- Open Access
- Total Downloads : 11
- Authors : Soren A . Davidsen, M . Padmavathamma, E . Sreedevi, M . Nagaraju Naik
- Paper ID : IJERTCONV2IS15033
- Volume & Issue : NCDMA – 2014 (Volume 2 – Issue 15)
- Published (First Online): 30-07-2018
- ISSN (Online) : 2278-0181
- Publisher Name : IJERT
- License: This work is licensed under a Creative Commons Attribution 4.0 International License
Smart-Phone Based Blood-Pressure Devices: A Review
SÃ¸ren A . Davidsen
Dept. of Computer Science,
E . Sreedevi
Dept. of Computer Science, Sri Venkateswara University, Tirupati517501, India
AbstractHypertension is a dangerous condition, but simple to diagnose by measuring the blood pressure. Smart- phones are ubiquitous, so a blood pressure device comprising only the smart-phone can make diagnosis commonly available without the need to procure access to a special blood pressure device. This paper reviews recent advances in blood-pressure measurements, specifically with focus on those implemented for smart-phones. Finally, future possible directions are mentioned.
Keywordssmart phone, blood pressure, medical device, sensor
Hypertension is a dangerous condition, which can lead to cardiovascular disceases, most commonly known is stroke in the heart. In 1990, India saw 1.3 million deaths from stroke, directly related to hypertension, and research suggests this number will only increase in the coming years . Diagnosing hypertension in the population is the first and most important step in order to decrease the risk of hypertension-related diseases. To diagnose hypertension is relatively simple, as it requires primarily knowledge of a persons blood pressure (see Tab. I), thus depends on the availability of a device to measure the blood pressure.
Prof. M . Padmavathamma
Professor and Head
M . Nagaraju Naik
Dept. of Computer Science, Sri Venkateswara University,
Bluetooth enabled ECG sensor. (b) Microphone-based digital Source: Clockwise.de stethoscope. Source: 
Fig. 1: External sensor examples.
listen for blood flow (so-called Korotkoff sounds); 2) the oscillometric method, typically used in-home, follows same principle, but uses electronic analysis of the blood flow instead of human listening.
Both the common methods are cuff-based, meaning they
require inflation and pressure applied. This is often disturbing the patient, leading to an overestimation of blood pressure. Thus, cuff-less methods are sought for ease and possibility of continous monitoring.
Measurement of blood pressure has long been an active field of research. Recently methods have been developed with
focus on digital sensors, signal processing and machine
Blood pressure category (mmHg) (mmHg) Normal < 120 and < 80
Prehypertension 120 139 or 80 89
High Blood Pressure
learning. The remainder of the paper is organized as follows: Section II discus which sensors are used together with their
respective signal processing methods. Section III discus the
(Hypertension) Stage 1 High Blood Pressure
140 159 or
methods used for producing BP output from available sensor
(Hypertension) Stage 2 > 160 or > 100
Hypertensive Crisis > 180 or > 110
TABLE I: Blood pressure categories. Source: American Heart Association
Blood pressure is typically recorded as two numbers, written as a ratio like this: Systolic/Diastolic.
The top number (systolic), which is also the higher of the two numbers, measures the pressure in the arteries when the heart beats (when the heart muscle contracts). The bottom number (diastolic), which is also the lower of the two numbers, measures the pressure in the arteries between heartbeats (when the heart muscle is resting between beats and refilling with blood).
Common blood pressure methods are: 1) the auscultatory method used in a clinical setting where a stethoscope and a sphygmomanometer are used to pressurize the upper arm, and
inputs. Finally, in Section IV, future improvements are men-
tioned, and the paper is concluded.
SENSORS AND SIGNAL PROCESSING
Observing the patient requires input, either specified by the patient herself, or by the use of sensors. Traditional sensors such as ECG and digital stethoscope (see Fig. 1) have been used together with smart-phones for measuring BP .
For the smart-phone based BP device, several sensors have been proposed, which are available in most of todays smart- phones:
Accelerometer, senses movement of the phone. Microphone, audio sensor.
Touch-screen, senses finger-movement and finger-pressure on the screen. Camera, visual sensor.
PPG sensor. The camera records (b) Microphone sensor. The micro- oxidation of the fingertip. The LED phone is positioned near the heart illuminates the fingertip in order to for better signal.
produce a better signal.
