example, "Flow_valid=false"). This is indicated by the arrow labeled "No" in Figure 12. The determination that flow_valid is false can be sent from the determination block 148 to the display, which may display "INVALID BLOOD FLOW". SQI flow If 1 is assigned to it, the determination block 148 determines that flow_valid is true (for example, "Flow_valid=true"). This is indicated by the arrow labeled "Yes" in Figure 12. The determination that flow_valid is true can be sent from the determination block 148 to the display, which may display "VALID BLOOD FLOW". The flow module 38 sends an instruction indicating whether flow_valid is true or false to CO AE It can also output to module 40, CO AE Module 40 is a valid flow processed It can be used as input.

[0082] If flow_valid is true, flow processed It is of high quality and acceptable for further processing and analysis. Effective flow processed It can also be used to improve patient care itself. If flow_valid is false, flow processed It is of low quality and will be excluded from further processing and analysis.

[0083] Flow module 38 is RVP waveform to processed Convert to that flow processed Determine whether it is of high quality. processed It can be used independently in patient care settings or in conjunction with other modules to generate cardiac output.

[0084] Figure 14 is a schematic block diagram showing the autoencoder model 144 of the flow module 38. The autoencoder model 144 includes an input 150, a filter 152, a latent space 154, a filter 156, and an output 158. The input 150 and filter 152 constitute the encoder 160, and the filter 156 and output 158 ​​constitute the decoder 162.

[0085] Input 150 of the autoencoder model 144 is RVP waveform This is a 10-second sample. By using a 10-second sample, the sample size is large enough for the waveform to be stable and more accurate, but the sample is continuous RVP waveform It is small enough to represent RVP. waveform The 10-second sample can be approximately 10-second samples (9.5 to 10.5 seconds) or any other suitable time sample. The autoencoder model 144 can receive 10-second samples on a rolling basis; for example, the sample input 150 is a 10-second sample, but can be input to the autoencoder model 144 every 2 seconds. The input 150 is encoded through the filter 152 of the autoencoder model 144 into a condensed version of the data from the input 150 in latent space 154. Latent space 154 stores the condensed data. The condensed data in latent space 154 is then decoded through the filter 156 of the autoencoder model 144 to become output 158. Output 158 ​​of the autoencoder model 144 is flow raw This is a 10-second sample. The autoencoder model 144 is RVP as described below with respect to Figure 16. waveform from flow raw It is trained to produce the following. Therefore, the encoder 160 of the autoencoder model 144 is RVP via the filter 152. waveform The part of the autoencoder model 144 encodes or compresses the input 150 into fewer variables that contain all the important information of the input 150. The decoder 162 of the autoencoder model 144 is flow rawTo produce output 158, such compressed encoded information is received and expanded via filter 156.

[0086] Traditionally, blood flow measurements from patients 16 are unavailable to healthcare professionals 18 due to the invasive nature of measuring human blood flow. Furthermore, conventional autoencoder models are trained to learn a latent space representation of the input, and the compressed input from the latent space is extended to the same output as the input. For example, a conventional autoencoder may have an RVP waveform input and produce an RVP waveform output.

[0087] The autoencoder model 144 is a machine learning model. More specifically, the autoencoder model 144 is RVP, as will be discussed below with respect to Figure 16. waveform to raw It is a deep learning-based model trained to use a neural network architecture to convert to RVP. Autoencoder model 144 is RVP waveform to raw It can be converted to a robust flow processed To bring about flow through the flow filter 146 raw It can process this. Therefore, the autoencoder model 144 provides a practical and minimally invasive method for generating estimates of continuous blood flow measurements of patient 16. As a result, the blood flow measurements of patient 16 are available for improved patient care. processed CO LR Module 42 and CO filtered Module 44 can also be used to generate the cardiac output of patient 16, as will be discussed below.

[0088] Figure 15 is a schematic diagram showing filters 152 and 156 of the encoder 160 and decoder 162 of the autoencoder model 144, respectively. The filter 152 of encoder 160 comprises multiple filters 152, and the filter 156 of encoder 162 comprises multiple filters 156.

[0089] Filter 152 is applied to input 150 of the autoencoder model 144. Filter 152 is RVP waveform This is a mathematical operation that transforms input 150 into compressed data. Each data sample of input 150 is reduced by a given coefficient by each filter 152 until the fully compressed data reaches the latent space 154. Filter 156 is a flow raw The compressed information in the latent space 154 is applied to produce the output 158. Data samples are expanded by each filter 156 with a given coefficient until the output 158 ​​is reached. The algorithm within the autoencoder model 144 automatically generates the architecture of filters 152 and 156, which is a convolutional neural network with automatically selected convolutions and pooling based on the input 150 and the desired output 158. A complex nonlinear relationship exists between the filtered data in each filtering layer 152 and 156. In the example in Figure 15, filters 152 and 156 have an 8-16-32 architecture. The first layer of filter 152 has 8 filters, the second layer of filter 152 has 16 filters, and the third layer of filter 152 has 32 filters. The first layer of filter 156 has 32 filters, the second layer of filter 156 has 16 filters, and the third layer of filter 156 has 8 filters. In the example in Figure 15, the autoencoder model 144 has 12,000 parameters and requires 112 convolution operations. The parameters are determined when the autoencoder model 144 is trained.

[0090] Filter 152 is RVP waveform It enables compression and encoding of the compressed data. Filter 156 processes the flow of the compressed data. raw This enables extension and decoding to RVP. As a result, filters 152 and 156 of the trained autoencoder model 144 are RVP waveform to raw It can be converted to [this].

[0091] Figure 16 shows human RVP waveform Based on human flow raw This is a flowchart showing process 164 for training the autoencoder model 144 to predict the flow rate. Process 164 for training the autoencoder model 144 includes steps 166-178. Once the autoencoder model 144 is trained through process 164, it is ready for use in the flow module 38, as discussed above.

[0092] In step 166, the autoencoder model 144 is trained using the measured RVP waveform of the animal and the measured blood flow rate of the animal. Both the animal's RVP waveform and the animal's blood flow rate are known in step 166. Since directly measuring blood flow rate in humans is too invasive, the information is derived from animal experiments by directly measuring blood flow rate in the same animal against a known RVP waveform of the animal. The autoencoder model 144 is trained using the animal's RVP waveform as input and the measured animal's blood flow rate as output to obtain an animal-trained autoencoder model 144.

[0093] In step 168, the animal's RVP waveform is input to the animal-trained autoencoder model 144 to generate the predicted animal blood flow. The known animal's RVP waveform is input to the animal-trained autoencoder model 144, and the waveform of the predicted animal blood flow is output from the autoencoder model 144.

[0094] In step 170, the measured animal blood flow is compared to the predicted animal blood flow to validate the training of the autoencoder model 144. Since both the animal's RVP waveform and the corresponding waveform of the measured animal blood flow are known, the predicted animal blood flow from the trained autoencoder model 144 can be compared with the known directly measured animal blood flow with respect to various parts of the animal's RVP waveform. Thus, the predicted animal blood flow waveform from the autoencoder model 144 is compared with the directly measured animal blood flow waveform. To determine the performance of the autoencoder model 144, metrics such as mean squared error (MSE), mean absolute error (MAE), correlation, and Brand-Altman analysis are used for comparison.

[0095] In step 172, the predicted animal blood flow from the trained autoencoder model 144 is determined to be valid. If the error between the predicted animal blood flow from the trained autoencoder model 144 and the directly measured animal blood flow is less than a preset threshold, the training of the autoencoder model 144 is determined to be valid. If the training of the autoencoder model 144 is determined to be valid, process 164 proceeds to step 174. If the error between the predicted animal blood flow from the trained autoencoder model 144 and the directly measured animal blood flow exceeds a preset threshold, the training of the autoencoder model 144 is determined to be invalid. If the training of the autoencoder model 144 is determined to be invalid, process 164 returns to step 166 and starts training the autoencoder model 144 again. The MSE or MAE index is iteratively calculated during training to adjust the model parameters until convergence occurs (e.g., to a minimum MAE or MSE).

[0096] In step 174, in order to generate human raw blood flow, human RVP waveformThis is input to the animal-trained autoencoder model 144. Human known RVP waveform However, the trained autoencoder model 144 of validated animals was input with raw human blood flow waveforms (i.e., "human flow"). raw The output from the autoencoder model 144 is ''. Therefore, step 174 is used to generate the shape of a raw human blood flow waveform. Since the animal training of the autoencoder model 144 has been validated, the shape of the raw human blood flow waveform should match that of a human blood flow (for example, the raw human blood flow waveform is RVP waveform However, if it indicates that the valve is open, it should be greater than zero, RVP waveform However, if the valve is closed, the value is close to zero.

[0097] In step 176, the raw human blood flow waveform is used to obtain the scaled human flow waveform. raw iCO is used to scale the waveform to generate it. Since the shape of the raw human blood flow waveform is accurate, the raw human blood flow waveform is used in step 176 for the flow over each heartbeat. raw The integral is scaled so that the average is the same as the iCO measured by catheter 54. Therefore, flow raw The amplitude of the waveform is adjusted. Therefore, the scaled flow of a human raw The accurately estimated flow raw This is generated.

[0098] In step 178, the autoencoder model 144 performs a human-scaled flow raw It is retrained using the human RVP waveform and scaled human flow to obtain a human-trained autoencoder model 144. raw The autoencoder model 144 is trained using the output. rawWhen retrained using, the autoencoder model 144 performs RVP on a human who is previously unknown. waveform From human flow raw It can be used to generate [something].

[0099] The animal data allows the autoencoder model 144 to be trained. As a result of training the autoencoder model 144 using the animal data, the autoencoder model 144 is RVP waveform Therefore, flow that cannot be measured in humans raw This allows for further analysis of the patient's condition and improved patient care.

[0100] CO AE Module (Figure 17) Figure 17 shows flow processed Autoencoder cardiac output CO2 AE Convert CO AE This is a schematic block diagram showing module 40.

[0101] CO AE Module 40 is flow processed The input and CO AE It has the output of flow. processed CO from flow module 38 AE Input to module 40, the flow module 38 receives the flow via autoencoder model 144. raw It generates and flows through the flow filter 146 raw to processed Filter by flow processed to CO AE Before sending to module 40, flow via determination block 148 processed It is determined that it is effective. The flow module 38 is CO AE Module 40 is (flow processed (In response to the instruction that it is valid) it can continue or (flow processed(In response to the instruction that it is invalid) the instruction to indicate whether flow_valid is true or false is also provided so that it cannot continue. AE Send to module 40. CO AE Module 40 is for patient 16 CO AE To derive flow processed Use this.

[0102] CO AE Module 40 is CO AE To generate flow processed Using a portion of the waveform, flow at heart rate levels processed The average value of the waveform is calculated. Therefore, flow processed The waveform is CO AE It is integrated to produce CO AE is flow processed This is a single value representing an estimate of cardiac output for the 10-second portion of flow. processed The integration of the waveform is performed every 10 seconds, CO AE This results in a continuous waveform. Samples are taken more frequently than every 10 seconds on a rolling basis. AE It can be input to module 40, in which case the sample is flow processed This is a 10-second sample, but every 2 seconds CO AE This can be input to module 40. processed By using a 10-second sample, the sample size is CO AE The waveform is large enough to be stable and more accurate, but the sample is CO AE It is small enough to represent a continuous waveform. AE It has a robust continuous waveform, CO AE The delay in CO is not too long in order to capture any changes in the cardiac output of patient 16. AE It is important to update this frequently, as this improves patient care.

[0103] Also, CO AE Module 40 is CO AE Relative RVP and RVP waveformSince it can be based on the form or shape and does not directly depend on the absolute pressure value, it can generally be described as "data-independent" with respect to the absolute pressure value (e.g., the absolute value of RVP). This is important because, in practice, the absolute pressure value is not always accurate. For example, healthcare workers may introduce some measurement error in RVP measurements when leveling a pressure transducer for a patient. Another example is raising or lowering a patient's bed for surgery or in the ICU, which can also change the absolute pressure value. AE Module 40 does not directly depend on absolute pressure values, therefore in these situations CO AE It can be estimated accurately. Furthermore, CO AE This allows for accurate calculation not only of low cardiac output but also of high cardiac output, which was previously impossible, further improving patient care. AE Module 40 is CO AE to CO LR Module 42 and CO filtered The output can be sent to module 44 and output device 102 (shown in Figure 6).

[0104] CO LR Modules (Figures 18-19) Figure 18 shows CO LR This is a schematic block diagram showing module 42. Figure 19 shows CO LR This is a schematic block diagram showing the reference and current features used as inputs to the regression model 180 in module 42. Figures 18 and 19 will be explained together. RVP features PAP features Demographic information, and SvO2 are used in RVP waveform and PAP waveform Based on the characteristics of CO, to estimate changes in cardiac output and / or cardiac output, LR This is input to module 42. Figures 18-19 show the CO2 model 180 and the cardiac output estimator 182. LR This shows module 42.

[0105] Regression model 180 is COLR This is the first submodule of module 42. Regression model 180 is RVP features PAP features It is configured to receive demographic information and SvO2 as input. The regression model 180 is an estimate of the change in cardiac output over time for patient 16 (shown in Figure 1), which is ΔCO2. LR The output is: Regression Model 180 is a conventional machine learning model that works well when the available data is not multimodal and is well distributed. For example, when compared to patient cardiac output data, changes in patient cardiac output data tend to be a more homogeneous dataset, so conventional machine learning techniques such as Regression Model 180 can be used to successfully learn across the entire population.

[0106] The regression model 180 is trained using a training dataset in which changes in cardiac output and corresponding changes in the features of the RVP and PAP waveforms are known for each patient. In examples where demographic information and / or SvO2 are used, the regression model 180 can be trained using a training dataset that also includes demographic information and / or SvO2 corresponding to known changes in cardiac output information.

[0107] Once the regression model 180 is trained, the regression model 180 is evaluated by the RVP feature module 32 (or validation module 36) as shown in Figure 6. waveform RVP derived from features It consumes. Regression model 180 is PAP by PAP feature module 34, as shown in Figure 6. waveform PAP derived from features You can also consume RVP. features and PAP features It is aggregated over a defined time window (or interval). For example, RVP features and PAP features This can be aggregated over a 10-second time window that includes multiple heartbeats. In other examples, RVP features and PAPfeatures This can be aggregated over longer or shorter time windows. The RVP feature module 32 and PAP feature module 34 (shown in Figures 1 and 6) can extract the features of each heartbeat within a time window and then calculate the average of each feature over the entire time window. The averaged features over the entire time window are consumed by the regression model 180. features and PAP features This can take the form of: In some examples, the regression model 180 is obtained from the averaged RVP obtained from good (i.e., valid or not discarded) heart rates within the time window determined by the validation module 36 (shown in Figure 9). features and PAP features It consumes only that.

