Lab on a Chip

CRITICAL REVIEW View Article Online

View Journal | View Issue

Review: imaging technologies for flow cytometry

This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.
Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

Cite this: Lab Chip, 2016, 16, 4639

Yuanyuan Han, Yi Gu, Alex Ce Zhang and Yu-Hwa Lo*

Received 23rd August 2016, High-throughput single cell imaging is a critical enabling and driving technology in molecular and cellular
Accepted 28th October 2016 biology, biotechnology, medicine and related areas. Imaging flow cytometry combines the single-cell im-
aging capabilities of microscopy with the high-throughput capabilities of conventional flow cytometry. Re-
DOI: 10.1039/c6lc01063f
cent advances in imaging flow cytometry are remarkably revolutionizing single-cell analysis. This article de-
www.rsc.org/loc scribes recent imaging flow cytometry technologies and their challenges.

1. Introduction convey certain messages about cells, such as cell size,

shape, morphology, and distribution or location of labeled
Flow cytometry is a widespread and powerful technique biomolecules within cells. As cellular morphology analysis
employed in cell counting, biomarker detection and cell plays an important role in various biological studies and
sorting. After the Coulter Principle was discovered in 1953 clinical diagnoses, such as cancer screening, conventional
and Fulwyler applied this principle to sort cells in 1965, op- flow cytometry is much anticipated to incorporate imaging
tical detections were soon adopted by flow cytometry and capabilities.16–18 Increasing flow cytometry's spatial resolu-
fluorescence-activated cell sorter (FACS) systems since the tion provides high-speed comprehensive analysis and in-
late 1960s.1,2 Soon afterwards, when false positive occurrence depth imagery of every individual cell. In addition, some er-
became a concern, imaging-in-flow techniques, including fly- roneous results yielded in conventional flow cytometry can
ing spot scanners, slit-imaging onto an array, laser strobes, be eliminated by acquiring and analyzing the cell images to,
and mirror tracking, were seen as invaluable in understand- for example, distinguish between cells, debris, and clusters
ing discrepancies in cell measurements due to cell orienta- of cells.
tion and dynamics in fluid flow.3–6 At the same time, be- In contrast to pure quantitative measurements provided
cause of flow cytometry's successful applications in by conventional flow cytometry, a technique invented in
immunology and thanks to the growing appreciation of re- the seventeenth century, microscopy allows capturing cell
searchers for the complexity of the immune system, further images that contain a wealth of information about a cell.
technological advances have focused on satisfying the in- While conventional flow cytometry measures forward
creasing need for polychromatic approaches to flow cytome- scattered light to estimate the relative cell size, microscopy
try; researchers have developed flow cytometry to simulta- yields the exact cell size through its bright field image.
neously measure 19 parameters—17 fluorescence and 2 Advances in the microscopy technique have realized both
scatter parameters in a high-speed manner.7 The broadly 3-dimensional and super resolutions that allow imaging of
useful technology—flow cytometry—has been evolving slowly the biological structure and function beyond the diffrac-
until imaging flow cytometry (IFC) became a resurgence of tion limit, producing extraordinarily detailed fluorescent
interest in the past decade.8,9 Due to its high-throughput cell images, but it can take as long as several minutes to
and multiparametric analysis, by supporting detection of sin- produce such high-content images. Automated microscopy
gle cell properties at rates from hundreds to 100 000 cells coupled with automated image analysis can be relatively
per second, conventional flow cytometry is an irreplaceable fast with a typical throughput of several hundreds of cells
cytologic instrumentation when a study of high-volume cell per second. The technology called laser scanning cytome-
populations and subpopulations needs to be performed.10–12 try (LSC) allows automated high-throughput image analysis
Meanwhile, due to the lack of spatial resolution in exchange to identify and measure cell properties with multiple spec-
for higher throughput, users have to make gating decisions tra and high spatial resolution for measuring dynamic
blind to some of the most informative and relevant sample processes.19–24 Although LSC and high-throughput micros-
attributes contained in cell images.13–15 Imaging is indubita- copy are suitable for studying complex biological pathways
bly indispensable for cell analysis because images effectively in a time-lapse manner, they are designed to work only
with adherent cells. There are also some bottlenecks for
Department of Electrical and Computer Engineering, University of California, San high-throughput microscopy systems, including mechanical
Diego, California, 92093, USA. E-mail: ylo@ucsd.edu stability when motorized stages and autofocus drives are

This journal is © The Royal Society of Chemistry 2016 Lab Chip, 2016, 16, 4639–4647 | 4639
 View Article Online

Critical review Lab on a Chip

required, image analysis problems caused by non-uniform 2.1. Camera-based IFC

illumination and cell-to-cell unmixing, sample and liquid CCDs and CMOSs have a dense array of individual sensors in
handling,25 and above all, low throughput and yield in a 2-dimensional (2D) arrangement. Therefore, in an IFC sys-
cell sorting. On the other hand, the development in mo- tem that uses CCD or CMOS as a detector, by employing
bile devices, especially in cell-phone cameras, has facili- wide-field illumination, 2D cell images can be produced as
tated several cost-effective and field-portable imaging tech- long as a sufficient number of photons are sensed within a
nologies, including lens-free optofluidic microscopy and given exposure time by those individual sensors of such cam-
cell-phone-based optofluidic fluorescent imaging cytome- era devices placed at the image plane. The conundrum in
try.26,27 These miniaturized and automated lab-on-a-chip
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

this case is to increase the speed of such imaging tools.