Fig. 2: Smart-phone sensors
Accelerometer as heart-beat sensor: The accelerometer provides the smart-phone with knowledge about motion, such as acceleration forces or rotation. In , the accelerometer was demonstrated to be able to measure the heart-beat at the heart, by strapping an iPhone to the patients chest. The accelerometer provides information about acceleration along X, Y and Z axis, and depending on the position of the patient, the heart-beat must be detected along the correct axis, or, possibly a combination thereof.  showed that the accuracy of detecting heart-beat was in [0:76; 0:98] when the patient was not moving. When moving (e.g. walking) the signal noise from stepping overshadowed the heart-beat.
Camera as photoplethysmography sensor: The camera sensor can be used as a photoplethysmography (PPG) sensor. PPG is sensing change in the oxidation of a body-part, as an effect of the pulse wave. PPG with the smart-phone is based on the flash LED and the camera. The LED illuminates the fingertip, providing a clearer view of the oxidation, and the camera as sensor, see Fig. 2. The change in illumination in the camera is the signal SP (t), as shown below:
where the resolution of the captured image is h w, Iij(t) is the
the microphone is prone to outside noise, this requires silence and signal filtering to extract the heart sound. Fig. 5(a) uses the microphone to measure pulse wave at the heart.
Processing sensor signals
Sensor signals are typically noisy. For example, an audio- waveform from a microphone encodes several frequencies from surrounding noise, and a plethysmographic sensor gives a subtlely different illumination for each frame, because of changed light conditions in the room of the patient.
The methods for measuring blood pressure need to know the pulse-wave, and processing the raw signal is done in order to get a smooth curve, where the pulse-wave can be confidently detected.
Band-pass filtering: Band-pass filtering is a combination of a low-pass filter and a high-pass filter. The low-pass filter allow wavesforms in the signal with a frequency below a specified threshold to be passed through without modification, while higher frequencies will be removed by filtering them out. A high-pass filter does the opposite, except allowing high frequencies to pass.
Depending on the application, typically band-pass filtering is used over audio or ECG waveforms where the interesting frequencies in the signal are known, and unwanted frequencies are likely to appear.
Smoothing: The purpose of smoothing is to minimize the effect of noise in the signal. Two types of smoothing are commonly used. Moving-average smoothing SM assigns the signal value to the average of a window of input signal values, defined as:
SM (t) = (w1S(t) +w2S(t 1) + ::: + wnS(t (n 1))) ; w =
(1/n1/n ) (2)
where S(t) is the input signal, and n is the window-size.
Exponential smoothing is another popular smoothing func- tion, which is defined recursively as:
S0 (t) = S(t) + (1 )S0 (t 1); S0 (0) = S(0)
pixel i; j in the camera image at t and green is a function to
E E E
extract the green color intensity from the pixel.
PPG is a very commonly used sensor ,  due to its robustness towards interfere from other sensors, however, smart-phones usually contains only one such sensor (LED is not available for front camera).
Touch screen as a strain sensor: The touch screen in smart-phones is commonly used as a navigational tool, by to measuring the position of the finger on the screen. However, they also allow to measure variation in pressure. This has been
again S(t) is the input signal and is the smoothing factor. Fig. 3 exemplifies the use of smoothing in an ECG signal.
Peak detection: Peak detection algorithms are deployed for analysis of the filtered and smoothed signal, to find the timing of the pulse-wave. Peak algorithms is a large topic and outside the scope here, however a commonly threshold based method is given .
The threshold method is based on the observation that between each peak, there must be a trough. Given a threshold each peak and trough can be defined as:
used in Fig. 5(b) where the finger is placed on top of the
screen. When the pulse wave reaches the finger, it gives a
XPj XTi +XPj
slight change in pressure, which is the output signal of this sensor.
Microphone as un-amplified stethoscope: The micro- phone in the smart-phone can be used as a stethoscope. As with a normal stethoscope, the faint heart sounds are
XTi XPjXTj \ XPj+1 XTi (5)
where XPj is the jth peak in X and XTi is the ith trough. Noise in the signal below the threshold does not trigger another peak or trough, and the definition ensures that a trough always follows a peak, and vice versa.
