Sensing Devices and Sensor Signal Processing for Remote ... IEEE PAPERS/embedded...Pierluigi Barba,...

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 62, NO. 3, MARCH 2013 553

Sensing Devices and Sensor Signal Processing forRemote Monitoring of Vital Signs in CHF Patients

Luca Fanucci, Member, IEEE, Sergio Saponara, Member, IEEE, Tony Bacchillone, Massimiliano Donati,Pierluigi Barba, Isabel Sánchez-Tato, and Cristina Carmona

Abstract—Nowadays, chronic heart failure (CHF) affects anever-growing segment of population, and it is among the majorcauses of hospitalization for elderly citizens. The actual out-of-hospital treatment model, based on periodic visits, has a lowcapability to detect signs of destabilization and leads to a highre-hospitalization rate. To this aim, in this paper, a complete andintegrated Information and Communication Technology system isdescribed enabling the CHF patients to daily collect vital signs athome and automatically send them to the Hospital InformationSystem, allowing the physicians to monitor their patients at dis-tance and take timely actions in case of necessity. A minimum setof vital parameters has been identified, consisting of electrocar-diogram, SpO2, blood pressure, and weight, measured through apool of wireless, non-invasive biomedical sensors. A multi-channelfront-end IC for cardiac sensor interfacing has been also devel-oped. Sensor data acquisition and signal processing are in chargeof an additional device, the home gateway. All signals are pro-cessed upon acquisition in order to assert if both punctual valuesand extracted trends lay in a safety zone established by thresholds.Per-patient personalized thresholds, required measurements andtransmission policy are allowed. As proved by first medical tests,the proposed telemedicine platform represents a valid support toearly detect the alterations in vital signs that precede the acutesyndromes, allowing early home interventions thus reducing thenumber of subsequent hospitalizations.

Index Terms—Biomedical instrumentation, chronic heart fail-ure (CHF), e-health, sensor signal processing, telemonitoring, vitalsigns sensors.

I. INTRODUCTION

CHRONIC HEART failure (CHF) represents one of themost relevant chronic disease in all industrialized coun-tries, affecting approximately 15 million people in Europe andmore than 5 million in the U.S., with a prevalence ranging from1% to 2% and an incidence of 3.6 million new cases each year inEurope and 550 000 cases in U.S. [1]–[3]. It is the leading causeof hospital admission particularly for older adults reaching

Manuscript received April 6, 2012; revised June 9, 2012; accepted July 16,2012. Date of publication October 5, 2012; date of current version February 5,2013. This work was partly supported by theAmbient Assisted Living Program.The Associate Editor coordinating the review process for this paper wasDr. Domenico Grimaldi.

L. Fanucci, S. Saponara, T. Bacchillone, and M. Donati are with theUniversity of Pisa and Consorzio Pisa Ricerche, 56125 Pisa, Italy (e-mail:[email protected]).

P. Barba is with the CAEN S.p.a., 55049 Viareggio, Italy (e-mail:[email protected]).

I. Sánchez-Tato and C. Carmona are with the Fundación CITIC, Parque Tec-nológico de Andalucía, ES-29590, Málaga, Spain (e-mail: [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TIM.2012.2218681

a prevalence of 1.3%, 1.5%, and 8.4% in 55–64 years old,65–74 years, and 75 years or older segments, respectively [2].Admission to hospital with heart failure has more than doubledin the last 20 years [1], and it is expected that CHF patientswill double in 2030. Hospital admissions caused by CHF resultin a large societal and economical issue, accounting for 2% ofall hospitalizations [4]. The CHF management accounts for 2%of the total healthcare expenditure [5], [6] and hospitalizationsrepresent more than two thirds of such expenditure [3].

The current healthcare model is mostly in-hospital basedand consists of periodic visits. Previous studies pointed outthat in patients with a discharge diagnosis of heart failure, theprobability of a readmission in the following 30 days is about0.25, with the readmission rate that approaches 45% within6 months [7]. It is acknowledged that changes in vital signsoften precede symptom worsening and clinical destabilization:indeed, a daily monitoring of some biological parameters wouldensure an early recognition of heart failure de-compensationsigns, allowing appropriate and timely interventions, likelyleading to a reduction in the number of re-hospitalizations.Due to lack of resources at medical facilities to support thiskind of follow-up, the use of Information and CommunicationTechnologies (ICT) has been identified by physicians and ad-ministrator as a possible valid support to overcome this limit.There is in literature some evidence that a multidisciplinarymanagement program [8], [9] including a home-based follow-up strategy can improve outcome of heart failure patients,including a reduction in mortality, hospital readmissions, andlengths of hospital stays, and increase patient satisfaction[10]–[12].

This paper represents an extension of [13] in which the sameauthors provide an overview of a flexible and high configurableplatform for domestic vital signs acquisition and processing,integrated with the Hospital Information System (HIS).

This work has been developed within the Health at Homeproject (H@H) of the Ambient Assisted Living Program(AAL). It takes into account the recent AAL Roadmap guide-lines [14], the future challenges in telecare [15], and somerecent studies conducted on AAL solutions [16], [17].

The H@H platform aims at connecting in-hospital care ofthe acute syndrome with out-of-hospital follow-up by patient/family caregiver, being directly integrated with the usualcardiology departmental HIS. Patients’ signs, symptoms, andraised alarms can be received by healthcare providers, andaggravations can be quickly detected and acted upon. Thanksto the collection of vital parameters at home, the sensor datasignal processing and the automatic data transmission to the

0018-9456/$31.00 © 2012 IEEE

554 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 62, NO. 3, MARCH 2013

Fig. 1. H@H System Architecture.

medical center, a more frequent (usually daily) assessment ofclinical status than in conventional practice is permitted [18].

After this introduction, Section II briefly describes the wholeH@H system architecture. The features of all involved sensorsare discussed in Section III, while Section IV is focused onthe developed front-end IC for cardiac sensor. The sensor dataprocessing strategy is described in Section V. Section VI andSection VII present the results of the demonstration phase andthe comparison with the state of the art. Finally, the conclusionsare drawn in Section VIII.

II. H@H TELECARE SYSTEM OVERVIEW

The H@H development Consortium is composed by indus-trial and research partners with qualified competences in sens-ing and data processing as well as very important healthcareproviders (Hospitales Virgen del Rocio, Spain; Dom KoperHospital in Slovenia and the research clinical center FondazioneGabriele Monasterio in Italy). The system requirements comedirectly from the long experience in the CHF field of theinvolved physicians. The resulting platform takes into consider-ation both medical expectations, patients’ features (elderly, withcomorbidity and cognitive deficit), and the progressive natureof the disease. For these reasons, we propose an intuitive homemonitoring system based on a configurable follow-up operatingprotocol (OP), integrated with the HIS of the cardiology depart-ment through a server software platform.

The complete H@H system has the client/server architectureshown in Fig. 1. The clients are typically located at patient’shome and consist of a set of wireless sensors to measure themain vital signs (see Table I) and an additional device, the homegateway (see Table II), that centralizes all computation andcommunication resources. These domestic subsystems havein turn a client/server structure where sensors are the clientsof the collection and transmission point represented by thehome gateway. The server platform, installed at health servicefacilities, accepts data from gateways making them available inthe HIS and finally allows the management of all patient’s datasince their enrolment.

TABLE ISENSING REQUIREMENTS AND VERSIONING

TABLE IIHOME GATEWAY HARDWARE RESOURCES

The home gateway [19] receives sensors data via point-to-point Bluetooth connections initiated by the sensing modules.Upon received, it processes all data to detect dangerous alter-ations and then forwards them to the hospital server throughADSL or mobile Broadband, to be further analyzed and flowedinto the HIS. In case of an alarm situation, caregivers orrelatives are contacted via SMS (i.e., reporting the abnormalvalues that lead to the alarm), and all pending data are sentto the server. Section V will explain in depth the sensors dataprocessing and the alarm detection task. The gateway normallyoperates connected to the power line, but the internal batteryensures about 5 h of autonomy in case of power failure. Thepeculiar features of the target patients require the design of

FANUCCI et al.: REMOTE MONITORING OF VITAL SIGNS IN CHF PATIENTS 555

an intuitive and simple home gateway user interface. Thisis able to display reminder messages, guide animations andsounds when a planned activity time is reached according tothe OP. Patient can read the last measured values and the statusof sensor battery charge. Green, yellow, and red backgroundcolors are used for information, warning, and error messages,respectively. To simplify the use of the gateway, a five-keymembrane keypad is provided (i.e., Yes, No, Alarm sending,and Up/Down scroll buttons).

The server platform is based on the webservices paradigm fordata reception and presentation and also for the interaction withthe cardiology department HIS and the patient’s informationmanagement. The user interface allows the clinicians to interactwith the system, also in mobility, using the web browser.

The follow-up OP consists of a formal XML specification ofthe list of daily actions to be performed during the monitoringperiod (i.e., types and frequencies of measurements and drugsassumptions), and it defines the behavior of the home gatewayin terms of alarm thresholds for each scheduled measures,transmission policy (i.e., daily, weekly, immediate send), andselectable symptoms. It is configurable according to the pa-tient’s needs and remotely updatable if required via the serverplatform. This novel concept embeds medical prescription forthe given patient in the home gateway.

