U

Ma

b

a

ARRAA

KHHBPI

1

vaisorstTrfi[p

tsslcmte

0h

Sensors and Actuators B 174 (2012) 10– 17

Contents lists available at SciVerse ScienceDirect

Sensors and Actuators B: Chemical

journa l h o me pa ge: www.elsev ier .com/ locate /snb

se of digital reflection devices for measurement using hue-based optical sensors

.M. Erenasa, K. Cantrellb, J. Ballesta-Clavera, I. de Orbe-Payáa, L.F. Capitán-Vallveya,∗

ECSens, Solid Phase Spectrometry Research Group, Department of Analytical Chemistry, Campus Fuentenueva, University of Granada, E-18071 Granada, SpainDepartment of Chemistry, The University of Portland, Portland, OR 97203, USA

r t i c l e i n f o

rticle history:eceived 6 March 2012eceived in revised form 26 July 2012ccepted 27 July 2012vailable online 13 August 2012

a b s t r a c t

In this paper the hue, or H component of the HSV (hue, saturation, value) color space, is used as analyticalparameter for bitonal optical sensors. This parameter is characterized by its robustness to variations in theimaging device, illuminant, and sensor membrane. It has a higher precision than either traditional opticalparameters, such as absorbance or transflectance, or other color parameters such as RGB coordinates.Here the hue is calculated from digital images obtained in reflection mode (digital camera) rather than

eywords: coordinateSV color spaceitonal optical sensorsotassium sensor

images in transmission mode (digital scanner) as demonstrated in previous work. The stability of the Hparameter, makes it possible to greatly simplify the procedure used for the optical sensors as well as touse ubiquitous digital cameras to acquire the image.

© 2012 Elsevier B.V. All rights reserved.

maging

. Introduction

In a previous paper we have discussed the use of a qualitativeariable to quantify analyte concentration in a sample by means of

sensor material which changes its properties through reaction ornteraction with the analyte if that qualitative variable can be repre-ented by a continuous function [1]. We are concerned with the usef color – qualities that are present or represented in visual expe-iences coming from the human perception of the electromagneticpectrum – as analytical parameter, which is typically used in quali-ative analysis resulting from separation or identification reactions.he specification of a color by a color space, a mathematical modelepresenting color as tuples of color components, permits their useor quantitative purposes using data from various imaging devicesncluding hand-held [2] and desktop scanners [3–5], CCD arrays6], video cameras [7,8], digital photographic cameras [9], mobilehone cameras [5,10–12] and digital color analyzers [13–15].

The HSV color space uses the classifying descriptors hue, satura-ion and value (brightness) [16]. An important feature of this colorpace is the representation of the cognitive color information in aingle parameter, the hue component or H parameter [17]. Hue isargely independent of variations in color intensity such as thoseoming from changes in colored reagent concentration or sensor

embrane thickness. The use of this qualitative signal as a quanti-

ative analytical signal is possible if the sensor involves a chemicalquilibrium with at least two distinct colors which is affected by the

∗ Corresponding author.E-mail address: lcapitan@ugr.es (L.F. Capitán-Vallvey).

925-4005/$ – see front matter © 2012 Elsevier B.V. All rights reserved.ttp://dx.doi.org/10.1016/j.snb.2012.07.100

analyte. Changes in the concentration of the analyte cause the sen-sor membrane to change its color and thus the hue (calculated fromthe digital image) also changes. The hue provides a 4- to 10-foldimprovement in terms of precision (as measured by the coefficientof variation, CV) [1] when compared to other optical signals suchas absorbance [18].

Different optical techniques have been used in quantitativesingle-use sensors, however the most widely utilized is reflectancespectrometry. Given the structure of different types of opticalsingle-use sensors [19], they are usually opaque, which is whythe typical measurement method is by diffuse reflectance spec-troscopy, with measurements at one or two wavelengths [20]. Thistechnique involves measuring the intensity that results from theinteraction of radiation with the reaction volume of the carrierwhen irradiation with electromagnetic radiation due to absorption,transmission and scattering by materials occurs. Generally, diffusereflection coefficients of the measuring area of a disposable sensorand reference are measured, and from these measurements, theKubelka–Munk function or its variations are calculated.

Measurement can be performed in two ways [21]. After reac-tion on the surface of the analytical element, the reflectance of thegenerated absorbing species in the sampling surface is determined.This approach has been the one adopted so far for monolayer fiber-impregnated single use sensors, and the simple diffuse reflectancedescribed by the phenomenological theory of Kubelka–Munk isused for quantitative analysis. In the second approach, the sam-

ple is applied to the top surface and the color is measured bymonitoring reflectance in the reverse surface of the element. Thisapproach is common with multilayer single-use sensors and makesuse of a combination of both diffuse and specular reflection in

and Ac

tcioct

mtoifi

satpncmsm

ccliamstapbbcf

2

2

wwp

bo(l(3tMoAoM

2ooo

M.M. Erenas et al. / Sensors

he different layers, reagent layer and spreading layer mainly, thatompose the analytical element [22]. In this last case the mathemat-cal treatment is more complicated involving diffuse reflectance,rdinary transmittance, and specular reflectance of hemispheri-ally distributed incident radiation, being typically described byhe Williams–Clapper transformation [23,24].

Among the monolayer single-use sensors that use membranesade up of a single layer, those with a homogenous layer are

he simplest. These materials achieve a homogenous distributionf components through impregnation of an adsorbent (fiber-mpregnated systems), deposition on a non-absorbent support,abrication of a film using a film-forming material, or chemicalmmobilization on a support or adsorption on a membrane.

