Portable Multiplexed Optical Detection for Point-Of-Care

Home , CIE 1931 color space

PORTABLE MULTIPLEXED OPTICAL DETECTION FOR POINT-OF-CARE

A dissertation submitted to the Division of Research and Advanced Studies of the University of Cincinnati in partial fulfillment of the requirements for the degree of

DOCTOR OF PHILOSOPHY

in the School of Electronic and Computing Systems of the College of Engineering and Applied Science

2012

Li Shen

M.S., Southeast University, Nanjing, China, 2008 B.E., Southeast University, Nanjing, China, 2005

Committee Chair: Ian Papautsky, Ph.D.

ABSTRACT

In this dissertation, a low-cost, portable and user-friendly optical detection system was developed for microfluidic based lab-on-a-chip (LOC) devices. The conventional analytical methods of measuring these devices require expensive benchtop instruments that are not suitable for point-of-care (POC) applications. The optical detection system developed in this work consists of three major components: a broadband emission white LED to excite multiple fluorephores, a wavelength-independent cross-polarization signal isolation scheme, and a CMOS image sensor for signal detection. The combination of the setup enables simultaneous detection of multiple fluorescent samples. As a proof of concept, the system was applied for optical oxygen detection. The oxygen concentration was indicated by the red luminescence emission intensity of platinum octaethylporphyrin (PtOEP). The sensitivity of the oxygen sensor reached

~41, comparable to the ~50 values reported by others using an external spectrometer. To achieve multi-color analysis, CIE 1931 color space based signal conversion technique was developed, being intuitive and user-friendly. This technique was used for analyzing images of pH and urine glucose colorimetric test strips taken by a camera phone. The linear response ranges are 1-12 and 0-60mM for pH and urine glucose, respectively. Finally, the optical detection system was applied for fluorescent micro-particle detection, achieving ~3µm special resolution. In the particle counting test, 98% and 85% accuracy were achieved in static and dynamic conditions, respectively. With further development and optimization, the optical detection system can be integrated into microfluidic LOC systems for POC applications.

ACKNOWLEDGMENTS

First and foremost, I would like to thank my advisor Dr. Ian Papautsky for his guidance and constant encouragement during this work. He is always patient and ready to provide innovative ideas and suggestions when I get lost or stuck in my project. Besides technical help,

Dr. Papautsky has supported me by providing research assistantship as well as motivating and inspiring me all the time during my stay at University of Cincinnati. I feel gratified to be his student.

My gratitude is also expressed to my committee professors, Dr. Fred Beyette, Dr. Jason

Heikenfeld, Dr. William Heineman, and Dr. David Klotzkin for their insight and suggestions. I would like to thank Dr. Josh Hagen at Air Force Research Laboratory for supporting and providing valuable advice. I would also like to express my appreciation to Jeff Simkins and Ron

Flenniken for their help with cleanroom processing.

I would like to thank the Air Force Office of Scientific Research, the Human

Effectiveness Directorate, the National Science Foundation, and the University of Cincinnati

Institute for Nanoscale Science and Technology for funding my research during these years.

Many thanks are in order to my fellow students at the Bio Micro Systems Laboratory group: Ali, Preetha, Choi, Mike, Taher, Yun, Jian, Xing, Xiao, Wenjing, Yuguang, Ananda,

Nivedita, for their continuous support and friendship over the years.

Last but not the least; I would like to thank my parents Yang Shen and Huiqin Zhou, and my wife Wen Li, for being always together with me, supporting me, and loving me.

TABLE OF CONTENTS

LIST OF FIGURES ...... viii

LIST OF TABLES ...... xi

CHAPTER 1 INTRODUCTION ...... 1

Motivation ...... 3 Scope of work ...... 3 Innovation and significance ...... 5 Chapter Summaries ...... 5

CHAPTER 2 FABRICATION OF A PORTABLE FLUORESCENCE DETECTION

SYSTEM ...... 7

Signal isolation for lab-on-a-chip devices ...... 8 Signal detection for lab-on-a-chip devices ...... 14 System assembly and testing ...... 21 Summary ...... 25

CHAPTER 3 SPECTRAL AND SPATIAL RESOLUTION AND APPLICATION TO

GAS SENSING ...... 26

Luminescent oxygen sensing ...... 27 Optimization of the oxygen sensor using ratiometric measurement ...... 33 Summary ...... 36

CHAPTER 4 ON-SITE COLORIMETRIC DETECTION USING CAMERA PHONE ...... 37

Introduction ...... 38 Cell Phone as Photodetector ...... 40 Measurements of pH and glucose ...... 46 Ambient Light Compensation ...... 49 Summary ...... 52

CHAPTER 5 STATIC AND DYNAMIC DETECTION OF PARTICLES IN

MICROCHANNELS ...... 55

Introduction ...... 55 Experimental setup and methods ...... 56 Particle imaging and counting ...... 58 Summary ...... 67

CHAPTER 6 CONCLUSIONS ...... 69

Summary ...... 69

REFERENCES ...... 72

vii

LIST OF FIGURES

Figure Page

1. Schematic of the cross-polarization filtering scheme ...... 11

2. Transmission of two polarizers placed in parallel (0°), 75°, and crossed (90°) orientations. Insets illustrate isolation of optical signal (Rhodamine B emission centered at 625 nm) from high background signal ...... 12

3. Overall transmittances of the sandwich structures when one of the five materials was clamped between two orthogonally oriented polarizers. The inset describes the light path during the test...... 13

4. Schematic of a CMOS image sensor in which photodetector array is covered by Bayer filters ...... 16

5. Response of the CMOS sensor to 620nm light as compared to a current measured by a silicon photodiode. The gamma correction function of the CMOS sensor distorts the linear response ...... 19

6. Effect of AWB to the image color and the RGB intensities ...... 20

7. Responsivity of the CMOS detector in each of its color channels. Inset images taken by the CMOS detector illustrate response at center wavelengths 460 nm, 540 nm and 620 nm...... 21

8. Schematic of the CMOS image sensor based fluorescence detection system. A broadband white LED was used as excitation and two orthogonally oriented polarizers functioned as wavelength selection ...... 22

9. (a) Simultaneous detection of Fluorescein and Rhodamine B solutions under excitation of a single white LED. (b) Intensity profile of the emission intensities in red and green channels of the CMOS array ...... 23

10. (a) Image of 1 µM, 10 µM, and 100 µM Rhodamine B solutions in PDMS channels. (b) Three dimensional surface plot of the emission intensities of the corresponding solutions. (c) Red channel response of the CMOS array to a dilution series of Rhodamine B...... 24

viii

11. Arrangement of the light source (LED), detector (CMOS), polarizers, oxygen sensitive PtOEP film in the portable oxygen sensor ...... 27

12. Picture of the oxygen measurement setup ...... 28

13. (a) Image of the PtOEP film. (b) The PtOEP absorption and emission spectra overlaid with the green LED emission and the red channel response of the CMOS detector ...... 29

14. (a) Emission intensity of PtOEP as a function of oxygen concentration. (b) Stern- Volmer plot of the oxygen sensor. Inset images illustrate emission of PtOEP at low and high oxygen concentrations...... 30

15. Response of the oxygen sensor exposed to an alternating streams of oxygen and nitrogen gases. The arrows indicate the time when the gases were switched (filled arrows indicate nitrogen; hollow arrows indicate oxygen) ...... 31

16. Stability of the RGB channels in the CMOS detector of the oxygen sensor in ambient air ...... 32

17. (a) Emission of PtOEP is quenched by oxygen while emission of the reference Rhodamine B does not change. CMOS sensor measured intensities in red channel at (b) low oxygen and (c) high oxygen ...... 34

18. PtOEP and Rhodamine B emission at different oxygen concentrations under fluctuating LED bias. (a) Uncompensated and (b) compensated. Insets illustrate corresponding Stern-Volmer plots ...... 35

19. The portable setup for colorimetric measurement using a cell phone ...... 38

20. (a) Grey scale images for camera sensitivity test. (b) Three dimensional surface intensity plot of the grey scale image. (c) The cell phone camera response to grey scale images. The responses of a high end Canon EOS T3i SLR camera and an HP color scanner were plotted for comparison ...... 41

21. (a) Color response of pH strip at different pH values. (b) Three dimensional surface intensity plot of the pH colors. (c) RGB intensities of the pH strips’ color at each pH value. None of the red, green or blue intensities reflect the pH values ...... 43

22. (a) The pH strip colors in CIE 1931 color space. (b) The 3D view of the relation between xy coordinates and the pH values. (c) Calibration curve of the cell phone reader response to different pH buffer solutions...... 47

23. (a) The urine glucose test strip colors in CIE 1931 color space. (b) The 3D view of the relation between xy coordinates and the glucose concentrations. (c) Calibration curve of the cell phone reader response to urine samples with different glucose concentrations ...... 48

24. The linear relationship of the reference patches intensities between 5000K fluorescent light and (a) 3500K fluorescent light, (b) sunshine, (c) shade, and (d) smartphone LED. The color intensities of the sample can be substituted to the fitted curve to derive the corresponding intensities at which the device was calibrated (5000K) ...... 50

25. Shift in color space coordinates of the glucose test strip images due to different ambient (3500K and 5000K). The ambient light correction technique effectively compensates the error caused by ambient light ...... 53

26. Schematic of the multiplexed fluorescent particle detection system. (b) Size comparison of the device and a cell phone ...... 57

27. (a) Illustration of micro-particle (20µm green and 15µm orange) imaging. Two greyscale images taken by a conventional fluorescence microscope using TRITC and FITC filters are displayed as comparison. (b) Images taken by the portable device with different zoom levels. (c) Images taken by the microscope using 4X, 10X, and 20X objectives. Note that only green particles are in the images ...... 60

28. Micro-particle images captured by (a) portable system and (b) microscope. (c) (d) The intensity profiles along the marked line 1 and 2 in (a)(b) ...... 61

29. (a) A typical image of 20.6µm, 15.4µm, 9.9µm, and 6.1µm particles and the corresponding magnified images of each particle. (b) The counting result of particles in an image. Each peak represents a single particle ...... 62

30. The histogram of the particles’ statistics in six images. Each peak represents the number of particles of the size. Insets are a comparison of the software- measurement and the actual values ...... 63

31. (a) The typical intensity histograms of the orange, green and blue particles in RGB channels of the CMOS array. (b)The image taken by the system including orange, green and blue particles and the deconvoluted sub-images ...... 64

32. Distribution of blue, green and orange emission particles in CIE 1931 color space ...... 66

33. (a) Scatter plot of a particle mixture sample in the 3D space illustrating four populations: green particles (9.9 µm), green particles (15.4µm), orange particles (15.4 µm), and green particles (20.6 µm). (b) Bubble plot of the particles. The bubble size indicates the concentration of the corresponding particles ...... 66

LIST OF TABLES

Table Page

1. Summary of different signal isolation methods ...... 14

2. Summary of different detection methods ...... 17

CHAPTER 1

INTRODUCTION

Microfluidic analysis systems are widely acknowledged as enabling technology in point- of-care (POC) diagnostics. Being disposable and affordable, such microfluidic systems are ideally suited for applications in clinical diagnostics [1-3], global health [4, 5], and personalized medicine [6, 7]. One example is the diagnosis of HIV, which is based on counting the number of

CD4 cells (white blood cells expressing surface protein CD4). By applying microfluidic technology, only a small volume of whole blood sample (e.g., from lancet puncture) is needed for analysis, which eliminates the need for a phlebotomist and biological waste disposal [2].

HIV diagnostic platforms using a flow cytometer to count the number of CD4 cells is promising to be used outside well-equipped clinical laboratories and hospitals [8-10]. Although the existing microfluidic diagnostic devices are usually compact in size, they need external instruments to detect the tiny amount of analytes, which makes the lab-on-a-chip (LOC) device actually a

“chip-in-a-lab.”

Detection methods in microfluidics can be categorized into three major branches: electrochemical [11-13], mass spectrometric [14, 15], and optical [16-18]. Electrochemical assays have been successful in blood chemistry, urinalysis, and small-molecule analysis. A representative example is the handheld iSTAT blood analyzer, which is capable of quantitatively analyzing 25 blood parameters. However, the background interference with non-specific redox species, the electrode performance fluctuation at different temperatures, and biofouling are

1 inevitable concerns [19]. Mass spectrometry (MS) demonstrates high selectivity in monitoring the trajectory of ions in electric/magnetic fields but it has difficulties in distinguishing stereoisomers [20]. Optical detection has been widely applied to quantitative proteomic and genomic diagnostics due to the ubiquity of optical instrumentation in research centers, hospitals, and laboratories [19, 21, 22]. The detection methods are either based on directly measuring absorbance [23] or interference of light [24, 25], or indirectly measuring the intensity of fluorescence [26, 27], chemiluminescence [28, 29], or surface plasmon resonance (SPR) [30, 31].

In addition to the aforementioned three common detection methods, the use of nuclear magnetic resonance (NMR) spectroscopy [32], acoustical [33], and magnetoresistive [34] approaches have been reported in microfluidic systems. Ultimately, optical detection is acknowledged as a non- invasive and sensitive technique to be integrated into microfluidic systems. Moreover, the cost per-test is potentially to be very low since the electrodes are out of touch with analytes and not integrated into the disposable. With the fast development of optoelectronic components and numerous innovations in microfluidic integrations, optical detection is becoming more practical for POC diagnostics.

Among the various optical detection methods, fluorescence and colorimetry measurements are the most straightforward and widely used in both research and clinical diagnostics. Fluorescence detection is an indirect method in which analytes are usually labeled with fluorescent materials; analyte is measured based on the fluorescence intensity. Colorimetry, as a subset of absorbance detection, is carried out based on the color change of labeling dyes. A variety of instruments, such as spectrophotometers, fluorometers, microscopes, tristimulus colorimeters, are accessible in well-equipped laboratories and hospitals. Moreover, highly selective and sensitive labeling techniques are already well-established from genomic, proteomic,

2 and cell analysis.

Motivation

Although optical detection in microfluidic devices is well-established as benchtop technique, challenges still exist in applying it to POC where instrumentation is not available, such as developing countries, home testing, and harsh natural environments. The small POC devices are usually surrounded by several cumbersome analysis instruments, such as microscopes, photomultipliers (PMTs) or spectrometers. To realize a real portable LOC device, the power-consuming and expensive instruments must be replaced by battery-powered, low-cost components while maintaining decent sensitivity and reliability. The low-cost substitutes, such as LEDs, laser diodes, and photodetectors, give researches more choices to miniaturize the system.

The broad goal of this work is to develop a portable optical detection platform using

CMOS image sensor for fluorescence and colorimetric based POC diagnostics. This integration of optical detection with LOC concept offers the combined advantages of optical sensing and miniaturization. The detection system is portable, inexpensive, low power consumption, and reliable. Moreover, the CMOS image sensor used in the work can detect hundreds or thousands of simultaneous reaction events at once, where different locations correspond to the pixels on the detection array. Being able to achieve the “sample in – results out” test, the system is highly user-friendly and can be operated by non-trained personnel. Ultimately, the developed system can be envisioned as a multi-analyte sensor, capable for POC diagnostic applications.

Scope of work

In this dissertation, a portable optical platform was developed for POC. Specifically, fluorescence detection was demonstrated for quantifying the concentration of gas, fluorescent

3 dye, and microparticles; colorimetric based detection was demonstrated for measuring pH and glucose.

