Slanted channel microfluidic chip for 3D fluorescence ... · 9. P. Sarder and A. Nehorai, “Deconvolution methods for 3-D ﬂuorescence microscopy images,” IEEE Signal Processing

Slanted channel microfluidic chip for 3Dfluorescence imaging of cells in flow

VEERENDRA KALYAN JAGANNADH,1 MARK. D. MACKENZIE,2 PARAMAPAL,3 AJOY K. KAR,2,*AND SAI SIVA GORTHI1

1Optics & Microfluidics Instrumentation Lab, Department of Instrumentation & Applied Physics, IndianInstitute of Science, Bangalore 560012, India2Institute of Photonics and Quantum Sciences, School of Engineering and Physical Sciences, Heriot WattUniversity, Edinburgh, EH14 4AS, UK3Robert Bosch Centre for Cyber-Physical Systems, Indian Institute of Science, Bangalore 560012, India.*[email protected]

Abstract: Three-dimensional cellular imaging techniques have become indispensable tools inbiological research and medical diagnostics. Conventional 3D imaging approaches employ focalstack collection to image different planes of the cell. In this work, we present the design andfabrication of a slanted channel microfluidic chip for 3D fluorescence imaging of cells in flow.The approach employs slanted microfluidic channels fabricated in glass using ultrafast laserinscription. The slanted nature of the microfluidic channels ensures that samples come into andgo out of focus, as they pass through the microscope imaging field of view. This novel approachenables the collection of focal stacks in a straight-forward and automated manner, even withoff-the-shelf microscopes that are not equipped with any motorized translation/rotation samplestages. The presented approach not only simplifies conventional focal stack collection, but alsoenhances the capabilities of a regular widefield fluorescence microscope to match the featuresof a sophisticated confocal microscope. We demonstrate the retrieval of sectioned slices ofmicrospheres and cells, with the use of computational algorithms to enhance the signal-to-noiseratio (SNR) in the collected raw images. The retrieved sectioned images have been used tovisualize fluorescent microspheres and bovine sperm cell nucleus in 3D while using a regularwidefield fluorescence microscope. We have been able to achieve sectioning of approximately200 slices per cell, which corresponds to a spatial translation of ∼ 15 nm per slice along theoptical axis of the microscope.

Published by The Optical Society under the terms of the Creative Commons Attribution 4.0 License. Furtherdistribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, andDOI.

OCIS codes: (180.0180) Microscopy; (180.2520) Fluorescence microscopy; (220.0220) Optical design and fabrication;

(220.4000) Microstructure fabrication; (320.0320) Ultrafast optics; (350.0350) Other areas of optics; (350.3390) Laser

materials processing.

References and links1. B. Alberts, D. Bray, K. Hopkin, A. Johnson, J. Lewis, M. Raff, K. Roberts, and P. Walter, Essential Cell Biology,

Fourth Edition (Garland Science, 2013).2. K. T. Patton and G. A. Thibodeau, Structure & Function of the Body (Elsevier Health Sciences, 2016).3. S. M. Rafelski and W. F. Marshall, “Building the cell: design principles of cellular architecture,” Nat Rev Mol Cell

Biol 9, 593–602 (2008).4. G. M. Church, M. B. Elowitz, C. D. Smolke, C. A. Voigt, and R. Weiss, “Realizing the potential of synthetic

biology,” Nat Rev Mol Cell Biol 15, 289–294 (2014).5. D. Zink, A. H. Fischer, and J. A. Nickerson, “Nuclear structure in cancer cells,” Nat Rev Cancer 4, 677–687 (2004).6. M. Saxena, G. Eluru, and S. S. Gorthi, “Structured illumination microscopy,” Adv. Opt. Photon. 7, 241–275 (2015).7. G. J. Brakenhoff, H. T. M. van der Voort, E. A. van Spronsen, W. a. M. Linnemans, and N. Nanninga, “Three-

dimensional chromatin distribution in neuroblastoma nuclei shown by confocal scanning laser microscopy,” Nature317, 748–749 (1985).

Vol. 24, No. 19 | 19 Sep 2016 | OPTICS EXPRESS 22144

#269179 Journal © 2016

http://dx.doi.org/10.1364/OE.24.022144 Received 23 Jun 2016; revised 10 Aug 2016; accepted 27 Aug 2016; published 15 Sep 2016

http://creativecommons.org/licenses/by/4.0/

8. J.-B. Sibarita, “Deconvolution Microscopy,” in “Microscopy Techniques,” J. Rietdorf, ed. (Springer BerlinHeidelberg, 2005), no. 95 in Advances in Biochemical Engineering, pp. 201–243. DOI: 10.1007/b102215.

9. P. Sarder and A. Nehorai, “Deconvolution methods for 3-D fluorescence microscopy images,” IEEE SignalProcessing Magazine 23, 32–45 (2006).

