Research Article Journal of the Optical Society of America B 1
Supercontinuum radiation in fluorescence microscopy
and biomedical imaging applications
CHETAN POUDEL
1
AND CLEMENS F. KAMINSKI
1,*
1
Department of Chemical Engineering and Biotechnology, University of Cambridge, United Kingdom.
*
Corresponding author: cfk23@cam.ac.uk
Compiled December 24, 2018
Compact, high brightness supercontinuum sources have already made a big impact in the fields of fluores-
cence microscopy and other biomedical imaging techniques, such as OCT and CARS. In this review, we
provide a brief overview on the generation and properties of supercontinuum radiation for imaging appli-
cations. We review specific uses of supercontinuum sources and their potential for imaging, but also their
limitations and caveats. We conclude with a review of recent advances in UV supercontinuum generation,
near-IR microscopy, exciting new potentials for the use of hollow core PCFs, on-chip supercontinuum
generation and technologies to improve supercontinuum stability for certain applications.
© 2018 Optical Society of America
http://dx.doi.org/10.1364/ao.XX.XXXXXX
1. INTRODUCTION
Fluorescence microscopy and biomedical imaging technologies
have made enormous progress over the past two decades and
an entirely new set of tools are available in the life sciences
that permit targeted structural and functional information to be
gained from biological systems. Exciting developments include
super-resolution methods [
1
–
4
] that go beyond the diffraction
limit in optical imaging as well as methods that allow one to
measure spectroscopic parameters, such as wavelength spec-
trum and fluorescence lifetime [
5
]. These techniques provide
information not only on the location of sub-cellular structures
but also on their function and environment. The field has been
massively boosted by huge technological developments in de-
tectors and light sources, but also molecular labeling methods to
permit specific staining of many different targets. The increasing
sophistication of bioimaging experiments has placed an ever
increasing demand on the enabling photonics technologies. One
of the most groundbreaking developments over the past two
decades in this context is the compact fiber-based supercontin-
uum (SC) source.
Exhaustive reviews on the topic of generating SC and its
subsequent properties already exist in the literature, for which
we refer the reader to [
6
–
11
]. The freedom in design afforded
by photonic crystal fibers (in tuning hole diameter, periodicity,
core size and material) has allowed scientists to study and use
the interplay between dispersion, nonlinearity, optical losses
and polarization effects [
9
] to generate tailored SC properties for
various applications. Additionally, a number of waveguiding
mechanisms can be exploited – using total internal reflection in
most solid core fibers, the photonic bandgap effect in hollow core
fibers [
12
], and anti-resonance in some recently reported hollow
fiber designs [
13
,
14
]. SC sources are often (erroneously) referred
to as ’white light lasers’. Temporally coherent and laser-like
SC can be generated through short pulse pumping, but many
of the most useful SC sources for biological applications come
from high power, incoherent supercontinua, that are driven by
longer pump pulses. Both types of SC radiation can provide
high spatial coherence as they are usually generated by the fun-
damental mode of optical fibers. Average output powers of
multiple Watts are routinely achieved in commercial devices,
making them suitable for many applications. SC sources have
been widely employed in applications ranging from studies of
fundamental processes [
11
] in physics, biology and chemistry
through absorption and excitation spectroscopy [
15
–
18
]; gener-
ating ultrashort femtosecond pulses to create optical clocks and
frequency combs [
19
] for precision frequency metrology (leading
to a Nobel prize in 2005); optical communication; atmospheric
science, and light detection and ranging (LIDAR) [
20
]; biomed-
ical imaging using optical coherence tomography (OCT) [
21
]
and coherent anti-Stokes Raman scattering (CARS) [
22
]; and live
cell imaging using various microscopy techniques [
23
]. In this
short review, we present aspects of SC sources most relevant for
the demands in microscopy (mostly fluorescence based modal-
ities) and for biomedical imaging applications. We begin with
the properties of SC that are critical in the context of imaging.
We then provide an overview of methodologies used to select
desired wavelengths, and go on to review the wide variety of
SC-enabled microscopies reported so far, which include wide-
Research Article Journal of the Optical Society of America B 2
field and laser scanning techniques, spectral imaging, lifetime
imaging, two-photon absorption, super-resolution and OCT and
CARS microscopies. We conclude with an outlook of current
challenges in the field, and recent developments that may enable
new applications in the future, including the use of gas-filled
hollow-core fibers to generate tunable UV radiation with high
pulse energies and peak powers.