platforms provide commendable image quality and even When applied in IFC that is on a cell-to-cell basis, the field of
Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

three-dimensional cell representations in some exam- view is typically only 10s μm by 10s μm, so the large number
ples.28,29 Yet, these systems are most suitable for a rela- of pixels in a CCD or CMOS camera is wisely re-arranged to
tively small sample volume and cannot match the high work in parallel, either spatially or temporally.
throughput of conventional flow cytometers. 2.1.1. ImageStream. The IFC developed by Millipore, e.g.
A parallel microfluidic cytometer (PMC) has been ImageStream and FlowSight, relies on high-speed CCD cam-
implemented to compromise between high throughput and eras that use the time delay and integration (TDI) technique,
high content, but has very limited 1-dimensional spatial reso- which is originally designed to image objects moving along
lution to resolve many sub-cellular components and struc- one axis at low light levels.32,39,40 This type of CCD provides
tures compared to 2-dimensional cell images.30,31 Recent ad- higher sensitivity by having multiple rows of sensors which
vances in imaging technologies, electronics, and digital shift their partial measurements to the adjacent row synchro-
computing have enabled imaging flow cytometry (IFC).32,33 nously with the motion of the moving cell image across the
As an integration of fluorescence microscopy and conven- array of sensors. Applying this reading out technique, one
tional flow cytometry, IFC combines flow cytometry's single- can detect weak fluorescence signals without motion blur
cell identification and high throughput with microscopy's cell caused by an increase in exposure time. The sensor arrays on
image acquisition. Therefore, it becomes an ideal approach CCD are divided into N columns to detect emission or
to simultaneously fulfill both analysis of morphological char- scattered light of N different spectral ranges from cells. Using
acteristics and phenotypic characterization of single cells spectral decomposition elements, 12 images per cell can be
within an enormous and heterogeneous population.34 Also, acquired simultaneously. Fig. 1 illustrates how the optics of
as the interest in performing IFC systems grows, the necessity ImageStream/FlowSight works. Because of the rich subcellu-
of combining this technique with cell sorting becomes lar information acquired by ImageStream, various analysis
evident.35–38 and machine learning algorithms can be applied to study cell
The basic idea behind IFC is scaling up flow cytometric phenotype and subgroup classification.
spatial resolution to analyze more properties of cells. IFC, in Because the translation of the cell is exactly synchronized
this review, aims at the fluidic-based platforms that have op- with the vertical charge transfer of each pixel on the CCD,
tical imaging functionality at informative spatial resolution using the TDI reading out technique requires a closely con-
while retaining the main features of conventional flow cytom- trolled fluidic system to ensure cells are centered and flow at
etry. This article aims at discussing recent IFC technologies
and also those advances developed targeting at IFC. The fol-
lowing sections will describe recent IFC technologies, review
their strengths and challenges, and also discuss the outlook
for IFC.

2. Technologies and methods

Taking both high throughput, i.e. high temporal resolution,
and high spatial resolution into account, signal detection is
the key challenge due to the fundamental trade-off between
acquisition speed, sensitivity, and amount of information.
Detectors used in imaging platforms can be divided into two
types: 1) multipixelated imaging devices, such as charge-
coupled device (CCD) and complementary metal-oxide-
semiconductor (CMOS), and 2) single pixel photodetectors,
such as photomultiplier tube (PMT) and avalanche photodi-
ode (APD). The rest of this section will discuss IFC platforms Fig. 1 Optics of ImageStream. Reproduced from ref. 40 with
that employ these two types of detectors separately. permission from the EMD Millipore Corporation.

4640 | Lab Chip, 2016, 16, 4639–4647 This journal is © The Royal Society of Chemistry 2016
 View Article Online

Lab on a Chip Critical review

a constant speed without rotation. This strict requirement erration caused by the diffractive lens. While field aberration
hinders the system to adopt a sorting mechanism, since any is relatively less severe in MIFC systems since cells can be hy-
minor fluidic disturbance from downstream cell sorting can drodynamically focused and the field of view does not need
cause imaging instability. One limitation of the system speed to be significantly larger than single cells, typically 20 μm,
is the inherent data downloading method of CCD: every unit chromatic aberration brought about by the diffractive lens
sensor passively collects incoming photons and stores makes MIFC work only in monochromatic mode. Light of
electrons till the whole line/array has been read out; accumu- various wavelengths has various light paths in an optical sys-
lated charges are transferred from the unit sensor to its tem that contains a diffractive lens. Its incompatibility with
neighbor before they are dumped into the charge amplifier to multi-mode and multi-spectral imaging becomes a hurdle of
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

be converted into voltage. This working scenario constrains this IFC technique.
Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

