sensors Article

Development of a Calibration Strip for Immunochromatographic Assay Detection Systems Yue-Ming Gao 1,2, *,† , Jian-Chong Wei 1,2,3,† , Peng-Un Mak 2,4 , Mang-I. Vai 2,3,4 , Min Du 1,2 and Sio-Hang Pun 3 1 2 3 4

* †

College of Physics and Information Engineering, Fuzhou University, Fuzhou 350116, China; [email protected] (J.-C.W.); [email protected] (M.D.) Key Lab of Medical Instrumentation & Pharmaceutical Technology of Fujian Province, Fuzhou 350116, China; [email protected] (P.-U.M.); [email protected] (M.-I.V.) State Key Laboratory of Analog and Mixed Signal VLSI, University of Macau, Macau 999078, China Department of Electrical and Computer Engineering, Faculty of Science and Technology, University of Macau, Macau 999078, China Correspondence: [email protected]; Tel./Fax: +86-591-8789-3239 These authors contributed equally to this work.

Academic Editor: Mun’delanji Vestergaard Received: 16 May 2016; Accepted: 21 June 2016; Published: 29 June 2016

Abstract: With many benefits and applications, immunochromatographic (ICG) assay detection systems have been reported on a great deal. However, the existing research mainly focuses on increasing the dynamic detection range or application fields. Calibration of the detection system, which has a great influence on the detection accuracy, has not been addressed properly. In this context, this work develops a calibration strip for ICG assay photoelectric detection systems. An image of the test strip is captured by an image acquisition device, followed by performing a fuzzy c-means (FCM) clustering algorithm and maximin-distance algorithm for image segmentation. Additionally, experiments are conducted to find the best characteristic quantity. By analyzing the linear coefficient, an average value of hue (H) at 14 min is chosen as the characteristic quantity and the empirical formula between H and optical density (OD) value is established. Therefore, H, saturation (S), and value (V) are calculated by a number of selected OD values. Then, H, S, and V values are transferred to the RGB color space and a high-resolution printer is used to print the strip images on cellulose nitrate membranes. Finally, verification of the printed calibration strips is conducted by analyzing the linear correlation between OD and the spectral reflectance, which shows a good linear correlation (R2 = 98.78%). Keywords: calibration strip; immunochromatographic (ICG) assay; fuzzy c-means (FCM) algorithm; maximin-distance algorithm

1. Introduction Lateral flow immunoassay, also known as immunochromatographic (ICG) assay, has been reported on a great deal for its several benefits—high sensitivity, ease of operation, low budget, etc. [1–4]. It utilizes antigen and antibody properties for the rapid detection of an analyte. Among the diverse labels of the antibody, colloidal gold particles are widely used [2,5–8]. Accordingly, colloidal gold-based ICG assay has been demonstrated to be potentially useful for medical diagnosis and detection of food safety, pathogen, drugs, environment, etc. [5,8–14]. Therefore, more and more research focuses on the applications of colloidal gold-based ICG assay from qualitative or semi–qualitative detection with the naked eye for precise quantitative detection. Based on the operating principles and the concerned hardware of the ICG assay detection system, they can be categorized into two groups, which are image processing detection systems Sensors 2016, 16, 1007; doi:10.3390/s16071007

www.mdpi.com/journal/sensors

Sensors 2016, 16, 1007

2 of 14

and photoelectric detection systems, respectively. Image processing detection systems use an image capture unit (camera or image scanner) to obtain an image of the whole test strip and performs the specific image processing algorithm to achieve the detection results. For example, Chia-Hsien et al. presents an optical inspection system based on the Taguchi method, which can achieve better linearity and decrease the standard deviation [15]. In photoelectric detection systems, a moving unit driven by a driving motor is used to scan the test strip and a photodiode is employed for photoelectric conversion. The scans can be performed very rapidly and it gets a 1-D signal along the scanning axis which results in a lower computational burden. In this context, several studies have been reported on photoelectric detection systems for ICG assays. For example, Gu et al. developed a portable fluorescence reader for the determination of C-reactive protein, which has a good sensitivity of 0.1 mg/L and linear dynamic range extended to 400 mg/L [16]. Yan et al. reports an ICG assay-based biosensor for rapid quantitative detection of Yersinia pestis [17]. Obviously, the existing research mainly focus on increasing the dynamic detection range or application fields. However, calibration of the detection system, which has a great influence on the detection accuracy, has not been addressed properly. Therefore, this work develops a printed calibration strip for the calibration of an ICG assay-based photoelectric detection system. Optical density (OD) value indicates the amount of light absorbed by a solution of organic molecules on the test strip measured by a spectrophotometer, which can be used to estimate the concentration of the colloidal gold particles on the test strip. Therefore, this work was based on analyzing features of the test strip by performing an image processing algorithm, which seeks the relation of the OD value and characteristic quantity of the test strip image. According to the obtained color information of hue (H), saturation (S), and value (V) of the test strips, the calibration strip is printed. Further, a photoelectric detection system tests the printed calibration strip for verification. The general steps of this work are described as follows: firstly, an image of ICG assay test strip is captured by an image acquisition device, followed by noise reduction using mean and median filters. Then, a fuzzy c-means (FCM) clustering algorithm and maximin-distance algorithm are proposed for image processing in the HSV color space, which extracts a test line of the strip image. In addition, experiments with different HCG solutions and different detection times are conducted to find the best characteristic quantity. By analyzing the linear coefficient, an average value of H at 14 min is chosen as the characteristic quantity for the calibration test strip and the empirical formula between H and OD values is obtained. Therefore, H is predicted by a number of selected OD values and S and V are calculated. Then, H, S, and V are transferred to the RGB color space and a high-resolution printer is used to print the RGB image of the test strip on cellulose nitrate membranes. Finally, verification of these printed calibration strips is performed by analyzing the linear correlation between OD and the spectral reflectance of the printed calibration strips. The rest of this paper is organized as follows: Section 2 introduces the quantitative detection system followed by methodology in Section 3. Section 4 presents the experimental results and discussions. Finally, the conclusions are drawn in Section 5. 2. Quantitative Detection System 2.1. Principle of Quantitative Detection The Beer-Lambert Law describes the relation between the attenuation of light and the properties of the material through which the light is traveling [18]. It is the basis for the principle of quantitative detection. By definition, it describes the relationship of A (the absorbance of solution), b (thickness of medium that absorbs the incident light), and c (concentration of solution). According to quantum theory, when a monochromatic parallel light irradiates a uniform medium of solution, the total absorbance of the medium is the sum of absorbance of every individual object. i.e.,: A“

m ÿ i “1

εi bci

(1)

Sensors 2016, 16, 1007

3 of 14

where εi and ci are molarity and constant, respectively. When there is only one kind of absorbent medium, Equation (1) can be simplified as: A “ εbc

(2)

