Biologically Inspired Motion Compensation and Real-Time Physiological Load Estimation Using a Dynamic Heart Rate Prediction Model

Abstract

The current invention pertains to a method whereby the accuracy of a heart rate prediction gathered from sensor data can be improved during periods when motion corrupts the signal. The model utilized can also be inverted to infer information on the physiological state of a subject, such as real-time energy utilization or physiological load. In addition, this method can also be used to segment the contribution of each energy system, namely the phosphagen system, anaerobic glycolysis and aerobic respiration, to the physiological load experienced by the user. At the core of this approach lies a model describing the dynamic adjustment of human heart rate under varying physiological demands.

Claims

1. A method for augmenting heart rate predictions determined from a heart rate signal using a dynamic heart rate model, the method comprising: (a) measurement of a motion signal from a motion capturing sensor; (b) measurement of a heart rate signal from a heart rate sensor; (c) application of a dynamic heart rate model which infers a heart rate from the motion signal and other parameters during periods when the heart rate signal is distorted; (d) transmitting the heart rate.

2. The dynamic heart rate model of claim 1, which may comprise an ordinary differential equation (ODE) model.

3. The parameters of claim 1, which may be inferred in conjunction with a probabilistic framework, such as Hidden Markov Models.

4. A system for augmenting heart rate predictions determined from a heart rate signal using a dynamic heart rate model, the system comprising: (a) a wearable device comprising a motion capturing sensor and a heart rate sensor; (b) measurement of a motion signal from the motion capturing sensor which may comprise an accelerometer; (c) measurement of a heart rate signal from the heart rate sensor which may comprise an electrocardiogram (ECG) or photoplethysmography (PPG) sensor; (d) application of a dynamic heart rate model which infers a heart rate from the motion signal and other parameters during periods when the heart rate signal is distorted; (e) transmitting the heart rate.

5. The dynamic heart rate model of claim 4, which may comprise an ordinary differential equation (ODE) model.

6. The parameters of claim 4, which may be inferred in conjunction with a probabilistic framework, such as Hidden Markov Models.

7. The system of claim 4 with the heart rate reported in its display

8. The system of claim 4, which can transmit the heart rate to a mobile electronic device, exemplified by a mobile phone.

9. The mobile electronic device of claim 8 configured to display the heart rate.

10. The system of claim 4 with the means to transmit the heart rate data wirelessly to a platform where said data can be stored, analyzed and viewed on client computing platforms, including but not limited to mobile computing devices, home computers or a wearable electronic device.

11. A method for inferring an instantaneous estimate of physiological load using a dynamic heart rate model, the method comprising: (a) measurement of a motion signal from a motion capturing sensor; (b) measurement of a heart rate signal from a heart rate sensor; (c) the application of a dynamic heart rate model to estimate the instantaneous physiological load; (e) transmitting the instantaneous physiological load estimate.

12. The dynamic heart rate model of claim 11, which may comprise an ordinary differential equation (ODE) model.

13. The parameters of claim 11, which may be inferred in conjunction with a probabilistic framework, such as Hidden Markov Models.

14. A system for inferring an instantaneous estimate of physiological load using a dynamic heart rate model, the system comprising: (a) a wearable device comprising a motion capturing sensor and a heart rate sensor; (b) measurement of a motion signal from the motion capturing sensor which may comprise an accelerometer; (c) measurement of a heart rate signal from the heart rate sensor which may comprise an electrocardiogram (ECG) or photoplethysmography (PPG) sensor; (d) the application of a dynamic heart rate model to estimate the instantaneous physiological load; (e) transmitting the instantaneous physiological load estimate.

15. The dynamic heart rate model of claim 14, which may comprise an ordinary differential equation (ODE) model.

16. The parameters of claim 14, which may be inferred in conjunction with a probabilistic framework, such as Hidden Markov Models.

17. The system of claim 14 with the instantaneous estimate of physiological load reported on its display.

18. The system of claim 14, that transmits the instantaneous estimate of physiological load to a mobile electronic device, exemplified by a mobile phone or directly to a cloud platform.

19. The mobile electronic device of claim 18 configured to display the instantaneous estimate of physiological load.

20. The system of claim 14 with the means to transmit the physiological load estimate data wirelessly to a platform where said data can be stored, analyzed and viewed on client computing platforms, including but not limited to mobile computing devices, home computers or a wearable electronic device.

21. A method for calculating the relative contribution of different biochemical energy systems to the instantaneous physiological load, the method comprising: (a) measurement of a motion signal from a motion capturing sensor; (b) measurement of a heart rate signal from a heart rate sensor; (c) the application of a dynamic heart rate model that infers heart rate from heart rate signals or motion signals and other parameters to estimate the instantaneous physiological load; (d) calculation of the relative contribution of different biochemical energy systems to the instantaneous physiological load estimate; (e) transmitting the relative biochemical energy system contribution to the instantaneous physiological load.

