Unsupervised real-time classification of arterial blood pressure signals

Abstract

A hemodynamic monitor is configured to receive a hemodynamic sensor signal representing an arterial blood pressure (ABP) of a patient. The hemodynamic monitor separates the received hemodynamic sensor signal into a plurality of heartbeat portions, each heartbeat portion representing the patient's ABP for one of a plurality of individual heartbeats. For each heartbeat portion, the hemodynamic monitor determines a set of coefficients representing frequency components of the corresponding heartbeat portion to produce a plurality of coefficient sets. Each coefficient set is normalized. A reference coefficient set is determined based on the normalized coefficient set. A quality indicator associated with the individual heartbeat is provided based on a comparison of the normalized coefficient set for the individual heartbeat with the reference coefficient set. The quality indicator is used to generate a modified hemodynamic sensor signal from which a hemodynamic parameter is derived.

Description

Technical Field

[0001] The present disclosure relates generally to arterial blood pressure monitoring and, more particularly, to classification of sensed arterial blood pressure signals. Background Art

[0002] Biomedical signal analysis techniques often utilize methods to detect and distinguish abnormal or noisy signals from normal (or expected) signals to reduce output errors. For example, arterial blood pressure (ABP) waveform signals are often used to determine hemodynamic parameters such as cardiac output (CO), vascular resistance, stroke volume, or other hemodynamic parameters that can be used to monitor and / or predict important physiological events (such as hypotension). Therefore, the presence of noise or other error artifacts in the sensed ABP waveform signal can adversely affect the accuracy and reliability of downstream hemodynamic processing operations.

[0003] Such ABP waveforms, while exhibiting quasi-periodic cyclic behavior, typically exhibit variations in one or more of amplitude, mean (skew), and period (stretch) from beat to beat due to natural and expected physiological effects. Therefore, direct comparison of the beat-to-beat portion of the ABP waveform to identify noise in the signal is complicated by naturally occurring variations in the ABP waveform. Summary of the Invention

[0004] In one example, a hemodynamic monitor includes a sensor interface, a heartbeat detection module, a model parameter module, a heartbeat classification module, and a hemodynamic processing module. The sensor interface receives a hemodynamic sensor signal from a hemodynamic sensor. The hemodynamic sensor signal represents a patient's arterial blood pressure (ABP). The heartbeat detection module separates the received hemodynamic sensor signal into a plurality of heartbeat components. Each heartbeat component represents the patient's ABP for one of a plurality of individual heartbeats of the patient. The model parameter module determines, for each of the plurality of heartbeat components, a coefficient set representing a frequency component of the corresponding heartbeat component to generate a plurality of coefficient sets. Each coefficient set includes the same number of coefficients. The model parameter module further normalizes each coefficient set to generate a plurality of normalized coefficient sets. The heartbeat classification module determines a reference coefficient set based on the plurality of normalized coefficient sets, and provides a quality indicator associated with the individual heartbeat based on a comparison of the normalized coefficient set for the individual heartbeat with the reference coefficient set. The hemodynamic processing module uses the quality indicator to generate a modified hemodynamic sensor signal, derives one or more hemodynamic parameters from the modified hemodynamic sensor signal, and outputs the one or more derived hemodynamic parameters.

[0005] In another example, a system includes a hemodynamic sensor and a hemodynamic monitor connected to the hemodynamic sensor. The hemodynamic sensor is configured to sense a patient's arterial blood pressure (ABP). The hemodynamic monitor includes a sensor interface, one or more processors, and a computer-readable memory. The sensor interface is configured to receive a hemodynamic sensor signal from the hemodynamic sensor, the hemodynamic sensor signal representing the patient's ABP sensed by the hemodynamic sensor. The computer-readable memory is encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to separate the received hemodynamic sensor signal into a plurality of heartbeat portions, each heartbeat portion representing the patient's ABP for one of a plurality of individual heartbeats of the patient. The computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to determine, for each of the plurality of heartbeat portions, a set of coefficients representing frequency components of the corresponding heartbeat portion to generate a plurality of coefficient sets. Each coefficient set includes the same number of coefficients. The computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to normalize each coefficient set to generate a plurality of normalized coefficient sets, determine a reference coefficient set based on the plurality of normalized coefficient sets, compare the normalized coefficient set for the individual heartbeat with the reference coefficient set, and determine a quality indicator associated with the individual heartbeat based on the comparison. The computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to generate a modified hemodynamic sensor signal using the quality indicator, derive one or more hemodynamic parameters from the modified hemodynamic sensor signal, and output the one or more derived hemodynamic parameters.

[0006] In another example, a method includes generating an analog hemodynamic sensor signal representing an arterial blood pressure (ABP) of a patient using a hemodynamic sensor, sampling the analog hemodynamic sensor signal at a defined sampling rate to generate a sampled hemodynamic sensor signal representing the patient's ABP, and separating the sampled hemodynamic sensor signal into a plurality of heartbeat portions, each heartbeat portion representing the patient's ABP for one of a plurality of individual heartbeats of the patient. The method further includes determining, for each of the plurality of heartbeat portions, a set of coefficients representing frequency components of the corresponding heartbeat portion to generate a plurality of coefficient sets. Each coefficient set includes the same number of coefficients. The method further includes normalizing each coefficient set to generate a plurality of normalized coefficient sets, determining a reference coefficient set based on the plurality of normalized coefficient sets, comparing the normalized coefficient set for the individual heartbeat with the reference coefficient set, and determining a quality indicator associated with the individual heartbeat based on the comparison. The method further includes generating a modified hemodynamic sensor signal using the quality indicator, deriving one or more hemodynamic parameters from the modified hemodynamic sensor signal, and outputting the derived hemodynamic parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 is a block diagram illustrating an example hemodynamic monitoring system that implements an unsupervised real-time classification algorithm for classifying arterial blood pressure waveforms received from a physiological sensor attached to a patient.

[0008] Figure 2 is a block diagram showing further details of a hemodynamic monitor for processing an arterial blood pressure waveform and providing quality indicators of individual heartbeat portions.

[0009] Figure 3 is a graph illustrating example arterial blood pressure waveform information that may be separated into individual heartbeat components.

[0010] Figure 4 is a flow chart illustrating example operation of a hemodynamic monitor to determine a set of frequency component coefficients representing the shape of an individual heartbeat portion of arterial blood pressure waveform information.

[0011] Figure 5 is a flow chart illustrating example operation of a hemodynamic monitor to classify individual heartbeat portions of an arterial blood pressure waveform to provide quality indicators of the individual heartbeat portions.

[0012] Figure 6 is a flow chart illustrating example operation of a hemodynamic monitor during an initial learning period of an unsupervised real-time classification algorithm. DETAILED DESCRIPTION

[0013] As described herein, a hemodynamic monitoring system implements an unsupervised real-time classification algorithm that classifies individual heartbeat portions of a sensed arterial blood pressure (ABP) waveform to provide a quality indicator for the associated heartbeat portion. The classification algorithm generates a set of normalized frequency component coefficients on a beat-by-beat basis that represent the shape of each individual heartbeat portion, independent of the scaling, mean (offset), and period (stretch) of the corresponding heartbeat portion. The set of normalized frequency component coefficients for a newly sensed individual heartbeat portion is compared with a reference set of coefficients developed by the classification algorithm. A quality indicator for each heartbeat portion is provided for use in downstream processing of the ABP waveform. In certain examples, the quality indicator classifies each individual heartbeat into one of two categories: a category eligible for downstream hemodynamic processing and a category ineligible for downstream processing. Thus, hemodynamic parameters (e.g., cardiac output, vascular resistance, stroke volume, or other parameters) are derived from those heartbeat portions classified as eligible for downstream processing. Portions of the ABP waveform classified as ineligible for downstream processing are ignored or otherwise excluded from hemodynamic processing operations. Thus, a hemodynamic monitoring system implementing the techniques of this disclosure may increase the accuracy and reliability of hemodynamic parameters derived from a patient's sensed ABP waveform.