Fig. 3: Prototype ECG signal noise removed by exponential smoothing
A method describes how input from sensors can be used to calculate the output BP. The methods found in literature are based on PWTT, machine learning, or hybrids thereof. In the following the PWTT method is described and variations in literature given, after which the machine learning and hybrid methods found are mentioned.
Pulse wave transit time
Blood pressure is measured as systolic blood pressure and
Fig. 4: Example PWTT measured as onset of pulse wave in the ECG signal at heart, and the photoplethysmography signal at fingertip. Source: 
2) Diastolic blood pressure: To estimate the diastolic blood pressure, the following relation between systolic blood pres- sure, diastolic blood pressure PD and pulse pressure PP is used:
PD = PS PP (8)
The pulse pressure can be calculated from knowing the heart rate HR (or, pulse rate) number of beats per minute as reported by the number of peaks per minute from either of the used sensors as follows:
BSA = 0:007184 Weight0:425 Height0:725 (9)
SV = 6:6 + 0:25(ET 25) 0:62HR (10)
+ 40:4BSA 0:51Age
diastolic blood pressure, and the pulse wave transit time
(PWTT) method is a relatively new method thereof, which has
(0:013Wt 0:007Age 0:004HR) + 1:307
now been validated quite well in the medical community , .
When the heart pumps it triggers a pulse wave which travels from the heart along the arterial tree. One of the parameters that decides the velocity of the pulse wave, is the arterial stiffness, i.e. the blood pressure. A shorter pulse wave transit time means a higher blood pressure. In order to measure PWTT the timing of the pulse wave must be measured at two points, one at the origin of the wave (the heart) and one away from the origin (typically the fingertip). An example is shown in Fig. 4, where ECG and photoplethysmography is used as pulse wave signals.
Several methods exists for estimating blood pressure from the measured PWTT f : PWTT ! P . A relatively complete method was given by , which requires only few other parameters than the PWTT. Their method is outlined below.
1) Systolic blood pressure: The change in systolic blood pressure has a linear relationship with the change in the pulse transit:
PS = 0:425PWTT (6)
By integration and experimental regression they found the following relation:
where ET is the time between first and second heart sound, Height, Weight and Age are features of the patient and Wt is the window size used in calculating HR.
While s given method is very complete, similar works ,  agree on the linear relationsip between PWTT and PS, but offer different methods of estimating actual parameters. does not estimate at all but simply assigns constants empirically, arguing that for practical uses it is good enough.
Machine learning methods
Machine learning methods have also been applied. Machine learning by nature does not measure BP, but predicts it. Machine learning in this context is a process in which a model is learned from examples, rather than defining the model in details. f : PWTT ! P is the model that should be learned.
Given a set of examples D and a hypothesis space H:
D = f(x1; y1); (x2; y2); :::; (xn; yn)g (X Y)n (12)
H fh : X ! Yg (13)
where each example (x; y) belongs to their respective domains X and Y, the learning algorithm selects an optimal hypothesis f ,
PS = 0:425 PWTT + 214 (7)
f 2 arg
pH ZX Y
l(h(x); y) L( ; )
Feature group Features Uses Patient body features Age, weight, height , 
Body mass index, waist, hip 
environmental Smoking habits 
Pulse wave features Position and amplitude of onset, Systolic , 
Peak, Point of Inflection, Dicrotic wave, sys-
tolic diastolic time difference
TABLE II: Features used in machine learning based BP prediction.
where l is a loss-function and L is an unknown data generating process representing the actual learning algorithm. The output function f will, given an unseen example predict the BP.
Examples of learning algorithms are: ANN (artificial neu- ral network), SVM (support vector machine) and BayesNet (Bayesian Network). As their working is a huge topic, please refer to  for further details.
Several ways of predicting are given:
Binary classification , : Predict whether the patient has high blood pressure or not, i.e. f : PWTT !
P; P 2 f0; 1g
Multi-class classification : Predict which hyper- tension group the patient is in, see e.g. Tab. I, for which P 2 f0; 1; 2; 3; 4g.
Regression : Predict the values for PS and PD of the patient, i.e. P 2 R
In the PWTT method, the PWTT itself is required, along with several contextual attributes of the patient. In a machine learning setting these are called features. Tab. ?? gives an overview of the features proposed in existing methods.
