A machine learning approach towards continuous measurement of pulmonary artery occlusion pressure without occlusion

Abstract

A system for generating an estimated pulmonary artery occlusion pressure (PAOP) measurement on an ongoing basis, without occlusion of pulmonary arteries, includes one or more sensing devices that produce one or more sensor signals. The system estimates the PAOP measurement based upon the one or more sensor signals from the one or more sensing devices, and displays the estimated PAOP measurement via a display device. In some examples, the system displays an indication of whether the estimated PAOP measurement of a patient is elevated, based upon the estimated PAOP measurement exceeding a predetermined threshold.

Description

[0001] Attorney Docket Number: P-31218.W001-B0968-P15068W001

[0002] A MACHINE LEARNING APPROACH TOWARDS CONTINUOUS MEASUREMENT OF PULMONARY ARTERY OCCLUSION PRESSURE WITHOUT OCCLUSION

[0003] CROSS-REFERENCE TO RELATED APPLICATION(S)

[0004] This application claims the benefit of U.S. Provisional Application No. 63 / 685,834, filed August 22, 2024, and entitled “A MACHINE LEARNING APPROACH TOWARDS CONTINUOUS MEASUREMENT OF PULMONARY ARTERY OCCLUSION PRESSURE WITHOUT OCCLUSION,” the disclosure of which is hereby incorporated by reference in its entirety.

[0005] BACKGROUND

[0006] Pulmonary artery occlusion pressure (PAOP) is a measurement taken within the pulmonary artery of the heart after occluding the artery by, for example, inflation of a balloon along a catheter. The resulting PAOP measurement is indicative of the function and health of the left side of the heart. Alternative methods for determining the function and health of the left side of the heart involve significantly more invasive procedures, such as puncturing a hole through the septum of the heart. Thus, occluding the pulmonary artery provides a less invasive way to evaluate the function and health of the left side. Occlusion, however, can still present dangers to the patient because it involves blocking the pulmonary artery for a specified period of time. Thus, a system for determining the PAOP which mitigates the dangers of occlusion is desirable.

[0007] SUMMARY

[0008] A system for generating an estimated pulmonary artery occlusion pressure (PAOP) measurement on an ongoing basis without occlusion of pulmonary arteries includes one or more sensing devices that produce one or more sensor signals. The system further includes a hardware unit that includes a processor, computer-readable memory, and a display device, wherein the computer-readable memory is encoded with instructions that, when executed by the processor, cause the system to perform the following steps. The system receives the one or more sensor signals from the one or more sensing devices. The system estimates the PAOP measurement based upon the one or more sensor signals to generate the estimated PAOP measurement. The system displays the estimated PAOP measurement via the display device.

[0009] A method for generating an estimated pulmonary artery occlusion pressure (PAOP) measurement on an ongoing basis without occlusion of pulmonary arteries includes receiving one or more sensor signals from one or more sensing devices. The Attorney Docket Number: P-31218.W001-B0968-P15068W001 method further includes estimating the PAOP measurement based upon the one or more sensor signals to generate the estimated PAOP measurement. The method further includes displaying, via a display device, the estimated PAOP measurement.

[0010] BRIEF DESCRIPTION OF THE DRAWINGS

[0011] FIG. 1 is a perspective view of a catheter positioned in a pulmonary artery to measure a pulmonary artery pressure (PAP) and a corresponding system for producing an estimated pulmonary artery occlusion pressure (PAOP) waveform.

[0012] FIG. 2 is an example waveform of the PAP including a PAOP obtained via prior art methods of measuring PAOP.

[0013] FIG. 3 is an architecture of an example machine learning model for producing an estimated PAOP measurement.

[0014] FIG. 4 is an architecture of an example selected features approach to producing the estimated PAOP measurement.

[0015] FIG. 5 is a method flowchart of a method for generating the PAOP measurement using a machine learning model.

[0016] FIG. 6 is a method flowchart of a method for generating the PAOP measurement using selected features.

[0017] DETAILED DESCRIPTION

[0018] The techniques of this disclosure relate to a machine learning approach for estimating a continuous pulmonary artery occlusion pressure (PAOP) waveform without necessitating occlusion. Typical methods for measuring PAOP include using a balloon catheter to occlude the pulmonary artery and recording the subsequent pulmonary artery pressure (PAP). The present disclosure alleviates the need for occlusion by feeding a ten- second interval of a PAP waveform to a machine learning model and receiving a PAOP estimation output. The machine learning model can extract machine learning derived features from the PAP waveform and use such features to produce a continuous estimated PAOP measurement. In another example, physiological features can be extracted from the PAP waveform can be selected and a regression can be used to produce the estimated PAOP measurement. The output in either the machine learning derived features case or the physiological features case can additionally or alternatively include a binary indication of whether the PAOP measurement is elevated based upon whether a PAOP measurement exceeds a predetermined threshold using logistic regression. The techniques of this disclosure alleviate the need for occluding the pulmonary artery and hence provide a safer alternative to measure PAOP. Attorney Docket Number: P-31218.W001-B0968-P15068W001