10. J. Mertz, “Optical sectioning microscopy with planar or structured illumination,” Nat Meth 8, 811–819 (2011).11. L. Shao, P. Kner, E. H. Rego, and M. G. L. Gustafsson, “Super-resolution 3d microscopy of live whole cells using

structured illumination,” Nat Meth 8, 1044–1046 (2011).12. C. J. R. Sheppard, “Three-dimensional phase imaging with the intensity transport equation,” Appl. Opt. 41, 5951–

5955 (2002).13. M. Mir, B. Bhaduri, R. Wang, R. Zhu, and G. Popescu, “Chapter 3 - Quantitative Phase Imaging,” in “Progress in

Optics,” , vol. 57 of Progress in Optics E. Wolf, ed. (Elsevier, 2012), pp. 133–217.14. D. A. Basiji, W. E. Ortyn, L. Liang, V. Venkatachalam, and P. Morrissey, “Cellular Image Analysis and Imaging by

Flow Cytometry,” Clinics in laboratory medicine 27, 653–viii (2007).15. K. Goda, A. Ayazi, D. R. Gossett, J. Sadasivam, C. K. Lonappan, E. Sollier, A. M. Fard, S. C. Hur, J. Adam,

C. Murray, C. Wang, N. Brackbill, D. D. Carlo, and B. Jalali, “High-throughput single-microparticle imaging flowanalyzer,” PNAS 109, 11630–11635 (2012).

16. E. Schonbrun, S. S. Gorthi, and D. Schaak, “Microfabricated multiple field of view imaging flow cytometry,” LabChip 12, 268–273 (2012).

17. Y. Sung, N. Lue, B. Hamza, J. Martel, D. Irimia, R. R. Dasari, W. Choi, Z. Yaqoob, and P. So, “Three-dimensionalholographic refractive-index measurement of continuously flowing cells in a microfluidic channel,” Phys. Rev.Applied 1, 014002 (2014).

18. V. K. Jagannadh, M. D. Mackenzie, P. Pal, A. K. Kar, and S. S. Gorthi, “Imaging Flow Cytometry With FemtosecondLaser-Micromachined Glass Microfluidic Channels,” IEEE J of Sel. Topics in Quantum Electron. 21, 370–375(2015).

19. N. C. Pégard and J. W. Fleischer, “Three-dimensional deconvolution microfluidic microscopy using a tilted channel,”J. Biomed. Opt 18, 040503–040503 (2013).

20. S. S. Gorthi and E. Schonbrun, “Phase imaging flow cytometry using a focus-stack collecting microscope,” Opt.Lett. 37, 707–709 (2012).

21. J. Wu, J. Li, and R. K. Y. Chan, “A light sheet based high throughput 3d-imaging flow cytometer for phytoplanktonanalysis,” Opt. Express 21, 14474–14480 (2013).

22. Y. Xia and G. M. Whitesides, “Soft lithography,” Ann. Rev. of Mater. Sci. 28, 153–184 (1998).23. J. C. Love, J. R. Anderson, and G. M. Whitesides, “Fabrication of Three-Dimensional Microfluidic Systems by Soft

Lithography,” MRS Bulletin 26, 523–528 (2001).24. D. Choudhury, J. R. Macdonald, and A. K. Kar, “Ultrafast laser inscription: perspectives on future integrated

applications,” Laser Photon. Rev. pp. 827–846 (2014).25. M. Beresna, M. Gecevic̆ius, and P. G. Kazansky, “Ultrafast laser direct writing and nanostructuring in transparent

materials,” Adv. Opt. Photon. 6, 293 (2014).26. K. Sugioka and Y. Cheng, “Ultrafast lasers-reliable tools for advanced materials processing,” Light Sci Appl 3, e149

(2014).27. F. He, Y. Liao, J. Lin, J. Song, L. Qiao, Y. Cheng, and K. Sugioka, “Femtosecond Laser Fabrication of

Monolithically Integrated Microfluidic Sensors in Glass,” Sensors 14, 19402–19440 (2014).28. P. Paié, F. Bragheri, R. M. Vazquez, and R. Osellame, “Straightforward 3d hydrodynamic focusing in femtosecond

laser fabricated microfluidic channels,” Lab Chip 14, 1826–1833 (2014).29. B.-B. Xu, Y.-L. Zhang, H. Xia, W.-F. Dong, H. Ding, and H.-B. Sun, “Fabrication and multifunction integration of

microfluidic chips by femtosecond laser direct writing,” Lab Chip 13, 1677–1690 (2013).30. D. Choudhury, D. Jaque, A. Rodenas, W. T. Ramsay, L. Paterson, and A. K. Kar, “Quantum dot enabled thermal

imaging of optofluidic devices,” Lab Chip 12, 2414–2420 (2012).31. D. Choudhury, W. T. Ramsay, R. Kiss, N. A. Willoughby, L. Paterson, and A. K. Kar, “A 3d mammalian cell

separator biochip,” Lab Chip 12, 948 (2012).32. P. Paié, F. Bragheri, A. Bassi, and R. Osellame, “Selective plane illumination microscopy on a chip,” Lab Chip

(2016).33. E. J. van Beers, L. Samsel, L. Mendelsohn, R. Saiyed, K. Y. Fertrin, C. A. Brantner, M. P. Daniels, J. Nichols, J. P.

McCoy, and G. J. Kato, “Imaging flow cytometry for automated detection of hypoxia-induced erythrocyte shapechange in sickle cell disease,” Am. J. Hematol. 89, 598–603 (2014).

34. L. M. Niswander, K. E. McGrath, J. C. Kennedy, and J. Palis, “Improved quantitative analysis of primary bonemarrow megakaryocytes utilizing imaging flow cytometry,” Cytometry A 85, 302–312 (2014).