The availability of multiple communication paths ensuresa good adaptability of the system in overall operating areasand improves the fault tolerance. As the coverage of GSMis close to 99%, the system reaches a very high degree ofconnectivity. In the worst case, the GPRS upload data rate of20 Kbps is sufficient to transmit all data in few minutes. Surely,better performance becomes available with EDGE and UMTS.Furthermore, the gateway is able to exploit the GSM network tosend SMSs to the physicians, patient’s relatives, and caregiversin case of alert situation.

Authentication, integrity, and confidentiality of the commu-nication are guaranteed by the HTTPS protocol. The use ofinternational standard for data communication, ANSI HL7-RIM Clinical Document Architecture v2 [20]–[22] and XML,improves the interoperability of the system as well as theintegration with existing HIS. All numeric and waveform obser-vations use SNOME CT [23] or LOINC [24] standards codes.

The proposed system is conceived to allow a better assess-ment of vital signs identified by clinicians as the most signifi-cant in CHF through one or few daily measurements, being incontrast to those systems that offer a continuous monitoring forlimited period. It does not introduce any remarkable overheadwith respect to regular activities of the medical staff. Indeed allsigns are flowed as row data into the patient’s electronic healthrecord, and, thanks to the provided automatic signal processingcapability, clinicians and caregivers are timely informed in caseof alarm detection. Moreover, H@H minimizes the impact onthe patient. The wireless biomedical sensors avoid connection-cable encumbrance. The number of sensing modules is min-imized, and the signal quality is not excessively dependenton transducer positioning. The domestic gateway reminds thescheduled activities, provides a graphical assisted procedurethat shows how to use the sensors, and acquires data withoutrequiring any preventive action to the user.

III. H@H SENSOR DEVICES

In general, biomedical sensors address the wearability/portability, non-invasivity, wireless communication, and batteryduration concerns in order to be easily used autonomouslyby the patients at home. Moreover, the system minimizes thenumber of devices and sensors/electrodes to be positioned onthe patient’s body (e.g., three recording sites electrocardiogram(ECG) instead of a more complex 12-lead ECG is adoptedto limit the effects of electrodes misplacement [25]). Themeasurement experience consists of wearing/using the sensorsperiodically, once or twice a day, only for the duration of theacquisition without any long-period application of the sensorsas in different solutions.

According to the analysis carried out by the clinicians, inter-esting vital parameters to monitor in a CHF patient are ECG,SpO2, weight, blood pressure (BP), chest impedance, respira-tion, and posture (see Table I for sampling details). To achievean asset for the implementation and to increase scalability ofthe system, the sensing modules have been clustered into twopossible configurations: basic and advanced. Basic partitioningis intended as the minimum set of requirements to ensure a com-plete and useful telecare system in CHF. As advanced, we referto additional features in order to widen the kind of CHF patientsto be possibly enrolled into the telemonitoring and to copewith other chronic diseases (i.e., chronic obstructive pulmonarydisease, diabetes). Since the basic configuration provides to theHIS the needed data set for accurate CHF telemonitoring, thenthe basic set in Table I is the one implemented for the medicaltechnology test reported in Section VI.

To achieve our goals, according to the basic configuration,the H@H system is formed by a bunch of three Bluetooth andbattery-powered sensing devices, the commercial standalonemodules UA-767BT arm cuff device for BP readings and UA-321PBT digital scale by A&D Medical and the new integratedECG-SpO2 module ad-hoc developed in the framework of thisproject. The latter is based on a new multi-channel front-endIC developed for cardiac sensor interfacing. The overall set ofdevices is shown in Fig. 2, where the sensors positioning andthe limited impact on the patient are also visible.

All sensing devices exploit the Bluetooth 2.0 wireless tech-nology for the communication with the home gateway. Theysend only raw data, and any form of signal processing is com-pletely demanded to the gateway. Class I transceivers, with near7 dBm of transmission power, ensure an appropriate coveragein the domestic environment.

The modules implement the Service Discovery Protocoland the Serial Port Profile (SPP) to discover and wirelesslycommunicate with the access point collection service, which isidentified in the home gateway, with optional link-level security(128 bit encryption). Each sensor acts as initiator of connectionsaccording to the following rules:

1) if it knows the default remote peer address (DRPA), itbegins the page procedure to establishes a point-to-pointconnection (GAP Connectable Mode enabled), sends alldata, and closes the connection;

2) if no DRPA exists, the module initiates the inquiry proce-dure for 10.24 s to discover remote peers in range (GAP


Fig. 2. Sensors Positioning.

General Discoverable Mode enabled) offering the SPPcollection service (i.e., gateways):

3) all peers identified in step 2 are sequentially asked for theauthentication PIN (pairing mechanism);

4) the device that matches the PIN receives data from thesensor, and in case of successful transmission, it becomesthe new DRPA;

5) if in the above step 1 the page procedure fails, steps 2 to4 are performed to identify the new DRPA.

The sensors are provided to the patients already configuredto work with the given home gateway, hence at runtime onlythe above step 1 is performed while the whole procedure isrequired once at configuration time. In this condition, a mea-surement simply consists of turning on the sensing device andwaiting until the end of the measurement process without anypreventive interactions with the gateway (e.g., device searches,PIN typing, connection opening), introducing also benefits inbattery saving. Actual gateway implementation deals with oneactive connection at a time, as required by the physicians whichdefined a time-division strategy for bio-signal acquisition. Si-multaneous connection attempts lead to discard attempts otherthan the first. Anyway, this meets the user requirements as thesystem guides the patient to follow the OP activities requestingactions one by one.

With respect to alternatives such as Zigbee or point-to-pointWiFi links, Bluetooth 2.0 provides the desired tradeoff amongavailable bandwidth, security and reliability of the connection,cost and power consumption of the node, and link distance. Itoperates at 2.4 GHz, does not require line of sight positioning ofthe units, and the frequency-hopping and fast acknowledgmentscheme improves the robustness of the link in noisy environ-ments. Indeed, the nominal data rate is around 2 Mbps against∼18 Kbps required by the ECG-SpO2 device. Zigbee, although

its very interesting for its power consumption, is still a newproduct with respect to the cheaper and more diffused Bluetoothtechnology.

A. ECG-SpO2 Module

The ECG-SpO2 module is a new sensing device, devel-oped ad hoc in the H@H project. Specifically, the moduleprovides electrocardiographic, pulse oximetry, and plethysmo-graphic measurements by means of proper sensors and elec-trodes placed on patient’s body, implementing non-invasivetechniques. It is hosted in a robust and small size ABS case(92× 150× 28 mm, 200 g) powered by an integrated recharge-able 3.7 V and 1700 mAh Li-Ion battery able to ensure 18 h ofcontinuous operability (i.e., more than 3 months of acquisitionsconsidering 5 min track twice a day).

The module uses an ECG patient trunk cable with fourleadwires: RA, LA, LL, and RL (neutral) to provide the elec-trocardiographic signal. The sensor for pulse oximetry andplethysmographic measurements is a classical fingerclip readertype to be applied at patient’s first finger. In accordance withthe physicians, the Einthoven’s 3 leads ECG configurationis considered sufficient for our purposes (e.g., detection ofheart rate (HR) and rhythm) and not excessively dependent onthe transducer positioning. All such signals are conditioned,digitalized and then packetized, and finally transmitted viaBluetooth protocol. The ECG-SpO2 device outputs digitalizedwaveforms of two standard limb leads, the oxygen saturationin the blood, the plethysmographic waveform, and the batterylevel. The final appearance of the sensing module and itsschematic view are shown in Fig. 3.

Integrating ECG and SpO2 functionalities in a novel singledevice reduces the number of sensing devices in the final systemand also enables to acquire synchronized ECG and SpO2 orplethysmogram traces. This allows a larger and more specificamount of information for advanced analysis and multi-sensordata fusion (e.g., the Pulse Transit Time estimation).

As reported in Fig. 3, the ECG-SpO2 module is realizedassembling its building blocks on a single printed circuit board.All communications within the board take place under the coor-dination of a dedicated firmware running on the MSP430F2418Ultra-Low Power Mixed Signal Controller, responsible formixing raw data before passing them to the Bluetooth interface.

The core of the ECG block is an ad-hoc developedapplication-specific integrated circuit (ASIC), very compactand high configurable, able to integrate the ECG functionalityinto portable and wearable devices. The technical specificationsof the chip, called CARDIC, are summarized in Table III, whilea detailed description of the chip architecture and performanceis provided in Section IV.

The ECG-SpO2 module also hosts the ChipOx OEM byEnvitec for pulsioximetry measuring, being fully configurable,highly dimension-contained and with low power consumption.Together with SpO2 readings, the device gives also HR digitaldata and a digitalized plethysmographic waveform (PPG). Therange of measure is from 45% to 100%, with an accuracy of1.5%–2% for oxygen saturation and 0–255 LSB at 100 Hz, withaccuracy > 6 ppm/LSB for plethysmography. It consumes up


Fig. 3. Architecture and final appearance of the ECG-SpO2 module.