the system's data access rate. On the other side, obtaining 2.1.3. Temporally coded excitation. The primary way to im-
enough sensitivity without any gain like electron multiplica- age weak cell fluorescence signals in a CMOS-based system is
tion also prevents the system to reach throughput higher to increase the exposure time, but consequential motion blur
than 3000 cells per second. can dramatically downgrade the image quality. Motion blur
2.1.2. Multiple field of view. Instead of faster sophisticated occurs when the exposure time is longer than the time it
photodetectors, a method named multi-field-of-view imaging takes the flowing cell to move a minimum resolvable dis-
flow cytometry (MIFC) was developed to image multiple tance. A technique that exploits temporally coded excitation
changes simultaneously to obtain high throughput. This effectively eliminates motion blur for fluorescence imaging of
method circumvents the trade-off between throughput and flowing cells.42 Instead of continuous illumination, a chopper
exposure time by projecting multiple fields of view onto the wheel is used to generate modulated excitation pulse se-
CMOS camera.41 The increase of throughput is, therefore, quences with a pseudo-random code (shown in Fig. 3). In
proportional to the degree of parallelization, i.e. the number this way, the captured images can be processed with a known
of parallel fluidic channels or the number of isolated fields code sequence and known point spread function and finally
of view, while the cell flow velocity is kept at a moderate level. be de-blurred using computational algorithms. Based on this
Due to microfabrication of multiple parallel microfluidic technique, a fluorescence image of cells moving faster than
channels and on-top N × M microlens arrays, the total field motion-blur velocity can be reconstructed. In addition, since
of view has to be very wide to cover N × M channels. Mean- the decoded image and moving velocity of the object are
while, a diffractive lens made of polydimethylsiloxane found computationally, the system can compensate for the
(PDMS) is used for its good monochromatic performance. effects of velocity variation in a laminar flow.
Fig. 2 shows the optical setup and the design of the lens
array.
Despite that having more parallel fields of view means 2.2. PMT-based IFC
higher throughput, it is not always extendable due to field ab- PMTs have superb sensitivity at photon counting levels and
high dynamic range because of their internal tandem
electron multiplication and gain adjustment. PMTs can also
provide higher bandwidth and lower dark noise than CCD/
CMOS cameras to support high-throughput IFC systems.
However, the readout of single-pixel PMTs presents the num-
ber of photons detected only in the time domain, which con-
tains no spatial information. In some high-speed microscopy

Fig. 2 Multiple field-of-view imaging flow cytometer. (a) Diffractive Fig. 3 Coded excitation fluorescence microscope. (a) Chopper wheel
lens wide-field imaging system. L1 collimates the LED. L2 is a con- that modulates the excitation beam with a pseudo-random code. (b) A
denser that images Iris 1 onto the object plane. DL is the diffractive microfluidic device imaged using a 40× fluorescence imaging micro-
lens providing multiple fields of view. L3 is a relay lens. (b) 16 object scope. (c) Raw blur encoded images captured by the camera and
planes. (c) 16 imaging planes. Sample: 3.5 μm latex beads. Reproduced decoded images after the computational approach. Reproduced from
from ref. 41 with permission from the Royal Society of Chemistry. ref. 42 with permission from the Optical Society.

This journal is © The Royal Society of Chemistry 2016 Lab Chip, 2016, 16, 4639–4647 | 4641
 View Article Online

Critical review Lab on a Chip

laser scanning cytometry techniques, PMTs are combined multiple wavelengths is also available for more efficient
with laser spot scanning to collect the entire light emitted or measurements.46
scattered by the illuminated cell and output a temporal sig- Besides producing bright-field images, STEAM can pro-
nal. Cell images are then retrieved by assembling the inten- duce phase-contrast images of cells by generating two 1D or-
sity signal in the time domain according to the laser scan- thogonally polarized spectral rainbows for illumination.47,48
ning position. Therefore, the overall throughput is limited by However, due to the high attenuation in the visible spectral
the speed of serial beam scanning. Instead of relying on me- range of one key optical component applied in the time-
chanical beam scanning, several techniques have been devel- stretching-based system—optical fiber—and the incoherence
oped to transfer spatial information to either the frequency of fluorescence, this technology has not been able to produce
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

domain or time domain in order to make use of PMT's extra fluorescence images, which is an essential function in nu-
Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

