Method for estimating hemodynamic parameters

Abstract

A method and system for deriving one or more hemodynamic parameters based on blood flow velocity and arterial diameter measures, or parameters proportional thereto, each sampled repeatedly or continuously over a period of time to obtain a data series (i.e., waveform) spanning a time window, which is preferably used in combination with at least one other physiological parameter, such as heart rate, to derive the one or more hemodynamic parameters. A transfer function or machine learning model is used to process the inputs to obtain the estimated hemodynamic parameters.

Description

Technical Field

[0001] The present invention relates to a method for non-invasive estimation of hemodynamic parameters.

Background Art

[0002] Patients in the intensive care unit (ICU) and patients in the operating room (OR) require monitoring of various physiological parameters. These include conventional vital sign monitoring such as heart rate (HR), respiratory rate (RR), arterial oxygen saturation (SpO2), body temperature, and sometimes invasive or non-invasive blood pressure (BP) measurement.

[0003] In addition, patients often undergo hemodynamic monitoring. Hemodynamic monitoring attempts to measure the forces that circulate blood within the body. This is, in effect, a performance indicator of the cardiovascular system. Typical central (central) hemodynamic parameters include cardiac output (CO), stroke volume (SV), and the variation in stroke volume.

[0004] Hemodynamic monitoring is required for the early detection, identification, and management of life-threatening clinical conditions such as sepsis and cardiogenic shock, and for evaluating the effectiveness of pharmaceutical interventions such as the administration of vasopressors.

[0005] There are a number of different methods for monitoring hemodynamic parameters. As an example, three clinically prevalent methods for monitoring cardiac output will be described.

[0006] The first example is the thermodilution method. This uses a Swan-Ganz pulmonary artery catheter (PAC). This method is widely believed to be the most clinically accurate method for evaluating central hemodynamics. This method is regarded as the "gold standard" for hemodynamic monitoring of critically ill cardiac patients and is usually the reference measurement against which new technologies are validated and compared. However, nevertheless, the Swan-Ganz approach still shows considerable inaccuracy, such that cardiac output measurements vary by up to 10 - 15%.

[0007] A second exemplary method for cardiac output monitoring that has been developed more recently is the PiCCO (pulse contour cardiac output) system. This is a less invasive alternative to PAC because the placement of the arterial line, which is placed in the axillary artery, brachial artery, femoral artery, or radial artery, is in a less demanding location. The PiCCO approach also requires a central venous catheter. The PiCCO approach is based on continuous cardiac output monitoring using pulse contour analysis. This includes using arterial waveform information to estimate CO and SV, and intermittent transpulmonary thermodilution for calibration purposes. In patients who already have a central arterial line, PiCCO requires only the insertion of an arterial catheter, making it less invasive than the SwanGanz approach. However, the accuracy of PiCCO measurements depends greatly on potential changes in systemic vascular resistance (SVR) and the time interval from the last calibration point (i.e., the time point from the last thermodilution). A rapid change in SVR due to the administration of vasoactive drugs can make PiCCO measurements inaccurate. Also, over time, SVR can change due to gradual autonomic regulation. For these reasons, intermittent calibration of the PiCCO system is required to obtain accurate measurements.

[0008] As a third example, minimally invasive (less invasive) and non-invasive methods for hemodynamic monitoring exist. These include the less invasive FloTrac® system and the non-invasive ClearSight® system. These each rely on pulse contour analysis but omit calibration. Therefore, they are not very trusted by clinicians. Such systems work better for tracking trends in hemodynamic parameters than for obtaining absolute measurements.

[0009] The FloTrac® approach employs an invasive peripheral arterial line, while the Clearsight® approach uses a finger cuff. Usually, the FloTrac® method has better performance than the ClearSight® method.

[0010] The main limitations of more accurate methods for determining hemodynamic parameters include the high invasiveness of the measurement method (which also increases the risk to the patient), and, particularly in the case of PAC, the high level of skill required to insert an arterial line. Furthermore, both the PAC and PiCCO approaches require intermittent calibration to obtain accurate measurements. This becomes a problem when thermodilution is used for calibration, as thermodilution cannot be performed frequently. Some PAC approaches include a heating element for heating a small amount of blood, enabling continuous thermodilution to calibrate the system.

Summary of the Invention

Problems to be Solved by the Invention

[0011] Regarding FloTrac and ClearSight, they are not calibrated and thus have low accuracy. This also means that they are not reliable for predicting the expected changes in SVR for clinically relevant events such as the administration of infusions or drugs.

[0012] Thus, existing hemodynamic monitoring techniques are either highly invasive, lack accuracy, or both. It would be beneficial to identify a non-invasive hemodynamic monitoring method that is more accurate than existing methods.

Means for Solving the Problems

[0013] The present invention is defined by the claims.

[0014] According to an example of one aspect of the present invention, there is provided a computer-implemented method (computer-implemented method) for deriving one or more hemodynamic parameters of a subject, the method comprising: - obtaining a blood flow velocity waveform representing the blood flow velocity at a measurement location of at least one blood vessel over a time window, preferably, the blood flow velocity waveform being at least partially based on Doppler ultrasound data obtained from the measurement location; - Obtaining an arterial diameter waveform representing the diameter of the at least one blood vessel at the measurement position over the time window, or a parameter directly proportional to the diameter, preferably, the diameter waveform is at least partially based on ultrasonic data from the measurement position; - Calculating a predetermined blood flow velocity parameter from the blood flow velocity waveform and calculating a predetermined arterial diameter parameter from the arterial diameter waveform; - Obtaining data representing at least one other predetermined physiological parameter regarding the subject over the time window; - Supplying the blood flow velocity parameter, the arterial diameter parameter, and the at least one other physiological parameter as a group of input parameters to a machine learning model, the machine learning model being configured to process the input parameters to generate an estimated value of one or more hemodynamic parameters as an output; and Generating a data output indicating the estimated one or more hemodynamic parameters output by the machine learning model; comprises.

[0015] Embodiments of the present invention are based on using a combination of ultrasonic acquisition parameters related to blood flow in an arterial branch of a subject to estimate hemodynamic parameters. This is based on research conducted by the present invention, which has found that blood flow velocity waveforms and arterial diameter waveforms acquired and processed as independent signal sources are well correlated with central hemodynamics. This method estimates central hemodynamics by using features related to blood flow parameters measured at the location of an artery (e.g., carotid artery) using a statistical or machine learning-based model. The statistical model or machine learning model embodies a predetermined functional relationship between the input and output hemodynamic parameters. This may be based on a supervised learning or training procedure. This may be based on a regression fitting procedure.

[0016] The above blood flow velocity waveform and arterial diameter waveform are, for example: - receiving data inputs representing the blood flow velocity waveform and the arterial diameter waveform, respectively, from a data store that stores measurement data derived previously (preferably, both waveforms are originally derived from ultrasonic data); or - receiving Doppler ultrasonic data of the at least one blood vessel and processing the Doppler ultrasonic data to derive a blood flow velocity waveform, and receiving ultrasonic imaging data representing the blood vessel at the measurement position and processing the ultrasonic imaging data to derive an arterial diameter; which can be obtained by the above.

[0017] The above ultrasonic data is received from an ultrasonic scanning device or from a data store that stores ultrasonic data acquired previously, for example. In either case, the diameter and velocity waveforms need to correspond to the same simultaneous time window. For example, both are derived from the same ultrasonic data set of the time window, or these two are derived from ultrasonic data sets recorded simultaneously covering the time window.

[0018] The hemodynamic parameter is a central hemodynamic parameter, for example, stroke volume, stroke volume variation or cardiac output.

[0019] The blood vessel is preferably an artery, for example, a peripheral artery. As an example, the blood vessel used can be the carotid artery.