[0108] Regression model 180 is RVP features and PAP features A group can be used in combination with the following. In one example, regression model 180 is RVP by RVP feature model 32. waveform All RVPs derived from features And, by PAP feature module 34, PAP waveform All PAP derived from features And can be used. In another example, regression model 180 uses the derived RVP features and PAP features Only a portion of it can be used. Furthermore, in other examples, regression model 180 is RVP features Use only PAP features Do not use it. Generally, regression model 180 uses individual RVP features and PAP features Any number or combination of these can be used. An example RVP used by regression model 180. featuresThis includes the following PAP values ​​used by regression model 180: pulse rate (i.e., calculation of heart rate based on pressure, although in other examples, heart rate referenced from an electrocardiogram or other means may also be used), maximum dP / dt, minimum dP / dt, systolic time, systolic blood pressure, end-systolic blood pressure, end-diastolic blood pressure, pulse pressure, and mean arterial pressure. features It also includes the following RVPs not specifically listed here, such as pulse rate, maximal dP / dt, minimum dP / dt, systolic time, systolic blood pressure, end-systolic blood pressure, end-diastolic blood pressure, pulse pressure, and mean arterial pressure. features and / or PAP features Furthermore, features such as pulse rate and systolic time can be used by regression model 180. For physiological reasons, RVP waveform or PAP waveform It is expected that similar values ​​will be obtained regardless of which of the following methods is used for derivation. Therefore, in some examples, regression model 180 is RVP features and PAP features Features that are predicted to be physiologically similar between the two are RVP waveform RVP derived from features And, PAP waveform PAP derived from features You can use only one source, not both, and the regression model 180 uses RVP for other features that are predicted to be physiologically different, such as maximum dP / dt and minimum dP / dt. features and PAP features Both can be used.

[0109] In some examples, the regression model 180 can also consume demographic information about patient 16. Such demographic information may include age, weight, height, sex (or gender), body mass index (BMI), and health status. The regression model 180 can receive demographic information about patient 16, for example, from system memory 22 (shown in Figure 1) where demographic information about patient 16 is stored (e.g., in a patient information database), or when it is entered by a healthcare professional 18 at the user interface 46 (shown in Figure 1). In some examples, the regression model 180 can also consume SvO2 data received via hemodynamic sensor 14B (shown in Figures 1 and 5). Overall, the RVP of patient 16 features PAP features Several variations of regression model 180 (i.e., more complex or less complex variations of regression model 180) using different combinations of demographic information and / or SvO2 are possible without changing the quality of regression model 180.

[0110] As shown in Figure 19, the regression model 180 has a reference time window W i The corresponding reference RVP feature ("RVP features_ref ) and the current time window W i+n The current RVP feature corresponding to (where "n" represents any integer value) ("RVP features_current Use ''). RVP features_ref is individual RVP features Constitutes the first set of RVP features_current is individual RVP features This constitutes the second set. Similarly, regression model 180 uses the reference time window W. i The corresponding reference PAP feature ("PAP features_ref ) and the current time window W i+n The current PAP features corresponding to "PAP features_current Use ''). PAP features_ref Individual PAP features The first set consists of PAP features_currentIndividual PAP features It constitutes the second set.

[0111] Reference time window W i ΔCO LR This is the initial time interval calculated by regression model 180. Reference time window W i ΔCO LR To calculate this, a relatively later time window (for example, the current time window W) i+n ) is compared with the reference time window W. In some examples, the reference time window W i This can be the time interval corresponding to when patient 16 was first admitted to the ICU, OR, or other patient care environment. In other examples, the reference time window W i This can be a time interval corresponding to a first known cardiac output measurement, such as the iCO measurement of patient 16. In other examples, the reference time window W i RVP waveform and / or PAP waveform This can be any appropriate time interval from which data was obtained.

[0112] CO LR Module 42 is a reference time window W i It can be configured to update to a new time interval. For example, the reference time window W i This can be updated periodically. In some examples, the reference time window W i RVP features and PAP features It can be updated after a set time has elapsed that it is considered too old. In some examples, the reference time window W i RVP waveform and / or PAP waveform This can be continuously updated in each time interval in which it is measured. That is, in such an example, the reference time window W i This is the current time window W i+n This is the time interval immediately preceding it, so the reference time window W i The reference time window W(i+n)-1 It can also be specified as follows: In other examples, the reference time window W i This can be updated at any time manually by a healthcare professional 18 via the user interface 46, or automatically according to instructions in the flow rate and cardiac output software code 30. In yet another example, the reference time window W i It will not be updated.

[0113] As shown in Figure 19, the reference time window W i The interval is 10 seconds, and RVP waveform This represents the 10-second portion. Reference time window W i The length (in units of time) is CO AE This can be determined based on the minimum requirements for the proper functioning of module 40 or other modules represented by the flow rate and cardiac output software code 30. Thus, the reference time window W i Other time intervals are also possible, such as longer or shorter than 10 seconds. Also, the reference time window W i and the current time window W i+n The current time window W allows for direct comparison with the current time window. i+n The length is the reference time window W i It is the same length as the current time window W. i+n It can also be set to a 10-second interval, RVP waveform It can represent different 10-second segments.

[0114] Current time window W i+n ΔCO LR This is the current time interval calculated by regression model 180. Current time window W i+n The reference time window W i This can be any time interval after the current time window W. i+nThis can accommodate real-time or near-real-time measurements from patient 16, thus enabling the regression model 18 to generate real-time or near-real-time estimates of changes in patient 16's cardiac output.

[0115] RVP features_ref and PAP features_ref Each of them is RVP features_current or PAP features_current It corresponds to each individual one of them. That is, RVP features_ref and PAP features_ref These are the RVP measured from the corresponding time window. features_current and PAP features_current It includes the same individual features as those used by regression model 180. For example, RVP used by regression model 180. features One of them is RVP, as shown in Figure 7. waveform The diastolic end blood pressure can be derived from 108. In such an example, the regression model 180 uses a reference time window W. i RVP inside waveform Diastolic end-blood pressure measurement derived from (i.e., diastolic end-blood pressure) w_i ) and the current time window W i+n Inside RVP waveform Diastolic end-blood pressure measurement derived from (i.e., diastolic end-blood pressure) w_i+n ) and use. In the same example, end-diastolic blood pressure w_i RVP features_ref It is one of the end-diastolic blood pressure w_i+n RVP features_current It is one of them.

[0116] Regression model 180 is ΔCO LR The RVP of patient 16 is multiplied by a coefficient to estimate it. features PAP features It is a linear model configured to include variables based on demographic information and / or SiO2. More specifically, the regression model 180 has a reference time window W i From the current time window W i+n ΔCO LRTo determine the reference time window W, i RVP during features_ref and PAP features_ref The individual characteristics and the current time window W i+n RVP during features_current and PAP features_current Use the changes between the corresponding individual features.

[0117] The variables used in regression model 180 can take several different forms. The variables used in regression model 180 are RVP features_ref PAP features_ref RVP features_current PAP features_current This may include one or more measured or received values ​​of demographic information and / or SvO2. The variables used in regression model 180 are RVP features_current or PAP features_current One of them, and RVP features_ref or PAP features_ref It can also include a difference variable that represents the difference between one of the corresponding features. The difference variable can also be expressed as a ratio obtained by dividing the difference between the current feature and the reference feature by the reference feature, so that the normalized variable results in a difference variable. For example, one difference variable in regression model 180 is: (HeartRate w_i+n -HeartRate w_i ) / HeartRate w_i It can be expressed as follows.

[0118] The example difference variable shown above is the current time window W. i+n Heart rate measurement and reference time window W i The difference between the measured heart rate and the reference time window W i Includes the value obtained by dividing by the heart rate measurement. RVP features_current or PAP features_current One of them, and RVP features_ref or PAP features_refFor any pair with one of the corresponding ones, a similar difference variable can be generated. The variables used in regression model 180 are RVP features_ref PAP features_ref RVP features_current PAP features_current It may further include demographic information, and a combination variable representing one or more measured or received values ​​of SvO2, and / or one or more difference variables. For example, one combination variable in regression model 180 is:

[0119]

number

[0120] It can be expressed as follows.

[0121] The example combination of variables shown above are reference pulse pressure and current pulse pressure, and the current time window W. i+n Measurement values, as well as the current time window W i+n Includes a difference variable using the measured systolic time of RVP. features_ref PAP features_ref RVP features_current PAP features_current Similar combination variables can be generated for any combination of demographic information, one or more measured or received values ​​of SvO2, and / or one or more difference variables.

[0122] The terms in regression model 180 include, or are formed from, the variables described above multiplied by predetermined coefficients. For example, the coefficients can be determined during the training process for regression model 180. Regression model 180 can contain any number of terms from 1 to n. Regression model 180 is ΔCO LR This is calculated as the sum of all terms. The general form of regression model 180 is as follows: ΔCO LR=(a1×del_feat1)+(a2×del_feat2)+(a3×del_feat3)+...+(a n ×ser_feat n ) (Formula 1) Here,

[0123]

number

[0124] That is the case.

[0125] In equation 1 above, from the variable del_feat1 to del_feat 11 RVP features Based on the variable. In other examples, the variable in Equation 1 is PAP features It can be based on the variables of, or RVP features PAP features , and / or based on demographic information.

[0126] The ΔCO represented by equation 1 above LR The reference time window W i From the initial cardiac output corresponding to the current time window W i+n This value represents the percentage change in cardiac output up to a certain point. From the variable del_feat1 to del_feat 11 Since is a normalized variable, ΔCO LR This represents a percentage change, not the magnitude of the change. ΔCO LR For example, this form may be particularly useful for healthcare professionals 18 when, for a specific use, the percentage change is more relevant than the magnitude of the change, or when the magnitude of the change in cardiac output may vary greatly within a patient population, but the percentage change is more consistent. ΔCO LR The reference time window W i The iCO measurement or CO corresponding to AE CO from Module 40 AE ΔCO LRBy multiplying by this, it can be converted into the magnitude of the change.

[0127] Alternatively, regression model 180 is RVP features_ref or PAP features_ref By including unnormalized variables in ΔCO, LR It can be configured to calculate as the magnitude of the change. For example, one denormalized difference variable in regression model 180 is (EndSystolicPressure w_i+n -EndSystolicPressure w_i ) It can be expressed as follows, and one denormalized combinatorial variable in regression model 180 is, (HeartRate w_i+n -HeartRate w_i )×(HeartRate w_i+n ×SystoleTime w_i+n ) It can be expressed as follows.

[0128] In another example, regression model 180 is ΔCO LR To calculate this, we can use a variable that is a combination of normalized and denormalized. For example, one such combination is:

[0129]

number

[0130] It can be expressed as follows.

[0131] ΔCO LR This is the output of regression model 180. ΔCO LR For patient 16, the reference time window W i and the current time window W i+n This shows how much cardiac output changes between [the specified point] and [the specified point]. ΔCO LRThis is independent and important information regarding the hemodynamic status of patient 16. To determine whether patient 16's cardiac output is increasing or decreasing, healthcare professional 18 will use CO LR ΔCO from Module 42 LR This can be used. For example, healthcare worker 18 may be performing an intervention on patient 16, such as administering an inotropic agent, and may want to know how much the cardiac output has changed, but may not need to know the actual value of the cardiac output at that point in time.

[0132] In some examples, CO LR Module 42, as shown in Figure 6, outputs ΔCO2 from the regression model 180 to the output device 102. LR It can output the ΔCO of patient 16. For example, output device 102 outputs the ΔCO of patient 16. LR A graph showing the value over time can be displayed. In some examples, CO LR Module 42 monitors the hemodynamic status of patient 16 using ΔCO2. LR The output is continuously output to output device 102. In some examples, CO is output from regression model 180. LR The cardiac output estimator 182 in module 42 receives ΔCO LR It can be passed.

[0133] The cardiac output estimator 182 is CO LR This is the second submodule of module 42. The cardiac output estimator 182 calculates ΔCO2. LR CO AE The device receives CCO and iCO as inputs. The cardiac output estimator 182 calculates the estimated cardiac output of patient 16, which is CO. LR The output is... The cardiac output estimator 182 uses the reference time window W i By using the corresponding cardiac output value (i.e., "reference cardiac output"), ΔCO LR CO LR Calculate.

[0134] The cardiac output estimator 182 calculates ΔCO2 from the regression model 180. LRIt receives. As explained above, ΔCO LR The reference time window W i From the current time window W i+n This shows the change in cardiac output up to CO. AE CO from module 40 AE It also receives. Specifically, the cardiac output estimator 182 receives the reference time window W i CO corresponding to AE The value of is used. The cardiac output estimator 182 is also configured to receive CCO and iCO measurements. Specifically, the cardiac output estimator 182 uses the reference time window W i Use the corresponding CCO or ICO measurement. Reference time window W i CO corresponding to AE The values ​​of CCO and iCO are, respectively, reference cardiac output. One of the reference cardiac output values ​​is used to initialize or calibrate the cardiac output estimator 182 (i.e., CO LR (Used to provide a starting point for calculating)

[0135] The cardiac output estimator 182 has the following general format: CO LR =RefCO w_i +(ΔCO LR ×RefCO w_i ) (Formula 2) According to CO LR We calculate this. Here, RefCO w_i This refers to the CCO measurement, the iCO measurement, or the reference time window W. i CO corresponding to AE This is the value of (i.e., the reference cardiac output), ΔCO LR This is the output from regression model 180, ΔCO LR This is the value.

[0136] In equation 2 above, RefCO w_i CO AE The duration, or reference time window W iThe corresponding CCO or ICO measurements can be used, but not all of them are necessary. For example, the reference time window W i If the corresponding iCO measurement is available, LR Module 42 can be configured to use iCO measurements. In other examples, CO LR Module 42 uses CO instead of CCO or iCO measurements. AE It can be configured to use [this feature].

[0137] As shown in Equation 2, the cardiac output estimator 182 is CO LR To form the CO2, the magnitude of the change in cardiac output is added to the reference cardiac output. LR This is the output of the cardiac output estimator 182. LR This is the current time window W i+n This is the cardiac output value of patient 16 corresponding to CO. LR Module 42, as shown in Figure 6, transmits CO2 from the cardiac output estimator 182 to the output device 102. LR It can output CO2. For example, output device 102 can output CO2 from patient 16. LR A graph showing the value over time can be displayed. In some examples, CO LR Module 42 monitors the hemodynamic status of patient 16 using CO LR The output is continuously output to the output device 102. In some examples, CO is output from the cardiac output estimator 182. filtered Module 44 has CO LR It is possible to pass this, which will be explained in detail below.