bandwidth that has not been fully utilized in conventional merous applications of cell analysis. Incorporating 1D fluo-
flow cytometers. rescence detection by employing additional lasers enables
2.2.1. STEAM. An ultrafast optical imaging modality the system to retain features of conventional flow cytometry,
named serial time-encoded amplified microscopy (STEAM) but such fluorescence signals do not benefit from the time-
has been built for blur-free imaging of cells flowing at high stretching technique.49,50 Another bottleneck of this tech-
speed.43–45 Different from the light sources, including light- nique, also due to the limited working spectral range up to
emitting diodes (LED), laser, and mercury lamps, used in the near-infrared regime (i.e. around 1000 nm), is that the
conventional flow cytometry or fluorescence microscopy, a spatial resolution does not go beyond the wavelength-related
mode-locked femtosecond pulse fiber laser is used to gener- diffraction limit.
ate illuminating near-infrared light with a wide spectral 2.2.2. FIRE. Adopting the schemes in radiofrequency com-
bandwidth of 10s nm centered at ∼1000 nm wavelength munications, an emerging technique named fluorescence im-
(Fig. 4). The ultrafast broadband laser pulses are spectrally aging using radiofrequency-tagged emission (FIRE) has made
encoded by using an optical spatial disperser so that a 1D or high-speed fluorescence imaging possible.51,52 The
2D spectral rainbow for illumination is generated. In this continuous-wave laser is converted into multiple intensity-
way, the one-to-one spatial-to-spectral relation is obtained. In modulated excitation beams by using an acousto-optic deflec-
other words, every individual point of the cell is illuminated tor (AOD) and an acousto-optic frequency shifter (AOFS) in
by light at a specific wavelength. After being recombined by the optical setup (Fig. 5). Every individual point of the cell
the spatial disperser and temporally stretched by a disper- within the imaging field of view is excited by light modulated
sive medium, typically a long optical fiber, the transmitted at a distinct radio frequency. Acousto-optic components dif-
rainbow is detected by a single-pixel photodetector. The fract light to different angles in 1D, and cells move at a cer-
output temporal waveform, therefore, represents the spatial tain speed along the direction orthogonal to the acousto-
information encoded by the cells being illuminated. IFC optically scanned beam in an IFC system. Thus, each pixel in
based on the time-stretching method can achieve through- a 2D cell image captured by FIRE is represented by a unique
put as high as 100 000 cells per second. In addition to one- combination of one radiofrequency and one time-point. The
to-one spatial–spectral mapping, encoding one location with resulting throughput is up to 50 000 cells per second. In addi-
tion, by increasing the bandwidth from that of the employed
AOD, the speed of this imaging technique is further exploited
to be limited by fluorescence lifetime.53
Because the time-domain signal from a FIRE system is a
Fourier superposition of the radiofrequency-tagged emission
from one row of pixels, a Fourier transform is required in
data processing. This can make realization of real-time pro-
cessing and instant result generation difficult for the system.
Besides, the operations of the key components in a FIRE sys-
tem—AOD and AOFS—are wavelength-dependent, which
makes multicolor imaging very challenging. In addition, be-
cause this radiofrequency tagging technique is only applied
to coherent light sources, the FIRE system is not suitable for
bright-field imaging where an LED source is preferred to
avoid speckle noise and interference.
2.2.3. Spatial–temporal transformation. Without sophisti-
Fig. 4 Schematic of the STEAM flow analyzer. Imaging and cated optical components for fluorescence excitation and de-
illumination optics takes blur-free images by encoding object location
tection or multipixelated photodetectors, a technique named
information into the spectral domain. ADFT converts spectral informa-
tion into the time domain through the time-stretching method and
spatial–temporal transformation is applied to retrofit a con-
processed using a real-time imaging processor. Reproduced from ref. ventional flow cytometer into an IFC system with minimal
45 with permission from PNAS. modification.54,55 The central, defining feature of this

4642 | Lab Chip, 2016, 16, 4639–4647 This journal is © The Royal Society of Chemistry 2016
 View Article Online

Lab on a Chip Critical review

This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.
Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

Fig. 5 FIRE microscopy. (a) Schematic diagram of the FIRE microscope. BS, beam splitter; AOD, acousto-optic deflector; AOFS, acousto-optic fre-
quency shifter; DM, dichroic mirror; EF, fluorescence emission filter; OL, objective lens; PMT, photomultiplier tube; DIG, 250 MS digital recording
oscilloscope; RS, resonant scanning mirror. Upper inset: the AOD produces a single diffracted first-order beam for each radiofrequency comb fre-
quency. Lower inset: beat frequency generation at the MZI output. (b) Gabor lattice diagram of FIRE's frequency-domain multiplexing approach.
Points in the same horizontal line are excited in parallel at distinct radiofrequencies. The horizontal line is scanned by a galvo-mirror to acquire a
2D image. Reproduced from ref. 51 with permission from the Nature Publishing Group.

technique is its ability to encode the time-domain signal image analysis, mainly for microscopy platforms, including
waveform with a specially designed spatial filter so that the CellProfiler and ImageJ, but in general, the pipelines of these
waveform consists of a sequence of patterns separated in the tools are for offline image analysis.62 Fortunately, some of
time domain. By inserting the spatial filter with a known pat-
tern in the image plane, fluorescence, transmission and
scattered light from different parts of the cell pass through
different slits at different times (Fig. 6). Cell images in multi-
modes, such as fluorescence and scattering, can be assembled
by using algorithms corresponding to the spatial filter used.
However, the throughput of the design is inversely propor-
tional to the length of the spatial filter, and the spatial filter
cannot be arbitrarily shortened in order to attain decent spa-
tial resolution. Without losing any features of conventional
flow cytometry, this technique enables the incorporation of
multi-mode and multi-spectral imaging capabilities. More-
over, the optical configuration of this IFC system allows the
use of disposable microfluidic devices;56–58 the involved com-
putation requires minimal time, and unlike the CCD-based
IFC, the data access time from the PMT does not limit the
system's overall imaging and processing throughput. There-
fore, this technique is well suited for high-throughput, real-
time image-based cell classification and sorting.
In addition, by adding more PMTs, dichroic mirrors and
more light sources, the system has been extended to work in
a multi-parameter manner—multicolor fluorescence, bright- Fig. 6 Implementation of spatial–temporal transformation-based IFC.
field and dark-field imaging. As shown in Fig. 7, two-color (a) Schematic diagram of the imaging flow cytometer system. DM, di-
fluorescence, transmission and backscattering cell images chroic mirror; SF, spatial filter; EF, emission filter; PMT, photomultiplier
tube. (b) Spatial filter design that has ten 100 μm by 1 mm slits posi-
are demonstrated.
tioned apart in a way that one is immediately after another in both the
x-direction and y-direction. (c) Experimental result: time-domain PMT
3. Challenges: high-throughput and output signal of fluorescent light from an A549 cell stained with
real-time image analysis CellTrace CFSE, corresponding original fluorescence image restored
using an algorithm, and corresponding resized fluorescence image to
show the real size of the cell. The numbered regions segmented by
Apart from the aforementioned challenges of each technique
dashed lines demonstrate the correspondence between the time-
(Table 1), the biggest challenge of IFC lies in acquiring, storing domain signal and the resulting image. Size is labelled in the figure.
and processing a massive number of cell images.59–61 There are Reproduced from ref. 54 with permission from the Nature Publishing
many software packages and tools for use in high-throughput Group.