35. Min Gu, Advanced Optical Imaging Theory, 75 (Springer, 2000).36. R. C. Gonzalez, Digital Image Processing (Pearson Education India, 2009).37. A. Mosleh, J. M. P. Langlois, and P. Green, “Image Deconvolution Ringing Artifact Detection and Removal via

PSF Frequency Analysis,” in “Computer Vision ECCV 2014,” D. Fleet, T. Pajdla, B. Schiele, and T. Tuytelaars,eds. (Springer International Publishing, 2014), no. 8692 in Lecture Notes in Computer Science, pp. 247–262. DOI:10.1007/978-3-319-10593-2_17.

38. N. G. Cassuto, A. Hazout, I. Hammoud, R. Balet, D. Bouret, Y. Barak, S. Jellad, J. M. Plouchart, J. Selva, and


C. Yazbeck, “Correlation between DNA defect and sperm-head morphology,” Reprod. Biomed. Online 24, 211–218 (2012).

39. T. G. Cooper, E. Noonan, S. von Eckardstein, J. Auger, H. W. G. Baker, H. M. Behre, T. B. Haugen, T. Kruger,C. Wang, M. T. Mbizvo, and K. M. Vogelsong, “World Health Organization reference values for human semencharacteristics,” Hum. Reprod. Update 16, 231–245 (2010).

40. G. Di Caprio, M. A. Ferrara, L. Miccio, F. Merola, P. Memmolo, P. Ferraro, and G. Coppola, “Holographic imagingof unlabelled sperm cells for semen analysis: a review,” J Biophotonics 8, 779–789 (2015).

41. M. A. Ferrara, G. Di Caprio, S. Managó, A. De Angelis, L. Sirleto, G. Coppola, and A. C. De Luca, “Label-freeimaging and biochemical characterization of bovine sperm cells,” Biosensors (Basel) 5, 141–157 (2015).

42. P. Memmolo, G. Di Caprio, C. Distante, M. Paturzo, R. Puglisi, D. Balduzzi, A. Galli, G. Coppola, andP. ao, “Identification of bovine sperm head for morphometry analysis in quantitative phase-contrast holographicmicroscopy,” Opt Express 19, 23215–23226 (2011).

43. M. Haifler, P. Girshovitz, G. Band, G. Dardikman, I. Madjar, and N. T. Shaked, “Interferometric phase microscopyfor label-free morphological evaluation of sperm cells,” Fertil. and Steril. 104, 43–47.e2 (2015).

44. X. Heng, D. Erickson, L. R. Baugh, Z. Yaqoob, P. W. Sternberg, D. Psaltis, and C. Yang, “Optofluidic microscopy-amethod for implementing a high resolution optical microscope on a chip,” Lab Chip 6, 1274–1276 (2006).

45. V. K. Jagannadh, B. P. Bhat, L. A. N. Julius, and S. S. Gorthi, “High-throughput miniaturized microfluidicmicroscopy with radially parallelized channel geometry,” Anal Bioanal Chem 408, 1909–1916 (2016).

1. Introduction

Cells, the fundamental building blocks of life constitute all living organisms from simple singlecelled organisms such as bacteria to vastly complex multicellular organisms like humans andwhales [1]. The extreme finesse of cellular functioning continues to elude scientists working indifferent fields related to biology and medicine, though our understanding is steadily increasing.Nature’s designs are all the more interesting, when it comes to the relationship between physicalstructure and the function of different cellular components. For example, the flagellum of asperm cell enables it to efficiently propel the cell through the female reproductive tract [2].The red blood cell is another example of design marvel, wherein the membrane deformabilityenables passage through the thinnest of capillaries to transport oxygen and carbon dioxide.Understanding design principles embedded in natural biology [3], is enabling the creation ofsynthetic cellular architectures, which are finding a wide range of applications in the fieldsof biotechnology and medicine [4]. From a medical perspective, understanding the structure-function relationship is leading to insights into disease mechanisms. These insights have enabledthe creation of effective diagnostics and therapeutics for diseases like cancer [5].

Our understanding of cellular morphological architectures has co-progressed with thedevelopments in 3D optical imaging techniques [6]. Most of the current commercially availableinstruments offer several methods for acquiring 3D spatio-morphological features of live cells.One of the most widely used methods for 3D imaging is optical sectioning using confocalmicroscopy. A good example is that of ascertaining the chromatin distribution in nuclei at asingle-cell level [7].

At the core of a confocal imaging experiment lies the process of focal stack collection [8, 9].Focal stack collection involves acquiring images of the sample, while different parts of thespecimen/cell are in the focal plane/depth of field (DOF) of the microscope objective. In otherwords, it is the process of scanning the sample along the optical axis of the microscope. Owingto the limited DOF of the microscope, different parts/sections of the biological tissue alongthe depth come into focus as the sample is scanned. The light from out-of-focus planes alsocontributes to the image formation; giving rise to defocus blur. This problem is mitigated bypositioning a pin-hole infront of the camera/detector. The pin-hole rejects/filters out light fromthe out-of-focus planes, thereby enhancing resolution. Whereas, in a conventional widefieldmicroscope, the defocus blur is computationally removed. The acquired focal stack of imagesis processed to reconstruct the 3D distribution of the fluorescence over the sample. This enablesthe assessment of fluorescence in different planes of the sample. Of late, several other methodsfor 3D fluorescence imaging have been developed, which include light sheet microscopy [10]


and structured illumination microscopy [6, 11]. These methods rely on illuminating the samplewith a sheet of light [10] or by patterning the illumination along with computational decoding[11] to extract slices/sections of the biological cells. The process of mechanical translationto acquire the focal stacks remains inherent even while employing of these sophisticatedtechniques.

