Development of a Continuous Blood Pressure Monitoring System

based on Pulse Transit Time and Hemodynamic Covariates

Yiming Zhang

, Congcong Zhou

, Zhongyi Huang

and Xuesong Ye

1,2

Department of Biomedical Engineering, Key Laboratory for Biomedical Engineering of Education Ministry,

Zhejiang University, Hangzhou 310027, China

Cyrus Tang Center for Sensor Materials and Applications, Zhejiang University, Hangzhou 310058, P.R. China

Keywords: Blood Pressure, Non-invasive, Pulse Transit Time, Synchronization, Wearable.

Abstract: There were many studies showing the relation between pulse transit time (PTT) and blood pressure (BP).

Besides, hemodynamic covariates may also contribute to BP values. Our previous study has proposed a BP

model based on PTT, HR, stiffness index (SIx) and descent time (DT), which has been validated on the Multi-

parameter Intelligent Monitoring for Intensive Care (MIMIC) database. In this article, we present a prototype

cuff-less monitoring device for non-invasive estimation of BP, which can obtain both electrocardiogram

(ECG) and finger plethysmograph (PPG) signals synchronously. The model proposed above has been

validated by using ECG and PPG records from 22 healthy subjects with no cardiovascular disease and

hypertension, and the error of BP estimation was 0.002±8.544 mmHg for SBP, 0.005±6.690 mmHg for

DBP. The reliability of this method in long-term BP monitoring was further verified by studying the data of

one individual for 28 days, while the error was 5.204±5.462 mmHg for SBP, 2.714±4.756 mmHg for DBP

without calibration. The results show that the model could estimate the BP value within the acceptable error

range based on this study, which is extremely close to AAMI's standard (5±8mmHg) and consistent with the

cuff-method. The proposed ultra-low power, wearable, time-synchronized prototype monitoring device with

an embedded hemodynamic covariate model, can measure SBP and DBP values accurately, which is expected

to estimate continuous blood pressure better.

1 INTRODUCTION

Blood pressure (BP) is the lateral pressure on the

vessel wall during blood flow. BP varies continuously

due to different factors such as emotion variation,

physical activities, medication, and disease. Many

studies have shown a significant correlation between

BP variability and cardiovascular mortality (Kikuya

et al., 2000). Continuous measurement of blood

pressure (BP) can dynamically monitor blood

pressure fluctuation, which has very important

practical significance and clinical value.

Although continuous BP can be accurately

measured by invasive methods (e.g. insertion of an

intra-arterial catheter), it could introduce risks to the

patient and workload for physicians (Fung, Dumont,

Ries, Mott, & Ansermino, 2005). Therefore, there is

an urgent requirement for non-invasive BP

measurement. Auscultation and oscillometry are the

two most widely used cuff-based ways to measure BP

noninvasively (Mukkamala et al., 2015). However,

they can provide only intermittent BP readings with

periodic cuff inflation and deflation. In particular,

cuffs are cumbersome, occlusive and time-consuming

to use, disruptive during ambulatory monitoring

(Josep et al., 2013; Peter, Noury, & Cerny, 2014).

Volume clamping and tonometry provide continuous

beat-to-beat BP monitoring. While they both require

relatively complex mechanical structures and are

highly sensitive to the sensor's position and precision

(Imholz, Wieling, van Montfrans, & Wesseling,

1998; Sato, Nishinaga, Kawamoto, Ozawa, &

Takatsuji, 1993).

According to the limitations, the estimation of BP

in successive cardiac cycles via pulse wave velocity

(PWV) or pulse transit time (PTT) has been

extensively proposed(Ding, Zhang, Liu, Dai, &

Tsang, 2016; Jernstedt & Newcomer, 1974; Mase,

Mattei, Cucino, Faes, & Nollo, 2011; Obrist et al.,

1978). The PTT is defined as the period spent by the

arterial pulse propagating from the heart to a

peripheral circulation and is indeed expected to be in

inverse relation with BP (Huynh, Jafari, & Chung,

2019; Josep et al., 2013; Mase et al., 2011).

Zhang, Y., Zhou, C., Huang, Z. and Ye, X.

Development of a Continuous Blood Pressure Monitoring System based on Pulse Transit Time and Hemodynamic Covariates.