This journal is © The Royal Society of Chemistry 2016 Lab Chip, 2016, 16, 4639–4647 | 4643
 View Article Online

Critical review Lab on a Chip

This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.
Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

Fig. 7 Demonstration of bright-field, two-color fluorescence and backscattering cell images. Scale bars are 5 μm. (a) Bright-field images of MDA-
MB-231 human breast cancer cells flowing at 0.2 m s−1. (b) Representative two-color fluorescence images of MDA-MB-231 human breast cancer
cells stained with CellTrace CFSE, with 1 μm fluorescent beads attached to the cell membrane, flowing in the microfluidic channel at 0.25 m s−1.47
(c) Representative fluorescence plus backscattering cell images from spatial filter-based imaging flow cytometry. All images are of A549 human
lung adenocarcinoma epithelial cells, stained with CellTrace CFSE, flowing at a velocity of 0.2 m s−1. Reproduced from ref. 54 with permission from
the Nature Publishing Group.

Table 1 Summary of imaging flow cytometry (IFC) techniques

Temporally coded Spatial–temporal

Features ImageStream Multiple-view excitation STEAM FIRE transformation
Illumination (spectral range) Laser LED Laser Pulse laser Laser Laser
LED LED (near-infrared) LED
Detector TDI CCD CMOS CMOS Single-pixel detector PMT PMT
Modality Fluorescence Transmission Fluorescence Transmission Fluorescence Fluorescence
Transmission Transmission Phase contrast Transmission
Side-scatter Back-scatter
Data transfer & computation Simple Simple Simple Complex Complex Simple
complexity
Inherit functions from FC Yes No Yes No Yes Yes
Compatibility with sorting No No No No No Yes
Image resolution Moderate Moderate High Moderate High Moderate
Throughput (cells per s) Up to 1000 ∼10 000 ∼2000 Up to 100 000 ∼50 000 >1000
Real-time processing solution Yes No No Yes No Yes
Reference 32 41 42 45 51 54

FC: conventional flow cytometry.

the difficult image processing problems in high-throughput ital image transportation and processing realized by the
microscopy and LSC can be avoided in IFC. For example, due back-end data handling unit. Assuming that a field-of-view of
to cell-to-cell and cell-to-substrate adhesion, strategies for im- 40 μm by 40 μm is represented by an image of 100 pixels by
age segmentation such as threshold, watershed and edge- 100 pixels, at least 100 MB data are generated in one second
detection have been under development for decades, but still at a throughput of 10 000 cells per second. Therefore, a test
a bottleneck of the automated image analysis in microscopy. of a few minutes can easily create a data file beyond 10s of
This problem has little effect on IFC where the cells are GB. For a possible solution, a compressive sensing theory-
placed in suspension and interrogated on a single-cell based method has recently been explored to build analog
basis.63–67 For blood cells, bone marrow cells, and many can- compression directly into the acquisition process so that the
cer cells flowing in the blood vessels, IFC is the most promis- sampling can be significantly more efficient.68,69 Some ma-
ing approach to study their morphological changes. chine learning techniques can also be applied to data pro-
Compared to the data format in conventional flow cytome- cessing for richer information carried out by IFC systems.66,70
try, including integral, peak and width of light intensity, cell Computational requirements for IFC platforms are un-
images produced by IFC are much more complex. Since IFC precedented. All IFC systems demonstrated to date perform
can produce thousands of multi-spectral cell images per sec- image analysis offline. In order to combine cell sorting with
ond, files generated by IFC can tremendously burden the dig- IFC to fully realize its tremendous potential, real-time image

4644 | Lab Chip, 2016, 16, 4639–4647 This journal is © The Royal Society of Chemistry 2016
 View Article Online