For fluorescence detection, the system design and the selection of the three key sensor components, including light source, signal isolation, and detector, focused on capability of multi- analyte detection. A broadband emission white LED was selected as light source to excite multiple fluorephores simultaneously. A wavelength-independent cross-polarization scheme, introduced by our group previously [16, 17], effectively filtered the excitation while maintaining a high transmittance to the signal light. For detection, a CMOS image sensor was used, in which millions of pixels detect fluorescent signals at different sites. As a result, the integrated system is capable of spectral and spatial detection. The performance of the system was validated with fluorescent dye detection, luminescent-based oxygen detection, and flow cytometry. In the above tests, the system achieved rapid and accurate measurement comparable to the results obtained using external instruments, which indicated future field applications, such as ELISA,

CD4+ counting, or environmental monitoring.

For colorimetric detection, the color response of pH and glucose test strips were quantified in CIE 1931 color space. A camera phone was used to capture the image of the strips and then image processing algorithms in Matlab (Mathworks, R2009b) enables one-click readout of the test results. Since the camera phone subscription is 6 billion as of the end of 2012 [35], it is feasible and valuable to explore the camera phone based diagnostics/telemedicine. With a quick snap and the following one-click readout of several key chemical levels in blood or urine, users are able to get an idea whether they should go to a hospital for more specific examination or they are fine. This rapid screening method not only helps the end users to keep tracking their health conditions but also saves the time of doctors, especially those who are facing lots of

4 patients in the developing areas.

Innovation and significance

This work demonstrates low-cost optical detection methods for microfluidic and paper microfluidic devices. Firstly, the high sensitivity achieved in oxygen detection indicates that the system can be applied to other intensity based applications for chemiluminescent immunoassay

[28] or capillary electrophoresis (CE) detection [36]. Secondly, the cell phone detection perfectly matches the low-cost paper microfluidic test strips and provides a fast, user-friendly health monitoring method. Using commercialized test strips, a number of key parameters can be measured, including glucose, bilirubin, pH, protein, ketone, specific gravity, urobilinogen nitrite, and leukocytes. Thus, people can easily track their health conditions using a cell phone/smart phone without going to the hospital. Further, for the first time, flow cytometry was demonstrated in a portable device with multiplexed capability using a low-cost CMOS array.

Chapter summaries

Following this introduction, Chapter 2 reviews various methods in portable optical detection systems for signal isolation and detection. The cross-polarization scheme and the

CMOS image sensor were picked and fully characterized. Chapter 3 describes the application of the developed system in luminescent oxygen sensing, in which the spectral and spatial resolution of the system was demonstrated. Chapter 4 introduces the colorimetric detection of pH and glucose using a camera phone. A light compensating method was explored and validated to calibrate errors caused by ambient light variation, which prevented the camera phone detection to be widely applicable. In chapter 5, the research extends to microparticle counting and flow cytomtery. Using the developed system, various microparticles with different size (6 µm-20 µm) and fluorescence (400 nm-620 nm) can be imaged in static or dynamic status. Finally, Chapter 6

5 summaries the work in this dissertation.

CHAPTER 2

FABRICATION OF A PORTABLE FLUORESCENCE DETECTION

SYSTEM

Fluorescence is the most commonly applied optical method for molecular sensing in microfluidic systems because of the well-developed instrumentation and techniques for conventional genomic and proteomic analyses that rely on highly selective and sensitive fluorescent labels. These optical detection systems typically require sophisticated benchtop instrumentation, such as spectrophotometers or lasers, and trained personnel. The need for portable and low-cost optical sensing has led to intense focus on miniaturization of sensor components.

A fluorescence detection system consists of three key components: light source, signal isolation, and detector. Fluorescent labels are typically excited by a laser due to the low divergence and narrow emission spectra band (0.5-3nm). However, the laser system is bulky and needs stable power supply, making it unsuitable for portable devices. The laser diodes, as an alternative, have been integrated into portable devices [37-40]. Furthermore, semiconductor and light-emitting diodes (LEDs) are more and more used for fluorescence excitation. Although

LEDs have relatively high divergence and broad emission, they are still gaining wide acceptance in portable devices because they are inexpensive and power saving. LEDs have been integrated in POC systems [19], immunosensors [41, 42] and chemical sensors [43, 44]. Signal isolation

7 and detection, however, remain active areas of research. Therefore, this chapter reviews the existing techniques and devices for these two components, as well as the advantages and shortcomings of each technique. Cross-polarization scheme and color CMOS image sensor were selected for signal isolation and detection, a combination that enables multiplexed fluorescence detection.

Signal isolation for lab-on-a-chip devices

Filtering of excitation light from signal light is critical in microscale fluorescent systems.

A common system design places an excitation LED orthogonal to the interrogation region and detector of the microfluidic system. This reduces system noise due to leakage of unabsorbed excitation light; however, such systems are bulky and complicated. The systems using a stacked arrangement where the light source is in-line with the interrogation region and detector is much more compact and preferable in portable systems. In such a system, the emission signal is drowned out by the leaking excitation light. An ideal filter blocks 100% of the excitation light and transmits 100% of the fluorescence signals, while maintaining a vertical absorption edge to the right of excitation wavelength and to the left of the signal spectrum. Such filter is not attainable in real devices because there are always non-vertical transitions from stopband to passband. A variety of techniques have been reported for filtering in these stacked microscale fluorescence systems, which are briefly reviewed below.

Interference filtering

The multi-layer interference filters were firstly used for LOC devices because they are traditionally used in macroscale spectrometers. These filters work as a mirror and reflect the unwanted light. Silicon dioxide, compound semiconductors, and metals can be used as filtering layers. The materials are deposited on the substrate using standard low temperature processes,

8 such as plasma enhanced chemical vapor deposition (PECVD) and molecular beam epitaxy

(MBE).

Burns et al. developed a highly integrated LOC device for DNA analysis using quarter- wave dielectric interference structure for signal isolation. The filter, deposited on a silicon photodiode, transmitted the fluorescence signal centered at 515 nm while rejected the excitation light under 500 nm. The LOC device consumes very little volumes of analyte and has an impressive limit of detection (LOD) of 10 ng/µL [45-47]. Other SiO2 or TiO2 based interference filter were also reported in the LOC systems and achieved 1 µM LOD for

Rhodamine 6G [48, 49]. Semiconductor based multi-layer interference filters were deposited by metalorganic chemical vapor deposition (MOCVD) on p-type AlGaAs substrate and achieved -

40dB extinction ratio [50-53]. The plasmonic interference filters are very promising in developing compact devices because different filter layers can be deposited on a single substrate

[54, 55].

The interference filters have several advantages. They can be easily integrated into microscale systems using standard low-temperature process and their cutoff wavelength is tunable with the layer arrangements. However, the cutoff wavelength is sensitive to the thickness of the layers: a few nanometers variation can cause up to 50 nm shift. Moreover, there are many difficulties in fabricating multiple filters on one substrate due to the non-uniformity when different etchants are used for the layer materials.

Absorption filtering

An absorption filter has a layer with strong absorption to the excitation light but relatively transparent to the emission signal light. Due to the single layer structure of the absorption filter, the fabrication is relatively simple compared to the multi-layer interference filters. Another

9 advantage of such filter is that they are independent of the incidence light angle. However, the working wavelength is determined by the optical property of the dye, making absorption filters less tunable to specific applications. There are two types of absorption materials: semiconductors and organic chromophores.

The semiconductor based bandgap filters have good absorption to photons with higher energy than their bandgap but pass the lower energy photons. Chediak et al. demonstrated a fluorometer device using heterogeneous CdS filters to detect green emission signals under blue excitation [56]. The CdS absorption filter achieved 60 dB rejection of the blue excitation (470 nm) and 4 dB (~40%) transmission of the green emission signal (513 nm). LOD of 0.12 µM has been reported in such device. Others have reported a poly-crystalline silicon thin film filter based fluorescence detection device with 35 dB signal to leakage excitation ratio [57]. The semiconductor filters have superior performance in filtering ultraviolet light.

Organic absorption filters rely on chromophore molecular to absorb light with specific wavelengths without emitting photons. By coating an 80 µm polycarbonate film onto an avalanche array detector, the fluorescence sensing device reaches an LOD of 25nM for fluorescein [58]. A reconfigurable filter array was developed by filling microfluidic channels with dye solutions [58-60]. The optical properties and concentrations of different dyes determine the cutoff wavelength and transmittance.

Cross-polarization filtering

The interference and absorption filtering approaches are wavelength-specific, making the system application-specific and far less versatile. In this work, signal isolation is achieved by cross-polarizing the excitation and emission light; an approach our group introduced recently [16,

17]. Two perpendicularly placed polarizers, with extinction ratio of 28 dB, are able to separate

Figure 1. Schematic of the cross-polarization filtering scheme. the emission light from the strong excitation light (Figure 1). Compared with using filters, this approach is inexpensive, easy to integrate, and is wavelength independent in the range of visible light.

The transmittance of two polarizers is tunable by adjusting the orientation angle. To demonstrate the signal isolation capability of the cross-polarization scheme, two pieces of polarizers were oriented at different angles (0°, 75°, and 90°) and the fluorescence of Rhodamine

B solution between them was captured using a regular point-and-shoot camera. Figure 2 illustrates the transmittance of two stacked polarizers in the range of 300 nm - 800 nm, measured using UV-Vis spectrophotometer (Varian, Cary 50/Bio). When the polarizers were oriented in parallel, ~40% of the light from 450 nm – 800 nm transmitted through the polarizers, resulting in a large background signal that masked the dye emission. At 75°, the transmittance dropped to ~7% and, consequently, the red emission of Rhodamine B became visible. When the polarizers were oriented orthogonally, the transmittance was below 0.04%, which was equivalent to an extinction ratio of ~28dB over the range of 300nm-730nm. As a result, the emission signal was separated from the strong excitation light. However, the actual extinction ratio was ~24dB in our applications because the existence of samples slightly changed the polarization of excitation light and caused higher leakage intensity. Thus, the optical properties of the sample, including

Figure 2. Transmission of two polarizers placed in parallel (0°), 75°, and crossed (90°) orientations. Insets illustrate isolation of optical signal (Rhodamine B emission centered at 625 nm) from high background signal. transmission and polarization, are worthwhile to be explored.

The cross-polarization signal isolation scheme is compatible with a variety of materials used in fabrication of LOC devices. I chose glass, PDMS, PMMA, COC, and polypropylene because they are widely used in prototype device fabrications or mass produce processes, such as injection molding and roll-to-roll processing. Moreover, these materials exhibited transparent windows in visible light from 400nm–800nm, while the average transmittance was above 90% so as not to attenuate the signal intensity. Figure 3 illustrates the overall transmittance of two cross- oriented polarizers with a piece of test material between them. When glass, PDMS, PMMA, and

COC was inserted, the transmittance did not increase compared to only polarizers which means they did not break the polarization and were fully compatible with the cross-polarization scheme.

Polypropylene slightly increased the transmittance by 1%. Considering the ~90% transmittance of polypropylene, the theoretical S/N is 19dB which is still practical for applications not exhibiting extremely low signal intensity. Overall, this cross-polarization scheme, providing a low noise window from 200nm-730nm for detection of sample emissions, is widely applicable in

12 existing microfluidic or medical devices.

Table 1 compares the signal isolation methods. The interference filters have highest selective over the excitation but they need MBE in fabrication to achieve accurate cutoff wavelength control. The absorption filter is relatively simple in fabrication and low in cost. The reconfigurable absorption based on microfluidic structure makes this method very versatile to different applications. However, these two methods are dependent on wavelength and cannot be

Figure 3. Overall transmittances of the sandwich structures when one of the five materials was clamped between two orthogonally oriented polarizers. The inset describes the light path during the test. used for simultaneous detection of multiple analytes with different absorption and emission spectra. The cross-polarization method, on the other hand, filters excitation light based on the polarization and works for visible light from 400nm – 730 nm. Although the signal to noise selectivity is relatively lower than the other two methods, the device based on polarization still reached 100 nM LOD for Rhodamine 6G [16].

Table 1. Summary of different signal isolation methods.

Interference Absorption Polarization

Fabrication Complex Simple Simple

Cost High Medium Low

Selectivity 40dB [50-53] 35dB [57] 28dB [16, 17]

LOD 1µM [48, 49] 25nM [58] 100nM [16, 17]

Multi-analyte No No Yes detection

Signal detection for lab-on-a-chip devices

A sensitive and reliable detector is critical to a portable sensor. Photomultiplier tubes

(PMTs) have been used as detectors in microfluidic systems for a long time. As a portable and low-cost alternative, photodiodes are gaining popularity, especially in the battery powered LOC systems. Finally, with the rapid development of semiconductor industry, the multi-pixel detectors, such as charge-coupled devices (CCDs) or CMOS arrays, are tending to gain interest.

Conventional photomultiplier tubes have been used for benchtop microfluidic systems

[36, 39]. A typical example was reported by Easley et al. for polymerase chain reaction (PCR) and CE detection. Sample DNA/RNA preparation and purification from whole blood and nasal

14 aspirate were integrated and only 10 µL sample solution is required [36]. However, the system employed an external high power laser and PMTs for fluorescence detection. Liu et al. reported oral swab and human bone extract using the developed system in which on-chip PCR and CE detection can be achieved [39]. Four dichroic filters and PMTs were assembled in the system for four-color fluorescence detection. The PMTs enables extremely low light detection in picomolar assays with a limit of detection down to single photon [26]. However, the price of PMTs is generally expensive and stable high voltage power supplies are necessary for the PMTs to work.

Thus, the PMT based detection system can be helpful in some circumstances but there is still a lot of work to do to miniaturize the size and lower the high energy consumption.

Avalanche photodiodes (APDs), electronically equivalent to PMTs, have been recently reported in microfluidics systems [27, 61]. Wu et al. demonstrated an LED-APD based LOC device for fluorescence detection and achieved an LOD of 0.2 nM for fluorescein [27]. Silicon photodiodes [62, 63] have also been used in POC systems due to low cost. Tu et al. has developed an optical immunosensors for human serum albumin (HAS) detection in which

CdSe/ZnS quantum dots (QDs) were used to label the analytes [64]. The system has achieved

3.2×10-5 mg/ml LOD when measuring QDs labeled HSA. Moreover, Pimentel et al. fabricated a silicon p-i-n photodiode for DNA measurement and achieved an LOD of 5 nM [65]. In the past, we [16, 17] and others [66, 67] have used organic photodiodes (OPDs) as low-cost integrated detection alternatives, being flexible and miniaturizable [68]. The LOD of the devices using

OPDs is usually in the range of 100 nM – 10 µM. Although OPDs are flexible and low-cost, they suffer from low power conversion efficiency [69] and short lifetime [70, 71]. Integration of these external components, however, presents alignment and assembly challenges.

The CCDs are generally used for image acquiring in microscope cameras or scanners.

Recently, lots of success examples demonstrate that these image sensors perform well in a LOC system as detector. Being able to detect thousands of events simultaneously, these image sensors are good for multiplexed fluorescence detection. Ymeti et al. developed a Young interferometer based immunosensors for HSV-1 virus detection [25]. A CCD camera was used to collect the interference patterns and references. The HSV-1 virus particles can be specifically detected at a very low concentration of 850 particles/ml. Another prototype device using CCD was developed and enhanced with surface plasmon resonance (SPR) for rapid bio-detections [72, 73]. The LOD of an analytical system developed by Yu et al. has reached 47 nM (3µg/L) for copper [74].

Recently, CMOS image sensors have attracted interest as detectors for POC systems because of the high-speed readout and low power consumption. Gamal and Eltoukhy [75] recently reviewed developments in CMOS image sensors and suggested applications to LOC as one of the future directions. CMOS image sensors, commonly used today in digital cameras, are built using the same process as ICs and consist of an array of photodetectors covered by a mosaic of microscale band-pass (Bayer) filters (Figure 4). Raw data taken by a CMOS image sensor are grids of three primary colors: red, green and blue (RGB). Each pixel contains the intensity information of the corresponding color. The intensity of each color is described within a given range from a minimum to a maximum which are dependent on the resolution of the analog to

Figure 4. Schematic of a CMOS image sensor in which photodetector array is covered by Bayer filters.

