Broadly Available Imaging Devices Enable High-Quality Low-Cost Photometry

This paper demonstrates that, for applications in resource-limited environments, expensive microplate spectrophotometers—which are used in many central laboratories for parallel measurement of absorbance of samples—can be replaced by photometers based on inexpensive and ubiquitous, consumer electronic devices (e.g., scanners and cell-phone cameras). Two devices—i) a flatbed scanner operating in transmittance mode and ii) a camera-based photometer constructed from a cell phone camera, a planar light source, and a cardboard box—demonstrate the concept. These devices illuminate samples in microtiter plates from one side, and use the RGB-based imaging sensors of the scanner/camera to measure the light transmitted to the other side. The broadband absorbance of samples (RGB-resolved absorbance) can be calculated using the RGB color values of only three pixels per micro-well. Rigorous theoretical analysis establishes a well-defined relationship between the absorbance spectrum of a sample and its corresponding RGB-resolved absorbance. The linearity and precision of measurements performed with these low-cost photometers on different dyes that absorb across the range of the visible spectrum, and chromogenic products of assays (e.g., enzymatic, ELISA) demonstrates that these low-cost photometers can be used reliably in a broad range of chemical and biochemical analyses. The ability to perform accurate measurements of absorbance on liquid samples, in parallel and at low cost, would enable testing, typically reserved for well-equipped clinics and laboratories, to be performed in circumstances where resources and expertise are limited.