Lab on a Chip Critical review

construction and analysis is required. Hence, the ability to 5 J. L. Cambier, D. B. Kay and L. L. Wheeless, A
produce, measure, and analyze cell images, and sort cells in multidimensional slit-scan flow system, J. Histochem.
a real-time manner will be the next major milestone for IFC. Cytochem., 1979, 27, 321–324.
Possible approaches to extract cell characteristics in real time 6 L. A. Kamentsky, Future Directions for Flow Cytometry,
include the use of field-programmable gate arrays (FPGA) or J. Histochem. Cytochem., 1979, 27, 1649–1651.
graphics processing units (GPU) to implement various image 7 S. P. Perfetto, P. K. Chattopadhyay and M. Roederer,
processing and machine learning algorithms. Seventeen-colour flow cytometry: unravelling the immune
system, Nat. Rev. Immunol., 2004, 4, 648–655.
4. Conclusion 8 L. A. Herzenberg, et al., The history and future of the
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

Fluorescence Activated Cell Sorter and flow cytometry: A

Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

With the advent of big-data era, life scientists started to grap- view from Stanford, Clin. Chem., 2002, 48, 1819–1827.
ple with massive volumes of data. Recent advances in high- 9 N. S. Barteneva, E. Fasler-Kan and I. A. Vorobjev, Imaging
throughput IFC allow continuous high-throughput capture of Flow Cytometry: Coping with Heterogeneity in Biological
cell images. Various technologies have made the generation Systems, J. Histochem. Cytochem., 2012, 60, 723–733.
of multiparametric imaging files highly feasible although effi- 10 S. H. Cho, et al., Review Article: Recent advancements in
cient analysis and utilization of this huge amount of data re- optofluidic flow cytometer, Biomicrofluidics, 2010, 4, 43001.
main a challenge. Multimodality (i.e. transmission, scattering 11 L. Jiang, et al., Monitoring the progression of cell death and
and fluorescence), functional flexibility (i.e. operation as con- the disassembly of dying cells by flow cytometry, Nat.
ventional flow cytometry or IFC at users' choice), and compat- Protoc., 2016, 11, 655–663.
ibility with cell sorting are the three principal areas of devel- 12 O. D. Laerum and T. Farsund, Clinical application of flow
opment for IFC to gain wide acceptance as a workhorse for cytometry: a review, Cytometry, 1981, 2, 1–13.
biomedical research and clinical applications. 13 E. Lugli, M. Roederer and A. Cossarizza, Data analysis in
flow cytometry: the future just started, Cytometry, Part A,
Acknowledgements 2010, 77, 705–713.
14 N. Aghaeepour, et al., Critical assessment of automated flow
This work was performed in part at the San Diego Nanotech- cytometry data analysis techniques, Nat. Methods, 2013, 10,
nology Infrastructure (SDNI) of UCSD, a member of the Na- 228–238.
tional Nanotechnology Coordinated Infrastructure, which is 15 C. E. Pedreira, E. S. Costa, Q. Lecrevisse, J. J. M. van Dongen
supported by the National Science Foundation (Grant ECCS- and A. Orfao, Overview of clinical flow cytometry data
1542148). Research reported in this publication was analysis: Recent advances and future challenges, Trends
supported by the National Institute of General Medical Sci- Biotechnol., 2013, 31, 415–425.
ences (NIGMS) of the National Institutes of Health under 16 J. V. Forment and S. P. Jackson, A flow cytometry-based
award number R21GM107977 and the National Institute of method to simplify the analysis and quantification of pro-
Biomedical Imaging and Bioengineering (NIBIB) of the Na- tein association to chromatin in mammalian cells, Nat.
tional Institutes of Health under award number Protoc., 2015, 10, 1297–1307.
R43EB021129. The content is solely the responsibility of the 17 L. Samsel, et al., Imaging flow cytometry for morphologic
authors and does not necessarily represent the official views and phenotypic characterization of rare circulating
of the National Institutes of Health. Yuhwa Lo has an equity endothelial cells, Cytometry, Part B, 2013, 84, 379–389.
interest in NanoCellect, Inc., a company that may potentially 18 S. E. Headland, H. R. Jones, A. S. V. D'Sa, M. Perretti and
benefit from the research results, and also serves as a mem- L. V. Norling, Cutting-edge analysis of extracellular
ber of the company's Scientific Advisory Board. microparticles using ImageStreamIJX) imaging flow
cytometry, Sci. Rep., 2014, 4, 5237.
References 19 Z. Darzynkiewicz, E. Bedner, X. Li, W. Gorczyca and M. R.
Melamed, Laser-scanning cytometry: a new instrumentation
1 M. J. Fulwyler, Electronic separation of biological cells by with many applications, Exp. Cell Res., 1999, 249, 1–12.
volume, Science, 1965, 150, 910–911. 20 P. Pozarowski, E. Holden and Z. Darzynkiewicz, Laser
2 H. R. Hulett, W. A. Bonner, J. Barrett and L. A. Herzenberg, Scanning Cytometry: Principles and Applications, Methods
Cell Sorting: Automated Separation of Mammalian Cells as a Mol. Biol., 2006, 319, 165–192.
Function of Intracellular Fluorescence, Science, 1969, 166, 21 M. M. Harnett, Laser scanning cytometry: understanding the
747–749. immune system in situ, Nat. Rev. Immunol., 2007, 7, 897–904.
3 D. B. Kay, J. L. Cambier and L. L. Wheeless, Imaging in flow, 22 M. Henriksen, Quantitative imaging cytometry: instrumentation
J. Histochem. Cytochem., 1979, 27, 329–334. of choice for automated cellular and tissue analysis, Nat.
4 V. Kachel, G. Benker, K. Lichtnau, G. Valet and E. Glossner, Methods, 2010, 7, i–ii.
Fast imaging in flow: a means of combining flow-cytometry 23 P. Pozarowski, E. Holden and Z. Darzynkiewicz, Laser
and image analysis, J. Histochem. Cytochem., 1979, 27, Scanning Cytometry: Principles and Applications–An Update,
335–341. Methods Mol. Biol., 2013, 931, 187–212.

