Sensors and Actuators B 156 (2011) 350359Contents lists available at ScienceDirectSensors and Actuators B: Chemicalj our nal homepage: www. el sevi er . com/ l ocat e/ snbMobile phone platform as portable chemical analyzerAntonio Garcaa, M.M. Erenasb, Eugenio D. Marinettoa, Carlos A. Abada,Ignacio de Orbe-Payab, Alberto J. Palmaa, Luis F. Capitn-Vallveyb,aECsens., Department of Electronics and Computer Technology, Campus Fuentenueva, Faculty of Sciences, University of Granada, E-18071 Granada, SpainbECsens., Department of Analytical Chemistry, Campus Fuentenueva, Faculty of Sciences, University of Granada, E-18071 Granada, Spainarti cle i nfoArticle history:Received 8 October 2010Received in revised form 13 April 2011Accepted 17 April 2011Available online 27 April 2011Keywords:Mobile phonePortable instrumentationImagingDisposable optical sensorPotassium determinationabstractIn this work, a mobile phone platform for portable chemical analysis is presented. This platform is basedon the use of the built-in camera for capturing the image of a single-use colorimetric chemical sensor,while a custom-developed software application processes this image for obtaining its characteristic H(hue) value, which is related to analyte concentration. This software application is optimized for mobilephone usage, thus preserving battery life and targeting reduced computation time through a customizedimage processing scheme including a modied monodimensional edge detection algorithm. Meanwhile,theinuenceofphysicalandchemicalfactorshasbeencharacterized, withresultsshowingthatthepresented platform provides accurate results even when variations on distance from phone to sensor,image focusing, or image centering are induced. In the same way, factors such as indicator concentrationandmembranethicknesshavebeenshowntohavenegligibleeffectsontheobtainedHvalues. Thecalibration and testing procedures have shown that the presented platformis able to provide a detectionlimit of 3.1105Min a range of 3.11050.1Mwith a relative standard deviation for inter-membranereproducibility lower than 1.6% for potassium concentration determination in solution. 2011 Elsevier B.V. All rights reserved.1. IntroductionAs mobile phone capabilities and processing power areincreased, the number of applications and the size of the associ-ated market have steadily grown. The current typical mobile phoneincludeshigh-resolutiondigitalcamera, usuallyabove3Mpixelswith autofocus and digital zoom, dedicated low-power high-performance processor [1], with running frequencies up to 1GHz,and sophisticated operating systems [2,3], whichusually offer mul-titasking, Java support, options for installingandrunningexternallydevelopedapplications, etc. Thus, consumershavenowintheirhands a tool as powerful as some low-cost personal computers ornetbooks, but for the inherent limitations of a hand-held device,basically screen and keyboard size and capabilities.Inthiscontext, mobilephoneshavebeenusedassupportinghardware in different applications in chemical analysis, but theircapabilities may be exploited further by taking full advantage ofboth the hardware and software associated to mobile phones, so itis possibletotransformthemintovaluableinstruments. Most of theanalytical uses rely on the concept of electronic and mobile health(e-health or m-health) or simply telemedicine, referred to the useof technology and mobile devices to improve the availability andCorresponding author.E-mail address: lcapitan@ugr.es (L.F. Capitn-Vallvey).quality of health care. Usually, an analytical module is connectedto a phone that collects the health information data and commu-nicates it to a health network platform [48]. Their most commonuse is simply as a device for imaging and transmissionof pictures toa central server, which translates latent information into chemicaldata.Therefore, mobile phones have been used to provide a way toeasily collect, transmit, and organize data along with different ana-lytical procedures, as it is the case in lateral ow immunoassays.In that assays, an immunochromatographic step is applied throughimmobilizing the antibodies and/or oligonucleotides at predeter-mined locations on the membrane (capture zones) that permit toreadvisuallyorbyimagingdevices, thatis, bymeansofoneormore optically readable lines on a test strip. Cocaine and its majormetabolite, benzoylecgonine, can be determined by imaging on aphonecamerathelateral owimmunoassaysthattransmittheimage to a central computer server, where a quantitative ratiomet-ric pixel density analysis (QRPDA) is performed [9]. Other similarprocedures have been patented based on the same concept, withthe phone acting as a capturing device that does not compute theresult but just send data to a server [10,11]. Another portable andquantitative assays are based on microuidic devices, as the paper-baseddeviceproposedforthesimultaneousanalysisofglucoseandproteininurine, alongwithacellularphonethatacquiresimages and transmits digital information to an off-site laboratory,wherethedataareanalyzedandtheresultsoftheanalysisare0925-4005/$ see front matter 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.snb.2011.04.045A. Garca et al. / Sensors and Actuators B 156 (2011) 350359 351returned [12]. A further advance is the 3D microuidic paper ana-lytical devices (PADs) fabricated by layering paper patterned intohydrophilicchannelsandhydrophobicwallsandtapepatternedwith holes that connect channels in different layers of paper, withapplications in telemedicine through the help of a phone cameraas in the prior case [13]. A low-cost system for sensing bad-smell,oriented to the monitoring of living environments, was describedbased on the well known gas detector tubes in which the readingat a certain time of the length of the discoloration layer permits theestimation of gas concentration. The combination of up to six gasdetector tubes with a mobile phone to acquire and transmit imagesandanexternal computerallowstocalculatethediscolorationspeed through brightness changes and, thus, the measurement ofthe gas distribution [14,15]. A last example of the use of mobilephones for imagingandtransmittingis avisiblesensor arraysystemproposed for simultaneous multiple SNP (single nucleotide poly-morphisms) genotyping based on alkaline phosphatase-mediatedprecipitation of coloured chemical substrates. The recording of theimage by a mobile phone equipped with a digital camera permitsthe quantication in a desktop computer [16].AdifferentapproachispresentedbyFilippiniandIqbal[17],whomake use of spectral ngerprinting(computer screenphotoas-sistedtechnique CSPT) for chemical sensing applications basedonastandard mobile phone, where the screen is programmed as a lightsource in combination with the phones camera, also transmittingto a remote