[0020] The blood flow velocity waveform means a waveform of the blood flow velocity as a function of time over a time window. The waveform can be represented, for example, in the form of a data series of blood flow velocity values at intervals over the time window. The diameter waveform means a waveform of the blood vessel diameter or a parameter proportional thereto as a function of time over a time window. This can be represented, for example, in the form of a data series of blood vessel diameter values (or values proportional thereto) at intervals over the time window.

[0021] "Arterial diameter" means the blood vessel diameter or a certain multiple or coefficient thereof, such as the arterial radius, cross-sectional area, or arterial circumference. Generally, this parameter can be any dimensional parameter of the blood vessel cross-section, that is, a parameter indicating the size of the width of the lumen of the blood vessel.

[0022] The blood flow velocity parameter simply means a parameter derived from the blood flow velocity waveform (processing). The arterial diameter parameter simply means a parameter derived from the arterial diameter waveform (e.g., processing). For example, each of these parameters can be a statistical parameter of the corresponding waveform, such as the average value, the area under the waveform (e.g., integration), the decile range, the range, the median, normalization, or any continuous combination of these operations performed on the waveform, for example. These can be called, for example, a first parameter derived / extracted from the blood flow velocity waveform and a second parameter derived / extracted from the arterial diameter waveform, among other things.

[0023] A machine learning algorithm is a self-training algorithm that processes input data to generate or predict output data. Appropriate machine learning algorithms for use in the present invention will be apparent to those skilled in the art. Examples of appropriate machine learning algorithms include linear regression algorithms, decision tree algorithms, and artificial neural networks. Other machine learning algorithms such as logistic regression, support vector machines, or naive Bayesian models are appropriate alternatives.

[0024] In some embodiments, the method may include receiving Doppler ultrasound data at the measurement location of the at least one blood vessel and processing the Doppler ultrasound data to derive a blood flow velocity waveform.

[0025] In some embodiments, the method may include receiving ultrasound data at the measurement location of the at least one blood vessel and processing the ultrasound data to derive an arterial diameter waveform.

[0026] The ultrasonic data described above may be ultrasonic image data. The data may be, for example, B-mode ultrasonic data from which the diameter can be extracted using an automatic segmentation procedure. In other examples, the diameter can be estimated using A-line ultrasonic data.

[0027] The Doppler ultrasonic data may be pulse wave Doppler data.

[0028] The at least one other parameter is a biological parameter of the subject. The parameter may be a physiological parameter. In some embodiments, the step of obtaining the at least one other physiological parameter comprises: receiving ultrasonic data of the at least one blood vessel; and processing the ultrasonic data to derive the at least one predetermined other parameter.

[0029] For example, the at least one predetermined other parameter may be derived from the same ultrasonic data used to derive one or both of the arterial blood flow velocity and the arterial diameter waveform.

[0030] Additionally or alternatively, the step of obtaining the at least one other physiological parameter may include receiving sensor signals from one or more physiological parameter sensors such as a heart rate / pulse rate sensor, a blood oxygen concentration meter, or a blood pressure sensor.

[0031] In some embodiments, the calculated blood flow velocity parameter may have at least one of: the interquartile range of the velocity waveform over the time window; the average value of the blood flow velocity over the time window; the average value of the peak systolic velocity over the time window; the average value of the blood flow velocity over the time window normalized by the number of cardiac cycles spanning the time window; and the integral of the velocity waveform with respect to time over the time window normalized by the number of cardiac cycles spanning the time window.

[0032] As will be described hereinafter in the present disclosure, these parameters have been found to have a particularly good correlation with central hemodynamic parameters by experiments.

[0033] In some embodiments, the arterial diameter parameter may have at least one of: an average value of the arterial diameter over the time window; and an average value of the cross-sectional area of the at least one blood vessel over the time window.

[0034] In some embodiments, the at least one other physiological parameter may include the subject's heart rate and / or a parameter derived from the heart rate. In some embodiments, the method may include receiving Doppler ultrasound data of the at least one blood vessel and processing the Doppler ultrasound data to derive a measure of the subject's heart rate.

[0035] Preferably, the other physiological parameter is derived using the same ultrasound data as that used to derive the blood flow velocity waveform.

[0036] In some embodiments, the at least one other physiological parameter may include a plurality of physiological parameters. In some embodiments, the at least one other physiological parameter may include a set of at least six physiological parameters.

[0037] In some embodiments, the step of obtaining the at least one other physiological parameter may include processing the velocity waveform and the arterial diameter waveform to derive an arterial stroke volume waveform. The step of obtaining the at least one other physiological parameter may further include processing the arterial stroke volume waveform to derive at least one of: an area under the arterial stroke volume waveform over the time window normalized by the number of cardiac cycles spanning the time window; and an average of the arterial stroke volume waveform over the time window normalized by the number of cardiac cycles spanning the time window, optionally.

[0038] In particular, the step of deriving the arterial stroke volume waveform may include a step of processing the velocity waveform to derive the area under the waveform over a single cardiac cycle (e.g., through calculation of the velocity-time integral of the waveform over a single complete cardiac cycle), and a step of processing the diameter waveform to derive a cross-sectional area waveform over the same cardiac cycle. The single stroke volume waveform may be derived based on the area under the velocity waveform and based on the cross-sectional area waveform. For example, the average cross-sectional area of the cardiac cycle is derived and multiplied by the velocity-time integral result.

[0039] In some embodiments, the step of obtaining the at least one other parameter may include: a step of processing the arterial diameter waveform to derive an arterial cross-sectional area waveform for the time window; a step of deriving an arterial blood flow waveform for the time window based on processing of the velocity waveform and the arterial cross-sectional area waveform; and a step of processing the arterial blood flow waveform to derive an average arterial blood flow value over the time window.

[0040] Arterial blood flow refers to the volume of blood per unit time. The arterial flow waveform can be derived as the product of the arterial cross-sectional area waveform and the arterial blood flow velocity waveform.

[0041] In some embodiments, the at least one other physiological parameter may include one or more vital signs, such as heart rate, respiratory rate, and / or blood pressure. The parameter may additionally or alternatively include one or more demographic parameters such as age, gender, and / or body mass index. These can help improve the accuracy in deriving central hemodynamic parameters.

[0042] In some embodiments, the one or more hemodynamic parameters may include one or more of cardiac output, stroke volume, and stroke volume variation.

[0043] Preferably, the aforementioned time window spans at least one cardiac cycle, and more preferably spans a plurality of cardiac cycles.

[0044] In some embodiments, the machine learning algorithm is a multi-parameter linear regression model.

[0045] An example according to another aspect of the present invention is a computer program product. The computer program product has a computer-readable medium, in which a computer-readable code is embodied. The computer-readable code is configured to cause a suitable computer or processor to execute the method according to any example or embodiment described above or below, or according to any claim of the present application when executed by the computer or processor.

[0046] An example according to another aspect of the present invention provides a processing device having an input / output unit; and one or more processors. The one or more processors are configured to: obtain a blood flow velocity waveform representing the blood flow velocity over a time window at a measurement position of at least one blood vessel, preferably, the blood flow velocity waveform is at least partially based on the Doppler ultrasound data at the measurement position; obtain an arterial diameter waveform representing the diameter of the at least one blood vessel or a parameter proportional to the diameter at the measurement position over the time window, preferably, the arterial diameter waveform is at least partially based on the ultrasonic imaging data at the measurement position; calculate a predetermined blood flow velocity parameter from the blood flow velocity waveform and calculate a predetermined arterial diameter parameter from the arterial diameter waveform; obtain data representing the values of at least one other predetermined physiological parameter regarding the subject over the time window; supply the blood flow velocity parameter, the arterial diameter parameter and the at least one other physiological parameter as a group of input parameters to a machine learning model, the machine learning model receiving the group of input parameters and processing the parameters to generate an estimated value of one or more hemodynamic parameters; and generate a data output indicating the estimated one or more hemodynamic parameters output by the machine learning model.