In photoelectric detection systems for colloidal gold-based ICG assay, the test line on the detection strip can be regarded as a thin layer of solution with a certain thickness b. Therefore, Equation (2) is simplified as: A “ k1 c (3) where k1 “ εb is constant. This indicates that A (the absorbance of test line) is proportional to c (concentration of solution). Thus, when stable parallel light irradiates the test line of the strip, the darker the color of test line is, the larger A is, and the smaller the intensity of reflective light, whereas the lighter the color the test line is, the smaller A is, and the larger the intensity of reflective light. Collected by optical fiber, reflective light is focused on a photodiode which transfers optical signals into electric signals. Finally, the test strip can be quantitatively detected by analyzing the electric signals. Conversely, in image processing detection system for colloidal gold-based ICG assay, when stable parallel light irradiates the surface of the medium, there is no light reflecting or penetrating in the ideal case, which means that all light is absorbed by the medium. Therefore, A is approximately equal to the integral optical density (IOD). IOD is given by: IOD “

N ÿ i “1

ODpiq “

N ÿ

φ lg p0iq i “1 φ

(4)

where OD(i) is the IOD of pixel i, φ0 is the reflective optical density of zero concentration of the solution, φpiq is the reflective optical density of pixel i, and N is the total number of image pixels. In the ideal case, both the background of the strip image and regions outside the test line are white, which indicates that all incident light is reflected. Thus, the density of incident light is equal to the density of reflected light. Additionally, for CCD or CMOS image sensors with linear photoelectric characteristics, the output current of sensors is proportional to the optical density of incident light. Hence: IOD “

N ÿ

N N ÿ ÿ φ G0 I0 lg p0iq “ lg piq lg piq “ I G i “1 φ i“1 i“1

(5)

where I(i) and G(i) are the output current and gray value of pixel i, respectively, I0 and G0 are the output current and gray value of the strip background, respectively. Thus, the concentration of solution can be calculated by measuring the gray value of test line and background of the image, which indicates that the test strip can be quantitatively detected by performing specific image processing algorithms. 2.2. Photoelectric Detection System As aforementioned in Section 1, the photoelectric detection system is superior to image processing detection systems in detection speed and computational burden, which attracts many researchers. Figure 1a displays the schematic diagram of photoelectric detection system, including a mechanical module, a photoelectric module, and a central board. In the mechanical module, the test strip is placed on the mechanical stage, which is driven in and out by a driving motor. In the optical module, two LEDs irradiate light on the test strip, as indicated by green arrows. An optical fiber is placed vertically above the test strip, through which the reflective light is focused on a photodiode for photoelectric conversion, as indicated by red purple arrows. Following that, electrical signal is transferred to a digital signal by an A/D conversion unit. The central board, which is equipped with a high-performance embedded processor, controls the whole detection procedure and performs the signal processing algorithm. The 3D structure of photoelectric detection system is shown in Figure 1b, in which major

Sensors 2016, 16, 1007 Sensors 2016, 16, 1007

4 of 14 4 of 14

components the test strip, photoelectric mechanical module, stage, central board,stage, and 1b, in whichincluding major components including the testmodule, strip, photoelectric mechanical driving module. central board, and driving module.

(a)

(b)

Figure1.1. (a) (a) Schematic Schematic diagram diagram of of photoelectric photoelectric detection detection system; system; and and (b) (b) the the 3D 3D structure structure of of the the Figure photoelectric detection system; with 1: test strip; 2: photoelectric module; 3: mechanical stage; photoelectric detection system; with 1: test strip; 2: photoelectric module; 3: mechanical stage; 4: central4: centraland board; and 5: module. driving module. board; 5: driving

The selection of an LED light source is based on the principle of complementary color. By The selection of an LED light source is based on the principle of complementary color. analyzing the test line absorption spectrum of the colloidal gold ICG assay test strip, the maximum By analyzing the test line absorption spectrum of the colloidal gold ICG assay test strip, the maximum absorption wavelength is found at 525 nm, which is within the green light wavelength range (500~560 absorption wavelength is found at 525 nm, which is within the green light wavelength range nm). The complementary color of green is red-purple. Therefore, according to the principle of (500~560 nm). The complementary color of green is red-purple. Therefore, according to the principle of complementary color, the test line of the strip absorbs green light and shows red-purple. In this complementary color, the test line of the strip absorbs green light and shows red-purple. In this regard, regard, a green LED is selected to get the highest excitation energy. Furthermore, the linear working a green LED is selected to get the highest excitation energy. Furthermore, the linear working range of range of photoelectric sensors, the transmission characteristics of the optical fiber, as well as the photoelectric sensors, the transmission characteristics of the optical fiber, as well as the parameters of parameters of the filter and amplifying circuit in the central board, have certain differences during the filter and amplifying circuit in the central board, have certain differences during manufacturing. manufacturing. They may even drift with different rules after a long time of operation, e.g., variation They may even drift with different rules after a long time of operation, e.g., variation of the temperature of the temperature and humidness, which have a great impact on detection accuracy and repeatability and humidness, which have a great impact on detection accuracy and repeatability of the photoelectric of the photoelectric detection system. Therefore, calibration of the detection system plays a significant detection system. Therefore, calibration of the detection system plays a significant role before the role before the detection, and how to extract the property of the ICG test strips for designing the detection, and how to extract the property of the ICG test strips for designing the calibration strip in calibration strip in the next step becomes the primary problem. the next step becomes the primary problem. 3.3. Methodology Methodology

3.1. Image ImageAcquisition Acquisitionand andProcessing Processing 3.1. 3.1.1. 3.1.1. Structure Structureof ofTest TestStrip Strip In Ingeneral, general,an anICG ICGassay assaytest teststrip stripconsists consistsof ofthree threesections, sections,including includingthe thesample samplepad, pad,analytical analytical membrane, pad, as shown in Figure 2. The2.sample pad includes a samplea hole, through membrane,and andabsorption absorption pad, as shown in Figure The sample pad includes sample hole, which thewhich test samples placedare byplaced drops. by In drops. a conjugate pad, colloidal nanoparticles are used through the testare samples In a conjugate pad, gold colloidal gold nanoparticles as markers formarkers the specific target antigen. On the analytical membrane, there are a T (Test) arethe used as the for the specific target antigen. On the analytical membrane, thereline areand aT C(Test) (Control) line C where the antibody is placed. The T line indicates theTconcentration the test sample, line and (Control) line where the antibody is placed. The line indicatesof the concentration whereas thesample, C line confirms of the test. absorption pad,The which is locatedpad, at the otheris of the test whereas the validity C line confirms theThe validity of the test. absorption which located attest the strip, othercreates end ofcapillary the test strip, action. Asto the testfrom samples starts to flow end of the action.creates As thecapillary test samples starts flow the sample pad to from the sample padastoindicated the absorption as immunoreaction indicated by the arrow, the the absorption pad, by the pad, arrow, occursimmunoreaction in the conjugate occurs pad to in form conjugate pad to form the conjugated particles (colloidal gold-labelled antigen-antibody complex). the conjugated particles (colloidal gold-labelled antigen-antibody complex). Then, the conjugated Then, the conjugated wick along the analytical membrane where another immunoreaction particles wick along theparticles analytical membrane where another immunoreaction occurs on the T line and occurs on the line and theparticles C line. The of the particles will continue their journey until they reach the C line. TheT rest of the willrest continue their journey until they reach the absorption pad. the absorption pad.can Finally, test results canthe be color interpreted of the T line and C line. Finally, test results be interpreted from of the Tfrom line the andcolor C line.

Sensors 2016, 16, 1007

5 of 14

Sensors 2016, 16, 1007

5 of 14

Sensors 2016, 16, 1007

5 of 14

Figure 2. Schematic diagrams of the ICG assay test strip.

