i
Heart Rate Variability
Biofeedback Application for
Android
A N D R E A S B E R N D T S S O N
Master of Science Thesis in Medical Engineering Stockholm 2013
Heart Rate Variability Biofeedback Application
for Android
A Master of Science Thesis Performed at STH in Flemingsberg, Stockholm
ANDREAS BERNDTSSON
Master’s Thesis at STH Supervisor: Farhad Abtahi Examiner: Kaj Lindecrantz
TRITA-STH. EX 2013-114
Abstract
Heart rate variability (HRV) is the variations in time between consecutive heart beats, and reflects the functioning of the autonomic nervous system. Not only is HRV a good marker for many physiological disorders, but it is well known that HRV can be altered consciously by different approaches even though it is controlled by the autonomic nervous system. Respiration is an important factor in modulating HRV and this property is utilized in HRV biofeedback, which is a method that aims at increasing heart rate variability. HRV biofeedback systems typically measures heart rate variability and display the parameters on a screen, enabling the user to gain control and increase heart rate variations. In this thesis a software for biofeedback of heart rate variability is presented. The software was implemented for Android and runs on a tablet computer to make the biofeedback system portable and more accessible than most other biofeedback systems. The developed software has proven to be fully functional in real-time providing the user with reliable information. A small pilot study on healthy volunteers has also been made to evaluate the effects of the biofeedback training. These measurements give a preliminary indication that biofeedback session with the proposed solution increases HRV. However, a more comprehensive study with a larger population needs to be carried out in order to confidently confirm the positive effects of biofeedback sessions with the software.
Sammanfattning
Heart rate variability (HRV) är variationerna i tid mellan två efterföljande hjärtslag, och återspeglar autonomiska nervsystemets funktion. HRV är en tydlig markör for många sjukdomar, men det är också välkänt att HRV kan påverkas medvetet trots att det styrs av autonomiska nervsystemet. Andning är en viktig påverkande faktor av HRV och denna egenskap utnyttjas i HRV biofeedback, som är en teknik som syftar till att öka HRV. Typiska system för HRV biofeedback mäter variationerna i hjärtfrekvens och visar upp informationen på en display, vilket låter användaren ta kontroll över denna parameter och öka HRV. I denna uppsats presenteras ett program för biofeedback av HRV. Mjukvaran har implementerats för Android och körs på en surfplatta för att skapa ett biofeedbacksystem som är portabelt och där tillgängligheten är hög, till skillnad från de flesta andra biofeedback system som är beroende av en dator. Programmet som utvecklats har visat sig vara fullt funktionellt i realtid och visar upp pålitliga parametrar för användaren. En förstudie har även utförts för att utvärdera effekterna vid användning av programmet. Dessa mätningar indikerar att biofeedbackträning med den föreslagna lösningen ökar HRV efter användning. En mer omfattande studie med fler personer bör dock genomföras för att ge en tydligare bild av effekterna av träning med detta program.
Contents
1 Introduction 1
1.1 Background and Problem definitions . . . 1
1.2 Thesis Objectives . . . 2
1.3 Thesis Outline . . . 2
2 Theory 3 2.1 Cardiac Physiology . . . 3
2.2 Nervous System . . . 4
2.3 Heart Rate Variability . . . 4
2.3.1 Time-Domain Methods . . . 5
2.3.2 Frequency-Domain Methods . . . 6
2.4 Biofeedback . . . 6
2.4.1 HRV Biofeedback . . . 8
3 HRV Biofeedback Solution 11 3.1 System Overview . . . 11
3.1.1 Bluetooth Device . . . 12
3.1.2 Mobile Device . . . 12
3.2 Signal Processing . . . 13
3.2.1 Peak Detection . . . 13
3.2.2 The Lomb Periodogram . . . 13
3.3 Software Development . . . 14
3.3.1 Data Management . . . 16
3.3.2 User Interface . . . 19
3.4 Using the HRV Biofeedback Software . . . 21
3.5 Experimental Setup . . . 22
4 Results and Discussion 24 4.1 Results . . . 24
4.2 Discussion . . . 25
5 Conclusions 28 5.1 Summary . . . 28
5.2 Suggested Future Work . . . 28
CONTENTS
Bibliography 29
A User Guide 31
List of Figures
2.1 The ECG waveform for two heart beats. . . . 3
2.2 ECG recording showing the variation in time between consecutive heart beats. 5 2.3 ECG (upper plot), RR interval (middle plot) and HRi (lower plot) for 30 sec- onds of recording. . . . 7
2.4 Heart rate and blood pressure reactions to stimuli if the baroreflex does not work. Taken from [5]. . . . 8
2.5 Heart rate and blood pressure reactions to stimuli if the baroreflex works. Taken from [5]. . . . 9
2.6 Heart rate and blood pressure oscillations elicited by the stimulus of respiration. Taken from [5]. . . . 9
3.1 System overview of the biofeedback solution.. . . 11
3.2 Block diagram of Pan and Tompkins peak detection algorithm. . . . 13
3.3 Flowchart for the different threads. . . . 15
3.4 Flowchart of the signal processing in the Bluetooth thread. . . . 17
3.5 Flowchart for ECG thread. . . . 18
3.6 The first view of the biofeedback application. . . . 20
3.7 The second view of the biofeedback application. . . . 21
4.1 Power spectral density for the five-minute recordings. . . . 26
4.2 Tachograms for subject 1 before and after biofeedback. . . . 27
List of Tables
2.1 Frequency domain measures of HRV for short-term recordings (5 min).
Adapted from [7]. . . 7 4.1 Test results from measurements of four test subjects. M1 and M2 de-
notes measurement before and after biofeedback training respectively.
TP denotes LF + HF. . . 25
Abbreviations and units
ANS Autonomic nervous system ECG Electrocardiogram
FFT Fast Fourier transform GUI Graphical user interface HF High frequency
HRi Instantaneous heart rate HRV Heart rate variability
IDE Integrated development environment LF Low frequency
OS Operating system
RFT Resonant frequency training RSA Respiratory sinus arrhythmia SDK Software development kit
SDNN Standard deviation of NN intervals SNS Parasympathetic nervous system SNS Sympathetic nervous system TEB Thoracic electrical bioimpedance UI User interface
VLF Very low frequency
Chapter 1
Introduction
Mobile devices such as smart phones and tablet computers have over the past years invaded our lives. Due to the fast performance improvement of these devices many of the computer’s previous main tasks can today be carried out by the mobile device.