[0047] Examples according to other aspects of the present invention provide a system. The system has a processing device according to any example or embodiment described in the present disclosure or according to any claim of the present application. The system further has an ultrasonic scanning device including at least one transducer unit for acquiring ultrasonic echo signal data of at least one blood vessel of a subject, and a processing unit for processing the echo signal data to derive Doppler ultrasonic data and ultrasonic image data. The input / output unit of the processing device is operably coupled to the output of the ultrasonic scanning device for receiving the Doppler ultrasonic data and the ultrasonic image data.

[0048] The transducer unit may be, for example, an ultrasonic probe.

[0049] The above and other aspects of the present invention will become apparent from the embodiments described later and will be elucidated with reference to these embodiments.

[0050] For a better understanding of the present invention and to more clearly show how the present invention can be implemented, the accompanying drawings will be referred to by way of example only.

Brief Description of the Drawings

[0051]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

[0052] The present invention will be described with reference to the drawings.

[0053] It should be understood that the detailed description and specific examples, while indicating exemplary embodiments of the apparatus, systems and methods, are for the purpose of illustration only and are not intended to limit the scope of the invention. The features, aspects, and advantages of the apparatus, systems, and methods of the present invention will become better understood from the following description, appended claims, and accompanying drawings. It should be understood that the figures are merely schematic and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the figures to indicate the same or similar parts.

[0054] The present invention provides a method for deriving one or more hemodynamic parameters based on blood flow velocity and arterial diameter (arterial size) measurements that are repeatedly or continuously sampled over a period of time to obtain a data series (i.e., waveform) for each time window. The method preferably uses at least one other physiological parameter, such as heart rate, in combination to derive one or more hemodynamic parameters. A transfer function or machine learning model is used to process the inputs to obtain an estimated hemodynamic parameter.

[0055] To better understand the invention, first, a background explanation of the research conducted by the inventor is outlined, which conveys the background of the development of the invention described in the claims.

[0056] This research was based on the hypothesis that arterial flow measurements related to arterial diameter and blood flow velocity are essentially correlated with changes in central hemodynamic parameters such as cardiac output (CO), stroke volume (SV), and stroke volume variation (SVV). The aim was to estimate central hemodynamics by developing an appropriate transfer function using, for example, ultrasonic (US)-derived parameters obtained in the carotid artery. Another established hemodynamic method, including PiCCO and Flotrac (described above), served as a reference standard.

[0057] A reference clinical dataset of previously acquired patient measurements was used. This included 187 ultrasonic measurements in the carotid artery. For each of these, there were hemodynamic reference measurements of PiCCO (n = 140), FloTrac (n = 99), and ClearSight (n = 73) from 15, 9, and 12 patients respectively.

[0058] For each of the above ultrasonic measurements, US B-mode and US pulsed wave Doppler (PWD) data series were available, enabling quasi-continuous measurements of arterial diameter (from B-mode data) and velocity waveforms (from PWD data). Additionally, the velocity waveform could be optionally used to estimate the heart rate on a beat-by-beat basis. A plurality of parameters were derived from the above-described US-based parameters (diameter, velocity waveform, and US-based heart rate).

[0059] A list of exemplary parameters that can be derived from arterial diameter and arterial velocity waveforms is shown in Table 1 below.

[0060] [Table 1] TIFF0007701387000002.tif125170

[0061] The relationships and dependencies between these different parameters are shown in Figure 6. An arrow from the first parameter to the second parameter indicates that the second parameter is at least partially dependent on the first parameter (i.e., the second parameter can be derived at least partially based on the first parameter). All features depend on one or more of the raw US parameters (arterial diameter and velocity waveform).

[0062] During the study, the US measurement data for each data entry typically spanned a period (time window) of about 30 - 60 seconds and thus spanned multiple heartbeats. To minimize the measurement error of the derived parameters, all US-based parameters shown in Table 1 were calculated by averaging the data from one period. For example, the average (or median) can be that of all PSVs within one period, or of a specific landmark derived from the waveform such as the average value of the entire waveform like the average of the velocity waveform of that period. Similarly, the hemodynamic (HDM) reference measurements (PiCCO, FloTrac) are derived based on calculating the average or median of the HDM reference variables (CO, SV, SVV) at each period corresponding to the stored US measurements. The average and median values of the HDM reference variables are equivalent, and all analyses in the study were performed using the median.

[0063] The patient study data was analyzed using a linear multi-parameter transfer function with different ultrasound-derived parameters as inputs and clinical hemodynamic measurements as reference measurements (ground truth). The accuracy of the proposed method was evaluated on a test dataset by calculating the R 2 and RMSE (root mean square error) for CO and SV derived from the transfer function with respect to CO and SV from different clinical reference standards. This was performed for different possible combinations of parameters and for the successful results for each evaluated one.

[0064] As an example, the correlation between carotid flow (quantity per unit time) and PiCCO is relatively low with an R 2 of only 0.22. This result is shown in Figure 1. This means that blood flow itself is not a good parameter for estimating central CO. This finding indicates that a multi-parameter analysis approach is needed to improve the goodness-of-fit of the transfer function for non-invasively estimating central hemodynamics.

[0065] To estimate measures of central hemodynamics, a multi-parameter approach based on linear regression modeling was used. Multiple different combinations of features derived from blood flow-related parameters extracted from ultrasound data were trained or fitted (applied) to the combinations or parameters associated with each transfer function or model, and then tested by evaluating their goodness-of-fit with grand true hemodynamic parameters (obtained using the PiCCO, Flotrac, and / or Clearsight methods).

[0066] The analytical approach for deriving the transfer function is outlined below.

[0067] First, using the reference dataset described above, all available measurements for all patients combined were randomly split into a training set and a test set of equal size. As an example, 140 US measurements where PiCCO served as the reference HDM scale were split into a training set and a test set, each of size 70.

[0068] Second, for the selection of the ultrasonic derivation features (e.g., the selection of those outlined in Table 1), the regression model was fitted once for one of the HDM reference measurements (e.g., CO, SV) using the data corresponding to the training set. The performance was then evaluated for both the training set and the test set. Regression fitting is a well-known technical process, and those skilled in the art will readily recognize methods for implementing it, such as the least squares method, the gradient descent method, etc. The performance metrics included the goodness of fit (R 2 ) of the linear regression and the root mean square error (RMSE) obtained from the correlation scatter plot, as well as the reproducibility coefficient (rpc) from the Bland - Altman plot. The reproducibility coefficient rpc = 1.96×SD, where SD is the standard deviation.

[0069] Third, the second step was repeated 100 times with a random selection of the measurements that each functioned as the training set and the test set (corresponding to double - repeated cross - validation), and these were then used to calculate the mean value and the standard deviation (an estimate of the variance) of the performance metric (e.g., R 2 ).

[0070] Figure 2 outlines the steps of an exemplary method according to one or more embodiments of the present invention. This method can be implemented by a computer. This method is for deriving one or more hemodynamic parameters of a subject.

[0071] The method includes step 12 of acquiring a blood flow velocity waveform representing the blood flow velocity at the measurement position of at least one blood vessel over a time window. The blood flow velocity waveform is based on Doppler ultrasound data acquired from the measurement position.

[0072] In some embodiments, the blood vessel can be a peripheral artery. A peripheral artery means an artery outside the heart and brain, e.g., in the neck, arm, leg, hand, or foot.