[0019] FIG. 1 is perspective view of system 100, including one or more sensing devices for producing one or more sensor signals, such as catheter 102 positioned in pulmonary artery 104 to measure a pulmonary artery pressure (PAP), and corresponding hardware unit 112 to derive the PAOP waveform based upon the PAP received from catheter 102. System 100 can optionally include electrocardiogram (ECG) sensor 120 and / or right atrial sensor 122 to collect an ECG signal and a right atrial pressure signal respectively in order to enhance the estimation of the PAOP measurement. FIG. 1 depicts catheter 102, pulmonary artery 104, right atrium 106, right ventricle 108, pulmonary artery branch 110, and hardware unit 112. Hardware unit 112 further includes processor 114, memory 116, and display device 118. Catheter 102 is operably connected to hardware unit 112. The hardware unit can be connected with an infusion device 124. The device 124 can include one or more infusion pumps. The infusion device 124 can delivery one or more of a plurality of drugs to a patient. The drugs can alter a hemodynamic state of the patient. The drugs can include fluid, vasodilators, inotropes, or other drugs or combination drugs. The processor 114 can generate a control signal. The control signal can operate the infusion device 124 to dispense one or more of the drugs to the patient. The control signal can include dosage information of the one or more drugs calculated based on the estimated PAOP measurement, as described further below.

[0020] In operation, catheter 102 is fed through right atrium 106 and right ventricle 108 such that it is positioned within pulmonary artery 104 near pulmonary artery branch 110. Catheter 102 is used to measure various hemodynamic parameters such as pulmonary artery pressure (PAP). Catheter 102 measures the various hemodynamic parameters via various hemodynamic sensors attached to catheter 102 (not pictured). The various hemodynamic sensors can be attached to a patient to sense hemodynamic data representative of a PAP waveform, right ventricular pressure (RVP) waveform, blood oxygen saturation, and / or cardiac output of the patient. The preceding are merely intended to be examples, and it is understood that the hemodynamic sensors attached to catheter 102 can measure additional or alternative parameters. The techniques of this disclosure involve sensing the PAP waveform via a pressure sensor attached to catheter 102.

[0021] As depicted, catheter 102 does not comprise a balloon element. In prior art PAOP waveform measurement methods, a pulmonary artery catheter may include a balloon element which inflates, thereby allowing a clinician to measure the PAOP measurement directly from the pulmonary artery catheter. The techniques of this disclosure utilize physiological derived signal features, machine learning derived signal features and Attorney Docket Number: P-31218.W001-B0968-P15068W001 machine learning algorithms (further described below with respect to FIGS. 3 and 4) to estimate the PAOP measurement based upon the hemodynamic parameters measured by catheter 102, and hence alleviate the need for a balloon element to occlude pulmonary artery 104. Although no balloon element is depicted along catheter 102, it is understood that catheter 102 can include a balloon element that remains uninflated when catheter 102 is positioned in pulmonary artery 104.

[0022] Upon measuring the hemodynamic parameters, data from catheter 102 can be transmitted to hardware unit 1 12. Processor 1 14 can, by executing one or more instructions encoded within memory 116, receive the data (e.g., including the PAP signal) from catheter 102. Processor 114 can further, by executing one or more instructions encoded within memory 116, extract one or more features from the data. In some examples, such as those described with respect to FIG. 3, processor 114 can use one or more machine learning derived features computed from the received data. In other examples, such as those described with respect to FIG. 4, physiological features can be selected, and hence processor 114 extracts the selected physiological features from the received data. Processor 114 can, under either approach or a combined approach, generate an estimated PAOP measurement based on the derived / extracted features. The estimated PAOP measurement can be displayed via display device 118.

[0023] The configuration of system 100 is advantageous as it does not require occlusion of the pulmonary artery, which can be potentially dangerous to a patient, and can be used to generate a continuous PAOP measurement. Further, occluding the pulmonary artery can require supervision of a clinician as opposed to being done by nursing staff. As such, the configuration of system 100 can alleviate the need for clinicians to oversee obtaining a PAOP waveform.

[0024] FIG. 2 depicts example waveform 200 of the PAP including a pulmonary artery occlusion pressure waveform (PAOP) obtained via prior art methods of measuring PAOP by occluding the pulmonary artery. Waveform 200 includes first PAP section 202, PAOP section 204, and second PAP section 206. As described with respect to FIG. 1, prior art methods for determining PAOP include using a balloon catheter to temporarily occlude the pulmonary artery. Thus, in obtaining the PAOP, a catheter is inserted within the pulmonary artery to measure the PAP with a balloon uninflated, as indicated by first PAP section 202 of waveform 200. The balloon at the tip of the catheter is then inflated to temporarily occlude the pulmonary artery, thus allowing the PAOP to be measured, as Attorney Docket Number: P-31218.W001-B0968-P15068W001 indicated by PAOP section 204 of waveform 200. The balloon is then deflated, thus allowing flow through the pulmonary artery and resulting in second PAP section 206.

[0025] According to the techniques of this disclosure, a PAOP measurement (i.e., as depicted in PAOP section 204) can be estimated via machine learning modelling and / or physiological modelling applied to hemodynamic parameters that are measured without occluding the pulmonary artery, so that occlusion of the pulmonary artery is not required. Thus, based on the techniques of this disclosure, a continuous estimated PAOP measurement, akin to PAOP section 204 can be produced as a result of PAP waveform data. In some embodiments, ten seconds of historical PAP data is used to generate the estimated PAOP measurement. Additional details regarding the methods of producing the estimated PAOP measurement from the PAP data are described herein in the description of FIG. 3.