[0138] CO LR Module 42 is RVP waveform Features (and optionally PAP) waveform Cardiac output is estimated from the characteristics of the autoencoder model 144 shown in Figures 12-15 and the CO shown in Figure 17. AEThis is a method for estimating cardiac output for patient 16, which differs from the method described above using module 40. This adds a level of redundancy within the hemodynamic monitoring system 10 for estimating cardiac output. LR Module 42 measures cardiac output (CO2). LR This enables the second calculation of CO in Figures 20A to 22. filtered As described in more detail below with reference to Module 44, CO is used to produce more accurate cardiac output estimates as a result. AE In addition, CO LR The CO described above can be used. AE Similar to module 40, another advantage here is CO LR Module 42 enables the determination of continuous cardiac output. LR CO is a continuous cardiac output estimate because it can be updated more frequently on a rolling basis, such as every 10 seconds or every 2 seconds. LR This is a more continuous estimate compared to iCO, which is measured only intermittently. LR Continuous cardiac output estimates, such as those mentioned above, can capture transient or rapid changes in the cardiac output of patients, which improves patient care.

[0139] Also, the CO2 explained above AE Similar to module 40, CO AE Module 40 is CO AE CO calculated using LR Therefore, relative RVP and RVP waveform It can also be based on the form or shape, and does not directly depend on the absolute pressure value, so it can be described as "data-independent" with respect to the absolute pressure value (e.g., the absolute value of RVP). LRThe time-based variables (e.g., heart rate, systolic time) and normalized pressure-based variables (e.g., systolic blood pressure, pulse pressure) used in Module 42 are not affected by errors or biases in absolute pressure values. This is important because, in practice, absolute pressure values ​​are not always accurate. For example, healthcare professionals may introduce some measurement error in RVP measurements when leveling a pressure transducer for a patient. Another example is raising or lowering a patient's bed for surgery or in the ICU, which can also alter absolute pressure values. LR Module 42 can be configured to not directly depend on absolute pressure values, so in these situations CO LR It can be estimated accurately. Furthermore, CO LR Module 42 may contain independent and useful information about patient 16 ΔCO LR It can also be calculated.

[0140] CO filtered Modules (Figures 20A-22) Figures 20A and 20B show CO filtered The following shows the variations of the Kalman filter algorithm in Module 44. The first variation of the Kalman filter algorithm is configured to output a filtered estimate of cardiac output, as shown in Figure 20A. The second variation of the Kalman filter algorithm is configured to output a filtered estimate of the change in cardiac output, rather than the cardiac output itself, as shown in Figure 20B.

[0141] Figure 20A shows CO2 containing the filter submodule 184A. filtered This is a schematic block diagram showing module 44. As shown in Figure 20A, CO filtered Module 44 includes a filter submodule 184A which contains a prediction block 186A and an update block 188A, and ΔCO AE Includes submodule 189A. AE CO LR , CCO, iCO, and ΔCOLR CO filtered To estimate CO filtered It is input to module 44. More specifically, CO AE ΔCO AE Input to submodule 189A, ΔCO AE Submodule 189A controls the autoencoder change in cardiac output ("ΔCO AE Outputs ) ). Linear regression change in cardiac output ΔCO LR and / or ΔCO AE This is input to the prediction block 186A of the filter submodule 184A. AE CO LR CCO and iCO are input to the update block 188A of the filter submodule 184A. Filtered cardiac output CO filtered This is output from filter submodule 184A.

[0142] The filter submodule 184A is CO filtered This is the first variation of the Kalman filter algorithm from Module 44. (CO filtered A second variation of the Kalman filter algorithm in module 44 is described below with reference to Figure 20B. Filter submodule 184A contains instructions within the code for implementing a Kalman filter algorithm that combines (or filters) one or more estimates of cardiac output into a single, more robust filtered estimate of cardiac output.

[0143] Generally, the Kalman filter algorithm can predict the current state of an input variable and then update that prediction using additional information, such as from the measurement input, to output a filtered estimate. The filter submodule 184A is configured to predict the current state of the patient's cardiac output. The filter submodule 184A also consumes the measurement input over time. As shown in Figure 20A, the measurement input that can be consumed by the filter submodule 184A is the CO2, which is the estimated cardiac output estimated by the flow rate and cardiac output software code 30. AE and CO LR This includes, for example, CCO and iCO, which are measurements obtained via catheter 54. Each measurement input may contain statistical noise and other inaccuracies, which can be represented as corresponding inaccuracies in the Kalman filter algorithm of filter submodule 184A. Filter submodule 184A is CO filtered Outputs.

[0144] ΔCO AE Submodule 189A is associated with the CO2 filter submodule 184A. filtered This is an additional submodule of module 44. ΔCO AE Submodule 189A is a ΔCO2 input for filter submodule 184A. AE Calculate ΔCO AE Submodule 189A is CO AE CO is the input from module 40. AE It receives ΔCO. AE Submodule 189A is CO AE It outputs CO AE This is consumed by the filter submodule 184A in prediction block 186A. For example, ΔCO AE Submodule 189A has CO2 at at least two time points. AE It can receive ΔCO. AE Submodule 189A is ΔCO AE CO AEThe slope between values ​​can be calculated. Therefore, ΔCO AE CO is calculated in different ways. LR ΔCO from Module 42 LR It may be different.

[0145] The Kalman filter algorithm of filter submodule 184A can be conceptualized as two distinct phases or steps, including a prediction phase and an update phase. Prediction block 186A represents the prediction phase of filter submodule 184A. As shown in Figure 20A, ΔCO LR and / or ΔCO AE This is input to prediction block 186A. The prediction phase of filter submodule 184A will be explained in more detail below with reference to Figure 21A. Update block 188A represents the update phase of filter submodule 184A. As shown in Figure 20A, CO AE CO LR The measurement input, including CCO and iCO, is then input to update block 188A. The update phase of filter submodule 184A will be described in more detail below with reference to Figure 21B.

[0146] In prediction block 186A (i.e., in the prediction phase), the filter submodule 184A uses the historical filtered estimate of cardiac output from the Kalman filter algorithm (e.g., CO, which will be described in more detail below). filtered (Past values ​​of) and ΔCO LR and / or ΔCO AEBased on (or another predictive model, as described in more detail below), the cardiac output for the current time step is predicted along with the corresponding predictive uncertainty, resulting in a predicted estimate of cardiac output. The predicted estimate of cardiac output generally does not directly include information from the current measurement input. However, the prediction phase of the Kalman filter algorithm in filter submodule 184A can be initialized using one of the measurement inputs for the first iteration of the algorithm. For example, the prediction phase can be initialized using an iCO measurement. After initialization, the prediction phase is performed using the previously filtered estimate, as described in more detail below. The Kalman filter algorithm can also be reinitialized based on predefined rules and triggers, such as the elapsed time since the last initialization or the elapsed time between the two most recent consecutive measurements.

[0147] In update block 188A (i.e., during the update phase), the filter submodule 184A receives one or more of the measurement inputs (e.g., CO AE CO LR The filter submodule 184A receives and consumes one or more of the CCO, iCO, and the corresponding uncertainty of the measurement (or estimate) for the current time step. When the filter submodule 184A receives the measurement input, the predicted estimate of cardiac output is updated using a weighted average of the prediction and estimate. Estimates with higher certainty and lower uncertainty are given greater weights. The weighted average results in a filtered estimate of cardiac output that falls between the predicted estimate of cardiac output and the measurement input, and has better estimated uncertainty than either one alone. This process is repeated at every time step (i.e., every iteration of the Kalman filter algorithm of the filter submodule 184A), and the filtered estimate of cardiac output, and its uncertainty from the previous time step, informs the prediction phase in the next iteration (i.e., the current time step).

[0148] Typically, the prediction and update phases alternate, with the predicted cardiac output estimate advancing at each time step until the next measurement input is received, after which the update phase incorporates the measurement input. That is, the Kalman filter algorithm of the filter submodule 184A is configured such that each iteration of the Kalman filter algorithm can include both a prediction and an update phase. However, if a measurement input is unavailable at a particular time step, the corresponding update phase can be skipped, allowing multiple prediction phases to run consecutively. Similarly, if multiple measurement inputs are received at the same time step (e.g., CO AE CO LR To further refine the filtered estimate of cardiac output (cardiac output estimates and / or measurements from multiple sources, such as two or more of the CCO and iCO), multiple update phases can be performed sequentially. The filter submodule 184A can include update phases for any one source or any combination of sources of measurement inputs. In other words, the filter submodule 184A is configured such that the total number of update phases per iteration of the Kalman filter algorithm (which may be zero in some cases) depends on which of the measurement inputs are available during that iteration. As will be further described below with reference to Figure 21C, the filter submodule 184A produces a filtered estimate of cardiac output that is generally more accurate than any single measurement input.

[0149] The Kalman filter algorithm of the filter sub-module 184A is recursive. That is, the filter sub-module 184A uses only the filtered estimate of the cardiac output from the previous time step, the uncertainty associated with the previous filtered estimate of the cardiac output, and the current measurement input to calculate the filtered estimate of the cardiac output for the current time step, and no other past information is required. For each time step, the previous filtered estimate of the cardiac output is fed back to the prediction block 186A to provide information for the next prediction of the filter sub-module 184A. Thus, information regarding the previous measurement inputs and their respective uncertainties is still carried over to the next prediction because the previous filtered estimate of the cardiac output is refined based on the measurement input from that time step.

[0150] The final filtered estimate of the cardiac output generated by the filter sub-module 184A at each time step is the value of CO filtered . CO filtered is the filtered representation of the right ventricular cardiac output ("RVCO"). CO filtered is the output of the filter sub-module 184A and is thus one possible output of the CO filtered module 44. In some examples, the CO filtered module 44 can output CO filtered from the filter sub-module 184A to the output device 102, as shown in FIG. 6. For example, the output device 102 can display a graph showing the values of the patient 16's CO filtered over time. In some examples, the CO filtered module 44 continuously outputs CO filtered to the output device 102 to monitor the hemodynamic state of the patient 16.

[0151] FIG. 20B is a schematic block diagram showing the CO filtered module 44 including the filter sub-module 184B. As shown in FIG. 20B, the COfiltered Module 44 includes a filter submodule 184B which contains a prediction block 186B and an update block 188B, and ΔCO AE Includes submodule 189B. AE and ΔCO LR ΔCO filtered To estimate CO filtered It is input to module 44. More specifically, CO AE ΔCO AE Input to submodule 189B, ΔCO AE Submodule 189B is ΔCO AE Outputs ΔCO. LR and ΔCO AE This is input to the update block 188B of the filter submodule 184B. The change in filtered cardiac output ΔCO filtered This is output from filter submodule 184B.

[0152] The filter submodule 184B is CO filtered This is a second variation of the Kalman filter algorithm in module 44. Filter submodule 184B includes instructions within the code to implement a Kalman filter algorithm that combines (or filters) one or more estimates of the change in cardiac output into a single, more robust filtered estimate of the change in cardiac output.

[0153] Filter submodule 184B is functionally similar to filter submodule 184A described above with reference to Figure 20A, except that filter submodule 184B is configured to filter multiple changes in cardiac output values ​​instead of multiple cardiac output values. For brevity, the features and functions of filter submodule 184B, which are largely the same as filter submodule 184A (shown in Figure 20A), will not be fully repeated in this section except to the extent necessary to explain the differences.

[0154] The filter submodule 184B is configured to predict the current state of changes in patient 16's cardiac output. The filter submodule 184B also consumes measurement inputs over time. As shown in Figure 20B, the measurement inputs that can be consumed by the filter submodule 184B are the respective estimates of changes in cardiac output, ΔCO2, estimated by the flow rate and cardiac output software code 30. AE and ΔCO LR This includes. Each measurement input may contain statistical noise and other inaccuracies, which can be represented as corresponding inaccuracies in the Kalman filter algorithm of filter submodule 184B. Filter submodule 184B is ΔCO filtered Outputs.

[0155] ΔCO AE Submodule 189B is associated with the CO2 filter submodule 184B. filtered This is an additional submodule of module 44. Refer to Figure 20A for the ΔCO described above. AE Similar to submodule 189A, ΔCO AE Submodule 189B is a ΔCO2 input for filter submodule 184B. AE Calculate ΔCO AE Submodule 189B is CO AE CO is the input from module 40. AE It receives ΔCO. AE Submodule 189B is ΔCO AE Outputs ΔCO AE This is consumed by the filter submodule 184B in update block 188B. For example, ΔCO AE Submodule 189B is CO at at least two points in time. AE The value can be received. Next, ΔCO AE Submodule 189B is ΔCO AE CO AE The slope between values ​​can be calculated. Therefore, ΔCO AE CO is calculated in different ways. LRΔCO from Module 42 LR It may be different.

[0156] Prediction block 186B represents the prediction phase of filter submodule 184B. In prediction block 186B (i.e., in the prediction phase), filter submodule 184B obtains historical filtered estimates of changes in cardiac output from the Kalman filter algorithm (e.g., ΔCO2, as will be explained in more detail below). filtered Based on past values ​​of ΔCO2, the change in cardiac output for the current time step is predicted along with the corresponding predictive uncertainty, resulting in a predicted estimate of the change in cardiac output. The predicted estimate of the change in cardiac output generally does not directly include information from the current measurement input. However, the prediction phase of the Kalman filter algorithm in filter submodule 184B can be initialized using one of the measurement inputs for the first iteration of the algorithm. For example, the prediction phase can be initialized using ΔCO2. AE or ΔCO LR It can be initialized using any of the following values. After initialization, the prediction phase is performed using the previously filtered estimates, as described in more detail below. The Kalman filter algorithm can also be reinitialized based on predefined rules and triggers, such as the time elapsed since the last initialization or the time elapsed between the last two consecutive measurements.

[0157] Update block 188B represents the update phase of filter submodule 184B. As shown in Figure 20B, ΔCO AE and ΔCO LR The measurement input, including ΔCO, is input to update block 188B. In update block 188B (i.e., in the update phase), the filter submodule 184B receives one or more of the measurement inputs (e.g., ΔCO). AE and ΔCO LRThe filter submodule 184B receives and consumes one or more of the measurement inputs and the corresponding uncertainty of the measurement (or estimate) for the current time step. When the filter submodule 184B receives a measurement input, the predicted estimate of the change in cardiac output is updated using a weighted average of the prediction and the estimate. Estimates with higher certainty and lower uncertainty are given greater weights. The weighted average results in a filtered estimate of the change in cardiac output that is positioned between the predicted estimate of the change in cardiac output and the measurement input, and has better estimated uncertainty than either one alone. This process is repeated for every time step (i.e., every iteration of the Kalman filter algorithm of the filter submodule 184B), and the filtered estimate of the change in cardiac output, and its uncertainty from the previous time step, informs the prediction phase in the next iteration (i.e., the current time step). Similar to the filter submodule 184A described above with reference to Figure 20A, the prediction and update phases of the filter submodule 184B can be performed alternately, or, in other examples, the update phase can be skipped or repeated depending on the availability of the measurement input. The filter submodule 184B generates filtered estimates of changes in cardiac output that are generally more accurate than any one of the measurement inputs.