[0073] The blood flow velocity waveform can be represented by a data series of blood flow velocity measurement samples at regular time intervals over a certain time window (or period). The blood flow velocity waveform can be received, for example, from a data store or an ultrasonic scanning device, or can be derived based on received Doppler ultrasonic data, such as pulse wave Doppler data. The method is performed in real time together with ultrasonic data collection, or at a later time point based on previously acquired ultrasonic measurement data.

[0074] The method further has step 14 of obtaining an arterial diameter waveform representing the diameter of the at least one blood vessel at the measurement position over the time window. The diameter waveform is based on ultrasonic data from the measurement position. The ultrasonic data can be, for example, B-mode, C-mode or A-line data. The extraction of the diameter measurement value is based on an automatic segmentation algorithm or other image processing algorithms. A person skilled in this field will immediately notice appropriate technical means for extracting the diameter measurement value from ultrasonic data.

[0075] The method further has step 16 of calculating predetermined blood flow velocity parameters from the above blood flow velocity waveform and calculating predetermined arterial diameter parameters from the above arterial diameter waveform.

[0076] The method further has step 18 of obtaining data representing at least one other predetermined physiological parameter of the subject over the time window. The other physiological parameter can be the heart rate of the subject or a parameter derived therefrom. The other physiological parameter can be derived based on one or more of the arterial velocity waveform, arterial diameter waveform and heart rate measurement value.

[0077] The method further has step 20 of supplying the blood flow velocity parameter, the arterial diameter parameter and the at least one other physiological parameter as a set of inputs to a statistical or machine learning model, and the model is trained to process the input parameters to generate an estimated value of one or more hemodynamic parameters as an output.

[0078] The method further includes step 22 of generating a data output indicative of one or more estimated hemodynamic parameters output by the model. In some examples, this data output can be communicated to a user interface, such as a patient monitoring system. The data output can be displayed using a display device. The data output can be communicated to a data storage unit for later retrieval. The data output can be transmitted to a remote computer or remote data store via a network or Internet link.

[0079] There are various options for obtaining the above data.

[0080] In some examples, the method can include receiving Doppler ultrasound data at a measurement location of the at least one blood vessel and processing the Doppler ultrasound data to derive a blood flow velocity waveform. In other examples, the velocity waveform can be received from an external source.

[0081] In some examples, the method can include receiving ultrasound data at a measurement location of the at least one blood vessel and processing the ultrasound data to derive an arterial diameter waveform. In other examples, the diameter waveform can be received from an external source.

[0082] Regarding the at least one other physiological parameter, the method can include receiving the parameter from an external source, such as a measurement sensor or a patient monitoring system or subsystem. The method can include deriving or calculating the parameter from the received sensed data. In some examples, the parameter can be derived from ultrasound data. For example, in some embodiments, the method can include receiving ultrasound data of the at least one blood vessel; and processing the ultrasound data to derive the at least one predetermined other physiological parameter.

[0083] Figure 3 shows an exemplary system 40 according to one aspect of the present invention. This system includes an input / output unit 34 and a processing device 32 having one or more processors 36. According to other aspects of the present invention, the processing device may be provided by the system itself.

[0084] One or more processors 36 of the processing device are configured to execute the steps of the methods according to any of the embodiments outlined above or described in the present disclosure or any of the claims of the present application.

[0085] With respect to system 40, the system may further include an ultrasonic scanning device 50 comprising at least one transducer unit 54 for acquiring ultrasonic echo signal data of at least one blood vessel of a subject, and a processing unit 52 for processing the echo to derive Doppler ultrasonic data and ultrasonic spatial data (e.g., B-mode data or A-line data).

[0086] Figure 4 schematically shows an exemplary pulse wave Doppler (PWD) envelope derived from ultrasonic echo data. In this example, the PWD envelope spans four cardiac cycles.

[0087] Figure 5 schematically shows an exemplary display output of an ultrasonic scanning device that simultaneously generates (dual mode) B-mode and PWD ultrasonic data. In this example, the ultrasonic data represents the carotid artery. The diameter of the artery is automatically extracted by a segmentation algorithm implemented by the ultrasonic scanning device.

[0088] There are various options related to the calculated blood flow velocity parameters, arterial diameter parameters, and other physiological parameters that serve as inputs to the machine learning model.

[0089] Regarding the blood flow velocity parameter and the diameter parameter, what these labels mean is that these parameters are calculated or extracted from the blood flow velocity waveform and the arterial diameter waveform. In other words, these are parameters derived from blood flow velocity and parameters derived from arterial diameter. These may alternatively be referred to as, for example, the first, second, and third parameters.

[0090] Different options for different parameters will be described with reference to Table 1 which lists a number of different parameters.

[0091] Regarding the blood flow velocity parameter, according to a non - limiting group of examples, this is: - The decile range of the velocity waveform over a time window (IDRVelWave in Table 1); - The average value of the blood flow velocity over a time frame (meanVelWave in Table 1); - The average value of the peak systolic velocity over a time frame (PSV in Table 1); - The average value of the blood flow velocity over a time window, normalized by the number of cardiac cycles spanning the time window (meanVelNormPerBeat in Table 1); and - The integral of the velocity waveform with respect to time over a time window, normalized by the number of cardiac cycles spanning the time window (VTINormPerBeat in Table 1); can be any one or more of these.

[0092] In some embodiments, combinations or two or more of these parameters can be used as inputs to the transfer function.

[0093] Regarding the arterial diameter parameter, according to a non - limiting series of examples, this is: - The average value of the arterial diameter over a time window (Dia in Table 1); and - The average value of the cross - sectional area of at least one blood vessel over a time window (CSArea in Table 1); can be one or more of these.

[0094] With respect to said at least one other physiological parameter, this may include, in some examples, the subject's heart rate (HR in Table 1). In some embodiments, the method may include receiving Doppler ultrasound data of at least one blood vessel and processing the Doppler ultrasound data to derive a measure of the subject's heart rate.

[0095] In some examples, said at least one other parameter may include a parameter derived from processing of the stroke volume waveform. For example, the method may include processing the velocity waveform and the arterial diameter waveform to derive an arterial stroke volume waveform. The arterial stroke volume waveform can be calculated by processing the velocity waveform to derive the area under the waveform over a single cardiac cycle (e.g., via calculation of the velocity-time integral of the waveform over a single complete cardiac cycle) and processing the diameter waveform to derive a cross-sectional area waveform over the same cardiac cycle. The stroke volume waveform is derived based on the area under the velocity waveform and based on the cross-sectional area waveform.

[0096] As an example, said at least one other parameter may include the area under the arterial stroke volume waveform over a time window, normalized by the number of cardiac cycles spanning the time window.

[0097] In addition or alternatively, said at least one other parameter may include, optionally, the average of the arterial stroke volume waveform over a time window, normalized by the number of cardiac cycles spanning the time window.

[0098] In some examples, said at least one other parameter may include the mean arterial blood flow value (volume flow per unit time) for a time window. This can be calculated by processing the arterial diameter waveform to derive an arterial cross-sectional area waveform for the time window; deriving an arterial blood flow waveform for the time window based on calculating the product of the velocity waveform and the arterial cross-sectional area waveform; and processing the arterial flow waveform to derive the value of the mean arterial flow over the time window.

[0099] In some examples, the at least one other physiological parameter may include one or more vital signs, such as heart rate, respiratory rate, and / or blood pressure. In some examples, these are obtained from sensor signals received from one or more physiological parameter sensors.

[0100] In some embodiments, the machine learning model can be configured to receive, as other inputs, one or more demographic characteristics of the subject, such as age, gender, and / or body mass index (BMI).