Several hybrids and alternative methods have been pro- posed, which may be interesting for future smart-phone based BP devices. Some work has been focused on new sensors directly measuring the BP:  present a sensor, which uses a specially designed strain sensor to measure BP. The sensor is small and cuff-less meaning it would fit in a smart-phone design. Likewise,  proposed an extended PWTT method based on an ICG sensor to measure volume more precisely at the heart, together with PPG at the fingertip. Their ICG sensor is bluetooth enabled and can be used with smart-phones.
Without additional sensors,  has proposed a PWTT method which used oly the pulse wave at the fingertip to es- timate PWTT by considering peaks in the twice-differentiated pulse wave and a learning phase where PWTT is measured at heart-height and one arm-length below heart-height, by moving the finger up and down.
DISCUSSION AND CONCLUSIONS
In this section the accuracy of current methods is evaluated, based on which future directions are discussed, and the paper is concluded.
The new BP measurement devices show promising reported accuracy in measuring values consistent with traditional BP measurement (such as a sphygmomanometer or a cuff-based oscillatory device). For the methods mentioned in this paper, the accuracy has been collected into the overview shown in Tab. III.
Method Sensors Reported accuracy Pulse wave transit time
(PWTT) PPG and Microphone 0:9
PPG and digital stetho- [0:94; 1:0]
2 x PPG  [0:94; 0:99]
PPG and ECG  [0:91; 0:98]
Differentiated PWTT  PPG [0:83; 0:92]
Pulse wave PPG ,  [0:88; 0:97]
Pulse wave and patient features PPG  [0:75; 0:95]
Patient features  (Nil) [0:72; 0:85]
TABLE III: Overview: Reported accuracies for discussed methods and sensors. Accuracy means correlation with tra-ditional BP measurement.
The accuracy for all methods is within [0:72; 1:0]. The most accurate methods are based on external sensors , , but alternative methods considering only a single PPG sensor along with contextual features are also showing good accuracy ,  under ideal conditions.
While the reported accuracies are already quite good, there is still room for improvement under less-than-ideal conditions. Adding more sensors allows to measure the heart-beat and finger pulse more precisely. In the smart-phone based device it is however not always a solution, as the sensors signals may overlap.
Human computer/device interface
All blood pressure devices display only the same three outputs: 1) pulse; 2) diastolic pressure; 3) systolic pressure (see two example Android based applications in Fig. 5). For the health-care professional, this information is essential and based on the training, a proper diagnosis can be made. For the patient the story may be different, remembering the limits for a normal condition, or, when a visit to the doctor is imminent, may not be possible.
Another interesting direction in human computer/device interaction is that some methods may offer continuous mea- surement of BP. This could give room for research into methods of alerting the patient (or doctor), should a given BP event occur.
Blood pressure variability
It has been shown that the blood pressure is not static, but rather continuous . While standard deviation of the accu- racy is not mentioned here (lack of data in the referred papers), methods with a wide interval suggests unstable performance, possibly based on other factors, as suggested by the blood pressure variability.
Source: Instantbloodpres- (b) Source: Purepush.org sure.com
Fig. 5: Screen-shots from two Android blood-pressure mea- surement applications.
The use of machine learning methods and environmental sensors (including Internet, such as Facebook or Twitter inte- gration) in this area could be interesting, to give a more precise measurement of blood pressure, considering the variability.
In conclusion, this paper has introduced the problem of hypertension, and reviewed current trends in smart-phone based blood-pressure devices to make hypertension monitoring easier. Used input sensors have been described and methods reviewed; found to be of good accuracy to benefit the hyperten-sion community. Finally, possible future directions have been given.
R. Gupta, Trends in hypertension epidemiology in india, Journal of Human Hypertension, vol. 18, no. 2, pp. 7378, 2004.
I. P. Singh, S. K. Mishra, and R. Daman, Innovative low cost mhealth tool kit, in Proc. 1st Int. Conf. on Intelligent Infrastructure, 2012.
V. Chandrasekaran, R. Dantu, S. Jonnada, S. Thiyagaraja, and K. Subbu, Cuffless differential blood pressure estimation using smart phones,
IEEE Transactions on Biomedical Engineering, vol. 60, no. 4, pp.
1080 1089, April 2013.
H. M. A. Younessi, M. A. Khalilzadeh, and S. Joharinia, Continuous and cuffless blood pressure monitoring using ecg and spo2 signals, Journal of Biomedical Physics and Engineering, vol. 4, no. 1, pp. 27 32, 2014.