Examples of reflectance-based sensors include a calcium sen-or fabricated by electrostatic immobilization of calcichrome atn anionic polymer membrane [25], a mercury sensor that useshe intensely colored reaction product with tyrodine adsorbed onolycaproamide membrane [26], and a test paper for the determi-ation of aluminum based on a cellulose paper impregnated withhromeazurol S [27]. In addition to different multilayer sensorsainly used for the clinical laboratory [24,28], many lateral-flow

ensors, typically based on immunoassay, have been proposedainly for drugs and pesticides [29–31].We are interested in quantitative, disposable sensors that use

hromaticity characteristics other than the typically used RGBhannel intensity or a combination of such parameters [32] as ana-ytical parameters. Our focus is on the hue, H value, obtained frommages of the sensors acquired by different imaging devices suchs scanners and digital photographic cameras working in reflectionode. This approach is of particular interest for membrane optical

ensors, especially disposable materials, due to the improvement inheir precision. Here the properties of the H coordinate are studieds an analytical parameter using a disposable K(I) sensor developedreviously [18]. Two models, one based on reflection and anotherased on transflexion, are used to demonstrate that the H value cane used as a parameter to follow the behavior of the sensors and toompare results with other possible analytical parameters derivedrom images of these sensor membranes.

. Experimental

.1. Reagents and membrane preparation

The potassium standard solutions were prepared by exacteighing of analytical reagent grade dry KCl and dissolving it inater. All the standard solutions and samples were buffered usingH 9.0 Tris(hydroxymethyl)aminomethane 1 M buffer.

The chemicals used to prepare the potassium sensitive mem-ranes were high molecular weight polyvinyl chloride (PVC),-nitro-phenyloctylether (NPOE) as plasticizer, tetrahydrofuranTHF) and the ionophore dibenzo18-crown-6-ether (DB18C6), theipophilic anionic salt potassium tetrakis (4-chlorophenyl)borateTCPB), and the chromoionophore 1,2-benzo-7-(diethylamino)--(octadecanoylimino)phenoxazine (liphophilized Nile Blue). Allhe reagents were purchased from Sigma–Aldrich (Química S.A.,

adrid, Spain). To prepare potassium sensitive membranes sheetsf Mylar-type polyester (Goodfellow, Cambridge, UK) were used.ll chemicals used were of analytical-reagent grade. Reverse-smosis quality water type III (Milli-RO 12 plus Milli-Q station fromillipore) was used throughout.The membranes were prepared from cocktails composed by

6.0 mg of PVC, 63.0 mg of NPOE, 0.8 mg of DB18C6, 1.3 mgf liphophilized Nile Blue and 1.1 mg of TCPB, all in 1.0 mLf freshly distilled THF. Sensing membranes were preparedver 14 mm × 40 mm × 0.5 mm sheets of polyester with a spin

tuators B 174 (2012) 10– 17 11

coating technique (WS-400Ez-6NPP-Lite Single Spin Processor,Laurell Tech., USA) using two different volumes of cocktail (15 �Land 20 �L) to prepare membranes with different thickness. The redviolet round-shape membrane has a diameter of 12 mm.

2.2. Measurement procedure

In all cases the membranes were activated by introductioninto 10−2 M HCl solution for 3 min. Once activated, the membranewas introduced and equilibrated without shaking in the standardor sample solution for 3 min. Next, the sensing membrane wasremoved from the solution, and the image was obtained as indi-cated in the next section. Additionally some membranes weretreated and measured after each equilibration with 10−2 M HCl,potassium standards, and 10−2 M NaOH in order to calculate therobustness coefficient Rg.

The concentration of potassium tested for different purposesranged between 10−9 and 0.1 M buffered with pH 9 Tris at a finalbuffer concentration of 0.02 M. The natural water samples wereprepared by combining 49 mL of water sample and 1 mL of pH 9Tris buffer.

2.3. Imaging devices, instruments and software

Different devices were used to image the potassium mem-branes, namely a scanner, a single-lens reflex digital photographiccamera, a compact camera, and a mobile phone camera. Thescanner was a CanoScan 8800f flatbed scanner (Canon USA, Inc.,Lake Success, NY, USA) with a white LED light source and 48-bit color detection, working in reflection mode and saving theimages in TIFF (Tagged Image File Format) format. The SLR cam-era used was a Canon EOS 500D (Digital Rebel T1i) digital SLRcamera with an EF-S 60 mm f/2.8 macro-lens. The detector is a15.50-megapixel single-plate CMOS sensor with RGB primary colorfilters. Images were taken in JPG format. The compact camerawas a Sony DSC T90 (Sony Corporation, Minato, Tokyo, Japan)with 10.5-megapixel resolution and equipped with Carl Zeiss andVario-Tessar 3.5–4.6/6.18–24.7 optic and 4× optical zoom and setat maximum resolution. The images obtained with the compactcamera were in JPG format and 10.2-megapixel resolution. The inte-grated camera of a HTC Diamond (High Tech Computer Corporation,Taoyuan, Taiwan) mobile phone was also used to acquire imagesfrom the sensors. This camera was equipped with a 3.2-megapixelCMOS sensor, F2.8 lens, and an autofocus function. Images from thephone camera were stored in JPG format. An Ocean Optics USB2000UV-Vis spectrophotometer (Ocean Optics, Dunedin, FL) was useto obtain the spectrum of a light tent used for all digital cameraimages.

All the images acquired with cameras were obtained inside of a30′′ optical tent, PBL photo studio light tent, placed in a sun-lit roomand illuminated with 3 daylight 600 W lamps (Steve Kaeser Photo-graphic Lighting and Accessories, Ventura, CA, USA). To assure thatthe conditions when the images are taken are always the same, thesensing membranes, after equilibration with standards or samples,were introduced in a homemade support made of white reflec-tive material. The sensors were placed in a fixed vertical positioninside the tent while the camera was on a tripod to maintain aconstant distance between the camera and sensors as shown inFig. 1.

The images were processed with a set of scripts and functionsdeveloped in Matlab r2007b (MathWorks, Natick, MA, USA). Sta-

tistical calculations were performed with Statgraphics softwarepackage (Manugistics Inc. and Statistical Graphics Corporation,USA, 1992), and Microsoft Excel (Microsoft Corp., Redmond, WA,USA) was used for general calculations.