[0101] Table 1 provides a non-limiting set of exemplary parameters, some or all of which may be selected as inputs to be obtained and supplied to a machine learning model for deriving one or more hemodynamic parameters.

[0102] With respect to the one or more hemodynamic parameters derived, these may include, by way of non-limiting example, one or more of cardiac output, stroke volume, and stroke volume variability.

[0103] As described above, the processing of the parameters for deriving one or more hemodynamic parameters is performed using a machine learning model having one or more machine learning algorithms trained to map a predetermined set of input parameters to a set of one or more hemodynamic parameters.

[0104] A machine learning algorithm is any self-training algorithm that processes input data to generate or predict output data. Here, the input data includes pre-selected blood flow velocity parameters, arterial diameter parameters, and other parameters, while the output data includes one or more hemodynamic parameters.

[0105] Appropriate machine learning algorithms employed in the present invention will be apparent to those skilled in the art. Examples of appropriate machine learning algorithms include linear regression algorithms, decision tree algorithms, and artificial neural networks. Other machine learning algorithms such as logistic regression, support vector machines, and naive Bayesian models are also suitable alternatives.

[0106] In the examples shown below, a machine learning algorithm in the form of a multi-parameter linear regression model is used to demonstrate the principle of the concept of the present invention. However, it should also be understood that in each example, the machine learning model can be replaced with different types of machine learning models without affecting the advantageous technical effects.

[0107] Specifically with respect to the multi-parameter linear regression model, the model first constructs a model or algorithm that incorporates each of the desired input parameters as corresponding coefficients or weightings with the (independent variable) parameters of the model, and then trains the model based on a training data set, thereby fitting the model coefficients or weightings to provide the best fit between the input parameters of the training data set and the corresponding output parameters of the training data set, and is thus established. The desired input parameters form the independent variables of the model, while the target hemodynamic parameters are the dependent variables of the model. The model represents that the relevant hemodynamic parameters are estimated as a linear sum (intercept) of constant terms, and each of the dependent variables is multiplied by the corresponding weighting or coefficient.

[0108] The training data set will have training input data entries and corresponding training output data entries. The training input data entries in this case correspond to exemplary values of pre-selected blood flow velocity parameters, arterial diameter parameters, and other physiological parameters. The training output data entries correspond to one or more pre-determined hemodynamic parameters.

[0109] To generate the predicted output data entries, an initialized machine learning algorithm is applied to each input data entry. The error between the predicted output data entry and the corresponding training output data entry is used to modify the machine learning algorithm. This process is repeated until the error converges and the predicted output data entries are sufficiently similar (e.g., ±1%) to the training output data entries. This process is commonly known as a supervised learning technique.

[0110] In the case of a multi-parameter regression model, the training process is a process of fitting the weights / coefficients of the model to the training data set. Once the training or fitting process is completed, the model is deployed using the weights or coefficients obtained in the training or fitting process to map the input parameters (independent variables) to the output hemodynamic parameters.

[0111] The performance or accuracy of the generated machine learning model can be evaluated by running the model against a test data set after training and evaluating the error between the output prediction values generated by the model and the actual ground truth values. For example, in the case of a linear regression model, the performance metrics may include the goodness of fit of the linear regression (R 2 ), the root mean square error (RMSE) obtained from the correlation scatter plot, and the reproducibility coefficient (rpc) obtained from the Bland-Altman plot.

[0112] Here, a plurality of specific examples representing preferred embodiments of the present invention will be described.

[0113] According to one or more embodiments, the parameters used as inputs for deriving hemodynamic parameters are: - Any of the blood flow velocity-related parameters listed in Table 1 above, preferably derived from pulse wave Doppler (PWD) data, such as PSV, VTINormPerBeat, meanVelNormPerBeat, meanVelWave, and / or IDRVelWave; - For example, the average diameter over a time window (Dia in Table 1), derived from B-mode ultrasound data; and - For example, the average heart rate over a time window (HR in Table 1), derived from pulse wave Doppler (PWD) data; are included.

[0114] In a preferred example, the blood flow velocity parameter used is IDRVelWave (the decile range of the velocity waveform over a time window). These three features are derived from different sources and provide complementary information for estimating CO and SV.

[0115] In the test, a group of different multi-parameter linear regression models were generated, each configured to map this series of inputs to one of cardiac output (CO) and stroke volume (SV), and these models were trained using a training data set. Table 2 below shows a summary of the performance statistics of the models for the test data set based on comparison with hemodynamic reference values ("HDM Reference").

[0116] For each model, Table 2 lists the input parameters to which the model was fitted to accept, i.e., the independent parameters of the model. In each case, the output value of the model, i.e., the dependent variable, is the specific hemodynamic parameter listed in the HDM reference entry, i.e., CO or SV.

[0117] In this example, the PiCCO and FloTrac methods served as reference measurements (ground truths) for both the training data set and the test data set. Performance metrics (e.g., R 2 ) were calculated as the mean and standard deviation (SD) of 100 runs of training and test sets randomly selected from the available data sets.

[0118]

Table 2

[0119] R for the test 2 is used as the main performance metrics followed by RMSE and rpc. In particular, the SD of the estimated metrics and the differences between the performance metrics of the training and test sets indicate overfitting of the model. For example, a relatively small decrease in R 2 from the training set to the test set indicates little evidence of overfitting. In other words, the model is relatively robust and will function well for equivalent patient populations.

[0120] Figures 7 (left) and 7 (right) each show a correlation plot and a Bland-Altman plot regarding the performance of the transfer function shown in column 2 of Table 2 when applied to the dataset while PiCCO measurements function as a reference.

[0121] According to the second embodiment, for example, a larger number of input parameters can be used, incorporating several blood flow velocity-derived parameters, arterial diameter parameters, and / or HR-related parameters. These are physiologically preferred features. Compared with the above-described embodiment, more features are incorporated to improve the performance of the model or transfer function. Since the combined parameters provide complementary information, the performance is improved. Further, some of the combined parameters are more robust measures of the underlying signal (blood flow velocity) (e.g., IDRVelWave compared to PSV) and are robust to artifacts in the signal.

[0122] Table 3 below shows a summary of a series of models constructed according to this embodiment, showing performance statistics when applied to a test dataset based on comparison with hemodynamic reference values ("HDM reference"). Each model was constructed to map a series of input parameters to either CO or SV.

[0123] In this example, the PiCCO and FloTrac methods functioned as reference measurements (ground truth) for both the training dataset and the test dataset. Performance metrics (e.g., R 2 ) were calculated as the mean and standard deviation (SD) of 100 runs of training and test sets randomly selected from the available data pool.

[0124]

Table 3

[0125] The labels of the input parameters used in row 4 correspond to those described in Table 1 above.

[0126] By way of example, Figure 8 shows the performance of the model shown in column 1 of Table 3 configured to generate an estimated value of cardiac output (CO) using the PiCCO method metric as the ground truth reference. Regarding the performance of the transfer function, Figure 8 (left) shows a correlation plot and Figure 8 (right) shows a Bland-Altman plot. Figure 9 shows the performance of the model shown in column 2 of Table 3 configured to generate an estimated value of cardiac output (CO) using the FloTrac method metric as the ground truth reference. Figure 9 (left) and Figure 9 (right) show a correlation plot and a Bland-Altman plot regarding the performance of the transfer function.

[0127] Comparing the results of the model in Table 2 with those in Table 3, for example, it can be seen that the PiCCO reference CO transfer function improved in performance (R 2 increased) by incorporating eight features for three. This is quantitative evidence that multiple features extracted from the PWD velocity waveform (e.g., PSV, meanVelWave, and IDRVelWave) can provide complementary information that improves the performance of the machine learning model. Furthermore, the performance of the model is comparable to or better than clinical standards such as FloTrac and ClearSight.