INTRODUCTION
Absorbance spectroscopy is the most common analytical technique used for chemical and biochemical analyses. Solution-based test kits (e.g., tube test kits and microtiter plate ELISA kits) are commercially available for hundreds of important analytes (e.g., metabolites, proteins, environmental pollutants), can provide accurate and precise results, and are inexpensive (the cost-per-sample of these test kits is typically < $5). Although these characteristics make absorbance spectroscopy attractive for use in laboratories and clinics across the world, the required instrumentation (e.g., spectrophotometers and microplate spectrophotometers) can cost from $2,000 to $50,000, depending on their specifications, versatility, and throughput.
Laboratories and clinics in low-and middle-income countries typically i) cannot afford this expensive instrumentation; ii) do not have access to the components, personnel, and expertise required to maintain the equipment; and/or iii) lack the infrastructure (e.g., continuous and stable electrical power) necessary to keep it running. 1 A simple, low-cost alternative to conventional spectrophotometric equipment would greatly reduce the barriers to providing modern medical and environmental testing to low-and middle-income countries.
Here, we report the adaptive use of ubiquitous consumer electronic devices to perform low-cost (< US$100) photometry suitable for use in resource-limited environments. We demonstrate that i) a commercially-available flatbed scanner with a transmittance mode for imaging photographic films, and ii) a portable, camera-based photometer-constructed from a paper box that houses a white, LED-based, planar light source for sample illumination, and a cell-phone camera (which is provided by the user) as a detector-can be used to perform sensitive absorbance measurements on multiple liquid samples stored in a microtiter well plate, in parallel, with accuracy and precision comparable to microplate spectrophotometers.
To develop low-cost photometers for measuring the absorbance of liquid samples, other efforts have typically used colored, light-emitting diodes (LEDs) to illuminate a liquid sample from one side, and a photodetector (photoresistor, phototransistor, or photodiode) on the other side to detect the light transmitted through the sample. [20][21][22][23][24][25][26][27][28][29] For example, we and others have constructed low-cost photometers based on this principle to measure i) antibodies against HIV in serum, 21 ii) hemoglobin [22][23] in blood, and iii) fluoride 27 and mercury 29 in water samples.
Although low-cost photometers based on this approach have filled important needs in specific analytical applications (e.g., for measuring hemoglobin 22 ), they have two important limitations: i) they are not sufficiently versatile to perform measurements at multiple wavelengths; that is, they can only perform measurements at the specific wavelengths defined by the LED used, and ii) they are incapable of high-throughput measurements using microtiter plates (which typically have 96 wells).
To address the limitations of systems that use narrowband LEDs for illumination, others have performed optical measurements of liquid samples using broadband light emitted by flatbed scanners [30][31] or digital displays (e.g., computer monitors, 32-36 tablet computers, 37 and digital photo displays 26 ), and used the image sensors found in flatbed scanners, 30 [30][31] The reflectancebased approach, however, is problematic because the variability of the color values across the sample-due to variability and complexity of the optical path length of reflected light-greatly complicates the calculation of absorbance and the precision of measurements. An alternative approach uses the light transmitted through the sample to measure absorbance. Fillipini and coworkers were the first to use consumer electronic devices to measure the transmittance of liquid samples. 32 They used screens from computers (CRT or LCD) to display a controlled sequence of colors that illuminates the sample from one side, while a web camera recorded the transmitted light from the opposite side. A Matlab-based program decomposed the video file into individual frames to extract the information of light intensity as gray level. This approach, and other similar approaches that have used consumer electronic devices, [30][31][32][33][34][35][36][37] have two important drawbacks: i) they require custom software to extract the intensity of the light from the image of a sample and ii) they require digital displays, which are expensive, and, in some cases, not portable (e.g., monitors), as light sources. Furthermore, although transmittance should, in principle, yield more precise and better-understood results than reflectance, previous efforts have not established a clear and well-defined relationship between the analytical signal and absorbance spectrum. This paper has three objectives: i) to outline the conditions required for performing highquality, high-throughput measurement of absorbance using inexpensive photometers based on broadly available consumer electronic devices (flatbed scanners and cell-phone cameras), ii) to establish a well-defined relationship between the absorbance of light (as measured by these commercial image sensors) and the absorbance spectra of samples (as measured by a laboratory 6 spectrophotometer) and iii) to demonstrate the applicability of inexpensive photometers based on broadly available consumer electronic devices in a broad range of clinical, environmental, and chemical analyses.
In this work, we used two low-cost photometers-a flatbed scanner and a camera-based photometer-that employ white, broadband LEDs as light sources and CCD/CMOS image sensors as photodetectors. By placing a microtiter plate (containing the samples) between the light source and the photodetector, we captured the images of the samples. In this configuration, the light source illuminates the sample with visible light (~ 400 -700 nm) from one side, and the pixel sensors of the imaging sensor digitize the intensity of the light, transmitted to the other side, into three color channels: red (R), green (G), and blue (B). An image of a microtiter plate contains images of several samples that can be analyzed separately, and therefore, multiple liquid samples can be analyzed, in parallel. By using the color values of a sample, we calculated the broadband "RGB-resolved absorbance" of the sample. We compared the "RGB-resolved absorbance" of solutions of 11 dyes to the peak absorbance measured using a microplate spectrophotometer, and we established a rigorous relationship between the "RGB-resolved absorbance" of a sample and the measured absorbance spectrum of an analyte. We demonstrated that the flatbed scanner and the camera-based photometer could replace microplate spectrophotometers in a broad range of biochemical and environmental assays and enable highquality photometric assays to be performed in low-resource and remote settings.

EXPERIMENTAL SECTION
Materials and Chemicals. We purchased a flatbed scanner (Epson Perfection V500 photo, Epson; $90 USD) from www.amazon.com and a low-cost, planar light source (edge-lit 7 LED backlight for iPhone 5; $6 USD) from www.Tmart.com. We also used a cell-phone camera (OPTIMUS F3 4G LTE, LG). Absorbance measurements were performed using a microplate spectrophotometer (SpectraMax M2, Molecular Devices). All chemicals and reagents were used as received without further purification (see Supporting Information for the suppliers). We used Image J ® or Microsoft Paint ® to read the color intensity (RGB values) of a pixel of an image.
Configuration of the Devices. The flatbed scanner employed a CCD image sensor in a strip, and operated in two different modes: reflectance and transmittance. The transmittance mode is designed for imaging photographic films, and uses a built-in transparency unit that employs a planar light source in a strip ( Figure 1A). Both the light source of the transparency unit and the image sensor scan across the imaging surface of the scanner simultaneously to capture an image. We placed a 96-well microtiter plate containing samples into the flatbed scanner (that is, between the imaging surface of the scanner and the transparency unit) ( Figure   1A). The imaging area of the flatbed scanner in transmittance mode ( Figure 2B shows how we arranged a microtiter plate (with samples) above the planar light source, which was large enough to illuminate 32 wells of a standard 96-microtiter plate. In this arrangement, we required three separate images (see Figure Figure 2D), which relies on ambient light (i.e., no planar light source or box). We read the color intensity (RGB values) of all the pixels in each well and estimated the distribution of the RGB values of the pixels in each well.