TABLE IIIASIC CARDICTECHNICAL SPECIFICATION

to 25 mA at 3.3 Volt of power supply, and it communicatesover an UART-TTL with a baud rate of 9600. Its dimensionsare 31× 14× 5 mm.

The Bluetooth communication leverages the OEMSPA312iby Connect Blue, which is a small size module based on thePhillips BGB203 system in package. The BGB203 has on-chipSRAM and FLASH stacked in the same package. The chip iscompliant with the 2.0 standard and the Class I specification.

B. UA-767BT BP Monitor

Regarding non-invasive BP measuring, the market offer andthe cost effectiveness led to the choice of a commercial stand-alone solution by A&D Medical (see Fig. 4). This arm cuffedautomatic meter, based on the oscillometric method, is alreadyused in various telemedicine systems as it is certified formedical use and can be employed also at home. It is equippedwith Bluetooth class I communication capabilities in order tosend the acquired systolic and diastolic values to a base station,and it is able to store up to 30 measurements waiting fortransmission. It outputs also HR frequency. All values have8 bit precision; the sensibility range and the accuracy are


Fig. 4. UA-767BT blood pressure monitor and UA-321PBT digital scale.

20–280 mmHg and +/−3 mmHg, respectively. Dimensions are147× 64× 110 mm with about 0.3 Kg of weight. The powersupplied by 4 × 1.5 Volt AA batteries ensures near 6 months ofoperability in normal OP (i.e., two or three measures per day).

C. UA-321PBT Digital Scale

The market of digital scales offers interesting Bluetoothenabled products allowing for wireless transmission of multiplereadings, thus avoiding cable encumbrance. These devices areuser-friendly, with a low-profile design, and are not expensive,commonly used in fitness and telecare systems. The A&D scaleshown in Fig. 4 measures only weight and belongs to the sameseries as the UA-767BT, so they have the same Bluetooth datacommunication specification. The current reading is visualizedon an on-board LCD display and sent to the base station. Allvalues are expressed with 5 bytes, and a simple conversionformula is needed. It has a maximum capacity of 200 Kg and anaccuracy of +/−0.1 Kg. The dimensions and weight are 300×300× 30 mm and 1.0 Kg, respectively. The battery life of the4 AA batteries ensures approximately 1000 measurements (i.e.,more than 1 year in classical OP).

IV. IC FRONT-END FOR CARDIAC SENSORS

CARDIC is an ASIC designed and manufactured in orderto be very compact and fully configurable for multi-sensorintegration in wearable and portable medical devices. It ensuresmultiparameters medical acquisition, offering a cost-effectivesolution able to meet requirements coming from both monitor-ing and diagnostic demand. Indeed, as in the automotive case[26] the medical monitoring field has a potential large volumemarket justifying the development of a mixed-signal IC forhigh-performance sensor front-end.

The chip is developed in AMS CMOS 0.8 μm CXZ 2 MetalLayers 2 Poly technology and is fully assembled in a 14×14× 1.4 mm TQFP 128 package. The core architecture hasbeen implemented taking into account the typical constraintsof biomedical signals monitoring: like standard ECG, BP, andbody temperature. Typical power supply voltage is 5 V. Digitalin and output pins are compatible with 3.3 V power supplylevel. The operating temperature is from 0 ◦C to 70 ◦C. Theoverall IC characteristics are reported in Table III.

Referring to the IC functional block in Fig. 5, the followingmain parts can be pointed out as a description of the chipfeatures and functionalities:

• a fully configurable multi-channel ECG block includingeight input differential channels for signal conditioning(amplification, filtering, offset regulation) of 3-5-12-leads

ECG systems, an adder generating the Central TerminalPoint (CTP) for prechordial leads, a right leg (RL) driverand a SHIELD driver in order to reduce the CommonMode 50 Hz noise, a Pace Maker detector section for pacepulses detection on a dedicated pin. All is in conformity toIEC 60601-2-51 2003-02 standard;

• one analog channel designed for decoupling and amplify-ing a BP signal coming from an optional external pressuresensor;

• one analog channel able to process a temperature signalprovided by an external temperature sensor;

• one programmable analog MUX for switching among thechannels to be converted by the ADC;

• one 16 bit (12 bit linearity) cyclic algorithmic ADC toconvert the voltages from the analog channels;

• a serial peripheral interface (SPI) to configure the chipsettings and for data readout;

• one battery channel for monitoring the battery status.

The structure of a single ECG channel is shown in Fig. 6. Itis mainly composed by an instrumentation amplifier (IA) withhigh common mode rejection ratio (CMRR), a programmablegain amplifier (PGA), a BUFFER, and an offset regulator block.The IA has a high CMRR, 100 dB typical, 92 dB minimum,in order to reduce environmental electric interferences, like the50 Hz noise from the industrial network, always present in bothelectrodes, that are connected with human body, and ground.The power supply rejection ratio is 100 dB typical, 96 dBminimum. The first-order high-pass filter, obtained through anexternal capacitor, removes low-frequency baseline wanderingthat is so common in ECG circuits (usually due to electrodes).An anti-aliasing filter is obtained with external RC network.The PGA stage can provide four gains, that are 18, 24, 36 and48, for standard ECG measurement. The IA has a gain of 15.6set with an external resistor of 1.2 MΩ. Therefore, the total gaincan be configured up to roughly 700. The third stage is a buffersection with fast settling which sends out the signal to the ADC.Each ECG channel has some switching MUXs placed in theinput section, and it is configurable via SPI command. The ar-chitecture of the ECG channel allows to implement the baselinefast restoring, useful any time variations in the baseline of thesignals due to artefacts cause a temporary saturation of the IA.Offset regulator and gains regulator procedures are also avail-able for each channels. If the ECG measurement is chosen, acyclic procedure is activated, and the analog MUX connects thissection to the A/D converter. Within 100 μs (worst case), all thechannels are processed. Each ECG processing channel has beenalso characterized in terms of noise: the input referred noisemeasured in the range 0.1 Hz to 150 Hz is within 10 μVrms.

The Adder block is used to obtain the CTP signal referenceto be used as inverting input for each channel that receives apre-cordial signal. A driven RL circuit (RL Driver block) helpsto set the common mode, and it is safer than connecting the RLto voltage reference. The circuitry able to drive in active modethe shield of the ECG cables (SHIELD Driver block) helps toreduce the Common Mode 50 Hz noise.

The pacemaker detector block (see Fig. 7) provides bandpass filtering on the ECG signal, full wave rectification for


Fig. 5. CARDIC chip schematic view.

Fig. 6. ECG signal channel for analog preprocessing.

detecting pacemaker pulses of either polarities, peak detec-tion on the filtered and rectified signals, and discriminationrelative to a programmable threshold level. Once a pace-

maker pulse is detected, it provides on the dedicated outputdigital pin a pulse whose duration is set by an external RCnetwork.


Fig. 7. Pace maker detector block diagram.

The BP section includes two channels. An analog channelthat converts a DC single ended, ground referred voltage toa differential voltage suitable to be converted by the on chipADC, and it amplifies this signal. Input signal ranges between0.5 V (0 mmHg) and 1.7 V (300 mmHg). The offset voltage forsingle-ended/differential conversion is provided by the internalreferences generator block. Another analog channel amplifiesthe ac component of the input pressure signal and translatesthe dc component around a reference voltage. Extraction ofac component is assured by two external capacitors and oneexternal resistor. If the BP measurement is chosen, a cyclicprocedure is activated, and the MUX connects this section tothe A/D converter.

The skin temperature measurement channel uses a NTC re-sistor as input sensor: the NTC resistor has a nonlinear voltage-temperature characteristic so an additional resistive network,shown in Fig. 8, is used to improve the linearity of the responseaccording to the following equation:

RLIN (T ) = (RNTC(T ) +R3) ‖R1. (1)

The chosen values in Fig. 8 (R1 = 18 kΩ, R2 = 3.75 kΩ,R3 = 1 kΩ, NTC with 10 kΩ at 25 ◦C and beta of 3976,VREF_TMP_IN1&2 of 3.5 V and 1.5 V, respectively) optimizethe response linearity of the acquisition system in the range32 ◦C to 46 ◦C, including the human body temperatures.

The ADC structure used in this design is a cyclic/algorithmicarchitecture with 1.5 bit per cycle, made by three main blocks:a sample and hold stage, comparators stage and a residuemultiplying DAC stage. ADC needs 16 clock cycles (1 MHz)

to produce a 16 bits output code. ADC code output is seriallytransmitted on output pin. The ADC has an INL of ±1 LSBand a DNL of 0.75 LSB (worst case). The offset and error gainof the ADC are below 0.5% and 1.5% of the full scale range,respectively. Finally, the battery channel in Fig. 5 is used tomonitor the battery status. To be noted that integrated bandgapcircuits are used to internally generate reference levels from1.1 V to 3.5 V.

In the H@H telemonitoring platform, the CARDIC IC hasbeen configured so that the ECG is acquired involving onlythree channels and the RL driver while the optional BP channelis not used (BP is monitored as discussed in Section III).Although temperature has not been identified by the cliniciansas a key parameter for CHF, the presence of this channel allowsalso for body temperature monitoring useful for upgrades ofthe H@H platform to other disease in order to create a multi-purpose configurable monitoring system.