[0128] As described above, the method according to an embodiment of the present invention may have an advantageous use for continuous or ongoing monitoring of a subject's hemodynamics. The method includes generating a data output indicating one or more estimated hemodynamic parameters generated by a model and transmitting the data output to a patient monitoring system, the patient monitoring system comprising a display device and being configured to display a visual representation of the derived one or more hemodynamic parameters on the display device. The patient monitoring system is additionally or alternatively configured to locally or remotely store or cache the derived hemodynamic parameters described above. The system can export the parameters to a remote system such as, for example, a facility network or server.

[0129] According to another aspect of the present invention, a method of providing a machine learning model for deriving one or more hemodynamic parameters is provided. The method includes receiving, as input, a predetermined set of parameters and generating an initial machine learning model configured to process the received parameters to generate an estimated value of one or more hemodynamic parameters. The predetermined input parameters include: a parameter calculated from a blood flow velocity waveform representing the blood flow velocity at a measurement location of at least one blood vessel over a time window; a parameter extracted from an arterial diameter waveform representing the diameter of the at least one blood vessel at the measurement location over the time window; and at least one other predetermined physiological parameter regarding the subject over the time window.

[0130] The method further includes providing a training data set including a plurality of training input data entries and corresponding plurality of training output data entries, each of the training input data entries including a value for each of the set of predetermined input parameters, and each of the training output data entries including a corresponding value for the one or more hemodynamic parameters.

[0131] The method further includes applying the machine learning algorithm to the training input data entries and adjusting internal parameters of the machine learning model to minimize an error between the generated output of the model and the training output data entries.

[0132] As described above, certain embodiments may include an ultrasonic scanning device or means for processing ultrasonic echo data to derive other data.

[0133] As a further more detailed explanation, the general operation of an exemplary ultrasonic system will be described with reference to FIG. 10.

[0134] The system includes an array transducer probe 104 having a transducer array 106 for transmitting ultrasonic waves and receiving echo information. The transducer array 106 has CMUT transducers; piezoelectric transducers formed of materials such as PZT or PVDF; or some other suitable transducer technology. In this example, the transducer array 106 is a two-dimensional array of transducers 108 that can scan either a 2D plane or a three-dimensional volume of the region of interest. In other examples, the transducer array may be a 1D array.

[0135] The transducer array 106 is coupled to a microbeamformer 112 that controls reception of signals by the transducer elements. The microbeamformer performs at least partial beamforming of signals received by sub-arrays (commonly referred to as “groups” or “patches”) of transducers as described in U.S. Pat. Nos. 5,997,479 (Savord et al.), 6,013,032 (Savord), and 6,623,432 (Powers et al.).

[0136] Note that the above microbeamformer is generally completely optional. Further, the system includes a transmit / receive (T / R) switch 116 that couples to the microbeamformer 112 while switching the array between transmit and receive modes to protect the main beamformer 120 from high energy transmit signals when the microbeamformer is not used and the transducer array is operated directly by the main system beamformer. Transmission of ultrasonic beams from the transducer array 106 is directed by the T / R switch 116 and the transducer controller 118 that couples to the microbeamformer and is coupled to a main transmit beamformer (not shown), and the controller can receive input from user operations of a user interface or control panel 138. The controller 118 can include a transmit circuit configured to drive the transducer elements of the array 106 (either directly or via the microbeamformer) during the transmit mode.

[0137] The functions of the control panel 138 in this exemplary system can be readily accomplished by an ultrasonic controller unit according to an embodiment of the present invention.

[0138] In a typical line-by-line imaging sequence, the beamforming system within the probe operates as follows. During transmission, the beamformer (either a microbeamformer or a main system beamformer depending on the implementation) drives the transducer array or a sub-aperture of the transducer array. The sub-aperture can be a one-dimensional line of transducers or a two-dimensional patch of transducers within a larger array. During the transmit mode, focusing and steering of the ultrasonic beam generated by the array or sub-aperture of the array are controlled as described below.

[0139] When a backscattered echo signal is received from a subject, the received signal undergoes a receive beamforming process (described later) to align the received signal. If sub-apertures are used, the sub-apertures are then shifted, for example, by only one transducer element. The shifted sub-aperture is then activated, and this process is repeated until all the transducer elements of the transducer array are activated.

[0140] For each line (or sub-aperture), the overall received signal used to form the relevant line of the final ultrasonic image will be the sum of the voltage signals measured by the transducer elements of a given sub-aperture during the receive period. The resulting line signal obtained according to the following beamforming process is typically called high-frequency (RF) data. Each line signal (RF data set) generated by the various sub-apertures then undergoes additional processing to generate the lines of the final ultrasonic image. The change in the amplitude of the line signal over time contributes to the change in the brightness of the ultrasonic image with depth, and the high-amplitude peaks will correspond to bright pixels (or sets of pixels) in the final image. Peaks that appear near the start of the line signal represent echoes from shallow structures, while peaks that appear increasingly later in the line signal represent echoes from structures at increasing depths within the subject.

[0141] One of the functions controlled by the transducer controller 118 is the direction in which the beam is steered and focused. The beam can be steered straight ahead (orthogonally) from the transducer array or at different angles for a wider field of view. The steering and focusing of the transmit beam can be controlled as a function of the drive times of the transducer elements.

[0142] In general ultrasonic data acquisition, two methods can be distinguished: plane wave imaging and "beam steering" imaging. These two methods are distinguished by the presence of beamforming in the transmit ("beam steering" imaging) and / or receive modes (plane wave imaging and "beam steering" imaging).

[0143] Turning first to the focusing function, by driving all the transducer elements simultaneously, the transducer array generates a plane wave that diverges as it propagates through the subject. In this case, the ultrasonic beam remains unfocused. By introducing a position-dependent time delay in the driving of the transducers, the wavefront of the beam is converged to a desired point called the focus zone. The focus zone is defined as the point where the lateral beam width is less than half of the transmit beam width. In this way, the lateral resolution of the final ultrasonic image is improved.

[0144] For example, when the time delay activates the transducer elements in a sequence starting from the outermost element of the transducer array and ending at the central element(s), a focus zone is formed at a given distance from the probe aligned with the central element(s). The distance of the focus zone from the probe varies depending on the time delay between successive activations of the transducer elements. After the beam passes through the focus zone, the beam begins to diverge and forms a far-field imaging zone. Note that in the case of a focus zone located close to the transducer array, the ultrasonic beam diverges rapidly in the far field, resulting in beam-width artifacts in the final image. Usually, the near field located between the transducer array and the focus zone shows little detail due to the large overlap of the ultrasonic beams. Thus, changing the position of the focus zone can lead to a significant change in the quality of the final image.

[0145] Note that in the transmit mode, only one focus can be defined unless the ultrasonic image is divided into multiple focus zones, each of which can have a different transmit focus.

[0146] Furthermore, when receiving echo signals from within a subject, in order to perform receive focusing, the reverse of the above-described process can be executed. In other words, the incoming signal can be received by the transducer element and can undergo an electronic time delay before being passed to the system for signal processing. The simplest example of this is called delay-and-sum beamforming. It is also possible to dynamically adjust the receive focusing of the transducer array as a function of time.

[0147] Turning now to the function of beam steering, the correct application of time delays to the transducer elements imparts a desired angle to the ultrasonic beam as it exits the transducer array. For example, by activating the transducers on the first side of the transducer array and subsequently activating the remaining transducers in an order that ends on the opposite side of the array, the wavefront of the beam will be tilted towards the second side. The magnitude of the steering angle with respect to the normal of the transducer array depends on the magnitude of the time delay during the activation of subsequent transducer elements.