Absorbance Measurements of Chromogenic Compounds. We prepared solutions of 11
chromogenic compounds (disperse orange 3, methyl orange, fluorescein, 2,2-diphenyl-1picrylhydrazyl (DPPH), eosin Y, rhodamine B, trypan blue, prussian blue, malachite green, methylene blue, chlorophyll b) at different concentrations (detailed in the Supporting Information) and measured the absorption spectra A(λ) and peak absorbance Apeak of these solutions using the microplate spectrophotometer. Importantly, these compounds exhibit absorbance peaks with very different spectral characteristics (e.g., shape, position, and intensity of the spectral peak) that span the entire visible spectrum (see Figure  more suitable for absorbance peaks between 400-505 nm, the green color value is more suitable for peaks between 505-580 nm and the red value for absorbance peaks between 580-700 nm. We used the above guidelines for the selection of the most appropriate color value (R, G, or B) to be used for the estimation of the RGB-resolved absorbance values of each solution.

Applications in Biochemical and Environmental Analyses.
To demonstrate the use of the flatbed scanner and the camera-based photometer in laboratory and clinical applications, we compared the performance of these devices vs. a microplate spectrophotometer in six photometric assays for the determination of: i) lactic acid in human serum using an enzymatic assay; ii) low-density lipoproteins (LDL) in human serum using a competitive ELISA; iii) anti-Treponema pallidum antibodies in human serum using an indirect ELISA; iv) total protein in solutions of antibodies using the Bicinchoninic acid (BCA) assay; v) hemoglobin in whole blood using the Cyanomethemoglobin assay; and vi) nitrite ion in samples of water using the Griess assay (see Supporting Information for details). For all assays, we used commercially available test kits or ready-to-use reagents (see the experimental details in the Supporting Information).

RESULTS AND DISCUSSION.
Comparison between Transmittance and Reflectance Measurements. Previous approaches to low-cost optical analysis of samples in microtiter plates have used flatbed scanners in reflectance mode. [32][33] Using reflected light to estimate absorbance, however, is problematic because the path that reflected light takes through each sample is complex. Depending on the geometry and divergence angle of the light source, the many different air/liquid/plastic interfaces present around the sample scatter the light at multiple locations and in different ways. This scattering cause the colors (RGB values) of the image to be non-uniform across each well. 32 This variation in the pixel values originates from the angled configuration of the light source and the photodetector used by the scanner when imaging in reflectance mode. 18 In an attempt to compensate for these artifacts, approaches based on reflectance-mode imaging use custom software to spatially average the RGB values of all pixels within of each well, and then to calibrate the relationship between the average RGB values and the concentration of the analyte. [32][33] Establishing a well-understood relationship between these mean values and the concentration of the analyte, however, remains complicated for two reasons: i) the definition of "mean" value is ambiguous because there is typically a multimodal distribution of the RGB color values across each well and ii) the complex and highly variable optical path-length precludes the establishment of a well-defined analytical relationship between the intensity of light that reaches the detector and the concentration of the analyte.
A better, simpler approach to perform low-cost, high-quality absorbance measurements is to detect the light transmitted through a sample. In this case, the alignment between the light source and the detector establishes an optical path length that is simple, well-defined, and spatially uniform across the plate. To reproduce colors better (e.g., more as they appear in real life), digital imaging sensors typically impose a nonlinear "gamma" correction before each color value ! is recorded, according to the relation, ! = ! ( ! ) ! ! , where β k is a linear correction factor and γ k is an exponential correction factor, which can be different for each color channel. The value of each color channel encoded into the captured image is therefore described by For a sample with absorbance spectrum A(λ) and for a light source with spectrum L(λ), the intensity of the spectral intensity reaching the detector is = ⋅ 10 !!(!) . The set of the recorded RGB values of each pixel of the image of the sample is therefore given by Using the equation for estimating standard absorbance ( !"#$ = −log !" Supporting Information for a detailed definition), we define the RGB-resolved absorbance A k of the sample (with measured color value ! (!) ) to a blank solution (with measured color value ! (!) ) by Equation 3.

