Graduate Category: Interdisciplinary Topics Degree Level: Ph.D ...
Graduate Category: Interdisciplinary Topics Degree Level: Ph.D. Abstract ID# 595
Combined Time-Frequency Calculation of pNN50 Metric From Noisy Heart Rate Measurement [1] Payam Parsinejad
Yolanda Rodriguez-Vaqueiro
Jose Angel Martinez-Lorenzo
Problem Definition
Rifat Sipahi (PI)
pNN50 using Combined Time-Frequency (STFT) (Cntd.)
• pNN50 is a metric derived from heart rate (HR) measurements
(f) Estimate the variations of NN intervals:
• Conventionally, it is calculated from successive time periods in peak-to-peak occurrences in HR. • In the case of noisy measurements of HR, however, peak-to-peak detection may not be reliable.
Objective
where 𝐓𝐢 𝐭 instantanouse period at time instant 𝒕, or the reciprocal instantaneous frequency 𝒇𝐢 𝐭 ; and 𝚫𝐭 = 𝐍𝐍 ≈ 𝐓𝐢 . hence,
• A combined time-frequency domain analysis (Short Time Fourier Transform) is presented. • GOAL: To more accurately extract pNN50 metric from noisy HR data. • TEST: An experimental measurement with added noise is used as a benchmark problem to demonstrate the effectiveness of the approach with noticeable improvement over the conventional time domain peak-to-peak detection algorithm.
(g) Calculate pNN50 using:
Validating Our Method in Calculating pNN50
Why Heart Rate (HR), and Heart Rate Variability (HRV)?
pNN50 Time Domain (Ground Truth)
• Auto Nervous System (ANS) responds to Stress and mental workload changes [4,7].
• HR, and heart Rate Variability (HRV)
affective states of human, and mental workload [4-7].
Heart Rate (HR), Heart Rate Variability (HRV) Measurement (1) Estimating the IBI: (a) Peak finding algorithm
IBI
(b) Calculating the time differences between consecutive peaks
• Experiment [2] • 12 subjects
• Stress and Mental Workload changes affect Heart Rate (HR) and hence Inter-Beat Interval (IBI).
White
White
White
White
60 sec. REST 1
Green
Add Noise 10dB
Red
60 sec. EASY 1
DIFFICULT
60 sec.
pNN50 Time Domain Compare the results
60 sec. REST 2
Refference
pNN50 Time-Frequency Domain
EASY 2
sdfsdfg
60 sec.
pNN50 is calculated for each segment of the game, i.e.,Rest1, Easy 1, Difficult, Rest2, and Easy 2
(2) HR, HRV Analysis [7]: (a) Frequency Domain:
Result FFT of IBI.
(b) Time Domain:
Time-Frequency Domain
IBI related metrics: HR, RMSSD, pNN50, etc.
Time Domain Rectangular
Calculating pNN50 (Conventional Method in time Domain) [7] 20.27 ± 2.67
• Temporal changes of the normal-to-normal (NN) heartbeat intervals. • pNN50 decreases when mental workload (or stress) level increases.
6.91 ± 0.8
Hamming
Chebyshev
5.13 ± 1.64 5.51 ± 1.80
pNN50 Average Error Across all Subjects
pNN50 for one subject’s HR data in time domain Rectangular
Hamming
Chebyshev
where #NN50 is the number of 𝚫𝐍𝐍 > 50ms, and
Error in Estimating IBI [3-7] Incorrect IBI Estimation
Suggested Solution
The suggested solution are: HR Sensor and Movement
Restrict Human Motion
Data Acquisition Error
Reduce Noise (filtering HR Signal)
Peak Detection Error (Noise)
Design Better Peak Detection Algorithm
pNN50 for one subject’s HR data in time-frequency domain
• post-processing methods.
Conclusion
• The entire HR data is required.
• Highest amount error in time domain not reliable on noisy signals • Are not universal.
pNN50 using Combined Time-Frequency (STFT) (a) Average IBI:
• Best setting
𝑪𝟏 = 𝟎. 𝟖 & 𝑪𝟐 = 𝟏. 𝟎 in time-frequency domain (STFT Algorithm) Hamming & Chebyshev
• Across all subjects (12), error in Time Domain is significantly higher than STFT regardless of window shape (P < 0.01).
where is weighted average of dominant component with threshold (𝑪𝟏 ), and (b)
• STFT algorithm outperforms the conventional pNN50 calculation across all the subjects.