Apart from confocal fluorescence microscopy, some quantitative phase imaging methods alsoemploy focal stacking. In these methods, the brightfield focal stack images are used to solvethe transport of intensity equation (TIE) [12, 13] to generate a quantitative phase map of thebiological cell. The volumetric or thickness information of the cell can be extracted from thequantitative phase map. These techniques enable live-cell imaging, as they do not require usageof chemical agents like stains; which tend to affect normal cell physiology.

In general, the most common approach to acquire a focal stack is the use of high-precisionpiezo-translation stage. Some of these translation stages tend to have a minimum incrementalmotion of about 50 nm (AG-LS25, Newport Inc.). Most of the current high-end commercialmicroscopes have similar translation stages and enable automated focal stack acquisition.

On the other hand, some optofluidic imaging techniques known as Imaging Flow Cytometry /Microfluidic Microscopy [14–20] leverage fluidic motion and its alignment with imaging opticsto enable novel imaging modalities. Some of these techniques have been shown to be capableof 3D imaging. These approaches essentially involve aligning the microfluidic device with theoptical axis at a predefined tilt [19, 20]. As the cells pass through the straight microfluidicchannel, the tilted placement of the whole device with respect to the imaging axis, ensures thatthe cells pass through the focal plane of the objective. These approaches have been shown tobe capable of 3D imaging of cells in brightfield [19], fluorescence [21] and quantitative phaseimaging [20]. Although simplified, the regular usage of microfluidic focal stack collection inconjunction with conventional microscopy systems still requires some expertise with optics.This is due to the fact that, positioning an element at a precise tilt would require operationalknowledge of opto-mechanical components. Further, the integration of additional rotationaldegrees of freedom into conventional microscope systems is not straightforward and can beprohibitively expensive. Most of the off-the-shelf microscopes may not provide enough gapbetween the objective and the sample holder to incorporate/add an additional opto-mechanicalcomponent.

A more elegant approach would be to fabricate a small microfluidic device, consistingof channels with an ‘out-of-plane’ tilt. This would allow for the device to be flat-mountedonto any of the existing microscope stages, yet at the same time have the required slant forfluidic focal stack collection. However, the incorporation of a fine oblique angle slope in themicrofluidic channel is quite challenging with conventional layer-by-layer type techniques likesoft lithography [22, 23]. In these techniques, the 3D structure is brought about by modifyingdifferent layers of the master mould. As the material modification happens in a layer-by-layermanner, fabricating a fine slope within the channel using these techniques would be quitechallenging.

In contrast to the conventional techniques, ultra-fast laser inscription (ULI) [24–26] enablesmaterial modification with inherent 3D capabilities. In ULI, the focal spot of an ultrafast laser isscanned through the volume of the material to modify properties of irradiated areas and therebyenable creation of arbitrary 3D structures. The power of ULI can be leveraged to implementcomplicated microfluidic architectures [27–32]; thereby drastically simplifying conventionalexperiments.

In this work, we demonstrate for the first time use of ULI for fabrication of a focal stackcollecting microfluidic device in glass. The device incorporates microfluidic channels with afine ‘out-of-plane’ slant/slope. The presence of this fine slant within the microfluidic channelenables collection of focal stacks of cells passing through them. In essence, the cells traverse


through the focal plane of the objective, enabling acquisition of cell images at differentdepths/focus in successive frames. This approach of slanted channel acquisition enablesautomated focal stack collection, without the need for high precision translation stages andvery high-speed cameras. Conventional optical sectioning methods, like confocal microscopyare inherently slow due to the point-based scanning approach. Whereas the presented widefieldapproach enables 3D imaging of cells, while they are in flow and so is amenable for high-throughput measurements. The acquired focal stacks have been processed using computationalalgorithms. We demonstrate the use of computational photon reassignment algorithms to restoresectional images of cells passing through the microfluidic device. As all the planes of thecell contribute to image formation in the case of each slice image, information from eachslice is recorded multiple times. This has been leveraged to significantly enhance the signal-to-noise (SNR), by computationally reassigning the photons. Thus, even while recording theimage at a lower exposure time, we demonstrate an increase in the effective exposure timeof the images. The retrieved slice images were used to reconstruct 3D morphology of thecells. The presented device allows for seamless integration of the fluidic focal stacking andsubsequent 3D imaging modality into conventional widefield microscopy systems, which arenot equipped with expensive precision automated translation stages. Moreover, the recent pasthas witnessed the early stage evolution of inherently slow conventional microscopic imagingtowards a high-throughput flow-based imaging modality. As opposed to the now conventionalsmear based imaging approaches wherein cells are stationary, flow-based high-throughputimaging modalities will eventually become the norm. This is evident from some of the recentlydemonstrated applications of imaging flow cytometry [33, 34]. The inherent slow speed ofmechanical translation in comparison to particle motion speed and acquisition rate make focalstack acquisition a challenging task in flow-based scenarios, even while using expensive, highprecision translation stages. The solution proposed here forms a key element in furthering thedevelopment of flow based 3D imaging modalities which rely on focal stack acquisition.