Since the mobility of these devices creates whole new possibilities a vast increase in number of available applications has been observed over the last years. Both old and new markets have been eager to explore how to utilize this new technology to provide new and enhanced services. The medical field is one of many areas that has found its way to the mobile applications arena. Mobile medical applications provide new and cost-effective ways to improve health.
1.1 Background and Problem definitions
The heart does not work like a metronome but instead, the instantaneous heart rate varies around the average heart rate. Heart rate variability (HRV) is the beat-to-beat variations in time between two consecutive heart beats. HRV provides information about the functioning of the autonomic nervous system which controls the heart rate. Specific alterations in HRV have been shown to correlate with a number of diseases and generally HRV provides information about cardiovascular health [7]. Not only is HRV an interesting parameter for the assessment of related diseases, but it has also been shown that HRV can be altered by different approaches.
Biofeedback is a method for improving biological functions by receiving information about the function in question. Biofeedback typically measures a physiological parameter and displays it back to the user on a screen. Most biofeedback systems rely on a computer to process and display the information. HRV biofeedback is a technique which targets to control the variations of the heart rate and the aim is to increase these variations. Since HRV is under influence of the autonomic nervous system many biological components affects it and several protocols for HRV biofeedback exists. However, the most promising approach seems to be resonant
CHAPTER 1. INTRODUCTION
frequency training [13]. This technique is based on the fact that heart rate is affected by our breathing and that low respiration rates increases HRV. Biofeedback for HRV can be used in a broad range of environments such as clinical, workplace and sports for improvement of health and performance [8]. For example, research by Sutarto et al. [13] has shown that HRV biofeedback improves cognitive functions and can therefore be used to increase performance at workplaces.
In order for biofeedback to become a more frequently used method it has to be easy to use and accessibility has to be high. A natural step for biofeedback is to utilize the fast improvement of today’s mobile devices, thus allowing biofeedback systems to become mobile.
1.2 Thesis Objectives
This thesis aims at developing a prototype software for HRV biofeedback. A previ- ously developed Bluetooth device will be used to measure both ECG and respiration.
The software will be implemented on a mobile device that will process the collected data and provide feedback to the user. The main part of the project will focus on implementing methods to analyse and visualize HRV and the respiration signal. The real-time visualization of the data is critical to the application, in order to function desirable. A successful prototype would be fully functional in real-time, provide the user with correct parameters and produce the expected results presented in the theory chapter.
1.3 Thesis Outline
In chapter 2 the theory this thesis is based on is presented. A short review of cardiac physiology and the autonomic nervous system is given. This is followed by a summary of heart rate variability, explaining standards of measurement and physiological interpretations. The last section of this chapter gives a brief introduction to HRV biofeedback training and specifically resonant frequency training.
Chapter 3 presents the proposed solution for this thesis. A brief description of the system overview is followed by the mathematics needed for the signal processing of the application. This is followed by an explanation of the software and how the application was implemented. The next section in this chapter is a description of how to use the biofeedback software, and the last section presents the experimental setup that was used to evaluate the effects of the biofeedback system.
In chapter 4 the results of this thesis is evaluated, and test measurements will be presented. The last chapter contains a summary and suggestions for future works.
Chapter 2
Theory
This chapter provides the basic theory needed to gain insight into the biofeedback solution. A brief review of the heart’s physiology and the nervous system is given.
This is followed by a more thorough introduction to heart rate variability and HRV biofeedback training.
2.1 Cardiac Physiology
The heart is a muscle that pumps blood throughout the blood vessels through repeated contractions, providing the body with oxygen and nutrients. Each con- traction is controlled by a series of electrical signals that is spread through the walls of the heart making the cardiac muscle cells to contract in a coordinated manner. These electrical signals can be measured by applying electrodes to the skin and record the voltage changes over the chest. This kind of recording is called electrocardiogram (ECG) (see figure 2.1).
Figure 2.1. The ECG waveform for two heart beats. This figure illustrates the most important characteristics of the ECG including P wave, QRS complex, T wave and the RR interval between two R peaks.
CHAPTER 2. THEORY
The figure displays the ECG waveform for two heart beats. For a normal heart beat the ECG wave form will have several characteristic parts, which includes a P wave, a QRS complex and a T-wave (and also a small U wave not included in this figure).
Each segment of the ECG wave form corresponds to a certain event in the cardiac cycle. For example, the QRS complex reflects depolarization of the ventricles [12].
The highest point of the R wave is called the R peak and the time between two consecutive R peaks is known as the RR interval, which gives the time between two heart beats.
2.2 Nervous System
The human nervous system can be divided into the somatic and the autonomic nervous system. The somatic nervous system is associated with voluntary control of the body by skeletal muscles. The autonomic nervous system (ANS) is responsible for involuntary actions in the body. The ANS serves as a control system regulating for example heart rate, respiration rate, digestion, perspiration, pupil dilation. The ANS can be further divided into two subsystems, the sympathetic nervous system (SNS) and the parasympathetic nervous system (PSNS). These two branches of the ANS have in many cases opposite effects, meaning that one branch activates a physiological response and the other inhibits it. The SNS is activated in physically demanding and stressful situations causing physiological responses also known as the "fight or flight" response. Sympathetic activation leads to responses like increase in heart rate, increase in blood pressure, pupil dilation, increased sweating. The PSNS is activated during rest and counteracts the responses described by the SNS, thus lowering heart rate, blood pressure etc. Parasympathetic activation allow the body to restore its energy levels and is therefore often referred to as "rest and digest".
Note that both branches of ANS always are at some level active, but one or the other more or less dominant.
The heart receives inputs from both sympathetic and parasympathetic pathways.
The sympathetic nervous system innervates the sinoatrial node (SA node) by spinal nerves which are responsible for increasing the heart rate. The parasympathetic nervous system on the other hand, innervates the SA node by the vagus nerve and causes the heart to decrease the heart rate. Thus, like in many other parts of the body SNS and PSNS counteracts each other and the sum of the sympathetic and parasympathetic inputs will influence the heart rate to either increase or decrease [10]. This interaction between the two branches of ANS is often referred to as sympathovagal balance.