[0014] Figure 1 is a block diagram of a hemodynamic monitoring system 10 including a hemodynamic monitor 12 that implements an unsupervised real-time classification algorithm for processing and classifying arterial blood pressure (ABP) waveforms received from physiological sensors 14 and / or 16 attached to a patient 18. Figure 1 As shown, the system 10 may further involve client devices 20A-20N that are communicatively connected to the hemodynamic monitor 12 via a communication network 22. The hemodynamic monitor 12 includes a display 24, processor(s) 26, a computer-readable memory 28, input element(s) 30, communication interface(s) 32, sensor interface(s) 34, an arterial blood pressure (ABP) waveform processing module 36, and a hemodynamic processing module 38.

[0015] like Figure 1 As shown, physiological sensors 14 and 16 can be attached to a patient 18 to sense the ABP waveform of the patient 18 and transmit the sensed ABP waveform signal to the hemodynamic monitor 12. The physiological sensor 14 can include, for example, one or more cuffs (e.g., one or more finger cuffs) or other peripheral arterial pressure sensors that can be used to sense peripheral arterial blood pressure. For example, the physiological sensor 14 can perform real-time finger pressure measurement using a volume clamp method at a sampling rate of, for example, 1000 Hz to provide the sensed ABP waveform to the hemodynamic monitor 12.

[0016] Physiological sensor 16 may be, for example, a pulmonary artery catheter (PAC), such as a Swan-Ganz catheter. Such a catheter may be inserted into the pulmonary artery of patient 18 to detect direct, simultaneous measurements of pressure in the right atrium, right ventricle, pulmonary artery, and filling pressure of the left atrium of patient 18 using a hot wire located on the catheter and the principle of thermodilution. Figure 1 The example of FIG1 shows two physiological sensors 14 and 16 sensing ABP waveform information of the patient 18, but it should be understood that in other examples, a single physiological sensor (i.e., one of the physiological sensors 14 and 16) can be used to sense the ABP waveform and provide the waveform information to the hemodynamic monitor 12. In still other examples, more than two physiological sensors can be used to sense and provide ABP waveform information.

[0017] like Figure 1 As shown, the hemodynamic monitor 12 may include one or more sensor interfaces 34 that communicate with the physiological sensors 14 and 16 to receive information from one or more of the physiological sensors 14 and 16 (and, in some examples, transmit information to one or more of the physiological sensors 14 and 16). The sensor interface 34 may communicate with the physiological sensors 14 and 16 via a wired or wireless connection, or both. The physiological sensors 14 and 16 may generate analog hemodynamic sensor signals that represent the ABP waveform of the patient 18. In some examples, the sensor interface 34 may include an analog-to-digital converter or other equivalent discrete or integrated logic circuitry to sample the analog sensor signals at a defined (e.g., consistent) rate (such as one hundred hertz or other sampling rate) to generate discrete-time ABP waveform information.

[0018] like Figure 1As shown, the hemodynamic monitor 12 may further include a display 24, one or more processors 26, a computer-readable memory 28, and an input element 30. The display 24 may be an electronic visual display presenting a graphical user interface for displaying information representing measurement data received from the physiological sensors 14 and 16. This information may take the form of, for example, waveforms, numerical indications, categorical indications, alarms, or other information. The input element 30 may include various physical and / or graphical control elements, such as physical and / or virtual buttons, knobs, sliders, or other control elements that can be manipulated (e.g., by a user) to affect the operation of the hemodynamic monitor 12. For example, the input element 30 may include physical and / or graphical control elements that enable a user to interact with the hemodynamic monitor 12 to select information graphically presented at the display 24, to configure operating parameters of the hemodynamic monitor 12 (or other components of the system 10), to acknowledge alarms generated by the hemodynamic monitor 12, or to otherwise affect the operation of components of the system 10. In some examples, display 24 may be a touch-sensitive display that presents one or more graphical control elements of input element 30 and enables user interaction in the form of gesture input (e.g., touch gestures, swipe gestures, pinch gestures, or other gesture inputs).

[0019] In one example, the one or more processors 26 are configured to implement functions and / or processing instructions for execution within the hemodynamic monitor 12. For example, the one or more processors 26 may be capable of processing instructions stored in the computer-readable memory 28. Examples of the one or more processors 26 may include any one or more of a microprocessor, a controller, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other equivalent discrete or integrated logic circuitry.

[0020] The computer-readable memory 28 can be configured to store information within the hemodynamic monitor 12 during operation. In some examples, the computer-readable memory 28 is described as a computer-readable storage medium. In some examples, the computer-readable storage medium may include a non-transitory medium. The term "non-transitory" may indicate that the storage medium is not embodied in a carrier wave or propagating signal. In some examples, a non-transitory storage medium may store data that may change over time (e.g., in RAM or cache). In some examples, the computer-readable memory 28 is described as a volatile memory, meaning that the computer-readable memory 28 does not retain the stored contents when the power to the hemodynamic monitor 12 is turned off. Examples of volatile memory may include random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), and other forms of volatile memory. In some examples, the computer-readable memory 28 is used to store program instructions executed by one or more processors 26. In one example, the computer-readable memory 28 is used by software or applications running on the hemodynamic monitor 12 (e.g., software programs that implement various aspects of the ABP waveform processing module 36 and / or the hemodynamic processing module 38) to temporarily store information during program execution. In some examples, the computer-readable memory 28 also includes a non-volatile storage element. Examples of such non-volatile storage elements can include, but are not limited to, a magnetic hard drive, an optical disk, flash memory, or an electrically programmable programmable memory (EPROM) or an electrically erasable programmable programmable memory (EEPROM).

[0021] like Figure 1 As shown, the hemodynamic monitor 12 may include one or more communication interfaces 32 that enable direct or indirect communication with one or more remote client computing systems, such as client devices 20A-20N communicatively coupled to the hemodynamic monitor 12 via a wired and / or wireless communication network 22. For example, the client devices 20A-20N, which may include any number of client devices, may access information generated by the hemodynamic monitor 12 to remotely monitor hemodynamic parameters of the patient 18, transmit commands or other information to the hemodynamic monitor 12, or otherwise interact with components of the system 10.

[0022] like Figure 1 As shown, the sensor interface 34 provides ABP waveform information (ie, analog and / or discrete time waveforms) to the ABP waveform processing module 36. In some examples, such as Figure 1 In the illustrated example, the sensor interface 34 also provides the ABP waveform information to the hemodynamic processing module 38 , although in other examples, the ABP waveform information may be provided to the hemodynamic processing module 38 via the ABP waveform processing module 36 .

[0023] The sampled ABP waveform information received by the ABP waveform processing module 36 via the sensor interface 34 represents the sensed ABP waveform of the patient 18 at discrete times (i.e., sampled at a consistent rate such as 100 Hz). Such an ABP waveform can be considered quasi-periodic because the waveform exhibits clearly delineated cyclic behavior, but with beat-to-beat variations in shape and period. Some variation in the amplitude and period modulation of the ABP waveform is considered normal and is caused by physiological factors of the human body, such as respiratory modulation and baroreflex regulation. For example, abnormal variations in the amplitude and period of the ABP waveform can be introduced by: mechanical coupling of the physiological sensors 14 and / or 16 due to patient movement, surgical tube dynamics, and vascular clamping; electrical interference from electrocautery instruments or nonlinear distortion of analog circuits; and physiological dysfunctions such as ectopic beats, irregular heart patterns, or atrial / ventricular fibrillation. Such abnormal variations in the ABP waveform signal can represent noise artifacts that negatively impact downstream processing of the ABP waveform for hemodynamic monitoring and parameter estimation.