site for data evaluation. The use of a mobile phone forsignal processing is proposed in some cases but not connected toimage data. As an example, the inclusion of a fuel cell sensor forbreathethanolinamobilephonepermitstomeasurethebloodethanol content [18].In this paper, a more complete use of a mobile phone as an ana-lytical tool is presented. Concretely, for this proof of concept studywe use a potassium sensor previously studied by us [19,20]. Thisoptical sensing membrane for potassiumis based on ion-exchangeequilibrium between an aqueous solution containing the analyteandaplasticizedPVCsensingmembranecontainingionophorethatdrivesthereactionandtheH+-selectivechromoionophore.The more analyte enters into the bulk membrane, the more chro-moionophore is deprotonated and then greater is the hue changeof the membrane from the acidic form, blue for the lipophilic Nilebluechromoionophoreusedinthismembrane, tothemagentacolour of the basic formof the chromoionophore. The ion activitiesratio aH+/aK+in aqueous phase is related through an equilibriumconstant with the experimental parameter, a relative optical sig-nal, whichcanbebuiltfromabsorbance[21], uorescence[22],reectance [23], transectance [24], refractive index [25], or colourcoordinateascouldbeoneRGBchannel [26], givingrisetoasigmoid-shaperesponsefunction. Previously, wehavedemon-strated that the use of colour, namely through hue-oriented colourspace, as is the HSV, offers several advantages as analytical param-eter, becauseitrepresentsthecolourinformationinonesingleparameter, the coordinate H(hue). In case of bitonal optical sensorsthat produce a colour change and hence a Hchange by reaction, theuse of this coordinate yields a substantial improvement inprecisionand is better suited for quantication [20].In this framework of image analysis as analytical tool, this paperexaminesthepossibilitiesandlimitationsofamobilephoneasacompleteanalytical instrument, alongwithsingle-useopticalsensors, for imaging, extracting the analytical parameter and cal-culating the analyte concentration. As it will be described in thefollowing, a custom image processing application has been devel-oped for mobile phone usage. This paper is organized as follows: inSection2themembranepreparationandthesoftwareplatformdevelopmentaredescribed, indicatingthemeasurementproce-dureandthealgorithmsforimageprocessingandcoordinateHcalculation. After that, in Section 3, the software application is rsttested with low resolution images preloaded in the mobile phone.After that, the whole measurement procedure is implemented inthemobilephone, usingthebuilt-incameraforimageacquisi-tion and then processing these images for computing the H valuesand thus analyte concentration. This allows characterizing the pre-sented platform attending to both physical and chemical factors.Later, the calibration and validation procedures are described and,nally the main conclusions are summarized.2. Experimental2.1. Instrument and softwareWhilethereareotheroptionsfordevelopingsoftwareappli-cationsformobilephones, suchasplatform-specicPhytonorC++ [27] implementations, JavaME has been selected for a num-berofreasons:itisanopensourcelanguage, withavarietyoffree development environments [28], simulators and other toolsbeingavailable;developmentofJava-basedapplicationsisplat-formindependent, providingthat anadequate Java Virtual Machineisavailableforthetargeteddevice, whichisthecaseformostcomputer operating systems, mobile phones andPDAs, anda grow-ing number of devices such as household appliances, automotivedevices, etc. Thus, Java-based applications can be used on almostanyplatform, orevencanbeeasilymigratedif anyspecicordevice-dependent API (application programming interface) has tobeused. Ontheotherhand, whilethemarketsharesofiPhonedevices and Android-based platforms are steadily growing, Nokiaprovides a variety of devices in the sector and is still the mobilephone market leader. Moreover, Nokia provides a handful of devel-opment resources for its Symbian-based devices, thus allowing theeasy generation of applications for this type of devices.Inthisway, thedevicesusedasanalyticalinstrumentswerethe mobile phones Nokia E65, Nokia 6110 and Nokia N73 (Espoo,Finland), which run Symbian S60 operating system. The rst twowere only used for the processing of preloaded images for appli-cationdebuggingandtesting, whiletheN73wasthemaintestplatformforthecompletemeasurementprocess, imagecaptur-ing and subsequent concentration determination. The integratedcameras of the phones were used to acquire images fromthe sens-ing membranes, consisting on CMOS sensors ranging from 2.0 to3.2Mpixels with autofocus function in the case of the N73 model.JavaME was selected for the development of the portable analy-sisplatform, withNetbeansalongNokiasS60PlatformSDKforJava as development platform. Apart from the mentioned mobilephones, thephoneemulatorintheSDKwasalsousedforsoft-ware debugging and testing, while Matlab software was used forthe development and tuning of the image processing techniquesdescribed below. In case of images not captured by the phone cam-era, a commercial ScanMaker i900 scanner (Microtek, Taiwan) wasusedforacquisitionanddigitalization, witha300dpiresolutionand 24-to-48 bits of colour.2.2. Reagents and membrane preparationThe potassiumstandardsolutions were preparedby exactweighingofanalytical reagentgradedryKCl anddissolutioninwater. All the standard solutions and samples were buffered usingpH 9.0 Tris(hydroxymethyl)aminomethane 1M buffer.The chemicals used to prepare the potassium sensitive mem-branes weredibenzo18-crown-6-ether (DB18C6) as ionophore,potassiumtetrakis (4-chlorophenyl)borate (TCPB) as lipophilicanionic salt, 1,2-benzo-7-(diethylamino)-3-(octadecanoylimino)phenoxazine (liphophilized Nile blue) as chromoionophore, highmolecular weight polyvinyl chloride (PVC) as membrane polymer,352 A. Garca et al. / Sensors and Actuators B 156 (2011) 350359o-nitro-phenyloctylether (NPOE) as plasticizer andtetrahydro-furan (THF) as solvent. All the reagents were obtained fromSigmaAldrich(Madrid, Spain). Sheets of Mylar-typepolyester(Goodfellow, Cambridge, UK) wereusedas plasticsupport forpreparing the membranes. Reverse-osmosis quality water type III(Milli-RO12 plus Milli-QstationfromMillipore) was used through-out.Themembraneswerepreparedfromcocktailscomposedby26.0mg of PVC, 63.0mg of NPOE, 0.8mg of DB18C6, 1.3mgof liphophilizedNileblueand1.1mgof TCPB, all in1.0mLoffreshlydistilledTHF. Sensingmembranes were preparedover14mm40mm0.5mmsheetsof polyester withaspincoat-ing technique (WS-400Ez-6NPP-Lite Single Spin Processor, LaurellTech., USA) using 20L of the cocktail. The red violet round-shapemembrane has a diameter of 12mm.2.3. Measurement procedureThe procedure of use of the membranes was the same