2.3 Heart Rate Variability
Heart rate variability (HRV) is the variations in time between two consecutive heart beats. With an ECG recording HRV can be evaluated by taking the time interval
2.3. HEART RATE VARIABILITY
0.923 sec 0.860 sec
0.980 sec
Figure 2.2. ECG recording showing the variation in time between consecutive heart beats.
between successive R peaks (see figure 2.2). This time interval is called RR intervals, and can also be interpreted as instantaneous heart rate (HRi) by HRi=60/RR beats per minute.
Different HRV characteristics have shown to be associated with several diseases and disorders. HRV differences have been used to detect autonomic neuropathy in patients with diabetes [7]. It has also been shown that changes in HRV has a high correlation with mortality after myocardial infarction [14]. Furthermore, HRV has proven to provide information about the sympathetic and parasympathetic branches of the autonomic nervous system which has strong correlations to cardiovascular health [7].
In 1996 a task force was formed by the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. This task force estab- lished a set of standards of measurement and recommendations for the physiologi- cal interpretation of HRV [7]. The next two sections will summarize some of these standards.
2.3.1 Time-Domain Methods
According to Malik et al. [7] HRV measurements should be categorized into short- term recordings (≤ 5 min), and long-term recordings (24 h). This thesis will be re- stricted to analyzing short-term recordings. Sometimes the term normal-to-normal (NN) interval is used instead of RR interval. NN intervals also refers to the time between adjacent R peaks (or QRS complexes) but emphasizes that the beat in fact
CHAPTER 2. THEORY
is a normal beat resulting from a sinus node depolarization. Figure 2.3 illustrates an example for a 30 seconds segment of an HRV measurement. The upper plot show ECG and the middle plot show the RR intervals extracted from the ECG signal. Plots illustrating RR intervals against time are also known as tachograms.
The lower figure illustrates the instantaneous heart rate (HRi). Note that this is an inverted tachogram, thus the middle plot and the lower plot contains the same information.
Examples of time-domain variables when measuring HRV are mean RR interval, mean heart rate, difference between the longest and the shortest RR interval. One important parameter is the standard deviation of NN intervals (SDNN). For an N samples long series of RR intervals:
SDN N =
s 1
N − 1 XN
j=1(RRj− ¯RR)2, (2.1) where ¯RR is the mean value of the RR intervals. It is important to point out that comparisons of SDNN between different recordings must be of the same length, since HRV increases with the length of the recording [7]. Another parameter is NN50 which is the number of pairs of adjacent NN intervals that differ more than 50 ms. pNN50 is equal to NN50 divided by the total number of NN intervals, thus expressing the percentage of NN50.
2.3.2 Frequency-Domain Methods
In the frequency-domain, power spectral density (PSD) estimations of HRV pro- vides other sets of information. In short-term recordings the spectrum is divided into three main spectral components: very low frequency (VLF), low frequency (LF) and high frequency (HF). The VLF band comprise frequencies ≤ 0.04 Hz, LF ranges between 0.04-0.15 Hz and HF consists of frequencies 0.15-0.4 Hz. Typ- ical frequency-domain measures from short-term recordings are given in table 2.1 [7]. The physiologic interpretation of the VLF component is not well defined and should be neglected when analysing short-term recordings. The HF component is affected by vagal stimulation, thus representing parasympathetic activation. The physiologic interpretation of the LF component has been a matter of debate where some authors claim that LF is a pure marker for sympathetic activity while others view it as a representation of both sympathetic and parasympathetic activity. As a result, the LF/HF ratio is considered by many researchers to reflect sympathova- gal balance, providing a measure of the opposing forces between sympathetic and parasympathetic modulation.
2.4 Biofeedback
Biofeedback is the method of gaining awareness and improving health by providing physiological information from the body. Biofeedback training aims at affect or
2.4. BIOFEEDBACK
125 130 135 140 145 150 155
400 500 600 700
ECG (µV)
125 130 135 140 145 150 155
0.6 0.65 0.7 0.75 0.8
RR interval (s)
125 130 135 140 145 150 155
75 80 85 90 95
Time (s) HRi (bpm)
Figure 2.3. ECG (upper plot), RR interval (middle plot) and HRi(lower plot) for 30 seconds of recording.
Table 2.1. Frequency domain measures of HRV for short-term recordings (5 min).
Adapted from [7].
Variable Units Description Frequency range
5 min total power ms2 The variance of NN intervals approx. ≤ 0.4 Hz VLF ms2 Power in very low frequency range ≤ 0.4 Hz
LF ms2 Power in low frequency range 0.04-0.15 Hz LF norm n.u. LF power in normalized units
LF/(Total Power - VLF) x 100
HF ms2 Power in high frequency range 0.15-0.4 Hz HF norm n.u. HF power in normalized units
HF/(Total Power - VLF) x 100 LF/HF Ratio LF [ms2]/HF [ms2]
improve different selected biological parameters. Typical biofeedback parameters include muscle tension, skin temperature, heart rate, respiration, blood pressure, heart rate variability, blood flow and brain electrical activity. Research by Yucha et al. [15] have shown that biofeedback training is an effective treatment for a variety of physical and psychological diseases. A few examples of disorders where biofeedback has proven to be a successful treatment include: ADHD, chronic pain, epilepsy, headache and hypertension.
CHAPTER 2. THEORY
2.4.1 HRV Biofeedback
HRV biofeedback is a method that aims at maximizing the heart rate variability.
Several protocols for HRV biofeedback exists, but the most promising approach seems to be resonant frequency training (RFT) [13]. The typical procedure for HRV biofeedback involves encouraging the subject to first put himself in a relaxed and positive state of mind, and then the actual training begins. This procedure engages the subject to breathe at a specific respiratory rate called the resonant frequency. It has been shown that due to the nature of the baroreflex, each person has a resonant frequency for breathing where the blood pressure and heart rate oscillates in synchronization which maximizes HRV [5].