[0158] The Kalman filter algorithm of the filter submodule 184B is recursive. That is, the filter submodule 184B uses only the filtered estimate of the change in cardiac output from the previous time step, the uncertainty associated with the previous filtered estimate of the change in cardiac output, and the current measurement input to compute a filtered estimate of the change in cardiac output for the current time step. For each time step, the previous filtered estimate of the change in cardiac output is fed back to the prediction block 186B to inform the next prediction of the filter submodule 184B. In this way, information about the previous measurement inputs and their respective uncertainties is still carried over to the next prediction, since the previous filtered estimate of the change in cardiac output is refined based on the measurement input from that time step.

[0159] The final filtered estimate of the change in cardiac output generated by the filter submodule 184B at each time step is ΔCO filtered This is the value of the filtered change in cardiac output ΔCO₂. filtered This is the output of filter submodule 184B, and therefore, CO filtered Another possible output of module 44. In some examples, CO filtered Module 44, as shown in Figure 6, outputs ΔCO2 from the filter submodule 184B to the output device 102. filtered It can output the following. For example, the output device 102 can output the ΔCO2 of patient 16 over time. filtered A graph showing the value of CO can be displayed. In some examples, CO filtered Module 44 monitors the hemodynamic status of patient 16 using ΔCO2. filtered This is continuously output to output device 102.

[0160] Although shown as separate examples in Figures 20A and 20B, CO filteredModule 44 can be configured to include either or both of filter submodule 184A and filter submodule 184B. In one example, CO filtered Module 44 contains only the filter submodule 184A and is therefore configured to filter cardiac output values. In another example, CO filtered Module 44 contains only the filter submodule 184B and is therefore configured to filter changes in cardiac output values. In yet another example, CO filtered Module 44 includes filter submodule 184A and filter submodule 184B, and is therefore configured to filter both cardiac output values ​​and changes in cardiac output.

[0161] Referring together to Figures 20A and 20B, the Kalman filter algorithm of filter submodules 184A and 184B becomes a powerful estimator when there are multiple noisy estimates and / or measurements for a variable such as cardiac output or changes in cardiac output. Due to the characteristics of the Kalman filter algorithm, including the prediction and update phases, filter submodules 184A and 184B effectively handle the uncertainty caused by noisy measurement inputs, resulting in a more accurate CO2. filtered and ΔCO filtered Furthermore, filter submodules 184A and 184B can generate more accurate CO2 values ​​for estimated and / or measured values ​​of cardiac output and changes in cardiac output received from various sources. filtered and ΔCO filtered To integrate them, we can work with these estimates and / or measurements.

[0162] Filter sub-modules 184A and 184B can also be allowed to receive estimated and / or measured values of asynchronous or sporadic cardiac output or changes in cardiac output from various sources. That is, filter sub-modules 184A and 184B do not require measurement inputs from all possible sources at each time step and can even be allowed not to receive any measurement inputs for a particular time step. For example, filter sub-module 184A may have received measurement inputs from both CO AE module 40 and CO LR module 42. Then, at a certain time step, the flow module 38 may determine that the autoencoder model 144 (shown in FIG. 12) is no longer generating good values. In that case, CO AE is not calculated for that time step and is not supplied to filter sub-module 184A. In such an example, the Kalman filter algorithm of filter sub-module 184A can automatically adapt to this change and continue to generate CO LR using only the measurement inputs from CO filtered module 42. The Kalman filter algorithm can also automatically re-adapt to use both the measurement inputs from CO AE module 40 and the measurement inputs from CO LR module 42 at a future time point. Similarly, when additional sources of estimated and / or measured values of cardiac output or changes in cardiac output (e.g., measurements from additional devices such as echocardiograms or different devices, or estimates from other models) are added to the hemodynamic monitoring system 10, these can be easily integrated into the Kalman filter algorithm as an additional update phase.

[0163] Using the Kalman filter algorithm in filter submodules 184A and 184B is highly computationally efficient because the Kalman filter algorithm is recursive. Similarly, the recursive nature of the Kalman filter algorithm means that it does not integrate information in isolation. Instead, the Kalman filter algorithm allows filter submodules 184A and 184B to intelligently combine information using the context of what the filtered estimate of cardiac output or change in cardiac output was in previous time steps. Thus, CO filtered and ΔCO filtered This can be used to obtain a more accurate estimate of the cardiac output of patient 16. Furthermore, the CO2 described above AE Module 40 and CO LR Similar to module 42, another advantage here is the CO2 filter submodule 184A. filtered Module 44 enables the determination of continuous cardiac output. filtered CO is a continuous cardiac output estimate because it can be updated more frequently on a rolling basis, such as every 10 seconds or every 2 seconds. filtered This is a more continuous estimate compared to iCO, which is measured only intermittently. filtered Continuous cardiac output estimates, such as those mentioned above, can capture transient or rapid changes in the cardiac output of patients, which improves patient care.

[0164] Also, the CO2 explained above AE Module 40 and CO LR Similar to module 42, CO filtered Module 44 is CO AE CO based on filtered Therefore, relative RVP and RVP waveform It can also be based on the form or shape, CO LR CO based on filteredSimilarly, it may not be affected by absolute pressure values, so it can be described as "data-independent" with respect to absolute pressure values ​​(e.g., absolute value of RVP). This is important because, in practice, absolute pressure values ​​are not always accurate. For example, healthcare workers may introduce some measurement error in RVP measurements when leveling a pressure transducer for a patient, and as another example, raising and lowering a patient's bed for surgery or in the ICU can also change absolute pressure values. CO2 using filter submodule 184A filtered Module 44 can be configured to not directly depend on absolute pressure values, so in these situations CO filtered It can be generated accurately.

[0165] Figure 21A shows CO filtered This graph shows the predicted value p for module 44. Figure 21B shows CO over time. filtered This is a graph showing the measured value m of module 44. Figure 21C shows CO filtered This is a graph showing the filtered value f over time from module 44. Figures 21A to 21C will be discussed together. In summary, Figures 21A to 21C are graphs corresponding to the function of filter submodule 184A (shown in Figure 20A), and each of Figures 21A to 21C shows the cardiac output value over time. However, it should be understood that the concepts explained for Figures 21A to 21C are generally applicable to the function of filter submodule 184B (shown in Figure 20B), and the graphs in Figures 21A to 21C can instead represent the change in cardiac output value over time.

[0166] Each of the graphs in Figures 21A to 21C shows the same arbitrary time steps n, n-1, n-2, n-3, and n-4 along the x-axis. Each of the time steps n, n-1, n-2, n-3, and n-4 corresponds to each iteration of the Kalman filter algorithm of the filter submodule 184A. That is, time step n represents the current iteration of the filter submodule 184A. Time steps n-1, n-2, n-3, and n-4 represent the time steps preceding time step n. For example, time step n-1 represents the time step immediately preceding time step n, time step n-2 represents the time step immediately preceding time step n-1 (and two iterations prior to time step n), and so on. As will be explained in more detail below, each of the graphs in Figures 21A to 21C also shows the cardiac output values ​​corresponding to time steps n, n-1, n-2, n-3, and n-4. The cardiac output values ​​graphed in Figures 21A to 21C are shown in liters per minute (L / min), but in other examples, they can be expressed in other velocity units.

[0167] Figure 21A is a graph corresponding to the function of prediction block 186A shown in Figure 20A. In Figure 21A, cardiac output values ​​are shown for each of the time steps n, n-1, n-2, n-3, and n-4. The predicted value p is the cardiac output value corresponding to the current time step n. The predicted value p is an exemplary value of the predicted estimate of cardiac output predicted by prediction block 186A. The cardiac output values ​​corresponding to time steps n, n-1, n-2, n-3, and n-4 are used by prediction block 186A in the current time step n (or used in previous time steps) to generate the predicted value p corresponding to the current time step n. filteredThis can be a previous value. If any of the time steps n, n-1, n-2, n-3, and n-4 represent the first time step in which the predicted value p was calculated, then the cardiac output value in a previous time step relative to the first time step can instead be the measurement input value used for initialization. The predicted value p at the current time step n is shown connected by a dashed line to the previous value in Figure 21A, as the predicted value p is a predicted estimate of cardiac output used internally within the filter submodule 184A. Unlike the other cardiac output values ​​shown in Figure 21A for previous time steps, the predicted value p is generally not the final filtered output of the filter submodule 184A.

[0168] As shown in Figure 21A, the predicted value p is equal to the corresponding uncertainty e p Saturates. Uncertainty e p This represents the uncertainty associated with the predicted value p as an estimate of the standard deviation. The uncertainty in the predicted value p can be configurable or predetermined for each application of the filter submodule 184A and may depend on the type of forecast used. The arrow extending from the predicted value p is proportional to the magnitude of the uncertainty in the predicted value p. In one non-restrictive example, the predicted value p is associated with an uncertainty e of ±0.4 L / min. p Since it may have such a possibility, if the predicted value p is 4.0 L / min, uncertainty e p The range indicated by the arrow is 3.6 to 4.4 L / min, indicating that there is approximately a 68% probability that the true value falls within this range.

[0169] The prediction phase of the Kalman filter algorithm in filter submodule 184A can be configured as one of several different prediction models for generating the predicted value p. In one example, the prediction phase can use extrapolation to compute the predicted value p. For example, the prediction phase can use linear extrapolation based on cardiac output values ​​at time steps n-1 and n-2 to determine the value of the predicted value p at the current time step n. Using linear extrapolation in the prediction phase means that the change in cardiac output (i.e., the slope) between the previous two time steps is carried over as the same up to the current time step n. That is, the slope between cardiac output values ​​between time steps n-2 and n-1 is considered to be the same between time steps n-1 and n. In other examples, the prediction phase can use extrapolation techniques by incorporating additional past cardiac output values ​​(e.g., time steps n-3, n-4, etc.) using other types of fitting functions, such as cubic fit. For example, prediction block 186A uses CO to determine the predicted value p at the current time step n. filtered We can take the last four values ​​and fit a cubic function to those points. In yet another example, such as an example where the system is stable, prediction block 186A is CO at time step n-1. filtered Using the value, the predicted value p at the current time n is calculated as the CO of time step n-1. filtered It can be extrapolated from the value of as being unchanged (i.e., the same).

[0170] Alternatively, the prediction phase calculates the predicted value p using ΔCO LR and / or ΔCO AE This can be used. For example, prediction block 186A is CO at time step n-1. filtered Starting with the value of , to determine the predicted value p at the current time step n, ΔCO of the current time step n LR or ΔCO AEThe values ​​can be added. In another example, prediction block 186A determines the predicted value p at the current time step n, and CO at time step n-1. filtered The value of ΔCO for the current time step n is used. LR and ΔCO AE The average (for example, ΔCO LR and ΔCO AE For each of these, a weighted average based on a specified or predetermined weight such as 0.5 and 0.5 can be added. In yet another example, prediction block 186A predicts CO at time step n-1. filtered The value of ΔCO for the current time step n is LR The first predicted value is determined by adding the values ​​of and the CO at time step n-1. filtered The value of ΔCO for the current time step n is AE A second predicted value can be determined by adding the values ​​of , and then the predicted value p can be determined as the average (e.g., weighted average) of the first and second predicted values. Further details and examples of weighting the input are described below with reference to Figure 21B, which is also applicable. In one non-restrictive example, ΔCO LR and ΔCO AE The specified or predetermined weights associated with ΔCO can be based on the relative accuracy (or uncertainty) associated with the two models (i.e., regression model 180 shown in Figures 18-19 and autoencoder model 144 shown in Figures 12-15). In yet another example, ΔCO LR and ΔCO AE The weights associated with the CO can change over time depending on the relative signal quality metrics associated with such methods. Compared to extrapolation methods, these predictive models use the CO of the immediately preceding time step (e.g., time step n-1) rather than one or more previous time steps. filtered Only the value of is needed. Instead of extrapolation, ΔCO LR and / or ΔCO AEThe trade-off of using this is that the quality of predictions depends to some extent on the capabilities of the regression model 180 and the autoencoder model 144. For example, if the reference features used in the regression model 180 become outdated, ΔCO LR Because quality may be reduced, RVP features_ref and PAP features_ref If it has been updated more recently, ΔCO LR Predictions using this method may be improved. Prediction block 186B does not receive any changes in cardiac output values ​​as input, so ΔCO LR and / or ΔCO AE It should be noted that the prediction model using this method is generally used only in prediction block 186A of filter submodule 184A (shown in Figure 20A) and not in prediction block 186B of filter submodule 184B (shown in Figure 20B). That is, prediction block 186B makes predictions using extrapolation, while prediction block 186A either uses extrapolation or ΔCO LR and / or ΔCO AE Predictions can be made using this method.

[0171] Each of the predictive models described above makes a prediction about the current value of patient 16's cardiac output by extrapolation, or by ΔCO2. LR The rate of change in cardiac output is used. Furthermore, each predictive model is based on the fundamental assumption that the change in cardiac output within a measurement time window (e.g., 10-second intervals) can be represented by a single aggregate or average value.

[0172] Figure 21B is a graph corresponding to the function of update block 188A shown in Figure 20A. In Figure 21B, cardiac output values ​​are shown for each of the time steps n, n-1, n-2, n-3, and n-4. Measured value m is the cardiac output value corresponding to the current time step n. Measured value m is an exemplary value of the measurement input received by the filter submodule 184A in update block 188A. As described above, the measurement input is CO AE CO LRThis can be any one (or more, but for simplification only, one is shown in Figure 21B) of CCO, or iCO. The measured value m is used by update block 188A to update the predicted value p for the current time step n. The cardiac output values ​​corresponding to time steps n, n-1, n-2, n-3, and n-4 are the previous measured input values ​​used by update block 188A in the previous time step to update the previous predicted estimate of cardiac output for the respective time step.

[0173] Figure 21B shows CO AE Module 40, CO LR This shows cardiac output values ​​from a single source, such as Module 42, CCO, or iCO. That is, the measured value m is CO AE CO LR It is one of CCO or iCO. In this example, the Kalman filter algorithm of filter submodule 184A has only one update phase using the measured value m. In other examples where cardiac output values ​​are received from multiple sources, there are multiple measured values ​​m in the current time step n.