12 M.M. Erenas et al. / Sensors and Ac

Fp

2

mbccco

2

t(cbBmaa

2

mviitsbt(wh

2

ibiTB

[HC ] X − X

ig. 1. Experimental setup for the membrane image acquisition with the mobilehone camera.

.4. Matlab processing

All digital images were processed using the Matlab program-ing environment. The basic steps image processing include: (1)

alancing (color balance and normalize the red, green, and bluehannels), (2) masking (determine which pixels in the image areolored), (3) slicing (separate regions in the image that containolored pixels), and (4) analyzing (calculate the most frequentlyccurring value of the hue in each colored region of the image).

.4.1. BalancingIn the balancing step the red, green, and blue channels are

reated independently. First, the most frequently occurring valuemode) in the entire image is determined separately for each colorhannel. Because most of the pixels in the image are from theackground, this value represents the “white” value for R, G, or

respectively. The RGB value of each pixel is then divided by theode from the appropriate color channel. The result of this step is

color-balanced image in which the RGB values range from 0 to 1nd white pixels have a value of 1 in each color channel.

.4.2. MaskingIn the masking step the maximum and minimum value is deter-

ined for each pixel in the image by comparing the R, G, and Balues of the pixel. The difference between the maximum and min-mum value is used to determine how “colored” a pixel within themage is. If the difference in the color channels is greater than ahreshold value, the pixel is classified as colored. If the difference ismaller than the threshold the pixel is classified as gray (includinglack and white pixels as well). The threshold used for all images inhis study was 30% of the maximum difference in the entire imagethe most “colored” pixel). The result of this step is a logical maskith the same dimensions as the original image in which each pixelas a value of 1 if it is colored and 0 if it is gray.

.4.3. SlicingIn the slicing step the logical mask produced in the previous step

s searched for both rows and columns where there is a transition

etween lines (either horizontal rows or vertical columns) contain-

ng no colored pixels and lines containing at least one colored pixel.hese transitional rows and columns then define a colored block.locks that do not meet a minimum size requirement (20,000 pixels

tuators B 174 (2012) 10– 17

in most images used in this study but adjusted depending on imageresolution) are discarded. Each colored block is then extracted as anindividual picture. The result of this step is a set of cropped rectan-gular sub-images each containing the data for an individual circularmembrane.

2.4.4. AnalyzingIn the analyzing step the sub-images produced in the previ-

ous step are first rebalanced. This process is identical to the colorbalance and normalization done in the beginning of the imageprocessing except that only the white pixels in the sub-image areused to calculate the value for each color channel’s mode. The orig-inal values of R, G, and B are then re-normalized using these newvalues. (Note: Although the normalized values calculated in thebeginning of the image processing are replaced during this step,balancing is performed the first time in order for the masking stepto correctly identify colored pixels. In images with poor white bal-ance, these pixels can be misclassified as colored if the balancingis not done before masking.). The result of this step is the mostfrequently occurring value for RGB and Hue for each sub-imagecontaining a membrane.

3. Rationale

Previous work demonstrated that the H parameter is suitablefor monitoring the behavior of a bitonal optical sensor used forpotassium determination [1,33,34]. This sensor was based on anionophore–chromoionophore mechanism, were an ion-exchangeis performed inside the membrane, analyte ions entering in the sen-sor membrane result in protons leaving the membrane and causinga color change (Eq. (1) barred species means membrane phase):

L̄+ HC+ + R− + K+Kexch←→KL+ + C̄ + R− + H+ (1)

The membrane is composed of an ionophore, L, which forms acomplex with the analyte and retains it in the sensor; a chro-moionophore, C, which is an indicator that changes its color bydeprotonation; and a highly lipophilic anion, R−, for ion-exchangepurposes. All materials are contained in a plasticized PVC matrix.Based on this complexation reaction as more of the analyte goesinto the bulk membrane, the chromoionophore is deprotonatedand the hue of membrane changes from the acidic form, HC+, tothe basic form, C.

The ratio of the ion activities in the aqueous phase, aK+/aH+ ,is related to the equilibrium constant Kexch and the experimentalparameter ̨ (see Eq. (3)) through the sigmoidal-shape responsefunction as show in Eq. (2):

aK+

aH+= 1

Kexch

(˛

1 − ˛

)CR − (1 − ˛)CC

(CL − (CR − (1 − ˛)CC))(2)

where CL, CC and CR are the analytical concentrations ofionophore (DB18C6), chromoionophore (liphophilized Nile Blue),and lipophilic anion (TCPB) respectively [35]. The relationshipshown in Eq. (2) assumes that the stoichiometric factor for the ana-lyte:ionophore complex and the charge of the analyte are both 1.The extent of the ion-exchange process, i.e., the recognition process,is typically observed by the normalized parameter 1 − ̨ obtainedfrom absorbance, fluorescence or other optical parameters accord-ing to Eq. (3), where X is the value for a given measurement, andXHC+ and XC are the values coming from fully protonated and depro-tonated chromoionophore species.