Substitution of Equation 2 into Equation 3 yields Equation 4, which relates A k to A S (λ) and
A B (λ).
It is important to note that this general definition applies to all imaging equipment, not only broadband RGB-based sensors (see Supporting Information for details).
Through extensive characterization, we estimated S k (λ) and γ k for all the color channels of the flatbed scanner and camera-based photometer (see Supporting Information for details).
Using Equation 4 and i) the L(λ) of the transparency unit of the flatbed scanner and the planar light source of the camera-based photometer, ii) the A(λ) of 11 different dyes measured at a range of concentrations by the microplate spectrophotometer, and iii) the extracted values of S k (λ) and γ k , we estimated the expected A k for all the standard compounds. We also calculated the measured A k for all the standard compounds and we compared them to the expected A k . Figure  Because the value of A peak is always, by definition, greater than the value of A k , all slopes of the correlation lines were always greater than unity. Equation 4 shows that this behavior occurs for two important reasons: the polychromatic nature of RGB-resolved absorbance and the gamma correction. Relative to a standard narrowband measurement at λ peak , broadening the optical bandwidth of the measurement and/or misaligning λ peak with respect to the peak value of ( ) ⋅ ! ( ) can only serve to decrease the sensitivity to changes in concentration. This effect necessarily includes regions where A(λ) < A peak . In this case, the light that does not interact much (or at all) with the analyte will contribute more to the raw RGB color values ζ κ than light transmitted near λ peak ; in this case, the ratio between the raw RGB values, captured from the light transmitted through the sample relative to the blank, will tend to unity ( ! ! ! ! → 1 ! ) , and therefore, A k < A peak for any absorbance spectrum A(λ).
For example, solutions of rhodamine B and eosin Y that exhibit similar values of A peak as well as overall spectral shape A(λ) (shown in Figure S-5 in the Supporting Information) may still exhibit very different values of green-resolved absorbance A G (~ 3-4x difference) because, compared to eosin Y, the position of λ peak of rhodamine B is better-aligned with the peak of ( ) ⋅ ! ( ).
The nonlinear gamma correction that all imaging equipment imposes also serves to distort the magnitude of A k . In particular, because γ k < 1 for all the devices that we characterized, the gamma corrections always tended to further reduce A k compared to a narrowband measurement These results demonstrate that RGB-based photometers, such as the ones we demonstrated, can be useful for absorbance measurements, regardless of the shape, position, or intensity of the peak absorbance A peak of the sample. The ultimate sensitivity of the measurements to changes in concentration of the sample, however, will depend strongly on the position and the shape of A peak relative to the spectral peaks of ( ) ⋅ ! ( ) for a chosen source of illumination and imaging sensor. The better the alignment of A peak , for a chosen sample, with the peaks of ⋅ , for each color channel, the more sensitive the measurements will be. Importantly, despite the reduced sensitivity of the different RGB-based measurements relative to those performed with a microplate spectrophotometer, we do not observe a corresponding decrease in precision or limitof-detection. We conclude, therefore, that the sensitivity of RGB-based measurements will not limit the applicability of RGB-based photometers. In the case that increased sensitivity is desired, however, there are two strategies to achieve it. i) When developing an assay, choosing or engineering dyes with narrow peaks that are centered on the peaks of independent-as may be the case if some or all of the absorption peaks occur within the same color channel-or if discrimination between more than three compounds is desired, it may be possible to satisfy these condition by adding extra degrees of freedom, for example, by comparing multiple measurements with and without narrow-band color filters.