2. Universal fluidic focal stack collection

2.1. Device design

As mentioned earlier, the process of focal stacking involves leveraging fluidic motion to movethe sample into and out of the depth-of-field of an imaging system. The schematic of thepresented approach is shown in Fig. 1(a). As can be seen in the schematic, the surface of themicrofluidic device is flat and is parallel to the nose of the microscope objective. Whereas,the subsurface microfluidic channel present within the device is non-parallel and is tilted withrespect to the focal plane of the objective. The flat geometry of the overall device enablesmounting onto any microscope sample stage, quite similar to positioning a microscope slide.This channel geometry presents an elegant solution to the problem of focal stacking and enablesimplementation of computational imaging approaches even with off-the-shelf microscopes,without the need for modifying its opto-mechanical architecture. ULI has been employed inthis work for fabrication of such a structure with great ease. The angle of tilt is an importantdesign parameter, as it determines the pitch of focal stack for a given flow rate and acquisitionfrequency or frame rate. The angle has to be chosen appropriately for a given size of the celland the length across which the sample traverses within the microscope field-of-view (FOV).

In order to collect the focal stack of a given cell, it is essential that the vertical motion itundergoes be more than sufficient to fully scan it along the depth dimension. The schematicrepresentation of the motion of the sample is shown in Fig. 1(b). For a given tilt (θ) and FOVwidth (l) along the direction of flow, the amount of vertical motion (d) that the cell/specimenwould undergo with respect to the objective’s focal plane is given by

d = l × tan(θ) (1)


(b)

(a)

OpticalAxis

DOF

FieldofViewWidth(l)

Fig. 1. (a) Schematic Design of the Proposed ULI based Universal Focal Stack CollectionDevice. (b) Schematic representation of motion of the cell through the focal stack collectiondevice within the microscope FOV.

The FOV width (l) of a given microscope is fixed, depending upon the magnification employedand also the size of the image sensor. The user has to choose the tilt as per the size of thesample, being investigated. For a given cell size (s), the vertical motion should be at least morethan the size of the cell. In general, it is ideal to choose the depth of the microfluidic channelto be at least twice the size of the cell. This ensures that the cells pass through without anyobstruction or clogging. In the absence of a flow focusing mechanism (in the depth direction),the cell can be present anywhere within the depth of the channel. Hence, we suggest that thedesign incorporates vertical motion, which is more than twice the size of the cell; giving rise torelation (2).

θ ≥ arctan(2 × sl) (2)

This would ensure that the complete depth of the cell would be scanned, while it traverses theFOV. For example, in order to acquire the focal stack of cell, which is 20 μm in diameter - overa FOV which has a width of 300 μm, the tilt should be ideally close to 7.5◦, so that the entiredepth dimension of the cell is covered.

2.2. Device fabrication

As per the schematic shown in Fig. 1, microfluidic devices were fabricated in fused silica. TheULI technique employs focused ultrashort pulses to locally modify material properties. As themodification happens at the focal spot of a laser which can be located inside a transparentmaterial positioned on a translation stage, the technique enables a superior degree of designfreedom, than that offered by techniques such as PDMS lithography. When an ultrafast laserpulse is focused into a tight spot, the material absorbs through a nonlinear process. This non-


Fig. 2. Images of focal stack collection glass microfluidic device after etching. (a)Microscopic Top view of the device (b) Microscopic side view of a device having channelswith a 1 degree tilt. (c) Microscopic side view of a device having channels with 5 degreetilt. (d) Top view of the device displaying it’s lateral width. (e) Side of the device, showingit’s thickness.

linear process results in the formation of nanogratings which enhance the etch rate of thematerial when placed in hydrofluoric acid (HF). The selective enhancement of etch rate at theirradiated locations allows for the formation of structures within the sample.

In these experiments, an IMRA μ-Jewel femtosecond laser set to a repetition rate of 500kHz, pulse length of 360 fs and power of 350 mW was focused onto the samples mounted onan Aerotech XYZ stage, through a 0.4 NA objective. Following irradiation, the samples wereetched for approximately 13 hours in 10% HF. Tubing from Upchurch Scientific of outer andinner diameter of 360 μm and 100 μm respectively was bonded to the inlets using UV curingglue (Thorlabs MIL-A-3920).

The microfluidic devices used in these experiments consist of a series of channels with an ‘outof plane’ tilt with a width of ∼20 μm and height of ∼31 μm. The inlets and outlets were createdon the sides of the device to facilitate ease of experimenting with a conventional microscope.The imaging region is fabricated beneath the surface close to the device boundary, so as toensure compatibility with low working-distance objectives. The inlet and the imaging regionare connected via a tilted channel as shown in Fig. 2(b).

At the imaging region, the channel bifurcates into eight channels, each fabricated with a fineslant. To showcase various possibilities, we have fabricated microfluidic devices with differenttilts. Side views for 1◦ and 5◦ tilted device have been shown in Fig. 2(b) and (c) respectively.As it is evident from the images, the slant is very fine and is not noticeable for 1◦ tilted device;whereas the slant is noticeable with the 5◦ tilted device. The design ensures that as the cellspass through these channels, the cells come into and go out-of-focus of the imaging system.This allows for acquisition of focal stacks with great ease. Fig. 2(d) and (e) show images of


the microfluidic device, after insertion of tubing. The device has a lateral dimension of about4.5 mm and is only 2 mm thick. The extremely small form factor of the microfluidic devicesensures incorporation into conventional microscopes with great ease.