in all thecases. Firstofall, themembraneswereactivatedbyintroducingin0.01MHClsolutionfor3min. Onceactivated, themembranewas introduced and equilibrated without shaking in the standardor sample solution for other 3min. Next, the sensing membranewas removed fromthe solution and placed in the holder, obtainingthe image with the mobile phone as indicated in the next section.Theconcentrationofpotassiumtestedfordifferentpurposesrangedbetween107and0.1Minallcases, bufferedwithpH9Tris at a nal buffer concentration of 0.02M. In the case of the realwater samples, they were prepared taking 49mL of water sampleand 1mL of pH 9 Tris 1M buffer solution.Themembraneimages wereobtainedusingboththecom-mercial scanner mentioned above, and the built-in camera of themobile phone at the highest possible resolution, setting automaticwhite balance and automatic congurations for any other cameraparameters. All the images were saved in JPEG (joint photographicexperts group) format, as it is the default format in the tested Nokiamobile phones. The images acquired with the phone cameras wereobtained inside an optical PBL Photo studio light tent 30

, illumi-nated with three daylight (D65) 600Wlamps that were distributedaround the tent as shown in Fig. 1. This assures a homogeneousillumination inside the tent. D65 is a commonly used standard illu-minant dened by the International Commission on Illumination(CIE). The sensing membranes were introduced after equilibrationwith standards or samples in a homemade support made of whitereective material. Thus, the membranes were placedina xedver-tical positioninsidethetent inorder tomaintainaconstant distancebetween the camera and the sensing membranes. The measuredilluminance at the plane of the sensor in the experimental setupwas2600lumen/m2. ThecamerawasxedinthedesiredplaceinsidethetentusingtweezersandthedistancefromthesensorFig. 1. Schematic view of the experimental setup for the membrane image acqui-sitionwithmobilephonecameras. Illuminationlamps, membraneandphonepositioning are shown.to the front of the phone was measured using a calibrated ruler. Inthis way, it is assured that the conditions for imaging are alwaysthe same.2.4. Mobile phone software developmentInthefollowing, theimageprocessingdevelopedforphoneapplicationisdescribed, aswell astheimplementationdetails,while further sections will discuss the obtained results.2.4.1. Image processingAs it has been mentioned in Section 1, tone (H coordinate in theHSV colour space) is the image characteristic that will be related tothe analyte concentration. While the RGB colour space is the mostwidelyutilizedinchemicalanalysis[2933], forthepurposeofconcentration determination the HSVspace, obtained froma trans-formation of the RGB coordinates, is better suited [20] and thus willbe used, specically the H(hue) component. Concretely, HSVis eas-ily obtained from normalized RGB coordinates, whose maximumandminimumvaluesaredenotedasmaxandmin, respectively,just applying the following non-lineal transformations:H =0 if max = min60oG Bmax minifmax = RandG B60oG Bmax min +360oifmax = RandG > B60oB Rmax min +120oifmax = G60oB Rmax min +240oifmax = B(1)S =0 if max = minmax minmax +minif V 0.5max min2 (max +min)if V > 0.5(2)V = max (3)It must be noted, from Eqs. (1)(3), that these transformationsareindependentoftheimageorthewayithasbeenacquired.Moreover, while RGB components are related to the contributionof each primary colour to each pixel, the meaning of HSV coordi-nates [34] is quite different and thus the Hvalue, which is an angle,contains most of the colour information. Because of this, the Hcom-ponent [19,20] has been used for concentration determination incolorimetric analysis.Thegeneral schemeoftheimplementedprocessingrequirestorst transformthesensorimage, usuallyrepresentedintheRGB space, to HSV coordinates; after that, an image histogram isobtained, which will determine the membrane characteristic H asthe maximum of this histogram, which corresponds to the modeof the H distribution. This H value will then be related to concen-tration, as described below. Fig. 2 shows an example of histogramon the H component for a membrane image transformed fromRGBto HSV with no prior processing. It is evident that this histogramallows the determination of a precise dominant value; however,contributions from background pixels, edges, etc., are still presentand may be removed for better H determination, which requiressome image preprocessing. Although this processing is quite sim-ple in its fundamentals, its computational complexity may be notas simple, even more as complex or large images have to be con-sidered. Whilethisisnot aproblemforoff-lineprocessingonanexternal computer, thelimitationsof aportable, hand-held,battery-operated mobile phone have to been taken into account.A. Garca et al. / Sensors and Actuators B 156 (2011) 350359 353Fig. 2. Example of H histogram for unprocessed membrane images.Inthisway, reducingthecomputational costofimageprocess-ing is crucial, not only in terms of computation time but also inorder to preserve energy consumption and thus battery life. This ismade possible by the application of border or edge detection tech-niques, whose aim is to differentiate the image areas containinguseful information from those that do not contribute to establisha proper value for the H coordinate. In this way, the portion of theimage to be analyzed as described above is reduced, thus limitingthe image processing to be carried out. While a wide range of edgedetection techniques are available [34], the selected technique hasto be light enough in order to not suppose a larger computationalcost than this saved by its application.Edgedetectiontechniques areusuallybasedontheuseof masks,whichallowsearchingfor suddenchanges of acertainimageparameter, i.e., a border or edge, through the denition of gradi-entsoverthisparameter. Thus, differentmasks, combinedwiththe mathematical operations associated to the mask application,denedifferent typeof edgestobefound. However, themereapplication of the mask does not sufce and usually further pro-cessing is required, with the application of directional masks forimproving the results, which is referred as thresholding. For thedevelopment of this work, three border detection methods havebeen considered: differential detection [34], Prewitt operator [35],and Sobel operator [36]. The rst method relies on the applicationof four 33 masks for obtaining image derivatives at each pixelon the vertical and horizontal directions, as well as on both diago-nals, with a further lineal combination of these derivative images.Thus, the computational costs associatedtothis methodmake it notadequate for its use ina mobile phone. Onthe other hand, the appli-cation of the Prewitt operator requires only the use