[0024] As further described below, the ABP waveform processing module 36 implements an unsupervised real-time classification algorithm that provides a quality indicator associated with an individual heartbeat that is used by the hemodynamic processing module 38 for determining and monitoring hemodynamic parameters such as cardiac output (CO), vascular resistance, stroke volume, or other hemodynamic parameters. In some examples, the quality indicator for each individual heartbeat classifies the heartbeat as eligible for downstream processing or ineligible for downstream processing. In other examples, the quality indicator provides a quantitative indication of the extent to which the individual heartbeat deviates from a representation of a reference heartbeat of the patient 18 developed by the unsupervised classification algorithm.

[0025] As described herein, the ABP waveform processing module 36 separates the received ABP waveform signal into heartbeat segments, each segment representing the ABP of the patient 18 for an individual heartbeat. For each individual heartbeat segment, the ABP waveform processing module 36 determines a set of coefficients representing the frequency components of the individual heartbeat segment, such as by using a finite Fourier series expansion of a discrete-time representation of the individual heartbeat segment. The ABP waveform processing module 36 generates the same number of coefficients for each heartbeat segment, regardless of the period of the individual heartbeat segment. Thus, each frequency coefficient set is a non-uniform discrete-time vector parameter sequence that represents the shape of the individual heartbeat segment and is independent of the period of the individual heartbeat segment. The ABP waveform processing module 36 further normalizes each frequency component coefficient set to generate a normalized coefficient set for each individual heartbeat segment that represents the shape of the heartbeat segment independent of the scaling, averaging (offset), and period (stretch) of the corresponding heartbeat segment.

[0026] The normalized frequency coefficient sets for the individual heartbeat portions are provided as input to an unsupervised real-time clustering algorithm implemented by the ABP waveform processing module 36, which uses multiple normalized coefficient sets (i.e., associated with multiple individual heartbeat portions) to determine a reference coefficient set. The reference coefficient set represents a central reference for a "normal" clustering group or class of normalized coefficient sets (i.e., representing a typical or other expected heartbeat for a patient).

[0027] The normalized coefficients for the newly sensed individual heartbeat portion are compared to the reference set of coefficients to generate a quality indicator for the newly sensed individual heartbeat portion. In some examples, the quality indicator may classify the newly sensed individual heartbeat portion as eligible for downstream processing by the hemodynamic processing module 38 or as ineligible for downstream processing. In other examples, the quality indicator may provide a quantitative measure of the deviation of the newly sensed individual heartbeat portion from the reference set of coefficients (i.e., a central reference cluster group or category).

[0028] The ABP waveform processing module 36 provides a quality indicator (e.g., a classification) associated with each individual heartbeat portion to the hemodynamic processing module 38. The hemodynamic processing module 38 utilizes the quality indicator to determine and / or monitor hemodynamic parameters based on the ABP waveform signal. For example, the hemodynamic processing module 38 may use the quality indicator to generate a modified hemodynamic sensor signal from which one or more hemodynamic parameters are derived and output for, e.g., display or further analysis operations. For example, in some examples, the hemodynamic processing module 38 may generate a modified hemodynamic sensor signal to include individual heartbeat portions that are indicated as eligible for downstream processing. The hemodynamic processing module 38 may generate a modified hemodynamic sensor signal to exclude individual heartbeat portions that are indicated as ineligible for downstream processing, thereby reducing the presence of noise or other error artifacts within the ABP signal used for hemodynamic parameter monitoring and / or processing.

[0029] Thus, a hemodynamic monitor 12 implementing the techniques described herein can provide quality indicators for individual beat portions of a patient's 18 sensed ABP signal. The quality indicators can be used to classify individual beat portions of the sensed ABP waveform as eligible or ineligible for downstream processing, thereby increasing the accuracy and reliability of hemodynamic parameters derived from the ABP waveform signal. The classification algorithm utilizes frequency parameters representing the shape of the ABP waveform that are independent of the scaling, averaging, and period of each individual beat portion of the ABP waveform, thereby increasing the consistency of classification and reducing the operational complexity associated with time-domain analysis of individual beat portions associated with naturally occurring variations in the cycle of an individual beat. Furthermore, the unsupervised nature of the clustering algorithm enables efficient real-time classification of individual beats based on the patient's previously sensed ABP waveform, without requiring extensive reference data that may be difficult to obtain and may not be representative of a particular patient's typical ABP waveform. Thus, the techniques described herein can increase the accuracy and reliability of hemodynamic parameters derived from a patient's sensed ABP waveform.

[0030] Figure 2 is a block diagram illustrating further details of the ABP waveform processing module 36 providing quality indicators of individual heartbeat portions of the sensed ABP waveform. Figure 2 As shown, the ABP waveform processing module 36 includes a heartbeat detection module 40 , a model parameter module 42 and a heartbeat classification module 44 .

[0031] The heartbeat detection module 40 receives the physiological sensors 14 and 16 ( Figure 1 ) as input. In some examples, the received ABP waveform signal may be a signal received by the sensor interface 34 ( Figure 1 ) is a discrete-time signal sampled at a consistent rate (e.g., one hundred hertz). In other examples, the received ABP waveform signal may be an analog ABP waveform signal, in which case the heartbeat detection module 40 may sample the received analog signal to generate a discrete-time ABP waveform.

[0032] The heartbeat detection module 40 identifies heartbeat portions of the received ABP waveform signal, each heartbeat portion representing a heartbeat of the patient 18 ( Figure 1) of the ABP waveform. For example, the heartbeat detection module 40 may identify individual heartbeat portions of the received ABP waveform based on the maximum derivative of the ABP waveform signal or (one or more) other features that identify a portion of the received ABP waveform signal corresponding to an individual heartbeat from the beginning of the systolic rise to the end of the diastolic rise. Various detection algorithms for identifying individual heartbeats within an ABP waveform are known in the art, and generally, the heartbeat detection module 40 may utilize any detection algorithm suitable for identifying individual heartbeat portions of a received ABP waveform signal.

[0033] The heartbeat detection module 40 separates the received ABP waveform signal into individual heartbeat portions, which are provided to the model parameter module 42 along with the discrete-time ABP waveform. For example, the heartbeat detection module 40 may identify, for each individual heartbeat portion, a start index and an end index of the discrete-time ABP waveform signal. Such indices (i.e., the start index and the end index) may take the form of a unique ordered index of the discrete-time ABP waveform that identifies the temporal sequence of sample values ​​within the discrete-time ABP waveform, an absolute time value associated with each start index and end index, a relative time value associated with each start index and end index, or other indications that identify an individual heartbeat portion within the discrete-time ABP waveform.

[0034] Model parameter module 42 receives from heartbeat detection module 40 an indication of the discrete-time ABP waveform and the location of individual heartbeat segments within the discrete-time ABP waveform. As described further below, model parameter module 42 determines, for each individual heartbeat segment, a set of coefficients representing the frequency components of the individual heartbeat segment. For example, the set of coefficients may be determined using a finite Fourier series expansion that produces frequency coefficients representing the shape of the individual heartbeat segment. Rather than performing the Fourier series expansion with a constant period, model parameter module 42 determines the frequency coefficients based on the period of the individual heartbeat segment. Model parameter module 42 may determine the period of the individual heartbeat segment based on the indication of the start and end of the individual heartbeat segment within the discrete-time ABP waveform provided by heartbeat detection module 40. For example, model parameter module 42 may identify the period of each individual heartbeat segment by multiplying the number of samples included in the individual heartbeat segment by a defined sampling rate used to generate the discrete-time ABP waveform. In other examples, such as when heartbeat detection module 40 provides an indication of the start time and end time (i.e., relative or absolute time) of each individual heartbeat portion, model parameter module 42 may determine the period of the individual heartbeat portion as the difference between the end time and start time indicated by heartbeat detection module 40.