[0174] As shown in Figure 21B, the measured value m is equal to the corresponding uncertainty e m Saturates. Uncertainty e m This represents the uncertainty in the measured value m as an estimate of the standard deviation. The uncertainty in the measured value m can be configurable or predetermined for each application of the filter submodule 184A and may depend on the type of measurement input used. The arrow extending from the measured value m is proportional to the magnitude of the uncertainty in the measured value m. In one non-restrictive example, CO AE CO from Module 40 AE The associated uncertainty e is 0.5 L / min. m (CO AE If the measured value m is 5.0 L / min, uncertainty e mThe range indicated by the arrow is 4.5 to 5.5 L / min, indicating that there is approximately a 68% probability that the true value falls within this range. Uncertainty e m The uncertainty e depends on how each uncertainty is determined. p It may be the same, or it may be different.

[0175] In the update phase of the Kalman filter algorithm in filter submodule 184A, the predicted value p and the measured value m are dynamically or adaptively weighted to generate the filtered value f, as shown in Figure 21C. This weighting is based on uncertainty e p or uncertainty e m This can result in uncertainty related to the filtered value f at the current time step n, which is predicted to be equal to or less than any of the following.

[0176] When the Kalman filter algorithm of filter submodule 184A is initialized, uncertainty e p and uncertainty e m Multiple sources of measurement input (e.g., CO) are specified. AE Module 40, CO LR If Module 42, CCO, and / or iCO) exist, each source has different associated uncertainties e m It may have the uncertainty e of the predicted value p. p This includes the uncertainty associated with the prediction model and the CO used to determine the predicted value p. filtered This includes any uncertainty related to past values ​​of . Thus, after the filter submodule 184A determines the predicted value p for the current time step n and receives the measured value m, the filter submodule 184A determines the corresponding uncertainty e p and uncertainty e m It also contains information regarding the following: The predicted value p and the measured value m are, respectively, associated with their respective uncertainties e. p and e mThe weighting is based on the following: The one with greater uncertainty between the predicted value p and the measured value m is given a smaller weight, and the one with less uncertainty between the predicted value p and the measured value m is given a larger weight. Using the above example to illustrate, uncertainty e p The flow rate is 0.4 L / min, and the uncertainty e m If the flow rate is 0.5 L / min, the predicted value p is given a greater weight, and the measured value m is given a smaller weight.

[0177] Uncertainty e p and uncertainty e m Each of these can be fixed or modulated over time. For example, uncertainty e m As explained above with reference to Figure 9, SQI RVP Or SQI PAP It can be scaled based on SQI. RVP If the value is between zero and 1 (indicating signal quality within a specified range, 0 to 1), then the flow rate and cardiac output software code 30 (shown in Figure 1) corresponds to the RVP waveform Using data CO AE and CO LR We can proceed to estimate the following. However, COAE and COLR associated with a signal quality index of zero and COAE associated with a signal quality index of 1 AE and CO LR and (SQI RVP =0 and SQI RVP Regarding =1), both are valid, but different uncertainties may exist. For example, SQI RVP If = 0, the uncertainty in the measurement may be low, so the uncertainty e related to the measurement m is m You can multiply it by 1 (or leave it unchanged). In the same example, SQI RVP When = 1, the uncertainty in the measurement may be high, so in order to increase the uncertainty, the uncertainty e related to the measurement m is m This can be multiplied by some coefficient, such as 2 or greater. In other examples, uncertainty ep and / or uncertainty e m SQI RVP and SQI PAP It can be scaled based on coefficients other than those mentioned above.

[0178] Figure 21C is a graph corresponding to the function of the filter submodule 184A, which uses information from the prediction block 186A and update block 188A, as shown in Figure 20A and explained above with reference to Figures 21A and 21B. In Figure 21C, cardiac output values ​​are shown for each of the time steps n, n-1, n-2, n-3, and n-4. The filtered value f is the cardiac output value corresponding to the current time step n. The filtered value f is the CO output output by the filter submodule 184A for the current time step n. filtered This is an exemplary value. The filtered value f is also used by prediction block 186A in the prediction for the next time step. The cardiac output values ​​corresponding to time steps n, n-1, n-2, n-3, and n-4 are the same as the previous CO values ​​output by filter submodule 184A in the previous time step. filtered It is a value.

[0179] The filtered value f, and its associated uncertainty e f This involves the expected value p, the measured value m, and the associated uncertainty e. p and e m Using and , it is calculated according to the following general format.

[0180]

number

[0181] Here, e f ≤e m , e f ≤e p , and 1 / e f 2 =(1 / ep 2 )+(1 / e m 2 ) (Equation 4) Furthermore, Equation 3 can be simplified and rewritten to clearly show the update of the predicted value p using the measured value m (i.e., the measured input in the Kalman filter algorithm).

[0182]

number

[0183] Equation 5 is, f=p+(K×(mp)) (Equation 6) It can also be expressed as follows: Here, K is

[0184]

number

[0185] And, e f ≤e m , and e f ≤e p This means that two different sources of measurement input (e.g., CO) are used simultaneously. LR and CO AE To extend Equation 3 to show two simultaneous measurements m from ), we use

[0186]

number

[0187] It can be rewritten as follows: Here, m_1 is CO AE CO LR The first measurement from the first source of the measurement input, such as CCO or iCO, m_2 is CO AE CO LRA second measurement from a second source of measurement input, such as CCO or iCO, e f ≤e m_1 , e f ≤e m_2 , e f ≤e p , and 1 / e f 2 =(1 / e p 2 )+(1 / e m_1 2 ) + (1 / e m_2 2 ) (Equation 8) That is the case.

[0188] As shown in Equation 3, the predicted value p is weighted by the respective contributions of the measured value m to the overall noise, and the measured value m is weighted by the respective contributions of the predicted value p to the overall noise. In other words, if either the predicted value p or the measured value m has greater uncertainty, the other value is given greater weight. To explain this, the predicted value p has uncertainty e p If = 0 (no uncertainty), then equation 3 is that the predicted value p is 1 (e m / e m The measured value m is weighted by (0 / e) and is zero (0 / e m It is simplified to be weighted by ). Generally, the filtered value f is different from both the predicted value p and the measured value m. However, if the measured value m is the same as the predicted value p, then the filtered value f will also be the same, regardless of the weighting in Equation 3.

[0189] Equation 7 shows how the filtered value f is calculated when there are two simultaneous measurements m, such as from two different sources of measurement input. Equation 7 can be further extended to include any number of measurements m (e.g., m_1, m_2, ... m_n). For example, the filtered value f is CO AE CO LRThis can be calculated using the measured values ​​m from each of the available sources of measurement input in the system, including CCO and iCO.

[0190] As shown in Figure 21C, the filtered value f is the corresponding uncertainty e f Saturates. Uncertainty e f This represents the uncertainty in the filtered value f. The arrow extending from the filtered value f is proportional to the magnitude of the uncertainty in the filtered value f. In the example containing a single measurement m, the uncertainty e f This can be determined using Equation 4 above. In an example involving multiple measured values ​​m, the uncertainty e f This can be determined using equation 8 above. As explained above, uncertainty e f Ideally, each uncertainty e p and uncertainty e m It is smaller than both of them. Therefore, uncertainty e f The arrows indicating uncertainty e p and uncertainty e m Shorter than the arrow indicating the difference. Uncertainty e f is uncertainty e p and uncertainty e m Since it is smaller than the predicted value p or the measured value m alone, the filtered value f is more accurate. Therefore, CO filtered Module 44 is the CO for the current time step n. AE or CO LR CO may be more accurate than either one alone. filtered It can output.

[0191] Figure 22 shows the CO2 levels for patient 16 in the pre-bypass state, post-bypass state, and in the ICU state. AE Module 40, CO LR Module 42, and CO filtered This graph compares exemplary cardiac output values ​​from module 44 with exemplary iCO values ​​over time. AE Module 40, CO LR Module 42, COfiltered Module 44 and iCO are different sources of cardiac output within the hemodynamic monitoring system 10 shown in Figure 1. In Figure 22, CO LR CO from module 42 LR Value (shown as a white square), CO AE CO from Module 40 AE Value (shown as a black circle), CO filtered CO from Module 44 filtered The values ​​(shown as a solid line) and the iCO values ​​(shown as black asterisks) are shown over time.

[0192] As shown in Figure 22, the exemplary cardiac output graph is divided into three parts, including a first part 190, a second part 192, and a third part 194. The first, second, and third parts 190, 192, and 194 are not connected to each other, as they are separated by arbitrary time quantities represented by the space along the x-axis between the first part 190 and the second part 192, and between the second part 192 and the third part 194. The first part 190 corresponds to the pre-bypass state of patient 16. The first part 190 includes exemplary cardiac output values ​​measured or acquired for patient 16 before bypass surgery. The second part 192 corresponds to the post-bypass state of patient 16. The second part 192 includes exemplary cardiac output values ​​measured or acquired for patient 16 after bypass surgery. The third part 194 corresponds to the ICU state of patient 16. The third part 194 includes exemplary cardiac output values ​​measured or obtained for patient 16 during the period that patient 16 was hospitalized in the ICU.

[0193] Figure 22 shows CO filtered CO from Module 44 filtered The value is CO LR CO from module 42 LR Value, CO AE CO from Module 40 AE This indicates that the value is more continuous and less noisy than the iCO value. LR CO AEThe values ​​of iCO, at each point in time, may be synchronous, asynchronous, tend to be in the same direction, and / or tend to be in different directions.

[0194] For example, in one region slightly before the middle along the second part 192 (viewed from left to right), CO LR and CO AE The values ​​tend to diverge or move in the opposite direction. Across the remainder of the second part 192 (and also much of the first part 190 and the third part 194), CO LR and CO AE While they tend to move in roughly the same direction, this is not always the case. As shown in Figure 22, CO filtered The value of CO is in the central region along the second part 192. LR and CO AE It falls between the values ​​that deviate from it. This is CO filtered Module 44 intelligently weights the estimated cardiac output, CO filtered This is because the Kalman filter algorithm is used to integrate it. Therefore, CO filtered The values ​​are less noisy than the other cardiac output values ​​shown in Figure 22.

[0195] As another example, CO LR CO AE There are several points in time when the iCO value is not received. LR CO AE The time points on the graph in Figure 22 where one or more of the iCOs are missing represent asynchronous cardiac output estimation activity from various sources of cardiac output in the hemodynamic monitoring system 10. However, CO filtered At these points in time, CO filtered It is still available from module 44. filtered Module 44 is generally always CO filtered Since it can generate values, CO filtered The graph is more continuous than the graph for the other cardiac output sources shown in Figure 22.

[0196] Exemplary implementation forms and alternative examples (Figure 23) Referring to Figures 1 to 22, many different configurations of the hemodynamic monitoring system 10 are possible using the combinations of components and modules described above. That is, the hemodynamic monitoring system 10 by the technique described herein can be implemented in many different ways, including using various combinations of hemodynamic sensors 14 to acquire hemodynamic data about patient 16 (shown in Figure 1). Therefore, some of the inputs and outputs shown in Figure 6 are arbitrary. Depending on the hemodynamic data available from patient 16, or depending on the desired output, the RVP feature module 32, PAP feature module 34, validation module 36, flow module 38, CO2 can be used to estimate the blood flow rate, cardiac output, and / or changes in cardiac output of patient 16, as shown in Figure 1. AE Module 40, CO LR Module 42, and CO filtered Various combinations of module 44 can be implemented together.

[0197] For example, PAP waveform One or more of the following may be unavailable from patient 16, or may be ignored for other reasons. PAP waveform In cases where it is not available or not used, the PAP feature module 34 is also not used, and PAP features This is verification module 36 or CO LR It is not passed to module 42. Instead, only the RVP feature module 32 is used, and the validation module 36 and CO LR Only module 42 is RVP features Received. In cases where CCO and iCO are unavailable or not used, the measurements of CCO and iCO are CO LR Module 42 or CO filtered It is not passed to module 44. Instead, CO LR Module 42 is CO AE CO from Module 40 AEOnly use CO filtered Module 44 is CO AE CO LR , and ΔCO LR Use some combination of these.

[0198] Furthermore, the estimated cardiac output of patient 16 was CO AE Module 40, CO LR Module 42, and CO filtered Output can be taken from any one or more of the modules 44. Similarly, the estimated flow processed CO AE To estimate CO AE In addition to being passed to module 40, or instead, it can be output from flow module 38. Similarly, the estimated change in cardiac output is ΔCO filtered To estimate CO filtered In addition to being passed to module 44, or instead, CO LR It can be output from module 42.

[0199] Figure 23 is a schematic block diagram showing the inputs and outputs for each module of an exemplary implementation of the hemodynamic monitoring system 10. Figure 23 shows the RVP feature module 202, the validation module 204, the flow module 206, and CO AE Module 208 and CO LR Module 210 and CO filtered Module 212 and output device 214 are shown. RVP feature module 202, verification module 204, flow module 206, CO AE Module 208, CO LR Module 210, CO filtered Module 212 and output device 214 are generally similar to the components or modules with the same names described above with reference to Figures 1 to 22, and one possible configuration of these components and modules is described here.

[0200] As shown in Figure 23, the RVP feature module 202 is RVPwaveform It receives as input. The RVP feature module 202 is RVP waveform From RVP features Extract the following. RVP feature module 202 is RVP features The output is sent to verification module 204. RVP feature module 202 outputs RVP features to CO LR It also outputs to module 210.

[0201] Verification module 204 is RVP waveform and RVP features It receives and as input. RVP features This is supplied from the RVP feature module 202 to the validation module 204. The validation module 204 checks whether RVP_valid is true or negative (i.e., RVP waveform An instruction indicating whether RVP_valid is valid or not is output to the flow module 206. The instruction RVP_valid is true (for example, RVP_valid=true) waveform This means that the corresponding hemodynamic data is valid for further processing by the flow module 206. The instruction that RVP_valid is false (e.g., RVP_valid=false) means that RVP waveform This means that the corresponding hemodynamic data is not valid for further processing. Validation module 204 also smooths the RVP waveform This is output to the flow module 206.

[0202] Flow module 206 is smoothed from verification module 204 RVP waveform It receives as input. The flow module 206 checks whether flow_valid is true or negative (i.e., flow processed CO AE Output to module 208. The instruction that flow_valid is true (for example, flow_valid=true) indicates the estimated blood flow waveform flow processed CO AEThis means that it is valid for further processing by module 208. The instruction that flow_valid is false (e.g., flow_valid=false) means that the estimated blood flow waveform flow processed This means that it is not effective for further processing. Flow module 206 is flow processed to CO AE Further output is sent to module 208. Flow module 206 also outputs flow processed The output is sent to processing device 214.