of two of theprevious masks, vertical and horizontal, later averaging the result-ing images and performing a histogram extension for improvingcontrast. Althoughthismethodimpliesareductionincomputa-Fig. 3. Use of V coordinates for delimiting membrane borders. Membrane image in V coordinates (center) and V values vs pixel position for two perpendicular lines passingthrough the image center (bottom and left) showing inner edges determination with magenta lines.354 A. Garca et al. / Sensors and Actuators B 156 (2011) 350359tional costs when compared to the differential detection, both ofthem separate the membrane from the background but do not iso-late the membrane border from the membrane body itself. On theother hand, the Sobel operator [36] comprises two masks, for verti-cal and horizontal edge detection, respectively, which do computethe rst-order derivative combined with a Gaussian smoothing l-ter. This lter allows reducing noise effects during edge detection.While the computational complexity of this method is comparableto that of the Prewitt operator, the obtained results are more ade-quate for the proposedmobile phone chemical analysis application.However, acompletebidimensionalprocessingisstillcomputa-tionally too intense, so an ad hoc monodimensional adaptation oftheSobel edgedetectionhasbeendevelopedformobilephoneimplementation.For the adaptation of the Sobel edge detection, the rst issue tosolve is the selection of the adequate space for image processing.While edge detection is usually carried out over a grey-scale image,thetransformationfromRGBtogrey-scalewouldrequireextracomputation. Thus, inthisapplication, theV(value)coordinate,which corresponds to image brightness and is equivalent to a grey-scale representation, has been selected for edge detection. In thisway, V values are available once the image has been transformedfromRGB to HSV, so no extra computation is required. Meanwhile,the edge detection will be started looking for the central vectorsof the membrane image in both directions, since they provide theareas with most signicative information within, as illustrated inFig. 3. As it can be seen, apart fromlimiting the process to computa-tions over vectors (pixel columns or rows in the image), the centralmembrane section is clearly delimited by two valleys associated tothe membrane border in the V coordinate representation.However, giventhe characteristics of the membrane images andthe variable width of the border, it is necessary to rene the edgedetection, since the shape of the valleys in Fig. 3 do not allow foran immediate decision on edge positioning. Concretely, it wouldbe possible to use a monodimensional 3-element Sobel operator,[ 2 0 2 ], but this may not properly process the central vectorsif the membrane border is too thick, with a correspondingly widevalley in Fig. 3. Thus, a modied monodimensional Sobel operatoris proposed:ModiedSobel operator = [ 2 0 0 0 2 ] (4)that compromises 5elements andallows theidenticationofthicker or out of focus edges.Both central vectors are processed applying this modied Sobeloperator, andtheaverageandvariancevaluesarecomputedforthe resulting vectors. In this way, the edges of the area where theH coordinate will be determined are dened as the pixels with VcoordinatevaluesnotseparatedfromtheaverageVvaluemorethan the variance. This is illustrated in Fig. 4, where an exampleof original vertical central vector (in red) and the processed vector(in blue) are shown, also displaying the average V value and thevariance limits.Once the edge of the useful membrane area has been dened forboth vertical and horizontal directions using the modied monodi-mensional Sobel edge detection, the histogramof the H coordinateis performed. ThemembraneH-valueis thencomputedas themodeof this histogram.2.4.2. Platform implementationHavinginmindbothsoftwareandhardwareconsiderations,Java 2 Micro Edition (J2ME) [37] was selected for the developmentof the portable analysis platform, while Nokia S60 Symbian-baseddevices were targeted. Thus, Netbeans [38] along with Nokias S60PlatformSDK for Java [28] was chosen as development platform. Inthis way, the developedapplications have beentestedonthe deviceemulator includedinthe SDKas well as ona variety of Nokia mobileFig. 4. Original V central vertical vector (in red), processed vector with the modi-ed Sobel operator (in blue), average processed V value (green) and variance limits(magenta). Arrows indicateedgelimits (For interpretationof thereferences tocolourin this gure legend, the reader is referred to the web version of this article.).phones available in the laboratory, including several models of theE and N series, mentioned above, and some older models. Whilethese older models do not include the same versions of the oper-ating system and the same processors than their E- and N-seriescounterparts, they were used as a benchmark for the portabilityof the developed applications, since they were able to run everyapplication tested providing identical results to those provided bythe emulators and the E- and N-series devices.Fig. 5 shows several typical screen captures of the developedsoftwareapplicationfortwosensorsequilibratedwithsolutionswith different potassium concentration, illustrating its main fea-tures. Thesensor imagecaptures aredisplayedinthecentralpictures and, in the right, the result screens are showed, compris-ing the cut out images with the proposed modied Sobel algorithmand the calculated H coordinates. Comparing both sensor images(before and after edge detection), it can be observed that the edgedetection procedure is a proper working tool for this task.3. Results and discussion3.1. Preliminary resultsThe rst step for testing the signal processing implementationinthemobilephonewastouseimagesofthesamesingle-usepotassiumsensoracquiredwithanexternalscannerworkingintransmission mode, thus loading them into the phone memory forprocessing by the software application. Therefore, at this point, thecamera of the phone was not used to acquire these images in orderto separate the inuence of the camera in the nal analysis. More-over, in this rst version of the software, the image size was limitedto 150150pixels.Inthisway, asetofvedifferentmembraneswasprepared,equilibratedwithsolutions of knownpotassiumconcentrationintherangefrom107Mto101M. Thesemembranes werescanned and processed to obtain lowresolution images. Theresulting images were processedona Nokia E65inorder toobtain the corresponding normalizedH values. These results aredepictedinFig. 6, wheresymbolsshowaveragedvaluesfortheve replicas and error bars are two times the standard deviations.Fig. 6showsthetypical sigmoid-trendresponseof thiskindofionophorechromoionophore sensors, as mentioned above. In fact,a sigmoidfunctionhas beenproposedtomodel these averageddataA. Garca et al. / Sensors and Actuators B 156 (2011) 350359 355Fig. 5. Screen captures of the developed software application showing: (left) the presentation screens, (center) the screens with acquired sensor images and, (right) nalscreens with processed images and H coordinate results, for two different potassium concentrations.for calibration purposes:H(x) = A2+A1A21 +e(xA0)/A3(5)where A0A3are tting parameters and x represents the decimallogarithmof the analyte concentration in molar units. In Fig. 6, a tbetween experimental H values and this model is represented by asolid line.Fig. 6. H coordinatesvs potassium concentration. Symbols with error bars showH values obtained from the mobile phone with preloaded images obtained fromscannerworkingintransmissionmode, andsolidlinecorrespondstoasigmoidmodelization.Therefore, the calibration function is given by,log[K(I)] = A0+A3 log