+

1 − ̨ =

[C]0= C

XHC+ − XC(3)

The 1 − ̨ value can be calculated from the experimentally deter-mined H coordinate using the three (H, HHC+ and HC) values. The

and Actuators B 174 (2012) 10– 17 13

sv1ibm

mostatti

A

Irmttttwab

R

Aai

R

Birfemg

R

w

cUw8e

X

Y

Z

Tpatt

M.M. Erenas et al. / Sensors

hape of the sigmoid indicated by Eq. (2) is the same using H (whicharies between HHC+ and HC of 0.56 and 0.82 respectively) instead

− ̨ (which varies between 0 and 1), in a similar way as observedn a previous work [1]. In the rest of the study we replace 1 − ˛y H in order to simplify the calculation and requiring only oneeasurement per sensor instead three.The colorimetric behavior of the sensor membrane can be

odeled using the absorptivity spectra of the acid and base formsf the chromoionophore (εa(�) and εb(�) respectively), the CIEtandard D65 illuminant (P0(�)), and the color matching func-ions for the CIE XYZ color space (Sx(�), SY(�), and SZ(�)). First, thebsorbance of the membrane at a particular wavelength, Amembrane,hat contains a fraction of the total concentration (CC) in the depro-onated form ( ̨ in Eq. (2)) can be calculated from Eq. (4), where bs the thickness of the membrane.

membrane = εabcc(1 − ˛) + εbbcc ̨ (4)

n this study two well established models for the fraction of lighteflected back to the imaging device were considered. In the firstodel the background is assumed to be a perfect reflector in which

he incident photons must travel through the membrane, strikehe background, and then travel back to the imaging device. In thisransmission model the fraction of light returning is the same as theransmittance where the average pathlength (b in Eq. (4)) is some-here between the film thicknesses and twice the film thickness,

nd the fraction of light returning to the imaging device (R) is giveny Eq. (5).

= 10−Amembrane (5)

lternatively, the fraction of reflected light (R) can be calculatedccording to Kubelka–Munk relationship given in Eq. (6) where ks the absorption coefficient and s is the scattering coefficient.

= 1 + k

s−

√k

s

(2 + k

s

)(6)

ecause k/s is directly proportional to absorbance, and the scatter-ng coefficient is considered to be independent of wavelength, theeflectance of the semi-transparent membrane can be estimatedrom Amembrane through the introduction of the adjustable param-ter ω, which is proportional to the ratio of the thickness of theembrane (b) and the effective scattering coefficient of the back-

round (s). This relationship is given in Eq. (7).

= 1 + Amembraneω −√

Amembraneω(2 + Amembraneω) (7)

here ω ≈ b/s.Thus the fraction of reflected light at a single wavelength can be

alculated using either Eq. (5) (transmission) or Eq. (7) (reflectance).sing both of the models to calculate R, the XYZ tristimulus valuesere then calculated by numerically integrating (from 360 nm to

30 nm in 5 nm increments via the trapezoid rule) the intensity ofach color channel according to Eqs. (8)–(10).

=∫

�

P0(��)RX (��)d (8)

=∫

�

P0(��)RY (��)d (9)

=∫

�

P0(��)RZ (��)d (10)

he XYZ values were then normalized by dividing each by the white

oint value, which was calculated by assuming that the light is notttenuated by the chromoionophore (XYZ is 0.9505, 1, 1.0890 forhe standard D65 illuminant). The sRGB tristimulus values werehen calculated from XYZ by linear matrix multiplication followed

Fig. 2. Comparison of experimental data and theoretical models.

by gamma correction. The hue was calculated from the RGB valuesvia Eq. (11) where R, G, and B are the RGB tristimulus values [36].

H = (G − B/maxchannel − minchannel) + 06

; if max = Ra

H = (B − R/maxchannel − minchannel) + 26

; if max = G

H = (R − G/maxchannel − minchannel) + 46

; if max = B

a if H is less than 0 then add 1 to H

(11)

Using this procedure it was possible to calculate the values of Husing ̨ values between 0 and 1 in increments of 0.01. Similarly,the associated potassium activity for the ̨ values were calculatedvia Eq. (2) assuming the experimental value of Kexch is 3.85 × 10−8

and that CL = CC = CR is 0.0236 M. Fig. 2 shows the values for the Hcalculated in this manner plotted as a function of the logarithmof potassium activity. Also displayed in Fig. 2 are the experimen-tally determined values for the H of the sensing membrane asdetermined by flatbed scanner (+ symbols) and reflex camera (Osymbols). A few things are evident in this figure. The calculated val-ues for H are close to the observed values despite several adjustableparameters which all affect the predicted response. The trans-mission model, R calculated via Eq. (5), gives a better fit to theexperimental data (standard error of the fit = 0.0186) than reflec-tion model (standard error 0.0295), but the differences are minorand may be attributed to these adjustable parameters such as thescattering coefficient or Kexch. In addition, the experimental valuesfor the scanner and reflex camera are very similar. The differences inthe experimental measurements from these devices are primarilyattributed to differences in lighting.

The effect of lighting conditions can be investigated by chang-ing the illuminant, represented as P0(��) in Eqs. (8)–(10). The D65illuminant is representative of daylight illumination at noon, E is anequal energy illuminant in which the spectral power is the same atall wavelengths, and A is the CIE standard illuminant for incandes-cent/tungsten light. In this study the spectrum of the light tent usedfor all cameras images was measured with a spectrophotometer.The light intensities were then converted to 5 nm equally spacedvalues between 360 and 830 nm. Fig. 3 shows the predicted valuesfor the H using these four light sources (D65, E, A and light tentspectrum) as a function of ˛; additionally, the white points of eachilluminant are given in Table 1. The interpretation of these results

fits with intuition; the color of the sensor membranes is not exactlythe same when they viewed under different types of illumination.The difference between an “ideal” source and daylight (the solidand dotted lines in Fig. 3) is negligible. But when viewed under the

14 M.M. Erenas et al. / Sensors and Actuators B 174 (2012) 10– 17

ialoto

4

htocts(

tF

R

wesasSipvi

4

c

TW

0.5

0.6

0.7

0.8

0.9

Compact camera

Reflex camera

Mobile phone camera

Scanner

Fig. 3. Variation of H value as a function of ̨ depending on illumination.

llumination of the light tent, the H of the indicator membrane at half deprotonation is different; it should appear more “blue” andess “magenta”. Thus modeling predicts that, at intermediate valuesf alpha, there will be a greater variability in the measured hue ofhe sensor membrane as the lighting conditions (or white balancef the image) change.