22. The dynamic heart rate model of claim 21, which may comprise an ordinary differential equation (ODE) model.

23. The parameters of claim 21, which may be inferred in conjunction with a probabilistic framework, such as Hidden Markov Models.

24. The energy systems of claim 23, which may be one or more of the following groups: phosphagen system, anaerobic glycolysis and aerobic respiration.

25. A system for calculating the relative contribution of different biochemical energy systems to the instantaneous physiological load estimate, the system comprising: (a) a wearable device comprising a motion capturing sensor and a heart rate sensor; (b) measurement of a motion signal from the motion capturing sensor which may comprise an accelerometer; (c) measurement of a heart rate signal from the heart rate sensor which may comprise an electrocardiogram (ECG) or photoplethysmography (PPG) sensor; (d) the application of a dynamic heart rate model to estimate the instantaneous physiological load; (e) calculation of the relative contribution of different biochemical energy systems to the instantaneous physiological load estimate; (f) transmission of the relative contribution of different biochemical energy systems to the instantaneous physiological load.

26. The dynamic heart rate model of claim 25, which may comprise an ordinary differential equation (ODE) model.

27. The parameters of claim 25, which may be inferred in conjunction with a probabilistic framework, such as Hidden Markov Models.

28. The energy systems of claim 25, which may be one or more of the following groups: phosphagen system, anaerobic glycolysis and aerobic respiration.

29. The system of claim 25 with the relative contribution of different biochemical energy systems to the instantaneous physiological load reported on its display.

30. The system of claim 25, that transmits the relative contribution of different biochemical energy systems to the instantaneous physiological load to a mobile electronic device, exemplified by a mobile phone or directly to a cloud platform.

31. The mobile electronic device of claim 25 configured to display the relative contribution of different biochemical energy systems to the instantaneous physiological load.

32. The system of claim 25 with the means to transmit the relative contribution of different biochemical energy systems to the instantaneous physiological load data wirelessly to a platform where said data can be stored, analyzed and viewed on client computing platforms, including but not limited to mobile computing devices, home computers or a wearable electronic device.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The preferred embodiments of the invention will be described by way of example only, with reference to the accompanying drawings:

[0020] FIG. 1: A depiction of the output from a simple model mapping physiological load to heart rate changes.

[0021] FIG. 2: A representation of the mapping of heart rate changes to physiological load and the inferred load difference that should be made during a tandem cycling and jogging session.

[0022] FIG. 3: A depiction of the different activity to physiological load mappings for data gathered from a tandem cycling and jogging session.

[0023] FIG. 4: A depiction of the corrected physiological load mappings based on the dynamic heart rate model combined with a probabilistic inference method (HMM).

[0024] FIG. 5: A representation of the intersection of periodic cadence noise with the heart rate signal.

[0025] FIG. 6: A graph showing heart rate data for two tandem jogging sessions at different exercise intensities.

[0026] FIG. 7: A representation of the inferred physiological load for the two jogging sessions of differing intensity as shown in FIG. 6.

[0027] FIG. 8: The output for a simple model of the three different energy systems under full physiological load.

[0028] FIG. 9: A representation of the application of the energy system model to the physiological load estimated in FIG. 7.

[0029] FIG. 10: A representation of the segmentation of energy utilization for the physiological load estimated in FIG. 7.

[0030] FIG. 11 shows a basic embodiment of the invention in the context of mobile and Internet technologies.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0031] The following detailed description and drawings describe different aspects of the current invention. The description and drawings serve to enable one skilled in the art to fully understand the current invention and are not intended to limit the scope of the invention in any manner. Before the present methods and systems are disclosed and described, it is to be understood that the methods and systems are not limited to special methods, special components, or to particular implementations. It is also to be understood that the terminology used herein is for the purpose of describing particular aspects only and is not intended to be limiting.

[0032] The premise of the current invention is demonstrated using a simple example model. The model is defined mathematically, some of its basic behaviors are demonstrated and in addition, the novel ways in which it can be applied are also presented. The model takes some measure of physical activity level as input—in this case this is demonstrated using the readings from an accelerometer placed on the upper arm of a test subject. For this exemplary embodiment it is supposed that the maximum acceleration vector that can be measured has a magnitude that is six times the magnitude of gravitational acceleration (6G). 1G is then subtracted due to gravity, the absolute value is taken (as upward acceleration could result in negative acceleration values) and this is resealed to a percentage value of the maximum acceleration recorded over a small time window. When viewed it would be typical to see percentage values close to zero when a subject is resting, whereas a jogging subject would generate values that are typically in the tens of percents. This percentage value is termed the measured activity level (MA), and this example is stated simply for demonstration purposes to cover the general process of converting physical movement related signals into an estimate of physical activity level.