Y.-H. Lin, Z.-M. Lin, Y.-T. Hsu, and H.-Y. Ku, A mobile intelligent blood pressure monitor based on the android smartphone, in Security- Enriched Urban Computing and Smart Grid, 2011, vol. 223, pp. 178 187.
A. Rashkovska, I. Tomasic, and R. Trobec, A telemedicine application: Ecg data from wireless body sensors on a smartphone, in Proc. 34th International Convention MIPRO, May 2011, pp. 262265.
M. Wu and G. Der-Khachadourian, Digital stethoscope: A portable, electronic auscultation device, 2012. [Online]. Avail-able: http://people.ece.cornell.edu/land/courses/ece4760/FinalProjects/ s2012/myw9 gdd9/myw9 gdd9/
S. Kwon, J. Lee, G. S. Chung, and K.-S. Park, Validation of heart rate extraction through an iphone accelerometer, in Proc. Annual Int. Conf. of the IEEE Engineering in Medicine and Biology Society (EMBC), Aug 2011, pp. 52605263.
A. D. Choudhury, A. Visvanathan, R. Banerjee, A. Sinha, A. Pal,
C. Bhaumik, and A. Kumar, Heartsense: Estimating blood pressure and ecg from photoplethysmograph using smart phones, in Proc. 11th ACM Conf. on Embedded Networked Sensor Systems, 2013, pp. 88:1 88:2.
H. Li and H. Zhao, Systolic blood pressure estimation using android smart phones, in Proc. 6th Int. Conf. on Biomedical Engineering and Informatics (BMEI), Dec 2013, pp. 260264.
E. Monte-Moreno, Non-invasive estimate of blood glucose and blood pressure from a photoplethysmograph by means of machine learning techniques, Artificial Intelligence in Medicine, vol. 53, pp. 127138, 2011.
S. Song, J. Cho, H. Oh, J. Lee, and I. Kim, Estimation of blood pressure using photoplethysmography on the wrist, Computers in Cardiology, vol. 36, pp. 741744, 2009.
N. R. Gaddum, J. Alastruey, P. Beerbaum, P. Chowienczyk, and
T. Schaeffter, A technical assessment of pulse wave velocity algorithms applied to non-invasive arterial waveforms, Annals of Biomedical Engineering, vol. 41, no. 12, pp. 26172629, 2013.
M. L. Jacobson, Auto-threshold peak detection in physiological sig- nals, in Proc. 23rd Ann. Int. Conf. of the IEEE in Medicine and Biology, 2001, pp. 21942195.
H. Gesche, D. Grosskurth, G. Kuchler,Â¨ and A. Patzak, Continuous blood pressure measurement by using the pulse transit time: comparison to a cuff-based method, European Journal of Applied Physiology, vol. 112, no. 1, pp. 309315, 2012.
Data Mining. Morgan Kaufmann Publishers, 2011.
H. F. Golino, L. S. de Brito Amaral, S. F. P. Duarte, C. M. A. Gomes,
T. de Jesus Soares, L. A. dos Reis, and J. Santos, Predicting increased blood pressure using machine learning, Journal of Obesity, vol. 2014, p. 637635, 2014.
V. G. Almeida, J. Vieira, P. Santos, T. Pereira, H. C. Pereira, C. Correia,
M. Pego, and J. Cardoso, Machine learning techniques for arterial pressure waveform analysis, Journal of Personalized Medicine, 2013.
T.-Y. Tu, P.-P. Chao, and Y.-P. Lee, A new non-invasive cuff-less blood pressure sensor, in Proc. IEEE Sensors, Nov 2013, pp. 14.
J. Sola, M. Proenca, D. Ferrario, J.-A. Porchet, A. Falhi, O. Grossen- bacher, Y. Allemann, S. Rimoldi, and C. Sartori, Noninvasive and nonocclusive blood pressure estimation via achest sensor, IEEE Transactions on Biomedical Engineering, vol. 60, no. 12, pp. 3505 3513, Dec 2013.
Instant blood pressure, 2014. [Online]. Available: http:
J. S. Floras, Blood pressure variability: A novel and important risk factor, Canadian Journal of Cardiology, vol. 29, pp. 557563, 2013.