A1A2H A21

(6)Eq. (6) has been used to recover the concentration values, whichhavebeencomparedtothetheoretical dataintherangefrom105to 101M, where there was more H value variation. In thisrange, themaximumdeviationbetweentheoretical andexperi-mental data is 4% of the logarithm of the potassium concentration,which is a very acceptable agreement for this rst software versionwith low resolution images.3.2. Image acquisition by the phone cameraOnce the software application has been shown to produce reli-ableresults,thecompletemobilephoneplatformistested.Asamobilephoneisnotinitiallyintendedasananalyticaltool, itisnecessary to demonstrate its suitability for obtaining valid analyti-cal information from an image of the sensor acquired directly withthe phone camera. Thus, the most relevant issues that inuence thephysical measuring conditions, such as sensor to phone distance,focusing, centering of the sensor in the image, and hand-held oper-ation of the phone, are addressed. Additionally, chemical factorssuch as membrane thickness and composition are also taken intoaccount.356 A. Garca et al. / Sensors and Actuators B 156 (2011) 350359Fig. 7. Hcoordinates as a function of the membrane-phone distance for three differ-ent sensors. Symbols show experimental H values obtained from the mobile phoneand solid lines correspond to the averaged H value for each membrane. The RSD forthe results obtained for each sensing membrane are 0.75%, 0.68% and 0.76%.3.2.1. Inuence of physical factorsAs described in Section 2.3, the phone and the membrane wereplacedinanoptical tent inordertomaintaintheilluminationconditionsasclosertodaylight,controlledandhomogeneousaspossible, since daylight shouldbe the typical illuminationfor insitumeasurement conditions. This experimental setup allowed hold-ing the phone and the membrane at controlled distances. Thus, 11different positions were tested, from 5cm to 55cm, in 5cm inter-vals, and for each distance 3 images were captured for 3 differentmembranesequilibratedwith5105, 5104and5103Mpotassium, respectively. Theseimageswereprocessedwiththesoftware application on the same phone used for image acquisi-tion, andthecorrespondingHvalueswereobtained, whicharedisplayed in Fig. 7 as a function of the distance from the phone tothemembraneandforthesetofimagesobtainedwiththeN73phone. Itcanbeobservedthatthisdistancedoesnotaffecttheobtained H value in the studied range. The values of relative stan-dard deviation (RSD), calculated for each membrane using all thedistance data, were 0.75%, 0.68%and 0.76%, respectively. This resultwasexpectedbecausetheHvalueobtainedforthemembranesimages is independent of the area of the image containing mem-brane information, whichdepends onthe distance fromthe camerato the sensor. When this distance is increased, the size of the mem-braneimagewithinthetotalimageareadecreases, sothereareless pixels containing useful information. However, this reductionin the number of signicant pixels available to the software appli-cation has not lead to changes of the H value. Numerically, there isa reduction in the number of useful pixels of a factor of 100 whenmoving the phone away from the membrane, from 5cm to 50cmdistances. This shows that the software is robust enough to over-come this effect andthat any inducedvariationis negligible. It mustbe noted that H values in Fig. 7 do not match with the calibrationcurve of Fig. 6, due to the different origin of images. Now, imagesare directly acquiredby the mobile phone camera, thus inreectionmode, while preloaded images used for software development andderivation of data in Fig. 6 were obtained by a scanner in transmis-sion mode. Below, a complete calibration curve will be calculatedusing mobile phone camera images.Other factor that is related to the distance from the camera tothe membrane is the focusing of the image. It must be noted thatusual phones includenot sosophisticatedoptics, someof themevenwith xed focus, so proper focusing is not possible for distancesshorter than a predened length. Even for phones with camerasfeaturing autofocus, as it is the case for the N73 device, focus rangestarts at 10cm, while the other phones available for this study withxed focus cameras feature typical minimumfocus distances up to20cm. Thus, data in Fig. 7 for distances below 10cm correspond tounfocusedimages, so it must be notedthat the software applicationis extracting correct H values (0.588 for 5cm and 0.593 for 10cm)even in these conditions. It is worth noting that this is an importantcontribution of the proposed software application, since distancesaround 10cm are a natural election for hand-held operation of thephone, while they also maximize the size of the membrane withinthe capturedimage, i.e., the number of pixels containing membraneinformation. In this way, even if the image may not be properlyfocused at this distance, depending on the camera characteristics,the obtained H value will be still valid.In order to test how hand-held operation may affect the per-formanceoftheanalytical platform, otherfactor, inadditiontodistance, that should be studied when the image is acquired is cen-tering of the sensor image within the whole picture. In this context,displaying the whole circular sensor image in the picture should bethe ideal option, but hand-held operation of the mobile phone mayresult in pictures containing only portions of the membrane image.For this purpose, four membranes were captured in different posi-tions withinthe picture, starting by placing the sensor image onthepicture center and then gradually displacing the membrane imageso this occupies less area of the total image, as illustrated in Fig. 8.The results show that the H values obtained when the membraneis not centered are the same as when it is centered. Concretely, therelative standard deviation of the results obtained when the mem-brane image is displaced from