16 digital converter. This approach of using CMOS detector as detector has been successfully applied for bioassays [76] and fluorescence-based cell detection [77-80] and the LOD can be as low as 10pM [18].

Table 2 compares different detectors. The PMTs and APDs are very sensitive to even extreme low light intensity but their costs are high. The Si PDs and OPDs are lower in cost with moderate sensitivity. However, there are difficulties in alignment of the detector and the sample assay. The image sensors (CCD and CMOS array) provide multi-analyte detection capability and are low costs while maintaining good sensitivity.

The choice to use a CMOS image detector as opposed to a CCD was based on the sensor needs [75]. CCD image sensors have better low-light performance. However, since most camera control functions need to be integrated off-chip, the CCD-based detection system

Table 2. Summary of different detection methods.

Detector Cost LOD Multi-analyte detection

PMT, APD High 0.2 nM [27] No

Si PD Low 5 nM DNA [65] No

OPD Medium 100 nM [16] No

CCD High 47 nM [74] Yes

CMOS Low 10 pM [18] Yes

17 can increase very rapidly in size and cost. The CMOS array integrates sensing with A/D converter and gain amplifier down to pixel level, making it customizable for a particular application [75]. Although these features make CMOS sensors popular with the consumer market, their impact must be fully understood for application to the light intensity-modulated optical sensing. Thus, in the next set of experiments we manually controlled these functions (by writing the registers through the standard serial camera control bus interface) to investigate capabilities of a consumer CMOS sensor.

In photography, gamma correction is used to compensate non-linear decoding process of displays by encoding the signal inversely. The gamma correction is given by

(1)

where Iout is the displayed intensity, Iin is the detected intensity, and the exponent γ is the gamma correction index. The Imax is the upper limit which depends on the color depth of the image sensor. The OV9810 is an 8-bit color image sensor so its Imax is 255. The default value of γ is

1/2.2 which makes the response of the CMOS detector non-linear to the light intensity [81].

Turning off the gamma correction by setting γ = 1 leads to the CMOS detector exhibiting perfect linearity satisfying a fundamental requirement as a detector. Figure 5 illustrates the red channel response to monochromatic 620 nm light generated by biasing LED from 2.6 V to 2.88 V to yield different intensities directly measured by a silicon photodiode. The blue and green channels were tested in the same way and they demonstrated good linearity as well. The effect of gamma correction is clearly evident. As these results indicate, gamma correction must be turned off for the detector to yield a linear response over the entire sensor range. However,

Figure 5. Response of the CMOS sensor to 620nm light as compared to a current measured by a silicon photodiode. The gamma correction function of the CMOS sensor distorts the linear response. gamma could provide additional signal amplification for low-light conditions if sensor range is limited.

Auto white balance (AWB) is another integrated CMOS feature that is commonly found in consumer cameras but can have substantial effect on the CMOS spectral response. AWB is designed to provide good color reproduction by amplifying the detected RGB signals at different ratios. Several aspects in image acquisition and signal processing make such color correction essential – including a mismatch in viewing conditions of acquisition and display, or fundamental differences between the human eye and the acquisition sensor [81]. Figure 6 illustrates images of three primary colors taken by the CMOS detector. A monochromator was used to select light at 460 nm, 540 nm and 620 nm, corresponding to the response peaks of RGB channels of the CMOS detector. When AWB is off, the RGB intensity is dominated by the corresponding color. The small quantity of response in other two channels is due to the imperfect filtration efficiency and spectral overlap of the Bayer filters. When activated, the

AWB increases the digital gains in the three channels separately, especially the non-dominated

Figure 6. Effect of AWB to the image color and the RGB intensities. colors. It makes the images bright and pleasing to our eyes. However, the deviation of the RGB values from the real values following the auto-adjustment skews the measurement from being accurate when the CMOS detector is used as a sensor.

The spectral response of the CMOS chip RGB channels was tested to examine the color discriminating capability. We used a monochrometer to select wavelength of illumination light and recorded the spectral response in the 430-680 nm range in 10 nm step increments. As illustrated in Figure 7, each channel of the CMOS detector exhibits a strong response to the corresponding wavelength range, as expected. The blue pixels exhibit a peak response at 460 nm

+/-30 nm, while the green and red pixels peak at 540 nm +/-40 nm and 600 nm +/-50 nm respectively. This test confirms that the CMOS detector is inherently discriminating to the RGB colors and demonstrates the spectral response range of each color pixel type. While Bayer filters permit color discrimination in the CMOS detector, response of each color pixel type does overlap. In a consumer camera this is desirable as the overlap permits capture of colors that fall between the standard RGB bins, such as yellow at 580 nm. In sensor applications, however, when individual color channels in the CMOS detector are used to distinguish between the excitation and emission of analyte, the spectral overlap leads to noise. For example, a sensor based on detection of fluorescein isothiocyanate (FITC) – a fluorescent dye commonly used in

Figure 7. Responsivity of the CMOS detector in each of its color channels. Inset images taken by the CMOS detector illustrate response at center wavelengths 460 nm, 540 nm and 620 nm. bioapplications – will detect emission signal in the green channel (521 nm) and unabsorbed excitation light (488 nm) in the blue channel. The overlap between the two channels will lead to high background noise as a fraction of the excitation light will be picked-up by the green channel.

These results are helpful in the development of sensors that take advantage of the CMOS detector’s color discriminating capability.

System assembly and testing

The schematic diagram of the fluorescence detection system is illustrated in Figure 8, with a broadband white LED light source (Bivar, R20WHT-F-0160), two polarizers (Edmund

Optics, NT45667), and a CMOS image sensor connected with a low cost 20X objective. The

CMOS image sensor with a USB interface (Aptina, MT9P031) has a 5.70mm × 4.28mm imaging area (optical format 1/2.5 in). The active array size was 2592(H) × 1944(V) pixels (5 megapixels), each approximately 2.2µm × 2.2µm. The objective with RMS threads was connected to the C-mount on the board of the CMOS array via an adapter (Thorlabs, RMSA5), then placed at the bottom of the device stack as detector. The white LED light source was biased

Figure 8. Schematic of the CMOS image sensor based fluorescence detection system. A broadband white LED was used as excitation and two orthogonally oriented polarizers functioned as wavelength selection. at 3.0 V (~16 mA forward current). The excitation light was linearly polarized after passing the first polarizer and then blocked by the second orthogonally oriented polarizer while the emission light with random polarization can pass through the second polarizer and be collected by the detector. Samples dye solutions were loaded into polydimethylsiloxane (PDMS) channels bonded on a 1” × 3” glass slide placed on the sample holder plane.

The multi-color detection capability of the portable system was demonstrated by imaging the fluorescence of Fluorescein and Rhodamine B. Figure 9(a) illustrates the fluorescence of

1mM Fluorescein and Rhodamine B solution in 1mm (W) × 70µm (H) PDMS channels. The white LED emission excited both dyes and was adequately filtered by the polarizers. The sharp transitions in the intensity profiles in Figure 9(b) indicate no strong cross talk between the two dye solutions. Moreover, the dark background and the low intensity level indicate that the polarizers effectively block the broadband white LED excitation. Although the color of the fluorescein emission was dominated by the green channel, there was still relatively weak

22 response in the red channel due to the emission spectra of dyes having distributions over a range that extends to all color channels. In general, the intensities in the main color channel reflect the dye concentration of the solution.

Quantitative measurement of the CMOS array response to dye concentration from 1µM to 100µM was performed using a dilution series of Rhodamine B. Figure 10(a) illustrates the fluorescence of 1µM, 10µM, and 100µM Rhodamine B solutions with increased emission. The emission uniformity can be observed in the three dimensional surface plot in Figure 10(b) which shows the intensities detected by each pixel. Unlike the single pixel photo diodes, misalignment between channels and detector does not cause significant error because the detection is based on the average intensities of a large amount of pixels. Moreover, any high noise pulse caused by dusts can be excluded when selecting the analysis area. Figure 10(c) illustrates the red channel

Figure 9. (a) Simultaneous detection of Fluorescein and Rhodamine B solutions under excitation of a single white LED. (b) Intensity profile of the emission intensities in red and green channels of the CMOS array.

Figure 10. (a) Image of 1 µM, 10 µM, and 100 µM Rhodamine B solutions in PDMS channels. (b) Three dimensional surface plot of the emission intensities of the corresponding solutions. (c) Red channel response of the CMOS array to a dilution series of Rhodamine B. intensities of the dilution series of Rhodamine B. The measured intensities were proportional to the dye concentration which is consistent with the Beer-Lambert’s law at low concentration. The dotted line illustrates the background level measured with only water in channel. The S/N ratio was estimated to be 1.1 at 1µM to 3.1 at 100µM concentration. The LOD can be optimized by using a CMOS array with larger pixels since the sensitivity is quadratically proportional to the pixel size.

Summary

Fluorescence detection is an important analytical tool in different areas, including biological research and environmental monitoring. The need of POC for first responders has led to intense focus on miniaturization of portable detection systems. Among the various choices of light filters and detectors, the cross-polarization and CMOS image sensor has been chosen to build the portable system. Both polarizers and CMOS sensors are inexpensive, small in size, and easy to integrate. Most importantly, the cross-polarization scheme effectively filters light even if the excitation and emission overlap. The extinction ratio is 28 dB in the range of 400nm –

730nm. Thus, a broadband light source, such a white LED or a halogen lamp, can be used to excite multiple lumophores simultaneously. Then the CMOS array, covered by color discriminating Bayer filters, is used to collect the emission light and achieved multi-color fluorescence detection. The LOD is 1 µM for the system. The combination of the wavelength independent cross-polarization scheme and the color discriminating CMOS array provides excellent spectral resolution, as well as spatial resolution, which enables multiplexed fluorescence detection. Although the LOD is still higher than the nanomolar LOD achieved using conventional photodiodes, the performance of CMOS based sensing systems is improving steadily and may soon become cost-effective alternative detectors for LOC applications.

CHAPTER 3

SPECTRAL AND SPATIAL RESOLUTION AND APPLICATION TO

GAS SENSING

The significance of oxygen in different areas, such as clinical diagnosis [82], biological research [83] and environmental monitoring [84], has led to high interest in sensors for measuring oxygen concentration. Oxygen is typically measured using an electrochemical Clark electrode which is easily calibrated and requires relatively inexpensive instrumentation.

However, Clark electrodes consume oxygen and are easily poisoned by various organic compounds. Moreover, they suffer from biofouling, which is a potentially fatal shortcoming in biomedical applications [85]. Given these limitations, much research has been focused on optical methods of measuring oxygen. Such sensors are typically based on quenching of luminescence emitted by oxygen-sensitive dyes [86]. Compared with electrochemical sensors, optical oxygen sensors offer quick response time, no oxygen consumption, and high sensitivity

[87-90]. Despite these virtues, the need for cumbersome and expensive instrument components, such as spectrophotometers or photomultiplier tubes, impede application of such sensors in cost- sensitive, compact, or in vivo environments. The need for portable and low-cost oxygen sensors has led to intense focus on miniaturization of sensor components. Thus, the CMOS image sensor based detection system fits the requirement very well. By applying the developed system in

Chapter 1 to oxygen detection, the light sensitivity and the spatial resolution of this approach has

26 been demonstrated.

Luminescent oxygen sensing

In the test, a CMOS image sensor with a USB interface (OmniVision, OV9810) was used.

The sensor has a packaged footprint of 8.195 mm × 7.535 mm, with 6.160 mm × 4.606 mm imaging area. The active array size was 3488 × 2616 pixels (9 megapixel resolution), each approximately 1.75 μm × 1.75 μm. The sensor was placed at the bottom of the device stack as the detector and was controlled by the supplied software. Captured raw images were split to three grayscale images (each representing a separate RGB color) and quantitatively analyzed using ImageJ (ver. 1.43u).

The sensor is based on the oxygen sensitive platinum octaethylporphyrin (PtOEP) encapsulated in ethyl cellulose (EC). A green LED (Bivar R20GRN-F-0160) with a central emission wavelength of λex=520 nm was used as the excitation light source (Figure 11). The excitation light was linearly polarized by passing through polarizer 1 (NT45667, Edmund

Optics). The red emission of the PtOEP film (λem=644 nm) passed through polarizer 2. The

Figure 11. Arrangement of the light source (LED), detector (CMOS), polarizers, oxygen sensitive PtOEP film in the portable oxygen sensor.

27 polarized excitation light not absorbed by the PtOEP film was blocked by polarizer 2. When cross-polarized, the extinction ratio of excitation light to leakage light is 28dB [17]. The oxygen sensitive film was prepared by completely dissolving 10 mg of PtOEP (PTO534, Frontier

Scientific) in 10 mL tetrahydrofuran. The mixture was added to a polymer solution prepared by dissolving 3 g EC (48% ethoxyl) in the mixture of toluene (24 mL) and ethanol (6 mL).

For oxygen measurements, all automatic CMOS image sensor functions, such as exposure time control, gamma correction, and auto white balance (AWB), were set to manual.

Figure 12 illustrates the experimental setup. Two flow meters controlled the ratio of oxygen and nitrogen supplied from the gas tanks. A mixing chamber was used to pre-mix the gases prior to introduction to the testing chamber. The sensor located in the center of the testing chamber; the chamber oxygen concentration was indicated by a commercial oxygen sensor (Nuvair, Pro O2 remote analyzer) as reference. During measurements, oxygen concentration in the chamber was changed by adjusting input gas flow rates. Signals and background were obtained by taking images of the PtOEP film and a bare glass wafer. The values in the red channel were used for

Figure 12. Picture of the oxygen measurement setup.

28 oxygen measurement. By subtracting the background from the signal, the emission intensity of

PtOEP at different oxygen concentration was obtained.

The green LED light source and the red channel in the CMOS sensor were selected according to the absorption and emission spectra of the PtOEP film, responsively. The absorption and emission of

PtOEP are centered at 535 nm and 640 nm when encapsulated in EC [91]. As Figure 13 illustrates, there is a significant overlap between the green LED emission (520nm) and the absorption of PtOEP. The red channel, which has strong response to red light, was used to detect the PtOEP emission. The response of red channel cuts off at around 580 nm, resulting in further filtering of the excitation light. Moreover, the green channel responds to the LED emission well but does not respond to the intensity variation of the red PtOEP emission. Thus, it is possible to monitor the intensity of the light source using the green channel and calibrate the measured result accordingly.

This oxygen sensor based on CMOS detector exhibited high sensitivity as well as good linearity in the Stern-Volmer analysis. Figure 14(a) illustrates the steady state emission intensities of PtOEP sensor subject to different oxygen levels. The red pixel values decrease

Figure 13. (a) Image of the PtOEP film. (b) The PtOEP absorption and emission spectra overlaid with the green LED emission and the red channel response of the CMOS detector.

29 significantly with the increasing oxygen concentration as the quenching effect becomes stronger.

The intensities in green and blue channels are much more stable because these two channels are inert to the red emission of PtOEP which varies with the oxygen concentration. The standard approach to characterizing luminescence quenching based oxygen sensors is using Stern-Volmer analysis in which,

[ ] (2)

where I0 and I are the emission intensities in the absence and presence of oxygen at concentration of [O2], respectively, and KSV is Stern-Volmer constant. Figure 14(b) illustrates a typical Stern-

Volmer plot of the developed oxygen sensor. The high correlation coefficient (0.9982) indicates a linear relationship between the oxygen concentration and the ratio of emission intensities in the absence (I0) and presence (I100) of oxygen. The ratio I0/I100 as sensitivity of the oxygen sensor is estimated to be ~41. This result is comparable to the ~50 values reported by others using an external spectrophotometer [87, 89] and is much higher than that of an integrated lab-on-a-chip

Figure 14. (a) Emission intensity of PtOEP as a function of oxygen concentration. (b) Stern- Volmer plot of the oxygen sensor. Inset images illustrate emission of PtOEP at low and high oxygen concentrations.