The baroreflex is a reflex that regulates blood pressure, by providing a negative feedback loop to the heart. Baroreceptors are stretch-sensitive receptors that senses changes in blood pressure. Baroreceptors are located at different locations in the circulatory system but the most sensitive ones are found in the carotid sinuses and the aortic arch. When blood pressure rises the baroreceptors signals to the autonomic nervous system to decrease the heart rate in order for the blood pressure to reduce. Oppositely, when blood pressure falls ANS is ordered to increase the heart rate which causes blood pressure to rise.
Figure 2.4. Heart rate and blood pressure reactions to stimuli if the baroreflex does not work. Taken from [5].
Lehrer et al. [5] illustrates this phenomenon and explains how it relates to HRV.
Figure 2.4 shows how the system would look like without the baroreflex. An increase in heart rate would result in an elevated blood pressure which is delayed by about 5 seconds. This delay results from inertia in the vascular system so blood pressure changes are slower in taller people than in short people because of larger amount of blood. Without the baroreflex the elevated blood pressure would after some time
2.4. BIOFEEDBACK
Figure 2.5. Heart rate and blood pressure reactions to stimuli if the baroreflex works. Taken from [5].
Figure 2.6. Heart rate and blood pressure oscillations elicited by the stimulus of respiration. Taken from [5].
subside. In figure 2.5 a normal functioning baroreflex is depicted. An elevation in blood pressure is immediately followed by a decrease in heart rate. The resulting fall in blood pressure about 5 seconds later causes a new change in heart rate (this time an increase) which in turn causes a new change in blood pressure. This way
CHAPTER 2. THEORY
stimulus on the baroreflex creates an oscillatory behavior in blood pressure. The baroreflex is constantly effected by stimulus. Each time we inhale we get an increase in heart rate and each time we exhale the heart rate decreases. This phenomenon is called respiratory sinus arrhythmia (RSA) and is understood to be the result of cardiac vagal efferent modulation created by respiration. [4]. Now, if we start to breathe in a pace according to the oscillations in blood pressure the heart rate and the blood pressure will oscillate at the same frequency and will be in a 180◦ phase relationship. This will enhance the amplitude of the blood pressure variations and thereby increasing HRV. The period for one oscillation will be the 5 second delay multiplied by two which gives a period of around 10 seconds, which corresponds to a frequency of approximately 0.1 Hz and a respiration rate of 6 breaths per minute.
By training HRV biofeedback the baroreflex will be exercised and becomes more efficient. Lehrer et al. found that HRV biofeedback training daily for about three months in healthy people results in a long-term increase in the resting baroreflex gain, causing a larger response in heart rate for each mm Hg change in blood pressure. In other words, modulation of blood pressure becomes more efficient [6].
Chapter 3
HRV Biofeedback Solution
3.1 System Overview
SUBJECT
BLUE TOOTH DEVICE
ANDROID DEVICE
E C G
E C G
ECG RESP
Figure 3.1. System overview of the biofeedback solution. ECG and respiration are measured by the Bluetooth device and sent by Bluetooth to the mobile device.
CHAPTER 3. HRV BIOFEEDBACK SOLUTION
The proposed Biofeedback solution consists of a Bluetooth device and an Android application running on a Sony Xperia Tablet Z (see figure 3.1). The Bluetooth device is connected to the user with electrodes and measures ECG and respiration (Thoracic electrical bioimpedance). The Bluetooth device passes the measured data continuously by Bluetooth connection to the Xperia tablet. The HRV Biofeedback software processes the data and provides a graphical user interface (GUI), presenting the parameters of interest back to the user.
3.1.1 Bluetooth Device
The hand-held Bluetooth device measures both ECG and Thoracic electrical bio- impedance (TEB). These two signals are measured by electrodes attached to the skin and connected by cables to the Bluetooth device. The ECG signal is measured across the thorax by two electrodes and are used to extract the RR intervals. The TEB signal provides monitoring of the respiration. In this thesis the terms TEB or respiration signal are used interchangeably. The TEB is measured by four electrodes across the thorax. The TEB measuring is carried out by a high precision impedance converter system named AD5933 from Analog Devices Inc.[1]. This system together with a synchronized ECG make up the total system.
The Bluetooth device is set at a sampling frequency of 200 Hz, and sends the data continuously by Bluetooth connection to the mobile device. To be able to measure TEB the device needs to be calibrated first. This is done with a 50 Ω resistor and is carried out through the software.
3.1.2 Mobile Device
The developed software for this thesis runs on a mobile device, Sony Xperia Tablet Z. Running the software on a mobile device provides a convenient way to use the software not being dependent of a computer which is the case with many other biofeedback systems. The Xperia Tablet Z runs with the operating system (OS) Android 4.2 Jellybean. It has a 10.1" screen which makes it easy to visualize all the important parameters in the software. The software should function on all android devices with Android 3.0 or higher.
Putting aside the aforementioned advantages with using a mobile device, there are of course some drawbacks. Despite the fast improvement of today’s mobile devices, the performance is not yet comparable with a computer. This yields a challenge for developing the software, considering the real-time processing of large amounts of data, and heavy computations for extracting the desired parameters.
3.2. SIGNAL PROCESSING
3.2 Signal Processing
The central part of this application is the signal processing. In order for the appli- cation as well as the biofeedback system as a whole to work satisfactory, the signal processing of the incoming data need not only to be reliant, but also efficient enough to provide the user with real-time data with a minimal delay. In every step during implementation a Matlab prototype has been used to confirm the reliability of the signal processing parts of the software. The two signals used are ECG and respira- tion. However, the ECG is the signal of interest for analysing purposes. The ECG is used to extract the RR interval to be able to calculate the Heart rate variability (HRV) and related parameters both in time-domain and frequency-domain.
3.2.1 Peak Detection
The first step towards HRV extraction is to detect the R peaks in the ECG signal.
To accomplish this the QRS detection algorithms by Pan and Tompkins[9] have been implemented. These algorithms consists of six steps in order to retrieve the R peaks from the raw ECG signal (see figure 3.2). When an R peak has been detected the RR interval is calculated by counting the time since the previous R peak.
Low pass filter
High pass filter
Derivative filter
Squaring function Moving
window integration Thresholding
and peak detection ECG
R peaks
Figure 3.2. Block diagram of Pan and Tompkins peak detection algorithm.