Figure 2. Schematic diagrams of the ICG assay test strip.

3.1.2. Image Acquisition Device Figure 2. Schematic diagrams of the ICG assay test strip.

3.1.2. Image Acquisition Device

The image acquisition device is mainly composed of three parts—CMOS, zoom lens, and LED 3.1.2. Image Acquisition Device

lightimage source,acquisition as shown in device Figure 3a. The LEDcomposed light sourceof is three designed as a cyclic structure improve The is mainly parts—CMOS, zoomto lens, and LED Theas image device is is mainly composed ofis three zoom lens, and quality of theacquisition captured image. It placed directly above theparts—CMOS, test strip a distance of 39.00 cm. light the source, shown in Figure 3a. The LED light source designed as aatcyclic structure to LED improve light source, shown in image. Figure 3a. The LED source is designed as astrip cyclic to improve As indicated by solid black dot in 3b, light there are 8above LEDs in inner loop and 16 LEDs the cm. the quality of theas captured ItFigure is placed directly thethe test atstructure a distance ofin39.00 the quality of the captureda image. It is placed directly above is theselected test strip at a distance of 39.00 cm. outer loop. Additionally, 10-bit ADC CMOS image sensor to capture the strip image. As indicated by solid black dot in Figure 3b, there are 8 LEDs in the inner loop and 16 LEDs in the outer As indicated by solid black in source, Figure 3b, there 8 LEDs in adjust the inner 16 LEDs in the Between the CMOS and LEDdot light there is aare zoom lens to the loop focaland length. Obviously, loop. Additionally, a 10-bit ADC CMOS image sensor is selected to capture the strip image. Between outer loop. Additionally, a 10-bit ADC CMOS image sensor is selected to capture the strip image. external natural lights have a great influence on the quality of the acquired image. Therefore, this the CMOS and light source, there is a zoom lens to adjust focal Obviously, external Between theLED CMOS LEDbox light there a zoom lens tothe adjust thelength. focal Obviously, device works inside and a black to source, eliminate the is outside interference. Finally, the length. acquired image is natural lights have a great influence on the quality of the acquired image. Therefore, this device works external natural lights have a great influence on the quality of the acquired image. Therefore, this transmitted to computer by USB for further processing. insidedevice a black box to eliminate the outside interference. Finally, the acquired image is transmitted works inside a black box to eliminate the outside interference. Finally, the acquired image is to transmitted to for computer USB for further processing. computer by USB furtherbyprocessing.

(a)

(b)

Figure 3. (a) Structure of the image acquisition device; and (b) structure of the light source. (a) (b)

3.1.3. Image Processing Figure 3. (a) Structure of the image acquisition device; and (b) structure of the light source. Figure 3. (a) Structure of the image acquisition device; and (b) structure of the light source. The acquired image of the test strip can be segmented into three parts, which are strip shell, 3.1.3. Image Processing background, and test line part. Nevertheless, only the test line part contains the detection 3.1.3. Image Processing The acquired image of the test strip can betest segmented into three parts, which are strip shell, information. Therefore, image segmentation and line extraction should be conducted. The acquired image ofline the test can representative be segmented intoline three parts, which strip background, test Nevertheless, only the color test part contains theare detection Basically, and RGB, HSV, andpart. YUVstrip are three spaces which are commonly usedshell, information. Therefore, image segmentation and test line extraction should be conducted. background, and test line part. Nevertheless, only the test line part contains the detection information. in the image processing field [19]. However, RGB and YUV are mainly applied in raw data and coding Basically, RGB,HSV HSV, YUV representative spaces which are HSV commonly used and standards, whereas is and more closer tothree human perceptions [20]. In this regard, the color space Therefore, image segmentation and testare line extraction shouldcolor be conducted. Basically, RGB, HSV, in the image processing field [19]. However, RGB and YUV are mainly applied in raw data and coding selected to perform the fuzzy algorithm and algorithm YUV is are three representative colorc-means spaces (FCM) whichclustering are commonly used in maximin-distance the image processing field [19]. standards, whereas HSV is more closer to human perceptions [20]. In this regard, the HSV color space for strip image However, RGB andsegmentation. YUV are mainly applied in raw data and coding standards, whereas HSV is more is selected to perform the algorithm, fuzzy c-means (FCM) algorithm and maximin-distance algorithm The FCM clustering which is anclustering unsupervised been widely closer to human perceptions [20]. In this regard, the HSV colorclustering space is technique, selected tohas perform the fuzzy for strip image segmentation. used in biomedical image segmentation [21–25]. In the FCM clustering algorithm, fuzzy data is c-means (FCM) clustering algorithm and maximin-distance algorithm for strip image segmentation. The FCM algorithm, which is an clustering technique, has been widely classified into aclustering proper subset by minimizing theunsupervised objective function. Assuming the given sample set The FCM clusteringimage algorithm, which is an unsupervised clustering technique,fuzzy has been widely s used in biomedical segmentation [21–25]. In the FCM clustering algorithm, data is X = x1 , x2 , , xn   R ,segmentation sample dimension isIns, the sample amount is n, and c (1given < c < n) is subset used is in biomedical imagesubset [21–25]. FCM clustering algorithm, fuzzy data is classified into a proper by space minimizing the objective function. Assuming the sample set amount classifying. thisminimizing case, the FCM be described as follows: classified thecan objective function. Assuming the given sample set is X into = xafter ,ax2proper , , xn  subset  Rs In by 1 is , sample space dimension is s, sample amount is n, and c (1 < c < n) is subset s X “ tx1 , x2 , ¨ ¨ ¨ , xn u Ă R , sample space dimension is s, sample amount is n, and c (1 < c < n) is subset amount after classifying. In this case, the FCM can be described as follows:

amount after classifying. In this case, the FCM can be described as follows:

Sensors 2016, 16, 1007

6 of 14

c ÿ n ÿ

uijm d2ij

(6)

uij “ 1,

1ďjďn

(7)

uij ą 0,

1ďiďc

(8)

JpU, Vq “

i“1 j“1

where

c ÿ i “1 n ÿ j “1

uij ě 0, 1 ď i ď c, 1 ď j ď n

(9)

In Equation (6) to Equation (9), m (m > 1) is the fuzzy parameter, U “ uij is a c ˆ n fuzzy partition matrix, uij indicates the membership value of xj that belongs to class i, V “ rv1 , v2 , ¨ ¨ ¨ , vc s is a s ˆ c matrix which is composed of c clustering center vectors, and dij “ ||x j ´ vi || is the distance between sample point xj and center point vi . Therefore, the fuzzy clustering algorithm is evolved into the optimization of the restrained argument of pU, Vq. Then, the iterative equation is acquired by the necessary conditions of the extreme point. řn

j “1

v i “ řn

uijm x j

j “1

uijm

,

i “ 1, 2, ¨ ¨ ¨ , c

(10)

ˇ ( Assuming Ij “ pi, jqˇx j “ vi , 1 ď i ď c , if Ij “ ∅, then » uij “ –

c ÿ r “1

˜

dij drj

¸

2 m´1

fi´1 fl

i “ 1, 2, ¨ ¨ ¨ , c; j “ 1, 2, ¨ ¨ ¨ , n

(11)