[0148] Furthermore, it is also possible to focus the steered beam, in which case the total time delay applied to each transducer element is the sum of both the focusing time delay and the steering time delay. In this case, the transducer array is called a phased array.

[0149] In the case of a CMUT transducer that requires a DC bias voltage for activation, the transducer controller 118 can be coupled to control a DC bias control unit 145 for the transducer array. The DC bias control unit 145 sets the DC bias voltage applied to the CMUT transducer elements.

[0150] For each transducer element of the transducer array, an analog ultrasonic signal, typically called channel data, is input to the system by the receiving channel. In the receiving channel, a signal that is partially beamformed from the channel data is generated by the microbeamformer 112 and then passed to the main receiving beamformer 120. In this beamformer, the signals that are partially beamformed from the individual patches of the transducer are combined into a fully beamformed signal called RF data. The beamforming process performed at each stage may be performed as described above or may include additional functions. For example, the main beamformer 120 has 128 channels, each of which receives a signal that is partially beamformed from a patch of dozens or hundreds of transducer elements. In this way, the signals received by the thousands of transducers of the transducer array efficiently contribute to a single beamformed signal.

[0151] The beamformed received signal is coupled to the signal processor 122. The signal processor 122 may process the received echo signal in various ways such as bandpass filtering; decimation (subsampling); I and Q component separation; and harmonic signal separation that enables the separation of linear and non-linear signals and the identification of non-linear (harmonics of the fundamental frequency) echo signals returned from tissue and microbubbles. The signal processor can also perform additional signal enhancements such as speckle reduction, signal synthesis, and noise removal. The bandpass filter in the signal processor can be a tracking filter, and its passband slides from a higher frequency band to a lower frequency band as the depth from which the echo signal is received increases, thereby eliminating noise at higher frequencies from deeper depths that usually lack anatomical information.

[0152] The beamformers for transmission and reception may be implemented with different hardware and may have different functions. Of course, the receiver beamformer is designed taking into account the characteristics of the transmitter beamformer. In FIG. 10, for simplicity, only the receiver beamformers 112, 120 are shown. A complete system would also have a transmission chain with a transmit microbeamformer and a main transmit beamformer.

[0153] The function of the microbeamformer 112 is to provide an initial combination of signals in order to reduce the number of analog signal paths. This is typically performed in the analog domain.

[0154] Final beamforming is performed by the main beamformer 120, typically after digitization.

[0155] The transmit and receive channels use the same transducer array 106 with a fixed frequency band. However, the bandwidth occupied by the transmit pulse can vary depending on the transmit beamforming used. The receive channel can either capture the full transducer bandwidth (this is the classical approach), or use bandpass processing to extract only the bandwidth containing the desired information (e.g., harmonics of the main harmonic).

[0156] The RF signal can then be coupled to a B-mode (i.e., luminance mode, or 2D imaging mode) processor 126 and a Doppler processor 128. The B-mode processor 126 performs amplitude detection on the received ultrasonic signal for imaging of structures within the body such as organ tissue and blood vessels. In the case of line-by-line imaging, each line (beam) is represented by the associated RF signal, and its amplitude is used to generate a luminance value to be assigned to a pixel in the B-mode image. The exact position of a pixel within the image is determined by the position of the associated amplitude measurement along the RF signal and the RF signal line (beam) number. A B-mode image of such a structure can be formed in the harmonic or fundamental image mode, or in both combinations, as described in U.S. Patent No. 6,283,919 (Roundhill et al.) and U.S. Patent No. 6,458,083 (Jago et al.). The Doppler processor 128 processes temporally different signals resulting from tissue movement and blood flow for detection of moving substances such as the flow of blood cells within the image field. The Doppler processor 128 typically includes a wall filter having parameters set to pass or reject echoes returned from selected types of substances within the body.

[0157] The structural and motion signals generated by the B-mode processor and the Doppler processor are coupled to a scan converter 132 and a multi-planar reformatting unit 144. The scan converter 132 arranges the received echo signals of the spatial relationship into a desired image format. In other words, the scan converter functions to convert RF data from a cylindrical coordinate system to a Cartesian coordinate system suitable for displaying an ultrasonic image on the image display 140. In the case of B-mode imaging, the luminance of a pixel at a given coordinate is proportional to the amplitude of the RF signal received from that position. For example, the scan converter can arrange the echo signals in a two-dimensional (2D) sector format, or a pyramidal three-dimensional (3D) image. The scan converter can overlay a color corresponding to the motion at a point within the image field on the B-mode structural image, in which case the Doppler estimated velocity generates a given color. The combined B-mode structural image and color Doppler image depict the motion of tissue and blood flow within the structural image field. The multi-planar reformatting unit converts the echoes received from points within a common plane in a volume region of the body into an ultrasonic image of that plane, as described in U.S. Patent No. 6,443,896 (Detmer). A volume renderer 142 converts the echo signals of a 3D data set into a projected 3D image as seen from a given reference point, as described in U.S. Patent No. 6,530,885 (Entrekin et al.).

[0158] The 2D or 3D image is coupled from the scan converter 132, the multi-planar reformatting unit 144, and the volume renderer 142 to an image processor 130 for further enhancement, buffering, and temporary storage for optional display on the image display 140. The image processor can be adapted to remove specific imaging artifacts from the final ultrasonic image, such as acoustic shadows caused by strong attenuators or reflections; post-enhancement caused by weak attenuators; reverberation artifacts, for example, when a highly reflective tissue interface is located in the immediate vicinity. Further, the image processor can be adapted to handle specific speckle reduction functions to improve the contrast of the final ultrasonic image.

[0159] In addition to being used for imaging, the blood flow values generated by the Doppler processor 128 and the tissue structure information generated by the B-mode processor 126 are coupled to the quantification processor 134. In addition to structural measurements such as the size of the organ and the gestational age, the quantification processor also generates metrics for different flow conditions such as the volumetric velocity of blood flow. The quantification processor receives inputs from the user control panel 138, such as points within the anatomical structure of the image for which measurements are to be taken.

[0160] The output data from the quantification processor is coupled to the graphics processor 136 for the reproduction of measurement graphics and values associated with the images on the display 140, and for the audio output from the display device 140. The graphics processor 136 can also generate graphic overlays for display with the ultrasonic images. These graphic overlays can include standard identification information such as the patient's name, the date and time of the image, and imaging parameters. For these purposes, the graphics processor receives inputs such as the patient's name from the user interface 138. The user interface is also coupled to the transmission controller 118 to control the generation of ultrasonic signals from the transducer array 106, and thus to control the transducer array and the images generated by the ultrasonic system. The transmission control function of the controller 118 is just one of the functions performed. The controller 118 also takes into account the operating mode (given by the user), as well as the corresponding required transmission configuration and the band-pass configuration in the receiver analog / digital converters. The controller 118 can be a state machine having a fixed state.

[0161] The user interface is also coupled to the multi-planar re-formatter 144 for the selection and control of the planes of a plurality of multi-planar re-formatted (MPR) images that can be used to perform measurements quantified in the image fields of the MPR images.

[0162] The above-described ultrasonic system can be operably coupled to the aforementioned processing device 32. The processing device can receive Doppler ultrasonic data and spatial ultrasonic data (e.g., B-mode) from the ultrasonic system. For example, in some instances, the ultrasonic system can be used to implement the ultrasonic sensing device 50 of the system 40 shown in FIG. 3.

[0163] The embodiments of the present invention described above use a processing device. The processing device can generally have a single processor or multiple processors. The processing device can be arranged within a single housing, structure, or unit, or can be distributed among multiple different devices, structures, or units. Thus, a reference to a processing configuration adapted or configured to perform a particular step or task can correspond to the step or task being performed alone or in combination by any one or more of a plurality of processing components. One of ordinary skill in the art will understand how to implement such a distributed processing configuration. The processing device includes a communication module or input / output section for receiving data and outputting data to other components.