DOI: 10.5220/0008944800330039

In Proceedings of the 13th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2020) - Volume 1: BIODEVICES, pages 33-39

ISBN: 978-989-758-398-8; ISSN: 2184-4305

Furthermore, many methods with additional

covariates (e.g. heart rate, pulse wave characteristic

parameters) have been recommended to improve the

precision of predicted BP (Cattivelli & Garudadri,

2009; Jadooei, Zaderykhin, & Shulgin, 2013; Lin et

al., 2015).

In our previous study, we proposed a BP model

based on PTT and other hemodynamic covariates

(Feng, Huang, Zhou, & Ye, 2018). We derived the

relationship between BP and PTT and presented an

improved method utilizing PTT, HR, stiffness index

(SIx) and descent time (DT) to prompt a better BP

estimation based on MIMIC database. The results

demonstrated that this method had the potential to

continuously track BP with higher accuracy and less

calibration frequency. While the limitation is that

model-based data mainly focus on the patient in the

intensive care unit, lacked the verification of healthy

individuals in the normal living and working

conditions.

In this paper, we introduced a reliable hardware

system to validate the blood pressure estimation

algorithms and models proposed in previous studies

from multiple data sets of healthy individuals. The

system was intended to detect both ECG and PPG

signals synchronously, which enabled continuous

cuff-less blood pressure evaluation. We established a

common model and studied the universal

applicability, expected to achieve a general

estimation of population blood pressure by only

changing the individual parameters of the model. In

particular, we investigated the accuracy and

reliability of long-term BP monitoring without

calibration by collecting data of a specific individual

for 28 days. Subsequently, we validated our approach

according to the AAMI standard.

2 MATERIALS AND METHODS

2.1 System Overview

Figure 1 depicts a schematic view of the system

developed, including a wireless signal detection

system, a cuff blood pressure reference instrument

and a host computer (PC).

The signal detection system acts as a slave device

to synchronously collect ECG and PPG signals and

transmit data via Bluetooth; the PC acts as a host for

receiving data, performing signal processing and BP

calculations. To evaluate and calibrate our results, the

Omron-8713 electronic sphygmomanometer was

used to measure cuff blood pressure as a reference.

The developed device is about 4 cm long and 3.5 cm

wide, which was smaller than a typical prototype

(Austad et al., 2016; Kim et al., 2013).

Figure 1: System overview.

Figure 2 describes the block diagram of the signal

detection system, which involves an ECG module

(AD8232), a PPG module (MAX30102), a

microcontroller with Bluetooth combined (NRF52832)

and power management unit. The controller collects

the ECG data of the AD8232 at a sampling frequency

of 200 Hz and reads the PPG data of the IIC interface

at the same rate to guarantee time synchronization.

Figure 2: The block diagram of signal detection system.

The ECG module obtains the lead I configuration

ECG signal from the three electrodes (RA, LA, and

RL), and the electrode feeds the original signal into

the ECG analog acquisition front end through the lead

wire. A reflective optoelectronic sensor with

embedded red and infrared emitters and

phototransistors is used to acquire the PPG signal at

the fingertips. The microcontroller pre-processes and

transmits the two signals, and then the PC performs

filtering, feature extraction, and calculation of SBP

and DBP. In this study, PTT was calculated from the

peak of each R-wave in the ECG to the maximum

value of the PPG.

NRF52832

ADC

AD8232

Electrode

MAX30102

IIC

Power management unit

Battery TPS737

3.3V

BIODEVICES 2020 - 13th International Conference on Biomedical Electronics and Devices

In our previous research, we found that discrete

systems bring some uncontrollable delays(Espina,

Falck, Muehlsteff, & Aubert, 2006). Since PTT is a

time-sensitive parameter, the delay will cause a

deviation in the blood pressure estimation. We use the

same one microcontroller to pick up ECG and PPG

signals (with Bluetooth device integrated inside the

microcontroller), which ensured synchronization.

2.2 BP Model Description

Our previous study has attempted to derive the

relationship between BP and PTT. This method has

been verified to perform better on the MIMIC

database than the two representative studies (Chen,

Kobayashi, Ichikawa, Takeuchi, & Togawa, 2000;

Poon & Zhang, 2005).