3.2.2 The Lomb Periodogram
The frequency-domain parameters in this software is calculated with a rather un- conventional method. In digital signal processing, spectral components are often computed with Fast Fourier Transform (FFT) or similar methods. The time se- ries which in this case is the tachogram (RR intervals vs time) have a problematic property, it is unevenly sampled. Calculations with the FFT requires the signal to be evenly sampled. Therefore, in order to use FFT on the tachogram, one has to first interpolate the tachogram and evenly resample the signal. This procedure
CHAPTER 3. HRV BIOFEEDBACK SOLUTION
would yield large operations and considering the limited performance capacity of the mobile device, using FFT in this system is not optimal. Instead, there is another method more appropriate to the unevenly nature of the tachogram, called the Lomb periodogram. It has been shown that the Lomb periodogram is better at estimating HRV frequency-domain parameters than FFT [3].
Consider a discrete time-series with N observed samples x0, x1, ..., xN −1, at times t0, t1, ..., tN −1. In order to estimate the power spectral density (PSD), the mean value ¯x and the variance σ2 first have to be defined by standard calculations:
x =¯ 1 N
N −1
X
j=0
xj (3.1)
σ2= 1 N − 1
N −1
X
j=0
(xj − ¯x) (3.2)
The normalized PSD is then calculated with the Lomb periodogram [3]:
P (ω) = 1 2σ2
[Pj(xj− ¯x) cos(ω(tj− τ ))]2 P
jcos2(ω(tj− τ )) +[Pj(xj− ¯x) sin(ω(tj− τ ))]2 P
jsin2(ω(tj− τ ))
! (3.3)
where:
τ = tan−1 P
jsin(2ωtj) 2ωPjcos(2ωtj)
!
(3.4) By multiplying by variance the absolute values are calculated. The Lomb method is based on a minimization of the squared error between xj and a sinusoidal function.
The parameter τ makes the periodogram time-invariant, which means identical PSD estimations are calculated no matter were in time the samples are located [11].
3.3 Software Development
The software in this thesis is developed for Android, which is a Linux-based op- erating system primarily designed for mobile devices such as mobile phones and tablet computers. Android applications are primarily written in the Java program- ming language. Before starting developing, the Android Software Development Kit (SDK) needs to be installed. The Android SDK comprise a set of development tools, including software libraries, debugger, emulator, documentation, sample code and tutorials. To make development easier it is recommended to use an integrated development environment (IDE). In this project I have used Eclipse, which is the officially supported IDE using the Android Development Tools.
3.3. SOFTWARE DEVELOPMENT
ECG Thread Frequency
Intent Service
Bluetooth Thread
TEB
Thread UI Thread
Time-Domain Plot Threads*
PSD Plot Thread
Start
Start Start
Start
ECG
TEB
Peak Detection
Handler ECG,
Update Array
Start
Sleep 50ms
Copy Static Array
Notify Observer Plot Redraw
Handler
Start Measure
Read Data
RR interval, Update
Array Start
PSD Calculations
PSD
Start
Handler
Update Array
Notify Observer Plot Redraw
PSD Param-
eters
Handler
Update Text
Update Progress Bar End
TEB, Udate Array
Update Text
Time- Domain Parameters
Handler
Update Progress Bar Time-Domain
Calculations
Figure 3.3. Flowchart for the different threads. Note that the time-domain plots consists of three different threads (ECG, TEB and tachogram), each working as illustrated in the figure.
CHAPTER 3. HRV BIOFEEDBACK SOLUTION
3.3.1 Data Management
When an application is launched, the system creates a thread of execution called the main thread or the UI thread (User Interface thread). The UI thread is re- sponsible for user interface events such as drawing events, buttons etc. Performing heavy operations on the UI thread leads to poor performance and might even block the UI causing the application to hang from a user’s point of view. [2]. To manage the real-time processing and visualization of data this application is accomplished by multithreading. A thread is a concurrent unit of execution. By implementing new threads responsible for different tasks (also called working threads) the appli- cation is allowed to perform these tasks in parallel making the software much more efficient, and at the same time releasing the UI thread from heavy work keeping the application responsive to user interactions. Figure 3.3 illustrates an overview of the implemented threads, and also a simplified flowchart to show how the data is processed both inside and between the threads. When the application is launched the UI thread is started and the user interface comes to the foreground. By hitting the button Start the other threads are initialized and started (except for Frequency intent service which are discussed later). Handler is a class that allows messages to be sent between threads. Each handler instance corresponds to one thread and handles a message queue of incoming messages. In other words, a handler waits for incoming messages and process them as they arrive. I will describe these threads in more detail now.
The User Interface Thread
The UI thread is started when the application is launched (see figure 3.3). One of the fundamental rules for Android development is to always handle UI events from the UI thread [2]. This means that all the parameters we want to display to the user on the screen has to be updated from this thread. This is managed by implementing handlers or observers depending on the case.
Bluetooth Thread
The Bluetooth thread is responsible for all Bluetooth communication. This thread establishes contact automatically with the Bluetooth device and stay connected until the application is closed. When the user hit Start button a message is sent to the Bluetooth device which in turn starts measuring. The Bluetooth device sends the data in packages of 8 bytes containing one sample of ECG and one sample of TEB. Figure 3.4 illustrates the basic steps for reading a data package. The first byte in each package is a start indicator with value 255. When a start byte has been read the two subsequent bytes contain the ECG sample. The first byte being the least significant byte and the second being the most significant byte. These two bytes gets read as an integer value. The next four bytes contains the TEB as a 32-bit float value. The last byte contains a Cyclic redundancy check (CRC) to
3.3. SOFTWARE DEVELOPMENT
Read byte from Bluetooth stream
If byte == 255
Read ECG, 2 bytes
Read TEB, 4 bytes
Calculate CRC value
Read CRC byte
If CRC is correct Send ECG sample to ECG thread
Send TEB sample to TEB thread True
False
False True
Figure 3.4. Flowchart of the signal processing in the Bluetooth thread.
verify that the data has been correctly processed. If the CRC is correct the ECG sample is sent to the ECG thread and the TEB is sent to the TEB thread.