[0035] Model parameter module 42 determines a set of frequency component coefficients for each individual heartbeat portion based in part on the period of the individual heartbeat portion. Each set of frequency component coefficients includes the same number of coefficients. Thus, each set of frequency component coefficients represents the shape of the associated individual heartbeat portion using the same number of frequency component coefficients, independent of the period of the corresponding heartbeat portion.

[0036] The model parameter module 42 normalizes each frequency component coefficient set to produce a normalized frequency component coefficient set for each individual heartbeat portion that is independent of the scaling, mean value, and period of the corresponding heartbeat portion. The normalized frequency component coefficient set is provided by the model parameter module 42 to the heartbeat classification module 44.

[0037] The heartbeat classification module 44 implements an unsupervised real-time clustering algorithm that classifies each individual heartbeat portion based on a comparison of a received set of normalized frequency component coefficients with a reference set of frequency component coefficients. The reference set of frequency component coefficients represents a central reference for "normal" clustering groups or categories of normalized coefficient sets and may be developed by the heartbeat classification module 44 during an initial learning phase of operation, as further described below. The heartbeat classification module 44 compares the normalized reference set of coefficients for a newly sensed individual heartbeat portion with the reference set of frequency component coefficients on a beat-by-beat basis to generate a quality indicator associated with the individual heartbeat portion.

[0038] In some examples, the heartbeat classification module 44 generates a quality indicator that identifies two sets (or classes) of individual heartbeat portions: those that are similar to a reference set of frequency component coefficients (e.g., within a threshold deviation therefrom) and that are classified as eligible for processing by the hemodynamic processing module 38 ( Figure 1 ) for downstream processing; and individual heartbeat portions associated with frequency component coefficients that are not similar to a reference set of frequency component coefficients (e.g., exceed a threshold deviation) and are classified as ineligible for downstream processing by the hemodynamic processing module 38 ( Figure 1 ) for downstream processing. In other examples, heartbeat classification module 44 may generate quality indicators that identify more than two sets or categories of individual heartbeat portions, such as by comparing the deviation between the frequency component coefficients for the individual heartbeat portions and a reference set of frequency component coefficients to a plurality of threshold deviations, each threshold deviation quantifying the degree to which the frequency component coefficients for the individual heartbeat portions deviate from the reference set of frequency component coefficients.

[0039] The heartbeat classification module 44 sends a message to the hemodynamic processing module 38 ( Figure 1) provides an indication of the individual heartbeat portions (e.g., the location of the individual heartbeat portions within the discrete-time ABP waveform signal) and a quality indicator associated with the individual heartbeat portions. Thus, the ABP waveform processing module 36 identifies a quality indicator (e.g., a classification) for each individual heartbeat portion, which is used to increase the accuracy of downstream hemodynamic parameter determinations.

[0040] Figure 3 is a graph showing an example ABP waveform 46, which may be Figure 2 The heartbeat detection module 40 separates the ABP waveform 46 into individual heartbeat components. Figure 1 ) of the ABP, which is plotted in millimeters of mercury (mmHg) as pressure and in minutes as time.

[0041] like Figure 3 As shown, the ABP waveform 46 includes a pressure waveform that can be separated into a plurality of individual heartbeat portions from the beginning of the systolic pressure rise to the end of the diastolic pressure. As further shown, the ABP waveform 46 includes regions A, B, and C. In this example, regions A, B, and C include false pressure measurements or other error artifacts that cause signal noise in the form of changes in the period, amplitude, and shape of the ABP waveform 46. Therefore, the heartbeat detection module 40 ( Figure 2 ) Downstream use of the individual heartbeat portions identified within regions A, B, and C can result in downstream hemodynamic parameter determination (e.g., by Figure 1 The accuracy of the hemodynamic processing module 38) is reduced.

[0042] As further described below, the heartbeat classification module 44 ( Figure 2 ) generates quality indicators for individual heartbeat portions of the ABP waveform 46, including those identified within regions A, B, and C. In some examples, the quality indicators may classify the individual heartbeat portions as eligible for downstream processing or ineligible for downstream processing. Figure 3 In the example of FIG4 , one or more individual beat portions identified within regions A, B, and C may be classified by the beat classification module 44 as ineligible for downstream processing, and individual beat portions identified within the ABP waveform 46 outside of regions A, B, and C may be classified by the beat classification module 44 as eligible for downstream processing. Thus, as described further below, the beat classification module 44 may provide quality indicators associated with individual beat portions of the ABP waveform that may be used to reduce the amount of noise or other error artifacts within the ABP waveform used for hemodynamic parameter determination or other monitoring techniques.

[0043] Figure 4 It shows Figure 2Flowchart of an example operation of the model parameter module 42 to determine a set of frequency component coefficients representing the shape of an individual heartbeat portion. Figure 1 The hemodynamic monitoring system 10 and the above Figure 1 and Figure 2 Example operations are described in the context of the ABP waveform processing module 36 .

[0044] Receive individual heartbeat portions of the ABP waveform signal (step 48). For example, the model parameter module 42 may receive an indication of the location of the individual heartbeat portion within the sensed ABP waveform signal from the heartbeat detection module 40 (e.g., via a discrete-time index). Determine the period of the individual heartbeat portion (step 50). For example, the model parameter module 42 may receive an indication of the start and end of the individual heartbeat portion within the discrete-time ABP waveform signal from the heartbeat detection module 40, such as an index within the discrete-time signal corresponding to the start of the individual heartbeat portion (i.e., the start of the systolic pressure rise) and an index corresponding to the end of the individual heartbeat portion (i.e., the end of the diastolic pressure).

[0045] A frequency component coefficient set is determined for each individual heartbeat portion (step 52). For example, the finite Fourier series expansion of the received individual heartbeat portion can be expressed according to the following equation:

[0046]

[0047] Where: n is the index number of the individual heartbeat portion within the ABP waveform;

[0048] s n (t) is the ABP waveform signal for the nth heartbeat index number;

[0049] w n (t) represents the model error for the nth heartbeat index number;

[0050] l is an index ranging from -M to +M;

[0051] M corresponds to the highest frequency coefficient;

[0052] c l [n] is the frequency component coefficient for the nth heartbeat part;

[0053] t s [n] is the start time of the nth beat portion within the ABP waveform; and

[0054] T[n] is the period of the nth heartbeat part.

[0055] Therefore, the frequency component coefficients representing the shape of the nth individual heartbeat portion can be determined according to the following equation:

[0056]

[0057] Where: n is the index number of the individual heartbeat portion within the ABP waveform;

[0058] l is an index ranging from -M to +M;

[0059] c l [n] is the frequency component coefficient for the nth heartbeat part;

[0060] T[n] is the period of the nth heartbeat part;

[0061] s n (t) is the ABP waveform signal for the nth heartbeat index number; and

[0062] t s [n] is the start time of the nth heartbeat portion within the ABP waveform.

[0063] The model parameter module 42 thus determines a set of coefficients representing the amplitude and phase of the frequency components of the received individual heartbeat portions. The determined set of frequency component coefficients for the nth individual heartbeat portion can be expressed in vector notation according to the following equation:

[0064] c[n]=[c1[n] ... c M [n]] T (Equation 3)

[0065] In some examples, model parameter module 42 may insert the average of the determined coefficients and the period of the individual heartbeat portion into the frequency component coefficient vector. In such an example, the determined set of frequency component coefficients for the nth individual heartbeat portion may be expressed in vector notation according to the following equation:

[0066] c[n]=[c0[n],c1[n],...,c M+1 [n]] T (Equation 4)

[0067] Where: c0[n] is the coefficient c1[n] to c M The average value of [n], and c M+1 [n] is the period of the nth individual heartbeat portion. Although in the above example equation 4, the frequency component coefficients c1[n] to c M The average value of [n] is inserted at the starting index c0[n], and the period of the nth individual heartbeat part is inserted at the ending index c M+1 [n], but it will be appreciated that the mean and period may be inserted at any defined position in the set of frequency component coefficients.