In addition to pulse wave transit time (PTT), the

BP model proposed in our previous studies introduces

some other hemodynamic parameters. Among

them, the heart rate (HR) is affected by the regulation

of baroreflex, the latter can adjust the short-term

regulation of BP and prevent wide fluctuations; the

stiffness index (SIx) exhibits the time interval from

the main wave peak to a later diastolic peak, indicates

the arterial stiffness; the descent time (DT) is

interpreted as the duration from the onset of the

dicrotic notch to the end of the diastole，which is

associated to the ventricular diastolic phase.

Figure 3: The BP modelling process.

The basic process of blood pressure modelling is

shown in Figure 3. We perform correlation analysis on

the standardized characteristic parameters and select

features according to the degree of relevance.

Afterward, features with high partial correlation were

selected for multiple linear regression to obtain the BP

model. Then, we established a regression equation

(Eqs.1 and Eqs.2) for SBP and DBP (Feng et al., 2018).

SBP  























(1)

DBP  























(2)

where a, b, c, and d are coefficients.

2.3 Experimental Protocols

A total of 22 healthy subjects (15 males and 7 females)

were recruited for a 4-week experiment, and data were

collected 1 to 3 times per day for each subject. The

subjects all aged between 22 and 50, with an average

age of 24. Participants were requested not to drink

alcoholic and caffeinated drinks for at least 4 hours

before the test. Measurements were carried out under

certain temperature (24 ℃ ~27 ℃ ) and humidity

conditions (50%~60%) compensated by air condition.

(a)

(b)

Figure 4: experimental set-up: sit position (a), stand

position (b).

The protocol involved tests in two postures:

sitting and standing; each data acquisition session

took 5 minutes. Specific steps of the protocol are

shown in Figure 4. Before starting the test, the

subjects were requested to rest for 5 minutes, and the

initial SBP1, DBP1, and HR1 were measured with an

Omron sphygmomanometer. Then, the ECG and PPG

signals were collected simultaneously for 2 minutes

using the developed monitoring system, and then

SBP2, DBP2, and HR2 were measured. The average

Correlat ion

analysis

Filter

features

Multiplelinear

regression

Features BPmo del

Development of a Continuous Blood Pressure Monitoring System based on Pulse Transit Time and Hemodynamic Covariates

of the two indications measured by Omron was taken

as the average blood pressure reference value within

2 minutes. After five minutes of rest, the above steps

were repeated in a standing position to obtain a 2

minute ECG, PPG signal and reference SBP, DBP,

HR. When measuring BP with Omron

sphygmomanometer, we ensure that the cuff is

always at the same level as the heart. If the current

two blood pressure values differ by >5 mmHg, the

blood pressure values are repeatedly measured after 2

minutes and the average of three measurements is

taken. Figure 5 shows the process.

Figure 5: experimental protocols.

2.4 Data Analysis

2.4.1 Preprocessing

To enhance processing efficiency, we separated the

raw signal into 30s segments. Since the original signal

has been preprocessed by the hardware system, we

used the wavelet transform to eliminate the residual

baseline drift and artifact on PC. It has been pointed

out that discrete wavelet decomposition provides

better phase response and computational

efficiency(Kachuee, Kiani, Mohammadzade, &

Shabany, 2017). The ECG and PPG segments were

decomposed into eight levels by the Daubechies db5

wavelet function, the fifth-order contour component

was removed from the ECG signal, while the seventh-

order contour component was removed from the PPG

signal, thereby selecting a suitable approximation

layer to reconstruct the signal.

2.4.2 Feature Extraction

Based on our previous research, we extracted the

parameters required for blood pressure modelling and

validated the proposed model. For the ECG signal, we

extract the R-peak of each cardiac cycle through the

sliding window and the dynamic adaptive threshold

and then calculated the HR from the adjacent R–R

interval.

The maximum value of the PPG waveform was

extracted as the pulse onset point in the same cardiac

cycle. Hence, PTT can be obtained by calculating the

time interval between the R peak of the ECG signal

and the corresponding feature point of the PPG signal.

Furthermore, we also extracted several

morphological parameters mentioned above and

normalized all features.

3 RESULTS

Based on the relationship among BP, PTT and

hemodynamic covariates obtained in our previous

studies, we have established systolic and diastolic

blood pressure models, respectively. We verified the

accuracy of the proposed BP model by comparing the

mean error and root-mean-square error between the

estimated BP value and the reference value of the

Omron sphygmomanometer. Ultimately, we assessed

the consistency of the two methods by Bland-Altman

analysis.