ECG Thread
Figure 3.5 illustrates the ECG thread more in detail. The ECG thread receives one sample at the time. Each ECG sample gets stored in an array for the possibility to save the raw data. Then the ECG sample gets processed by implementation of Pan and Tompkins peak detection algorithms [9] described in section 3.2.1. When
CHAPTER 3. HRV BIOFEEDBACK SOLUTION
Peak detection
Detected peak?
No
Yes Calculate RR interval
Respiratory
cycle finished? Yes
No Update raw ECG
array
Update RR interval array
Start Frequency Intent Service
Update heart rate, send to UI thread Update RRmax-
RRmin
Update filtered ECG array
Send RRmax-RRmin to UI thread Receive ECG sample
from Bluetooth thread
Figure 3.5. Flowchart for ECG thread.
an ECG sample processing results in a detected peak the peak is first stored in an array which also is plotted in the ECG plot. Then the RR interval is calculated by counting the time since the previous peak, and subsequently stored. The RR interval is sent to Frequency intent service which will do the frequency domain calculations. Then the average heart rate is updated by the new RR interval. After that the difference between the maximum and minimum RR interval is updated.
This value is calculated for each respiratory cycle and is used in the biofeedback training as a measure of HRV oscillations.
3.3. SOFTWARE DEVELOPMENT
TEB Thread
The TEB thread is responsible for handling the respiration signal. Each TEB sample gets stored in a static array just like the ECG signal. The respiration plot also includes a pacer which the user is supposed to breathe in accordance to. This pacer appears as a sine curve in the plot and is controlled from this thread by adapting to the respiration signal.
Frequency Intent Service
The spectral estimations are handled in a different way compared to the other threads. An Android class called IntentService is implemented. This class also runs on its own thread but gets instantiated and subsequently terminated each time it is performed. The advantage of this is that memory used for calculating the spectral components are freed up when it is finished. When Frequency intent service executes the power spectrum is calculated by an implementation of the Lomb periodogram, which is sent to the PSD plot thread. Also the sum of the power in each frequency band is calculated which is sent to the UI thread for updating text fields and progress bar.
Plot Threads
Figure 3.3 illustrates the principle scheme for the plotting. The time-domain plots includes three different plots, including ECG, TEB and RR interval. These plots all work on their own background thread but follows the same principle. ECG, TEB and RR interval are stored in static arrays which means these arrays are accessible from other threads. For example, the TEB samples are stored in a static array in the TEB thread. By declaring this array as static it can be accessed by the TEB plot thread. When measuring starts the TEB thread enters a loop that waits for 50 milliseconds, then copy the static array from the TEB thread and notify observer in the UI thread which redraws the TEB plot. The PSD plot works different from the time-domain plots. Since the spectral data is calculated in an Intent Service all variables are destroyed when the Intent Service is finished. My solution to this is to pass the spectral components in an array to a handler in the PSD plot thread.
So the handler waits for data and when it is received the PSD is updated and the corresponding observer in the UI thread is notified.
3.3.2 User Interface
The design of the user interface (UI) is a critical part for this software in order to not only provide a clear representation of the parameters, but also motivate the user to achieve the desired goals. In Android development, the layout is created in separate files written in XML (Extensible Markup Language). The UI in this application is mainly built up of plots, text views, progress bars and buttons. The plots are implemented by using the open source library AndroidPlot. Four different
CHAPTER 3. HRV BIOFEEDBACK SOLUTION
Figure 3.6. The first view. This is the default view when launching the appli- cation, showing from top to bottom: ECG, respiration, tachogram and power spectral density. Bottom right corner shows the power in different frequency bands. The maximum peak in the power spectrum at 0.1 Hz corresponds to a respiration rate at 6 breaths per minute.
plots are used providing: ECG (filtered), respiration, tachogram and power spectral density. The UI consists of two different views that can be switched between at any time by clicking switch view in the action bar at the top of the screen.
First View
The first view (see figure 3.6), and also the default view when launching the appli- cation, has been used during development as a control view to make sure that the parameters are correctly estimated. This view is not intended for biofeedback train- ing but might be useful for detecting errors during measurement such as electrode displacement. The uppermost plot show the ECG and also the detected R peaks as dots. The second plot from the top shows the respiration signal along with a pacer.
The pacer is a sine curve further described in the next section. The third plot from the top shows the tachogram where each dot corresponds to an RR interval. The plot in the bottom shows the power spectrum for the tachogram. The text views in the bottom right shows the power distribution in the different frequency bands.
3.4. USING THE HRV BIOFEEDBACK SOFTWARE
Figure 3.7. The second view. This is the view used for biofeedback training.
Second View
The second view is the actual biofeedback training view (see figure 3.7). The plots that are being used in this view are the respiration and tachogram. The plot in the top is the respiration plot. The sine curve is the pacer which is able to change to different respiration rates, and also adapts to the maximum and minimum values of the respiration signal. The default setting for the pacer is set at a respiration rate of 6 breaths per minute, and by clicking on the plus or minus buttons the user can increase or decrease the pacer. A beep sound is also generated each time the pacer reaches a maximum or a minimum value. This sound therefore signals to the user the either inhale or exhale. This sound can be switched off by pressing Sound Off. Three different progress bars are being used. The uppermost bar presents RR amplitude which display the difference between the maximum and the minimum RR interval within each respiratory cycle. This bar expands to the right when RR amplitude gets larger. The middle bar represents the normalized power in the low frequency band. The lower bar is representing normalized power in the high frequency band.
3.4 Using the HRV Biofeedback Software
As described in chapter 2, the ultimate objective of HRV biofeedback training is to increase the variations in RR intervals, or in other words create large amplitudes in
CHAPTER 3. HRV BIOFEEDBACK SOLUTION
the tachogram plot. In the biofeedback training view the user can see the amplitude of the tachogram translated into a progress bar, and the goal is to fill the bar as far as possible to the right. This is achieved by breathing at the resonant frequency, which is aided with a green sine curve with the same frequency which acts as a pacer in the respiration plot. The user should breathe accordingly to the pacer in order to reach the resonant frequency. The pacer is by default set at 0.1 Hz which corresponds to a respiration rate at 6 breaths per minute. However, as the resonant frequency does not need to be exactly 0.1 Hz, the user is able change the rate of the pacer increasing or decreasing the respiration rate. The subject is thereby able to test which respiration rate produces the highest RR interval oscillations, finding the resonant frequency.