V. H@H SENSOR SIGNAL PROCESSING

Sensor data signal processing, implemented in the homegateway, assumes an important role within remote health mon-itoring system, being in charge of detecting as soon as possiblealterations in the vital parameters that are potentially dangerousfor the patient. The aim is to extract the meaningful informationin the fastest way. In this way, the physicians are timely alertedif anything out of the ordinary is found, and they have in theHIS all patient’s data to plan the following actions.

The processing core is an Intel ATOM N450 at 1.66 GHz,providing a computational capability of about 5000 MIPS, with


Fig. 8. Temperature channel block diagram.

512 KB L2 on-chip cache memory and off-chip 1 GB DDR2memory. Its computational and storage capability are enoughfor a real-time processing of data collected from the sensors.

Data processing involves three main steps:

1) preprocessing: raw data provided by the sensors are fil-tered to remove noise and main interferences (i.e., powerline or movements artefacts). Extraction and gatheringof derived information are also accomplished (i.e., HRfrom ECG track). Resulting low quality signals arisingfrom sensors misplacement or corrupted by heavy motionartefacts are detected and the system asks for the mea-surement repetition;

2) analysis: the results of the previous block are comparedwith the thresholds that establish the admissibility rangefor punctual values or trends over a medium period.All values must lay within the safe zone defined by theclinicians. Maximum, minimum, and average values arechecked by an expert system;

3) false positive avoidance: to reduce the number of alarmsgenerated for example due to temporary stress, in caseof abnormal value, the same measurement is shortlydeferred and only if confirmed the alarm is raised. Bothobservations are stored and tagged accordingly.

The signal processing chain is implemented in C languageleveraging, respectively VSIPL library, libbluetooth, libSSL,libXML, libgtk to support filters implementation, Bluetoothmanagement, security, XML parsing, and graphical develop-ment. The main concerns of the system, data acquisition, pro-cessing algorithms, data transmission, data presentation, andscheduling of activities were implemented in separated threadsto ensure a high modularity and the maximum flexibility. Apermanent storage of pending data waiting for transmission

avoids that power supply failures will result in data loss. Thegateway provides also the possibility to store in a local databaseall collected vital signs occupying less than 1 Gb for one yearobservation.

A. ECG

ECG signal is, obviously, one of the most significant andreliable sources of information for CHF patients monitoring.Expert cardiologists within the project have considered thatspecial attention should be given to four signs of heartdeterioration:

• Abnormal heart frequency, above 120 or under 50 beatsper minutes

• Emergence of atrial fibrillation (AFIB) episodes• QRS complex of more than 120 ms with complete left

bundle branch block morphology• Signs of myocardial ischemia

Early detection of these symptoms is decisive in the preven-tion of cardiac threats. Home monitoring, such as the one devel-oped in this project, allows the physicians to periodically checkthe ECG of the patients without unnecessary visits neither tothe patient’s home nor to the clinic. However, this can still beimproved. Making a prior study of the ECG at the patient’shouse makes it possible to repeat those measures that obtainedunexpected results and to raise early alarms for the physiciansto check certain patient’s ECG sooner. At this project, the ECGanalysis is separated in two parts. The first one extracts basicfeatures for HR calculation or respiration analysis. The secondpart uses the extracted information to evaluate the presence ofAFIB episodes.


Fig. 9. ECG main points and signal processing for QRS detection.

1) Basic Feature Extraction: The starting point for any ECGanalysis is feature extraction, particularly the QRS complexdetection that is nearly always the reference point within theECG signal. A first derivative-based algorithm [27] combinedwith a rule-based system [28] is used for the QRS complexdetection. Fig. 9 shows the main points of the ECG and thesteps of the processing algorithm. The signal is filtered usingan optimum Kaiser filter band pass 85th-order symmetric FIRin order to keep just the central frequencies (8 to 30 Hz) wherethe QRS complex information lays. This kind of filter has linearphase, constant delay, and it is reasonably sized. Afterwards,the signal is differentiated, squared, and averaged by usinga rectangular sliding window. Comparing the averaged signalagainst a threshold creates a set of windows that allow recog-nizing the R peaks in the filtered signal (maximum positivewithin the window). The R peak has still to pass through arule-based system that evaluates whether the detected QRS isa valid QRS complex or not basing on the distance in timebetween consecutive peaks: i.e., if two peaks are closer than therefractory period of the myocardium (200 ms), one of them isdiscarded. Then, the T-wave discrimination is performed, beingstricter about those peaks situated 200–360 ms later than anaccepted R peak. The rule system is also used to check forpossible missed peaks when the current RR interval is 1.5 timesthe previous RR interval. In addition, a 50 Hz notch filter isapplied to the signal. This filtering does not affect the detection,but it improves the legibility of the track for further medicalreview. Fig. 10(a) and (b) show the raw and filtered signal.There is also a cubic spline data interpolation algorithm thatextracts the envelope of the R peaks as an indication of therespiratory activity. Fig. 10(c) shows the envelope signal ofthe peaks. Maximum, minimum, and average HRs along thetrack are calculated (RR interval) and analyzed using a 30 swindow that is shifted along the time axis on 5 s length steps.The VSIPL library linear FIR function was used to implementthe Kaiser filtering, and additional IIR filtering and cubic splineinterpolation functions have been developed based on the basicfunctions of the library.

2) Abnormal Heart Frequency: At the basic configuration,heart frequency is calculated using a 30 s window shifted

along the time axis on 5 s length steps. For each step, thenumber of beats within the window is counted, and that value isextrapolated to a 60 s window to the beats per minute value.Maximum, minimum, and average HRs along the track arecalculated. If these values go higher or lower than the limitsestablished by the physicians, an alarm is raised.

3) AFIB: AFIB is one of the most common arrhythmias,particularly between elder people. It is the cause of approx-imately one third of hospitalizations for cardiac rhythm dis-turbances. An estimated 2.3 million people in North Americaand 4.5 million people in the European Union suffer fromAFIB, and with the aging of the population, the number ofpatients suffering from this condition is definitively going toincrease. AFIB is a heart condition where the activation inthe atria is fully irregular or chaotic. This might be caused byan irregular focal triggering mechanism involving automaticcells or multiple reentrant wavelets that randomly excite tis-sue that has previously just been activated by the same oranother wavelet. Irregular atrial activity is reflected in theECG as the absence of P waves before the QRS complexand fluctuating waveforms in the baseline. Organized ventricleactivity allows QRS complex to maintain its usual shape, butwith irregular rates so RR interval will show sharp variations,see Fig. 11.

There are multiple approaches to AFIB detection. Many ofthem are based on RR interval variability and its statistics: in[29], the authors compare the standard density histogram ofthe RR interval length (RR) and RR difference (ΔRR) withpreviously compiled standard density histograms segments dur-ing AFIB. In [30], the previously calculated histograms of thenormalized ΔRR for AFIB and non-AFIB episodes are usedto calculate the probability that a sequence of consecutive RRis part of an AFIB episode. They compare the histograms ofthe normalized ΔRR for both AFIB and non-AFIB episodesand demonstrate that they can be approximated by Laplace andGaussian distributions. In [31], a linear discriminant classifierfed with RR intervals statistics is used. A method based on therandom characteristics of the AFIB intervals and the ShannonEntropy is discussed in [32]. Both [29] and [31] need AFIBtemplates or training data while [32] is a purely statistic methodthat does not need any training data.

Another common approach is to separate the atrial activityfrom the ventricular activity in order to analyze the atrialbehavior. It is necessary to take into account that atrial andventricular activity occurs in the same frequencies and some-times, as during AFIB, at the same time. There are multipledocumented methods to do this. Some are based on detect-ing and subtracting the QRS complex. Others are based onrepresenting the ECG signal in a distributed way (principalcomponent analysis (PCA), wavelets, etc.) that allows the atrialactivity to be separated. [33] presents a solution based onPCA (Karhunen-Lòeve Transform). After calculating first 12coefficients of KLT of the V1 lead, it concludes that coefficients1–2 contain ventricular information, 3–8 contain atrial activityinformation, and 9–12 contain noise. [34] reconstructs the atrialactivity using components 3–8 and after makes a study infrequency of that activity. [34] presents a similar solution basedon wavelets. Atrial activity is mainly contained in 4–9 Hz


Fig. 10. (a) ECG signal; (b) filtered signal plus R peaks; (c) R envelope signal.

Fig. 11. AFIB detection.

frequencies and can be isolated using certain coefficients of thestationary wavelet transform (SWT). Other authors mix classicRR interval analysis with additional indicators to decrease thenumber of false positives detected with the previous methods.[35] proposes a method based on RR intervals mixing it withP-wave detection. It also includes a hysteresis counter to de-termine the beginning of an episode only if several consecutivesets of parameters have been classified as AFIB. It uses Markovprocess modeling to analyze RR interval regularity. P wavedetection considers P wave location (PR interval variation) andmorphology.