Applications in Biochemical and Environmental Analyses.
We compared the performance of the flatbed scanner and camera-based photometer against a microplate spectrophotometer (used as the laboratory standard) as detectors in six important assays for the determination of: i) lactic acid in human serum; ii) low-density lipoproteins (LDL) in human serum; iii) anti-Treponema pallidum antibodies in human serum; iv) total protein in solutions of antibodies; v) hemoglobin in whole blood; and vi) nitrite ion in samples of water. We chose these assays because: i) they are commonly used in biochemical or environmental analyses and ii) they are representative of common chemical assays (e.g., enzymatic assay, immunoassay, and chromogenic assay). The lactate assay is used to measure the concentration of lactic acid in serum and is representative of other assays to detect common serum components (e.g., glucose, creatinine, alkaline phosphatase). The LDL ELISA is used to measure the concentration of lowdensity lipoproteins in serum and is representative of other immunoassays of common biomarkers (e.g., carcinoembryonic antigen, troponin). The anti-Treponema pallidum IgG ELISA is used to detect syphilis and is representative of other qualitative ELISA tests to detect infectious diseases (e.g., dengue fever, Ebola, malaria). The Cyanmethemoglobin assay is the gold-standard method to detect hemoglobin in blood. The Bicinchoninic acid (BCA) assay is the gold-standard method to measure the total protein concentration of purified biological samples.
The Griess assay is used to measure the concentration of nitrite ions in water samples and is representative of other simple chromogenic assays for the detection of pollutants in environmental samples (e.g., chlorine, metal ions). Using a t-test, we found that the results we obtained using the flatbed scanner and the camera-based photometer were not significantly different than those obtained with the microplate spectrophotometer. From these data, we conclude that these lowcost photometers can replace a microplate spectrophotometer in many, routine photometric assays. Using the flatbed scanner as a photometer could be very useful in clinics that perform many different analyses on dozens of samples per day. For in-field absorbance measurements on multiple samples, the camera-based photometer may prove more suitable due its portability (especially because of the lightness, low power consumption, and mobility of a cell phone compared to a scanner). After using either low-cost photometer to capture images of samples, a user could analyze the images locally and/or send them to a central location for remote analysis, interpretation, and logging.

CONCLUSIONS
Although devices for digital imaging have been used to measure absorbance in the past, fundamental questions about the quality and the meaning of these optical measurements have not previously been answered. Our optical model, experiments, and analyses yield four conclusions 20 about RGB-based measurements of absorbance. i) Transmittance-mode imaging is superior to reflectance-mode imaging because the color-uniformity of neighboring pixels in transmittance measurements enables any software for imaging be used for reading the RGB values, and therefore the absorbance, easily and accurately. ii) The absorbance spectrum of any sample, and therefore, the peak absorbance value collected by narrow-band measurements, is analytically related to the broadband absorbance measured by an RGB-based image sensor. iii) The highest sensitivity measurements are obtained from the RGB color channel that best overlaps with the peak absorbance of the sample. iv) The accuracy, precision, and linearity of these RGB-based measurements of absorbance are comparable to measurements of the peak absorbance performed by conventional (and expensive) laboratory-based equipment in a wide range of routine photometric assays.
These results show that either a flatbed scanner for imaging photographic films, or a camera-based photometer (assembled from a planar light source, a cardboard box and a cellphone camera), can be used as low-cost photometers without major modifications or complex training of operators. Four important characteristics show that these photometers-both based on ubiquitous, low-cost reliable consumer electronic devices-are effective for all common, arraybased spectrophotometric assays. i) They are versatile enough to detect analytes that have very different absorption spectra at different wavelengths within the visible spectrum, regardless of the wavelength of the peak absorbance or the shape of the peak. ii) They are capable of highthroughput measurement of many samples at a time , depending on the configuration and the device), because every single pixel sensor of the imaging device works as a single detector.
Independent photometric measurements can, therefore, be performed simultaneously. iii) They are affordable and can be easily replaced. iv) They are portable and lightweight.