30 sensor (~1.4) reported recently [92]. The insets in Figure 14(b) illustrate images taken at low and high oxygen concentrations. As emission of PtOEP is strongly quenched by oxygen, little emission light is observed and the corresponding image was dark (right inset). In low oxygen conditions, however, the emission of PtOEP film increases and the image becomes red.

The oxygen sensor exhibited reversible quenching and good repeatability when exposed to an alternating atmosphere of oxygen and nitrogen. A typical dynamic response of the sensor is illustrated in Figure 15. The response time of an optical oxygen sensor is considered as the time of 90% total intensity change when switching between oxygen and nitrogen conditions.

The response time of the oxygen sensor in this work was ~4 s upon switching from nitrogen to oxygen and the reverse switch time, from oxygen to nitrogen, was ~80 s. The asymmetric response curve with a slow O2 to N2 response compared to N2 to O2 response is typical of the luminescent thin film oxygen sensors [93]. The response is comparable to the reported sensor using polystyrene as lumophore-encapsulating matrix (18 s / 60 s) [87]. We believe the response time is limited by the mass-transport in the gas test chamber, which can take 22-60 seconds to fill,

Figure 15. Response of the oxygen sensor exposed to an alternating streams of oxygen and nitrogen gases. The arrows indicate the time when the gases were switched (filled arrows indicate nitrogen; hollow arrows indicate oxygen).

31 depending on input gas flow rate (1-8 liters per minute). Reducing the test chamber volume will improve sensor response time in gas switching test. A number of methods may also be used to improve the film response time, such as using Pt porphyrins bearing side groups designed to increase solubility at higher concentration in a thinner layer [94] or use a material with high diffusivity to oxygen [89].

The oxygen sensor demonstrated excellent reliability after 30 hours exposure to a green

LED in ambient air (21% oxygen condition). Figure 16 illustrates intensities of the RGB channels during the test. At 21% O2, emission of PtOEP was not strong but very stable during the 1800-min test. The signals fluctuated slightly initially but stabilized during the first ten minutes. The fluctuation may be caused by the LED or the power supply. This result suggests that a ten-minute warm-up time is necessary for the system to reach a stable condition for accurate measurement. After the warm-up, the mean value of the intensities in red channels is

16.5 with a standard deviation of 0.33, resulting in a coefficient of variation of only 0.02. Based on these results, the oxygen sensor is reliable and can be used for continuous oxygen sensing

Figure 16. Stability of the RGB channels in the CMOS detector of the oxygen sensor in ambient air.

32 applications.

Optimization of the oxygen sensor using ratiometric measurement

The oxygen sensor exhibited high sensitivity when the light source was powered by the stable external power supply. However, such a power supply is not always available outside the laboratory and batteries are used in most portable devices. While batteries are much smaller in size and are light in weight, their output decreases over time with use or even due to storage.

The lower driving voltage causes lower light source emission; as a result, the PtOEP emission is a function of the oxygen concentration and the light source intensity. Thus, an oxygen insensitive reference should be integrated to monitor the light source intensity. In the next experiment, Rhodamine B was integrated as a reference. It was sealed in a glass capillary to enhance its stability. In the test for the sensor performance under an unstable light source, the forward voltage of the LED was manually varied by ±0.02 V during oxygen measurements. The emission intensities of the Rhodamine B reference were standardized and the emission intensities of the PtOEP were calibrated accordingly.

The Rhodamine B emission does not change with the oxygen while the PtOEP emission is sensitive to oxygen. The intensity profiles of two images taken at high and low oxygen are illustrated in Figure 17(a). The PtOEP emission decreases significantly from 190 to 20 a.u. as expected when the oxygen concentration increases. On the other hand, the emission of

Rhodamine B remained constant. The sensor images in Figure 17(b) and (c) clearly illustrate the quenching effect of oxygen on PtOEP. The black tape between the PtOEP and Rhodamine B blocks the bleeding light from PtOEP at low oxygen concentrations and thus enhances the stability of the Rhodamine B reference signal.

The integrated reference makes the oxygen sensor stable and accurate because any

33 fluctuation of the power supply can be observed from the emission intensity of Rhodamine B and then negated. To demonstrate this, the LED bias was intentionally varied by ±0.02V, less than

1%, during oxygen measurement. The oxygen sensor was first characterized with LED biased at

3.0 V using a Keithley 6487 voltage source. Figure 18(a) illustrates the resulting Rhodamine B and PtOEP emission intensities which deviate from the calibration curve (dotted lines). After the

Rhodamine B intensities were compensated to the calibration curve, the PtOEP intensities were calibrated using the same ratio (Figure 18(b)). The sensor with an integrated reference

Figure 17. (a) Emission of PtOEP is quenched by oxygen while emission of the reference Rhodamine B does not change. CMOS sensor measured intensities in red channel at (b) low oxygen and (c) high oxygen.

Figure 18. PtOEP and Rhodamine B emission at different oxygen concentrations under fluctuating LED bias. (a) Uncompensated and (b) compensated. Insets illustrate corresponding Stern-Volmer plots.

35 maintains good linearity in response even if the power supply is not stable. The inset images in

Figure 18 illustrate the Stern-Volmer plots based on the calibrated PtOEP intensities. The coefficient of determination (R2) of 0.88 demonstrates the good linearity of the sensor. Without the Rhodamine B reference, the R2 decreases to 0.54 under less than 1% of power supply output fluctuation.

Summary

In this chapter, a portable optical oxygen sensor using a low-cost CMOS image sensor as detector and a cross-polarization scheme as signal isolation was demonstrated. The sensor exhibited sensitivities comparable to that of macroscale benchtop sensor systems. The integrated oxygen insensitive reference was monitored simultaneously by the CMOS sensor, permitting compensation when the light source is not stable. Overall, the described approach of using the wavelength-independent polarizer scheme and color-recognizable CMOS image sensor clearly demonstrate suitability for reliable, on-site sensor applications.

CHAPTER 4

ON-SITE COLORIMETRIC DETECTION USING CAMERA PHONE

This chapter introduces a novel approach of using a smartphone to quantify colors of colorimetric diagnostic sensors. The RGB intensities of the color image taken by a smartphone camera were converted to chromaticity values and used to construct calibration curves of analyte concentrations. A reference test strip with seven greyscale and five color reference segments was designed to stabilize the auto-white balance (AWB) of the smartphone camera and to compensate the measurement errors caused by variation in ambient light. We demonstrated applications of the approach in multiple colorimetric sensors, including measurements of pH and urine glucose, and achieved good accuracy (R2>0.99). The linear response ranges are 1-12 and

0-60mM for pH and urine glucose, respectively. These results are comparable to those of reported devices using a desktop scanner or silicon photodetectors. The ambient light compensating technique enables self-calibration at different light sources, such as sun light, fluorescent light bulb, or smartphone LED. Finally, the developed algorithm, including RGB conversion function, concentration calibration and measurement, and ambient light compensation, enables user friendly one-click reading, which makes the approach operable without any professional training or complex instrumentation.

Introduction

Since the introduction of paper-based microfluidic analytical devices [95, 96], much progress has been made over the past several years towards inexpensive and quick POC assays, including HIV microfluidic chips [97, 98], paper based ELISA assays [99, 100], and other low- cost colorimetric diagnostic assays [101-103]. These paper microfluidic assays are gaining popularity as a simple and fast way of disease screening in resource limited environments [104,

105]. Although the colorimetric results of these assays can be viewed by naked eyes, the amount of the analyte cannot be precisely quantified [106]. Promising colorimetric detection results have been demonstrated using video cameras [107], digital color analyzers [108], scanners [109] or custom portable readers [102], although the acquired images must be transferred to a computer for image processing and analysis. These methods all require specific instrumentation for analysis, whereas smartphones and tablets are becoming ubiquitous.

Recent rapid developments in hardware and software of smartphones and tablets have led to a potentially new and exciting approach for quantitative POC detection. Several investigators

Figure 19. The portable setup for colorimetric measurement using a cell phone.

38 have recently successfully demonstrated the use of camera phones for onsite diagnoses in dermatology [110] and ophthalmology [111]. Figure 19 illustrates the basic components for colorimetric POC analysis: a camera phone, colorimetric test strips, and a specially designed reference paper. Though cell phones have been applied to colorimetric diagnostic detection

[112-114], the demonstrated detection only works for single color assays. Therefore, the capability of taking color images of the camera phones has not yet been fully exploited, which limits the application in commercially available colorimetric paper strips for pH, glucose, protein, to name a few.

This work introduces a novel approach of using the International Commission on

Illumination (CIE) 1931 color space to quantify multiple colorimetric paper test strips. We demonstrate that chromaticity based color quantification combined with a conventional cell phone for instant reading of multiple elements in colorimetric diagnostics is feasible in POC applications. Moreover, using our newly developed light compensating technique, measurement errors caused by the variation in ambient light and camera to camera differences can be compensated with high accuracy. Cell phones have yet to gain popularity for colorimetric detection due to three key challenges. First, integrated color balancing functions of a conventional cell phone are optimized for photography in high ambient light, and are not suited for images when accurate quantitative measurements must be performed. Second, lighting conditions during imaging can be difficult to control, especially outside of controlled environments such as a laboratory. Third, analysis of images can be difficult especially when small color changes are present, and RGB values alone are not necessarily sufficient. Herein, we show that these challenges can all be addressed, as we have demonstrated using simple model assays – pH paper and urine glucose test strip. Thus, a smartphone can provide a viable and

39 simple solution to quantitative POC analysis.

Cell Phone as Photodetector

Mainstream smartphones, including iPhones, Android phones, or Windows phones, are equipped with a dual-core 1-1.5Ghz processor. In our proposed application, the light data processing load is accessible by any entry level smart phone equipped with a single core 600

MHz processor. While the processing ability is not the bottle neck of the application, we must consider adaptability and durability. Since Android is an open source operating system, manufactures have more space in choosing hardware and the layout design. The Casio Gz’One

Commando is a good example. It is equipped with an 800 MHz processor and a 5-megapixel camera. More importantly, it has passed eleven MIL-STD-810G tests, including drop, immersion, temperature, etc., showing operability in tough environments for first responders or military applications.

We first evaluated the sensitivity of the smartphone camera to assess its suitability as photodetector and compared its performance with other more expensive detectors [109, 115].

We took images of 10 greyscale standards (Labsphere) and then compare the intensities in each image. In grey scale, intensity of zero is considered as black while the maximum value (255 for the Commando whose color depth is 8bit) is white, as is illustrated in Figure 20(a). A desktop scanner (HP Scanjet N6310 document flatbed scanner) and an SLR camera (Canon EOS T3i) with higher color depth (12 bits and 22 bits) were used for sensitivity comparison. All the intensities are standardized into 0-255 scale. The three dimensional surface plot in Figure 20(b) illustrates the trend of greyscale intensities. The image of the grey scale standards was taken by the Casio Commando smartphone, Canon EOS T3i, and HP scanner. Figure 20(c) illustrates the mean intensities of different greyscale regions in each image, in which all the devices have

Figure 20. (a) Grey scale images for camera sensitivity test. (b) Three dimensional surface intensity plot of the grey scale image. (c) The cell phone camera response to grey scale images. The responses of a high end Canon EOS T3i SLR camera and an HP color scanner were plotted for comparison.

41 shown a linear response. The differences between the sample intensities and the actually measured intensities are determined by the material property because its surface absorbs light in a specific wavelength range while reflects all others. The light collected by the camera or the scanner is the unabsorbed part of the ambient light. The test using the HP scanner was carried using the internal light source of the scanner which is brighter than the light bulbs in the camera test, resulting in the higher values in intensities. According to the comparison, the sensitivity of a smartphone camera is comparable to a high end SLR camera or a scanner when the pictures are taken with enough ambient light. Herein, we use Casio Commando smartphone for the following tests.

Smartphone cameras use CMOS arrays, which are low cost but have a range of automated functions integrated, such as auto-gain and Auto White Balance (AWB). These automatic functions are designed to provide good color reproduction by adjusting the detected

RGB signals at different ratios. The resulting adjusted images are brighter and more pleasing to our eyes, making the automatic function popular for non-professional photographers. However, these functions change the RGB values and skew the measurement from being consistent when the smartphone camera is used as a sensor. A clear illustration of this can be observed from imaging pH paper, which turns colors from red to dark blue according to pH value of the test solution. The RGB values of images when pH increased from 1 to 12 are plotted in Figure 21, illustrating that RGB intensities do not exhibit any discernible trend and thus are difficult to correlate with test solution pH values. Some investigators attempted to address this by using hue values [115], by taking a ratio between red and green channels [114], or using subtractive

CMYK and hue-based HLS values [116]. However, these methods are not sensitive to dark colors and are application specific [117]. Another way to address this challenge is to use a

Figure 21. (a) Color response of pH strip at different pH values. (b) Three dimensional surface intensity plot of the pH colors. (c) RGB intensities of the pH strips’ color at each pH value. None of the red, green or blue intensities reflect the pH values.

CMOS image array and its developer board which provides access to most of the array’s configuration functions. Indeed, we used this approach in our previous work with optical oxygen sensors and fluorescence detection [118].

To overcome errors caused by the smartphone camera and the variation in ambient light, we designed a color chart containing 12 color regions with known color intensities to stabilize the camera functions and compensate ambient light differences in data processing. This color chart in our test is based on the commonly used color-rendition chart in photography in which 24 regions are included [119]. We simplified the chart to 12 regions of which seven are in grayscale and the rest are in blue (short wavelength), green, yellow, orange, and red (long wavelength). In our test, the existence of the simplified color-rendition chart greatly reduces the effect of automatic camera functions, making the images more replicable. Moreover, the ambient light compensating test proved that it was sufficient to use only 12 reference regions.

We used CIE color space to code the colorimetric image to overcome inadequacies of the simple RGB analysis. The CIE system is the most recognized method in which color is represented [120], with tristimulus values X, Y, and Z characterizing the emission color of luminescence data across the wavelength range of visible light [121]. Specifically, we used the

CIE 1931 xyY variant, in which the Y parameter represents luminance (brightness) of a color and the derived parameters x and y determine the chromaticity of a color (two of the three normalized values are functions of all three tristimulus values X, Y, and Z). The xy chromaticity diagram is easy to understand and suitable for quantitative analysis of colors [122]. Moreover, the CIE

1931 system can be used to predict the outcome of mixture of two colors. The mixed color lies along the straight line connecting the two points of the original colors on the xy chromaticity diagram. The ratio of the two original colors determines the position of the mixed color [123].

This can be potentially useful in more complicated colorimetric assays [122]. Notably, the hue and saturation of a color, based on which the widely used HSV and HSL models were defined, can be derived from its location on the xy diagram [124]. Considering all the assets of the CIE

1931 xyY color space, we chose this system in our work for quantifying colors.