For this project, a variation of the approach in [35] has beenadopted. RR interval variation and PR interval variation areused to detect the presence of AFIB episodes. As shown inFig. 12, each of them is modeled as a separate three stateMarkov process. Each interval is evaluated to decide whetherit is short (S), regular (R), or long (L) with respect to the mean

of the interval length (IntLen); interval means (IntMeam) willbe determined recursively following the relation

IntMean=(mw)·IntMean(i− 1) + (1−mw)·IntLen(i).(2)

The mean weight parameter determines how sensible shouldthe mean be to the variations of the actual interval. Any Markovprocess is characterized by two transition probability matrixes,see Table IV, where the probability of transition from one stateto another is represented for each of the studied cases (AFIB ornon AFIB beat).

The best feature of Markov processing is its capability toevaluate a sequence of events before reaching a conclusion.In this case, it recursively evaluates the probability of a beatbelonging to an AFIB episode considering the state transitionsof the previous chainLength beats

Initialcondition : proAFIB(i)=1; proNonAFIB(i)=1;

recursivecondition : for k ∈ (2, chainLength)

proAFIB(i)=proAFIB(i)

· transMatAF (RRcode(k), RRcode(k−1))

proNonAFIB(i)=proNonAFIB(i)

· transMatNonAF

×(RRcode(k), RRcode(k−1))

Finally, the Markov score for a single beat will be

score(i) = log

(proAFIB(i)

proNonAF (i)

). (3)

Beat to beat AFIB detection has a huge problem with veryshort detected episodes. A three-beat fibrillation episode can


Fig. 12. AFIB detection flow chart.

TABLE IVTRANSITION PROBABILITY MATRIXES (TRANSMAT) FOR RR INTERVAL

TABLE VSTATISTIC RESULTS (SE: sensitivity = TP/(TP + FN), PPV:

Positive predictive Value = TP/(TP + FP))

hardly be of any interest. In order to avoid false positive detec-tions, two symmetric independent hysteresis counters have beenimplemented. They evaluate separately the possibility of a beatbeing AFIB giving a score from −10 to 10. Table V summarizesthe results obtained with an external evaluator (Epicmp functionincluded in physioTools [40]) excluding from the statisticsall episodes shorter than 1 min, as they are discarded inthe application. Sensitivity gives an idea of the accuracy ofthe system, while positive predicted value gives an idea of therobustness of the system against false alarms (false positives).The obtained results show the robustness of the system. It isimportant to remark that this algorithm is only used to giveadditional information and warnings to the medical personal ofthe project, not for diagnose.

B. SpO2 and Plethysmographic Wave

Oxygen saturation indicates the percentage of haemoglobinmolecules in the arterial blood which are saturated with oxygen.Concerning SpO2 signal analysis, normal values in healthyadults range from 94% to 100%, while in CHF patients, oxy-

genation level outside this interval could be assumed as warningsituation. A frequent monitoring is necessary to avoid thesechanges to go undetected. Typically, SpO2 is acquired onceor twice per day, and each acquisition lasts some minutes. Theanalysis step for SpO2 has in the worst case a complexity linearwith the number of samples both in computation and memory.It involves maximum and minimum levels over the track andalso the average value, extracted by an on-line digital low-passFIR filter. An example of trend of the average SpO2 value alonga week is shown in Fig. 13. In this case, all values are includedin the safety zone and no alarms are raised.

The plethysmographic wave is a simple way to detect bloodvolume changes in the microvascular bed of tissue. The wave-form is composed by two main parts: a pulsatile physiologicalwaveform due to cardiac synchronous changes in the bloodvolume with each heart beat and a slowly varying (“DC”)baseline with various lower frequency components attributedto respiration [36]. The availability of this signal in the systemrepresents another source for the HR calculation and moreoverleaves open the possibility to extract information about therespiration frequency in non-invasive fashion. There is also inliterature some evidence that the BP could be extracted fromthis signal [37].

C. BP

BP is the pressure exerted by circulating blood upon thewalls of blood vessels. During each heartbeat, BP varies be-tween a maximum (systolic) and a minimum (diastolic) level.In this segment of population, the abnormality of punctualvalues of BP and its variability in a short period are the mainmanifestations of cardiac instability. For these reasons, moremeasurements per day are suggested. The systolic and diastolicpunctual values provided by the sensor are analyzed to find un-der or over threshold situations, and the general trends of bothparameters are verified looking for suspicious variability. Thecomplexity in terms of memory and computation is linear withthe number of values considered. An example of observationof BP, one month long, is shown in Fig. 14 along with thesafety thresholds. It is possible to observe an under thresholdand an over threshold situation, respectively, for the diastolicand systolic parameters.


Fig. 13. Example of SpO2 average level trend (29th of June to 6th of July).

Fig. 14. Example of blood pressure trends, comparison thresholds defined in the OP are also visible (1 month).

Fig. 15. Example of weight trend from 20th to 28th July.

D. Weight

This parameter is related to the body mass. As far as weightsignal processing, it is very easy to measure as well as very ef-fective in the CHF management. A rapid gain of this parametermeans fluid retention in the patient, and it is one of the cardinaland dangerous manifestation of CHF. Frequent monitoring ofthe weight trend allows for simple detection of a gain. Increasesof 1 Kg in a day or 3 Kg in a week is generally considered asalarm situation. The system is able to store a sufficient series ofpunctual values sent by the scale in order to extract the trend.To avoid impossible values of weight, i.e., other people usesthe scale, a difference of more than 3 kg respect to the lastobservation leads to discard it. Fig. 15 shows an example ofweight trend where the signal processing step finds out two

dangerous situations: a gain of 1.4 Kg within 24 h and thesecond one due to 3.0 Kg gain in the last 7 days.

VI. TESTING AND RESULTS

Exhaustive testing of HW/SW platforms is a key issue fortheir adoption in telemedicine systems [38], [39]. Each devel-oped component passed the unit test. From the hardware pointof view, the novel ECG-SpO2 module was subjected to theelectrical certification, and the home gateway was tested afterthe replacement of the native keyboard with the custom mem-brane keypad. The operational correctness of the software ele-ments has been verified using conventional software validationprocedures. Both the sensor firmware and the multithreading


application software executed on the home gateway have beenheavily tested, with particular attention to the concurrence andthe responsivity. Moreover, all the signal processing algorithmswere tested making use of signal simulators (i.e., ECG simu-lator) and manually generated values. Additionally, the QRSdetection algorithm has been tested using PhysioToolkit andPhysioNet databases [40], obtaining a 99.7% of sensitivity and99.8% of positive predictive value on MIT_BIH_Arrythmiadatabase [41] and also on MIT_BIH AFIB [41] obtaining a99.54% of sensitivity and 99.36% of positive predictive value.The following integration test verified the end-to-end commu-nication in the system, and the overall system-level interactionsof the different HW and SW parts, also with the execution of ad-hoc prepared scenarios and the involvement of healthy users totest if the behavior of the systems meets the expected one. An-other important phase before the final technical demonstrationwas conducted enrolling for a month two patients, minimallyaware of ICT and younger than the CHF average age. The firstimpressions of physicians and patients coming from these testswere taken into account to tune the final version of the system.

Once completed the unit and system prototype, a techni-cal validation of the system has been implemented involving30 patients with CHF disease in NYHA class III and IV, withan average age of 62 years and recently hospitalized for HF.The size of the set of CHF-affected patients is similar to thoseof other works published in literature about ICT systems forCHF monitoring [42]. The minimum period of monitoringwas one month. All devices were provided in a single bag,with the dimension of those carrying 15′′ laptop computer,and an overall weight of 2, 8 kg to be transportable every-where. Patients were enrolled in the study at time of dischargefrom the hospital where they were admitted for acute heartfailure or during a routine ambulatory visit. Main inclusioncriteria were diagnosis of heart failure, class NYHA ≥ III, atleast one hospitalization for acute heart failure in the previous6 months, agreement to take part in the study. Acute coronarysyndrome within 3 months before the enrolment was the onlyexclusion criterion. Physicians that took part to the study wereall cardiologist, involved in the management of patients withheart failure. They also checked out information arrival in HIS,evaluating the quality and coherence of data collected andthe relevance of the alarms. The quality and the reliability ofacquired signals and generate alarms, the robustness of datatransmission, and the system effectiveness from the medicalperspective were evaluated as a key point of system function-alities. On the other hand, the ergonomic of patient’s interfaceswas evaluated, as well as the general end-user usability ofthe sensing devices and the home gateway. A specific testingprotocol and a questionnaire have been developed to gatherpatients, caregivers, and physicians feedbacks and validate thesystem.

The results show a very limited number of activity misses(< 3%), mostly in the first days of monitoring, confirming alsothe property of such system to improve the therapy compliance.Moreover, the number of false positive alarms is less than 5%.No connectivity and transmission problems, including data lost,occurred. All end-users reported a positive feedback and goodsatisfaction level in the final questionnaire. The scores reached

TABLE VIPHYSICIANS’ AGGREGATION FEEDBACKS

Fig. 16. Feedbacks aggregation result of patients.

in each macro-parameters are shown in Table VI and Fig. 16,respectively, for medical staff and patients. Physicians reportedthat the use of this platform does not load up in a significant waytheir regular activity, but represents a valid means to controlat distance the evolution of the followed patients thanks to thehigh quality of acquired signals and alarm detection capability.All physicians involved in the demonstration are definitivelyin favor of the adoption of the H@H system. The 89% of thepatients report a very high satisfaction level, highlighting thefriendliness of the solution and the easiness to follow the dailytherapy.