If Ij ‰ ∅, then uij is the arbitrary non-negative real number that satisfies the following condition: c ÿ

uij “ 1; uij “ 0, dij ‰ 0

(12)

i“1

The equation of membership degree shows the mapping relationship from point to set and the membership degree can be updated by Equation (13). Now, the implementation steps of the FCM algorithm are: firstly, initialize the clustering center or membership degree matrix; then, Equations (10) and (13) are iterated until the inequality (14) is satisfied. Specific steps of the FCM algorithm are explained in Table 1. $ « 2 ff´1 c ´ d ¯ m´1 ’ ř ’ ij ’ ’ , Ij “ ∅ & drj uij “

r “1

1 ’ , I ‰ ∅, i P I j ’ ’ |I | j ’ % j 0, Ij ‰ ∅, i R Ij

||V pkq ´ V pk´1q || ď ε, k ě 1

(13)

(14)

Sensors 2016, 16, 1007

7 of 14

V ( k )  V ( k 1)   , k  1 Sensors 2016, 16, 1007

(14) 7 of 14

Table 1. Specific steps of the FCM clustering algorithm. Table 1. Specific steps of the FCM clustering algorithm.

Step 1: Set clustering number c and fuzzy exponent m; initialize center of clustering V(°); set  times k. convergence accuracy Step 1: Set clustering number and c anditeration fuzzy exponent m; initialize center of clustering V (˝) ; set convergence ( ) accuracy and iteration k. Step 2: εCalculate V ° times according to Equation (13). (k + 1) according to Equation (10). (˝) Step 3: Let k = k + 1, calculate V Step 2: Calculate V according to Equation (13). + 1) according Step Repeat 2 andV3(k until inequality (14) is satisfied. Step 3: 4: Let k = k + Steps 1, calculate to Equation (10). Step 4: Repeat Steps 2 and 3 until inequality (14) is satisfied.

A maximin-distance algorithm is a simple heuristic procedure which can be used in initializing the cluster center in the algorithm FCM algorithm to assure stability of the result to avoid random A maximin-distance is a simple heuristic procedure which canand be used in initializing initialization [26–29]. The specific steps of algorithm is described in Table 2. the cluster center in the FCM algorithm to assure stability of the result and to avoid random initialization [26–29]. The specific steps of algorithm is described in Table 2. Table 2. Procedure of the maximin-distance algorithm. Table 2. Procedure of the maximin-distance algorithm.

X  x1 , x2 ,...xn  Step 1: Assuming the dataset X is composed of n vectors, i.e., ; aibitrarily Step 1: one Assuming is dataset composed n vectors, i.e., X “ vtx , x2 , . v. .1 x=n x u;1.aibitrarily select one select vectorthe (x1dataset ) from X the as of the first clustering 1,1e.g., vector (x1 ) from the dataset as the first clustering v1 , e.g., v1 = x1 . Step 2: Calculate the distances between v1 and all other points in the set, find the point with Step 2: Calculate the distances the largest distance and set between as v2. v1 and all other points in the set, find the point with the largest distance and set as v2 . Step 3: Calculate the distances between the remaining vectors of X and the known Step 3: Calculate the distances between the remaining vectorsas ofaXgroup. and theThen, knownselect clustering clustering centers, and choose the minimum distances the centers, and choose the minimum distances as a group. Then, select the maximum distance in this group. If the maximum distance in this group. If the maximum is larger than the given threshold maximum is larger than the given threshold m|Z2 ´ Z1 |, this point will be set as a new clustering center. m |Z 2  Z0.5 | ď m ď 1. Generally, 1 , this point will be set as a new clustering center. Generally, 0.5  m  1 . Step Step 3 until the the acquired maximum distance does not satisfy condition of creating a Step4:4:Repeat Repeat Step 3 until acquired maximum distance does notthe satisfy the condition new center or the value of the clustering center reaches the desired number. of creating a new center or the value of the clustering center reaches the desired number. 3.2. Development Development and and Verification Verification of of the the Calibration 3.2. Calibration Strip Strip Development and steps of the strip are shown The most important andverification verification steps of calibration the calibration strip are Figure shown4.Figure 4. The most part of the development of the calibration strip is searching the relationship between the OD value important part of the development of the calibration strip is searching the relationship between the and value characteristic quantitiesquantities of the test of strip However, in colloidal ICG assay,ICG the OD and characteristic theimage. test strip image. However, ingold-based colloidal gold-based binding of antigen and antibody is a dynamic process. Therefore, color depth of the test line of the assay, the binding of antigen and antibody is a dynamic process. Therefore, color depth of the test stripofisthe changing time because of because capillarity siphonand action, i.e., action, there isi.e., no exact the line strip isover changing over time of and capillarity siphon there end is nofor exact chromatographic process. Certainly, selection of the detection time has a great on theimpact detection end for the chromatographic process. Certainly, selection of the detection timeimpact has a great on accuracy. In this regard, experiments with different concentrations of solutions and different detection the detection accuracy. In this regard, experiments with different concentrations of solutions and times are detection conductedtimes to find best characteristic of the strip image. By the linear different arethe conducted to find thequantities best characteristic quantities ofanalyzing the strip image. By coefficient, the empirical formula between OD value and characteristic quantities of the strip image is analyzing the linear coefficient, the empirical formula between OD value and characteristic quantities obtained, by which H, S, and V are calculated. In order to print the strip image with a high-resolution of the strip image is obtained, by which H, S, and V are calculated. In order to print the strip image printer, the obtained H, S, andthe V values areH, transferred to theare RGB color space. Finally, the printed with a high-resolution printer, obtained S, and V values transferred to the RGB color space. calibration strip is produced on cellulose nitrate membranes. Finally, the printed calibration strip is produced on cellulose nitrate membranes.

Figure Figure 4. 4. Development Development and and verification verification steps steps of of the the calibration calibration strip. strip.

Sensors 2016, 16, 1007

8 of 14

Sensors 2016, 16, 1007

8 of 14

As aforementioned in Section 2, the principle of quantitative detection is based on the As aforementioned in Section 2, the principle of quantitative detection is based on the BeerBeer-Lambert Law, which indicates that the color depth of the test line in different concentrations Lambert Law, which indicates that the color depth of the test line in different concentrations is linear is linear with the absorbance of the test line. Therefore, verification of the printed calibration strips with the absorbance of the test line. Therefore, verification of the printed calibration strips can be can be evaluated by the absorption spectral peak value theline. test In line. Incontext, this context, verification of evaluated by the absorption spectral peak value of the of test this verification of the the calibration strip is conducted by analyzing the linear correlation between the OD value and the calibration strip is conducted by analyzing the linear correlation between the OD value and the spectral reflectance reflectance of of the the printed printed calibration calibration strip. strip. spectral 4. Experimental Results and Discussion 4. Experimental Results and Discussion 4.1. Experiment of Test Line Extraction 4.1. Experiment of Test Line Extraction Basically, there is mainly Gaussian noise and impulse noise in the original image. Thus, a mean Basically, therefilter is mainly Gaussian to noise and noise impulse noise in the originalthe image. Thus, a mean filter and median are performed reduce before implementing test line extraction filter and median filter are performed to reduce noise before implementing the test line extraction ˝ algorithm. In order to improve processing efficiency, H, S, and V are divided by 6 , 0.0625, and 0.0625, algorithm. In order to improve processing efficiency, H, S, and are divided by 6°, 0.0625, 0.0625, respectively, as described in Equation (15). This reduces the Vimage dimensions withoutand losing the respectively, as described in Equation (15). This reduces the image dimensions without losing the main features of the image [30,31]. The three-dimensional h’ s’ v’ histogram of the test strip image main features of [30,31]. The three-dimensional h’ s’ v’ histogram of the test strip image (concentration is the 350 image mIU/mL) is shown in Figure 5a. (concentration is 350 mIU/mL) is shown in Figure 5a. h1 “ H{6o ; s1 “ S{0.0625; v1 “ V{0.0625 h '  H / 6o ；s '  S / 0.0625；v '  V / 0.0625