[0026] FIG. 3 depicts an architecture of machine learning model 300. Machine learning model 300 is an example machine learning model for producing an estimated PAOP measurement. Machine learning model 300 can be housed within memory 116 of system 100 and machine learning model 300 can be executed by processor 114 of system 100. Machine learning model 300 includes input 302, convolutional layers 304, demographics 306, regularization techniques 308, fully connected layer 310, fully connected layer 312, and PAOP output 314.

[0027] In operation, input 302 is a ten-second sample of hemodynamic data received from catheter 102. The hemodynamic data can include, for example, a PAP waveform, an RVP waveform, cardiac output data, or any combination thereof. In an example embodiment, a ten-second sample of a PAP waveform is measured, for example, by catheter 102 of system 100 (FIG. 1). The ten-second sample of the PAP waveform may be about a ten-second sample (e.g., 9.5 seconds to 10.5 seconds) or any other suitable time sample. Convolutional layers 304 can receive ten-second samples on a rolling basis, for example input 302 of samples are ten-second samples but may be input into convolutional layers 304 every two seconds, thus allowing a continuous output (i.e., PAOP output 314).

[0028] In some examples, additional inputs can be combined with the ten-second PAP waveform. In one such example, an ECG signal can be measured via ECG sensor 120. The ECG signal can be combined with the PAP signal in order to create a combined PAP and ECG signal. In such an example, the combined PAP and ECG signal can then be fed to convolutional layers 304 such that features are extracted from the combined PAP and ECG waveform. In another example, a right atrial signal can be measured via right atrial Attorney Docket Number: P-31218.W001-B0968-P15068W001 sensor 122. The right atrial signal can be combined with the PAP signal in order to create a combined PAP and right atrial signal. In such an example, the combined PAP and right atrial signal can then be fed to convolutional layers 304 such that features are extracted from the combined PAP and right atrial waveform.

[0029] Convolutional layers 304 receive input 302, which can be a PAP waveform, or a combined waveform as described above. In the depicted embodiment, convolutional layers 304 are made up of four convolutional layers with 128, 64, 32, and 16 layers respectively. It is understood that this is merely one example of a machine learning model framework, and that any number of suitable layers and corresponding filters can be used. Convolutional layers 304 operate by using a series of filters to compute a sliding dot product of a convolutional layer (e.g., derived from input 302) and a filter, wherein the filter is configured to have a matrix size and a stride length. The resulting dot product can be indicative of extracted features from input 302. Thus, successive convolutional layers, such as convolutional layers 304, can continually refine input 302 to pare down the extracted features from input 302.

[0030] The output of convolutional layers 304 can then be flattened, for example, to reduce the dimensionality of the output of convolutional layers 304 such that it is suitable for additional processing by fully connected layer 310. Demographics 306 can be included as an input into fully connected layer 310. Such demographic information can include patient age, weight, height, sex (or gender), body mass index (BMI), existing medical condition(s), prior medical condition(s), etc. Demographics 306 can be retrieved either from a digital memory (e.g., in a patient information database) or can be entered by, for example, a healthcare worker. In some embodiments, machine learning model 300 operates without demographics 306.

[0031] Prior to fully connected layer 310 processing the flattened output from convolutional layers 304 and, in some cases, demographics 306, regularization techniques 308 can be applied. Regularization techniques 308 can include, for example, LI regularization, dropout regularization, a rectified linear unit activation function, and / or a MaxPool function.

[0032] LI regularization, also referred to as lasso regression, can be used to mitigate overfitting wherein the model develops patterns specific to training data which may not generalize well to new unseen data. LI regularization can include adding a penalty to the objective function of a model (i.e., a function used within fully connected layer 310 and fully connected layer 312) equal to the absolute value of the weights of the model. Attorney Docket Number: P-31218.W001-B0968-P15068W001

[0033] Using such a regularization technique can lead to a sparse model where some weights are equal to zero. LI regularization can be particularly useful to the techniques of this disclosure as many unnecessary features can be eliminated. Additionally or alternatively, dropout regularization can also be used. Dropout regularization can be used to mitigate overfitting by randomly excluding a certain percentage of neurons during training, thereby improving the generalization ability of the model.

[0034] In addition to regularization techniques 308, a rectified linear unit activation function can be used to introduce nonlinearity into the model. Further, MaxPooling can be used to reduce the spatial dimensions of the output of convolutional layers 304. MaxPooling can include dividing the input matrix (i.e., a matrix of the output of convolutional layers 304) into non-overlapping regions, and outputting a maximum value of each region. MaxPooling can allow for enhanced training speed and noise suppression within machine learning model 300.

[0035] It is understood by those of skill in the art that the described regularization techniques are intended to be examples, and it is understood that other regularization techniques can be used to provide various advantages to machine learning model 300.

[0036] Fully connected layer 310 can apply linear and non-linear transformations to the input received from convolutional layers 304 combined with, in some examples, demographics 306. Fully connected layer 310 can output derived features. In the depicted example, fully connected layer 310 can output 485 features, though it is understood that this is merely an example. Fully connected layer 312 can then further reduce the number of features. In the depicted example, fully connected layer 312 can output 36 features, though it is understood that this is merely an example and fully connected layer 312 can be designed to output any number of features.

[0037] Machine learning model 300 can then generate a PAOP output 314 based upon the derived features of fully connected layer 312. The derived features can, in one example, be derived based upon first PAP section 202 and second PAP section 206 of FIG. 2. More specifically, machine learning model 300 can be trained to identify features within first PAP section 202 and second PAP section 206 which will estimate a PAOP measurement. Thus, such features can be derived from the 10-second sample of the PAOP waveform via input 302. Thereafter, the derived features are used to generate PAOP output 314.