. Results and discussion

Once the relationship between H and potassium concentrationas been theoretically demonstrated using a model that fits withhe experimental data, we studied the influence of different factorsn the performance of the analytical parameter H. These studiesan be divided in three different groups: (a) dependence of H on theype of device used to image the membrane, (b) membrane factorsuch as thickness of the sensor and chromoionophore amount, andc) dependence on the illuminant used.

Additionally, it was necessary to define a coefficient to measurehe robustness of the H parameter as these conditions were varied.or this purpose the robustness coefficient, Rb, was defined as:

b = Maxavg − Minavg

spooled(12)

here Maxavg is the average H value for all membranes that werequilibrated to the solution that gave the largest H value (NaOHolution, chromoionophore fully deprotonated), and Minavg is theverage H value for all membranes in the solution that gave themallest value (HCl solution, chromoionophore fully protonated).pooled is the pooled standard deviation of all the membranes stud-ed in each case. A greater the Rb coefficient indicates a more robustarameter because it implies lower Spooled and/or higher range ofariation of the signal; thus larger values indicate a parameter thats better suited for quantification.

.1. Device dependence

The devices used were a scanner, a reflex digital photographicamera, a compact digital camera, and a mobile phone. With these

able 1hite points of various illuminants.

Illuminant X Y Z

D65 0.9505 1 1.0890E 1.0001 1 1.0003Light tent 0.9177 1 0.7869A 1.0985 1 0.3558

-2.0 -4.0 -6.0 -8.0 -10.0

Fig. 4. H dependence with potassium concentration for different imaging devices.

devices the sensor was tested using 11 different potassium solu-tions ranged from 10−9 M to 10−3 M. Fig. 4 shows the resultsobtained from different devices plotting the average of three repli-cates per device and solution. The H values obtained for eachsolution are comparable, showing that results do not depend on thedevice used to digitize the membrane. Also, the responses obtainedfrom each device are very similar and define the same sigmoidcurve.

If we consider the precision obtained for each potassium con-centration tested independently of the device used (Table 2), anaverage coefficient of variation CV of 0.76% was obtained consider-ing images coming from the scanner and the two cameras and anaverage CV of 2.0% if all devices are considered.

Furthermore, the robustness coefficient, Rg, was calculated forthe values of H parameter and other potential parameters such asR, G, and B channels as shown in Table 3. This coefficient was esti-mated using all of the data from the scanner and the two camerastogether as a single set because all points were similar and therewas a large number of replicates (99). The value using the H param-eter was 102 which is much greater than that obtained with the Gchannel, 24, which confirms that H is a parameter more suitableand robust than the others. In the case of mobile phone camera(Table 4) the Rg coefficient is higher for H, 32, with the exceptionof R channel, 38.

4.2. Membrane factors

The manufacture of the membranes introduces lot-to-lot vari-ation, and hence increases the uncertainty in the measurements.The two main differences between individual membranes are thethickness and the amount of chromoionophore in the membrane.To study this variability, a set of sensors with different thicknesseswas prepared with 15 �L (2.0 �m thickness) and 20 �L (2.7 �m)of cocktail, and another set was prepared with a reagent cocktaildiluted by a factor of 2 with THF.

Working with the same set of potassium solutions abovedescribed, the results obtained using sensors with the two differentthickness studied (198 sensors) in terms of average CV was 0.74%.Considering the Rg coefficient, the H parameter presents the largestvalue (55) compared to the next closest value for the R channel (8)as shown in Table 3. In the case of mobile phone camera, the Rg coef-ficient calculated for H is 23 compared to the value for R channel(7) as shown in Table 4. Therefore, the differences observed usingsensors with different thickness are negligible.

The precision of the measurements in terms of CV using sensorswith the two different amount of chromoionophore tested (198membranes) was 1.4%. Considering the Rg, again the value from H(25) is higher than the obtained for B channel (6) (Table 3). The

M.M. Erenas et al. / Sensors and Actuators B 174 (2012) 10– 17 15

Table 2H values for different devices.

log [K(I)] H value CV

Compact cameraa Reflex cameraa Scannera

−9.0 0.554 0.551 0.545 0.82%−8.0 0.549 0.550 0.545 0.46%−7.0 0.551 0.550 0.546 0.50%−6.3 0.549 0.551 0.547 0.42%−6.0 0.550 0.553 0.548 0.46%−5.3 0.568 0.561 0.554 1.24%−5.0 0.566 0.569 0.569 0.35%−4.3 0.595 0.600 0.602 0.55%−4.0 0.623 0.632 0.643 1.60%−3.3 0.718 0.729 0.734 1.14%−3.0 0.840 0.826 0.831 0.86%

Average CV 0.76%

a 3 replicates per solution.

Table 3Influence of different factors on the Rg coefficient.

Parameter Rg coefficient

K(I) testa Thicknessa Concentrationa Concentration and thicknessa White balanceb

R 17 8 5 5 16G 24 7 6 5 12B 20 7 6 5 12H 102 55 25 27 57

a Calculated from reflex camera, compact camera and scanner data as a whole.b Calculated from compact camera.

Table 4Influence of different factors on the Rg coefficient using a mobile phone camera.

Rg coefficient

K(I) test Thickness Concentration Concentration and thickness

R 38 7 10 64 32 2

37 22

vtH

fdpat((

4

iitid

(wsb

p

G 8 3

B 6 2

H 32 23

alue of Rg with the mobile phone was 37 for H (Table 4), betterhan the one for R channel (10). In conclusion, the tonal parameter

is robust to variations of chromoionophore amount.Finally, the Rg coefficient was calculated considering all data

rom this study (normal cocktail/20 �L, normal cocktail/15 �L andiluted cocktail/20 �L with a total of 297 membranes). The Harameter maintains its robustness in terms of Rg, 27 for camerasnd scanner (Table 3), and 37 for mobile phone (Table 4) as opposedo the RGB channels which were 5, 5, 5 for scanner and camerasTable 3) and 10, 4, 2 for the mobile phone camera respectivelyTable 4).