[0203] CO AE Module 208 is flow processed It receives as input. flow processed CO2 is released from flow module 206. AE It is supplied to module 208. AE Module 208 also determines whether flow_valid is true or false, and accordingly CO AE Module 208 receives instructions from flow module 206 indicating whether it can proceed. AE Module 208 is CO AE to CO LR Module 210 and CO filtered Output to module 212.

[0204] CO LR Module 210 is RVP features and CO AE It receives and as input. RVP features This is from RVP feature module 202 to CO LR It is supplied to module 210. AE CO AE CO from module 208 LR It is supplied to module 210. LR Module 210 is CO LR and ΔCO LR and CO filtered Output to module 212.

[0205] CO filtered Module 212 is CO AECO LR , and ΔCO LR It receives CO as input. AE CO AE CO from module 208 filtered It is supplied to module 212. LR and ΔCO LR CO LR CO from module 210 filtered It is supplied to module 212. filtered Module 212 is CO filtered This is output to output device 214.

[0206] As shown in Figure 23, the output device 214 is connected to the flow module 206 and CO filtered It receives the output from module 212. More specifically, output device 214 receives the flow processed and CO filtered The system receives the following outputs. Each of these outputs can be displayed as a corresponding graph representing the value over time (for example, via display 24 as shown in Figure 1).

[0207] Any of the various systems, devices, and apparatus in this disclosure can be sterilized (e.g., using heat, radiation, ethylene oxide, hydrogen peroxide, etc.) to ensure that they are safe for use in patients, and the methods described herein may include sterilization (e.g., using heat, radiation, ethylene oxide, hydrogen peroxide, etc.) of the relevant systems, devices, and apparatus.

[0208] Therapeutic techniques, methods, steps, etc., described or suggested in this specification or the references incorporated herein may be performed on living animals or on non-living simulations such as corpses, corpse hearts, anthropomorphic ghosts, or simulators (in which body parts, tissues, etc., are simulated).

[0209] Discussion of possible examples The following is a non-exclusive description of possible examples of the present invention.

[0210] A system for determining a patient's hemodynamic state includes a first hemodynamic sensor and a display. The first hemodynamic sensor continuously generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which instructions are encoded, when executed by one or more processors, to cause the system to receive the first hemodynamic sensor signal representing the patient's right ventricular pressure waveform, to convert the patient's right ventricular pressure waveform into an estimate of the blood flow waveform, and to extract features from the patient's right ventricular pressure waveform. The instructions further cause the system to estimate the patient's cardiac output based on the patient's right ventricular pressure waveform or features extracted from the patient's right ventricular pressure waveform, and to output the patient's blood flow waveform and the patient's cardiac output to the display.

[0211] The system described in the preceding paragraph may optionally, additionally and / or alternatively, include any one or more of the following features, configurations, and / or additional components:

[0212] The system may further include a catheter connected to a first hemodynamic sensor, and when a command is executed by one or more processors, the system may further cause the system to compare features extracted from the patient's right ventricular pressure waveform with one or more specified ranges of features, in order to determine if the catheter is properly positioned if the values ​​of the features extracted from the patient's right ventricular pressure waveform fall within one or more specified ranges of values, or to determine if the catheter is mispositioned if the values ​​of the features extracted from the patient's right ventricular pressure waveform fall within one or more specified ranges of values, and to output an indication to a display or a module of the system indicating whether the catheter is properly positioned in the patient based on whether the values ​​of the features extracted from the patient's right ventricular pressure waveform fall within one or more specified ranges of values.

[0213] One or more specified value ranges for features can correspond to the physiological limits of each feature extracted from the patient's right ventricular pressure waveform, plus the standard deviation of each feature.

[0214] The specified value range can be used to identify noise, under-attenuation signals, or over-attenuation signals in the patient's right ventricular pressure waveform.

[0215] The system can continuously monitor the placement of the catheter.

[0216] The correct placement of the catheter can be determined by the quality of the right ventricular pressure waveform.

[0217] When the instruction is executed by one or more processors, the system may further filter the data from the patient's right ventricular pressure waveform by excluding data from portions of the right ventricular pressure waveform that contain features extracted from the patient's right ventricular pressure waveform that are not within one or more specified value ranges.

[0218] A portion of the right ventricular pressure waveform may be a right ventricular pressure waveform heartbeat that includes features extracted from the right ventricular pressure waveform of a patient whose values ​​are not within one or more specified ranges.

[0219] Features extracted from a patient's right ventricular pressure waveform may include one or more of the following: end-diastolic blood pressure, maximal systolic blood pressure, minimum diastolic blood pressure, end-systolic blood pressure, percentage change in maximal blood pressure over time, percentage change in minimum blood pressure over time, diastolic gradient, and right ventricular pulse pressure.

[0220] When the instruction is executed by one or more processors, the system can further assign a signal quality index to the right ventricular pressure waveform based on whether the values ​​of the features extracted from the patient's right ventricular pressure waveform fall within one or more specified value ranges, and the signal quality index indicates the quality of the right ventricular pressure waveform.

[0221] Signal quality indicators can be assigned to the right ventricular pressure waveform based on the characteristic values ​​from 10-second intervals of the right ventricular pressure waveform.

[0222] The signal quality index can range from 0 to 5, where 0 is the highest quality and 5 is the lowest quality. The signal quality index can be based on how closely the features extracted from the patient's right ventricular pressure waveform are within one or more specified value ranges.

[0223] When the instruction is executed by one or more processors, the system can further cause the system to determine whether the patient's right ventricular pressure waveform is valid based on a signal quality index of the patient's right ventricular pressure waveform. A valid right ventricular pressure waveform indicates that the catheter is properly positioned in the patient, while an invalid right ventricular pressure waveform indicates that the catheter is not properly positioned in the patient, is not properly connected to the system, or is experiencing a signal quality problem.

[0224] The patient's right ventricular pressure waveform can be converted into an estimate of the blood flow waveform using a machine learning model.

[0225] Machine learning models can be models based on deep learning that use neural network architectures.

[0226] The machine learning model can be an autoencoder model, and using an autoencoder model may include inputting a patient's right ventricular pressure waveform, encoding the right ventricular pressure waveform into condensed data through a first set of filters, storing the condensed data in latent space, decoding the condensed data through a second set of filters, and outputting a waveform of the patient's blood flow.

[0227] A 10-second sample of the patient's right ventricular pressure waveform can be input into the autoencoder model on a rolling basis.

[0228] The machine learning model can be an autoencoder model.

[0229] Autoencoder models can be trained using animal data.

[0230] Once the instructions are executed by one or more processors, the system can then be used to train an autoencoder model.

[0231] Training an autoencoder model may include training the autoencoder model using measured animal right ventricular pressure waveforms and measured animal blood flow to obtain a trained autoencoder model of an animal; inputting the animal right ventricular pressure waveforms into the trained autoencoder model of an animal to generate predicted animal blood flow; comparing the measured animal blood flow to predicted animal blood flow to validate the trained autoencoder model of an animal; determining whether the predicted animal blood flow from the trained autoencoder model of an animal is valid or invalid; inputting human right ventricular pressure waveforms into the trained autoencoder model of an animal to generate raw human blood flow; scaling human raw blood flow using intermittent cardiac output to generate scaled human raw blood flow; and retraining the trained autoencoder model of an animal using human right ventricular pressure waveforms and scaled human raw blood flow to obtain a trained autoencoder model of an animal.

[0232] The autoencoder model can estimate the raw blood flow waveform of a patient, and the estimated blood flow waveform can be used as an estimate of the processed blood flow waveform of the patient.

[0233] The raw blood flow waveform of a patient can be filtered to remove artifacts or physiological inaccuracies from the raw blood flow waveform, resulting in a processed blood flow waveform of the patient.

[0234] Negative flow rates, flow values ​​in the thousands, rectangular flow rates, excessive flow rate increases, or excessively different flow rates per heartbeat can be removed from the raw blood flow waveform to obtain the processed blood flow waveform of the patient.

[0235] The estimated blood flow waveform can be the processed blood flow waveform, which is filtered through a flow filter to reflect the physiological prediction of blood flow.

[0236] When the instruction is executed by one or more processors, the system can further smooth the waveform of the patient's blood flow.

[0237] When the instruction is executed by one or more processors, the system can further cause the system to integrate the waveform of the patient's blood flow in order to estimate the patient's cardiac output.

[0238] When the instruction is executed by one or more processors, the system can further cause the system to assign a signal quality index to the patient's blood flow waveform based on the amount of error data detected in the patient's blood flow waveform, and the signal quality index indicates the quality of the patient's blood flow waveform.

[0239] Signal quality indicators can be assigned to the patient's blood flow waveform based on an analysis of the patient's blood flow waveform at 10-second intervals.

[0240] When the instruction is executed by one or more processors, the system can further cause the system to determine whether the patient's blood flow waveform is valid based on a signal quality index of the patient's blood flow waveform, where a valid patient blood flow waveform indicates that the patient's blood flow waveform is of high quality and can be used for further analysis, and an invalid patient blood flow waveform indicates that the patient's blood flow waveform is of low quality and cannot be used for further analysis.

[0241] The patient's cardiac output can be estimated using a regression model based on features extracted from the patient's right ventricular pressure waveform.

[0242] Features extracted from the right ventricular pressure waveform may include one or more of the following: pulse rate, percentage change in maximum blood pressure over time during systolic elevation (dP / dt), minimum dP / dt during post-systolic relaxation, systolic duration, systolic blood pressure, end-systolic blood pressure, end-diastolic blood pressure, pulse pressure, and mean arterial pressure.

[0243] Features extracted from a patient's right ventricular pressure waveform may include a reference feature corresponding to a reference time window and a current feature corresponding to the current time window.

[0244] The reference time window and the current time window can each be a 10-second interval representing a 10-second portion of the patient's right ventricular pressure waveform.

[0245] The reference time window can correspond to the time interval since the patient was admitted to the patient care environment, or to a time interval associated with the patient's known cardiac output value, while the current time window can correspond to real-time or near-real-time measurements from the patient.

[0246] Reference features and reference time windows can be updated periodically or continuously.

[0247] Reference features and reference time windows can be updated based on the passage of a set amount of time.

[0248] A regression model can use the changes between individual features of the reference feature and their corresponding individual features of the current feature to determine the change in cardiac output from a reference time window to the current time window.

[0249] The change in cardiac output determined using a regression model can be input to the cardiac output estimator submodule, which then calculates cardiac output based on the change in cardiac output by adding the magnitude of the change in cardiac output to a reference cardiac output value corresponding to a reference time window, so that the cardiac output calculated by the cardiac output estimator submodule corresponds to the current time window.

[0250] A regression model may include one or more variables, which may include one or more first variables, each first variable representing a measure of a reference feature or current feature; one or more second difference variables, each second difference variable representing the difference between an individual feature of the reference feature and a corresponding individual feature of the current feature; one or more third combination variables, each third combination variable representing a combination of one or more first variables and / or one or more second difference variables; or any combination of one or more first variables, one or more second difference variables, and / or one or more third combination variables.

[0251] A regression model may include one or more terms, each of which is formed by multiplying one or more variables by a predetermined coefficient, and the change in cardiac output can be calculated using the regression model as the sum of all one or more terms.

[0252] The change in cardiac output calculated using a regression model can be input to the cardiac output estimator submodule, which then calculates cardiac output based on the change in cardiac output by adding the magnitude of the change in cardiac output to a reference cardiac output value corresponding to a reference time window, so that the cardiac output calculated by the cardiac output estimator submodule corresponds to the current time window.

[0253] Cardiac output estimated using a regression model can be a continuous estimate of the patient's cardiac output.

[0254] Regression models can be considered machine learning models.

[0255] The regression model can be a linear regression model.

[0256] The patient's cardiac output can be a filtered cardiac output estimated by filtering one or more cardiac output estimates using a Kalman filter algorithm, at least one of which is determined using the patient's right ventricular pressure waveform or features extracted from the patient's right ventricular pressure waveform.

[0257] The Kalman filter algorithm can be configured such that each iteration of the Kalman filter algorithm includes a prediction phase and one or more update phases, and the prediction phase can be performed alternately with the one or more update phases.

[0258] The prediction phase can predict a predicted cardiac output estimate corresponding to the current time step, and each of the one or more update phases can consume one or more measurement inputs corresponding to the current time step, and the predicted cardiac output estimate can be updated using a weighted average of the predicted cardiac output estimate and each of the one or more measurement inputs.

[0259] One or more measurement inputs may include one or more of the following: autoencoder cardiac output estimated by an autoencoder model, linear regression cardiac output estimated using a regression model, continuous cardiac output acquired via a catheter-based thermal filament, and intermittent cardiac output acquired via a catheter-based thermistor after administration of a fluid bolus.

[0260] The prediction phase can predict a predicted cardiac output estimate based on one or more previous filtered estimates of cardiac output corresponding to previous time steps of the Kalman filter algorithm.

[0261] The predicted estimate of cardiac output may include a corresponding predictive uncertainty, and each of the one or more measurement inputs may include a corresponding measurement uncertainty, and the corresponding predictive uncertainty and the corresponding measurement uncertainty can be predetermined, and the predicted uncertainty of cardiac output and each of the one or more measurement inputs can be weighted by the corresponding predictive uncertainty and the corresponding measurement uncertainty, respectively, such that greater uncertainties are given smaller weights and smaller uncertainties are given larger weights.

[0262] The corresponding predictive uncertainty and corresponding measurement uncertainty can be fixed over time.

[0263] The corresponding predictive uncertainty and the corresponding measurement uncertainty can be modulated over time.

[0264] The corresponding measurement uncertainty can be scaled based on the signal quality index of the patient's right ventricular pressure waveform.

[0265] The corresponding uncertainty in filtered cardiac output may be smaller than the corresponding predictive uncertainty and corresponding measurement uncertainty.

[0266] The Kalman filter algorithm can be configured such that the number of update phases in each iteration of the Kalman filter algorithm depends on which of the one or more measurement inputs is available for the corresponding iteration, and the Kalman filter algorithm can be configured to allow asynchronous reception of one or more measurement inputs and / or additional sources of measurement inputs.

[0267] The Kalman filter algorithm is recursive and can use one or more filtered previous estimates of cardiac output corresponding to previous time steps of the Kalman filter algorithm and one or more measurement inputs corresponding to the current time step to generate filtered cardiac output for the current time step.

[0268] The filtered cardiac output estimated using the Kalman filter algorithm can be used as a continuous estimate of the patient's cardiac output.

[0269] The first hemodynamic sensor can be connected to a hemodynamic monitor that includes a display, one or more processors, and computer-readable memory.