[0038] In some examples, PAOP output 314 is a continuous measurement that estimates the pulmonary artery occlusion pressure as a function of time. In other examples, Attorney Docket Number: P-31218.W001-B0968-P15068W001

[0039] PAOP output 314 is a binary prediction indicative of whether the PAOP measurement is elevated. A PAOP is determined to be elevated when the PAOP exceeds a predetermined threshold. In one example, the predetermined threshold is 18 mmHg. Thus, PAOP output 314 would indicate that the PAOP measurement is elevated when PAOP measurement exceeds 18 mmHg. In some examples, PAOP output 314 includes both the continuous estimated PAOP measurement as a function of time and the binary indication of whether the estimated PAOP measurement is elevated.

[0040] Machine learning model 300 provides various advantages. Machine learning model 300 uses machine learning techniques to derive features from input 302 (e.g., a ten-second sample of a PAP waveform) to estimate PAOP output 314. Generating an estimated PAOP measurement relieves the need for a balloon catheter to physically occlude a pulmonary artery. Further, machine learning model 300 can enhance the accuracy of the estimated PAOP output 314 by combining the ECG and / or right atrial signals with the ten-second sample of the PAP waveform within input 302. Additionally, machine learning model 300 applies regularization techniques to better stabilize PAOP output 314. Specifically, the combination of the LI regularization and the dropout regularization were found to have beneficial effects on the stability of PAOP output 314.

[0041] FIG. 4 depicts an architecture of system 400, which is an approach where selected features (either machine learning derived features obtained using a system like machine learning model 300 described above, or physiological features), can be used to estimate whether a PAOP measurement is elevated. System 400 can be housed within memory 116 of system 100 and system 400 can be executed by processor 114 of system 100. System 400 includes input 402, selected features module 404, demographics 406, regression (logistic or standard) module 408, and PAOP output 414. Input 402, demographics 406, and PAOP output 414 are akin to input 302, demographics 306, and PAOP output 314 of FIG. 3, and hence are incremented by 100 with respect to their reference numerals to demonstrate the similarity.

[0042] In operation, input 402 is a ten-second sample of a PAP waveform as measured, for example, by catheter 102 of system 100 (FIG. 1). The ten-second sample of the PAP waveform may be about a ten-second sample (e.g., 9.5 seconds to 10.5 seconds) or any other suitable time sample. In some examples, additional inputs can be combined with the ten-second PAP waveform. In one such example, an ECG signal can be measured via an ECG sensor. The ECG signal can be combined with the PAP signal in order to create a combined PAP and ECG signal. In such an example, the combined PAP and ECG signal Attorney Docket Number: P-31218.W001-B0968-P15068W001 can then be fed to selected features module 404 such that features are extracted from the combined PAP and ECG waveform. In another example, a right atrial signal can be measured via a right atrial sensor. The right atrial signal can be combined with the PAP signal in order to create a combined PAP and right atrial signal.

[0043] Selected features module 404 receives input 402 (e.g., a ten-second sample of a PAP waveform as measured, for example, by catheter 102). In some examples, selected features module 404 includes methods in code for extracting features from a PAP waveform. The methods in code can be stored in memory 116 and executable by processor 114 of system 100. For example, selected features model 404 can extract features for each beat in the time window and then compute an average for each feature for the entire time window. The features to be extracted from input 402 can be selected features that are determined by a user of system 400. Thus, for example, the user can select features including a PAP diastolic feature. The user selected features can further include pulse rate, maximum dP / dt, minimum dP / dt, systole time, systolic pressure, end systolic pressure, end diastolic pressure, pulse pressure, and mean pressure. Such features can also be machine learning derived, such as by being extracted from machine learning model 300 (FIG. 3). Such features can then be extracted from input 402, and provided to regression (logistic or standard) module 408 as an input.

[0044] Demographics 406 can also be included as an input into regression (logistic or standard) module 408. Such demographic information can include patient age, weight, height, sex (or gender), body mass index (BMI), existing medical condition(s), prior medical condition(s), etc. Demographics 406 can be retrieved either from a digital memory (e.g., in a patient information database) or can be entered by, for example, a healthcare worker. In some embodiments, system 400 operates without demographics 406.

[0045] Regression (logistic or standard) module 408 receives inputs from selected features module 404 and, in some examples, demographics 406 as inputs. Regression module 408 can be a traditional machine learning model which is trained via training data including PAP signals and demographic information. Regression module 408 produces PAOP output 414.

[0046] In some examples, PAOP output 414 is a continuous measurement depicting an estimated pulmonary artery occlusion pressure as a function of time; in such examples, regression module 408 will be a standard regression module. In other examples, PAOP output 414 is a binary estimation indicative of whether the PAOP is elevated; in such examples, regression module 408 will be a logistic regression module. A PAOP is Attorney Docket Number: P-31218.W001-B0968-P15068W001 determined to be elevated when the PAOP exceeds a predetermined threshold. In one example, the predetermined threshold is 18 mmHg. Thus, PAOP output 414 would indicate that the PAOP is elevated when PAOP exceeds 18 mmHg. In some examples, PAOP output 414 includes both the continuous estimated PAOP as a function of time and the binary prediction indicative of whether the PAOP is elevated. In such a case, regression module 408 could include a standard regression module (for a continuous output) as well as a logistic regression module (for a binary output).