3. Characterization of focal stack acquisition with fluorescent microspheres

MicroscopeObjective

Microfluidic Device

(a) (b)

Fig. 3. S1. (a) Conventional wide-field fluorescence microscope fitted with digital camera.(b) Zoomed in view of the sample stage, showing the flat-mounted focus stack collectingmicrofluidic device.

The proposed approach for focal stack collection was first characterized by collectingfocal stacks of fluorescent microspheres. In these experiments, we have used fluorescenceisothiocyante (FITC) labelled microspheres of diameter - 4 μm (Thermofisher, F-8859). Thefluorescent dye (FITC) is distributed uniformly over the complete volume of these microspheres.When such a bead is optically sectioned, it is expected that slices, which have a circular disc likeshape are observed. When a sphere is sliced at different locations perpendicular to the axis, thediameter of the slice would vary with respect to its distance from the center. The central slicewould have the highest diameter, and the diameter would decrease as we move away from thecenter of the sphere. Further, as the dye is uniformly distributed it is expected that all the sliceswould have uniform intensity.

The glass microfluidic device was flat mounted onto the stage of a conventional widefieldfluorescence microscope as shown in Fig. 3. As shown in the figure, the experimental setup doesnot require anything else apart from a regular scientific digital microscope. In our experiments,we have employed the Nikon Eclipse Ci for acquiring focal stacks of fluorescent microspheresand cells. A suspension of fluorescence isothiocyante (FITC) labeled microspheres of diameter- 4 μm (Thermofisher, F-8859) was pumped through the microfluidic device. The microfluidicdevice is flat mounted on a conventional widefield fluorescence microscope and viewed under40X magnification (Nikon Plan Fluor, NA = 0.75). As the material (glass) of the microfluidicdevice does not significantly impact the imaging quality, it is expected that the system wouldenable a resolution of about 0.37 μm in the lateral direction. As per the Rayleigh criterion, thedepth resolution is about 1.9 μm. The flow of the suspension was facilitated by a syringe pump(NE-1000, New Era Pump Systems). Videos of flow were acquired using Andor Zyla 5.5 cameramounted onto the microscope. The exposure time was set to 500 μs and the frame rate was setto 200 frames per second (fps).

The acquired videos were processed in MATLAB to extract the focal stacks correspondingto individual beads. Each given frame of the video would consist of single image of the bead,focused at a particular depth. As the bead traverses from one end to the other, it is imaged atdifferent depths. When the bead first enters the field of view, it is focused close to the edge.


(a)(b)(c)

Fig. 4. (a) Raw, (b) Deconvolved Focal stack of a 4 μm fluorescent microsphere. (c) Slicesof fluorescent microsphere, after removal of artifacts. Length of Scale bar is 10 μm.

When it reaches the central portion of the FOV, the center of the bead comes into focus. Theextracted raw focal stack for a given bead is shown in Fig. 4(a). These raw focal stack ofimages were processed using deconvolution to computationally restore sectional images of thebeads. In order to deconvolve the acquired raw focal stacks, the point-spread funtion (PSF)of the system has to be known. In general, there are three approaches to determine the PSF:Experimental, Analytical and Computational [9]. Experimental determination would be quitechallenging owing to the difficulty in recording the PSF using point-like objects in the flow-based scenario at hand. Several analytical models for calculating the PSF are available [35].The available models have been developed for conventional slide based imaging and do nottake into account the curvature of the microfluidic channel, which is relevant to the currentscenario. Hence, we chose the computational approach to determine the PSF. The computationalapproach involves the use of blind deconvolution algorithms to estimate the PSF of the systemfrom the acquired images. Blind deconvolution algorithms iteratively estimate both the blur-freeimage and system PSF simultaneously. Most Blind deconvolution algorithms work by removingthe out-of-focus light, rather than reassigning the light into the proper planes as in the case ofrestoration algorithms like Wiener deconvolution [36]. Thus, the results are expected to be betterwhen, PSF estimated (using blind deconvolution or otherwise) is used for photon reassignmentwith restoration algorithms.

As mentioned earlier, the diameter of the section is expected to increase as we move from oneend of the bead to its center. However, due to the presence of out-of-focus blur in the images,the variation in diameter is not evident in the raw images of the focal stack.

To restore, the sectional images of the microsphere, we have first estimated the systemPSF using Lucy-Richardson maximum likelihood blind deconvolution algorithm [9]. Followingwhich, the estimated PSF was used to process the raw focal stack with Wiener deconvolution[36]. The deconvolved images have been shown in Fig. 4(b). The reduction/change in thediameter of the slice as we move from one plane to the next is quite evident in the case ofdeconvolved images. The variation in the diameter of the slices can be observed from theintensity profiles across a given slice. The deconvolved images display wave-like artifacts/circular ring, seen around the slice commonly referred to as ringing artifacts. Ringing artifactsoccur due to Gibbs phenomenon, wherein the Fourier sum overshoots at discontinuities. Theseeffects are more pronounced, when blind deconvolution algorithms are used to estimate the PSF.For the image stacks acquired using our system, the optimal number of iterations to estimatethe PSF was 5 [37]. The deconvolved images were post-processed to minimize artifacts. Thepost-processed images have been shown in Fig. 4(c). Comparison between the intensity profilesfor two different slices: Edge Slice which corresponds to the slice with the least diameter andthe center Slice, corresponding to highest diameter have been shown in Fig. 5.