References
(c) Assuming: Window shape: Rectangular, Hamming, and Chebyshev. Window size (𝜶) = 2 * IBI (minimum size to calculate pNN50!). Window overlap (ov) = 𝑪𝟐 . IBI (𝑪𝟐 = 𝟏 : 50% window size). (d) Calculate Short Time Fourier Transform (STFT) (time and frequency)
[1] Parsinejad, p., Rodriguez-Vaqueiro, Y., Martinez-Lorenzo, J. A. & Sipahi, R. “Combined Time-Frequency Calculation of pNN50 Metric From Noisy Heart Rate Measurement." Proceedings of ASME Dynamic Systems and Control Conference (DSCC). 2014. [2] Parsinejad, p., Sipahi, R. "A Touchscreen Game to Induce Mental Workload on Human Subjects." 40th Bioengineering Conference (NEBEC). IEEE, 2014. [3] Picard, R. W., 1997. Affective computing. MIT press. [4] Hoover, A., Singh, A., Fishel-Brown, S., and Muth, E., 2012. “Real-time detection of workload changes using heart rate variability”. Biomedical Signal Processing and Control, 7(4), pp. 333–341. [5] Kohler, B.-U., Hennig, C., and Orglmeister, R., 2002. “The principles of software qrs detection”. Engineering in Medicine and Biology Magazine, IEEE, 21(1), pp. 42–57. [6] Berntson, G. G., and Stowell, J. R., 1998. “Ecg artifacts and heart period variability: don’t miss a beat!”. Psychophysiology, 35(1), pp. 127–132. [7] Malik, M., Cripps, T., Farrell, T., and Camm, A., 1989. “Prognostic value of heart rate variability after myocardial infarction. a comparison of different dataprocessing methods”. Medical and Biological Engineering and Computing,27(6), pp. 603–611.
Acknowledgment The work of R. Sipahi is supported by DARPA N66001‐11‐1‐4161 grant. IRB#: 11-11-19
(e) Extract the instantaneous frequency:
Any opinions presented on this poster are those of authors, not those of the funding agency.
Combined Time-Frequency Calculation of pNN50 Metric From Noisy Heart Rate Measurement [1] Payam Parsinejad
Yolanda Rodriguez-Vaqueiro
Jose Angel Martinez-Lorenzo
Problem Definition
Rifat Sipahi (PI)
pNN50 using Combined Time-Frequency (STFT) (Cntd.)
• pNN50 is a metric derived from heart rate (HR) measurements
(f) Estimate the variations of NN intervals:
• Conventionally, it is calculated from successive time periods in peak-to-peak occurrences in HR. • In the case of noisy measurements of HR, however, peak-to-peak detection may not be reliable.
Objective
where 𝐓𝐢 𝐭 instantanouse period at time instant 𝒕, or the reciprocal instantaneous frequency 𝒇𝐢 𝐭 ; and 𝚫𝐭 = 𝐍𝐍 ≈ 𝐓𝐢 . hence,
• A combined time-frequency domain analysis (Short Time Fourier Transform) is presented. • GOAL: To more accurately extract pNN50 metric from noisy HR data. • TEST: An experimental measurement with added noise is used as a benchmark problem to demonstrate the effectiveness of the approach with noticeable improvement over the conventional time domain peak-to-peak detection algorithm.
(g) Calculate pNN50 using:
Validating Our Method in Calculating pNN50
Why Heart Rate (HR), and Heart Rate Variability (HRV)?
pNN50 Time Domain (Ground Truth)
• Auto Nervous System (ANS) responds to Stress and mental workload changes [4,7].
• HR, and heart Rate Variability (HRV)
affective states of human, and mental workload [4-7].
Heart Rate (HR), Heart Rate Variability (HRV) Measurement (1) Estimating the IBI: (a) Peak finding algorithm
IBI
(b) Calculating the time differences between consecutive peaks
• Experiment [2] • 12 subjects
• Stress and Mental Workload changes affect Heart Rate (HR) and hence Inter-Beat Interval (IBI).