[0068] Normalize the frequency component coefficient sets for the individual heartbeat portions (step 54). For example, model parameter module 42 may normalize the frequency component coefficient sets representing the shapes of the individual heartbeat portions expressed above in vector notation with respect to Equation 3 using division by the 2-norm of the frequency component coefficient sets or other normalization techniques. In some examples, such as when model parameter module 42 expresses the frequency component coefficient sets in vector notation according to Equation 4 above (i.e., including the average value of the frequency component coefficients and the period of the individual heartbeat portions), model parameter module 42 may normalize the frequency component coefficient sets using only those elements of the set that represent the shapes of the individual heartbeat portions (i.e., excluding elements corresponding to the average value of the coefficients and the period of the individual heartbeat portions). For example, model parameter module 42 may normalize the frequency component coefficient sets for the nth individual heartbeat portion according to the following equation:

[0069]

[0070] Where: c s [n] is the set of normalization coefficients for the nth heartbeat part;

[0071] c0[n] is the average value of the frequency component coefficients before normalization;

[0072] T[n] is the period of the nth heartbeat portion; and

[0073] ||c[n]|| s is the set of frequency coefficients [c1[n]…c M The 2-norm of [n].

[0074] The 2-norm of a set of frequency component coefficients can be determined according to the following equation:

[0075]

[0076] The normalized frequency component coefficient sets for the individual heartbeat portions are provided to the heartbeat classification module 44 (step 56). For example, the model parameter module 42 may provide the frequency component coefficient sets c s [n], as described above with respect to Equation 5 Provided to the heartbeat classification module 44 for classification and determination of quality indicators associated with the individual heartbeat portions.

[0077] Although Figure 4 The example of describes a single iteration of the model parameter module 42 to determine the frequency component coefficients representing the shape of an individual heartbeat portion, but it should be understood that Figure 4 The example operations of may be iteratively performed on a beat-by-beat basis to provide a plurality of sets of normalized frequency component coefficients for a plurality of individual heartbeats for classification by the heartbeat classification module 44 .

[0078] The model parameter module 42 determines each frequency component coefficient set (and therefore, each normalized frequency component coefficient set) to have the same number of coefficients, such as 128 coefficients or another number of coefficients. Each frequency component coefficient set is determined based on the period of the associated individual heartbeat portion. Therefore, even though the period of the individual heartbeat portion may differ due to expected physiological behavior, the model parameter module 42 provides a normalized frequency coefficient set for each individual heartbeat portion that uses the same number of frequency component coefficients to represent the shape of the individual heartbeat portion to facilitate comparison of the shapes of multiple individual heartbeat portions by the heartbeat classification module 44 or other downstream processing / classification. In addition, the normalized frequency component coefficient set represents the shape of the individual heartbeat portion independently of the scaling (amplitude), average value (offset), and period (stretch) of the corresponding individual heartbeat portion, thereby further facilitating comparison and classification of individual heartbeat portions by taking into account natural (or normal) deviations in the amplitude, offset, and period of the individual heartbeat portion due to physiological factors (such as respiratory regulation and baroreflex regulation of the human body).

[0079] Figure 5 It shows Figure 2 A flowchart of an example operation of the heartbeat classification module 44 to classify individual heartbeat portions of an ABP waveform to provide quality indicators associated with the individual heartbeat portions. For purposes of clarity and ease of discussion, the following is a flowchart of an example operation of the heartbeat classification module 44 to classify individual heartbeat portions of an ABP waveform to provide quality indicators associated with the individual heartbeat portions. Figure 1 The hemodynamic monitoring system 10 and the above Figure 1 and Figure 2 Example operations are described in the context of the ABP waveform processing module 36 .

[0080] Receive a set of normalized frequency component coefficients for an individual heartbeat portion (step 58). For example, the heartbeat classification module 44 may receive the set of normalized frequency component coefficients c from the model parameter module 42. s [n], which can be described in vector notation as

[0081] The set of normalized frequency component coefficients for the individual heartbeat portions is compared to a reference set of frequency component coefficients (step 60). For example, the heartbeat classification module 44 may generate a central reference for a "normal" (i.e., representative of a patient's typical or other expected heartbeat) cluster group or class using, for example, iterative estimation of the mean (or other central tendency) of the received set of normalized frequency component coefficients, as further described below.

[0082] The heartbeat classification module 44 can use the determined center reference as a reference normalized frequency component coefficient set and can compare the received normalized frequency component coefficient set for an individual heartbeat portion with the reference frequency component coefficient set to identify the extent to which the received normalized frequency component coefficient set (and therefore, the shape of the individual heartbeat portion) deviates from the reference frequency component coefficient set (i.e., representing an "average" or "normal" heartbeat of the patient). For example, the heartbeat classification module 44 can compare the received normalized frequency component coefficient set for the nth individual heartbeat portion with the reference frequency component coefficient set according to the following equation:

[0083]

[0084] Where: c s [n] is the received set of normalized frequency component coefficients;

[0085] is the reference normalized frequency component coefficient set; and

[0086] W is the weight vector, expressed as [w0, w1, ..., w M+1 ] T .

[0087] Thus, the heartbeat classification module 44 may use a weighted Euclidean norm to determine the deviation (or distance) between the received normalized set of frequency component coefficients and the reference set of frequency component coefficients, as expressed in Equation 7 above, although other distance functions or deviation techniques are possible.

[0088] The individual elements of the weight vector W (expressed as [w0, w1, ..., w M+1 ] T ) to suit a particular application. In some examples, for example, an individual heartbeat with irregular long or short cycles due to physiological conditions may be considered "normal" (or compliant), in which case the weight vector w is reduced. M+1 The values ​​of the elements of (i.e., the periods corresponding to the individual heartbeat parts) are appropriate. In the example where individual heartbeats with irregularly long or short periods due to physiological conditions are considered "abnormal" (or irregular), the weight vector w is increased. M+1 The value of the element of is appropriate. Similarly, the average value of the frequency component coefficients for individual heartbeat portions, represented by the normalized frequency component coefficients c0[n] and weighted by the weight vector element w0, can change due to "normal" (or expected) physiological reasons. Therefore, reducing the value of the weight vector element w0 can reduce the impact of unexpected or otherwise rapidly changing values ​​of the average coefficient value c0[n] (or penalize these values). Increasing the value of the weight vector element w0 can increase (or emphasize) the impact of the value of the average coefficient value c0[n].

[0089] Based on the comparison of the normalized frequency component coefficient set with the reference frequency component coefficient set, a quality indicator associated with the individual heartbeat portion is provided (step 62). For example, in an example where the heartbeat classification module 44 classifies the individual heartbeat portion as belonging to one of two separate categories (or groups), namely, a category that is eligible for downstream processing or a category that is not eligible for downstream processing, the heartbeat classification module 44 may calculate the determined distance between the normalized frequency component coefficient set and the reference frequency component coefficient set according to the following equation: or other deviations to compare with the threshold deviation:

[0090]

[0091] where β is the threshold bias parameter.

[0092] The threshold deviation parameter β can be an empirically determined value, such as a value between 0 and 1 (or other range), which separates individual heartbeat portions associated with normalized frequency component coefficients that are similar to the reference frequency component coefficient set (i.e., less than the threshold deviation parameter β) from individual heartbeat portions associated with normalized frequency component coefficients that are not similar to the reference frequency component coefficient set (i.e., greater than the threshold deviation parameter β). In such an example, the heartbeat classification module 44 can provide a quality indicator representing the classification of the individual heartbeat portions, such as providing a value of 0 (or other defined value) in response to determining that the deviation (or distance) between the normalized frequency component coefficient set and the reference frequency component coefficient set is less than the threshold deviation parameter (e.g., β), or providing a value of 1 (or other defined value) in response to determining that the deviation is not less than (e.g., greater than or equal to) the threshold deviation parameter (e.g., β).