Conversion of the smartphone camera image data into the CIE 1931 xyY color space involves three simple steps, in which the color space terms are derived from RGB values. Since the smartphone camera used in our experiments utilizes sRGB color standard, as do most of the digital cameras, LCDs, and color printers [125], the nonlinear sRGB values must be converted to linear RGB values using

( ) (3)

in which Csrgb stands for Rsrgb, Gsrgb, and Bsrgb, and Clinear indicates Rlinear, Glinear, and Blinear. Then, the linear RGB values can be converted to trisimulus values X, Y, and Z using the following equations:

[ ] [ ] [ ] (4)

Finally, the chromaticity values x and y are obtained by

{ (5)

The new color space specified by x, y and Y is represented in a 2-D diagram, the Horse Shoe shaped Chromaticity diagram. The pure colors are located on the boundary curve from blue

(380nm) to red (700nm) while all the mixed colors, such as yellow and pink, are represented within the area enclosed by the curve. The position of a point in the diagram indicates the chromaticity of the corresponding color.

Measurements of pH and glucose

The described color quantification method can be applied to commercially available colorimetric test strips. We first demonstrated the approach using colorimetric pH indicator strips (Micro Essential Laboratory) which were dipped into pH buffer solutions and then imaged using the smartphone camera. The mean RGB intensities of the region of interest (ROI) were calculated and converted to chromaticity values x and y. Figure 22(a) illustrates that the 2-D diagram not only intuitively reflects the color change of the pH strip but also the corresponding pH change. Therefore, the pH value is a function of the chromaticity values x and y in the 3-D space, illustrated in Figure 22(b). To measure the pH value, chromaticity x and y obtained for various buffer solution with known pH are fitted into a 2nd order polynomial equation

(6)

The coefficient of determination of the fitting equation (R2 = 0.9874) indicates that the fitting model is accurate to act as a calibration curve. By substituting the x and y values to the calibrated equation, the corresponding pH value was obtained. Figure 22(c) illustrates the measured results of the pH test, compared to the actual buffer solution pH. The measurement of pH data gave a coefficient of determination of 0.9937 and a linear response range from 2 to 10.

The sensitivity of ~0.5 is limited by the pH strip but can be improved by using narrow range pH strips to get more accurate measurement, especially for pH around 7.4 for human blood and urine.

In the second demonstration of the approach, we used a calorimetric urine glucose test strip (Science Kit & Boreal Laboratories). When abnormal values of glucose are found during a urine test, further investigation is required to ascertain one’s true health status. A urine test strip is a quick and inexpensive way to check for glucose in urine, and is one of the most commonly used colorimetric test kit products that can be done in the privacy of one’s home. The normal glucose range in urine is 0-0.8 mM (0-15 mg/dL); when the glucose level exceeds the renal threshold of ~10 mM, glucose can be found in the urine. The test strip measurement range is 0-

111 mM (0-2000 mg/dL) and it changes color to blue-green when urine glucose concentration exceeds 5mM. Figure 23(a) illustrates the color response of the glucose test strip to different

Figure 22. (a) The pH strip colors in CIE 1931 color space. (b) The 3D view of the relation between xy coordinates and the pH values. (c) Calibration curve of the cell phone reader response to different pH buffer solutions. 47 urine glucose concentrations that are indicated in Figure 23(b). After color quantification, the glucose concentration can be expressed as

. (7)

Illustrated in Figure 23(c), the response has a linear range between 5mM and 110mM glucose, with an LOD of 5mM glucose. This result is comparable with that of the colorimetric reader using multiple photodetectors [102]. The linear range covers the range of 0-60mM urine glucose concentration used in clinical diagnosis [126]. The measurement accuracy of the whole system is dependent on both detector and the sensor. By utilizing the CIE color conversion and color reference standards, we have maximized the accuracy obtainable by a smartphone based detector.

Figure 23. (a) The urine glucose test strip colors in CIE 1931 color space. (b) The 3D view of the relation between xy coordinates and the glucose concentrations. (c) Calibration curve of the cell phone reader response to urine samples with different glucose concentrations.

Through continued development of colorimetric assays, the accuracy and sensitivity can be further enhanced.

We should note that the calibration curves developed for both pH and urine glucose test we report herein show dependence on the smartphone CMOS chip. Our tests with HTC and

BlackBerry phones show slight variations (~5%), indicating necessity of re-calibration for each new smartphone model used if higher precision is desired. We should also note that additional measurement errors can be caused by differences in ambient light conditions when test strips are imaged. Thus, we next examine the effect of ambient light and propose an approach for its compensation.

Ambient Light Compensation

While the benchtop results above demonstrate the capability of using a smartphone for quantitative colorimetric analysis, a critical challenge still prevents us from taking it out of the lab and into everyday use. This is due to the variation in ambient light conditions which are determined by the position of the light source, light temperature, or outdoor lighting environment.

An object absorbs light in a specific wavelength range while reflecting the rest. The CMOS pixel measured intensity of the reflected light is determined by many factors, such as the ambient light wavelength, the reflection, the color of the object and the RGB responsivity of the CMOS pixels. A practical way is to treat all these factors as a black box and build a mapping algorithm based on the measured RGB intensities of the references. After measuring the 12 reference colors, we found that the measured intensities between different ambient light conditions had a linear relationship. Figure 24 illustrates the red channel intensities of the reference colors at5000K fluorescent light, compared to other light conditions, such as sunshine,

Figure 24. The linear relationship of the reference patches intensities between 5000K fluorescent light and (a) 3500K fluorescent light, (b) sunshine, (c) shade, and (d) smartphone LED. The color intensities of the sample can be substituted to the fitted curve to derive the corresponding intensities at which the device was calibrated (5000K).

50 shade, smartphone LED, and 3500K fluorescence. The value of R2= 0.99 indicates excellent linearity of the fit. Therefore, we can use the reference paper with known colors to compensate the intensity differences caused by ambient light change. As is illustrated in Figure 24(a), the measured intensities at one light condition I1 can be mapped to another I2. In our test, we calibrated the test strips’ responses and the corresponding 3D fitting model using a 5000K fluorescent light source. Then we ran detection tests under other light conditions. By building up the compensating equations in the RGB channels, we can map any detected sample RGB intensities to what they should have been under 5000K exposure and calculate the corresponding chromaticity values. This color mapping method provides a good way to compensate for the error caused by the ambient light change, making the cellphone detection approach applicable to any lighting environment, from indoor to outdoor, from sunshine to cloudy atmospheres.

In these experiments, we calibrated the detection algorithm for urine glucose at 5000K ambient light condition. Then the measurement was repeated at 3500K ambient light condition.

Figure 25 illustrates the measured chromaticity coordinates of the urine test strips’ color response to samples with different glucose concentrations. The chromaticity values measured under 3500K ambient light exhibit an approximately 0.05 shift in x and 0.03 shift in y which then causes large error in glucose detection. After the processing, the different intensities caused by the ambient light were compensated, resulting in a new series of chromaticity coordinates that matches the calibration at 5000K very well. This light compensating method creates mapping of signals detected at any light conditions to the calibration light condition, enabling improvisatory calibration of the test. To get an accurate measurement, the image of the color reference and test strip should be taken with care. The method assumes that the ambient is uniformly shining on the sample so that the intensities of the reference colors can be used to build a precise conversion

51 curve for the unknown sample. If the smartphone is too close to the sample, it may block the light and generates shadow which breaks the uniformity and causes false measurement. Thus, the smartphone should be placed at a proper position and height, depending on the location of the camera, to get accurate readings.

Summary

This chapter demonstrated that the images taken by a smartphone can be used in colorimetric POC diagnostics. The CIE 1931 based processing approach is not only practical for smartphone images but also applicable for images taken by a microscope or a scanner.

According to the statistic of International Telecommunication Union (ITU), by the year of 2012 the mobile-cellular subscriptions reached 6 billion in which 75% are in developing countries. In addition, a 3-8 megapixel camera is a standard component for most of the smartphones. It will be very simple and convenient to read the levels of several key components in blood or urine by taking a snapshot of the colorimetric paper strips, followed by on board quantitative analysis using smartphone apps, which are currently being developed in our laboratory. This process does not need any trained personnel and can be accomplished in seconds by the user. Therefore, the camera phones are promising to play a central role in disease screening, especially in developing countries and resource limited settings. With the low-cost paper testing strips, the camera phone can generate instant reading of the results, giving the user or medical provider rapid decision making tools. The smartphone based colorimetric detection will not replace the traditional microscopic or spectroscopic based diagnosis but it provides a perfect way of first time screening of a large number of potential patients. Moreover, this method will also benefit people in developed countries who want to monitor their health conditions because with good 3G wireless coverage the smartphone can send high resolution images to a doctor for analysis.

Figure 25. Shift in color space coordinates of the glucose test strip images due to different ambient (3500K and 5000K). The ambient light correction technique effectively compensates the error caused by ambient light. The CIE 1931 color system projects all the human visible colors on a 2D plane, regardless of their brightness. Therefore, this method itself is resistant to the intensity change in the ambient light. However, the actual measurement may take place at any conditions, such as sunshine, home lamp, or smartphone LED. The difference in wavelength between the light conditions where the device is calibrated and the measurement place can cause huge error in the final readings. Using the light compensating technique, this problem can be overcome, moving the smartphone based platform further in practical diagnostic applications.

In summary, we have introduced a novel colorimetric detection method for smartphone based diagnosis. The CIE 1931 color space makes the measurement intuitive and independent to light brightness. Moreover, the light compensating technique improves the feasibility of the platform in various environments. Since the platform is robust and low-cost, it is affordable for most people. Most importantly, all the complicated processing is accomplished by the pre- installed program (app), even an untrained person can easily obtain the results by just one click.

This work is currently being pursued by our laboratory. We believe this technique can be broadly applied for POC diagnosis for any sensor systems that provide colorimetric response.

CHAPTER 5

STATIC AND DYNAMIC DETECTION OF PARTICLES IN

MICROCHANNELS

Introduction

Microscale investigations in medical sciences are mainly carried out using conventional microscopes. A fluorescent microscope typically contains a broadband light source, a set of objectives and filters, and specially designed mechanical holders and arms. These components make the microscope fragile and difficult to miniaturize. Most of the microscopes are equipped a high-end CCD detector that enables imaging or video capturing. These CCD detectors, however, need a stable power supply for continuous running. In most cases, neither the space nor the power supply is available. Thus, the conventional microscope configuration is not suitable for first responder POC in resource limited areas.

Many groups are working on portable microscopes. The optofluidic microscope (OFM) combines microfluidics with aperture-based optics [78, 127]. Two rows of apertures were formed by coating an opaque metal film onto a CMOS array surface. When the specimen flow across the apertures, its speed can be calculated and then used to reconstruct the image. The method is critically dependent on the constant flow rate and the specimen orientation which introduce challenges for POC applications [128]. The holographic microscope is another type of lensless microscope which achieves an impressive resolution of 1µm [129]. However, the

55 imaging process is based on complicated mathematical calculations compared to other on-chip imaging technologies. Moreover, color filters used in this device confines its flexibility in fluorescence detection. The work done by Matsunaga et al. concluded that the distance between the micro-specimen and the surface of CMOS array should be around 25µm to get clear image

[130]. This means all the regular sheet filters are not applicable. One solution is to deposit a thin color filtering layer on the CMOS array. This method, however, will limit the whole device in a specific application because the filter layer is not replaceable. Recently, cellphone based microscopy [131] and telemedicine [113]exhibits promising application in POC due to the powerful signal processing and wireless capability of the cellphone. However, most of the cellphones with a reasonable camera cost several hundred dollars while the CMOS array itself only cost about ~$20. Therefore, we focus our work on the CMOS array to reduce the cost of the whole device. Ultimately, equipping with wireless communication components can enable remote diagnostics using this device.

This chapter will demonstrate the use of the developed fluorescence system (Chapter 1) in static microscopic imaging of fluorescent microparticles. Taking advantage of the spectral and spatial resolution of the system, microparticles with different fluorescence and sizes can be detected simultaneously. Since the CMOS image sensor takes video of flowing microparticles, the system can be used for flow cytometry applications. A video analyzing program was developed in MATLAB, which enables “sample in – results out” process.

Experimental setup and methods

The schematic diagram of the fluorescence detection system is illustrated in Figure 26(a).

The broadband emission of the white LED (Bivar, R20WHT-F-0160), biased at 3.0 V, excites the particles simultaneously. The cross placed polarizers (Edmund Optics, NT45667) pass the

56 signal emission from the sample particles while block the white excitation. The CMOS sensor

(Aptina, MT9P031), connected with a 20X objective, collects the emission light from the particles and save the signal as an image or a video. The microparticle samples were loaded on a

1” ×3 ” glass slide and placed on the sample holder. Figure 26(b) illustrates the CMOS array board and the objective. The whole device is only 3” × 2” × 4”, including the light source and polarizers.

The green emission particles with 20.4µm diameter (catalog no. 19096), 9.9µm diameter

(catalog no. 18140), and 6.1µm diameter (catalog no. 17156) were purchased from Polysciences,

Inc. All 15.4µm diameter blue emission particles (Cat: F8837), green emission particles (Cat:

F21010), and orange emission particles (Cat: F8841) were purchased from Invitrogen. The particles, from 20.4µm to 6.1µm, were diluted in deionized (DI) water at concentrations of 6×104 particles/ml, 2×105 particles/ml, 6×105 particles/ml, 2×106 particles/ml, respectively.

To compare the spatial resolution of the portable system with a conventional fluorescence microscope, 1 µL of a 15.4 µm diameter particle solution and 10 µL of a 20.4 µm diameter particle solution were mixed and dipped onto a glass slide, then placed in a fume hood until the water evaporated. The prepared samples with aggregated particles were imaged using the

Figure 26. (a) Schematic of the multiplexed fluorescent particle detection system. (b) Size comparison of the device and a cell phone. 57 portable system and a conventional microscope with 4X, 10X, and 20X objectives. The intensity profiles of the images were analyzed using Matlab and the number of pixels was normalized to distance in micrometer.

In size based particle counting test, four solutions of 20.6µm, 15.4µm, 9.9µm, and 6.1µm particles were randomly mixed and imaged simultaneously. Matlab was used to convert the captured RGB images to greyscale and mark the particles. The pixel area of each particle was also recorded and then translated to the actually size in micrometer.

Particles of three emissions (blue, green, orange) were used to demonstrate the spectral capability. The RGB intensities of the multi-color particles were recorded and converted to CIE

1931 chromaticity values x and y. The intensities of RGB channels were set to a specific range to select the corresponding emission particles from the whole image. The RGB values detected by the camera were converted to CIE XYZ tristimulus values using equation (3)-(5) in Chapter 4.

Particle imaging and counting

Size based analysis (spatial resolution)

The portable device is capable for multi-color micro-particle imaging. Figure 27(a) illustrates the conventional microscope images of 20.6µm green (λex 480nm, λem 520nm) and

15.4µm orange (λex 540nm, λem 560nm) particles while the middle image captured by the portable system using cross-polarization includes all the particles in one image. The conventional fluorescence microscope used different filters for each type of the particles. The all-in-one imaging method saves time in detection by not requiring filter changes. Since there are no moving parts, the system is highly stable and solid, satisfying essential requirements for field test devices. Figure 27(b) illustrates the particle images at different zoom level. For comparison

58 purpose, the same sample was imaged using the microscope with 4X, 10X, and 20X objectives.

Since the conventional microscope used filters, we chose the “FITC” filter set to image the majority green particles. As a result, the orange emission particles turned out to be vacancies in

Figure 27(c). Compared with the optical magnification of the microscope, the particle images taken by the portable system still maintained good sharpness at high digital magnification primarily due to that the CMOS array we used in the portable system had 5 megapixels with

2.2µm × 2.2µm pixels size. When the optical magnification ratio of the system was 4, a 20µm particle could be magnified to 80µm which approximately covered 36 pixels. Thus, a total of

~1100 (1/4×π×362) pixels were enough to record the details of a single particle.