This journal is © The Royal Society of Chemistry 2016 Lab Chip, 2016, 16, 4639–4647 | 4645
 View Article Online

Critical review Lab on a Chip

24 J. A. Hutcheson, et al., A light sheet confocal microscope for 43 K. Goda, K. K. Tsia and B. Jalali, Serial time-encoded ampli-
image cytometry with a variable linear slit detector, Proc. fied imaging for real-time observation of fast dynamic phe-
SPIE, 2016, 9720, 97200U. nomena, Nature, 2009, 458, 1145–1149.
25 M. Oheim, Advances and challenges in high-throughput 44 K. Goda, A. Motafakker-Fard, K. K. Tsia and B. Jalali, Serial
microscopy for live-cell subcellular imaging, Expert Opin. Time Encoded Amplified Microscopy (STEAM) for high-
Drug Discovery, 2011, 6, 1299–1315. throughput detection of rare cells, in IEEE Photonics Society
26 H. Zhu, S. Mavandadi, A. F. Coskun, O. Yaglidere and A. Winter Topicals Meeting Series 2, TuB2.2, 2010.
Ozcan, Optofluidic fluorescent imaging cytometry on a cell 45 K. Goda, et al., High-throughput single-microparticle imaging
phone, Anal. Chem., 2011, 83, 6641–6647. flow analyzer, Proc. Natl. Acad. Sci. U. S. A., 2012, 109,
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

27 X. Cui, et al., Lensless high-resolution on-chip optofluidic 11630–11635.

Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

microscopes for Caenorhabditis elegans and cell imaging, 46 E. Huang, Q. Ma and Z. Liu, Ultrafast Imaging using
Proc. Natl. Acad. Sci. U. S. A., 2008, 105, 10670–10675. Spectral Resonance Modulation, Sci. Rep., 2016, 6, 25240.
28 Q. Wei, et al., On-chip cytometry using plasmonic nanoparti- 47 A. Mahjoubfar, C. Chen, K. R. Niazi, S. Rabizadeh and B.
cle enhanced lensfree holography, Sci. Rep., 2013, 3, 1699. Jalali, Label-free high-throughput cell screening in flow, Bio-
29 W. Bishara, S. O. Isikman and A. Ozcan, Lensfree optofluidic med. Opt. Express, 2013, 4, 1618–1625.
microscopy and tomography, Ann. Biomed. Eng., 2012, 40, 48 A. K. S. Lau, H. C. Shum, K. K. Y. Wong and K. K. Tsia,
251–262. Optofluidic time-stretch imaging – an emerging tool for
30 M. C. Cheung, B. McKenna, S. S. Wang, D. Wolf and D. J. high-throughput imaging flow cytometry, Lab Chip, 2016, 16,
Ehrlich, Image-based cell-resolved screening assays in flow, 1743–1756.
Cytometry, Part A, 2015, 87, 541–548. 49 M. Ugawa, et al., High-throughput optofluidic particle profil-
31 B. K. McKenna, J. G. Evans, M. C. Cheung and D. J. Ehrlich, ing with morphological and chemical specificity, Opt. Lett.,
A parallel microfluidic flow cytometer for high-content 2015, 40, 4803–4806.
screening, Nat. Methods, 2011, 8, 401–403. 50 C. Lei, et al., High-throughput time-stretch microscopy with
32 D. A. Basiji, W. E. Ortyn, L. Liang, V. Venkatachalam and P. morphological and chemical specificity, Proc. SPIE,
Morrissey, Cellular Image Analysis and Imaging by Flow 2016, 9720, 97200X.
Cytometry, Clin. Lab. Med., 2007, 27, 653–670. 51 E. D. Diebold, B. W. Buckley, D. R. Gossett and B. Jalali,
33 K. E. Mcgrath, T. P. Bushnell and J. Palis, Multispectral Digitally synthesized beat frequency multiplexing for sub-
Imaging of Hematopoietic Cells: Where Flow Meets millisecond fluorescence microscopy, Nat. Photonics,
Morphology, J. Immunol. Methods, 2008, 336, 91–97. 2013, 7, 806–810.
34 I. A. Vorobjev and N. S. Barteneva, Imaging Flow Cytometry 52 H. Mikami, H. Kobayashi, S. Hamad, Y. Ozeki and K. Goda,
Methods and protocols, Springer Science+Business Media, 2016. Ultrafast Confocal Fluorescence Microscopy by Frequency-
35 S. H. Cho, C. H. Chen, F. S. Tsai, J. M. Godin and Y.-H. Lo, Division-Multiplexed Multi-Line Focusing, in CLEO AM4O.5,
Human mammalian cell sorting using a highly integrated 2016.
micro-fabricated fluorescence-activated cell sorter (micro- 53 H. Mikami, et al., Enhanced speed in fluorescence imaging
FACS), Lab Chip, 2010, 10, 1567–1573. using beat frequency multiplexing, Proc. SPIE, 2016, 9720,
36 L. Schmid, D. A. Weitz and T. Franke, Sorting drops and 97200T.
cells with acoustics: acoustic microfluidic fluorescence- 54 Y. Han and Y.-H. Lo, Imaging Cells in Flow Cytometer Using
activated cell sorter, Lab Chip, 2014, 14, 3710–3718. Spatial–Temporal Transformation, Sci. Rep., 2015, 5, 13267.
37 L. Ren, et al., A high-throughput acoustic cell sorter, Lab 55 Y. Han and Y.-H. Lo, Imaging flow cytometer using
Chip, 2015, 15, 3870–3879. computation and spatially coded filter, Proc. SPIE,
38 C. W. Shields, C. D. Reyes and G. P. López, Microfluidic cell 2016, 9720, 972010.
sorting: a review of the advances in the separation of cells 56 J. Godin, et al., Microfluidics and photonics for bio-System-
from debulking to rare cell isolation, Lab Chip, 2015, 15, on-a-Chip: A review of advancements in technology towards
1230–1249. a microfluidic flow cytometry chip, J. Biophotonics, 2008, 1,
39 G. S. Elliott, Moving pictures: imaging flow cytometry for 355–376.
drug development, Comb. Chem. High Throughput Screening, 57 M. E. Piyasena and S. W. Graves, The intersection of flow
2009, 12, 849–859. cytometry with microfluidics and microfabrication, Lab Chip,
40 EMD Millipore, INSPIRE®, 18, 2012, at <http://www. 2014, 14, 1044–1059.
pedsresearch.org/_files/INSPIRE_User_Manual_MkII_082912. 58 V. K. Jagannadh, M. D. Mackenzie, P. Pal, A. K. Kar and S. S.
pdf>. Accessed August 2016. Gorthi, Imaging flow cytometry with femtosecond laser-
41 E. Schonbrun, S. S. Gorthi and D. Schaak, Microfabricated micromachined glass microfluidic channels, IEEE J. Sel. Top.
multiple field of view imaging flow cytometry, Lab Chip, Quantum Electron., 2015, 21, 6800106.
2012, 12, 268–273. 59 V. Marx, Biology: The big challenges of big data, Nature,
42 S. S. Gorthi, D. Schaak and E. Schonbrun, Fluorescence 2013, 498, 255–260.
imaging of flowing cells using a temporally coded excitation, 60 M. May, LIFE SCIENCE TECHNOLOGIES: Big biological
Opt. Express, 2013, 21, 5164–5170. impacts from big data, Science, 2014, 344, 1298–1300.