the center down to occupying only40% of the picture is 0.95%. The reason for this behaviour is that theHvalue was computed through its mode, so even when most of thepixels are almost white, corresponding to the image background,they do not affect the obtained result.Finally, some images were capturedby hand-holding the phone,thus completing the study on the inuence of physical factors onthe measurement process. Concretely, three membranes in threedifferent solutions were processed in this way, and the results interms of RSD are 0.78%, 0.77% and 0.92% for the same consideredconcentrations.3.2.2. Inuence of chemical factorsRegarding the chemical factors, the inuence of variationinsen-sor composition, that could be a simile of lot-to-lot variation, wasalso considered through the inuence of membrane thickness andindicator concentration. These deliberate changes were performedduring the preparation of the membrane in order to evaluate howrobust the analytical parameter H computed by the mobile phoneis, bymeasuringitsinsensitivitytoalterationsintheoperatingconditions.Theinuenceof thickness was studiedbypreparingaset of sens-ing membranes using different volumes of the same cocktail, from10 to 25L, but maintaining a constant spot diameter of 12mm.Thus, the calculated thickness varies between 1.4 and 3.5m. Theresultsfromthreedifferentmembranesofeachthickness, equi-librated with three differentK(I) concentrations each, show thatthe H value for each K(I) concentration is almost independent ofthe membrane thickness, and the results do not show bias whenthe membrane thickness increases. Considering the precision in Husing all the membranes of different thickness for the three K(I)solutions, the RSD values are 2.68, 1.34, and 1.33%.Regarding the amount of indicator in the membrane, the chro-moionophore in this case, several membranes were prepared withtwodifferent cocktails, onewiththecompositionindicatedinSection 2.2, and another with half of the concentration in all con-stituents previously used. Similarly to previous results, there is nobiaswiththeincreaseinindicatorconcentration, sincethehuevalue is maintained. The RSDs calculated from all the membranesA. Garca et al. / Sensors and Actuators B 156 (2011) 350359 357Fig. 8. Centered (left) and decentered (right) membrane images showing H valuesandpercentageof useful pixels for four different sensors. A: HC =0.58; HD =0.58; 59%;B: HC =0.57; HD =0.57; 51%; C: HC =0.58; HD =0.57; 43%; D: HC =0.58; HD =0.68; 26%.equilibrated in the same potassium solution before testing were2.45, 1.60, and 1.69% for each K(I) solution. Consequently, the smallvariations in membrane thickness and indicator concentration canbeminimizedbytheuseoftheHparameteracquiredwiththemobile phone platform.The independence of the results with respect to both physicaland chemical factors reinforce the fact that it is possible to use amobile phone as a platform for obtaining analytical information,even when obviously this type of device is not designed for thispurpose. This is possible because of the proposed custom softwareapplication, which is able to cope with all the factors studied, whileit features a computationally-reducedimage processingthat allowspreserving batterylife andreducedprocessing times, still providingreliable analytical information.Fig. 9. Calibration of the potassium concentration with experimental (symbols) Hvalues obtainedwiththeproposedmobilephoneplatform. Solidlineshows sigmoid-function model.3.3. Calibration with the mobile phoneAs we have prior demonstrated [20], the direct use of Hparame-ter for calibration enables simplifying the experimental procedure,avoiding the use of calibrants for the protonated and deprotonatedforms of the chromoionophore, needed for the calculation of thedegree of deprotonation, 1, the usual analytical parameter forthis type of sensors [21]. Thus, the use of H directly simplies theprocedure, only requiring one measurement, and avoids the mea-surement before and after equilibration with calibrants.For calibration purposes, 12 potassiumstandard solutionsbufferedatpH9.0withTris0.02Mbuffersolutionandconcen-trations ranging from 107to 0.1M were prepared and used forreacting with three different potassium sensing membranes. Afterthat, threepicturesof eachmembranewereobtainedwiththemobilephoneplacedintheoptical tentatadistanceof10cm,according to the procedure described above. The resulting Hvaluescomputedby the phone andplottedagainst logarithmof potassiumconcentration are shown in Fig. 9 as a sigmoid curve, as it is usualfor this class of sensing membranes.The analytical range can be estimated by tting a straight linetotheregionofmaximumslopeofthesigmoidforobtainingalinearfunction, H=1.17(0.04) +0.16(0.01)[K(I)]. Thisrelation-shiphas R2=0.9763andalimit of detectionof 3.16104M,obtainedfromtheintersectionofthelinearcalibrationfunctionand the linearized background at lowconcentration. This detectionlimit is higher than the calculated from absorbance measurements(1.25105M) [20].However, the dynamic range of the procedure can beincreasedbyttingresultstothesigmoidrelationshipgiveninTable 1Analytical specications of the mobile phone platformfor potassiumdetermination.Fitting parameter of the Sigmoid-function modelling ValuesA02.79893A10.58442A20.84908A30.30909R20.9947Analytical parameters ValuesDetection limit 3.1105MRange 3.11050.1MInter-membrane precision RSD5105M 1.13%5104M 1.51%5103M 1.53%358 A. Garca et al. / Sensors and Actuators B 156 (2011) 350359Table 2Validation test results. Phone vs atomic absorption spectroscopy potassium concentration values.Phone AAS p-valueMineral water 4.5105M0.41053.83105M0.011050.096Tap water 5105M1.01056.22105M0.011050.479Spring water 6.3105M0.81055.29105M0.051050.096Eq. (6), where the adjusting coefcients t well (A0=2.798;A1=0.584; A2=0.849; A3=0.309), also showing a good correlation(R2=0.9947). The use of this function for the Hparameter improvesthe analytical range and the detection limit of the procedure. Withthe use of the IUPAC criteria (3s) for the lower and upper limits,a wider measurement range (3.11051.0101M) is achieved,thus obtaining a detection limit, 3.1105M, similar to the cal-culated from absorbance measurements, 1.25105M, and lowerthan the obtained using the linear regression.Theprecisionof theprocedureusingdifferent single-usesensorswas studied at three potassium concentrations, namely 5105,5104and5103M, using10replicas. Theobtainedresults,also in terms of RSD, were 1.13%, 1.51% and 1.53% for the solutionsmentioned above, respectively. The results of inter-membrane pre-cisionareacceptableif twofactsareconsidered:rst, theRSDvalues are much lower than in the case of use of analytical param-etersbasedonintensitymeasurement(absorbance, reectance,transectance, uorescence, RGBcoordinate) and, second, thedevice