Due to the success of the H@H technology test under medi-cal control, a clinical validation, including an economical evalu-ation (with OPEX/CAPEX capital and operational expenditureanalysis), with more than 500 patients has been already plannedin the Italian Regional Tuscany Health System.

VII. COMPARISON WITH THE STATE OF THE ART

This section underlines the distinctive features of the H@Hplatform with respect to telemedicine systems specific for CHFor suitable for this kind of disease and with some dedicatedprototypes and projects. Systems based on telephone calls orweb portal [43] to report symptoms and measures have to bediscarded for scalability and usability issues.


Similar commercial systems in terms of supported vital signs[44]–[46] use proprietary data formats and dedicated databaseslimiting the degree of interoperability and integration withexisting departmental HIS, while they rely on dedicated callcenters. The user interfaces and the collection gateways (i.e.,PDA or smartphone) are in some cases quite complicated withrespect to the patient’s ability [45],[46]. Others involve 12 leadECG prone to mispositioning issues [47]. In literature, thereare systems where often one main parameter, pulse oximetryas in [48] or ECG only as in [49] or [50], is acquired (notenough for CHF). Despite the importance of the activity level,home monitoring using motion sensors [51], [52] is useful morefor presence or fall detection rather than for early diagnosis ofCHF. In [53], weight and BP are monitored, and the user has tomanually type the values in the gateway for sending, while [54]is based on a T-shirt equipped with sensing areas to measurethe ECG, temperature and activity level. In different ways,they result scarcely usable for elderly patients and provideless vital signs that identified in the medical requirements.Our architecture differs from similar ones such as [55]. In thatwork, only ECG and SpO2 are included that again are less thanthe minimum required by clinicians for clinical monitoring ofpatient with CHF. Additionally the user interface running onPC is not suitable for elderly people. All previous solutions donot include advanced signal processing functions in the homegateway for early warning of the remote system. Instead, theydemand all processing to the remote host, applying only noiseremoval on the captured signals.

The chipsets for ECG front-end and parts of the sig-nal processing (i.e., Texas Instruments [56], [57] and IMEC[58]–[61]) cannot be compared to H@H system since theyrepresent only a subsystem as the CARDIC subunit. CARDICoffers a multi-channel analog front-end IC for bio sensor in-terfacing. Texas Instruments has a family of IC ADS119x andADS129x and ADs1258/ADs1278 with up to eight channels(CMRR up to 100 dB, noise levels in the order of few μV),with PGA, 16 bit or 24 bit ADC integrating the circuitry neededto interface 1 or multiple ECG or EEG electrodes. However,some external components to build up a complete monitoringsystem (i.e., channels for external temperature sensor or BPsensors, pace maker detector) are missing. CARDIC insteadintegrates all missing elements (see Table III) reaching a deeperintegration level than the TI 16-bit ADC version. IMEC hasdeveloped several versions of ICs for ECG and EEG withintegrated signal processing in the analog or digital domain.Such systems are optimized in terms of low power consumption(less than 1 mW with designs realized in 0.5 μm or 180 nmwhile CARDIC is in 0.8 μm) but have only 1 channel andjust dedicated to ECG while CARDIC has 12 channels forinterfacing ECG electrodes or other sensors. Hence, IMECdesign is more suited for ultra low-power wireless HR mon-itoring system for wellness or general healthcare. CARDICis more suited for clinical applications where more channelsand sensors are required. H@H systems do not require ultralow power, and the advanced signal processing is demanded tothe home gateway, which is realized using a commercial off-the-shelf programmable core, the Atom processor, rather thandesigning application-specific DSP system integrating single-

chip microcontroller and hardware accelerator IP cells [62],[63]. Overall, the H@H system addresses the usability concernsthrough an easy-user interface designed to allow the patientto follow autonomously the therapy at home. It allows multi-parametric monitoring, according to the clinicians recommen-dations, based on easy-to-use sensing devices (i.e., 3-lead ECG,BP, weight in basic configuration) and flexibility, through theOP, to meet the individuality of the end users.

VIII. CONCLUSION

This work presents the requirements and the realization interms of sensing devices and sensor data signal processingof a complete and integrated ICT platform to improve theprovisioning of healthcare services for CHF patients. The H@Hsystem proposes an innovative home care model in order tosupport in integrated and coordinated fashion the whole processof the patient treatment, connecting in-hospital care with out-of-hospital follow up. With the remote monitoring, the medicalstaff can realize changes in the parameters of patients withoutfrequently visiting them and consequently they can take con-cerned action to prevent possible aggravations. The benefitsextend beyond the early detection of clinical exacerbation to op-timizing specialized resources scheduling and to reduce unnec-essary travels to hospital. The system definition was completelydriven by the end-users resulting in a platform particularlyeffective and practical with respect to other telemonitoring trialsand state-of-art products. One of the main system novelties isrepresented by the home gateway which embeds, through theso-called OP, the medical prescription for any given patient.Moreover, it locally performs all the sensor signal processingand alarm detection. The OP is configurable according to thepatient’s needs and remotely updatable if required via theserver platform. The basic sensors kit established by physiciansincludes two commercial devices for weight and BP and anew developed module for acquiring synchronized ECG andSpO2 track leveraging a multi-channel front-end IC for cardiacsensors. This again represents a novelty aspect with respectto the state-of-the-art together with the use of internationalstandards for data exchange to favor the integration/interactionof the platform with other systems or biomedical sensors. Theoverall cost of a kit, considering the first prototype and testphase of H@H with 30 kits produced, is close to 1000 euros.Adopting the H@H telemonitoring CHF system in a large scale,e.g., in national health system, the kit cost can be reduced inthe range of hundreds of euros. First technology assessmentin a real medical scenario with tens of patients affected byCHF disease NYHA class III and IV, under remote control forsome months of cardiologists using their usual HIS, proves theeffectiveness of the telemonitoring system from both patientsand caregivers point of view. Key points underlined by the end-users are the goodness of the sensor data analysis implementedat the home side, enabling early and accurate detection ofdestabilizations in vital signs and naturally the usability ofthe system. Summarizing the H@H system meets end-userexpectations, and, in particular, physician have already planneda clinical validation, including an economical evaluation, withmore than 500 patients in the Italian Regional Tuscany HealthSystem.


REFERENCES

[1] “SHAPE survey results to the general public,” in Proc. Annu. Congr. Eur.Soc. Cardiol. Vienna, Vienna, Austria, Sep. 2003.

[2] F. Zannad, N. Agrinier, and F. Alla, “Heart failure burden and therapy,”Europace, vol. 11, no. 5, pp. 1–9, Nov. 2009.

[3] V. Roger, A. S. Go, D. M. Lloyd-Jones, R. J. Adams, J. D. Berry,T. M. Brown, M. R. Carnethon, S. Dai, G. de Simone, E. S. Ford,C. S. Fox, H. J. Fullerton, C. Gillespie, K. J. Greenlund, S. M. Hailpern,J. A. Heit, P. M. Ho, V. J. Howard, B. M. Kissela, S. J. Kittner,D. T. Lackland, J. H. Lichtman, L. D. Lisbeth, D. M. Makuc,G. M. Marcus, A. Marelli, D. B. Matchar, M. M. McDermott,J. B. Meigs, C. S. Moy, D. Mozaffarian, M. E. Mussolino, G. Nichol,N. P. Paynter, W. D. Rosamond, P. D. Sorlie, R. S. Stafford, T. N. Turan,M. B. Turner, N. D. Wong, and J. Wylie-Rosett, “Heart disease and strokestatistics—2011 update: A report from the American Heart Association,”Circulation, vol. 123, no. 4, pp. 18–209, Feb. 2011.

[4] F. Alla, F. Zannad, and G. Filippatos, “Epidemiology of acute heart failuresyndromes,” Heart Fail. Rev., vol. 12, no. 2, pp. 91–95, Jun. 2007.

[5] C. Berry, D. Murdoch, and J. McMurray, “Economics of chronic heartfailure,” Eur. J. Heart Fail., vol. 3, no. 3, pp. 283–291, Jun. 2001.

[6] A. Bundkirchen and R. H. G. Schwinger, “Epidemiology and economicburden of chronic heart failure,” Eur. Heart J. Suppl., vol. 6, no. SD,pp. 57–60, Aug. 2004.

[7] J. Ross, J. Chen, Z. Lin, H. Bueno, J. P. Curtis, P. S. Keenan, S. L.Normand, G. Schreiner, J. A. Spertus, M. T. Vidán, Y. Wang, and H. M.Krumholz, “Recent national trends in readmission rates after heart failurehospitalization,” Circ., Heart Fail., vol. 3, no. 1, pp. 97–103, Jan. 2009.

[8] S. Stewart, “Financial aspects of heart failure programs of care,” Eur. J.Heart Fail., vol. 7, no. 3, pp. 423–428, Mar. 2005.