According to our experimental protocol, the

population universal model and the individual long-

term model were studied separately.

3.1 Universal BP Model

Data were collected for each individual for 1~3 times,

a total of 248 segments of data. During the experiment,

a total of 20 valid subjects and 180 data segments were

analyzed, while the signals of two subjects were

screened out for the low signal to noise ratio in the PPG

signals for a good estimation of PTT. Table 1

summarized the basic information about the subject.

Table 1: Subject statistics.

Subjects details

umber 22

Data sets 180 segments

Sex(m/f) 15/7

Age

Range: 22-50

mean+SD: 27.96+7.12

DBP mean+SD: 74.42+9.87

SBP mean+SD: 112.49+12.23

All valid data sets were divided into 10 parts by

using the method of 10-fold cross-validation, nine of

which were taken in turn as training data to establish

a BP model, and one was used as test data to estimate

the BP value. Then compared the value with reference

values measured by the Omron sphygmomanometer

to calculate errors. Table 2 shows the mean error and

root-mean-square of the results of the 10-fold cross-

validation. The average error was 0.002±8.544

mmHg for SBP, 0.005±6.690 mmHg for DBP, which

extremely approached the standard of AAMI (5±8

mmHg).

Start

Rest

Cuff

BP1

Stop

5min 1min

2min 1min 5min

1min

2min 1min

ECG,PPG-sit

(our system)

Cuff

BP2

Rest

Cuff

BP1

ECG,PPG-stand

(our system)

Cuff

BP2

Test1

Test2

BIODEVICES 2020 - 13th International Conference on Biomedical Electronics and Devices

Table 2: The results of 10-fold cross-validation.

SBP (mmHg) DBP (mmHg)

mean std mean std

1 0.0397 8.6899 0.0184 6.8063

2 -0.0133 8.5496 0.0366 6.6011

3 -0.0124 8.4692 -0.0054 6.6397

4 0.0216 8.7775 -0.0118 6.8001

5 -0.0252 8.3874 -0.0110 6.6463

6 -0.0389 8.4475 -0.0215 6.6103

7 0.0223 8.4548 0.0180 6.5604

8 0.0012 8.5908 -0.0095 6.8713

9 -0.0021 8.4272 -0.0008 6.5995

10 0.0274 8.6498 0.0421 6.7686

Mean 0.0020 8.5444 0.0055 6.6904

(a)

(b)

Figure 6: Bland-Altman analysis of SBP(a), DBP(b)

estimation for universal BP model.

The Bland-Altman plot of the SBP and DBP

estimation for our proposed method versus Omron

sphygmomanometer are given in Fig. 6. The x-axis

presents the average of the two methods, while the y-

axis shows the difference between them.

We observed that a total of 97.8% of the SBP, and

95% of the DBP measurements lie in the limits of

agreement (1.96  SD), indicating that the estimated

BP with the proposed method is in close agreement

with the Omron sphygmomanometer.

3.2 Individual Long-term BP Model

To investigate the validity of the evaluation of a

particular subject's BP model over several

consecutive days, further studies were performed on

a selected subject. Regression models related to PTT,

hemodynamic parameters, and BP values (SBP, DBP)

were constructed by collecting data throughout one

day at various time intervals during the subject's

regular working hours. To demonstrate the accuracy

and reliability of the model in long-term continuous

monitoring scenarios, we compared the estimated BP

values with the Omron sphygmomanometer.

Measurements were performed at three different

times of the day (10:00 am, 3:00 pm, 7:00 pm) based

on fluctuations in blood pressure, and the experiment

lasted for one month(29 days).

A total of 683 segments of data were collected for

the same individual, excluding invalid data caused by

collection failure. The individual BP model was

created using 24 data sets on the first day, and the

remaining 27 days of data were used to test the model

to verify its long-term effectiveness. Similarly, we

compared the estimated 28-day blood pressure with

the Omron sphygmomanometer, and the mean and

RMS of the statistical errors were shown in Table 3.

The average error was 5.204±5.462 mmHg for SBP,

2.714 ± 4.756 mmHg for DBP, which extremely

approached the standard of AAMI (5±8 mmHg).

Table 3: Individual model error analysis.