21
There are two potential disadvantages of using the RGB-resolved absorbance in place of the peak absorbance. i) Measurements of the RGB-resolved absorbance are less sensitive by approximately 1.3 -16x (depending on the device and selected color channel) than measurements of narrowband peak absorbance due to polychromatic nature of RGB-resolved absorbance and nonlinear gamma corrections. ii) The low spectral resolution (i.e., broad bandwidth) of RGB channels does not provide sufficient spectral information for unique discrimination between more than three compounds in a mixture, at a time. Adding color filters to the image sensor could improve both the sensitivity and selectivity of measurements by narrowing the bandwidths of the measurements around the absorption peaks of the dyes. To improve sensitivity and specificity, analytical laboratories could develop and validate low-cost kits specifically tailored for use with RGB-based photometers-that is, they could select or engineer dyes with absorption peaks that coincide best with the RGB channels of common image sensors than those in current test kits. In general, however, the reduced sensitivity of RGBresolved measurements relative to measurements of the peak absorbance by a laboratory spectrophotometer does not affect the quality (accuracy, precision, and linearity) of the results, and therefore, is not a practical limitation. The inability to identify unknown samples with complex spectra uniquely, or many samples in a mixture, is also not a major limitation because most applications in biochemical and environmental analysis involve the assay of a single dye with a well-known absorbance spectrum that exhibits a single peak.
These simple and affordable alternatives to conventional equipment are suitable for performing precise and accurate measurements of absorbance in resource-limited laboratories.
The camera-based photometer is particularly well-suited to in field measurements. A range of laboratories that cannot afford to buy a new microplate reader, or repair or replace a broken one, could use scanners for film imaging for research, education, or routine measurements. For example, laboratories that specialize in biochemistry and organic chemistry might use these devices to monitor chemical and biochemical processes. Overall, this approach to low-cost photometry has the potential to reduce the difficulty and cost of common analysis, and may facilitate the wider use of such analysis for screening of diseases (e.g., infectious diseases, chronic diseases, malnutrition), comprehensive testing of the quality of food and water (e.g., bacterial contamination, evaluation of authenticity), and enable simple environmental monitoring (e.g., heavy metals, poisonous gases).

Supporting Information
The supporting information section includes the following: details about the fabrication of the camera-based photometer; a comprehensive description of the experimental procedures; details about the estimation of the spectral sensitivity S(λ) and the gamma correction factors; definition of peak absorbance; reasoning why RGB color system is more suitable than grayscale or HSV; tables and figures of calibration lines (peak and RGB-resolved absorbance vs. concentration) and correlation lines (predicted vs. measured RGB-resolved absorbance and peak absorbance vs. RGB-resolved absorbance) for the 11 different dyes; tables of analytical characteristics of calibration lines and of the results for the six photometric assays; images of a microtiter plate inside the camera-based photometer; absorbance spectra of the 11 dyes and the chromogenic products of the six photometric assays; figures outlining the steps to estimate A k of a solution for the flatbed scanner and the camera-based photometer. This material is available free of charge via the Internet at http://pubs.acs.org.

Author Contributions
The manuscript was written through contributions of all authors. / All authors have given approval to the final version of the manuscript.

ACKNOWLEDGMENT
This work was supported by Bill and Melinda Gates Foundation (award number 51308) and by a contract from the Defense Threat Reduction Agency (award number HDTRA1-14-C-0037).