used as analytical instrument is a very simple camera inte-grated in a mobile phone that digitalizes the membrane images.Moreover, the membrane precision with the developed procedureis comparable to those from other methods, such as the use of ascanner and a computer for processing the images, yielding from0.14 to 0.75%, or the use of a scanner and custom-developed soft-ware, providing from 0.33 to 0.92%. Thus, the results provided bythe proposed mobile phone platformare slightly worse than previ-ously mentioned data, but clearly better than other usual methods,such as diode array spectrophotometers, which provides RSD from3.3to7.7%[19], transectancemeasurements, from2.4to7.9%[24], or a scanner using RGB colour space, 0.45 and 1.80% [26]. Asasummary, inTable1theanalyticalparametersforthemobilephone-based procedure are shown.3.4. Validation of the mobile phone-based procedureTo test the usefulness and validate the procedure, it was appliedto waters of diverse provenance (spring, mineral, tap) fromAndalu-sia, Spain, for measuringpotassiumcontent. Hvalues for thesensors were determined, and the concentration of potassium wascalculated directly using Has the analytical parameter. The methodwas validated by comparison with results from atomic absorptionspectroscopy. Theresults obtainedusingthedescribedproce-durewerecomparabletotheestablishedtechniques, asshownin Table 2, which includes the p-value in order to accept the nullhypothesis that conrms that there are not signicant differencesbetween results. This conrms the capability of the mobile phone-based procedure to analyze potassium using a single-use sensor,and the possibility of using optical sensors for relating hue changesto the analyte concentration.4. ConclusionsThis paper has demonstratedthecapabilityof amobilephoneforprocessing and obtaining analytical information, also presenting itto the user. It has beenmade possible throughthe development of acustom software application, which includes reduced and adaptedimage processing in order to save computation time and preservebattery life while still providing accurate analytical measurements.Such a platform, as it has been shown, allows a non-experienceduser to carry out the analysis of potassium in water, without therequirement of any chemical knowledge or laboratory equipment.Moreover, theinherentportabilityandbattery-basedoperationmakepossibleinsituanalysis withaportableinstrument, providinganalytical information and analyte concentration.The presented analytical platformhas been characterizedregardingbothphysical andchemical factors. Thus, ithasbeenprovedthat physical factors suchas adistancetothesensor,image centering, or even focusing, do not affect the computed Hvaluesandanalyteconcentrations. Thishasbeenmadepossibledue to the custom-developed image processing, which features amonodimensional edgedetectionprocedure, andtheuseofthehuecomponentforobtainingtheanalyteconcentration. Inthisway, the presentedplatformis well-suitedfor hand-heldoperation,enabling in situ chemical analysis with a tool as simple as a mobilephone. Meanwhile, thestudyofchemicalfactorshasconrmedthat membrane thickness and indicator concentration have no sig-nicant inuence on the potassium concentration determination.Thus, theperformanceofthepresentedplatformiscomparabletomostusualproceduresbasedonimageprocessingonoff-linecomputers.AcknowledgementsWe acknowledge nancial support fromthe Ministerio de Cien-ciaeInnovacin, DireccinGeneral deInvestigacinyGestindel PlanNacional de I+D+i (Spain) (Projects CTQ2009-14428-C02-01 and CTQ2009-14428-C02-02); and the Junta de Andaluca(Proyecto de Excelencia P08-FQM-3535). These projects were par-tially supported by European Regional Development Funds (ERDF).References[1] ARM, Cortex-A8 Technical Reference Manual, http://infocenter.arm.com/help/topic/com.arm.doc.ddi0344k/index.html, Cambridge, UK, 2010.[2] Android. Developers Guide, 2010.[3] Apple iOS Developer Center, Technical Documentation, 2010.[4] Y.T. Cheung, Test paper-type healthcare mobile phone, China Patent201388225 (2010).[5] Z. Dong, D. Pang, J. Li, Sports pose/blood sugar-monitoring mobile phone andon-line expert diagnosis system, China Patent 201467228 (2010).[6] Y. Meng, M. Liu, C. Ma, G. Qu, L. Du, Real-time blood sugar monitoring sys-temand method, blood sugar detection apparatus, and mobile phone terminal,China Patent 101647703 (2010).[7] Y. Zhao, Method for determining blood glucose content with mobile phone,China Patent 101637389 (2010).[8] E.H. Lee, M.S. Kim, C.H. Kim, Clinical chemistry measuring system capable ofbeing applied to portable analysis system by using mobile phone, and methodtherefor, KR Patent 2007006304 (2007).[9] B.A. Cadle, K.C. Rasmus, J.A. Varela, L.S. Leverich, C.E. ONeill, R.K. Bachtell, D.C.Cooper, Cellular phone-based image acquisition and quantitative ratiometricmethod for detecting cocaine and benzoylecgonine for biological and forensicapplications, Subst. Abuse: Res. Treat. 4 (2010) 21.[10] F. Levin, Optical analytical systemandmethods for thedeterminationofsubstances in liquids using photographic images, WO Patent 102008026803(2009).[11] D. Bremnes, Methodforsemi-quantitativeorquantitativeanaly. of assaysincludingimmunoassaysinwhichmethodcomprisesassigningabarcodecoded text segment on a test strip, WO Patent 2009054729 (2009).[12] A.W. Martinez, S.T. Phillips, E. Carrilho, S.W. Thomas, H. Sindi, G.M. Whitesides,Simple telemedicine for developing regions: camera phones and paper-basedmicrouidicdevicesforreal-time, off-sitediagnosis, Anal. Chem. 80(2008)3699.A. Garca et al. / Sensors and Actuators B 156 (2011) 350359 359[13] A.W. Martinez, S.T. Phillips, G.M. Whitesides, Three-dimensional microuidicdevices fabricated in layered paper and tape, Proc. Natl. Acad. Sci. U.S.A. 105(2008) 19606.[14] H.P. Ninh, Y. Tanaka, T. Nakamoto, K. Hamada, A bad-smell sensing networkusinggasdetectortubesandmobilephonecameras, Sens. ActuatorsB125(2007) 138.[15] T. Nakamoto, T. Ikeda, H. Hirano, T. Arimoto, Humidity compensation by neuralnetworkforbad-smellsensingsystemusinggasdetectortubeandbuilt-incamera, IEEE Sens. (2009) 281.[16] Y. Michikawa, T. Suga, Y. Ohtsuka, I. Matsumoto, A. Ishikawa, K. Ishikawa, M.Iwakawa, T. Imai, Visible genotype sensor array, Sensors 8 (2008) 2722.[17] Z. Iqbal, D. Filippini, Spectral ngerprinting on a standard mobile phone, J. Sens.(2010).[18] W. Koehn, Cellphonewithbreathalcoholanalyzer, USPatent2009325639(2009).[19] L.F. Capitan-Vallvey, M.D. Fernandez Ramos, M. Al Natsheh, Adisposablesingle-use optical sensor for potassium determination based on neutral ionophore,Sens. Actuators B 88 (2003) 217.[20] K. Cantrell, M.M. Erenas, I. Orbe-Paya, L.F. Capitan-Vallvey, Useof thehueparameter of the hue, saturation, value color space as a quantitative analyticalparameter for bitonal optical sensors, Anal. Chem. 82 (2010) 531.