.3. Illuminant dependence

The illumination is a factor that affects the H parameter as shownn the rationale section. In order to calculate the extent of thisnfluence, images were taken with the compact camera in whichhe white balance setting was changed while maintaining constantllumination, resulting in images appearing as though lighted withifferent illuminants.

The white balances selected were sunlight (5300K), cloudy6000K), lamp (3000K), and fluorescent (4000K) and all these setsere tested over the eleven standard potassium solutions. Fig. 5

hows the image of the same sensor obtained using different whitealances.

Results obtained from this study, using 132 membranes, dis-lays that a changing of the white balance has a small influence on

Fig. 5. Membrane images acquired using different white balances: (A) sunlight(5300K), (B) cloudy (6000K), (C) lamp (3000K), (D) fluorescent (4000K).

16 M.M. Erenas et al. / Sensors and Actuators B 174 (2012) 10– 17

Table 5Determination of potassium in different types of matrices using AAS as a reference method.

Samples Phone AAS p-Value

Mineral water 1 4.5 × 10−5 ± 0.3 × 10−5 M 6.80 × 10−5 ± 0.03 × 10−5 M 0.153−5 −5 −5 −5

tctti1nf

4

tpti

tttl3i

htw

H

Taudlt(

5r

Fc

Mineral water 2 5.2 ×10 ± 0.2 ×10 MMineral water 3 5.8 × 10−5 ± 0.4 × 10−5 M

Mineral water 4 5.5 × 10−5 ± 0.7 × 10−5 M

he H value, as it is demonstrated in terms CV, the mean of which,onsidering all the white balances and the potassium concentra-ions studied, is 1.7%. Also the Rg coefficient demonstrates againhat H is a robust and reliable parameter. The Rg value with the hues 57 which is greater than any RGB channel, which gave values of6, 12, and 12, respectively (Table 3). This results show that illumi-ation for sensing membrane imaging is not a very critical factor

or H parameter acquisition.

.4. Calibration

Once the robustness of H parameter was demonstrated in reflec-ion mode, we studied the calibration function for potassium. 11otassium standard solutions ranging from 10−7 M to 10−1 M withhree replicates each were used, and the sensor membrane wasmaged with the scanner and the two digital cameras.

The shape of the dependence as shown in Fig. 6 is sigmoidal, asheory predicts. As usual in this type of sensors the central zone ofhe sigmoid can be linearized for calibration purposes and the equa-ion H = 1.1 (±0.1) + 0.16 (±0.03) [K(I)]; R2 = 0.998 was obtained. Theimit of detection calculated as usual [37] from this regression was.5 × 10−4 M, a result that is worse than that obtained working from

mages acquired in transmission mode (3.6 × 10−5 M) [1].However, the high precision of the measurements due to the

igh robustness of the parameter allows for the use of a Boltzmannype equation (Eq. (13)) to enlarge the dynamic range using thehole calibration set.

= a2 +a2 − a1

1 + exp(([K(I)] − a0)/a3)(13)

he constants of the Boltzmann type equation were a0 = −2.965;1 = 0.867; a2 = 0.572; a3 = 0.280. The R2 of the fit was 0.993 andsing the IUPAC criteria to calculate the detection limit (3 s), aynamic range of 5.5 × 10−6 M to 1.0 × 10−1 M was found. The

imit of detection found is similar to that obtained working withransmission images using the same Boltzmann type equation

4.5 × 10−6 M) [1].

The CV of 3 different potassium concentrations (5.0 × 10−5, × 10−4 and 5 × 10−3 M, 9 replicates each) were 0.42, 1.2 and 0.55%,espectively; similar values to those calculated using a scanner in

0.5

0.6

0.7

0.8

0.9

-8.0 -6.0 -4.0 -2.0 0.0

H

logK(I)

ig. 6. Calibration curve using an average of scanner, reflex camera and compactamera data.

6.02 ×10 ± 0.03 ×10 M 0.0511.51 × 10−5 ± 0.02 × 10−5 M 0.0774.97 × 10−5 ± 0.04 × 10−5 M 0.494

transmission mode (from 0.14% to 0.75%) [1] and much better thanusing transflectance as analytical parameter (from 2.4% to 7.9%)[38].

To check the quality of the information obtained from theimages acquired in reflection mode, different mineral water ofdiverse provenance (Spain) were analyzed for potassium con-centration using atomic absorption spectroscopy as a referencemethod. The results obtained are given in Table 5 and are com-parable using the p-value test.

5. Conclusions

The use of the H parameter from imaging devices working inreflection mode has been demonstrated to model and calibrate apotassium optical sensor. To model the sensor behavior, two dif-ferent modeling approximations were considered, one based ona transmission model and another based on a reflection model.Although both gave reasonable predictions, the transmission-basedmodel gave better agreement with the experimental data.

The use of H color coordinate, as opposed to other opticalparameters such as absorbance, reflectance or transflectance, per-mits a considerable simplification of the procedure for using thesensing membranes, requiring only one measurement of the cali-bration standards or samples. The use of this qualitative parametergives good results for the quantitative determination of analytes bymeans bitonal optical sensors. Additionally, this parameter gives ahigh robustness and independence from variables such as the typeof imaging device, illuminant, membrane thickness, and amount ofchromoionophore in the sensing membrane.

Comparing results obtained using H in reflection mode withthose obtained working in transmission mode, a very similar limitof detection was found using the whole concentration range stud-ied modeled with a Boltzmann-type function (5.5 × 10−6 M forreflection mode and 4.5 × 10−6 M for transmission mode), althoughit worsens if only the central section of the sigmoid is consid-ered for calibration purposes (3.5 × 10−4 M for reflection mode and3.6 × 10−5 M for transmission mode). Considering the precision, it isbetter to use transmission mode (0.14–0.75% in transmission modeand 0.42–1.2% in reflection mode), although the value obtainedfrom reflection is still good for a single use sensor. Much of thisprecision is attributed to the consistency in lighting and white bal-ance that can be obtained using a single optical scanner as opposedthe heterogeneity from using various cameras.