[0033] If it is assumed that there is some mapping between this measured activity level and the physiological energy demand that a subject's body experiences, the measured activity level can be converted into an inferred physiological load value. When such a load is applied to an individual's physiology, the body reacts by increasing the heart rate and heart stroke volume to the point where the amount of oxygen delivered to the muscles matches the physiological load. For a certain sustainable physiological load, an individual will have a heart rate at which the supply of oxygen and demand for metabolic energy are equally matched. In this embodiment, the target heart rate is designated as the heart rate for a specific exercise at a constant load.

[0034] Conceivable values for the target heart rate range between a minimum measured at rest (rHR) and a maximum determined at peak exercise intensity. The physiological load of an exercise can be mapped to a target heart rate (tHR), in the simplest case by simply employing a linear equation, with constant k1 such as:

tHR=k1(MA)+rHR  (1)

[0035] In FIG. 1 equation 1 has been employed for two exercise sessions, one at half the maximal physiological load (50%) and the next at a full physiological load (100%). The target heart rate is indicated with dashed lines for rest at 60 bpm, at 120 bpm for the first exercise session and at 180 bpm for the second exercise session.

[0036] Following this, equation 2 describes how heart rate changes in time (sHR′(t)) to reach the target heart rate. In real exercise data, the relationship resembles an exponential decay of the difference between the current heart rate and target heart rate. This can be described using an ordinary differential equation where the heart rate changes in proportion to said difference.

sHR′(t)=k2(tHR−sHR)+rHR  (2)

[0037] The relaxation constant k2 of equation 2, is better described with two separate values, k2a and k2b, for instances where sHR<tHR and sHR>=tHR respectively, as the heart rate generally adapts faster to increased target HR values than decreased HR values. This provides a complete description for a simple instance of a dynamic heart rate model.

[0038] In FIG. 2 the model output for two simulated exercise sessions where the same physiological load was applied, first in a jogging and then a cycling session, is shown. In both cases the subject is faced with a full physiological load (100%) for 5 minutes, but the physical activity readings require different multipliers to arrive at 100%. In this case additional information is clearly needed in order to find the appropriate coefficient to map between the activity reading of the accelerometer and the physiological load that the subject experiences. If a gold-standard device such as an ECG heart rate monitor was used, this makes it possible to calculate the physiological load and the appropriate factor for mapping the activity measurement to heart rate, which would show a factor two difference for the time segment where the subject cycled compared to where the subject was jogging.

[0039] For applications where the sensor used to determine heart rate is susceptible to motion artifacts, such as PPG based technologies, heart rate predictions made during times of heavy signal distortion can be augmented by outputs from an accelerometer based HR prediction. Numerous statistical frameworks exist whereby noisy readings can be dramatically improved by making use of a physical model of the system and independent noise measurements. In such approaches, estimates of the internal state of the model are continually updated based on sensor readings when a clear signal is received, and the model becomes more autonomous and is relied on more heavily when the signal quality becomes poor.

[0040] One application of such a probabilistic framework could be a Hidden Markov Model, which is a statistical model containing observable quantities, as well as the hidden states of an underlying model. When combining the model discussed thus far with accelerometer readings, the activity measure and heart rate are both observables. As pointed out in FIG. 2, the mapping from physical activity measurements to the physiological load on a subject can vary significantly between different exercise modalities, but is generally similar within a session consisting of one exercise modality. The discrepancy in this mapping can be described simply as a hidden state in an HMM and the algorithms for inferring the most likely value for this discrepancy, such as the forward algorithm (for local real-time estimation) or the backward algorithm (for the most likely global estimation) are well established. Following on from this, an exemplary embodiment of how such an approach can be implemented to infer an instantaneous physiological load value for real data gathered from the cycling and running exercise session discussed earlier is provided.

[0041] In FIG. 3 the real data gathered from an exercise and jogging session similar to the one described earlier in FIG. 2 is shown. The lower curve in FIG. 3 shows the measured activity level according to a 6G triaxial accelerometer for which the total acceleration was determined and converted as described earlier to a percentage value to indicate a measured activity level. The upper curve in FIG. 3 shows the heart rate recorded during the exercise session. From the figure it is clear that although the two exercise sessions reached similar maximum heart rate values (around 160 bpm) after 5 minutes, the measured activity values are vastly different between the two (around 30% for cycling and over 90% for running). This is expected, knowing that the test subject's arms were swinging during the run, while they were rather stationary while gripping the bicycle's handle bars. In FIG. 4 it is shown that by using the dynamic heart rate prediction model discussed earlier together with the activity measurement added to the activity discrepancy state modeled in the HMM outlined above, it is possible to obtain a realistic physiological load value for both exercises sessions (around 85% for cycling and around 95% for jogging). The discrepancy curve also highlights the slightly elevated physiological load between and after exercise sessions, which can be attributed in part to a phenomenon known as Excess Post-Exercise Oxygen uptake (EPOC), whereby anaerobic energy systems are recharged to normal levels after exercise (Ie. the phosphagen system and lactic fermentation system). A more in depth analysis of these systems is provided in the next section.