[21] E. Bakker, P. Bhlmann, E. Pretsch, Carrier-based ion-selective electrodes andbulk optodes. 1. General characteristics, Chem. Rev. 97 (1997) 3083.[22] A. Ceresa, Y. Quin, S. Peper, E. Bakker, Mechanistic insights intothedevelopmentof optical chloride sensor based on the mercuracarborand-3 ionophore, Anal.Chem. 75 (2003) 133.[23] R.H. Ng, K.M. Sparks, B.E. Statland, Colorimetric determination of potassium inplasma and serum by reectance photometry with a dry-chemistry reagent,Clin. Chem. 38 (1992) 1371.[24] I.d. Orbe-Paya, M.M. Erenas, L.F. Capitan-Vallvey, Potassium disposable opticalsensor based on transectance measurements, Sens. Actuators B 127 (2007)586.[25] D. Freiner, R.E. Kunz, D. Citterio, U.E. Spichiger, M.T. Gale, Integrated opticalsensors based on refractometry of ion-selective membranes, Sens. Actuators B29 (1995) 277.[26] A. Lapresta-Fernandez, L.F. Capitan-Vallvey, Scanometric potassium determi-nation with ionophore-based disposable sensors, Sens. Actuators B 134 (2008)694.[27] E. Spence, Rapid Mobile Enterprise Development for Symbian OS: An introduc-tion to OPL Application Design and Programming, Wiley, New York, 2005.[28] Nokia. S60Platform: IntroductoryGuide. http://sw.nokia.com/id/fc17242f-9bb2-4509-b12c-e6b8206085b/S60 Platform Introductory Guide v1 6 en.pdf.[29] Y. Shishkin, S.G. Dmitrienko, O.M. Medvedeva, S.A. Badakova, L.N. Pyatkova,Use of a scanner and digital image-processing software for the quanticationof adsorbed substances, J. Anal. Chem. 59 (2004) 102.[30] N.C. Birch, D.F. Stickle, Example of use of a desktop scanner for data acquisitionin a colorimetric assay, Clin. Chim. Acta 333 (2003) 95.[31] A. Abbaspour, A. Khajehzadeh, A. Ghaffarinejad, Asimpleandcost-effectivemethod, as an appropriate alternative for visible spectrophotometry: develop-ment of a dopamine biosensor, Analyst 134 (2009) 1692.[32] M Kompany-Zareh, M. Mansourian, F. Ravaee, Simple method for colorimetricspot-testquantitativeanalysisofFe(III)usingacomputercontrolledhand-scanner, Anal.Chim.Acta 471 (2002) 97.[33] N. Maleki, A. Safavi, F. Sedaghatpour, Single-step calibration, prediction andreal samples data acquisition for articial neural network using a CCD camera,Talanta 64 (2004) 830.[34] S.E Umbaugh, Computer Imaging: Digital Image Analysis and Processing, CRCPress, 2005.[35] W. Dong, Z. Shisheng, Color image recognition method based on the prewittoperator, in: Proc. 2008 IEEE International Conference on Computer Scienceand Software Engineering, 2010, p. 170.[36] N. Kazakova, M. Margala, N.G. Durdle, Sobel edge detection processor for a real-time volume rendering system, in: Proc. 2004 IEEE International Symposiumon Circuits and Systems (ISCAS04), vol. 2, 2004, p. 913.[37] Sun Microsystems. Java platform. http://java.sun.com.[38] Sun Microsystems. N etBeans IDE 6.9.1. http://www.netbeans.org.BiographiesAntonio Garca received the MASc degree in Electronic Engineering, the MSc degreein Physics (majoring in electronics), and the PhD degree in Electronic Engineeringfrom the University of Granada (Granada, Spain) in 1995, 1997, and 1999, respec-tively. He was an associate professor at the Department of Computer Engineering,Universidad Autnoma de Madrid (Madrid, Spain) before joining the DepartmentofElectronicsandComputerTechnologyattheUniversityofGranada, wherehecurrently serves as a professor. He has authored more than 80 technical papers ininternational journals and conferences and serves regularly as a reviewer for sev-eralIEEEandIEEjournals. HeisalsopartoftheProgramCommitteeforseveralinternationalconferencesonprogrammablelogic. Hiscurrentresearchinterestsinclude the combination of digital and analog programmable technologies for smartinstrumentation, IP protection of VLSI and FPGA-based systems and low-power andhigh-performance VLSI signal processing systems.M.M. Erenas was born in Granada (Spain) in 1981. He received the MSc degree andthe PhD degree in Analytical Chemistry from the University of Granada (Granada,Spain) in 2004 and 2010, respectively. Currently he works as laboratory managerat ROVI Laboratories. His current research interest is the development of sensingphases for their use as chemical sensors in the determination of inorganic speciesin different matrices and multivariate calibration methods.Eugenio D. Marinetto was born in Granada, Spain, on March 24, 1986. After obtain-ing his Master Degree in Telecommunications Engineering fromGranada Universityin2009, he starteda Master Degree inBiomedical Engineering. Currently he is work-ing as a predoctoral researcher at the Imaging Laboratory of the Gregorio Mara nnHospital in Madrid. His main interest is imaging processing, mainly for MRI and CTsystems and computer guided surgery.Carlos A. Abad received his BSc in Telecommunication Engineering in 2010 fromUniversity of Granada. He is currently pursuing his MSc in Electrical Engineering atColumbia University.IgnaciodeOrbe-PayaisanassociateprofessorintheDepartmentofAnalyticalChemistry at the University of Granada (Spain). His main areas of research are thedevelopment of the sensing phases for their use as chemical sensors in the deter-minationofinorganicionsinseveralmatrices;multivariatecalibrationmethodsfor the quality control of pharmaceutical products and development of analyticalmethodology using solid-phase spectrometry.Alberto J. Palma was born in 1968 in Granada (Spain). He received the BS and MScdegrees in Physics (Electronics) in 1991 and the PhD degree in 1995 from the Uni-versityofGranada, Granada, Spain. HeiscurrentlyanassociateprofessorattheUniversity of Granada. Since 1992, he has been working on trapping of carriers indifferent electronic devices (diodes and MOS transistors) including characterizationand simulation of capture cross sections, random telegraph noise, and generation-recombinationnoise indevices. From2000, his current researchinterest is the studyof the application of MOS devices as radiation sensors and the electronic instru-mentation design directed to portable, low cost electronic systems in the elds ofchemical and physical sensors.Luis F. Capitn-Vallvey is a full professor of Analytical Chemistry at the Universityof Granada, received his BSc in Chemistry (1973) and PhD in Chemistry (1976) fromthe Faculty of Sciences, University of Granada (Spain). In 1983, he founded the SolidPhase Spectrometry Group (GSB) and in 2000, together with Prof. Palma, the inter-disciplinary group ECsens, which includes Chemists, Physicists and Electrical andComputer Engineers at the University of Granada. His current research interests arethe design, development and fabrication of sensors and portable instrumentationfor environmental, health and food analysis and monitoring.