The H parameter obtained from reflectance images presentsmajor advantages in term of simplicity of the measurement andthe availability of the equipment necessary for analysis. The possi-bility of using digital cameras or mobile phone cameras paves theway for in situ analysis of real samples by non-trained personnelusing these ubiquitous devices.

Acknowledgements

We acknowledge financial support from the Ministerio de Cien-

cia e Innovación, Dirección General de Investigación y Gestión delPlan Nacional de I + D + i (Spain) Projects CTQ2009-14428-C02-01and CTQ2009-14428-C02-02; and the Junta de Andalucía (Proyectode Excelencia P08-FQM-3535 and FQM-5974). These projects were

and Ac

p(

R

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

the Solid Phase Spectrometry group (GSB) and in 2000, together with Prof. Palma

M.M. Erenas et al. / Sensors

artially supported by European Regional Development FundsERDF).

eferences

[1] K. Cantrell, M.M. Erenas, I. de Orbe-Paya, L.F. Capitán-Vallvey, Use of the hueparameter of the hue, saturation, value color space as a quantitative analyticalparameter for bitonal optical sensors, Analytical Chemistry 82 (2010) 531.

[2] M. Kompany-Zareh, M. Mansourian, F. Ravaee, Simple method for colorimetricspot-test quantitative analysis of Fe(III) using a computer controlled hand-scanner, Analytica Chimica Acta 471 (2002) 97.

[3] J. Gabrielson, M. Hart, A. Jarelov, I. Kuhn, D. McKenzie, R. Mollby, Evaluationof redox indicators and the use of digital scanners and spectrophotometer forquantification of microbial growth in microplates, Journal of MicrobiologicalMethods 50 (2002) 63.

[4] A. Lapresta-Fernandez, L.F. Capitán-Vallvey, Scanometric potassium determi-nation with ionophore-based disposable sensors, Sensors and Actuators B 134(2008) 694.

[5] S. Capel-Cuevas, M.P. Cuellar, I. de Orbe-Paya, M.C. Pegalajar, L.F. Capitán-Vallvey, Full-range optical pH sensor array based on neural networks,Microchemical Journal 97 (2011) 225.

[6] J.J. Lavigne, S. Savoy, M.B. Clevenger, J.E. Ritchie, B. McDoniel, S.J. Yoo, E.V.Anslyn, J.T. McDevitt, J.B. Shear, D. Neikirk, Solution-based analysis of multipleanalytes by a sensor array: toward the development of an electronic tongue,Journal of the American Chemical Society 120 (1998) 6429.

[7] K. Tohda, M. Gratzl, Micro-miniature autonomous optical sensor array for mon-itoring ions and metabolites 1: design, fabrication, and data analysis, AnalyticalSciences 22 (2006) 383.

[8] C. Di Natale, E. Martinelli, R. Paolesse, A. D’Amico, D. Filippini, I. Lundstroem,An artificial olfaction system based on the optical imaging of a large array ofchemical reporters, Sensors and Actuators B 142 (2009) 412.

[9] V.V. Apyari, S.G. Dmitrienko, Using a digital camera and computer dataprocessing for the determination of organic substances with diazotizedpolyurethane foams, Journal of Analytical Chemistry 63 (2008) 530.

10] H.P. Ninh, Y. Tanaka, T. Nakamoto, K. Hamada, A bad-smell sensing networkusing gas detector tubes and mobile phone cameras, Sensors and Actuators B125 (2007) 138.

11] A.W. Martinez, S.T. Phillips, E. Carrilho, S.W. Thomas, H. Sindi III, G.M.Whitesides, Simple telemedicine for developing regions: camera phones andpaper-based microfluidic devices for real-time, off-site diagnosis, AnalyticalChemistry 80 (2008) 3699.

12] A. Garcia, M.M. Erenas, E.D. Marinetto, C.A. Abad, I. de Orbe-Paya, A.J. Palma, L.F.Capitán-Vallvey, Mobile phone platform as portable chemical analyzer, Sensorsand Actuators B 156 (2011) 350.

13] E. Hirayama, T. Sugiyama, H. Hisamoto, K. Suzuki, Visual and colorimetriclithium ion sensing based on digital color analysis, Analytical Chemistry 72(2000) 465.

14] K. Suzuki, E. Hirayama, T. Sugiyama, K. Yasuda, H. Okabe, D. Citterio, Ionophore-based lithium ion film optode realizing multiple color variations utilizing digitalcolor analysis, Analytical Chemistry 74 (2002) 5766.

15] Y. Suzuki, K. Suzuki, Optical sensors for ions and protein based on digital coloranalysis, in: Frontier in Chemical Sensors, Springer Series on Chemical Sensorsand Biosensors, vol. 3, 2005, p. 343.

16] S.D. Cotton, Colour, Colour Spaces and the Human Visual System, School ofComputer Science, University of Birmingham, Birmingham, UK, 1996.

17] G. Wyszecki, W.S. Stiles, Color Science: Concepts and Methods, QuantitativeData and Formulae, Wiley Classics Library, Denver, USA, 2000.

18] L.F. Capitán-Vallvey, M.D. Fernandez Ramos, M. Al-Natsheh, A disposablesingle-use optical sensor for potassium determination based on neutralionophore, Sensors and Actuators B 88 (2003) 217.

19] Yu. Zolotov, V.M. Ivanov, V.G. Amelin, Test methods for extra-laboratory anal-ysis, Trends in Analytical Chemistry 21 (2002) 302.