[0047] In some examples, the techniques of system 400 and machine learning model 300 can be combined. In such an example, machine learning model 300 can be used as described with respect to the description of FIG. 3 to produce a first estimated PAOP measurement based upon the one or more machine learning derived PAP features. System 400 can similarly produce a second estimated PAOP measurement based upon the one or more selected PAP features. The first estimated PAOP measurement and the second estimated PAOP measurement can then be combined (e.g., as a weighted average) and be displayed via display device 118 of system 100. In some examples, an ensemble model can be used to combine the first estimated PAOP measurement and the second estimated PAOP measurement.

[0048] System 400 performs an alternative method of outputting an estimated PAOP measurement, and hence has similar advantages to those described with respect to machine learning model 300. Specifically, system 400 performs a method for estimating PAOP measurement without the need for occluding the pulmonary artery. System 400 provides the additional advantage of operating based upon selected parameters derived from input 402. Such parameter selection allows for less intensive processing power than would be used, for example, by machine learning model 300 in the process of deriving features.

[0049] FIG. 5 is a method flowchart of method 500 for estimating the PAOP measurement using a machine learning model. Reference will be made to the reference numerals of FIG. 1 for clarity, though it is understood that the techniques of method 500 can apply to any of the various examples described in this disclosure.

[0050] Method 500 begins at step 502, wherein a PAP signal is received by processor 114 from catheter 102. The PAP signal can include a ten-second sample of the PAP waveform. The PAP signal can optionally be combined with an ECG signal and / or a right atrial signal. The ECG signal can be received from a connected ECG sensor and the right atrial signal can be received from a connected right atrial sensor. Attorney Docket Number: P-31218.W001-B0968-P15068W001

[0051] At step 504, processor 114 derives one or more features from the PAP signal via a machine learning model to generate one or more machine learning derived PAP features. The machine learning model can be a convolutional neural network, such as machine learning model 300, as depicted in FIG. 3.

[0052] At step 506, processor 114 generates an estimated PAOP measurement based upon the one or more machine learning derived PAP features. In some examples, processor 114 can also generate a binary indication of whether the PAOP is elevated (i.e., whether the PAOP exceeds a predetermine threshold). At step 508, the estimated PAOP measurement and / or the indication of an elevated PAOP is displayed via display device 118. At step 508, a control signal can be generated to control an infusion device based on the estimated PAOP, as described above in relation to infusion device 124.

[0053] FIG. 6 is a method flowchart of method 600 for estimating the PAOP measurement using selected PAP features. Reference will be made to the reference numerals of FIG. 1 for clarity, though it is understood that the techniques of method 600 can apply to any of the various examples described in this disclosure.

[0054] Method 600 begins at step 602, wherein a PAP signal is received from catheter 102 of system 100. The PAP signal can include a ten-second sample of the PAP waveform. The PAP signal can optionally be combined with an ECG signal and / or a right atrial signal. The ECG signal can be received from a connected ECG sensor and the right atrial signal can be received from a connected right atrial sensor.

[0055] At step 604, processor 114 extracts one or more physiological or machine learning derived features from the PAP signal via a selection to generate one or more selected PAP features. Selecting the PAP features can include executing methods in code (e.g., stored in memory 116) for extracting PAP features from the PAP signal.

[0056] At step 606, processor 114 applies a regression model to the one or more selected PAP features to generate an estimated PAOP measurement. In some examples, processor 114 can also generate a binary indication of whether the PAOP is elevated (i.e., whether the PAOP exceeds a predetermined threshold), such as by using a logistic regression model. At step 608, the estimated PAOP measurement and / or the indication of an elevated PAOP is displayed via display device 118. At step 608, a control signal can be generated to control an infusion device based on the estimated PAOP, as described above in relation to infusion device 124.

[0057] The techniques of this disclosure allow for estimating a PAOP measurement without necessitating occlusion. The present disclosure alleviates the need for occlusion by Attorney Docket Number: P-31218.W001-B0968-P15068W001 feeding a ten-second interval of a PAP waveform to a machine learning model and receiving a PAOP estimation output. In another example, the features extracted from the PAP waveform can be selected and a regression model can be used to produce the continuous estimated PAOP measurement. The output in either case can additionally or alternatively include a binary indication of whether the PAOP is elevated based upon whether a PAOP level exceeds a predetermined threshold. The techniques of this disclosure alleviate the need for occluding the pulmonary system and hence provide a safer alternative to producing the PAOP waveform.

[0058] Any of the various systems, devices, apparatuses, etc. (e.g., catheter 102) in this disclosure can be sterilized (e.g., with heat, radiation, ethylene oxide, hydrogen peroxide, etc.) to ensure they are safe for use with patients, and the methods herein can comprise sterilization of the associated system, device, apparatus, etc. (e.g., with heat, radiation, ethylene oxide, hydrogen peroxide, etc.).

[0059] DISCUSSION OF POSSIBLE EMBODIMENTS

[0060] The following are non-exclusive descriptions of possible embodiments of the present invention.

[0061] A system for generating an estimated pulmonary artery occlusion pressure (PAOP) measurement on an ongoing basis without occlusion of pulmonary arteries includes one or more sensing devices that produce one or more sensor signals. The system further includes a hardware unit including a processor, computer-readable memory, and a display device, wherein the computer-readable memory is encoded with instructions that, when executed by the processor, cause the system to perform the following steps. The system receives the one or more sensing signals from the one or more sensing devices. The system estimates the PAOP measurement based upon the one or more sensor signals to generate the estimated PAOP measurement. The system displays the estimated PAOP measurement via the display device.