White
White
White
White
60 sec. REST 1
Green
Add Noise 10dB
Red
60 sec. EASY 1
DIFFICULT
60 sec.
pNN50 Time Domain Compare the results
60 sec. REST 2
Refference
pNN50 Time-Frequency Domain
EASY 2
sdfsdfg
60 sec.
pNN50 is calculated for each segment of the game, i.e.,Rest1, Easy 1, Difficult, Rest2, and Easy 2
(2) HR, HRV Analysis [7]: (a) Frequency Domain:
Result FFT of IBI.
(b) Time Domain:
Time-Frequency Domain
IBI related metrics: HR, RMSSD, pNN50, etc.
Time Domain Rectangular
Calculating pNN50 (Conventional Method in time Domain) [7] 20.27 ± 2.67
• Temporal changes of the normal-to-normal (NN) heartbeat intervals. • pNN50 decreases when mental workload (or stress) level increases.
6.91 ± 0.8
Hamming
Chebyshev
5.13 ± 1.64 5.51 ± 1.80
pNN50 Average Error Across all Subjects
pNN50 for one subject’s HR data in time domain Rectangular
Hamming
Chebyshev
where #NN50 is the number of 𝚫𝐍𝐍 > 50ms, and
Error in Estimating IBI [3-7] Incorrect IBI Estimation
Suggested Solution
The suggested solution are: HR Sensor and Movement
Restrict Human Motion
Data Acquisition Error
Reduce Noise (filtering HR Signal)
Peak Detection Error (Noise)
Design Better Peak Detection Algorithm
pNN50 for one subject’s HR data in time-frequency domain
• post-processing methods.
Conclusion
• The entire HR data is required.
• Highest amount error in time domain not reliable on noisy signals • Are not universal.
pNN50 using Combined Time-Frequency (STFT) (a) Average IBI:
• Best setting
𝑪𝟏 = 𝟎. 𝟖 & 𝑪𝟐 = 𝟏. 𝟎 in time-frequency domain (STFT Algorithm) Hamming & Chebyshev
• Across all subjects (12), error in Time Domain is significantly higher than STFT regardless of window shape (P < 0.01).
where is weighted average of dominant component with threshold (𝑪𝟏 ), and (b)
• STFT algorithm outperforms the conventional pNN50 calculation across all the subjects.
References
(c) Assuming: Window shape: Rectangular, Hamming, and Chebyshev. Window size (𝜶) = 2 * IBI (minimum size to calculate pNN50!). Window overlap (ov) = 𝑪𝟐 . IBI (𝑪𝟐 = 𝟏 : 50% window size). (d) Calculate Short Time Fourier Transform (STFT) (time and frequency)
[1] Parsinejad, p., Rodriguez-Vaqueiro, Y., Martinez-Lorenzo, J. A. & Sipahi, R. “Combined Time-Frequency Calculation of pNN50 Metric From Noisy Heart Rate Measurement." Proceedings of ASME Dynamic Systems and Control Conference (DSCC). 2014. [2] Parsinejad, p., Sipahi, R. "A Touchscreen Game to Induce Mental Workload on Human Subjects." 40th Bioengineering Conference (NEBEC). IEEE, 2014. [3] Picard, R. W., 1997. Affective computing. MIT press. [4] Hoover, A., Singh, A., Fishel-Brown, S., and Muth, E., 2012. “Real-time detection of workload changes using heart rate variability”. Biomedical Signal Processing and Control, 7(4), pp. 333–341. [5] Kohler, B.-U., Hennig, C., and Orglmeister, R., 2002. “The principles of software qrs detection”. Engineering in Medicine and Biology Magazine, IEEE, 21(1), pp. 42–57. [6] Berntson, G. G., and Stowell, J. R., 1998. “Ecg artifacts and heart period variability: don’t miss a beat!”. Psychophysiology, 35(1), pp. 127–132. [7] Malik, M., Cripps, T., Farrell, T., and Camm, A., 1989. “Prognostic value of heart rate variability after myocardial infarction. a comparison of different dataprocessing methods”. Medical and Biological Engineering and Computing,27(6), pp. 603–611.
Acknowledgment The work of R. Sipahi is supported by DARPA N66001‐11‐1‐4161 grant. IRB#: 11-11-19
(e) Extract the instantaneous frequency:
Any opinions presented on this poster are those of authors, not those of the funding agency.