The loss of details like sharp boundary of the bead and uniform intensity, in the raw focalstack (Fig. 4(a)) is due to the presence of out-of-focus light overlaying on the in-focus lightin the same image. In other words, when a particular plane of bead is in the DOF, the lightfrom the planes below and above the in-focus plane is also acquired by the imaging system.Deconvolution restores the sectional images by computationally reassigning the out-of-focus


(a)

(b)

Size (in μm) along the diameter of slice

Inte

nsity

Fig. 5. Intensity Profiles drawn across the surface of two different slices of the fluorescentbead. (a) Image of the edge slice, the intensity profile shown is over the line across image.(b) Image of the Center slice, the intensity profile shown is over the line across image.

photons/light back into their respective planes [8]. As mentioned earlier, the device facilitatesimplementation of powerful computational imaging techniques to significantly enhance the thesignal-to-noise ratio (SNR). The enhancement in SNR is thus dependent on the number ofslices/stacks acquired for a given specimen/cell. This is due to the fact that collecting moreslices would enable reassignment of more light, thereby increasing the SNR. Conversely, notconsidering a selected number of slices from the originally recorded focal stack would result ina reduction in intensity of the deconvolved image(s). This is due to the reduction in light thatwould have been reassigned to the proper plane. In order to analyze the photon reassignmentcapability and the SNR enhancement, the deconvolution was performed taking only a few ofthe recorded slices into account. The intensity of a given restored section/slice would vary asa function of number of slices used for deconvolution. Fig. 6 shows a given sectional image,obtained by using different numbers of recorded slices for deconvolution. As evident fromthe figure, the intensity across the line profile along a given slice of bead is highest whenthe complete stack is used. The intensity reduces for subsequent reduction in the number ofslices used for reconstruction. As noticed, the SNR enhancement is dependent on the number


(a)

(b)

(c)

(d)

Fig. 6. Comparison of photon reassignment, when different number of slices are used fordeconvolution. The intensity profiles across the restored images when (a) Full stack (44slices), (b) 22 slices, (c) 15 slices were used for reconstruction. The (d) raw image of thesame plane and it’s corresponding intensity profile have been shown. Length of scale bar is10 μm.

of slices acquired per cell. Further, it follows that specimen with weaker fluorescence emissionwould require higher enhancement in SNR. Whereas, larger specimen with higher fluorescenceemission would require a lesser SNR enhancement and thereby would require lesser number offocal stack images (per cell) to be acquired. The proposed fine slant channel geometry offersgreat flexibility in terms of acquiring the required number of focus stack images. The slicingpitch can be very easily varied by changing the flow speed of the cell (flow rate) and the framerate of acquisition. Subsequently, the acquired slices can be used for SNR enhancement withthe implementation of a suitable photon-reassignment algorithm.

In conventional optical sectioning approaches, the out-of-focus light is physically rejectedby placing a pinhole in front of the detector. Whereas, in the presented approach, light frommultiple planes is collected multiple times so as to enable computational reassignment andincreases the effective exposure time; thereby, making an intelligent use of the out-of-focuslight present in each of the raw images in the acquired stack. Conventional confocal microscopy


Size (in μm)Size (in μm)

Size

(in

μm)

Fig. 7. 3D visualization of the 4 μm fluorescent microsphere.

2.5 3 3.5 4 4.5 5 5.50

5

10

15

20

25

30

Size (in micron)

Num

ber

of B

eads

Fig. 8. Histogram for the size of fluorescent microspheres as measured using the presentedslanted channel microfluidic chip.

involves scanning in a point-by-point manner. This makes for an inherently slow technique. Incontrast, the proposed approach is a widefield technique, wherein cells are imaged while theyare in flow. Thus, the aspects of speed and automation of the presented technique potentiallyenables high-throughput 3D imaging of cells like in imaging flow cytometry.

Further, the added advantage of the proposed method is the ability to visualize samples/cellsin 3D. The restored (deconvolved) focal stacks can be used to reconstruct a 3D image of thefluorescing regions. The 3D reconstruction of the fluorescent microsphere from the acquireddata (Fig. 4(a)) is shown in Fig. 7. The contours along the sphere represent the edges/borders


of the circular regions present in each slice/plane of the bead (Fig. 4(c)). A surface is drawnthrough these contours to obtain the 3D structure of the object. As expected, the structure of thereconstructed object is similar to that of a 4 μm diameter sphere. In order to ascertain the fidelityof shapes in the reconstructed images, we have estimated the sizes of fluorescent microspheres.The histogram plot shown in Fig. 8 corresponds to sizes measured for about 64 beads. As perthe datasheet of the manufacturer, the nominal size of the bead is 4.2 ± 0.2 μm. The size ofthe microspheres as measured by our system is about 4.1 ± 0.2 μm. Thus, the presented systemenables fairly accurate measurement of morphological sizes of microscale specimen.

4. 3D fluorescence imaging flow cytometry

(a)(b)(c)

Fig. 9. (a) Raw focal stack of Bovine sperm cell nucleus. (b) Deconvolved focal stackof Bovine sperm cell nucleus. (c) Slices of bovine sperm cell nucleus, after removal ofartifacts. Length of scale bar is 10 μm.