[0062] The system of the preceding paragraph can optionally include, additionally and / or alternatively, any one or more of the following features, configurations and / or additional components:

[0063] The one or more sensor signals may be indicative of a pulmonary artery pressure (PAP) signal, a right ventricular (RV) pressure signal, a right atrial (RA) pressure signal, an arterial blood pressure (ABP) signal and / or an electrocardiogram (ECG) signal.

[0064] The computer-readable memory may be further encoded with instructions that, when executed by the processor, cause the system to process the one or more sensor Attorney Docket Number: P-31218.W001-B0968-P15068W001 signals to extract clinical physiological features and / or machine learning derived features, and process the clinical physiological features and / or machine learning derived features into one or more models.

[0065] The one ore more models may comprise a regression model for estimating the PAOP measurement.

[0066] The one or more models may comprise a machine learning model for estimating the PAOP measurement.

[0067] The one or more models may comprise a combination of a machine learning model and a regression model for estimating the PAOP measurement.

[0068] The combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement may include a summation of the machine learning derived features multiplied by first weighting coefficients with the clinical physiological features multiplied by second weighting coefficients.

[0069] The computer-readable memory may be further encoded with instructions that, when executed by the processor, cause the system to receive patient demographic information, and apply the patient demographic information to the combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement via adjusting the first weighting coefficients and / or the second weighting coefficients.

[0070] The patient demographic information may be indicative of a PAOP measurement model to be used for estimating the PAOP measurement.

[0071] The patient demographic information may include age, weight, height, sex, body mass index (BMI), existing medical condition(s) and / or prior medical condition(s).

[0072] The machine learning model may include dropout regularization and / or L 1 regularization.

[0073] The combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement may be implemented via an ensemble model.

[0074] The clinical physiological features may include a pulmonary artery pressure (PAP) diastolic pressure.

[0075] The estimated PAOP measurements may be updated continuously at a clinically relevant update rate. Attorney Docket Number: P-31218.W001-B0968-P15068W001

[0076] The clinically relevant update rate may be between 10 seconds and 300 seconds.

[0077] The one or more sensor signals may include a time history of sensor data.

[0078] The time history of sensor data may be between 10 seconds and 300 seconds.

[0079] The computer-readable memory may be further encoded with instructions that, when executed by the processor, cause the system to evaluate whether the estimated PAOP measurement is elevated based upon whether the estimated PAOP measurement exceeds a predetermined PAOP elevated threshold, and display, via the display device, an indication of whether the estimated PAOP measurement is elevated.

[0080] The predetermined PAOP threshold may be between 12 and 20 millimeters of mercury (mmHg).

[0081] The computer-readable memory may be further encoded with instructions that, when executed by the processor, cause the system to receive an external PAOP waveform and / or external PAOP measurement from a user, and calibrate the estimated PAOP measurement based upon the external PAOP waveform and / or external PAOP measurement from the user.

[0082] The computer-readable memory may be further encoded with instructions that, when executed by the processor, cause the system to receive an internal PAOP waveform and / or internal PAOP measurement from the system, and calibrate the estimated PAOP measurement based upon the internal PAOP waveform and / or internal PAOP measurement from the system.

[0083] The one or more sensing devices may be sterilized.

[0084] A method for generating an estimated pulmonary artery occlusion pressure (PAOP) measurement on an ongoing basis without occlusion of pulmonary arteries includes receiving one or more sensor signals from one or more sensing devices. The method further includes estimating the PAOP measurement based upon the one or more sensor signals to generate the estimated PAOP measurement. The method further includes displaying, via the display device, the estimated PAOP measurement.

[0085] The one or more sensor signals may be indicative of a pulmonary artery pressure (PAP) signal, a right ventricular (RV) pressure signal, a right atrial (RA) pressure signal, an arterial blood pressure (ABP) signal and / or an electrocardiogram (ECG) signal.

[0086] The method may further include processing the one or more sensor signals to extract clinical physiological features and / or machine learning derived features, and Attorney Docket Number: P-31218.W001-B0968-P15068W001 processing the clinical physiological features and / or machine learning derived features into one or more models.

[0087] The one or more models may comprise a regression model for estimating the PAOP measurement.

[0088] The one or more models may comprise a machine learning model for estimating the PAOP measurement.

[0089] The one or more models may comprise a combination of a machine learning model and a regression model for estimating the PAOP measurement.

[0090] The combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement may include a summation of the machine learning derived features multiplied by first weighting coefficients with the clinical physiological features multiplied by second weighting coefficients.

[0091] The method may further include receiving patient demographic information, and applying the patient demographic information to the combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement via adjusting the first weighting coefficients and / or the second weighting coefficients.

[0092] The patient demographic information may be indicative of a PAOP measurement model to be used for estimating the PAOP measurement.

[0093] The patient demographic information may include age, weight, height, sex, body mass index (BMI), existing medical condition(s) and / or prior medical condition(s).

[0094] The machine learning model may include dropout regularization and / or LI regularization.

[0095] The combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement may be implemented via an ensemble model.

[0096] The clinical physiological features may include a pulmonary artery pressure (PAP) diastolic pressure.