[0164] One or more processors of the processing device can be configured in various ways using software and / or hardware to perform the various functions required. A processor typically uses one or more microprocessors that can be programmed using software (e.g., microcode) to perform the necessary functions. The processor can be implemented as a combination of dedicated hardware for performing some functions and one or more programmed microprocessors and associated circuitry for performing other functions.

[0165] Examples of circuits that can be used in various embodiments of the present disclosure include, but are not limited to, conventional microprocessors, application-specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs).

[0166] In various configurations, the processor can be associated with one or more storage media such as volatile and non-volatile computer memories like RAM, PROM, EPROM, and EEPROM. The storage media can be encoded with one or more programs that perform the required functions when executed on one or more processors and / or controllers. The various storage media can be fixed within the processor or controller or be portable such that the one or more stored programs can be loaded into the processor.

[0167] Variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention by considering the drawings, disclosure, and appended claims. In the claims, the term "comprising" does not exclude other elements or steps, and the singular does not exclude the plural.

[0168] A single processor or other unit can perform the functions of several items recited in the claims. The mere fact that certain means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be used advantageously. It should be noted that when the term "adapted" is used in the claims or description, it is intended to be equivalent to the term "configured".

[0169] The computer program can be stored / distributed not only by a suitable medium such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but also in other forms such as via the Internet or other wired or wireless communication systems.

[0170] Reference signs in the claims should not be construed as limiting the scope.

Claims

1. A computer-implemented method for deriving one or more central hemodynamic parameters of a subject, the computer-implemented method comprising: Obtaining a blood flow velocity waveform representing the blood flow velocity at a measurement location of at least one blood vessel over a time window, the blood flow velocity waveform being at least partially based on Doppler ultrasound data obtained from the measurement location; Obtaining an arterial diameter waveform representing the diameter of the at least one blood vessel or a parameter proportional to the diameter at the measurement location over the time window, the arterial diameter waveform being at least partially based on ultrasound data from the measurement location; Calculating a predetermined blood flow velocity parameter from the blood flow velocity waveform and calculating a predetermined arterial diameter parameter from the arterial diameter waveform; Obtaining data representing at least one other predetermined physiological parameter of the subject over the time window; Supplying the blood flow velocity parameter, the arterial diameter parameter, and the at least one other predetermined physiological parameter as a group of input parameters to a machine learning model, the machine learning model processing the input parameters to generate an estimated value of one or more central hemodynamic parameters as an output, and Generating a data output indicating the estimated one or more central hemodynamic parameters output by the machine learning model comprising: The one or more central hemodynamic parameters include one or more of cardiac output, stroke volume, and stroke volume variability; The computer-implemented method, wherein the at least one other predetermined physiological parameter includes one or more of vital signs of heart rate, respiratory rate, and blood pressure.

2. Receiving Doppler ultrasound data at the measurement location of the at least one blood vessel and processing the Doppler ultrasound data to derive the blood flow velocity waveform, and Receiving ultrasound data at the measurement location of the at least one blood vessel and processing the ultrasound data to derive the arterial diameter waveform The computer-implemented method according to claim 1, comprising:

3. The step of obtaining the at least one other physiological parameter comprises: Receiving ultrasound data of the at least one blood vessel, and The step of processing the ultrasonic data to derive the at least one other predetermined physiological parameter The computer-implemented method according to claim 1 or claim 2, having this step

4. The blood flow velocity parameter is The decile range of the blood flow velocity waveform over the time window The average value of the blood flow velocity over the time window The average value of the peak systolic velocity over the time window The average value of the blood flow velocity over the time window, normalized by the number of cardiac cycles spanning the time window, and The integral of the blood flow velocity waveform with respect to time over the time window, normalized by the number of cardiac cycles spanning the time window The computer-implemented method according to any one of claims 1 to 3, having at least one of the above

5. The arterial diameter parameter is The average value of the arterial diameter over the time window, and The average value of the cross-sectional area of the at least one blood vessel over the time window The computer-implemented method according to any one of claims 1 to 4, having at least one of the above

6. The at least one other predetermined physiological parameter includes the heart rate of the subject and / or a parameter derived from the heart rate. Optionally, the computer-implemented method includes receiving Doppler ultrasonic data of the at least one blood vessel and processing the Doppler ultrasonic data to derive a measure of the heart rate of the subject The computer-implemented method according to any one of claims 1 to 5, having this step The computer-implemented method according to any one of claims 1 to 5, having this step

7. The step of obtaining the at least one other physiological parameter includes The step of processing the blood flow velocity waveform and the arterial diameter waveform to derive an arterial stroke volume waveform, and Processing the arterial stroke volume waveform to Derive at least one of the area under the arterial stroke volume waveform over the time window, normalized by the number of cardiac cycles spanning the time window, and Optionally, the average of the arterial stroke volume waveform over the time window, normalized by the number of cardiac cycles spanning the time window The step of deriving at least one of the above The computer-implemented method according to any one of claims 1 to 6, having this step

8. The step of obtaining the at least one other parameter includes The step of processing the arterial diameter waveform to derive the arterial cross-sectional area waveform of the time window The step of deriving the arterial blood flow volume waveform of the time window based on the processing of the blood flow velocity waveform and the arterial cross-sectional area waveform, and The step of processing the arterial blood flow rate waveform to derive an average arterial blood flow rate value over the time window The computer-implemented method according to any one of claims 1 to 7, comprising this step

9. The computer-implemented method according to any one of claims 1 to 8, wherein the time window spans at least one cardiac cycle

10. The computer-implemented method according to any one of claims 1 to 9, wherein the machine learning model is a multi-parameter linear regression model

11. A computer-readable medium in which a computer-readable code is embodied, and when the computer-readable code is executed by a computer or a processor, the computer or the processor is caused to execute the computer-implemented method according to any one of claims 1 to 10

12. An input / output unit, and One or more processors A processing device having, wherein the one or more processors Acquire a blood flow velocity waveform representing the blood flow velocity over a time window at a measurement position of at least one blood vessel, wherein the blood flow velocity waveform is at least partially based on Doppler ultrasound data acquired from the measurement position Acquire an arterial diameter waveform representing the diameter of the at least one blood vessel or a parameter proportional to the diameter at the measurement position over the time window, wherein the arterial diameter waveform is at least partially based on ultrasonic imaging data of the measurement position Calculate a predetermined blood flow velocity parameter from the blood flow velocity waveform and calculate a predetermined arterial diameter parameter from the arterial diameter waveform Acquire data representing at least one other predetermined physiological parameter regarding the subject over the time window Supply the blood flow velocity parameter, the arterial diameter parameter, and the at least one other physiological parameter as a group of input parameters to a machine learning model, wherein the machine learning model receives the group of input parameters and processes the parameters to generate an estimated value of one or more central hemodynamic parameters, and Generate a data output indicating the estimated one or more central hemodynamic parameters output by the machine learning model The one or more central hemodynamic parameters include one or more of cardiac output, stroke volume, and stroke volume variation A processing device, wherein the at least one predetermined other physiological parameter includes one or more of vital signs of heart rate, respiratory rate, and blood pressure. **Claim 13** The processing device according to claim 12, and at least one transducer unit for acquiring ultrasonic echo signal data of at least one blood vessel of a subject, and a processing unit for processing the ultrasonic echo signal data to derive Doppler ultrasonic data and ultrasonic image data, A system having wherein the input / output unit of the processing device is operably coupled to an output of the ultrasonic scan device to receive the Doppler ultrasonic data and the ultrasonic image data. System.

Images