The next test was to explore the spatial resolution of the portable system. A 5µL particle solution was dipped onto a glass slide and left overnight until the particles were aggregated due to the water evaporation. This process allowed us to locate the particles physically close to one another. Figure 28(a) and (b) illustrates the image of a heap of particles (20.6µm green and

15.4µm orange) which were captured using the portable system and the microscope (20X objective, FITC filter sets), respectively. The intensity profiles along line 1 were plotted in

Figure 28(c) in which the intensity troughs at ~28µm were 38% and 75% lower than the peak values. Based on the commonly used Rayleigh criteria, two objects are regarded as being resolved if the intensity profile has an intensity minimum that is 20%-27% lower than the peak intensity. Thus, the spatial resolution of the portable system is estimated to be 3µm. As particles were closely adjacent to each other, their emission on the edge accumulates, forming the sharp pulses in Figure 28(d). Although the pulse of the red curve is broader than that of the blue curve, the three particles are still differentiable during the following particle counting experiment.

Figure 27. (a) Illustration of micro-particle (20µm green and 15µm orange) imaging. Two greyscale images taken by a conventional fluorescence microscope using TRITC and FITC filters are displayed as comparison. (b) Images taken by the portable device with different zoom levels. (c) Images taken by the microscope using 4X, 10X, and 20X objectives. Note that only green particles are in the images.

Figure 28. Micro-particle images captured by (a) portable system and (b) microscope. (c) (d) The intensity profiles along the marked line 1 and 2 in (a)(b).

The system demonstrated high accuracy in counting the number of particles and quantifying their sizes. Figure 29(a) illustrates the image of four sized fluorescent particles that are 20.6µm, 15.4µm, 9.9µm, and 6.1µm. The field of view of the system is about 1.5 mm × 2 mm. The particles with different sizes were digitally zoomed to further demonstrate the high spatial resolution of the portable system. Since particles exhibit clear outline and high contrast over the background intensity, particle counting can be carried out automatically using Matlab.

Figure 29(b) illustrates the auto counting results in which a peak stands for a count of a particle.

During the counting process, the pixel size of each particle was recorded as well, represented by the values on y-axis. There are clearly four plateaus in Figure 29(b) which indicates the four groups of particles with different sizes.

Figure 30 illustrates the histogram of the automatic counting results obtained after

Figure 29. (a) A typical image of 20.6µm, 15.4µm, 9.9µm, and 6.1µm particles and the corresponding magnified images of each particle. (b) The counting result of particles in an image. Each peak represents a single particle. counting 1200 particles. The numbers of the particles were also counted manually for comparison. The auto-counting results (1184) demonstrated high coincidence with the actual numbers of 1208. The accuracy was ~99% for 20.6µm and 15.4µm particles and ~95% for

9.9µm and 6.1µm particles, with an overall counting accuracy of 98%. The particle areas were then converted to diameters in micrometer using the following equation:

√ (8)

in which P is the pixel size of the CMOS array, M is the magnification ratio, and A is the pixel area. The particle size histogram in Figure 30 exhibits Gaussian distributions centered at

5.47µm, 9.58µm. 15.77µm and 20.09µm. The inset chart indicates high consistency between the measured results and the manufacturer’s data, compared to which the accuracy of size measurement for 20.6µm, 15.4µm and 9.9µm particles is ~97% and ~90% for 6.1µm particles.

The measured standard deviation around 0.5µm is a little higher than the actual sample’s value

Figure 30. The histogram of the particles’ statistics in six images. Each peak represents the number of particles of the size. Insets are a comparison of the software-measurement and the actual values. around 0.3µm but the 0.2 µm difference is negligible in practical applications since the blood cells are larger than 5 µm.

Color based analysis (spectral resolution)

The particle counting takes advantage of the high spatial resolution of the system to differentiate particles by their sizes and achieved high accuracy. However, cells in blood samples are not as uniform as the polymer particles and may have large overlap in size, which could result in discrepancies if the measurements are only based on size. Since the developed system is capable of imaging multi-color fluorescence, the RGB intensities for each particle were recorded during the counting process. Figure 31(a) illustrates the RGB intensity histograms of particles with three emission wavelengths centered at 415nm (blue), 540nm (yellow green), and

560nm (orange). The intensities in each channel exhibit narrow Gaussian distributions. As expected, the intensities of blue and green particles are dominant by the corresponding channel.

The orange particles have almost equally red and green intensity because the emission

63 wavelength of orange light also locates between the peaks of green and red channel of the CMOS array. Since there is no overlap between the color histograms in each channel, the three colors can be separated by setting proper threshold in each channel. Figure 31(a) illustrates the threshold ranges for green particles which selectively deconvolute the green particles from the whole image. Similarly, the particles with other two colors can be deconvoluted too, illustrated in Figure 31(b). Thus, even particles of identical sizes can still be recognized due to the spectral properties of the system.

The particle populations were further interpreted in CIE 1931 color space by converting the RGB intensities to the chromaticity values x and y (details have been described in Chapter 4).

Figure 32 illustrates the distribution of blue, green and orange particles in the color space. There are several advantages of using color space for particle analysis in our system. Firstly, it is an intuitionistic way of displaying the population of particles stained by different emission dyes.

Compared to using color space, it is much less comprehensible to use the abstract 3D RGB space

Figure 31. (a) The typical intensity histograms of the orange, green and blue particles in RGB channels of the CMOS array. (b)The image taken by the system including orange, green and blue particles and the deconvoluted sub-images. 64 to plot the particles’ distribution. Secondly, it is capable to accommodate more samples with different colors because the color space includes all the colors of visible light. Thus, all the particles stained by different dyes can be simply plotted in this 2D diagram. In our system, the purple and deep blue areas are not available because the shortest emission wavelength of the white LED (~420nm) does not excite dyes with shorter emission wavelengths. However, there is still plenty of room in the color space from blue to deep red that we can use for multiple samples.

Thirdly, this simple analysis method perfectly matches the single LED and single signal isolation structure of the system in which the CMOS array detects all multi-color particles simultaneously.

Combining the spatial and spectral information

All the sample particles can be analyzed and categorized according to the size and color information. The system records four parameters of each particle: RGB intensities and size. By converting the RGB intensities to xy chromaticity values, particles with different emission colors can be clearly distinguished and analyzed in the 2-D diagram. Then the size information can be plotted as z-axis to construct a complete 3-D space containing all the populations in the sample mixture. Figure 33(a) illustrates the distribution of four types of particles. The orange particles can be easily distinguished from the sample from their chromaticity values x&y. Although the

9.9µm 15.4 µm, and 20.4µm particles are identical in emission color and their positions on the color space overlapped, they can still be identified by size. The saved data in the 3D space can be used for further analysis. According to the measured data, the concentrations of 20.6µm

(green), 15.4µm (green), 15.4µm(orange), and 9.9µm (green) particles are 1.7×104/ml,

6.2×104/ml, 6.4×104/ml, 2.1×105and /ml, respectively. The concentration is indicated by the size of the bubbles in Figure 33(b). The results are very close to the prepared concentrations of

2×104/ml, 6.6×104/ml, 6.6×104/ml, 2×105and /ml.

Figure 32. Distribution of blue, green and orange emission particles in CIE 1931 color space.

Figure 33. (a) Scatter plot of a particle mixture sample in the 3D space illustrating four populations: green particles (9.9 µm), green particles (15.4µm), orange particles (15.4 µm), and green particles (20.6 µm). (b) Bubble plot of the particles. The bubble size indicates the concentration of the corresponding particles.

Summary

The multiplexed fluorescence detection system provides several advantages. First, by using the combination of color discriminating CMOS image sensor and cross-polarization signal isolation, there is no need to change light source or filters for different particles. Without such moving parts, the system is highly robust and stable, permitting on site applications. Second, all the samples can be imaged at once due to the high spectral and spatial resolution. This greatly simplifies the testing procedure to just one snapshot, making the device user-friendly to untrained personnel. Third, the use of CIE 1931 color system enables straightforward signal processing. By combining the chromaticity and size information of each particle, the scatter plot for the particle populations can be displayed in a 3-D space, being intuitive and easy to understand. Therefore, the low-cost, user-friendly fluorescence detection system can be used as an alternative for conventional fluorescent microscopes, especially in rural or developing areas.

Compared to the recently emerged camera phone based detection method [112, 113, 131,

132] our method of using a CMOS image sensor as detector does not provide the ability of wireless communication. However, the utilization of CMOS image sensor can help to keep the consistency in measurements. Manufactures use different camera chips for their cell phones and all the camera settings are not the same between different models. As a result, it is very difficult to establish a universal standard or protocol for cell phone based diagnostics. On the other hand, the low-cost CMOS-based devices can be mass produced specially for diagnosis rather than the omnipotent smartphones. All camera functions that may affect the measurement results can be controlled so as to keep the test consistency [118, 133].

As the system demonstrated promising results with micro-particles, it will be validated with fluorescent dye stained cells to further explore its feasibility in cell counting based disease

67 diagnostics. The pixel size of the CMOS sensor can be selected accordingly to the NA of objective and the size of sample cells to achieve a good balance between light sensitivity and spatial resolution. Moreover, the conversion of RGB intensity to CIE chromaticity can be coded in the image processor chip to achieve real time analysis, enabling faster diagnosis.

In summary, a prototype fluorescence detection system using a CMOS image sensor was developed. The novel cross-polarization scheme was used for signal isolation. The combination of color discriminating CMOS image sensor and wavelength independent cross-polarization realizes simultaneous multi-color multi-sample detection. The system demonstrated ~3µm in spatial resolution and 98% accuracy in particle counting. By converting the measured RGB intensity to the CIE 1931 color space coordinates all the particles with different colors can be displayed on a single 2D plane. When the size information is added, all the particle samples, including particles with different sizes and emission colors, can be accommodated and analyzed in the single 3D space using three parameters (chromaticity values xy and size). Thus, the easy- to-use signal processing algorithm is appropriate for our system. According to the polarization test, this detection system is also compatible to the most commonly used materials for medical devices. In the future, it is also possible to integrate wireless communication modules into the system to achieve telemedicine applications. The low-cost portable fluorescence detection can be useful for quantitative analysis of diagnostic assays developed for health applications, especially in resource limited areas.

CHAPTER 6

CONCLUSIONS

Summary

In this dissertation, an optical detection platform has been developed for POC applications. The overall objective was to build a portable, affordable, and user-friendly system.

Two widely used detection methods have been explored: fluorescence detection and colorimetric detection. The feasibility of using CMOS image sensors, either on a development board or integrated in a cell phone, as detector to acquire signal light has been studied. Promising results indicate future clinical applications for disease diagnostics or health examination.

In the fluorescence detection system, three key components were chosen for the benefit of multiplex detection: a broadband emission white LED light source, wavelength independent cross-polarization signal isolation, and a color discriminating CMOS image sensor. Using this setup, multiple fluorephores can be excited and detected at once. The cross-polarization scheme works well in filtering the broad white light, even when the excitation and emission spectra overlap. The compatibility of the system to mainstream medical materials has been validated using PMMA, glass, PDMS, COC, and polypropylene. The fluorescent dye measurement and oxygen sensing applications demonstrate the spectral and spatial resolution of the system. The

1µM LOD for fluorescent dye and the 41 sensitivity in oxygen sensing are much higher than the reported lab-on-a-chip sensor. The system is also capable for flow cytometry tests. By

69 combining size and color information, microparticles can be counted and categorized accordingly using the specially developed Matlab algorithm. The simultaneous counting result of more than 8000 micro-particles have shown accuracy over 98% and 85% in static and dynamic test, respectively. These results indicate the sensitivity and accuracy of the system is potentially capable for future clinical diagnostic detection.

In the cell phone based colorimetric detection, CIE 1931 color space was used to quantify the color responses of pH and glucose strips. Linear response ranges are 1-12 for pH detection and 0-60 mM for urine glucose detection. Using the ambient light compensating technique, this cell phone detection approach can be used in various light conditions, making it much more applicable and practical. Since cell phones are processed by most people in the world and a camera is a standard module in a phone, the cell phone based approach can be widely applied in disease screening or home health monitoring.

In this work, the cell phone based colorimetric detection requires that the ambient uniformly light the sample strip. Shadows of the tester or the cell phone may cause large measurement error. Therefore, background light detection and subtraction function will be helpful to further enhance the accuracy of the technique, making it more applicable. Moreover, the direct reflection of light source from smooth surfaces leads to high noise over the scatter light.

Since any surface reflection light is polarized, a low-cost polarizer can be used to eliminate such noise and thus improve the measurement accuracy.

The colorimetric paper strips are extremely inexpensive and easy to use but the sensitivity is relatively poor compared to conventional laboratory assays, such as ELISA and flow cytometry. The CMOS image sensor based system has already demonstrated good sensitivity and accuracy in fluorescent dye detection and micro-particle counting. To further

70 improve the LOD of the system, CMOS sensors with larger pixel size can be used, as the sensitivity is quadratically proportional to the pixel size. On the other hand, the throughput of microparticle counting can be improved by integrated a high speed CMOS sensor with frame rate larger than 1000 fps. The system can also be further validated with clinical assays and real cell counting. Last but not least, a battery powered microcontroller board is being developed to acquire digital signals from the CMOS sensor. This moving will finally eliminate the needs of a computer and achieve a pocket-sized device for various optical diagnostic applications.

REFERENCES

[1] A. M. Dupuy, S. Lehmann and J. P. Cristol, "Protein biochip systems for the clinical laboratory." Clin. Chem. Lab. Med., vol. 43, pp. 1291-1302, 2005.

[2] D. S. Boyle, K. R. Hawkins, M. S. Steele, M. Singhal and X. Cheng, "Emerging technologies for point-of-care CD4 T-lymphocyte counting," Trends Biotechnol., vol. 30, pp. 45-54, 2012.

[3] S. Park, Y. Zhang, S. Lin, T. -. Wang and S. Yang, "Advances in microfluidic PCR for point-of-care infectious disease diagnostics," Biotechnol. Adv., vol. 29, pp. 830-839, 2011.

[4] C. D. Chin, V. Linder and S. K. Sia, "Lab-on-a-chip devices for global health: Past studies and future opportunities," Lab Chip, vol. 7, pp. 41-57, 2007.

[5] P. Yager, T. Edwards, E. Fu, K. Helton, K. Nelson, M. R. Tam and B. H. Weigl, "Microfluidic diagnostic technologies for global public health," Nature, vol. 442, pp. 412-418, 2006.

[6] L. Bissonnette and M. G. Bergeron, "Next revolution in the molecular theranostics of infectious diseases: Microfabricated systems for personalized medicine," Expert Review of Molecular Diagnostics, vol. 6, pp. 433-450, 2006.

[7] M. -. Mohammed and M. P. Y. Desmulliez, "Lab-on-a-chip based immunosensor principles and technologies for the detection of cardiac biomarkers: A review," Lab Chip, vol. 11, pp. 569-595, 2011.

[8] D. A. Anderson, S. M. Crowe and M. Garcia, "Point-of-care testing," Current HIV/AIDS Reports, vol. 8, pp. 31-37, 2011.

[9] C. F. Gilks, S. Crowley, R. Ekpini, S. Gove, J. Perriens, Y. Souteyrand, D. Sutherland, M. Vitoria, T. Guerma and K. De Cock, "The WHO public-health approach to antiretroviral treatment against HIV in resource-limited settings," Lancet, vol. 368, pp. 505-510, 2006.

[10] S. Mtapuri-Zinyowera, M. Chideme, D. Mangwanya, O. Mugurungi, S. Gudukeya, K. Hatzold, A. Mangwiro, G. Bhattacharya, J. Lehe and T. Peter, "Evaluation of the PIMA point-of-care CD4 analyzer in VCT clinics in Zimbabwe." J. Acquir. Immune Defic. Syndr., vol. 55, pp. 1-7, 2010.