(15) (15)

Inthis thishistogram, histogram,the thelargest largest number numberof ofpixels pixelsisisselected selectedas asthe thefirst first clustering clustering center. center. Based Based on on In the aforementioned aforementioned maximin-distance maximin-distance algorithm, algorithm, the the clustering clustering number number c and clustering center V((°˝)) the are determined. determined. Then, Then,aaFCM FCMalgorithm algorithmisisimplemented implementedon onthe theHSV HSVcolor colorspace spaceto tosegment segmentthe thestrip strip are image.The Theclustering clusteringresult resultisisshown shownin inFigure Figure5b. 5b. Without Withoutinsufficient insufficient or or over over segmentation segmentationof of the the image. test line, the segmentation algorithm has high performance, as shown in Figure 6, where (a) is the test line, the segmentation algorithm has high performance, as shown in Figure 6, where (a) is the original image, image, (b) is the segmented original segmented test test line, line, (c) (c)isisthe theHSV HSVimage imagewith withadjusted adjustedbrightness, brightness,and and(d) (d)is resulting test line segmentation. isthe the resulting test line segmentation.

16

16

14

14

12

12 10 v'

v'

10 8

8

6

6

4

4

2 16

14

2 16 12

10

8 s'

6

4

2

0

25

(a)

30

35

40 h'

45

50

55

60

14

12

10

8 s'

6

4

2

0

25

30

35

45

40

50

55

h'

(b)

Figure and(b) (b)the theresult resultof ofFCM FCM clustering. clustering.  ■ is ' s ' v ' ;and Figure 5. 5. (a) (a)Three-dimensional Three-dimensionalhistogram histogramofofhh1s1v1; is the the position position of of the the clustering clustering center center and and the the other other five five symbols symbols are are five five different different clusterings. clusterings.

60

Sensors 2016, 16, 1007 Sensors 2016, 16, 1007 Sensors 2016, 16, 1007

9 of 14 9 of 14 9 of 14

(a) (a)

(b) (b)

(c) (c)

(d) (d)

Figure 6. Process image segmentation. Original image; (b) the test line; (c) HSV (a)(a) Original image; (b) the test line; HSV Figure 6. 6. Process Processofof ofimage imagesegmentation. segmentation. (a) Original image; (b) segmented the segmented segmented test(c) line; (c)image HSV image with adjusted brightness; and (d) the result of test line segmentation. with brightness; and (d)and the (d) result test line segmentation. imageadjusted with adjusted brightness; theof result of test line segmentation.

4.2. 4.2. Development Development of of the the Calibration Calibration Strip Strip HCG preliminary selection Down’s syndrome the HCG solution, solution, which which has has been been used used in in preliminary selection of of Down’s syndrome and and the diagnosis of early pregnancy, or eccyesis, is chosen as the reagent for extracting characteristic or eccyesis, is chosen as the reagent for extracting characteristic quantities. diagnosis ofofearly earlypregnancy, pregnancy, or eccyesis, is chosen as the reagent for extracting characteristic quantities. addition, test strips selected from kits (Xiamen Boson Biotech Ltd., In addition,In strips are diagnostic kits (Xiamen Biotech Ltd.,Co., Xiamen, quantities. Intest addition, test selected strips are arefrom selected from diagnostic diagnostic kits Boson (Xiamen BosonCo., Biotech Co., Ltd., Xiamen, China) for rapid quantitative determination of human choriogonadotropin (β-HCG) in the China) forChina) rapid for quantitative determination of human choriogonadotropin (β-HCG) in(β-HCG) the same in patch Xiamen, rapid quantitative determination of human choriogonadotropin the same patch and specification. The detection sensitivity of the strip is 10 mIU/mL. Additionally, the and specification. The detectionThe sensitivity of sensitivity the strip is of 10the mIU/mL. the concentration same patch and specification. detection strip isAdditionally, 10 mIU/mL. Additionally, the concentration of involves 10, 150, 200, 250, 300, 350, 400, 450, Generally, of diluents involves 10, 50, 100, 150, 200,100, 250, 300, 350, 400, Generally, the best concentration of diluents diluents involves 10, 50, 50, 100, 150, 200, 250, 300,450, 350,500 400,mIU/mL. 450, 500 500 mIU/mL. mIU/mL. Generally, the best is min~18 after sample dropped into sample pad. detection time is 10time min~18 aftermin the sample droppedis the sample pad. Therefore, the best detection detection time is 10 10min min~18 min after the thesolution sampleissolution solution isinto dropped into the the sample pad. Therefore, the image acquisition device captures the strip image at 10 min, 12 min, 14 min, 16 min, the image acquisition device captures the strip image at 10image min, 12 14 min, 16 min, and 16 18 min, min. Therefore, the image acquisition device captures the strip atmin, 10 min, 12 min, 14 min, and Then, value of acquired strip is measured by Then, OD value theOD acquired image is immediately measured by a quantitative detection and 18 18themin. min. Then, ofthe the OD valuestrip of the the acquired strip image image is immediately immediately measured by aa quantitative detection system (SWP-SC-2). To reduce concentration of system for ICG assay (SWP-SC-2). To assay reduce errors, every diluent is repeatedly quantitative detection system for for ICG ICG assay (SWP-SC-2). Toconcentration reduce errors, errors,ofevery every concentration of diluent is repeatedly measured three times. The test strips are shown in Figure 7. measured three times.measured The test strips shown Figure diluent is repeatedly three are times. The in test strips7.are shown in Figure 7.

Figure 7. Test strips for colloidal gold-based ICG assay. Figure Figure 7. 7. Test Test strips strips for for colloidal colloidal gold-based gold-based ICG ICG assay. assay.