[0270] The system may further include a second hemodynamic sensor that continuously generates a second hemodynamic sensor signal representing the patient's pulmonary artery pressure waveform, and when the instruction is executed by one or more processors, the system may further cause the system to receive the second hemodynamic sensor signal representing the patient's pulmonary artery pressure waveform, extract features from the patient's pulmonary artery pressure waveform, and estimate the patient's cardiac output based on the features extracted from the patient's right ventricular pressure waveform and the features extracted from the patient's pulmonary artery pressure waveform.

[0271] The system can continuously output the patient's blood flow waveform and cardiac output to a display.

[0272] A patient's cardiac output can be determined based on the relative change in the patient's right ventricular pressure.

[0273] The patient's cardiac output can be determined based on the morphology of the patient's right ventricular pressure waveform.

[0274] A patient's cardiac output can be determined based on both the relative change in the patient's right ventricular pressure and the morphology of the patient's right ventricular pressure waveform.

[0275] A system for determining a patient's hemodynamic status includes a first hemodynamic sensor, a catheter connected to the first hemodynamic sensor, and a display. The first hemodynamic sensor continuously generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which instructions, when executed by one or more processors, are encoded to cause the system to receive the first hemodynamic sensor signal representing the patient's right ventricular pressure waveform and to extract features from the patient's right ventricular pressure waveform. The instructions further cause the system to compare the features extracted from the patient's right ventricular pressure waveform with one or more specified ranges of features in order to determine that the catheter is correctly positioned if the values ​​of the features extracted from the patient's right ventricular pressure waveform fall within one or more specified ranges of features, or to determine that the catheter is incorrectly positioned if the values ​​of the features extracted from the patient's right ventricular pressure waveform fall within one or more specified ranges of features. The command further instructs the system to use a machine learning model to convert the patient's right ventricular pressure waveform into an estimate of the patient's blood flow waveform, and to integrate the patient's blood flow waveform to determine the patient's cardiac output. The command further instructs the system to use a regression model to estimate the change in the patient's cardiac output based on features extracted from the right ventricular pressure waveform, and to estimate the patient's filtered cardiac output by filtering the patient's cardiac output using a Kalman filter algorithm that uses the change in the patient's cardiac output. The command further instructs the system to output the patient's blood flow waveform and / or the patient's filtered cardiac output to the display.

[0276] The system described in the preceding paragraph may optionally, additionally and / or alternatively, include any one or more of the following features, configurations, and / or additional components:

[0277] When the command is executed by one or more processors, it can cause the system to output changes in the patient's cardiac output to a display.

[0278] A system for estimating a patient's blood flow includes a hemodynamic sensor and a display. The hemodynamic sensor continuously generates a hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which, when executed by one or more processors, instructions are encoded to cause the system to receive the hemodynamic sensor signal representing the patient's right ventricular pressure waveform and to use a machine learning model to convert the patient's right ventricular pressure waveform into an estimate of the patient's blood flow waveform. The instructions further cause the system to output the estimate of the patient's blood flow waveform to a display or a module for determining the patient's cardiac output.

[0279] The system described in the preceding paragraph may optionally, additionally and / or alternatively, include any one or more of the following features, configurations, and / or additional components:

[0280] When the instruction is executed by one or more processors, the system can further cause the system to integrate the waveform of the patient's blood flow in order to determine the patient's cardiac output.

[0281] Machine learning models can be models based on deep learning that use neural network architectures.

[0282] The machine learning model can be an autoencoder model.

[0283] The autoencoder model can estimate the raw blood flow waveform of a patient, and the estimated blood flow waveform can be used as an estimate of the processed blood flow waveform of the patient.

[0284] The patient's raw blood flow waveform can be filtered to remove artifacts or physiological inaccuracies from the raw waveform, resulting in the patient's processed blood flow waveform.

[0285] Negative flow rates, flow values ​​in the thousands, square flow rates, excessive flow increases, or excessively different flow rates per heartbeat can be removed from the raw blood flow waveform to obtain the processed blood flow waveform of the patient.

[0286] The estimated blood flow waveform can be the processed blood flow waveform, which is filtered through a flow filter to reflect the physiological prediction of blood flow.

[0287] When the instruction is executed by one or more processors, the system can further smooth the waveform of the patient's blood flow.

[0288] The system can continuously output estimated values ​​of the patient's blood flow waveform.

[0289] When the instruction is executed by one or more processors, the system can further filter data from the patient's blood flow waveform by excluding data from portions of the blood flow waveform that contain physiological inaccuracies.

[0290] A portion of the blood flow waveform can be a heart rate waveform of blood flow that includes data extracted from the blood flow waveform of patients whose values ​​are outside the specified range.

[0291] When the instruction is executed by one or more processors, the system can further cause the system to assign a signal quality index to the patient's blood flow waveform based on the amount of error data detected in the patient's blood flow waveform, and the signal quality index indicates the quality of the patient's blood flow waveform.

[0292] The amount of error data can be detected by comparing the blood flow waveform with a specified range of blood flow values.

[0293] Signal quality indicators can be assigned to the patient's blood flow waveform based on an analysis of the patient's blood flow waveform at 10-second intervals.

[0294] The signal quality index can range from 0 to 5, where 0 represents the highest quality and 5 represents the lowest quality.

[0295] When the instruction is executed by one or more processors, the system can further cause the system to determine whether the patient's blood flow waveform is valid based on a signal quality index of the patient's blood flow waveform, where a valid patient blood flow waveform indicates that the patient's blood flow waveform is of high quality and can be used for further analysis, and an invalid patient blood flow waveform indicates that the patient's blood flow waveform is of low quality and cannot be used for further analysis.

[0296] The effective waveform of the patient's blood flow can be used to measure the patient's cardiac output.

[0297] Machine learning models can be trained using animal data.

[0298] Once the instructions are executed by one or more processors, the system can further be used to train a machine learning model, which is an autoencoder model.

[0299] Training an autoencoder model may include training the autoencoder model using measured animal right ventricular pressure waveforms and measured animal blood flow to obtain a trained autoencoder model of an animal; inputting the animal right ventricular pressure waveforms into the trained autoencoder model of an animal to generate predicted animal blood flow; comparing the measured animal blood flow to predicted animal blood flow to validate the trained autoencoder model of an animal; determining whether the predicted animal blood flow from the trained autoencoder model of an animal is valid or invalid; inputting human right ventricular pressure waveforms into the trained autoencoder model of an animal to generate raw human blood flow; scaling human raw blood flow using intermittent cardiac output to generate scaled human raw blood flow; and retraining the trained autoencoder model of an animal using human right ventricular pressure waveforms and scaled human raw blood flow to obtain a trained autoencoder model of an animal.

[0300] The machine learning model can be an autoencoder model, and using an autoencoder model may include inputting a patient's right ventricular pressure waveform, encoding the right ventricular pressure waveform into condensed data through a first set of filters, storing the condensed data in latent space, decoding the condensed data through a second set of filters, and outputting a waveform of the patient's blood flow.

[0301] A 10-second sample of the patient's right ventricular pressure waveform can be input into a machine learning model on a rolling basis.

[0302] The hemodynamic sensor can be connected to a hemodynamic monitor that includes a display, one or more processors, and computer-readable memory.

[0303] The system can continuously output an estimated value of the patient's blood flow waveform to the display in order to monitor the patient's blood flow based on the patient's right ventricular pressure waveform.

[0304] A method for estimating a patient's blood flow includes the steps of: receiving sensed hemodynamic data representing the patient's right ventricular pressure waveform by a hemodynamic monitoring system; and using a machine learning model, converting the patient's right ventricular pressure waveform into an estimate of the patient's blood flow waveform by the hemodynamic monitoring system. The method further includes the step of outputting the estimate of the patient's blood flow waveform by the hemodynamic monitoring system to a display for monitoring the patient's blood flow based on the patient's right ventricular pressure waveform, or to a module for determining the patient's cardiac output by the hemodynamic monitoring system.

[0305] The methods described in the preceding paragraph may optionally, additionally and / or alternatively, include any one or more of the following actions, features, configurations, and / or additional components:

[0306] Machine learning models can be models based on deep learning that use neural network architectures.

[0307] The machine learning model can be an autoencoder model.

[0308] A method for training an autoencoder model to estimate patient blood flow includes the steps of training an autoencoder model using measured animal right ventricular pressure waveforms and measured animal blood flow to obtain a trained autoencoder model of an animal. The method further includes the steps of inputting measured animal right ventricular pressure waveforms into the trained autoencoder model of an animal to generate a predicted animal blood flow, and comparing the measured animal blood flow to a predicted animal blood flow to validate the trained autoencoder model of an animal. The method further includes the step of determining whether the predicted animal blood flow from the trained autoencoder model of an animal is valid or invalid. The method further includes the steps of inputting human right ventricular pressure waveforms into the trained autoencoder model of an animal to generate human raw blood flow, and scaling human raw blood flow using intermittent cardiac output to generate scaled human raw blood flow. The method further includes the step of retraining an animal-trained autoencoder model using human right ventricular pressure waveforms and human scaled raw blood flow to obtain a human-trained autoencoder model for estimating patient blood flow based on the patient's right ventricular pressure waveform.

[0309] A system for determining a patient's hemodynamic state includes a first hemodynamic sensor and a display. The first hemodynamic sensor continuously generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The system further includes one or more processors and a computer-readable memory in which instructions are encoded, when executed by one or more processors, to cause the system to receive the first hemodynamic sensor signal representing the patient's right ventricular pressure waveform and to extract features from the patient's right ventricular pressure waveform. The instructions further cause the system to estimate changes in cardiac output based on the features extracted from the patient's right ventricular pressure waveform using a regression model, and to output the changes in cardiac output to the display and / or to a cardiac output estimator submodule for calculating cardiac output.

[0310] The system described in the preceding paragraph may optionally, additionally and / or alternatively, include any one or more of the following features, configurations, and / or additional components:

[0311] Features extracted from the right ventricular pressure waveform may include one or more of the following: pulse rate, percentage change in maximum blood pressure over time during systolic elevation (dP / dt), minimum dP / dt during post-systolic relaxation, systolic duration, systolic blood pressure, end-systolic blood pressure, end-diastolic blood pressure, pulse pressure, and mean arterial pressure.

[0312] The system may further include a second hemodynamic sensor that continuously generates a second hemodynamic sensor signal representing the patient's pulmonary artery pressure waveform, and when the instruction is executed by one or more processors, the system may further cause the system to receive the second hemodynamic sensor signal representing the patient's pulmonary artery pressure waveform, extract features from the patient's pulmonary artery pressure waveform, and use a regression model to estimate changes in cardiac output based on the features extracted from the patient's pulmonary artery pressure waveform.

[0313] Once the instruction is executed by one or more processors, the system can further receive patient-related demographic information and use a regression model to estimate changes in cardiac output based on the patient-related demographic information.

[0314] Demographic information may include one or more of the patient's age, weight, height, sex, body mass index, and health status.

[0315] The system may further include a third hemodynamic sensor that continuously generates a third hemodynamic sensor signal representing the patient's blood oxygen saturation, and when an instruction is executed by one or more processors, the system may further cause the system to receive the third hemodynamic sensor signal representing the patient's blood oxygen saturation and use a regression model to estimate changes in cardiac output based on the third hemodynamic sensor signal.

[0316] Features extracted from a patient's right ventricular pressure waveform may include a reference feature corresponding to a reference time window and a current feature corresponding to the current time window.

[0317] The reference time window and the current time window can each be a 10-second interval representing a 10-second portion of the patient's right ventricular pressure waveform.

[0318] The reference time window can correspond to the time interval when the patient was admitted to the patient care environment, or to a time interval associated with the patient's known cardiac output value, while the current time window can correspond to real-time or near-real-time measurements from the patient.

[0319] Each feature of a reference feature can correspond to the corresponding feature of the current feature.

[0320] Reference features and reference time windows can be updated periodically or continuously.

[0321] Reference features and reference time windows can be updated based on the passage of a set amount of time.

[0322] A regression model can use the changes between individual features of the reference feature and their corresponding individual features of the current feature to determine the change in cardiac output from a reference time window to the current time window.

[0323] A regression model may include one or more variables, which may include one or more first variables, each first variable representing a measure of a reference feature or current feature; one or more second difference variables, each second difference variable representing the difference between an individual feature of the reference feature and a corresponding individual feature of the current feature; one or more third combination variables, each third combination variable representing a combination of one or more first variables and / or one or more second difference variables; or any combination of one or more first variables, one or more second difference variables, and / or one or more third combination variables.

[0324] A regression model may contain one or more terms, each of which is formed by one or more variables multiplied by a predetermined coefficient.

[0325] Changes in cardiac output can be calculated using a regression model as the sum of all one or more terms.

[0326] Changes in cardiac output can be defined as the percentage change in cardiac output from the reference time window to the current time window.

[0327] The display can show a graph of changes in cardiac output over time.

[0328] The cardiac output estimator submodule can receive changes in cardiac output estimated by a regression model, and can calculate cardiac output based on the changes in cardiac output estimated by the regression model.

[0329] When the instruction is executed by one or more processors, the system can further cause the cardiac output to be displayed on the screen.

[0330] The display can show a graph of cardiac output values ​​over time.

[0331] The cardiac output estimator submodule can further receive reference cardiac output values ​​corresponding to a reference time window, and can calculate cardiac output by adding the magnitude of the change in cardiac output estimated by the regression model to the reference cardiac output value corresponding to the reference time window.

[0332] The reference cardiac output value corresponding to the reference time window can be an autoencoder cardiac output value estimated by an autoencoder model, a continuous cardiac output measurement obtained via a catheter-based thermal filament, or an intermittent cardiac output measurement obtained via a catheter-based thermistor after the administration of a fluid bolus.

[0333] The cardiac output calculated by the cardiac output estimator submodule can correspond to the current time window.

[0334] The cardiac output calculated by the cardiac output estimator submodule can be used as a continuous estimate of the patient's cardiac output.

[0335] Regression models can be considered machine learning models.

[0336] The regression model can be a linear regression model.

[0337] The first hemodynamic sensor can be connected to a hemodynamic monitor which includes a display, one or more processors, and computer-readable memory.

[0338] The system can continuously display changes in cardiac output to monitor the patient's hemodynamic status.