[11] C. L. Do Lago, H. D. Torres da Silva, C. A. Neves, J. G. Alves Brito-Neto and J. A. Fracassi da Silva, "A dry process for production of microfluidic devices based on the lamination of laser-printed polyester films," Anal. Chem., vol. 75, pp. 3853-3858, 2003.

[12] I. Grabowska, M. Sajnoga, M. Juchniewicz, M. Chudy, A. Dybko and Z. Brzozka, "Microfluidic system with electrochemical and optical detection," Microelectron. Eng., vol. 84, pp. 1741-1743, 2007.

[13] N. Ohgami, S. Upadhyay, A. Kabata, K. Morimoto, H. Kusakabe and H. Suzuki, "Determination of the activities of glutamic oxaloacetic transaminase and glutamic pyruvic transaminase in a microfluidic system," Biosens. Bioelectron., vol. 22, pp. 1330- 1336, 2007.

[14] E. De Hoffmann, J. Charette and V. Stroobant, Mass Spectrometry: Principles and Applications, 1996.

[15] A. K. Naik, M. S. Hanay, W. K. Hiebert, X. L. Feng and M. L. Roukes, "Towards single- molecule nanomechanical mass spectrometry," Nat. Nanotechnol., vol. 4, pp. 445-450, 2009.

[16] A. Pais, A. Banerjee, D. Klotzkin and I. Papautsky, "High-sensitivity, disposable lab-on- a-chip with thin-film organic electronics for fluorescence detection," Lab Chip, vol. 8, pp. 794-800, 2008.

[17] A. Banerjee, A. Pais, I. Papautsky and D. Klotzkin, "A polarization isolation method for high-sensitivity, low-cost on-chip fluorescence detection for microfluidic lab-on-a-chip," IEEE Sens. J., vol. 8, pp. 621-627, 2008.

[18] T. C. D. Huang, S. Paul, P. Gong, R. Levicky, J. Kymissis, S. A. Amundson and K. L. Shepard, "Gene expression analysis with an integrated CMOS microarray by time- resolved fluorescence detection," Biosens. Bioelectron., vol. 26, pp. 2660-2665, 2011.

[19] F. B. Myers and L. P. Lee, "Innovations in optical microfluidic technologies for point-of- care diagnostics," Lab Chip, vol. 8, pp. 2015-2031, 2008.

[20] J. Wu and M. Gu. Microfluidic sensing: State of the art fabrication and detection techniques. J Biomed Opt, vol. 16, pp. 080901, 2011.

[21] B. Kuswandi, Nuriman, J. Huskens and W. Verboom, "Optical sensing systems for microfluidic devices: A review," Anal. Chim. Acta, vol. 601, pp. 141-155, 2007.

[22] K. B. Mogensen and J. P. Kutter, "Optical detection in microfluidic systems," Electrophoresis, vol. 30, pp. S92-S100, 2009.

[23] O. Gustafsson, K. B. Mogensen, P. D. Ohlsson, Y. Liu, S. C. Jacobson and J. P. Kutter, J Micromech Microengineering, vol. 18, 2008.

[24] J. Xu, D. Suarez and D. S. Gottfried, "Detection of avian influenza virus using an interferometric biosensor," Anal. Bioanal.Chem., vol. 389, pp. 1193-1199, 2007.

[25] A. Ymeti, J. Greve, P. V. Lambeck, T. Wink, S. W. F. M. Van Novell, T. A. M. Beumer, R. R. Wijn, R. G. Heideman, V. Subramaniam and J. S. Kanger, "Fast, ultrasensitive virus detection using a young interferometer sensor," Nano Letters, vol. 7, pp. 394-397, 2007.

[26] D. Brennan, J. Justice, B. Corbett, T. McCarthy and P. Galvin, "Emerging optofluidic technologies for point-of-care genetic analysis systems: A review," Anal. Bioanal.Chem., vol. 395, pp. 621-636, 2009.

[27] J. Wu, X. Liu, L. Wang, L. Dong and Q. Pu, "An economical fluorescence detector for lab-on-a-chip devices with a light emitting photodiode and a low-cost avalanche photodiode," Analyst, vol. 137, pp. 519-525, 2012.

[28] A. Bhattacharyya and C. M. Klapperich, "Design and testing of a disposable microfluidic chemiluminescent immunoassay for disease biomarkers in human serum samples," Biomed. Microdevices, vol. 9, pp. 245-251, 2007.

[29] E. Yacoub-George, W. Hell, L. Meixner, F. Wenninger, K. Bock, P. Lindner, H. Wolf, T. Kloth and K. A. Feller, "Automated 10-channel capillary chip immunodetector for biological agents detection," Biosens. Bioelectron., vol. 22, pp. 1368-1375, 2007.

[30] T. M. Chinowsky, J. G. Quinn, D. U. Bartholomew, R. Kaiser and J. L. Elkind, "Performance of the Spreeta 2000 integrated surface plasmon resonance affinity sensor," Sens. Actuators, B, vol. 91, pp. 266-274, 2003.

[31] D. Wei, O. A. Oyarzabal, T. -. Huang, S. Balasubramanian, S. Sista and A. L. Simonian, "Development of a surface plasmon resonance biosensor for the identification of Campylobacter jejuni," J. Microbiol. Methods, vol. 69, pp. 78-85, 2007.

[32] L. -. Bouchard, S. R. Burt, M. S. Anwar, K. V. Kovtunov, I. V. Koptyug and A. Pines, "NMR imaging of catalytic hydrogenation in microreactors with the use of para- hydrogen," Science, vol. 319, pp. 442-445, 2008.

[33] B. Godber, K. S. J. Thompson, M. Rehak, Y. Uludag, S. Kelling, A. Sleptsov, M. Frogley, K. Wiehler, C. Whalen and M. A. Cooper, "Direct quantification of analyte concentration by resonant acoustic profiling," Clin. Chem., vol. 51, pp. 1962-1972, 2005.

[34] M. Mujika, S. Arana, E. Castaño, M. Tijero, R. Vilares, J. M. Ruano-López, A. Cruz, L. Sainz and J. Berganza, "Magnetoresistive immunosensor for the detection of Escherichia coli O157:H7 including a microfluidic network," Biosens. Bioelectron., vol. 24, pp. 1253- 1258, 2009.

[35] International Telecommunication Union, "Mobile-cellular subscriptions, by level of development," Mobile-Cellular Subscriptions, by Level of Development, pp. http://www.itu.int/ITU-D/ict/statistics/index.html, 2012.

[36] C. J. Easley, J. M. Karlinsey, J. M. Bienvenue, L. A. Legendre, M. G. Roper, S. H. Feldman, M. A. Hughes, E. L. Hewlett, T. J. Merkel, J. P. Ferrance and J. P. Landers, "A fully integrated microfluidic genetic analysis system with sample-in-answer-out capability," Proc. Natl. Acad. Sci. U. S. A., vol. 103, pp. 19272-19277, 2006.

[37] A. E. Herr, A. V. Hatch, W. V. Giannobile, D. J. Throckmorton, H. M. Tran, J. S. Brennan and A. K. Singh, "Integrated microfluidic platform for oral diagnostics," Ann. N.Y. Acad. Sci., vol. 1098, pp. 362-374, 2007.

[38] D. Liu, X. Zhou, R. Zhong, N. Ye, G. Chang, W. Xiong, X. Mei and B. Lin, "Analysis of multiplex PCR fragments with PMMA microchip," Talanta, vol. 68, pp. 616-622, 2006.

[39] P. Liu, T. S. Seo, N. Beyor, K. -. Shin, J. R. Scherer and R. A. Mathies, "Integrated portable polymerase chain reaction-capillary electrophoresis microsystem for rapid forensic short tandem repeat typing," Anal. Chem., vol. 79, pp. 1881-1889, 2007.

[40] F. Xu, P. Datta, H. Wang, S. Gurung, M. Hashimoto, S. Wei, J. Goettert, R. L. McCarley and S. A. Soper, "Polymer microfluidic chips with integrated waveguides for reading microarrays," Anal. Chem., vol. 79, pp. 9007-9013, 2007.

[41] B. C. Heinze, J. -. Song, C. -. Lee, A. Najam and J. -. Yoon, "Microfluidic immunosensor for rapid and sensitive detection of bovine viral diarrhea virus," Sens. Actuators, B, vol. 138, pp. 491-496, 2009.

[42] D. B. Papkovsky, T. C. O'Riordan and G. G. Guilbault, "An immunosensor based on the glucose oxidase label and optical oxygen detection," Anal. Chem., vol. 71, pp. 1568- 1573, 1999.

[43] R. Irawan, S. C. Tjin, X. Fang and C. Y. Fu, "Integration of optical fiber light guide, fluorescence detection system, and multichannel disposable microfluidic chip," Biomed. Microdevices, vol. 9, pp. 413-419, 2007.

[44] M. Hajj-Hassan, T. Gonzalez, E. Ghafar-Zadeh, H. Djeghelian, V. Chodavarapu, M. Andrews and D. Therriault, "Direct-dispense polymeric waveguides platform for optical chemical sensors," Sensors, vol. 8, pp. 7636-7648, 2008.

[45] M. A. Burns, B. N. Johnson, S. N. Brahmasandra, K. Handique, J. R. Webster, M. Krishnan, T. S. Sammarco, P. M. Man, D. Jones, D. Heldsinger, C. H. Mastrangelo and D. T. Burke, "An integrated nanoliter DNA analysis device," Science, vol. 282, pp. 484- 487, 1998.

[46] V. Namasivayam, R. Lin, B. Johnson, S. Brahmasandra, Z. Razzacki, D. T. Burke and M. A. Burns, "Advances in on-chip photodetection for applications in miniaturized genetic analysis systems," J Micromech Microengineering, vol. 14, pp. 81-90, 2004.

[47] J. R. Webster, M. A. Burns, D. T. Burke and C. H. Mastrangelo, "Monolithic capillary electrophoresis device with integrated fluorescence detector," Anal. Chem., vol. 73, pp. 1622-1626, 2001.

[48] Y. -. Kim, K. -. Shin, J. -. Kang, E. -. Yang, K. -. Paek, D. -. Seo and B. -. Ju, "Poly(dimethylsiloxane)-based packaging technique for microchip fluorescence detection system applications," J Microelectromech Syst, vol. 15, pp. 1152-1158, 2006.

[49] K. -. Shin, Y. -. Kim, K. -. Paek, J. -. Park, E. -. Yang, T. -. Kim, J. -. Kang and B. -. Ju, "Characterization of an integrated fluorescence-detection hybrid device with photodiode and organic light-emitting diode," IEEE Electron Device Lett., vol. 27, pp. 746-748, 2006.

[50] T. Kamei, B. M. Paegel, J. R. Scherer, A. M. Skelley, R. A. Street and R. A. Mathies, "Integrated Hydrogenated Amorphous Si Photodiode Detector for Microfluidic Bioanalytical Devices," Anal. Chem., vol. 75, pp. 5300-5305, 2003.

[51] T. Kamei, B. M. Paegel, J. R. Scherer, A. M. Skelley, R. A. Street and R. A. Mathies, "Fusion of a-Si:H sensor technology with microfluidic bioanalytical devices," J. Non Cryst. Solids, vol. 338-340, pp. 715-719, 2004.

[52] E. Thrush, O. Levi, K. Wang, J. J. S. Harris and S. Smith, "Integrated semiconductor fluorescent detection system for biochip and biomedical applications," 2nd Annual International IEEE-EMBS Special Topic Conference on Microtechnologies in Medicine & Biology, Madison, Wisconsin, USA, 2002.

[53] E. Thrush, O. Levi, W. Ha, K. Wang, S. J. Smith and J. S. Harris Jr., "Integrated bio- fluorescence sensor," J. Chromatogr. A, vol. 1013, pp. 103-110, 2003.

[54] W. L. Barnes, A. Dereux and T. W. Ebbesen, "Surface plasmon subwavelength optics," Nature, vol. 424, pp. 824-830, 2003.

[55] J. A. Dionne, L. A. Sweatlock, H. A. Atwater and A. Polman, "Plasmon slot waveguides: Towards chip-scale propagation with subwavelength-scale localization," Phys. Rev. B: Condens. Matter, vol. 73, 2006.

[56] J. A. Chediak, Z. Luo, J. Seo, N. Cheung, L. P. Lee and T. D. Sands, "Heterogeneous integration of CdS filters with GaN LEDs for fluorescence detection microsystems," Sens.Actuators A, vol. 111, pp. 1-7, 2004.

[57] V. P. Iordanov, J. Bastemeijer, R. Ishihara, P. M. Sarro, A. Bossche and M. J. Vellekoop, "Filter-protected photodiodes for high-throughput enzymatic analysis," IEEE Sens. J., vol. 4, pp. 584-588, 2004.

[58] M. L. Chabinyc, D. T. Chiu, J. C. McDonald, A. D. Stroock, J. F. Christian, A. M. Karger and G. M. Whitesides, "An integrated fluorescence detection system in poly(dimethylsiloxane) for microfluidic applications," Anal. Chem., vol. 73, pp. 4491- 4498, 2001.

[59] C. Chen, D. Hirdes and A. Folch, "Gray-scale photolithography using microfluidic photomasks," Proc. Natl. Acad. Sci. U. S. A., vol. 100, pp. 1499-1504, 2003.

[60] O. J. A. Schueller, D. C. Duffy, J. A. Rogers, S. T. Brittain and G. M. Whitesides, "Reconfigurable diffraction gratings based on elastomeric microfluidic devices," Sens Actuators A, vol. 78, pp. 149-159, 1999.

[61] Y. Kostov, P. Harms, L. Randers-Eichhorn and G. Rao, "Low-cost microbioreactor for high-throughput bioprocessing," Biotechnol. Bioeng., vol. 72, pp. 346-352, 2001.

[62] C. McDonagh, C. Kolle, A. K. McEvoy, D. L. Dowling, A. A. Cafolla, S. J. Cullen and B. D. MacCraith, "Phase fluorometric dissolved oxygen sensor," Sens. Actuators, B, vol. 74, pp. 124-130, 2001.

[63] A. N. Watkins, B. R. Wenner, J. D. Jordan, W. Xu, J. N. Demas and F. V. Bright, "Portable, low-cost, solid-state luminescence-based O2 sensor," Appl. Spectrosc., vol. 52, pp. 750-754, 1998.

[64] M. -. Tu, Y. -. Chang, Y. -. Kang, H. -. Chang, P. Chang and T. -. Yew, "A quantum dot- based optical immunosensor for human serum albumin detection," Biosens. Bioelectron., vol. 34, pp. 286-290, 2012.

[65] A. C. Pimentel, R. Cabeça, M. Rodrigues, D. M. F. Prazeres, V. Chu and J. P. Conde, "Fluorescence detection of DNA hybridization using an integrated thin-film amorphous silicon n-i-p photodiode," Mater. Res. Soc. Symp. Proc., pp. 451-456, 2008.

[66] O. Hofmann, P. Miller, P. Sullivan, T. S. Jones, J. C. Demello, D. D. C. Bradley and A. J. Demello, "Thin-film organic photodiodes as integrated detectors for microscale chemiluminescence assays," Sens. Actuators, B, vol. 106, pp. 878-884, 2005.

[67] X. Wang, O. Hofmann, R. Das, E. M. Barrett, A. J. DeMello, J. C. DeMello and D. D. C. Bradley, "Integrated thin-film polymer/fullerene photodetectors for on-chip microfluidic chemiluminescence detection," Lab Chip, vol. 7, pp. 58-63, 2007.