[0097] The estimated PAOP measurements may be updated continuously at a clinically relevant update rate.

[0098] The clinically relevant update rate may be between 10 seconds and 300 seconds.

[0099] The one or more sensor signals may include a time history of sensor data. Attorney Docket Number: P-31218.W001-B0968-P15068W001

[0100] The time history of sensor data may be between 10 seconds and 300 seconds.

[0101] The method may further include evaluating whether the estimated PAOP measurement is elevated based upon whether the estimated PAOP measurement exceeds a predetermined PAOP elevated threshold, and displaying, via the display device, an indication of whether the estimated PAOP measurement is elevated.

[0102] The predetermined PAOP threshold may be between 12 and 20 millimeters of mercury (mmHg).

[0103] The method may further include receiving an external PAOP waveform and / or external PAOP measurement from a user, and calibrating the estimated PAOP measurement based upon the external PAOP waveform and / or external PAOP measurement from the user.

[0104] The method may further include receiving an internal PAOP waveform and / or internal PAOP measurement from the system, and calibrating the estimated PAOP measurement based upon the internal PAOP waveform and / or internal PAOP measurement from the system.

[0105] The method may further include sterilizing the one or more sensing devices.

[0106] The above method(s) can be performed on a living animal or on a simulation, such as on a cadaver, cadaver heart, anthropomorphic ghost, simulator (e.g., with body parts, heart, tissue, etc. being simulated).

[0107] While the invention has been described with reference to an exemplary embodiment(s), it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for elements thereof without departing from the scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from the essential scope thereof. Therefore, it is intended that the invention not be limited to the particular embodiment(s) disclosed, but that the invention will include all embodiments falling within the scope of the appended claims.

Claims

Attorney Docket Number: P-31218.W001-B0968-P15068W001CLAIMS:

1. A system for generating an estimated pulmonary artery occlusion pressure (PAOP) measurement on an ongoing basis without occlusion of pulmonary arteries, the system comprising: one or more sensing devices that produce one or more sensor signals; and a hardware unit comprising a processor, computer-readable memory, and a display device, wherein the computer-readable memory is encoded with instructions that, when executed by the processor, cause the system to: receive the one or more sensor signals from the one or more sensing devices; estimate the PAOP measurement based upon the one or more sensor signals to generate the estimated PAOP measurement; and display the estimated PAOP measurement via the display device.

2. The system of claim 1 , wherein the one or more sensor signals are indicative of a pulmonary artery pressure (PAP) signal, a right ventricular (RV) pressure signal, a right atrial (RA) pressure signal, an arterial blood pressure (ABP) signal and / or an electrocardiogram (ECG) signal.

3. The system of claim 1, wherein the computer-readable memory is further encoded with instructions that, when executed by the processor, cause the system to: process the one or more sensor signals to extract clinical physiological features and / or machine learning derived features; and process the clinical physiological features and / or machine learning derived features into one or more models.

4. The system of claim 3, wherein the one or more models comprise a regression model for estimating the PAOP measurement.

5. The system of claim 3, wherein the one or more models comprise a machine learning model for estimating the PAOP measurement.

6. The system of claim 3, wherein the one or more models comprise a combination of a machine learning model and a regression model for estimating the PAOP measurement.

7. The system of claim 6, wherein the combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement includes a summation of the machine learning derived featuresAttorney Docket Number: P-31218.W001-B0968-P15068W001 multiplied by first weighting coefficients with the clinical physiological features multiplied by second weighting coefficients.

8. The system of claim 7, wherein the computer-readable memory is further encoded with instructions that, when executed by the processor, cause the system to: receive patient demographic information; and apply the patient demographic information to the combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement via adjusting the first weighting coefficients and / or the second weighting coefficients.

9. The system of claim 8, wherein the patient demographic information is indicative of a PAOP measurement model to be used for estimating the PAOP measurement.

10. The system of claim 8, wherein the patient demographic information includes age, weight, height, sex, body mass index (BMI), existing medical condition(s) and / or prior medical condition(s).

11. The system of claim 6, wherein the machine learning model includes dropout regularization and / or LI regularization.

12. The system of claim 6, wherein the combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement is implemented via an ensemble model.

13. The system of claim 3, wherein the clinical physiological features include a pulmonary artery pressure (PAP) diastolic pressure.

14. The system of claim 1, wherein the estimated PAOP measurements are updated continuously at a clinically relevant update rate.

15. The system of claim 14, wherein the clinically relevant update rate is between 10 seconds and 300 seconds.

16. The system of claim 1, wherein the one or more sensor signals include a time history of sensor data.

17. The system of claim 16, wherein the time history of sensor data is between 10 seconds and 300 seconds.

18. The system of claim 1, wherein the computer-readable memory is further encoded with instructions that, when executed by the processor, cause the system to:Attorney Docket Number: P-31218.W001-B0968-P15068W001 evaluate whether the estimated PAOP measurement is elevated based upon whether the estimated PAOP measurement exceeds a predetermined PAOP elevated threshold; and display, via the display device, an indication of whether the estimated PAOP measurement is elevated.

19. The system of claim 18, wherein the predetermined PAOP threshold is between 12 and 20 millimeters of mercury (mmHg).

20. The system of claim 1 , wherein the computer-readable memory is further encoded with instructions that, when executed by the processor, cause the system to: receive an external PAOP waveform and / or external PAOP measurement from a user; and calibrate the estimated PAOP measurement based upon the external PAOP waveform and / or external PAOP measurement from the user.