20] R. Phillips, G. McGarraugh, F.A. Jurik, R.D. Underwood, Whole blood glucose teststrip, US Patent 5,563,042 (1996).

21] L.F. Capitán-Vallvey, in: C.A. Grimes, E.C. Dickey, M.V. Pishko (Eds.), Encyclo-pedia of Sensors, The Pennsylvania State University, Pennsylvania, USA, 2005,pp. 55–93.

22] J. Greyson, Problems and possibilities of chemistry on dry reagent carriers,Journal of Automatic Chemistry 3 (1981) 66.

23] F.C. Williams, F.R. Clapper, Journal of the Optical Society of America 43 (1953)595.

24] A. Zipp, W.E. Hornby, Solid-phase chemistry: its principles and applications inclinical analysis, Talanta 31 (1984) 863.

25] L.K. Chau, M.D. Porter, Optical sensor for calcium: performance, structureand reactivity of calciochrome immobilized at an anionic polymer, AnalyticalChemistry 62 (1990) 1964.

tuators B 174 (2012) 10– 17 17

26] R.F. Gur’eva, S.B. Savvin, A test method for the determination of mer-cury(I,II) on a polymer surface, Journal of Analytical Chemistry 58 (2003)990.

27] O.Yu. Nadzhafova, S.V. Lagodzinskaya, V. Sukhan, Test paper for the determi-nation of aluminium in solution, Journal of Analytical Chemistry 56 (2001)178.

28] O. Sonntag, Dry Chemistry. Analysis with Carrier-bound Reagents, Elsevier Sci-ence Ltd., 1993.

29] W.J. Gui, S.T. Wang, Y.R. Guo, G.N. Zhu, Development of a one-step strip forthe detection of triazophos residues in environmental samples, Analytical Bio-chemistry 377 (2008) 202.

30] M. Mathai, K.J. Siebert, S.T.A. Siebert, R.A. Durst, S.G. Reeves, Developmentof a signal measurement technique for application in multi-analyte lipo-some immunomigration assays, Food and Agricultural Immunology 10 (1998)195.

31] G.P. Zhang, X.N. Wang, J.F. Yang, Y.Y. Yang, G.X. Xing, Q.M. Li, D. Zhao, S.J.Chai, J.Q. Guo, Development of an immunochromatographic lateral flow teststrip for detection of beta-adrenergic agonist clenbuterol residues, Journal ofImmunological Methods 312 (2006) 27.

32] N.I. Ershova, V.M. Ivanov, Application of chromaticity characteristics for directdetermination of trace aluminum with Eriochrome cyanine R by diffuse reflec-tion spectroscopy, Analytica Chimica Acta 408 (2000) 141.

33] M.M. Erenas, O. Pineiro, M.C. Pegalajar, M.P. Cuellar, I. de Orbe-Paya, L.F.Capitán-Vallvey, A surface fit approach with a disposable optical tongue foralkaline ion analysis, Analytica Chimica Acta 694 (2011) 128.

34] M.M. Erenas, M.C. Pegalajar, M.P. Cuellar, I. de Orbe-Paya, L.F. Capitán-Vallvey,Disposable optical tongue for alkaline ion analysis, Sensors and Actuators BB156 (2011) 976.

35] E. Bakker, P. Bühlmann, E. Pretsch, Carrier-based ion-selective electrodesand bulk optodes. 1. General characteristics, Chemical Reviews 97 (1997)3083.

36] A.R. Smith, Color gamut transform pairs, in: Proceedings of the 5th AnnualConference on Computer Graphics and Interactive Techniques, 1978, p. 12.

37] L.F. Capitán-Vallvey, M.D. Fernandez-Ramos, P. Alvarez de Cienfuegos, Opti-cal test strip for calcium determination based on neutral ionophore, AnalyticaChimica Acta 451 (2002) 231.

38] I. de Orbe-Paya, M.M. Erenas, L.F. Capitán-Vallvey, Potassium disposable opticalsensor based on transflectance measurements, Sensors and Actuators B 127(2007) 586.

Biographies

Miguel Maria Erenas was born in 1981 in Granada (Spain). He received the MScdegree (2004) and the PhD degree in Analytical Chemistry (2011) from the Uni-versity of Granada (Granada, Spain) in 2003 and 2009, respectively. His researchinterests include the use of imaging along with disposable sensors, and their appli-cations to handheld instrumentation.

Kevin Cantrell received his MS (1998) and PhD (2001) from Oregon State Universityand his undergraduate degree from Furman University (1992). He joined the facultyof the University of Portland (Portland, OR, USA) in 2001 and is currently an AssociateProfessor there. His research interests include environmental chemistry, chemicalsensors, chemometrics, and automated methods of analysis.

Julio Ballesta-Claver was born in 1977 in Granada (Spain). He received the MScdegree in Chemistry (2000) and the PhD degree in Analytical Chemistry (2009),both from the University of Granada. He is currently working as a Researcher atthe ECsens group, Department of Analytical Chemistry, University of Granada. Hiscurrent research interests include electrochemiluminescence, chemical sensors andbiosensors, electroanalytical techniques and polymer sciences.

Ignacio de Orbe-Payá (Granada, 1960) is currently Associate Professor of theDepartment of Analytical Chemistry at the University of Granada (Spain). His mainareas of interest are the development of the sensing phases for multianalyte sensingusing multivariate calibration methods for environmental, biomedical and foodanalysis.

Luis Fermín Capitán-Vallvey is a full professor of Analytical Chemistry at the Uni-versity of Granada, received his BSc in Chemistry (1973) and PhD in Chemistry (1986)from the Faculty of Sciences, University of Granada (Spain). In 1983, he founded

López, the interdisciplinary group ECsens, which includes Chemists, Physicists andElectrical and Computer Engineers at the University of Granada. His current researchinterests are the design, development and fabrication of sensors and portable instru-mentation for environmental, health and food analysis and monitoring.