[68] T. N. Ng, W. S. Wong, M. L. Chabinyc, S. Sambandan and R. A. Street, "Flexible image sensor array with bulk heterojunction organic photodiode," Appl. Phys. Lett., vol. 92, pp. 213303, 2008.

[69] P. Peumans and S. R. Forrest, "Very-high-efficiency double-heterostructure copper phthalocyanine/C60 photovoltaic cells," Appl. Phys. Lett., vol. 79, pp. 126-128, 2001.

[70] P. Vivo, J. Jukola, M. Ojala, V. Chukharev and H. Lemmetyinen, "Influence of Alq3/Au cathode on stability and efficiency of a layered organic solar cell in air," Solar Energy Mater. Solar Cells, vol. 92, pp. 1416-1420, 2008.

[71] Q. L. Song, M. L. Wang, E. G. Obbard, X. Y. Sun, X. M. Ding, X. Y. Hou and C. M. Li, "Degradation of small-molecule organic solar cells," Appl. Phys. Lett., vol. 89, pp. 251118, 2006.

[72] T. M. Chinowsky, S. D. Soelberg, P. Baker, N. R. Swanson, P. Kauffman, A. Mactutis, M. S. Grow, R. Atmar, S. S. Yee and C. E. Furlong, "Portable 24-analyte surface

plasmon resonance instruments for rapid, versatile biodetection," Biosens. Bioelectron., vol. 22, pp. 2268-2275, 2007.

[73] E. Fu, T. Chinowsky, K. Nelson, K. Johnston, T. Edwards, K. Helton, M. Grow, J. W. Miller and P. Yager, "SPR imaging-based salivary diagnostics system for the detection of small molecule analytes," Ann. N.Y. Acad. Sci., vol. 1098, pp. 335-344, 2007.

[74] Y. -. Yu, Y. Jiang and R. -. He, "Development of a miniature analytical system in a lab- on-valve for determination of trace copper by bead injection spectroscopy," Talanta, vol. 88, pp. 352-357, 2012.

[75] A. El Gamal and H. Eltoukhy, "CMOS image sensors," IEEE Circuits Devices Mag., vol. 21, pp. 6-20, 2005.

[76] D. M. Vykoukal, G. P. Stone, P. R. C. Gascoyne, E. U. Alt and J. Vykoukal, "Quantitative detection of bioassays with a low-cost image-sensor array for integrated microsystems," Angew. Chem. Int. Ed., vol. 48, pp. 7649-7654, 2009.

[77] Y. Hosseini and K. V. I. S. Kaler, "Integrated CMOS optical sensor for cell detection and analysis," Sens. Actuators, A, vol. 157, pp. 1-8, 2010.

[78] X. Heng, D. Erickson, L. R. Baugh, Z. Yaqoob, P. W. Sternberg, D. Psaltis and C. Yang, "Optofluidic microscopy - A method for implementing a high resolution optical microscope on a chip," Lab Chip, vol. 6, pp. 1274-1276, 2006.

[79] G. Medoro, N. Manaresi, A. Leonardi, L. Altomare, M. Tartagni and R. Guerrieri, "A lab-on-a-chip for cell detection and manipulation," IEEE Sens. J., vol. 3, pp. 317-325, 2003.

[80] J. M. Song, M. Culha, P. M. Kasili, G. D. Griffin and T. Vo-Dinh, "A compact CMOS biochip immunosensor towards the detection of a single bacteria," Biosens. Bioelectron., vol. 20, pp. 2203-2209, 2005.

[81] G. Sharma. Digital Color Imaging Handbook 2003.

[82] B. G. Healey and D. R. Walt, "Improved fiber-optic chemical sensor for penicillin," Anal. Chem., vol. 67, pp. 4471-4476, 1995.

[83] K. Tsukada, S. Sakai, K. Hase and H. Minamitani, "Development of catheter-type optical oxygen sensor and applications to bioinstrumentation," Biosens. Bioelectron., vol. 18, pp. 1439-1445, 2003.

[84] E. Vander Donckt, B. Camerman, R. Herne and R. Vandeloise, "Fibre-optic oxygen sensor based on luminescence quenching of a Pt(II) complex embedded in polymer matrices," Sens. Actuators B, vol. 32, pp. 121-127, 1996.

[85] J. R. Bacon, "Determination of oxygen concentrations by luminescence quenching of a polymer-immobilized transition-metal complex," Anal. Chem., vol. 59, pp. 2780-2785, 1987.

[86] P. Hartmann, M. J. P. Leiner and M. E. Lippitsch, "Luminescence quenching behavior of an oxygen sensor based on a Ru(II) complex dissolved in polystyrene," Anal. Chem., vol. 67, pp. 88-93, 1995.

[87] S. -. Lee and I. Okura, "Porphyrin-doped sol-gel glass as a probe for oxygen sensing," Anal. Chim. Acta, vol. 342, pp. 181-188, 1997.

[88] P. Douglas and K. Eaton, "Response characteristics of thin film oxygen sensors, Pt and Pd octaethylporphyrins in polymer films," Sens. Actuators, B, vol. 82, pp. 200-208, 2002.

[89] T. -. Yeh, C. -. Chu and Y. -. Lo, "Highly sensitive optical fiber oxygen sensor using Pt(II) complex embedded in sol-gel matrices," Sens. Actuators B, vol. 119, pp. 701-707, 2006.

[90] D. A. Chang-Yen and B. K. Gale, "An integrated optical oxygen sensor fabricated using rapid-prototyping techniques," Lab Chip, vol. 3, pp. 297-301, 2003.

[91] S. R. Ricketts and P. Douglas, "A simple colorimetric luminescent oxygen sensor using a green LED with Pt octaethylporphyrin in ethyl cellulose as the oxygen-responsive element," Sens. Actuators B, vol. 135, pp. 46-51, 2008.

[92] E. Kraker, A. Haase, B. Lamprecht, G. Jakopic, C. Konrad and S. Köstler, "Integrated organic electronic based optochemical sensors using polarization filters," Appl. Phys. Lett., vol. 92, pp. 033302, 2008.

[93] R. Herne, J. Brocas and E. Vander Donckt, "The response time of optical sensors based on luminescence quenching," Anal. Chim. Acta, vol. 364, pp. 131-141, 1998.

[94] D. B. Papkovsky, G. V. Ponomarev, W. Trettnak and P. O'Leary, "Phosphorescent Complexes of Porphyrin Ketones: Optical Properties and Application to Oxygen Sensing," Anal. Chem., vol. 67, pp. 4112-4117, 1995.

[95] A. W. Martinez, S. T. Phillips, M. J. Butte and G. M. Whitesides, "Patterned paper as a platform for inexpensive, low-volume, portable bioassays," Angew Chem Int Edit, vol. 46, pp. 1318-1320, 2007.

[96] D. A. Bruzewicz, M. Reches and G. M. Whitesides, "Low-cost printing of poly(dimethylsiloxane) barriers to define microchannels in paper," Anal. Chem., vol. 80, pp. 3387-3392, 2008.

[97] S. Wang, F. Xu and U. Demirci, "Advances in developing HIV-1 viral load assays for resource-limited settings," Biotechnol. Adv., vol. 28, pp. 770-781, 2010.

[98] M. A. Alyassin, S. Moon, H. O. Keles, F. Manzur, R. L. Lin, E. Hæggstrom, D. R. Kuritzkes and U. Demirci, "Rapid automated cell quantification on HIV microfluidic devices," Lab Chip, vol. 9, pp. 3364-3369, 2009.

[99] S. Wang, L. Ge, X. Song, J. Yu, S. Ge, J. Huang and F. Zeng, "Paper-based chemiluminescence ELISA: Lab-on-paper based on chitosan modified paper device and wax-screen-printing," Biosens Bioelectron, vol. 31, pp. 212-218, 2012.

[100] A. W. Martinez, "Microfluidic paper-based analytical devices: From POCKET to paper- based ELISA," Bioanalysis, vol. 3, pp. 2589-2592, 2011.

[101] W. Wang, W. Y. Wu, W. Wang and J. J. Zhu, "Tree-shaped paper strip for semiquantitative colorimetric detection of protein with self-calibration," J Chromatogr A, vol. 1217, pp. 3896-3899, 2010.

[102] D. -. Lee, B. G. Jeon, C. Ihm, J. -. Park and M. Y. Jung, "A simple and smart telemedicine device for developing regions: A pocket-sized colorimetric reader," Lab Chip, vol. 11, pp. 120-126, 2011.

[103] M. A. Nash, J. M. Hoffman, D. Y. Stevens, A. S. Hoffman, P. S. Stayton and P. Yager, "Laboratory-scale protein striping system for patterning biomolecules onto paper-based immunochromatographic test strips," Lab Chip, vol. 10, pp. 2279-2282, 2010.

[104] W. G. Lee, Y. -. Kim, B. G. Chung, U. Demirci and A. Khademhosseini, "Nano/Microfluidics for diagnosis of infectious diseases in developing countries," Adv. Drug Deliv. Rev., vol. 62, pp. 449-457, 2010.

[105] A. W. Martinez, S. T. Phillips, G. M. Whitesides and E. Carrilho, "Diagnostics for the developing world: Microfluidic paper-based analytical devices," Anal. Chem., vol. 82, pp. 3-10, 2010.

[106] L. Yu, C. M. Li, Y. Liu, J. Gao, W. Wang and Y. Gan, "Flow-through functionalized PDMS microfluidic channels with dextran derivative for ELISAs," Lab Chip, vol. 9, pp. 1243-1247, 2009.

[107] K. Tohda and M. Gratzl, "Micro-miniature autonomous optical sensor array for monitoring ions and metabolites 1: Design, fabrication, and data analysis," Anal Sci, vol. 22, pp. 383-388, 2006.

[108] K. Suzuki, E. Hirayama, T. Sugiyama, K. Yasuda, H. Okabe and D. Citterio, "Ionophore- based lithium ion film optode realizing multiple color variations utilizing digital color analysis," Anal. Chem., vol. 74, pp. 5766-5773, 2002.

[109] D. J. Soldat, P. Barak and B. J. Lepore, "Microscale colorimetric analysis using a desktop scanner and automated digital image analysis," J. Chem. Educ., vol. 86, pp. 617-620, 2009.

[110] E. M. T. Wurm, R. Hofmann-Wellenhof, R. Wurm and H. P. Soyer, "Telemedicine and teledermatology: Past, present and future," JDDG, vol. 6, pp. 106-112, 2008.

[111] V. F. Pamplona, A. Mohan, M. M. Oliveira and R. Raskar, "Dual of Shack-Hartmann Optometry Using Mobile Phones," Proceedings of Frontiers in Optics, Adaptive Optics for the Eye (FTuB), 2010.

[112] S. Wang, X. Zhao, I. Khimji, R. Akbas, W. Qiu, D. Edwards, D. W. Cramer, B. Ye and U. Demirci, "Integration of cell phone imaging with microchip ELISA to detect ovarian cancer HE4 biomarker in urine at the point-of-care." Lab Chip, vol. 11, pp. 3411-3418, 2011.

[113] A. W. Martinez, S. T. Phillips, E. Carrilho, S. W. Thomas III, H. Sindi and G. M. Whitesides, "Simple telemedicine for developing regions: Camera phones and paper- based microfluidic devices for real-time, off-site diagnosis," Anal. Chem., vol. 80, pp. 3699-3707, 2008.

[114] A. García, M. M. Erenas, E. D. Marinetto, C. A. Abad, I. De Orbe-Paya, A. J. Palma and L. F. Capitán-Vallvey, "Mobile phone platform as portable chemical analyzer," Sens Actuators, B, vol. 156, pp. 350-359, 2011.

[115] K. Cantrell, M. M. Erenas, I. De Orbe-Payá and L. F. Capitán-Vallvey, "Use of the hue parameter of the hue, saturation, value color space as a quantitative analytical parameter for bitonal optical sensors," Anal. Chem., vol. 82, pp. 531-542, 2010.

[116] S. Paciornik, A. V. Yallouz, R. C. Campos and D. Gannerman, "Scanner image analysis in the quantification of mercury using spot-tests," J Brazil Chem Soc, vol. 17, pp. 156- 161, 2006.

[117] M. H. Saad, H. I. Saleh, H. Konbor and M. Ashour, "Image Retrieval Based on Integration Between YCbCr Color Histogram and Texture Feature," Int. J. Comput. Theory Eng., vol. 3, pp. 701-706, 2011.

[118] L. Shen, M. Ratterman, D. Klotzkin and I. Papautsky, "A CMOS optical detection system for point-of-use luminescent oxygen sensing," Sens. Actuators B, vol. 155, pp. 430-435, 2011.

[119] C. S. McCamy, H. Marcus and J. G. Davidson, "COLOR-RENDITION CHART." J Appl Photogr Eng, vol. 2, pp. 95-99, 1976.

[120] R. T. Marcus, "The measurement of color," Color for Science, Art and Technology, pp. 31-96, 1998.

[121] R. G. Kuehni, Color: An Introduction to Practice and Principles, 1997.

[122] B. C. Thompson, P. Schottland, K. Zong and J. R. Reynolds, "In situ colorimetric analysis of electrochromic polymers and devices," Chem Mater, vol. 12, pp. 1563-1571, 2000.

[123] C. Arbizzani, M. G. Cerroni and M. Mastragostino, "Polymer-based symmetric electrochromic devices," Solar Energy Mater. Solar Cells, vol. 56, pp. 205-211, 1999.

[124] G. Wyszecki and W. S. Stiles, Color Science: Concepts and Methods, Quantitative Data and Formulae, 1982.

[125] P. Green and L. MacDonald, Colour Engineering: Achieving Device Independent Colour, 2002.

[126] W. J. Marshall and S. K. Bangert, Clinical Biochemistry: Metabolic and Clinical Aspects, 1995.

[127] X. Cui, L. M. Lee, X. Heng, W. Zhong, P. W. Sternberg, D. Psaltis and C. Yang, "Lensless high-resolution on-chip optofluidic microscopes for Caenorhabditis elegans and cell imaging," Proc. Natl. Acad. Sci. U. S. A., vol. 105, pp. 10670-10675, 2008.

[128] U. A. Gurkan, S. Moon, H. Geckil, F. Xu, S. Wang, T. J. Lu and U. Demirci, "Miniaturized lensless imaging systems for cell and microorganism visualization in point-of-care testing," Biotechnol. J., vol. 6, pp. 138-149, 2011.

[129] W. Bishara, U. Sikora, O. Mudanyali, T. -. Su, O. Yaglidere, S. Luckhart and A. Ozcan, "Holographic pixel super-resolution in portable lensless on-chip microscopy using a fiber-optic array," Lab Chip, vol. 11, pp. 1276-1279, 2011.

[130] T. Tanaka, T. Saeki, Y. Sunaga and T. Matsunaga, "High-content analysis of single cells directly assembled on CMOS sensor based on color imaging," Biosens. Bioelectron., vol. 26, pp. 1460-1465, 2010.

[131] H. Zhu, O. Yaglidere, T. -. Su, D. Tseng and A. Ozcan, "Cost-effective and compact wide-field fluorescent imaging on a cell-phone," Lab Chip, vol. 11, pp. 315-322, 2011.

[132] H. Zhu, S. Mavandadi, A. F. Coskun, O. Yaglidere and A. Ozcan, "Optofluidic fluorescent imaging cytometry on a cell phone," Anal. Chem., vol. 83, pp. 6641-6647, 2011.

[133] L. Shen, M. Ratterman, D. Klotzkin and I. Papautsky, "Use of a low-cost CMOS detector and cross-polarization signal isolation for oxygen sensing," IEEE Sens. J., vol. 11, pp. 1359-1360, 2011.