[0339] A system for determining a patient's hemodynamic state includes a first hemodynamic sensor, a second hemodynamic sensor, and a display. The first hemodynamic sensor continuously generates a first hemodynamic sensor signal representing the patient's right ventricular pressure waveform. The second hemodynamic sensor continuously generates a second hemodynamic sensor signal representing the patient's pulmonary artery pressure waveform. The system further includes one or more processors and a computer-readable memory in which, when executed by one or more processors, instructions are encoded to cause the system to receive the first hemodynamic sensor signal representing the patient's right ventricular pressure waveform and to extract features from the patient's right ventricular pressure waveform. The instructions further cause the system to receive the second hemodynamic sensor signal representing the patient's pulmonary artery pressure waveform and to extract features from the patient's pulmonary artery pressure waveform. The command further instructs the system to use a regression model to estimate changes in cardiac output based on features extracted from the patient's right ventricular pressure waveform and features extracted from the patient's pulmonary artery pressure waveform, and to output the changes in cardiac output to the display and / or to the cardiac output estimator submodule for calculating cardiac output.

[0340] The system described in the preceding paragraph may optionally, additionally and / or alternatively, include any one or more of the following features, configurations, and / or additional components:

[0341] The system may further include a third hemodynamic sensor that continuously generates a third hemodynamic sensor signal representing the patient's blood oxygen saturation, and when an instruction is executed by one or more processors, the system may further cause the system to receive the third hemodynamic sensor signal representing the patient's blood oxygen saturation, receive demographic information related to the patient, and use a regression model to estimate changes in cardiac output based on the third hemodynamic sensor signal and the demographic information related to the patient.

[0342] A method for determining a patient's hemodynamic state includes the steps of: receiving sensed hemodynamic data representing the patient's right ventricular pressure waveform by a hemodynamic monitoring system; and performing waveform analysis of the hemodynamic data by the hemodynamic monitoring system to extract features from the patient's right ventricular pressure waveform. The method further includes the steps of: estimating changes in cardiac output based on features extracted from the patient's right ventricular pressure waveform using a regression model by the hemodynamic monitoring system; and outputting the changes in cardiac output to a display and / or cardiac output estimator submodule for calculating cardiac output by the hemodynamic monitoring system.

[0343] While the present invention has been described with reference to illustrative examples, it will be understood by those skilled in the art that various modifications can be made and elements of the invention can be replaced with equivalents without departing from the scope of the invention. In addition, many modifications can be made to adapt the teachings of the invention to specific situations or materials without departing from the essential scope of the invention. Therefore, the invention is not limited to the specific examples disclosed, and is intended to include all examples that fall within the scope of the appended claims. [Explanation of symbols]

[0344] 10. Hemodynamic monitoring system 12. Hemodynamic monitor 14. Hemodynamic sensors 14A Hemodynamic sensor, second pressure transducer, invasive hemodynamic sensor, pressure-sensing hemodynamic sensor, pressure transducer sensor, first pressure transducer 14B Hemodynamic Sensor, Module, Oxygen Saturation Measurement Module, Invasive Hemodynamic Sensor 14C hemodynamic sensor, implantable hemodynamic sensor 14D hemodynamic sensor, implantable hemodynamic sensor 16 patients 18 Healthcare workers 20 System Processors 22 System memory, computer-readable system memory 24 displays 26 Analog-to-Digital Converter (ADC), ADC 28. Digital-to-Analog Converter (DAC), DAC 30 Flow rate and cardiac output software code 32 RVP Feature Modules 34 PAP Feature Modules 36 Verification Modules 38 Flow Modules 40CO AE module 42 CO LR module 44CO filtered module 46 User Interface 48 control elements 50-Sensation Alarm 52 Input and / or Output (I / O) Connectors, I / O Connectors 54 Catheter 56 Sheath 58 lumens 60 Fluid Connectors 60A distal port connector 60B Proximal Injection Connector 60C Right Ventricular Pacing Connector 60D Balloon Connector 62 Optical Connectors 64 Thermistor Connector 66 Thermal Filament Connectors 68 ports 68A port, distal port 68B Proximal port, proximal injection port 68C Right ventricular port 70 balloons 72 Tip 74 syringes 76 Housing 78 Fluid input ports 80 Catheter-side fluid port 82 I / O Cables 84 I / O connectors 86 Housing 88 Protective Door 90 Cable 92 connectors 102 Output Devices 104 RVP waveform trace 106 indicators 108 indicators, diastolic end-level blood pressure 108E Exclusion of end-diastolic blood pressure 110 indicators, maximum systolic blood pressure 110E Maximum systolic blood pressure exclusion 112 indicators 114 indicators 116 indicators 118 indicators 120 indicators 122 PAP waveform trace 124 indicators 126 indicators, maximum systolic blood pressure 128 indicators 130 indicators 132 indicators 134 Signal Quality Detector 136 Judgment Block 137 RVP waveform trace 138 indicators, pulse running time 140 indicators, systolic gradient 142 indicators, mean arterial pressure 144 Autoencoder Models, Animal-Trained Autoencoder Models 146 Flow filter 148 Judgment Block 150 inputs 152 filters, filtering layers 154 Latent space 156 filters, filtering layers 158 Output 160 encoders 162 Decoder 180 Regression Models 182 Cardiac Output Estimator 184A Filter Submodule 184B Filter Submodule 186A Prediction Block 186B Prediction Block 188A Update Block 188B Update Block 189A ΔCO AE Submodule 189B ΔCO AE Submodule 190 Part 1 192 Part 2 194 Part 3 202 RVP Feature Modules 204 Verification Module 206 Flow Module 208 CO AE module 210 CO LR module 212 CO filtered module 214 Output Devices

Claims

1. A system for determining the hemodynamic status of a patient, A first hemodynamic sensor that continuously generates a first hemodynamic sensor signal representing the right ventricular pressure waveform of the patient, The display and One or more processors, When executed by the one or more processors, the system The first hemodynamic sensor signal, representing the right ventricular pressure waveform of the patient, is received. Convert the right ventricular pressure waveform of the aforementioned patient into an estimated value of the blood flow waveform. Features were extracted from the right ventricular pressure waveform of the aforementioned patient. The cardiac output of the patient is estimated based on the right ventricular pressure waveform of the patient or the features extracted from the right ventricular pressure waveform of the patient. The waveform of the patient's blood flow and the patient's cardiac output are output to the display. Instructions are encoded in computer-readable memory and A system equipped with these features.

2. The system further includes a catheter connected to the first hemodynamic sensor, and when the instruction is executed by the one or more processors, the system further includes: If the value of the feature extracted from the patient's right ventricular pressure waveform falls within one or more specified value ranges, it is determined that the catheter is correctly positioned. If the value of the feature extracted from the patient's right ventricular pressure waveform does not fall within the one or more specified value ranges, it is determined that the catheter is incorrectly positioned. To this end, the feature extracted from the patient's right ventricular pressure waveform is compared with the one or more specified value ranges of the feature. Based on whether the value of the feature extracted from the right ventricular pressure waveform of the patient falls within the one or more specified ranges, the system causes the display or a module of the system to output an instruction indicating whether the catheter is correctly positioned within the patient. The system according to claim 1.

3. The system according to claim 2, wherein when the instruction is executed by the one or more processors, the system further filters the data from the right ventricular pressure waveform of the patient by excluding data from a portion of the right ventricular pressure waveform that includes the features extracted from the right ventricular pressure waveform of the patient that are not within the one or more specified ranges of values, the portion of the right ventricular pressure waveform being a heartbeat of the right ventricular pressure waveform that includes the features extracted from the right ventricular pressure waveform of the patient that are not within the one or more specified ranges of values.

4. When the instruction is executed by one or more processors, the system further: A signal quality index is assigned to the right ventricular pressure waveform based on whether the value of the feature extracted from the right ventricular pressure waveform of the patient falls within one or more specified value ranges, wherein the signal quality index indicates the quality of the right ventricular pressure waveform. Based on the signal quality index of the patient's right ventricular pressure waveform, it is determined whether the patient's right ventricular pressure waveform is valid, wherein a valid right ventricular pressure waveform indicates that the catheter is correctly positioned within the patient, and an invalid right ventricular pressure waveform indicates that the catheter is not correctly positioned within the patient, or is not properly connected to the system, or a signal quality problem is occurring. The system according to claim 2, which causes the following to be performed.

5. The right ventricular pressure waveform of the patient is converted into an estimate of the blood flow waveform using a machine learning model, and the machine learning model is an autoencoder model, and the use of the autoencoder model is Inputting the right ventricular pressure waveform of the aforementioned patient, Encoding the right ventricular pressure waveform into condensed data via a first set of filters, The aforementioned condensed data is stored in latent space, The condensed data is decoded through a second set of filters, Outputting the waveform of the blood flow of the patient. The system according to claim 1, including the following:

6. The system according to claim 5, wherein the autoencoder model estimates the waveform of the patient's raw blood flow, the estimate of the blood flow waveform is an estimate of the patient's processed blood flow waveform, and the raw blood flow waveform is filtered to remove artifacts or physiological inaccuracies from the raw blood flow waveform so that the patient's raw blood flow waveform yields the patient's processed blood flow waveform.

7. The system according to claim 1, wherein, once the instruction is executed by the one or more processors, the system further causes the system to integrate the waveform of the patient's blood flow in order to estimate the patient's cardiac output.

8. When the instruction is executed by one or more processors, the system further: A signal quality index is assigned to the blood flow waveform of the patient based on the amount of error data detected in the blood flow waveform of the patient, wherein the signal quality index indicates the quality of the blood flow waveform of the patient. Based on the signal quality index of the patient's blood flow waveform, it is determined whether the patient's blood flow waveform is valid, wherein a valid blood flow waveform indicates that the patient's blood flow waveform is of high quality and can be used for further analysis, and an invalid blood flow waveform indicates that the patient's blood flow waveform is of low quality and cannot be used for further analysis. The system according to claim 1, which causes the following to be performed.

9. The cardiac output of the patient is estimated using a regression model based on the features extracted from the right ventricular pressure waveform of the patient. The features extracted from the right ventricular pressure waveform of the patient include a reference feature corresponding to a reference time window and a current feature corresponding to the current time window. The regression model uses the changes between the individual features of the reference feature and the corresponding individual features of the current feature to determine the change in cardiac output from the reference time window to the current time window. The change in cardiac output determined using the regression model is input to a cardiac output estimator submodule, and the cardiac output estimator submodule calculates the cardiac output based on the change in cardiac output by adding the magnitude of the change in cardiac output to a reference cardiac output value corresponding to the reference time window, such that the cardiac output calculated by the cardiac output estimator submodule corresponds to the current time window. The system according to claim 1.

10. The regression model includes one or more variables, and the one or more variables are One or more first variables, wherein each of the one or more first variables represents a measurement of the reference feature or the current feature, One or more second difference variables, wherein each of the one or more second difference variables represents the difference between the individual features of the reference feature and the corresponding individual features of the current feature, One or more third combination variables, wherein each of the one or more third combination variables represents one or more combinations of the one or more first variables and / or the one or more second difference variables, or Any combination of the one or more first variables, the one or more second difference variables, and / or the one or more third combination variables. The system according to claim 9, including the system described in claim 9.

11. The system according to claim 9, wherein the cardiac output estimated using the regression model is a continuous estimate of the patient's cardiac output.

12. The system according to claim 1, wherein the cardiac output of the patient is a filtered cardiac output estimated by filtering one or more cardiac output estimates using a Kalman filter algorithm, and at least one of the one or more cardiac output estimates is determined using the right ventricular pressure waveform of the patient or the features extracted from the right ventricular pressure waveform of the patient.

13. The Kalman filter algorithm is configured such that each iteration of the Kalman filter algorithm includes a prediction phase and one or more update phases, and the prediction phase is performed alternately with the one or more update phases. The aforementioned prediction phase predicts a predicted estimate of cardiac output corresponding to the current time step, Each of the one or more update phases consumes a measurement input corresponding to the current time step such that one or more measurement inputs are consumed. The predicted estimate of cardiac output is updated using a weighted average of the predicted estimate of cardiac output and each of the one or more measurement inputs. The one or more measurement inputs are Autoencoder cardiac output estimated by the autoencoder model, Linear regression cardiac output estimated using a regression model, Continuous cardiac output obtained via catheter-based thermal filament, and Intermittent cardiac output obtained via catheter-based thermistor after fluid bolus administration Including one or more of the following: The system according to claim 12.

14. The system according to claim 12, wherein the filtered cardiac output estimated using the Kalman filter algorithm is a continuous estimate of the patient's cardiac output.

15. The system according to claim 1, wherein the first hemodynamic sensor is connected to a hemodynamic monitor including the display, one or more processors, and computer-readable memory.

16. The system further comprises a second hemodynamic sensor that continuously generates a second hemodynamic sensor signal representing the pulmonary artery pressure waveform of the patient, When the instruction is executed by one or more processors, the system further: The second hemodynamic sensor signal, which represents the pulmonary artery pressure waveform of the patient, is received. Features are extracted from the pulmonary artery pressure waveform of the aforementioned patient. Based on the features extracted from the right ventricular pressure waveform of the patient and the features extracted from the pulmonary artery pressure waveform of the patient, the cardiac output of the patient is estimated. The system according to claim 1.

17. The system according to claim 1, wherein the system continuously outputs the waveform of the patient's blood flow and the patient's cardiac output to the display.

18. The system according to claim 1, wherein the cardiac output of the patient is based on a relative change in the right ventricular pressure of the patient, and / or the cardiac output of the patient is based on the shape of the right ventricular pressure waveform of the patient.

19. A system for determining the hemodynamic status of a patient, A first hemodynamic sensor that continuously generates a first hemodynamic sensor signal representing the right ventricular pressure waveform of the patient, A catheter connected to the first hemodynamic sensor, The display and One or more processors, When executed by the one or more processors, the system The first hemodynamic sensor signal, representing the right ventricular pressure waveform of the patient, is received. Features are extracted from the right ventricular pressure waveform of the aforementioned patient. If the value of the feature extracted from the patient's right ventricular pressure waveform falls within one or more specified value ranges, it is determined that the catheter is correctly positioned. If the value of the feature extracted from the patient's right ventricular pressure waveform does not fall within the one or more specified value ranges, it is determined that the catheter is incorrectly positioned. To this end, the feature extracted from the patient's right ventricular pressure waveform is compared with the one or more specified value ranges of the feature. Using a machine learning model, the right ventricular pressure waveform of the patient is converted into an estimated value of the patient's blood flow waveform. In order to determine the cardiac output of the patient, the waveform of the patient's blood flow is integrated, Using a regression model, the change in the patient's cardiac output is estimated based on the features extracted from the right ventricular pressure waveform. By filtering the patient's cardiac output using a Kalman filter algorithm that utilizes the changes in the patient's cardiac output, the filtered cardiac output of the patient is estimated. The waveform of the patient's blood flow and / or the filtered cardiac output of the patient are output to the display. Instructions are encoded in computer-readable memory and A system equipped with these features.

20. The system according to claim 19, wherein when the instruction is executed by the one or more processors, the system further causes the system to output the change in the patient's cardiac output to the display.

Images