As As the the light light source source of of image image acquisition acquisition device device is is not not exactly exactly stable, stable, which which results results in in the the As theoflight source of image acquisition device is not exactly stable,ofwhich results in the instability instability V in the HSV color space, four characteristic quantities the strip image are selected instability of V in the HSV color space, four characteristic quantities of the strip image are selected of V in the HSV color space, four characteristic quantities of the strip image areselected H and Hb 10 from H H 360 from H S S ,, and from H H and and S, S, which which are are oneone- or or two-dimensional two-dimensional mean mean values values of of H ,, 360  H ,, 10 and 2 S, which are oneor two-dimensional mean values of H, 360 ˆ H, 10 ˆ H ˆ S, and H ` p360 ˆ Sq2 . 2 2 2 2 H   H linear  (360 (360coefficient S S)) . The linear coefficient between characteristic quantities concentrations of The between these characteristic quantities and concentrations HCG, and OD . The linear coefficient between these these characteristic quantities and and of concentrations of HCG, and OD values, are shown in Tables 3 and 4, respectively. As shown, an average value of H at values, are OD shown in Tables 3 andin 4, Tables respectively. Asrespectively. shown, an average value H at 14value min has HCG, and values, are shown 3 and 4, As shown, anof average of Hthe at 14 min has the maximum linear coefficient (95.24%) with the concentration of HCG. Further, the 14 min has the maximum linear coefficient (95.24%) with the concentration of HCG. Further, the

Sensors 2016, 16, 1007

10 of 14

maximum linear coefficient (95.24%) with the concentration of HCG. Further, the linear coefficient between an average value of H at 14 min and OD value is over 98%. Thus, an average value of H at 14 min is chosen as the characteristic quantity for the calibration test strip. Table 3. Linear coefficient between characteristic quantities and concentration of HCG (%). Time (min) CQ H 360 ˆ H b 10 ˆ H ˆ S H 2 ` p360 ˆ Sq2

10

12

14

16

18

95.15 79.58 84.57

95.04 84.70 87.54

95.24 88.34 90.21

94.61 89.45 90.01

94.22 91.35 91.88

79.84

84.81

88.41

89.50

90.82

Table 4. Linear coefficient between characteristic quantities and OD value (%). Time (min) CQ H 360 ˆ H b 10 ˆ H ˆ S H 2 ` p360 ˆ Sq2

10

12

14

16

18

98.02 81.33 86.77

98.31 87.87 90.92

98.12 90.91 93.02

98.11 92.18 93.90

97.93 93.11 94.85

81.64

88.02

91.03

92.27

93.56

Now, the empirical formula between H and OD value is obtained by: H “ 10.153 ˆ OD ` 262.427

(16)

From Equation (16), the mean value of H can be calculated from the OD value. Selecting a number of OD values (1.5, 3, 4, 5, 6, 7, 8, 9, 9.5), the predicted mean values of H are shown in Table 5. Table 5. The predicted mean values of H. OD

H

1.5 3 4 5 6 7 8 9 9.5

277.657 292.886 303.039 313.192 323.345 333.498 343.651 353.804 358.881

In order to determine S and V, the pixels of pixij that satisfied the inequality (H ´ 0.25 ă H pixij ď H ` 0.25) are regarded as valid points. For example, if H = 303.039, the valid points of the segmented strip image is shown in Figure 8. Hence, S and V is calculated by Equation (17). S“

ÿ

S pixij {N; V “

ÿ

Vpixij {N

(17)

Sensors 2016, 16, 1007 Sensors 2016, 16, 1007 Sensors 2016, 16, 1007

11 of 14 11 of 14 11 of 14

Figure 8. Valid points of the segmented image. Figure 8. Valid points of the segmented image.

Then, the obtained H, S, and V values are transferred to the RGB color space, which is more Then, the the obtained obtained H, H, S, S, and values are are transferred transferred to to the the RGB RGB color color space, which is is more Then, and V V values space, which more suitable for printing. A high-resolution color inkjet printer is used to print the RGB image on cellulose suitable for printing. A high-resolution color inkjet printer is used to print the RGB image on cellulose suitable for printing. A high-resolution color inkjet printer is used to print the RGB image on cellulose nitrate membranes. The printed calibration test strips are shown in Figure 9. nitrate membranes. The printed printed calibration calibration test nitrate membranes. The test strips strips are are shown shown in in Figure Figure 9. 9.

Figure 9. Printed calibration test strips. Figure 9. Printed Figure 9. Printed calibration calibration test test strips. strips.

4.3. Verification of the Calibration Strip 4.3. Verification of the Calibration Strip 4.3. Verification of the Calibration Strip In the verification of the printed calibration strips, OD values for different concentrations of the In the verification of the printed calibration strips, OD values for different concentrations of the In the (the verification of the calibration strips, OD values different concentrations of the test strips same test kit printed as aforementioned in Section 4.2) is for read by a quantitative detection test strips (the same test kit as aforementioned in Section 4.2) is read by a quantitative detection test strips same test(SWP-SC-2), kit as aforementioned Section 4.2)4.14, is read by5.93, a quantitative system system for(the ICG assay which arein1.50, 2.64, 5.07, 6.43, 7.93,detection 8.93, and 9.87. system for ICG assay (SWP-SC-2), which are 1.50, 2.64, 4.14, 5.07, 5.93, 6.43, 7.93, 8.93, and 9.87. for ICG assay an (SWP-SC-2), which areSpectroradiometric 1.50, 2.64, 4.14, 5.07, 5.93, 6.43, 7.93, 8.93, (Optronic and 9.87. Additionally, Additionally, OL750 Automated Measurement System Laboratories Additionally, an OL750 Automated Spectroradiometric Measurement System (Optronic Laboratories an OL750 Automated Measurement Laboratories Orlando, Inc., Orlando, Florida,Spectroradiometric America) is used to measure theSystem spectral(Optronic reflectance of test lineInc., as displayed Inc., Orlando, Florida, America) is used to measure the spectral reflectance of test line as displayed Florida, is used to measureand the vertical spectraldirection reflectance of test line as displayed in Figure 10, in in FigureAmerica) 10, in which the horizontal refer to wavelength (500~600 nm, increase in Figure 10, in which the horizontal and vertical direction refer to wavelength (500~600 nm, increase which theand horizontal vertical direction referTherefore, to wavelength (500~600 nm, increase by 5ofnm) by 5 nm) spectraland reflectance, respectively. the average spectral reflectance the and test by 5 nm) and spectral reflectance, respectively. Therefore, the average spectral reflectance of the test spectral reflectance, respectively. Therefore, the average spectral reflectance of the test line is calculated, line is calculated, as listed in Table 6, where first column is the OD value and second column is the line is calculated, as listed in Table 6, where first column is the OD value and second column is the as listed spectral in Table reflectance 6, where first column is the(wavelength OD value and second is the spectral average of the test line from 500 tocolumn 600 nm). Theaverage linear fit of the average spectral reflectance of the test line (wavelength from 500 to 600 nm). The linear fit of the reflectance of thereflectance test line (wavelength fromin 500 to 600 The linear fit ofthat theOD average average spectral and OD is shown Figure 11,nm). which demonstrates of thespectral printed average spectral reflectance and OD is shown in Figure 11, which demonstrates that OD of the printed 2 = 98.78%). reflectance ODhas is shown in Figure which demonstrates that OD of the printed(Rcalibration strips In calibration and strips a good linear 11, correlation with the spectral reflectance 2 calibration strips has a good linear correlation with the2 spectral reflectance (R = 98.78%). In has a good linear correlation with the spectral reflectance (R = 98.78%). In conclusion, the developed conclusion, the developed printed calibration strip is effective for ICG assay photoelectric detection conclusion, the developed printed calibration strip is effective for ICG assay photoelectric detection printed systems.calibration strip is effective for ICG assay photoelectric detection systems. systems.