SBP (mmHg) DBP (mmHg)

mean std mean std

5.2043 5.4624 2.7141 4.7561

Figure 7: RMSE of BP estimated with the proposed

measurement system with the reference system over 28

days for a subject.

0 5 10 15 20 25 30

DBP

SBP

RMSE(mmHg)

Day

Development of a Continuous Blood Pressure Monitoring System based on Pulse Transit Time and Hemodynamic Covariates

(a)

(b)

Figure 8: Bland-Altman analysis of SBP(a), DBP(b)

estimation for individual long-term BP model.

Figure 7 illustrates the observed trend of the

estimated RMSE of estimated versus reference BP

value. The proposed method can track BP values with

lower RMSE values in 28 days with an error within

an acceptable range. The Bland-Altman plot of the

SBP and DBP estimation for the individual long-term

model we proposed versus the Omron

sphygmomanometer are given in Figure 8, which

indicates that the two methods are in close agreement.

4 DISCUSSION

This study introduced a reliable hardware system to

verify the proposed models on normal subjects. The

system guarantees high quality and strict time

synchronization of the original signal, thus

eliminating errors caused by noise and uncontrolled

delays. Moreover, high-quality signals can improve

the accuracy of feature extraction, which in turn

contributes to the estimation of blood pressure models.

However, there are some limitations to our study.

First, we recruited healthy young and middle-aged

people who were at higher risk of cardiovascular

disease due to high mental stress and daily work

intensity. Despite this, we still need data from older

people or patients with cardiovascular disease to

support the model. Second, the range of BP

fluctuations in the experimental protocol is not

obvious enough. Different BP perturbations must be

applied to rigorously evaluate the validity of the

proposed model, such as exercise test, cold pressor

test.

5 CONCLUSIONS

In this paper, we proposed a wearable signal detection

system that can simultaneously acquire ECG and PPG

signals, ensuring signal integrity and reliability.

Subsequently, we built a generic BP model and an

individual long-term BP model to calculate SBP and

DBP values, then compared the two values with the

Omron sphygmomanometer, which showed a good

agreement. Further, the errors between Omron and

our system of the two models (generic SBP: 0.002±

8.544 mmHg, DBP: 0.005±6.690 mmHg; individual

SBP: 5.204 ± 5.462 mmHg, DBP: 2.714 ± 4.756

mmHg) are close to the AAMI standard(5±8 mmHg).

The results indicate that our system has the potential

to continuously track BP for a long time without

calibration. Nonetheless, more validation in various

subjects and situations should be conducted with the

system.

ACKNOWLEDGEMENTS

This work was supported by the National Key R&D

Program of China (No.2017YFF0210803), this

research was also funded by China Postdoctoral

Science Foundation (No.2018M632456) and the

Fundamental Research Funds for the Central

Universities (No.2019FZA5015).

REFERENCES

Austad, H. O., Vedum, J., Røed, M. H., Dalgard, S., et al.

(2016). An Unobtrusive Wearable Device for

Ambulatory Monitoring of Pulse Transit Time to

Estimate Central Blood Pressure. International Joint

Conference on Biomedical Engineering Systems and

Technologies, BIOSTEC 2016, 179-186.

BIODEVICES 2020 - 13th International Conference on Biomedical Electronics and Devices

Cattivelli, F. S., & Garudadri, H. (2009). Noninvasive

Cuffless Estimation of Blood Pressure from Pulse

Arrival Time and Heart Rate with Adaptive Calibration.

2009 Sixth International Workshop on Wearable and

Implantable Body Sensor Networks, 114-119.

Chen, W., Kobayashi, T., Ichikawa, S., Takeuchi, Y., &

Togawa, T. (2000). Continuous estimation of systolic

blood pressure using the pulse arrival time and

intermittent calibration. Medical & Biological

Engineering & Computing, 38(5), 569-574.

Ding, X. R., Zhang, Y. T., Liu, J., Dai, W. X., & Tsang, H.

K. (2016). Continuous Cuffless Blood Pressure

Estimation Using Pulse Transit Time and

Photoplethysmogram Intensity Ratio. IEEE Trans

Biomed Eng, 63(5), 964-972.

Espina, J., Falck, T., Muehlsteff, J., & Aubert, X. (2006).