[9] F. McAlister, S. Stewart, S. Ferrua, and J. J. McMurray, “Multidisciplinarystrategies for the management of heart failure patients at high risk foradmission: A systematic review of randomized trials,” J. Amer. Coll.Cardiol., vol. 44, no. 4, pp. 810–819, Aug. 2004.

[10] E. Seto, “Cost comparison between telemonitoring and usual care ofheart failure: A systematic review,” Telemed. J. E-Health, vol. 14, no. 7,pp. 679–686, Sep. 2008.

[11] C. Klersy, A. De Silvestri, G. Gabutti, F. Regoli, and A. Auricchio, “Ameta-analysis of remote monitoring of heart failure patients,” J. Amer.Coll. Cardiol., vol. 54, no. 18, pp. 1683–1694, Oct. 2009.

[12] S. C. Inglis, R. A. Clark, F. A. McAlister, J. Ball, C. Lewinter,D. Cullington, S. Stewart, and J. G. Cleland, “Structured telephone sup-port or telemonitoring programmes for patients with chronic heart fail-ure,” Cochrane Lib., no. 8, p. CD 007 228, Jul.–Sep. 2010.

[13] S. Saponara, M. Donati, T. Bacchillone, L. Fanucci, I. Sanchez-Tato,C. Carmona, and P. Barba, “Remote monitoring of vital signs in patientswith chronic heart failure: Sensor devices and data analysis perspective,”in Proc. IEEE Sens. Symp. Appl., 2012, pp. 1–6.

[14] G. van den Broek, F. Cavallo, and C. Wehrmann, Ambient Assisted LivingRoadmap. Amsterdam, The Netherlands: IOS Press, Mar. 2010.

[15] S. J. Devaraj and K. Ezra, “Current trends and future challenges in wire-less telemedicine system,” in Proc. IEEE ICECT , 2011, pp. 417–421.

[16] C. Fabbricatore, M. Zucker, S. Ziganki, and A. P. Karduck, “Towards anunified architecture for smart home and ambient assisted living solutions:A focus on elderly people,” in Proc. IEEE DEST , Jun. 2011, pp. 305–311.

[17] A. Jara, M. A. Zamora, and A. F. G. Skarmeta, “An architecture forambient assisted living and health environments,” in Proc. Lecture NotesComput. Sci., 2009, vol. 5518, pp. 882–889.

[18] J. P. Riley and M. R. Cowie, “Telemonitoring in heart failure,” HeartEduc. Heart, vol. 95, no. 23, pp. 1964–1968, Dec. 2009.

[19] T. Bacchillone, M. Donati, S. Saponara, and L. Fanucci, “A flexible homegateway system for telecare of patients affected by chronic heart failure,”in Proc. IEEE Int. Symp. Med. Inf. Commun. Technol., 2011, pp. 139–142.

[20] R. Dolin, L. Alschuler, S. Boyer, C. Beebe, F. M. Behlen, P. V. Biron, andA. Shabo, “HL7 clinical document architecture, release 2,” J. Amer. Med.Inf. Assoc., vol. 13, no. 1, pp. 30–39, Jan./Feb. 2006.

[21] Implementation Guide for CDA Release 2.0 Personal Healthcare Moni-toring Report (PHMR), Health Level Seven Int., Ann Arbor, MI, 2008.[Online]. Available: http://www.hl7.org

[22] M. Yuksel and A. Dogac, “Interoperability of medical device informationand the clinical applications: An HL7 RMIM based on the ISO/IEEE11073 DIM,” IEEE Trans. Inf. Technol. Biomed., vol. 15, no. 4, pp. 557–566, Jul. 2011.

[23] Systemized Nomenclature of Medicine—Clinical Terms, Nat. Lib. Med.,Bethesda, MD, 2007. [Online]. Available: http://www.nlm.nih.gov

[24] Logical Observation Names and Identifier, Indianapolis, IN. [Online].Available: http://loinc.org/

[25] B. Schijvenaars, G. van Herpen, and J. A. Kors, “Intraindividual variabil-ity in electrocardiograms,” J. Electrocardiol., vol. 41, no. 3, pp. 190–196,May/Jun. 2008.

[26] S. Saponara, E. Petri, L. Fanucci, and P. Terreni, “Sensor modeling, low-complexity fusion algorithms, and mixed-signal IC prototyping for gasmeasures in low-emission vehicles,” IEEE Trans. Instrum. Meas., vol. 60,no. 2, pp. 372–384, Feb. 2011.

[27] P. Jiapu and W. J. Tompkins, “A real-time QRS detection algorithm,”IEEE Trans. Biomed. Eng., vol. BME-32, no. 3, pp. 230–236, Mar. 1985.

[28] N. Arzeno, Z. Deng, and C. Poon, “Analysis of first-derivative based QRSdetection algorithms,” IEEE Trans. Biomed. Eng., vol. 55, no. 2, pp. 478–484, Feb. 2008.

[29] K. Tateno and L. Glass, “Automatic detection of atrial fibrillation using thecoefficient of variation and density histograms of RR and ΔRR intervals,”Med. Biol. Eng. Comput., vol. 39, no. 6, pp. 664–671, Nov. 2001.

[30] A. Ghodrati and S. Marinello, “Statistical analysis of RR interval irreg-ularities for detection of atrial fibrillation,” in Proc. Comput. Cardiol.,2008, vol. 35, pp. 1057–1060.

[31] R. Shouldice, C. Heneghan, and P. de Chazal, “Automated detection ofparoxysmal atrial fibrillation from inter-heartbeats intervals,” in Proc.IEEE EMBC, 2007, pp. 686–689.

[32] S. Dash, K. H. Chon, S. Lu, and E. A. Raeder, “Automatic real time de-tection of atrial fibrillation,” Ann. Biomed. Eng., vol. 39, no. 9, pp. 1701–1709, Sep. 2009.

[33] I. R. Legarreta, “Component selection for principal component analysis-based extraction of atrial fibrillation,” in Proc. Comput. Cardiol., 2006,vol. 33, pp. 137–140.

[34] B. Weng, J. J. Wang, F. Michaud, and M. Blanco-Velasco, “Atrial fibrilla-tion detection using stationary wavelet transform analysis,” in Proc. IEEEEMBC, 2004, pp. 1128–1131.

[35] S. Babaeizadeh, R. E. Gregg, E. D. Helfenbein, J. M. Lindauer, and S. H.Zhou, “Improvements in atrial fibrillation detection for real-time moni-toring,” J. Electrocardiol., vol. 42, no. 6, pp. 522–526, Nov./Dec. 2009.

[36] J. Allen, “Photoplethysmography and its application in clinical phys-iological measurement,” Physiol. Meas., vol. 28, no. 3, pp. R1–R39,Mar. 2007.

[37] M. Asif-Ul-Hoque, “Measurement of blood pressure using photoplethys-mography,” in Proc. 13th Int. Conf. Comput. Model. Simul. (UKSim),2011, pp. 32–35.

[38] B.-K. Miller and W. MacCaull, “Toward web-based careflow managementsystems,” J. Emerg. Technol. Web Intell., vol. 1, no. 2, pp. 137–145,Nov. 2009.

[39] B. Chen, G. S. Avrunin, E. A. Henneman, L. Clarke, L. J. Osterweil, andP. L. Henneman, “Analyzing medical processes,” in Proc. ICSE, 2008,pp. 623–632.

[40] A. L. Goldberger, L. A. Amaral, L. Glass, J. M. Hausdorff, P. C. Ivanov,R. G. Mark, J. E. Mietus, G. B. Moody, C. K. Peng, and H. E. Stanley,“PhysioBank, PhysioToolkit, and PhysioNet: Components of a new re-search resource for complex physiologic signals,” Circulation, vol. 101,no. 23, pp. 215–220, Nov. 2000.

[41] “MIT-BIH Arrythmia Database and Atrial Fibrillation Database,”Harvard-MIT Div. Health Sci. Technol., Cambridge, MA, 2000. [Online].Available: http://www.physionet.org

[42] L. Pecchia et al., “Discrimination power of short-term heart rate variabil-ity measures for CHF assessment,” IEEE Trans. Inf. Technol. Biomed.,vol. 15, no. 1, pp. 40–46, Jan. 2011.

[43] Insight Telehealth Syst., Phoenixville, PA. [Online]. Available: http://www.itsmyhealthyheart.com

[44] Aerotel Medical System, Holon, Israel. [Online]. Available: http://www.aerotel.com

[45] Tunstall, Telgte, Germany. [Online]. Available: http://www.tunstall.it[46] Telbios, Milan, Italy. [Online]. Available: http://www.telbios.com[47] Parsys Telemedicine, Paris, France. [Online]. Available: http://www.

parsyssante.com[48] P. Crilly, E. T. Arakawa, D. L. Hedden, and T. L. Ferrell, “An integrated

pulse oximeter system for telemedicine applications,” in Proc. IEEEIMTC, 2007, pp. 102–104.

[49] C. De Capua, A. Meduri, and R. Morello, “A smart ECG measurementsystem based on web-service-oriented architecture for telemedicine ap-plication,” IEEE Trans. Instrum. Meas., vol. 59, no. 10, pp. 2530–2538,Oct. 2010.