Sensors 2016, 16, 1007 Sensors 2016, 16, 1007

12 of 14 12 of 14

Figure 10. The spectral reflectance of test line of the printed calibration strips. Table 6. The calculated average spectral reflectance of the test line. OD SR 1.50 94.83 2.64 77.49 4.14 67.05 5.07 53.87 5.93 50.21 6.43 45.06 7.93 33.02 8.93 18.05 Figure 10. 10. The The spectral spectral reflectance reflectance of of test line of of the printed calibration strips. Figure 9.87 test line 15.09the printed calibration strips. Table 6. The calculated average spectral reflectance of the test line. OD 1.50 2.64 4.14 5.07 5.93 6.43 7.93 8.93 9.87

SR 94.83 77.49 67.05 53.87 50.21 45.06 33.02 18.05 15.09

Figure 11. 11. Linear Linear fit fit of of the the spectral spectral reflectance. reflectance. Figure

5. Conclusions

Table 6. The calculated average spectral reflectance of the test line.

This work develops a calibration strip for immunochromatographic (ICG) assay photoelectric detection systems. An image of the testOD strip is captured SR by an image acquisition device. Mean and 1.50 in the acquired 94.83 image. Without insufficient or over median filters are used to reduce noise 2.64 77.49 segmentation, the proposed FCM algorithm and maximin-distance algorithm has a good 4.14 67.05 performance on extraction of the test line. with different HCG solution and 5.07 Further, experiments 53.87 different detection times are conducted5.93 to find the best characteristic quantity, which indicates that 50.21 6.43 45.06 the average value of H at 14 min has the best linear coefficient with a concentration of HCG (95.24%) 7.93 and OD value (98.12%). Therefore, the empirical formula33.02 between H and OD (optical density) values 8.93 fit of the spectral 18.05reflectance. Figure 11. Linear is established, by which H, S, and V are calculated. Additionally, H, S, and V values are transferred 9.87 15.09 to the RGB color space and a high-resolution printer is used to print the RGB images on cellulose 5. Conclusions nitrate membranes. Finally, OD and the spectral reflectance of the printed calibration strips are 5. Conclusions analyzed good linear correlation (R2 for = 98.78%), which indicates that the assay developed printed This with workadevelops a calibration strip immunochromatographic (ICG) photoelectric This work a calibration (ICG) assay photoelectric calibration stripdevelops is effective forofthe ofimmunochromatographic the ICG assay system. detection systems. An image thecalibration teststrip stripfor is captured by andetection image acquisition device. Mean and detection systems. image the test stripinis captured by animage. image Without acquisition device. Mean and median filters areAn used to of reduce noise the acquired insufficient or over median filters are used to reduce noise in the acquired image. Without insufficient or over segmentation, segmentation, the proposed FCM algorithm and maximin-distance algorithm has a good the proposed FCM algorithmofand has with a good performance extraction performance on extraction themaximin-distance test line. Further,algorithm experiments different HCG on solution and of the testdetection line. Further, with HCGcharacteristic solution andquantity, differentwhich detection timesthat are different timesexperiments are conducted to different find the best indicates conducted findof the quantity, which indicates the average value(95.24%) of H at the averagetovalue H best at 14characteristic min has the best linear coefficient with a that concentration of HCG 14 min the (98.12%). best linear coefficient a concentration of HCGH(95.24%) OD value (98.12%). and ODhas value Therefore, thewith empirical formula between and ODand (optical density) values Therefore, the empirical formula between H and OD (optical density) values is established, by which is established, by which H, S, and V are calculated. Additionally, H, S, and V values are transferred H, V are calculated. Additionally, H, S, and V values are transferred the RGB color and to S, theand RGB color space and a high-resolution printer is used to print thetoRGB images onspace cellulose anitrate high-resolution printer is used printthe thespectral RGB images on cellulose membranes. membranes. Finally, ODtoand reflectance of thenitrate printed calibrationFinally, strips OD are 2 analyzed with a good linear correlation (R = 98.78%), which indicates that the developed printed calibration strip is effective for the calibration of the ICG assay detection system.

Sensors 2016, 16, 1007

13 of 14

and the spectral reflectance of the printed calibration strips are analyzed with a good linear correlation (R2 = 98.78%), which indicates that the developed printed calibration strip is effective for the calibration of the ICG assay detection system. Acknowledgments: This work was supported by the Taiwan, Hong Kong and Macau Cooperation Project of Chinese Ministry of Science and Technology 2012DFM30040, the Science and Technology Development Fund of Macau, 047/2013/A2, the Major Industry-Academy Cooperation Project of Fujian Province, China, 2011Y4007 and the Major Project of Fujian Province, China, 2014YZ0001. Author Contributions: Yue-Ming Gao contributed to the idea of calibration strip; Jian-Chong Wei wrote the paper; Mang-I Vai and Min Du conceived and designed the experiments; Sio-Hang Pun performed the experiments; Peng-Un Mak analyzed the data. Conflicts of Interest: The authors declare no conflict of interest.

References 1.

2. 3. 4.

5. 6.

7.

8.

9. 10.

11.

12.

13. 14.