21. The system of claim 1, wherein the computer-readable memory is further encoded with instructions that, when executed by the processor, cause the system to: receive an internal PAOP waveform and / or internal PAOP measurement from the system; and calibrate the estimated PAOP measurement based upon the internal PAOP waveform and / or internal PAOP measurement from the system.

22. The system of claim 1, wherein the one or more sensing devices are sterilized.

23. A method for generating an estimated pulmonary artery occlusion pressure (PAOP) measurement on an ongoing basis without occlusion of pulmonary arteries, the method comprising: receiving one or more sensor signals from one or more sensing devices; estimating the PAOP measurement based upon the one or more sensor signals to generate the estimated PAOP measurement; and displaying the estimated PAOP measurement via a display device.

24. The method of claim 22, wherein the one or more sensor signals are indicative of a pulmonary artery pressure (PAP) signal, a right ventricular (RV) pressure signal, a right atrial (RA) pressure signal, an arterial blood pressure (ABP) signal and / or an electrocardiogram (ECG) signal.

25. The method of claim 23, further comprising:Attorney Docket Number: P-31218.W001-B0968-P15068W001 processing the one or more sensor signals to extract clinical physiological features and / or machine learning derived features; and processing the clinical physiological features and / or machine learning derived features into one or more models.

26. The method of claim 25, wherein the one or more models comprise a regression model for estimating the PAOP measurement.

27. The method of claim 25, wherein the one or more models comprise a machine learning model for estimating the PAOP measurement.

28. The method of claim 25, wherein the one or more models comprise a combination of a machine learning model and a regression model for estimating the PAOP measurement.

29. The method of claim 28, wherein the combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement includes a summation of the machine learning derived features multiplied by first weighting coefficients with the clinical physiological features multiplied by second weighting coefficients.

30. The method of claim 29, further comprising: receiving patient demographic information; and applying the patient demographic information to the combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement via adjusting the first weighting coefficients and / or the second weighting coefficients.

31. The method of claim 30, wherein the patient demographic information is indicative of a PAOP measurement model to be used for estimating the PAOP measurement.

32. The method of claim 30, wherein the patient demographic information includes age, weight, height, sex, body mass index (BMI), existing medical condition(s) and / or prior medical condition(s).

33. The method of claim 28, wherein the machine learning model includes dropout regularization and / or LI regularization.

34. The method of claim 28, wherein the combination of the machine learning model for estimating the PAOP measurement and the regression model for estimating the PAOP measurement is implemented via an ensemble model.Attorney Docket Number: P-31218.W001-B0968-P15068W00135. The method of claim 25, wherein the clinical physiological features include a pulmonary artery pressure (PAP) diastolic pressure.

36. The method of claim 23, wherein the estimated PAOP measurements are updated continuously at a clinically relevant update rate.

37. The method of claim 36, wherein the clinically relevant update rate is between 10 seconds and 300 seconds.

38. The method of claim 23, wherein the one or more sensor signals include a time history of sensor data.

39. The method of claim 38, wherein the time history of sensor data is between 10 seconds and 300 seconds.

40. The method of claim 23, further comprising: evaluating whether the estimated PAOP measurement is elevated based upon whether the estimated PAOP measurement exceeds a predetermined PAOP elevated threshold; and displaying, via the display device, an indication of whether the estimated PAOP measurement is elevated.

41. The method of claim 40, wherein the predetermined PAOP threshold is between 12 and 20 millimeters of mercury (mmHg).

42. The method of claim 23, further comprising: receiving an external PAOP waveform and / or external PAOP measurement from a user; and calibrating the estimated PAOP measurement based upon the external PAOP waveform and / or external PAOP measurement from the user.

43. The method of claim 23, further comprising: receiving an internal PAOP waveform and / or internal PAOP measurement; and calibrating the estimated PAOP measurement based upon the internal PAOP waveform and / or internal PAOP measurement.

44. The method of claim 23, further comprising: sterilizing the one or more sensing devices.

45. The system of any of claims 1-22, wherein the computer-readable memory is further encoded with instructions that, when executed by the processor, cause the system to:Attorney Docket Number: P-31218.W001-B0968-P15068W001 generate a control signal for an infusion device to deliver a first drug based on the estimated PAOP measurement.

46. The system of claim 45, further comprising: an infusion device for delivering the first drug, the infusion device configured to receive the control signal.

47. The system of Claims 45 or 46, wherein the first drug is an inert fluid, a vasopressor, or an inotrope.

48. A system for generating an estimated pulmonary artery occlusion pressure (PAOP) measurement on an ongoing basis without occlusion of pulmonary arteries, the system comprising: one or more sensing devices that produce one or more sensor signals; an infusion device containing a first drug; and a hardware unit comprising a processor, computer-readable memory, and a display device, wherein the computer-readable memory is encoded with instructions that, when executed by the processor, cause the system to: receive the one or more sensor signals from the one or more sensing devices; estimate the PAOP measurement based upon the one or more sensor signals to generate the estimated PAOP measurement; and generate a control signal for the infusion device to deliver the first drug based on the estimated PAOP measurement.

49. The system of Claim 48, further comprising any of the features of Claims 2-22.

50. The system of Claim 48, wherein the first drug is an inert fluid, a vasopressor, or an inotrope.

Images