Following the characterization of the proposed approach with microspheres / fluorescent-beads, we employed the microfluidic devices to perform three dimensional imaging flowcytometry. Using the devices, we have investigated the 3D morphology of bovine sperm cellnuclei. The nuclei of cells present in a suspension were stained using DAPI (4’,6-diamidino-2-phenylindole). As detailed in the previous section, the videos of the flow stream were acquiredusing a widefield digital fluorescence microscope. The flow rate was set to 30 μlhr−1 andthe frame rate was increased to 600 fps. The low flow rate together with the small channeldimensions ensures that the flow remains in the laminar regime. This prevents the cell fromrotating, which may otherwise impact the image acquisition. Also, the flow rate was optimizedto ensure that the motion-blur in the acquired images is minimal. The individual frames of thevideo were extracted and processed using Wiener deconvolution. The complete stack for a singlebovine sperm cell nucleus consisted of 226 slices; of which a few representative slices have beenshown in Fig. 9. The deconvolved stacks have been used to visualize the 3D morphology of thesperm cell nucleus (shown in Fig. 10). The increase in the number of slices acquired per cell isevident from the higher density of contour lines in the case of sperm cell reconstruction.

Due to their relevance in animal and human healthcare, sperm cell analytical techniqueshave gained significant research interest over the last few years. Morphology is one of themost critical parameters used to assess the health of the sperm cell [38, 39]. Recent researchefforts are being directed towards development of automated sperm morphological analysistechniques. The automation aspect plays a critical role in eliminating variation in testing due toenvironmental conditions and subjectiveness of the technician [40]. Several digital holographybased techniques for sperm cell 3D morphological assessment have been demonstrated [40–43].Although well proven, these techniques require implementation of optical interferometricsetups with a high degree of sophistication. In contrast, the presented approach enables3D morphological assessment, with a regular widefield fluorescence microscope. It does soby automating the process of focal stack collection in an elegant manner and subsequentcomputational reconstruction. The presented microfluidic devices can also be used for acquiringbrightfield focal stacks. These brightfield focal stacks can be processed using computationalmethods like transport of intensity equation, to generate a quantitative phase map [20].


Size

(in μm

)

Size (in μm)Size (in μm)

Fig. 10. 3D visualization of Bovine Sperm Cell nucleus.

Quantitative phase imaging enables retrieval of volumetric or thickness information withoutthe need for any labels thereby enabling live-cell imaging and assessment. Thus, the proposedautomated approach for focal stacking is equally applicable for label-free assessment of livecells in flow.

5. Conclusion

In this article, we have demonstrated an automated, elegant approach for focal stack collectionbased 3D fluorescence imaging using a generic widefield microscope. The presented approachinvolves the use of a small glass microfluidic device with slanted microfluidic channelsfabricated in it using ULI. The slanted nature of the channel forces the cell to moveperpendicularly to the focal plane of the objective as it traverses the FOV. This enables theacquisition of focal stacks, without the need for an expensive precision automated translationstage.

The proposed scheme enables acquisition of light from multiple planes, multiple times. Thisfacilitates the implementation of computational photon reassignment to make use of the out-of-focus light to enhance SNR, as well as retrieve sectional images of microscopic specimen.The retrieved sections/slices have been used for 3D visualization of cells and fluorescentmicrospheres. The approach has been shown to be capable of acquiring 200 slices per cell(Bovine sperm cell nucleus), amounting to imparting an equivalent mechanical translation


of 15 nm (per slice) along the optical axis. As demonstrated, the technique operates in awidefield, flow-based manner, which would enable higher 3D imaging throughputs as comparedto inherently slow, conventional methods. The presented architecture for the slanted channelmicrofluidic chip is a critical step towards realizing a functional prototype, which can enable 3Dfluorescence imaging of cells in flow using deconvolution microscopy. The generic nature of theapproach enables usage of this technique by biologists and non-optics experts as well, withoutthe need for setting up custom optical systems. We believe that the presented innovation woulddemocratize 3D imaging flow cytometry for usage in any biological lab, equipped with a genericwidefield fluorescence microscope. Also, the proposed design of slanted channel microfluidicdevices are aptly suitable for use with selective plane illumination microscopy (SPIM) systemsfor 3D fluorescence imaging. Usage of the proposed device with SPIM eliminates the need forcomputational reconstruction.

Further, the advent of portable optofluidic microscopy is enabling the implementation ofimaging based healthcare tools at the point-of-care [44, 45]. The integration of conventionalfocal stacking approaches in these portable systems is a significant challenge, owing to therequirement of additional mechanical components for imparting translation. Whereas with theuse of the present scheme of slanted glass microfluidic channels, focal stacking based 3Dimaging can be implemented in these point-of-care diagnostic devices as well. The presentedapproach has the potential to add another dimensionality in classification/analysis of cells inthe case of small-form factor devices, which have been otherwise limited to 2D morphology.We envision that such implementations would revolutionize 3D morphology based cellulardiagnostics in remote settings/point-of-care diagnostics.

Funding

UK India Education Research Initiative (UKIERI) (DST-2013-14/066); Engineering and Physical Sciences Research Council(EPSRC)(EM/M015130/1)(EM/M015130/1)(EP/J500227/1); Fast-track YoungScientist-Engineering Science Scheme of Science and Engineering Research Board(SERB)(SERB/F/2669/2014-2015).


Documents

Slanted channel microfluidic chip for 3D fluorescence ... · 9. P. Sarder and A. Nehorai, “Deconvolution methods for 3-D ﬂuorescence microscopy images,” IEEE Signal Processing