4646 | Lab Chip, 2016, 16, 4639–4647 This journal is © The Royal Society of Chemistry 2016
 View Article Online

Lab on a Chip Critical review

61 H. T. K. Tse, et al., Strategies for Implementing Hardware- 66 T. Blasi, et al., Label-free cell cycle analysis for high-
Assisted High-Throughput Cellular Image Analysis, J. Lab. throughput imaging flow cytometry, Nat. Commun., 2016, 7,
Autom., 2011, 16, 422–430. 10256.
62 R. Wollman and N. Stuurman, High throughput microscopy: 67 C. L. Chen, et al., Deep Learning in Label-free Cell Classifica-
from raw images to discoveries, J. Cell Sci., 2007, 120, tion, Sci. Rep., 2016, 6, 21471.
3715–3722. 68 B. T. Bosworth and M. A. Foster, High-speed ultrawideband
63 D. R. Gossett, et al., Hydrodynamic stretching of single cells photonically enabled compressed sensing of sparse radio
for large population mechanical phenotyping, Proc. Natl. frequency signals, Opt. Lett., 2013, 38, 4892–4895.
Acad. Sci. U. S. A., 2012, 109, 7630–7635. 69 B. T. Bosworth, et al., Continuous high-rate photonically-en-
This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

64 Y. Zheng, J. Nguyen, Y. Wei and Y. Sun, Recent advances in abled compressed sensing (CHiRP-CS) for high speed flow
Open Access Article. Published on 31 October 2016. Downloaded on 5/13/2019 12:12:20 PM.

microfluidic techniques for single-cell biophysical characteri- microscopy, Imaging Appl. Opt., OSA Technical Digest (on-
zation, Lab Chip, 2013, 13, 2464–2483. line), Optical Society of (2015).
65 K. Galler, K. Bräutigam, C. Grosse, J. Popp and U. 70 A. C. Zhang, et al., Computational cell analysis for label-free
Neugebauer, Making a big thing of a small cell–recent ad- detection of cell properties in a microfluidic laminar flow,
vances in single cell analysis, Analyst, 2014, 139, 1237–1273. Analyst, 2016, 141, 4142–4150.

This journal is © The Royal Society of Chemistry 2016 Lab Chip, 2016, 16, 4639–4647 | 4647