Cho, Y.-J.; Lee, D.-H.; Kim, D.-O.; Min, W.-K.; Bong, K.-T.; Lee, G.-G.; Seo, J.-H. Production of a monoclonal antibody against ochratoxin A and its application to immunochromatographic assay. J. Agric. Food Chem. 2005, 53, 8447–8451. [CrossRef] [PubMed] Shyu, R.-H.; Shyu, H.-F.; Liu, H.-W.; Tang, S.-S. Colloidal gold-based immunochromatographic assay for detection of ricin. Toxicon 2002, 40, 255–258. [CrossRef] Sun, X.; Zhao, X.; Tang, J.; Gu, X.; Zhou, J.; Chu, F.S. Development of an immunochromatographic assay for detection of aflatoxin B1 in foods. Food Control 2006, 17, 256–262. Yang, W.; Li, X.-B.; Liu, G.-W.; Zhang, B.-B.; Zhang, Y.; Kong, T.; Tang, J.-J.; Li, D.-N.; Wang, Z. A colloidal gold probe-based silver enhancement immunochromatographic assay for the rapid detection of abrin-a. Biosens. Bioelectron. 2011, 26, 3710–3713. [CrossRef] [PubMed] Chiao, D.-J.; Shyu, R.-H.; Hu, C.-S.; Chiang, H.-Y.; Tang, S.-S. Colloidal gold-based immunochromatographic assay for detection of botulinum neurotoxin type B. J. Chromatogr. B 2004, 809, 37–41. [CrossRef] [PubMed] Shim, W.; Yang, Z.; Kim, J.; Kim, J.; Kang, S.; Woo, G.; Chung, Y.; Eremin, S.A.; Chung, D. Development of immunochromatography strip-test using nanocolloidal gold-antibody probe for the rapid detection of aflatoxin B1 in grain and feed samples. J. Microbiol. Biotechnol. 2007, 17, 1629. [PubMed] Shim, W.-B.; Yang, Z.-Y.; Kim, J.-Y.; Choi, J.-G.; Je, J.-H.; Kang, S.-J.; Kolosova, A.Y.; Eremin, S.A.; Chung, D.-H. Immunochromatography using colloidal gold-antibody probe for the detection of atrazine in water samples. J. Agric. Food Chem. 2006, 54, 9728–9734. [CrossRef] [PubMed] Zhou, Y.; Pan, F.-G.; Li, Y.-S.; Zhang, Y.-Y.; Zhang, J.-H.; Lu, S.-Y.; Ren, H.-L.; Liu, Z.-S. Colloidal gold probe-based immunochromatographic assay for the rapid detection of brevetoxins in fishery product samples. Biosens. Bioelectron. 2009, 24, 2744–2747. [CrossRef] [PubMed] Guo, Y.-R.; Liu, S.-Y.; Gui, W.-J.; Zhu, G.-N. Gold immunochromatographic assay for simultaneous detection of carbofuran and triazophos in water samples. Anal. Biochem. 2009, 389, 32–39. [CrossRef] [PubMed] Ju, Y.; Hao, H.-J.; Xiong, G.-H.; Geng, H.-R.; Zheng, Y.-L.; Wang, J.; Cao, Y.; Yang, Y.-H.; Cai, X.-H.; Jiang, Y.-Q. Development of colloidal gold-based immunochromatographic assay for rapid detection of Streptococcus suis serotype 2. Vet. Immunol. Immunopathol. 2010, 133, 207–211. [CrossRef] [PubMed] Kumar, R.; Singh, C.K.; Kamle, S.; Sinha, R.P.; Bhatnagar, R.K.; Kachru, D.N. Development of nanocolloidal gold based immunochromatographic assay for rapid detection of transgenic vegetative insecticidal protein in genetically modified crops. Food Chem. 2010, 122, 1298–1303. [CrossRef] Meng, K.; Sun, W.; Zhao, P.; Zhang, L.; Cai, D.; Cheng, Z.; Guo, H.; Liu, J.; Yang, D.; Wang, S. Development of colloidal gold-based immunochromatographic assay for rapid detection of Mycoplasma suis in porcine plasma. Biosens. Bioelectron. 2014, 55, 396–399. [CrossRef] [PubMed] Shyu, R.-H.; Tang, S.-S.; Chiao, D.-J.; Hung, Y.-W. Gold nanoparticle-based lateral flow assay for detection of staphylococcal enterotoxin B. Food Chem. 2010, 118, 462–466. Sheng, W.; Li, Y.; Xu, X.; Yuan, M.; Wang, S. Enzyme-linked immunosorbent assay and colloidal gold-based immunochromatographic assay for several (fluoro) quinolones in milk. Microchim. Acta 2011, 173, 307–316. [CrossRef]

Sensors 2016, 16, 1007

15.

16. 17.

18.

19. 20.

21.

22.

23.

24.

25.

26. 27.

28. 29. 30.

31.

14 of 14

Yeh, C.-H.; Zhao, Z.-Q.; Shen, P.-L.; Lin, Y.-C. Optimization of an Optical Inspection System Based on the Taguchi Method for Quantitative Analysis of Point-of-Care Testing. Sensors 2014, 14, 16148–16158. [CrossRef] [PubMed] Gu, Y.; Yang, Y.; Zhang, J.; Ge, S.; Tang, Z.; Qiu, X. Point-of-care test for C-reactive protein by a fluorescence-based lateral flow immunoassay. Instrum. Sci. Technol. 2014, 42, 635–645. [CrossRef] Yan, Z.; Zhou, L.; Zhao, Y.; Wang, J.; Huang, L.; Hu, K.; Liu, H.; Wang, H.; Guo, Z.; Song, Y. Rapid quantitative detection of Yersinia pestis by lateral-flow immunoassay and up-converting phosphor technology-based biosensor. Sens. Actuators B Chem. 2006, 119, 656–663. [CrossRef] Vieira, G.P.; Perdigão, S.R.; Fiore, M.F.; Reis, B.F. Development of a high sensitive automatic setup for screening of microcystins in surface waters by employing a LED-based photometric detector. Sens. Actuators B Chem. 2012, 161, 422–428. [CrossRef] Phung, S.L.; Bouzerdoum, A.; Chai, D. Skin segmentation using color pixel classification: Analysis and comparison. IEEE Trans. Pattern Anal. Mach. Intell. 2005, 27, 148–154. [CrossRef] [PubMed] Sural, S.; Qian, G.; Pramanik, S. Segmentation and histogram generation using the HSV color space for image retrieval. In Proceedings of the 2002 IEEE International Conference on Image Processing, Rochester, NY, USA, 22–25 September 2002; Volume 2, pp. II-589–II-592. Akbari, H.; Kalkhoran, H.M.; Fatemizadeh, E. A robust FCM algorithm for image segmentation based on spatial information and Total Variation. In Proceedings of the 9th IEEE Iranian Conference on Machine Vision and Image Processing (MVIP), Teheran, Iran, 17–19 November 2015; pp. 180–184. Lan, T.; Xiao, Z.; Hu, C.; Ding, Y.; Qin, Z. MRI brain image segmentation based on Kerneled FCM algorithm and using image filtering method. In Proceedings of the 2014 IEEE International Conference on Audio, Language and Image Processing (ICALIP), Shanghai, China, 7–9 July 2014; pp. 511–515. Song, X.; Li, G.; Luo, L. Optimizing FCM for segmentation of image using Gbest-guided artificial bee colony algorithm. In Proceedings of the 2015 11th IEEE International Conference on Natural Computation (ICNC), Hunan, China, 15–17 August 2015; pp. 764–768. Tang, D.-Y.; Yang, J.; Huang, Y.-S. Double weighted FCM algorithm for color image segmentation. In Proceedings of the 2012 IEEE International Conference on Machine Learning and Cybernetics (ICMLC), Xi’an, China, 15–17 July 2012; pp. 1135–1138. Wang, P.; Wang, H. A modified FCM algorithm for MRI brain image segmentation. In Proceedings of the 2008 IEEE International Seminar on Future BioMedical Information Engineering (FBIE’08), Wuhan, China, 18 December 2008; pp. 26–29. Dimitriadou, E.; Barth, M.; Windischberger, C.; Hornik, K.; Moser, E. A quantitative comparison of functional MRI cluster analysis. Artif. Intell. Med. 2004, 31, 57–71. [CrossRef] [PubMed] Jarmasz, M.; Somorjai, R.L. Exploring regions of interest with cluster analysis (EROICA) using a spectral peak statistic for selecting and testing the significance of fMRI activation time-series. Artif. Intell. Med. 2002, 25, 45–67. [CrossRef] Pizzi, N.J.; Vivanco, R.A.; Somorjai, R.L. EvIdent™: A functional magnetic resonance image analysis system. Artif. Intell. Med. 2001, 21, 263–269. [CrossRef] Somorjai, R.L. Exploratory data analysis in functional neuroimaging. Artif. Intell. Med. 2002, 25, 1–3. [CrossRef] Cheng, B.; Zhuo, L.; Zhang, J. Comparative Study on Dimensionality Reduction in Large-Scale Image Retrieval. In Proceedings of the 2013 IEEE International Symposium on Multimedia (ISM), Anaheim, CA, USA, 9–11 December 2013; pp. 445–450. Lei, L.; Wang, X.; Yang, B.; Peng, J. Image dimensionality reduction based on the HSV feature. In Proceedings of the 2010 9th IEEE International Conference on Cognitive Informatics (ICCI), Beijing, China, 7–9 July 2010; pp. 127–131. © 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).