[0093] Thus, in some examples, the heartbeat classification module 44 may classify each individual heartbeat portion as belonging to one of two categories: a category that is eligible for downstream processing (i.e., similar to the reference heartbeat) or a category that is not eligible for downstream processing (i.e., not similar to the reference heartbeat). The quality indicator may be provided to, for example, the hemodynamic processing module 38 ( Figure 1) for use in hemodynamic parameter determination or other hemodynamic monitoring techniques. For example, the hemodynamic processing module 38 can utilize individual heartbeat portions that are classified as eligible for downstream processing related to hemodynamic parameter determination, such as determination of cardiac output (CO), vascular resistance, stroke volume, or other hemodynamic parameters. The hemodynamic processing module 38 can refrain from utilizing (e.g., by discarding or otherwise refraining from utilizing) individual heartbeat portions that are classified as ineligible for downstream processing. For example, the hemodynamic processing module 38 can discard (or ignore) portions of the received ABP waveform signal that are indicated as ineligible for downstream processing, thereby assembling a usable ABP waveform signal from an ordered combination of individual heartbeat portions that are indicated as eligible for downstream processing.

[0094] In some examples, the heartbeat classification module 44 may classify an individual heartbeat portion as belonging to one of three or more categories, such as three or more categories representing the extent to which the set of normalized frequency component coefficients for the individual heartbeat portion deviates from a reference set of frequency component coefficients. In such examples, the hemodynamic processing module 38 may utilize the classification of each individual heartbeat portion during hemodynamic parameter determination, such as by weighting the contribution of each defined category differently. For example, the hemodynamic processing module 38 may apply the greatest weight to the individual heartbeat portion associated with the normalized frequency component coefficients classified as being most similar to the reference set of frequency component coefficients (i.e., having the least deviation therefrom). Similarly, the hemodynamic processing module 38 may apply the least weight to the individual heartbeat portion associated with the normalized frequency component coefficients classified as being least similar to the reference set of frequency component coefficients (i.e., having the greatest deviation therefrom).

[0095] In some examples, the heartbeat classification module 44 may calculate the deviation (or distance) (e.g., distance The heartbeat classification module 44 may provide the hemodynamic processing module 38 with the actual value of the determined deviation for downstream processing. For example, the hemodynamic processing module 38 may utilize the actual value of the determined deviation as a coefficient (or gain) that affects the contribution of the individual heartbeat component to the determination of the hemodynamic parameter. In another example, the heartbeat classification module 44 may provide the deviation to the hemodynamic processing module 38. Thus, heartbeat classification module 44 can provide quality indicators associated with individual heartbeat portions that can be used by hemodynamic processing module 38 during downstream processing of the ABP waveform to determine one or more hemodynamic parameters.

[0096] The reference coefficient set is updated based on the quality indicator (step 64). For example, the heartbeat classification module 44 can determine the reference frequency component coefficient set using, for example, an iterative estimation of an average (or other central tendency) of multiple received normalized frequency component coefficient sets, as further described below. In some examples, in response to determining that the individual heartbeat portion is classified as eligible for downstream processing (e.g., the deviation between the frequency component coefficient set for the individual heartbeat portion and the reference frequency component coefficient set is less than a threshold deviation parameter β), the heartbeat classification module 44 updates the reference frequency component coefficient set using the received normalized frequency component coefficient set in the iterative estimation of the average. In response to determining that the individual heartbeat portion is not eligible for downstream processing (e.g., the deviation between the frequency component coefficient set for the individual heartbeat portion and the reference frequency component coefficient set is greater than or equal to the threshold deviation parameter β), the heartbeat classification module 44 can refrain from using the received normalized frequency component coefficient set for the individual heartbeat portion in the iterative estimation of the average (i.e., determining the reference frequency component coefficient set).

[0097] Thus, the heartbeat classification module 44 implements an unsupervised real-time classification algorithm that provides a quality indicator associated with each individual heartbeat portion on a beat-by-beat basis. The hemodynamic processing module 38 uses the quality indicator to identify those portions of the ABP waveform that are suitable for downstream hemodynamic parameter determination, thereby reducing the amount of noise or other artifacts included in the ABP waveform signal (from which the hemodynamic parameters are derived). The unsupervised classification algorithm updates the central reference (i.e., the set of reference frequency component coefficients) based on the classification, thereby adapting to the data generated by the physiological sensors (e.g., Figure 1 The ABP waveform signal sensed by one or more of the physiological sensors 14 and 16 is used to further increase the accuracy of the classification scheme.

[0098] Figure 6 It shows Figure 2 and Figure 5 FIG. 4 is a flow diagram of example operations of the heartbeat classification module 44 during an initial learning period of an unsupervised real-time classification algorithm. Figure 6 The example operations of may be performed by the heartbeat classification module 44, for example, during an initial phase of monitoring, to develop a reference set of frequency component coefficients used by a real-time classification algorithm to classify individual heartbeat portions. Figure 1 The hemodynamic monitoring system 10 and the above Figure 1 and Figure 2 Example operations are described in the context of the ABP waveform processing module 36 .

[0099] Receive the normalized frequency component coefficient set for the individual heartbeat portion (step 66). For example, the heartbeat classification module 44 may receive the normalized frequency component coefficient set c described above with respect to Equations 5 and 6 from the model parameter module 42. s [n].

[0100] Generate a reference frequency component coefficient set using the normalized frequency component coefficient set (step 68). For example, the heartbeat classification module 44 may determine the reference frequency component coefficient set for the nth individual heartbeat portion using iterative estimation of the average value of the received normalized frequency component coefficient set according to the following equation:

[0101]

[0102] Where: n is the index number of the individual heartbeat portion within the ABP waveform;

[0103] is the set of reference frequency component coefficients; and

[0104] α is the iterative averaging parameter.

[0105] The value of the iterative averaging parameter α can be determined empirically, such as a value greater than 0 and less than 1, which controls how much weight is given to past estimates relative to the latest input. Therefore, the value of the iterative averaging parameter α controls the reference frequency component coefficient set The speed of adaptation to changing heartbeat shapes (e.g., due to "normal" or expected physiological factors). Smaller values ​​of the iterative averaging parameter α give less weight to new values, resulting in a reference frequency component coefficient set of A larger value of the iterative averaging parameter α gives a larger weight to the new value, resulting in a slower adaptation of the reference frequency component coefficient set Faster adaptation.

[0106] A determination is made as to whether the initial learning period criteria are met (step 70). For example, in some examples, the heartbeat classification module 44 may determine that the initial learning period criteria are met in response to receiving a threshold number of normalized frequency component coefficient sets, such as ten sets, twenty sets, or another number of normalized frequency component coefficient sets.

[0107] In other examples, the heartbeat classification module 44 may determine that the initial learning period criterion is satisfied in response to determining that a threshold number of normalized frequency component coefficient sets that satisfy a threshold standard deviation criterion have been received. For example, the heartbeat classification module 44 may determine the reference frequency component coefficient set according to the following equation: Iterative estimation of the standard deviation of :

[0108]

[0109] Where: n is the index number of the individual heartbeat portion within the ABP waveform;

[0110] is the iterative estimate of the standard deviation;

[0111] α is the iterative averaging parameter;

[0112] c s [n] is the set of normalized frequency component coefficients; and is the set of reference frequency component coefficients.

[0113] In some examples, heartbeat detection module 44 may calculate the standard deviation The 2-norm or other normalization of the frequency component coefficients is compared with a threshold standard deviation criterion to determine whether the initial learning period criterion is met. The threshold standard deviation criterion can be an empirically determined value, such as a value equal to half of a threshold minimum number of normalized frequency component coefficient sets (e.g., ten, twenty, or another threshold value).