2. PROPERTIES OF SUPERCONTINUUM TAILORED
FOR IMAGING
Ranka’s[
25
] use of solid core PCF and a mode-locked Ti:sapphire
laser, and Birks’s report [
26
] on tapered optical fibers were the
first demonstrations to herald the promise of efficient, cost-
effective and routine generation of octave-spanning SC in the
laboratory. The creation of endlessly single-mode PCFs support-
ing all generated wavelengths within one fundamental guided
mode of the fiber [
27
] complemented these developments. Today,
after over forty years of theoretical and experimental research
into SC generation, the spectral broadening mechanisms are
understood and have been identified to be soliton dynamics,
self-phase modulation (SPM), four-wave mixing (FWM), mod-
ulation instability (MI), Raman self-frequency shift (RSFS) and
dispersive wave generation (DW) [
11
,
28
–
30
]. SC can today be
generated routinely using a range of fiber types, input pump
sources, pulse energies, and input pulse durations ranging from
femtoseconds (fs) to continuous wave sources [
7
]. SC generated
by pumping PCFs can be broadly categorized according to the
injected pulse durations, as A: ultrafast fs pulses and B: long
pulses (ps-ns pulses or continuous wave), summarized in Table
1. This can be further subdivided based on whether the SC gen-
erating medium is pumped in the normal dispersion (NDi) or
anomalous dispersion (ADi) regimes. The dispersion regimes
dictate propagation dynamics and SC generation mechanisms
inside the nonlinear fibers. Therefore, choosing the right regime
for desired output characteristics is critical, affecting not just the
obtainable bandwidth and pulse durations but also properties
such as the SC dynamics like coherence and shot-to-shot stability.
In what follows, we discuss how these properties (spectral band-
width, brightness, pulse energy, pulse duration, average power,
coherence, stability) can be controlled by selecting appropriate
input pulses and dispersion regimes for specific applications,
such as fluorescence microscopy.
SC generation over the last few decades was driven primarily
by a push to increase spectral bandwidth and create octave-
spanning SC encompassing the full visible range and NIR. An
efficient way of generating such broadband SC is by using fs
pulses from Ti:Sapphire (800nm) lasers to pump tailored PCFs.
These PCFs are fabricated specifically to place the input source
wavelengths in the ADi regime, as in Ranka’s experiment [
25
].
Owing to its enormous potential in optical imaging, this method
was picked up already in 2004 by various biophotonics labo-
ratories, reporting either the use of custom-built PCFs [
31
–
33
]
or tapered silica fibers [
34
] in the ADi regime with hundreds
of mW input power from
∼
80MHz Ti:Sapphire lasers. SC cov-
ered a broad wavelength spectrum from
∼
400nm to near IR
wavelengths of
∼
1000-1500nm because of efficient broadening
via soliton dynamics. The spectral power densities were fairly
low (less than 0.5mW/nm) but usable for general fluorescence
microscopy applications.
However, pumping in the ADi regime also leads to high in-
stability arising from shot noise in the input laser pulses, which
gets amplified stochastically by modulation instabilities [
7
]. This
leads to temporally incoherent SC light with uneven spectra and
large differences in spectral density profiles from shot-to-shot
(see Figure 1). SC instability does not necessarily pose a prob-
lem in general fluorescence microscopy applications since the
timescale of experiments is much longer than the instantaneous
fluctuations, causing them to be averaged over. Temporally
incoherent light is, in fact, often favored for imaging because
coherent waves can introduce unwanted speckle patterns in the
image. On the other hand, applications such as CARS need a
high degree of coherence, for which fs pumping of all-NDi fibers
is more suitable since modulation instability and soliton-related
effects do not occur in this case. This generates stable, flat and
coherent SC with much better signal-to-noise [
24
,
35
], albeit with
smaller spectral widths at comparable peak power [36].
A big limitation in using fs pumps is the maximum obtainable
SC power. The peak intensity damage threshold in microstruc-
tured fibers usually limits the maximal spectral power density
to
∼
0.5mW/nm. For applications requiring higher power densi-
ties, using ps-ns input pulses increases it to several mW/nm in
the visible range [
37
]. Of course, when longer pulses are used
as input pump sources, the instability problem is greater (both
in the NDi and ADi regimes): larger fluctuations are seen, and
the coherence of the SC is compromised. Despite the stability
issue, the massive development of fiber laser sources over the
last few decades has provided cost-effective, reliable means of
providing ps-ns input pulses for SC generation with low main-
tenance and smaller fingerprint than fs sources. Therefore, for
microscopy and bioimaging applications, generation of SC has
mostly switched over to using these ps-ns fiber laser pump
sources. The technology has also been commercialized by a few
companies making SC generation widely accessible through a
variety of compact turnkey sources, usable by general biology
laboratories [
22
]. Further efforts over the last decade have re-
sulted in SC sources with higher power, faster pulses and greater
range of wavelengths. As an example of commercial technology,
one of the widely successful incoherent SC source is based on
coupling ps pulses from a mode-locked (typically 40MHz) yt-
terbium fiber laser into an engineered PCF with a ZDW around
1050nm, yielding high average SC output powers at up to
∼
20W
with 6ps pulse widths, few hundred nJ maximum pulse en-
ergy and single-mode operation in the 400-2400nm range [
38
].
Groups working on higher average output from ps sources have
reached powers up to 39W, with 31.7mW/nm spectral power
density and good uniformity across the full visible spectral range
[39].