If the subject breaths consistently at the resonant frequency a peak in the HRV power spectrum will appear at that frequency, and also the power spectrum will be concentrated in the low frequency band. The second progress bar labelled LF corresponds to the normalized power in the low frequency band and therefore this bar also should be maximized, making the bar go right.
3.5 Experimental Setup
To evaluate the physiological effects of the proposed biofeedback solution, a small pilot study on healthy volunteers have been have been executed and analysed. The goal of these tests was to assess the impacts on heart rate variability by the HRV biofeedback system. The test subjects consisted of four healthy and non-medicating adults(1 female and 3 males), in the ages of 29 to 46 years old (mean age was 35.3).
Subject 1 and 2 perform physical exercise sporadically, and subject 3 and 4 exercise on a daily basis. The test procedure consisted of three steps:
1. 5 minutes of HRV recording
2. 10 minutes of HRV biofeedback training 3. 5 minutes of HRV recording
The whole test session was carried out in sitting position. All three steps were executed sequentially without any interruptions. The first five minutes were used to establish a baseline measurement of HRV. During this step the mobile device was not visible to the test subject and sound was turned off. The next step consisted of ten minutes of HRV biofeedback training according to the description in section 3.4.
The participants were instructed to start the session at 6 breaths per minute and were regularly asked if this respiration rate felt comfortable. If any discomfort was experienced the respiration rate was increased. After the ten minutes of biofeedback training a follow-up HRV measurement of five minutes was made (step 1 repeated).
3.5. EXPERIMENTAL SETUP
With these two five-minute recordings an analysis could be made to compare HRV metrics before and after a biofeedback session, and hopefully provide information about the effectiveness of the treatment.
Chapter 4
Results and Discussion
The developed software has been proven to work satisfactory in terms of signal pro- cessing and mathematical calculations. Comparisons with reliable Matlab functions verify that the software calculates and displays the parameters correctly. Also, the software is able to accomplish this in real time as intended.
4.1 Results
The results of the test measurements are presented in table 4.1. Columns M1 and M2 contains the results from the first and the second five-minute measurement respectively (before and after biofeedback training). By comparing column M1 with M2 for each test subject the changes of HRV parameters over the biofeedback session can be interpreted. Note that TP in this table denotes power in low frequency band plus power in high frequency band (LF + HF).
Table 4.1 shows that HRV in fact has increased after the biofeedback session. The standard deviation of NN intervals (SDNN) increased in all subjects, with large increases for subject 1 and 2, and more moderate increases for subject 3 and 4.
pNN50 and heart rate does not show any significant trends in these recordings. A large increase in LF power and total power is present in all test subjects, as expected.
Figure 4.1 illustrates the power spectral density (PSD) for all test subjects. The left column shows the PSD for the five-minute recording before biofeedback training, and the right column shows the PSD for the second recording. It is clear that both the LF component and the total power has increased after the biofeedback session, as expected. The second column shows that breathing at a respiration rate at 6 breaths per minute makes the HRV to oscillate at the same frequency, thus producing a peak in the PSD at 0.1 Hz, and furthermore that these oscillations are present after the biofeedback session has ended. As a complement to these analysis a more intuitive illustration of the changes of HRV is given in figure 4.2. This figure displays the tachogram for subject 1. The upper plot shows the tachogram for the
4.2. DISCUSSION
Table 4.1. Test results from measurements of four test subjects. M1 and M2 denotes measurement before and after biofeedback training respectively. TP denotes LF + HF.
Measures Subject 1 Subject 2 Subject 3 Subject 4
M1 M2 M1 M2 M1 M2 M1 M2
SDNN (ms) 47.4 71.7 58.5 88.7 48.0 50.9 53.3 56.4
pNN50 (%) 2.1 21.5 35.5 39.3 11.4 13.7 19.4 18.0
HR (bpm) 84 73 58 58 62 62 65 68
LF (ms2) 363 1957 514 2405 372 558 587 1389
HF (ms2) 130 475 545 567 375 436 167 135
TP (ms2) 493 2432 1059 2972 747 994 754 1524
first recording and the lower plot shows the second recording after biofeedback, illustrating an obvious increase in amplitude.
4.2 Discussion
The conclusions that can be made from these tests are that HRV is increased by the use of the proposed biofeedback solution. SDNN and total power is increased for all four test participants. However, while subject 1 and 2 shows significant increases, subject 3 and 4 shows more moderate increases, especially considering SDNN. A possible reason for this might be that subject 3 and 4 ought to be considered more well trained and had also exercised previously on the day of measurement. Since physical exercise increases HRV an explanation might be that the baseline HRV for subjects 3 and 4 already was relatively high for these individuals, and subsequently the biofeedback training had less effect.
It is important to point out that the protocol for this biofeedback session is not optimized. First of all, the resonant frequency was never determined in these tests, but instead all subjects used a respiration rate of 6 breaths per minute. This means that the optimal respiration rate for each participant was not established, and higher improvements of HRV should be expected if the resonant frequency is used. Also the length of the biofeedback session could be questionable, and longer sessions of training might provide greater results. Furthermore, like any other training our ability improves the more we train and test measurements over multiple sessions should also lead to greater increases of HRV.
CHAPTER 4. RESULTS AND DISCUSSION
0 0.1 0.2 0.3 0.4 0.5
0 0.02 0.04 0.06
PSD (s2 /Hz)
Subject 1 before biofeedback
0 0.1 0.2 0.3 0.4 0.5
0 0.02 0.04 0.06
Subject 1 after biofeedback
0 0.1 0.2 0.3 0.4 0.5
0 0.02 0.04 0.06 0.08
PSD (s2 /Hz)
Subject 2 before biofeedback
0 0.1 0.2 0.3 0.4 0.5
0 0.02 0.04 0.06 0.08
Subject 2 after biofeedback
0 0.1 0.2 0.3 0.4 0.5
0 0.005 0.01 0.015 0.02
PSD (s2 /Hz)
Subject 3 before biofeedback
0 0.1 0.2 0.3 0.4 0.5
0 0.005 0.01 0.015 0.02
Subject 3 after biofeedback
0 0.1 0.2 0.3 0.4 0.5
0 0.02 0.04 0.06
Frequency (Hz) PSD (s2 /Hz)
Subject 4 before biofeedback
0 0.1 0.2 0.3 0.4 0.5
0 0.02 0.04 0.06
Frequency (Hz) Subject 4 after biofeedback PSD
VLF LF HF
Figure 4.1. Power spectral density for the five-minute recordings. Left column illustrates PSD before biofeedback session and the right column shows PSD after biofeedback.