[0042] In addition to the hidden states chosen above to infer the true physiological load of an exercise, it is also possible to model hidden states wherein the motion distortion signal and heart rate signal are expected to occur at such similar frequencies that they cannot be separated from each other during signal processing (FIG. 5) when using commonly employed frequency domain methods such as the Fast Fourier Transform (FFT). These temporary situations are termed ‘cadence locks’ and by following only the accelerometer-based HR prediction during this period, a best guess of the likely heart rate trajectory can be provided. This predicted HR can also be used to improve detection of the first clearly measured HR reading after exiting this cadence lock state. Note that the accelerometer is used for obtaining both a measure of activity level and the running cadence in this example.

[0043] Up to this point, it has been demonstrated how a basic model that predicts dynamic changes in heart rate in response to different activity levels and thereby physiological loads can be used to either aid signal processing techniques in order to provide more accurate heart rate predictions or how it can be utilized to infer the physiological load for different exercise or rest states. A second use of this dynamic model includes using it in it's inverted form with HR predictions obtained from other algorithms. In FIG. 6 the HR obtained from an ECG-based device for two consecutive running sessions, the first being a shorter less intense run than the second, is depicted. Using the inverted dynamic heart rate model discussed earlier, it is possible to obtain an estimate for physiological load shown in FIG. 7, where two rectangular regions are shown for each running session, making clear the difference in time and intensity between the two exercises.

[0044] As outlined earlier, the current invention pertains to providing measurements of instantaneous activity levels as opposed to steady-state concepts such as the lactate threshold. It has already been demonstrated how estimates of instantaneous physiological load and thereby energy consumption can be obtained using measures of motion and heart rate activity. In this next section, the current invention further segments the estimated instantaneous activity level in terms of the different biochemical energy systems that contribute towards energy production in the body.

[0045] The energy system most directly linked to the muscle proteins that make movement possible is known as the phosphagen energy system. This group consists of molecules that can carry a high energy phosphate charge such as ATP and creatine-phosphate. Cells generally contain a tiny amount of these molecules, but can recharge them rapidly by breaking down glucose. The latter can be performed either in an oxygen dependent (aerobic respiration) or an oxygen independent manner (lactic fermentation). In the case of the latter, the glucose molecule is not broken down fully to CO.sub.2, but is instead converted to lactic acid, for which the accumulation capacity is limited. It is possible to model these processes mathematically to produce estimates for the activity of each of these processes at different times and different physiological loads. In FIG. 8 the degree to which each system is engaged at different time points given a full physiological load (100%), using a simple ODE model of the system, is shown. Using the instant activity levels calculated for the two running sessions shown in FIG. 7 as the physiological load value in this model, it is possible to predict the contribution of each energy system as shown in FIG. 9. Note how the phosphagen system is quick to respond but is soon exhausted, while anaerobic glycolysis is second to be engaged with a larger capacity to sustain exercise. Finally the aerobic system is the slowest, but only sustainable energy source for extended exercise sessions. Note also how the slower aerobic energy system trajectory in FIG. 9 closely follows the trajectory of the HR shown in the HR data, FIG. 6, since HR is closely coupled to the rate at which the body can supply oxygen to the muscles.

[0046] In FIG. 10 it is shown that the contribution of all three energy systems can be added together in such a way that the original physiological load estimated in FIG. 8 can be used to segment physiological load in terms of the contribution of each system. Note also how as expected the first brief run has a larger contribution from anaerobic energy systems than the longer sustained run and how between runs, the negative values for the phosphagen and anaerobic system fluxes indicate that the aerobic energy system is acting to recharge these reservoirs.

[0047] A basic embodiment of the inventions described above concerning motion compensated heart-rate calculation and instant physiological load estimation is demonstrated in FIG. 11, where 1 is a wearable electronic device containing the necessary sensor means to measure a pulse and motion signal. The wearable device optionally contains a display (2) and is capable of transmitting data to a mobile device (3) and or directly to an Internet based platform (4). The data can be stored and further processed on a server (6) for future retrieval and to be viewed on a computing platform exemplified by the personal computer (5), the mobile phone (3) and or wearable device (1).