[0114] In such an example, heartbeat detection module 44 may respond to determining that a minimum number of normalized frequency component coefficient sets (e.g., ten, twenty, or other minimum number of sets) has been received and the determined reference frequency component coefficient set The standard deviation of meets the threshold standard deviation criterion to determine whether the initial learning period criterion is met.

[0115] In response to determining that the initial learning period criteria are not met (the "No" branch of step 70), the heartbeat classification module 44 may continue to receive normalized frequency component coefficient sets for individual heartbeat portions during the initial learning period (step 66). In response to determining that the initial learning period criteria are met (the "Yes" branch of step 70), the heartbeat classification module 44 may exit the initial learning period phase (step 72) and may enter the real-time classification phase to classify the normalized frequency component coefficient sets to provide associated quality indicators, as described above with respect to Figure 5 The example operation is described.

[0116] Thus, a hemodynamic monitoring system 10 implementing the techniques of the present disclosure can implement an unsupervised, real-time classification algorithm that classifies and provides quality indicators associated with individual heartbeat portions of a patient's sensed ABP signal. The classification algorithm utilizes frequency parameters representing the shape of the ABP waveform that are independent of the scaling, average value, and period of each individual heartbeat portion of the ABP waveform, thereby increasing the consistency of the classification and reducing the operational complexity associated with time-domain analysis of individual heartbeat portions associated with naturally occurring variations in the cycle of an individual heartbeat. Thus, the techniques described herein can increase the accuracy and reliability of hemodynamic parameters derived from a patient's sensed ABP waveform.

[0117] Although the present invention has been described with reference to (one or more) exemplary embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for its elements without departing from the scope of the present invention. In addition, many modifications may be made to adapt specific circumstances or materials to the teachings of the present invention without departing from the basic scope of the present invention. Therefore, it is intended that the present invention is not limited to the disclosed (one or more) specific embodiments, but that the present invention will include all embodiments falling within the scope of the appended claims.

Claims

1. A hemodynamic monitor comprising: a sensor interface that receives a hemodynamic sensor signal from a hemodynamic sensor, the hemodynamic sensor signal representing the patient's arterial blood pressure (ABP); a heartbeat detection module that separates a received hemodynamic sensor signal into a plurality of heartbeat portions, each heartbeat portion representing the ABP of the patient for one of a plurality of individual heartbeats of the patient; Model parameter module, the model parameter module: determining, for each of the plurality of heartbeat portions, a coefficient set representing a frequency component of the corresponding heartbeat portion to generate a plurality of coefficient sets, each coefficient set including the same number of coefficients; and normalizing each coefficient set to generate a plurality of normalized coefficient sets; Heartbeat classification module, the heartbeat classification module: determining a reference coefficient set based on the plurality of normalized coefficient sets; and providing a quality indicator associated with the individual heartbeat based on a comparison of the set of normalized coefficients for the individual heartbeat with the reference set of coefficients; as well as A hemodynamics processing module, wherein the hemodynamics processing module: generating a modified hemodynamic sensor signal using the quality indicator; deriving one or more hemodynamic parameters from the modified hemodynamic sensor signal; as well as One or more derived hemodynamic parameters are output.

2. The hemodynamic monitor of claim 1, wherein the quality indicator classifies the individual heartbeat as eligible for downstream processing or ineligible for the downstream processing.

3. The hemodynamic monitor of claim 1 , wherein the heartbeat classification module generates the quality indicator based on the extent to which the set of normalized coefficients for the individual heartbeat deviates from the reference set of coefficients.

4. The hemodynamic monitor of claim 3 , wherein the heartbeat classification module determines the extent to which the set of normalized coefficients for the individual heartbeat deviates from the reference set of coefficients as a vector norm of a difference between the set of normalized coefficients for the individual heartbeat and the reference set of coefficients. The hemodynamic monitor of claim 4 , wherein the vector norm is a weighted vector norm.

6. The hemodynamic monitor of claim 1 , wherein the model parameter module determines, for each of the plurality of heartbeat portions, the set of coefficients representing the frequency components of the corresponding heartbeat portion by: for each of the plurality of heartbeat portions, identifying a period of the corresponding heartbeat portion; and The set of coefficients for the respective heartbeat portion is determined based on the identified period of the respective heartbeat portion.

7. The hemodynamic monitor of claim 6 , wherein the model parameter module determines the set of coefficients for the corresponding heartbeat portion based on the identified period of the corresponding heartbeat portion using a Fourier series expansion of the corresponding heartbeat portion as a function of the identified period of the corresponding heartbeat portion.

8. The hemodynamic monitor of claim 7 , wherein the model parameter module determines the set of coefficients according to the following equation using the Fourier series expansion of the corresponding heartbeat portion as a function of the identified period of the corresponding heartbeat portion: wherein n is an index number identifying the corresponding heartbeat portion within the hemodynamic sensor signal; Wherein l is an index ranging from the negative value of the highest frequency coefficient to the positive value of said highest frequency coefficient; where c l [n] is the set of coefficients for the corresponding heartbeat portion; wherein T[n] is the identified period of the corresponding heartbeat portion; where s n (t) is the hemodynamic sensor signal; and where t s [n] is the start time within the hemodynamic sensor signal of the corresponding heartbeat portion.

9. The hemodynamic monitor of claim 1 , wherein the model parameter module normalizes each coefficient set by dividing each coefficient from the corresponding coefficient set by a 2-norm of the corresponding coefficient set to generate the plurality of normalized coefficient sets.

10. The hemodynamic monitor of claim 1, wherein the heartbeat classification module determines the reference coefficient set based on the plurality of normalization coefficient sets using an iterative estimation of a mean value of the normalization coefficient sets.

11. The hemodynamic monitor of claim 10 , wherein the heartbeat classification module determines the reference coefficient set using the iterative estimation of the mean of the normalized coefficient set according to the following equation: in is the reference coefficient set for heartbeat index n; and where α is the iterative averaging parameter.

12. The hemodynamic monitor according to claim 11, wherein α is a number greater than 0 and less than 1.

13. The hemodynamic monitor of claim 1 , wherein the quality indicator classifies the individual heartbeat as eligible for downstream processing or ineligible for the downstream processing; and wherein the hemodynamic processing module generates the modified hemodynamic sensor signal using the quality indicator in the following manner: responsive to determining that the quality indicator classifies the individual heartbeat as eligible for the downstream processing, generating the modified hemodynamic sensor signal to include the individual heartbeat therein; and In response to determining that the quality indicator classifies the individual heartbeat as ineligible for the downstream processing, the modified hemodynamic sensor signal is generated to not include the individual heartbeat in the modified hemodynamic sensor signal.

14. A system comprising: a hemodynamic sensor configured to sense the patient's arterial blood pressure (ABP); as well as a hemodynamic monitor connected to the hemodynamic sensor, the hemodynamic monitor comprising: a sensor interface configured to receive a hemodynamic sensor signal from the hemodynamic sensor, the hemodynamic sensor signal representing the ABP of the patient sensed by the hemodynamic sensor; one or more processors; and a computer readable memory encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to: separating the received hemodynamic sensor signal into a plurality of heartbeat portions, each heartbeat portion representing the ABP of the patient for one of a plurality of individual heartbeats of the patient; determining, for each of the plurality of heartbeat portions, a coefficient set representing a frequency component of the corresponding heartbeat portion to generate a plurality of coefficient sets, each coefficient set including the same number of coefficients; normalizing each coefficient set to generate a plurality of normalized coefficient sets; determining a reference coefficient set based on the plurality of normalized coefficient sets; comparing the set of normalized coefficients for the individual heartbeat with the reference set of coefficients; determining a quality indicator associated with the individual heartbeat based on the comparison; generating a modified hemodynamic sensor signal using the quality indicator; deriving one or more hemodynamic parameters from the modified hemodynamic sensor signal; and One or more derived hemodynamic parameters are output.

15. The system of claim 14, wherein the quality indicator classifies the individual heartbeat as eligible for downstream processing or ineligible for the downstream processing.