Wireless Body Sensor Network for Continuous Cuff-

less Blood Pressure Monitoring. IEEE/EMBS

International Summer School on Medical Devices &

Biosensors.

Feng, J., Huang, Z., Zhou, C., & Ye, X. (2018). Study of

continuous blood pressure estimation based on pulse

transit time, heart rate and photoplethysmography-

derived hemodynamic covariates. Australas Phys Eng

Sci Med, 41(2), 403-413.

Fung, P., Dumont, G., Ries, C., Mott, C., & Ansermino, M.

(2005). Continuous noninvasive blood pressure

measurement by pulse transit time. International

Conference of the IEEE Engineering in Medicine &

Biology Society.

Huynh, T. H., Jafari, R., & Chung, W. Y. (2019).

Noninvasive Cuffless Blood Pressure Estimation Using

Pulse Transit Time and Impedance Plethysmography.

IEEE TRANSACTIONS ON BIOMEDICAL

ENGINEERING, 66(4), 967-976.

Imholz, B. P., Wieling, W., van Montfrans, G. A., &

Wesseling, K. H. (1998). Fifteen years experience with

finger arterial pressure monitoring: assessment of the

technology. Cardiovasc Res, 38(3), 605-616.

Jadooei, A., Zaderykhin, O., & Shulgin, V. I. (2013).

Adaptive algorithm for continuous monitoring of blood

pressure using a pulse transit time. Electronics &

Nanotechnology.

Jernstedt, G. C., & Newcomer, J. P. (1974). Blood-Pressure

and Pulse-Wave Velocity-Measurement for Operant-

Conditioning of Autonomic Responding. Behavior

Research Methods & Instrumentation, 6(4), 393-397.

Josep, S., Martin, P. A., Damien, F., Jacques-André, P., et

al. (2013). Noninvasive and Nonocclusive Blood

Pressure Estimation Via a Chest Sensor. 60(12), 3505-

3513.

Kachuee, M., Kiani, M. M., Mohammadzade, H., &

Shabany, M. (2017). Cuffless Blood Pressure

Estimation Algorithms for Continuous Health-Care

Monitoring. IEEE Trans Biomed Eng, 64(4), 859-869.

Kikuya, M., Hozawa, A., Ohokubo, T., Tsuji, I., et al.

(2000). Prognostic significance of blood pressure and

heart rate variabilities: the Ohasama study.

Hypertension, 36(5), 901-906.

Kim, H., Lee, H., Baek, H., Lee, W., et al. (2013). A

preliminary study of non-intrusive blood pressure

monitoring using portable device. International

Conference on Biomedical Electronics and Devices,

BIODEVICES 2013, 163-167.

Lin, H., Xu, W. Y., Guan, N., Ji, D., et al. (2015).

Noninvasive and Continuous Blood Pressure

Monitoring Using Wearable Body Sensor Networks.

IEEE Intelligent Systems, 30(6), 38-48.

Mase, M., Mattei, W., Cucino, R., Faes, L., & Nollo, G.

(2011). Feasibility of cuff-free measurement of systolic

and diastolic arterial blood pressure. J Electrocardiol,

44(2), 201-207.

Mukkamala, R., Hahn, J. O., Inan, O., Mestha, L., et al.

(2015). Towards Ubiquitous Blood Pressure

Monitoring via Pulse Transit Time: Theory and

Practice. 62(8), 1879-1901.

Obrist, P. A., Light, K. C., Mccubbin, J. A., Hutcheson, J.

S., et al. (1978). Pulse transit time: Relationship to

blood pressure. 10(5), 623-626.

Peter, L., Noury, N., & Cerny, M. (2014). A review of

methods for non-invasive and continuous blood

pressure monitoring: Pulse transit time method is

promising? IRBM, 35(5), 271-282.

Poon, C. C., & Zhang, Y. T. (2005). Cuff-less and

noninvasive measurements of arterial blood pressure by

pulse transit time. Conf Proc IEEE Eng Med Biol Soc,

6, 5877-5880.

Sato, T., Nishinaga, M., Kawamoto, A., Ozawa, T., &

Takatsuji, H. (1993). Accuracy of a continuous blood

pressure monitor based on arterial tonometry.

Hypertension, 21(6 Pt 1), 866-874.

Development of a Continuous Blood Pressure Monitoring System based on Pulse Transit Time and Hemodynamic Covariates