[50] Y.-W. Bai, C. Y. Cheng, C.-L. Lu, and Y.-S. Huang, “Design and imple-mentation of an embedded remote ECG measurement system,” in Proc.IEEE Instrum. Meas. Technol. Conf., 2001, pp. 1401–1406.

[51] M. Alhamid, J. Saboune, A. Alamri, and A. El Saddik, “Hamon: Anactivity recognition framework for health monitoring support at home,”in Proc. IEEE IM2TC, 2011, pp. 1–5.


[52] G. Gupta, “Design of a low-cost physiological parameter measurementand monitoring device,” in Proc. IEEE IMTC, 2007, pp. 1–6.

[53] A. Gund, I. Ekman, K. Lindecrantz, B. A. Sjoqvist, E. L. Staaf, andN. Thorneskold, “Desing evaluation of an home-based telecare system forChronic heart failure patients,” in Proc. IEEE Eng. Med. Biol Soc., 2008,pp. 5851–5854.

[54] E. Villalba, D. Salvi, M. Ottaviano, I. Peinado, M. T. Arredondo, andA. Akay, “Wearable and mobile system to manage remotely heart failure,”IEEE Trans. Inf. Technol. Biomed., vol. 13, no. 6, pp. 990–996, Nov. 2009.

[55] J. Yao, R. Schmitz, and S. Warren, “A wearable point-of-care system forhome use that incorporates plug-and-play and wireless standards,” IEEETrans. Inf. Technol. Biomed., vol. 9, no. 3, pp. 363–371, Sep. 2005.

[56] K. Soundarapandian and M. Berarducci, “Analog Front-End Designfor ECG Systems Using Delta-Sigma ADCs,” TI Rep. SB AA1 60A,pp. 1–11, Apr. 2010.

[57] “Low-Power, 8-Channel, 16-Bit Analog Front-End for BiopotentialMeasurements,” Rep. SB AS4 71C, pp. 1–76, Nov. 2011.

[58] R. F. Yazicioglu, S. Kim, T. Torfs, H. Kim, and C. Van Hoof, “A 30 μWanalog signal processor ASIC for portable biopotential signal monitor-ing,” IEEE J. Solid-State Circuits, vol. 46, no. 1, pp. 209–223, Jan. 2011.

[59] R. F. Yazicioglu, P. Merken, R. Puers, and C. Van Hoof, “A 60 μW60 nV/Hz readout front-end for portable biopotential acquisition systems,”IEEE J. Solid-State Circuits, vol. 42, no. 5, pp. 1100–1110, May 2007.

[60] H. De Groot, M. Ashouei, J. Penders, V. Pop, M. Vidojkovic,B. Gyselinckx, and F. Yazicioglu, “Human++: Key challenges and trade-offs in embedded system design for personal health care,” in Proc.Euromicro Conf. Digital Syst. Des., 2011, pp. 4–10.

[61] H. Kim, R. F. Yazicioglu, S. Kim, N. Van Helleputte, A. Artes,M. Konijnenburg, J. Huisken, J. Penders, and C. Van Hoof, “A config-urable and low-power mixed signal SoC for portable ECG monitoringapplications,” in Proc. IEEE VLSI Symp. Circuits, 2011, pp. 142–143.

[62] N. E. L’Insalata, S. Saponara, L. Fanucci, and P. Terreni, “Automaticsynthesis of cost effective FFT/FFT cores for VLSI OFDM systems,”IEICE Trans. Electron., vol. E91-C, no. 4, pp. 487–496, Apr. 2008.

[63] L. Fanucci, S. Saponara, and A. Morello, “Power optimization of an 8051-compliant IP microcontroller,” IEICE Trans. Electron., vol. E88-C, no. 4,pp. 597–600, Apr. 2005.

Luca Fanucci (M’95) received the Laurea and thePh.D. degrees in electronic engineering from theUniversity of Pisa, Pisa, Italy, in 1992 and 1996,respectively.

From 1992 to 1996, he was with the EuropeanSpace Agency—ESTEC, Noordwijk (NL), as a Re-search Fellow. From 1996 to 2004, he was a SeniorResearcher of the Italian National Research Councilin Pisa. He is a Professor of microelectronics systemsat the Faculty of Engineering of Pisa University. Hisresearch interests include several aspects of design

technologies for integrated circuits and electronic systems, with particularemphasis on system-level design, hardware/software codesign, and sensorconditioning and data fusion. The main applications areas are in the fieldof wireless communications, low-power multimedia, automotive, healthcare,ambient assisted living, and technical aids for independent living. He is acoauthor of more than 300 journal and conference papers and a coinventor ofmore than 30 patents. He served in several technical program committees ofinternational conferences. He was the Program Chair of DSD 2008, ApplicationTrack Chair at DATE from 2006 to 2010, and Tutorial Chairs at DATE in2011 and 2012. Currently, he is serving as the Vice-Program Chair at DATE2013. He is a Member of the Editorial Board of Elsevier Microprocessors andMicrosystems Journal and IOS Press Technology and Disability Journal.

Sergio Saponara (M’12) received the M.Sc., cumlaude, and the Ph.D. degrees in electronic engineer-ing from the University of Pisa, Pisa, Italy.

In 2002, he was with IMEC (B) as Marie CurieResearch Fellow. He is an Associate Professor atthe University of Pisa, Pisa in the field of elec-tronic circuits and systems and authored more than170 scientific publications and ten patents. He isan Associate Editor of the Journal Real-Time Im-age Processing, and he is a Reviewer of severalIEEE/Springer/Elsevier journals and a Member of

the committee of several IEEE/SPIE conferences.

Tony Bacchillone received the M.Sc. degree in com-puter engineering from the University of Pisa, Pisa,Italy, in 2007, and he is working toward the last yearof the Ph.D. program in information engineering atthe same university.

His research topics include intellectual propertiesconfigurability and packaging, code abstraction, anddesign-flow automation for digital design. Currently,he is with Consorzio Pisa Ricerche, Pisa, Italy,where he is involved with the Electronic Systemsand Microelectronic Division on several projects of

industrial relevance in the fields of very-large-scale integration architectures forhigh-speed digital communication systems, and hardware/software embeddedelectronic systems mainly concerning medical telemomitoring applications.

Massimiliano Donati received the M.Sc degree,cum laude, in computer engineering from the Uni-versity of Pisa, Pisa, Italy, in 2009. Since 2011,he has been a Ph.D student in the Department ofInformation Engineering of the same university.

His research interests include the embedded andpervasive systems, particularly for Ambient AssistedLiving and Assistive Technology; the sensor data sig-nal processing and the digital design. Currently, he isinvolved in various European and national projectson embedded systems and digital design of IPs for

space application. Finally, he is a coauthor of scientific papers that appeared inIEEE and SPIE conferences.

Pierluigi Barba received the M.Sc degree in elec-tronical engineering from the University of Pisa,Pisa, Italy, in 2008, with the specialization in bio-electronics, focusing on wearable biosensors to beapplied in biofeedback systems for telemonitoring.

He has been working for CAEN SpA Company,Viareggio Lucca, Italy since October 2008 as De-signer and Integrator of electronic portable devicesaimed at biomedical applications. He was involved inframeworks of national and European projects abouthomecare and clinical care. His professional experi-

ence is centered on front-end electronics design for non-invasive biosensors,system integration, firmware, and software programming.

Isabel Sánchez-Tato received the M.Sc degree intelecommunication engineering at the University ofMálaga, Málaga, Spain.

From 2004 to 2007 she worked on signal pro-cessing for communications at at4Wireless, Málaga.In 2007, she also worked as a Researcher for theDepartment of Electronic Technology at the Univer-sity of Málaga, Málaga, focusing on robotics andnavigation. During 2008, she worked on wirelesscommunications at picoChip Design Ltd., U.K. From2009 to 2012, she worked as a Researcher developing

algorithms and software for biosignal analysis at CITIC (ES).

Cristina Carmona received the B.Sc. degreein computer engineering from the University ofMalaga, Málaga, Spain, in 1998. After getting herB.Sc., she worked at Novasoft developing softwarefor hospitals and health centers. From 2001 to 2008,she worked at the University of Malaga, Malaga andreceived the M.Sc. degree in software engineeringand artificial intelligence in 2004 and the Ph.D.degree in computer science from this universityin 2011.

During those years, she worked with machinelearning algorithms and data mining techniques applied to student modeling ine-learning systems. From 2009, she is working as a Researcher at CITIC (ES).Her research interests include data analysis and user interface and interaction.

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 300 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages false /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 1200 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False

/Description > /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ > /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles false /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /DocumentCMYK /PreserveEditing true /UntaggedCMYKHandling /LeaveUntagged /UntaggedRGBHandling /UseDocumentProfile /UseDocumentBleed false >> ]>> setdistillerparams> setpagedevice

Sensing Devices and Sensor Signal Processing for Remote ... IEEE PAPERS/embedded...Pierluigi Barba,...

Documents

Transcript of Sensing Devices and Sensor Signal Processing for Remote ... IEEE PAPERS/embedded...Pierluigi Barba,...