16. The system of claim 14, wherein the computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to determine the quality indicator based on the extent to which the set of normalized coefficients for the individual heartbeat deviates from the reference set of coefficients.

17. The system of claim 16, wherein the computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to determine the extent to which the set of normalized coefficients for the individual heartbeat deviates from the reference set of coefficients as a vector norm of the difference between the set of normalized coefficients for the individual heartbeat and the reference set of coefficients.

18. The system of claim 17, wherein the vector norm is a weighted vector norm.

19. The system of claim 14, wherein the computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to determine, for each of the plurality of heartbeat portions, the set of coefficients representing frequency components of the respective heartbeat portion by causing the hemodynamic monitor to: for each of the plurality of heartbeat portions, identifying a period of the corresponding heartbeat portion; and The set of coefficients for the respective heartbeat portion is determined based on the identified period of the respective heartbeat portion.

20. The system of claim 19, wherein the computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to determine the set of coefficients for the corresponding heartbeat portion based on the identified period of the corresponding heartbeat portion by causing the hemodynamic monitor to determine the set of coefficients using a Fourier series expansion of the corresponding heartbeat portion as a function of the identified period of the corresponding heartbeat portion.

21. The system of claim 20, wherein the computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to determine the set of coefficients according to the following equation using the Fourier series expansion of the corresponding heartbeat portion as a function of the identified period of the corresponding heartbeat portion: wherein n is an index number identifying the corresponding heartbeat portion within the hemodynamic sensor signal; Wherein l is an index ranging from the negative value of the highest frequency coefficient to the positive value of said highest frequency coefficient; where c l [n] is the set of coefficients for the corresponding heartbeat portion; wherein T[n] is the identified period of the corresponding heartbeat portion; where s n (t) is the hemodynamic sensor signal; and where t s [n] is the start time within the hemodynamic sensor signal of the corresponding heartbeat portion.

22. The system of claim 14, wherein the computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to normalize each coefficient set to produce the plurality of normalized coefficient sets by causing the hemodynamic monitor to divide each coefficient from a corresponding coefficient set by a 2-norm of the corresponding coefficient set.

23. The system of claim 14, wherein the computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to determine the reference set of coefficients based on the plurality of normalized coefficient sets by causing the hemodynamic monitor to determine the reference set of coefficients using an iterative estimate of an average of the normalized coefficient sets.

24. The system of claim 23, wherein the computer-readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to use the iterative estimate of the mean of the normalized coefficient set to determine the reference coefficient set by causing the hemodynamic monitor to determine the reference coefficient set according to the following equation: in is the reference coefficient set for heartbeat index n; and where α is the iterative averaging parameter. The system of claim 24 , wherein α is a number greater than 0 and less than 1.

26. The system of claim 14, wherein the quality indicator classifies the individual heartbeat as eligible for downstream processing or ineligible for the downstream processing; and wherein the computer readable memory is further encoded with instructions that, when executed by the one or more processors, cause the hemodynamic monitor to use the quality indicator to generate the modified hemodynamic sensor signal by causing the hemodynamic monitor to: responsive to determining that the quality indicator classifies the individual heartbeat as eligible for the downstream processing, generating the modified hemodynamic sensor signal to include the individual heartbeat therein; and In response to determining that the quality indicator classifies the individual heartbeat as ineligible for the downstream processing, the modified hemodynamic sensor signal is generated to not include the individual heartbeat in the modified hemodynamic sensor signal.

27. A method comprising: generating, using a hemodynamic sensor, an analog hemodynamic sensor signal representative of the patient's arterial blood pressure (ABP); sampling the analog hemodynamic sensor signal at a defined sampling rate to produce a sampled hemodynamic sensor signal representative of the ABP of the patient; separating the sampled hemodynamic sensor signal into a plurality of heartbeat portions, each heartbeat portion representing the ABP of the patient for one of a plurality of individual heartbeats of the patient; determining, for each of the plurality of heartbeat portions, a coefficient set representing a frequency component of the corresponding heartbeat portion to generate a plurality of coefficient sets, each coefficient set including the same number of coefficients; normalizing each coefficient set to generate a plurality of normalized coefficient sets; determining a reference coefficient set based on the plurality of normalized coefficient sets; comparing the set of normalized coefficients for the individual heartbeat with the reference set of coefficients; determining a quality indicator associated with the individual heartbeat based on the comparison; generating a modified hemodynamic sensor signal using the quality indicator; deriving one or more hemodynamic parameters from the modified hemodynamic sensor signal; as well as Output derived hemodynamic parameters.

28. The method of claim 27, wherein the quality indicator classifies the individual heartbeat as eligible for downstream processing or ineligible for the downstream processing.

29. The method of claim 28, wherein determining the quality indicator associated with the individual heartbeat comprises determining the quality indicator based on a degree to which the set of normalized coefficients for the individual heartbeat deviates from the set of reference coefficients.

30. The method of claim 29, further comprising determining the extent to which the set of normalized coefficients for the individual heartbeat deviates from the reference set of coefficients as a vector norm of a difference between the set of normalized coefficients for the individual heartbeat and the reference set of coefficients.

31. The method of claim 30, wherein the vector norm is a weighted vector norm.

32. The method of claim 27, wherein determining, for each of the plurality of heartbeat portions, the set of coefficients representing the frequency components of the respective heartbeat portion comprises: for each of the plurality of heartbeat portions, identifying a period of the corresponding heartbeat portion; as well as The set of coefficients for the respective heartbeat portion is determined based on the identified period of the respective heartbeat portion.

33. A method according to claim 32, wherein determining the coefficient set for the corresponding heartbeat portion based on the identified period of the corresponding heartbeat portion includes determining the coefficient set using a Fourier series expansion of the corresponding heartbeat portion as a function of the identified period of the corresponding heartbeat portion.

34. The method of claim 33 , wherein determining the set of coefficients using the Fourier series expansion of the respective heartbeat portion as a function of the identified period of the respective heartbeat portion comprises determining the set of coefficients according to the following equation: wherein n is an index number identifying the corresponding heartbeat portion within the sampled hemodynamic sensor signal; Wherein l is an index ranging from the negative value of the highest frequency coefficient to the positive value of said highest frequency coefficient; where c l [n] is the set of coefficients for the corresponding heartbeat portion; wherein T[n] is the identified period of the corresponding heartbeat portion; where s n (t) is the sampled hemodynamic sensor signal; and where t s [n] is the start time within the sampled hemodynamic sensor signal of the corresponding heartbeat portion.

35. The method of claim 27, wherein normalizing each coefficient set to produce the plurality of normalized coefficient sets comprises dividing each coefficient from a corresponding coefficient set by a 2-norm of the corresponding coefficient set.

36. The method of claim 27, wherein determining the reference coefficient set based on the plurality of normalization coefficient sets comprises determining the reference coefficient set using an iterative estimate of an average of the normalization coefficient sets.

37. The method of claim 36, wherein determining the reference coefficient set using the iterative estimate of the mean of the normalized coefficient set comprises determining the reference coefficient set according to the following equation: in is the reference coefficient set for heartbeat index n; and where α is the iterative averaging parameter. The method according to claim 37 , wherein α is a number greater than 0 and less than 1.

39. The method of claim 27, wherein the quality indicator classifies the individual heartbeat as eligible for downstream processing or ineligible for the downstream processing; and Wherein generating the modified hemodynamic sensor signal using the quality indicator comprises: responsive to determining that the quality indicator classifies the individual heartbeat as eligible for the downstream processing, generating the modified hemodynamic sensor signal to include the individual heartbeat therein; as well as In response to determining that the quality indicator classifies the individual heartbeat as ineligible for the downstream processing, the modified hemodynamic sensor signal is generated to not include the individual heartbeat in the modified hemodynamic sensor signal.

Images