4.2. DISCUSSION
0 50 100 150 200 250 300
0.6 0.7 0.8 0.9 1 1.1
RR interval (s)
0 50 100 150 200 250 300
0.6 0.7 0.8 0.9 1 1.1
Time (s)
RR interval (s)
Figure 4.2. Tachograms for subject 1 before and after biofeedback training.
Upper plot illustrates RR intervals for the first five-minute recording and the lower plot shows RR intervals for the second recording after biofeedback session.
The performed tests indicates that the HRV biofeedback solution has positive effects on heart rate variability. However, the biofeedback protocol should be improved by optimizing the parameters discussed previously. A more comprehensive test consisting of a larger amount of test subjects will also reveal a deeper understanding of the effects from HRV biofeedback training.
Chapter 5
Conclusions
5.1 Summary
In this thesis a software for HRV biofeedback has been developed. The software has shown to provide accurate HRV parameters in real-time back to the user. By imple- menting the software on a tablet computer for Android the biofeedback solution is mobile and expands the possibilities for the use of biofeedback training. Performing the required signal processing and visualization in real-time on a mobile device has its limitations but has proven to be manageable. To evaluate the biofeedback sys- tem as a training method a set of test measurements have been performed. These measurements indicates that the use of this biofeedback solution increases HRV after training session, but more extensive tests should be performed with a larger population to provide better knowledge of the impacts from biofeedback sessions of the proposed system.
5.2 Suggested Future Work
For future work with the software I recommend to implement a function for finding the resonant frequency for each user. This can be achieved by implementing a procedure where the user breaths at different respiration rates in an appropriate interval, and measure which rate produces the highest HRV. More generally, new approaches for how to visualize the biological parameters should also be considered.
Perhaps the sound for aiding the respiration rate is enough and the screen can be used for other purposes. An example could be to implement games where the outcome depends on HRV parameters which would take the software to another level.
Bibliography
[1] Analog Devices. AD5933 Datasheet Document. www.analog.com/ad5933, 2013.
[2] Android Developers. Processes and Threads.
http://developer.android.com/guide/components/processes-and-threads.html, 2013.
[3] GD Clifford. Signal processing methods for heart rate variability. PhD thesis, Department of Engineering Science, University of Oxford, 2002.
[4] J Hayano, F Yasuma, A Okada, S Mukai, and T Fujinami. Respiratory sinus arrhythmia a phenomenon improving pulmonary gas exchange and circulatory efficiency. Circulation, 94(4):842–847, 1996.
[5] P Lehrer and E Vaschillo. The future of heart rate variability biofeedback.
Biofeedback, 36(1):11–14, 2008.
[6] P Lehrer, E Vaschillo, B Vaschillo, S Lu, D Eckberg, R Edelberg, WJ Shih, Y Lin, T Kuusela, K Tahvanainen, et al. Heart rate variability biofeedback increases baroreflex gain and peak expiratory flow. Psychosomatic Medicine, 65(5):796–805, 2003.
[7] M Malik, AJ Camm, JT Bigger, G Breithardt, S Cerutti, RJ Cohen, P Coumel, EL Fallen, HL Kennedy, RE Kleiger, et al. Heart rate variability: standards of measurement, physiological interpretation and clinical use. task force of the european society of cardiology and the north american society of pacing and electrophysiology. Circulation, 93(5):1043–1065, 1996.
[8] R McCraty and D Tomasino. Heart rhythm coherence feedback: A new tool for stress reduction, rehabilitation, and performance enhancement. In Proceedings of the First Baltic Forum on Neuronal Regulation and Biofeedback, pages 2–4, 2004.
[9] J Pan and W Tompkins. A real-time qrs detection algorithm. IEEE Transac- tions on Biomedical Engineering, BME-32(3):230–236, 1985.
[10] D Purves, GJ Augustine, D Fitzpatrick, WC Hall, AS LaMantia, JO McNa- mara, and LE White. Neuroscience. Sinauer Associates, 2008.
BIBLIOGRAPHY
[11] L Sörnmo and P Laguna. Bioelectrical signal processing in cardiac and neuro- logical applications [electronic resource]. Academic Press, 2005.
[12] G Stouffer. Practical ECG interpretation: clues to heart disease in young adults. John Wiley & Sons, 2009.
[13] AP Sutarto, MNA Wahab, and NM Zin. Heart rate variability (hrv) biofeed- back: A new training approach for operator’s performance enhancement. Jour- nal of industrial engineering and management, 3(1):176–198, 2010.
[14] MM Wolf, GA Varigos, D Hunt, and JG Sloman. Sinus arrhythmia in acute myocardial infarction. The Medical Journal of Australia, 2(2):52–53, 1978.
[15] C Yucha and D Montgomery. Evidence-based practice in biofeedback and neu- rofeedback. AAPB Wheat Ridge, CO, 2008.
Appendix A
User Guide
Following is a short guide for using the HRV biofeedback software.
1. Switch on Bluetooth device.
2. Start the HRV biofeedback application.
3. Connect a 50 Ω resistor to the TEB inputs/outputs (labelled RESP) on the Bluetooth device. Then press button CALIBRATE in the top right of the screen. This step only needs to be taken the first time after the device has been turned off.
4. Connect electrodes according to the figure on the device and connect the cables in the corresponding inputs/outputs.
5. Press Start button on the screen.
6. Press SWITCH VIEW button at the top of the screen and begin biofeedback training according to the previous section.
7. When finished, press Stop button. A dialog window will appear asking if the session should be saved. If you click Yes a new dialog window will appear.
Enter your name and click Save.
8. To start a new session, repeat the instructions from step 5. To close the application, click the home button at the bottom of the screen.