SC has also been generated using continuous wave sources
like fiber lasers using simple, cost-effective setups yielding
broad spectral profiles. Massive output power densities (10s of
mW/nm) can be generated from 5-50W pump sources and used
in applications where high average brightness is more important
than short-pulse characteristics, e.g. peak power. However, SC
generated from CW input undergoes significant intensity fluctu-
ations and has negligible temporal coherence due to modulation
instability [
7
]. The spectral coverage usually lies in the NIR with
bandwidth not large enough to extend into the visible range.
This is one of the biggest limitations of CW input for SC gen-
eration as obtaining output in the visible spectrum is of prime
importance in applications like fluorescence microscopy. Some
attempts have been made to increase bandwidth and simultane-
ously push it towards the visible range going down to 600nm
[
40
] but this required industrial class fiber lasers requiring enor-
mous pump powers of 400W, which is impractical for routine
use. Another concept for extending to shorter wavelengths in
Research Article Journal of the Optical Society of America B 3
Fig. 1.
Comparison of shot-to-shot differences in SC spectra obtained with femtosecond pumping in the normal dispersion (NDi
- left panels) and anomalous dispersion regimes (ADi - right panels). The red/white lines (top left) represent a long term average
(10,000 shots) of the all-NDi spectrum, where each individual shot trace (blue) mimics the shape of the average long-term spectrum.
This stability in the NDi regime can also be observed in the 200 consecutive single-shot spectra (bottom left). The green/white
lines (top right) are low pass filtered versions of individual shot-to-shot spectra, all of which show significant deviations from their
long term average (red/white line in the bottom) in the ADi regime SC. Consecutive single shot spectra also show large differences
(bottom right) in the ADi regime. Clearly, the dynamics of SC in the NDi regime are in stark contrast with dynamics in the ADi
regime. Adapted from [24].
Pulse duration Femtosecond pulses Pico-nanosecond pulses and continuous waves
Spectral width Broad spectrum (visible-NIR)
Broad spectrum when using pulsed input, some covering
400-2400nm; Mostly only IR spectrum for CW sources
Coherence and
stability
Coherent and stable (NDi regime);
Incoherent, unstable (ADi regime)
Usually incoherent, unstable but can be coherent with ps
sources (NDi regime); Large fluctuations, incoherent for CW
Spectral power
density
Low, around 0.5mW/nm
Moderate values for ps-ns pulses, around few mW/nm;
High power for CW: 10s of mW/nm
Pump peak
power
fs input: few kW ps-ns input: Usually 100s of W; CW input: Few W
Convenience
Usually very expensive, complex,
and difficult to maintain
Cost-effective and low maintenance. Widespread applica-
tions for pulsed output.
Table 1. General characteristics of different SC generated by altering pump sources.
CW SC was demonstrated through fiber tapering, causing dis-
persive waves to be further blue-shifted. Wavelength extending
to 650nm can thus be reached even with moderate pump pow-
ers of around 35W [
41
]. Finally, through Ge-doping and fiber
tapering, the first pure CW white-light SC was generated in 2012
with a spectrum spanning 470nm to more than 1750nm at 9.3W
power [42].
3. WAVELENGTH SELECTION SCHEMES
Perhaps the most useful property of a SC source in fluorescence
microscopy lies in its massively broadband wavelength spec-
trum. Traditionally, fluorescence studies of cellular processes
have used mercury and xenon arc lamps for broadband illumina-
tion but suffer from low illumination efficiency and low spatial
coherence, and therefore cannot be used to improve resolution
in scanning-microscopy applications. Diodes and monochro-
matic lasers provide bright illumination without the necessary
broad bandwidth. The handful of monochromatic lasers that are
commonly available pose an unnecessary restriction on the vast
available toolbox of excitable fluorophores, rendering only a few
fluorophores usable whose excitation wavelengths match the
fixed laser wavelengths. Even then, the most efficient excitation
Research Article Journal of the Optical Society of America B 4
Fig. 2.
Different schemes to select desired excitation bands from SC sources: (a) filter wheel (b) prism [
32
] (c) axial monochromator
[43] (d) LVTF [44] (e) AOTF [45] and (f) AOBS [46]. Their individual characteristics are summarized in Table 2.
Characteristics Filter wheel Prism Axial monochromator LVTF AOTF AOBS
High transmission + − + + − +
Tuning speed − − − − + +
Wide wavelength range + + + + + +
Tunable bandwidth − + − + − −
Simultaneous multi-color − − − − + +
Out-of-band suppression + + + − − −
Steep spectral edge + − − − − −
Polarization insensitive + − − + − +
Table 2. Wavelength selection schemes with their pros(+) and cons(-).
wavelength can lie between available laser wavelengths [23].
The increasing complexity of bioimaging experiments and
rising demand to simultaneously image in multiple colors ne-
cessitate flexible use of excitation wavelengths across the entire
visible range to avoid crosstalk. Previously, dye laser systems
with tunable wavelengths have been used [
47
] to excite multiple
fluorophores, but the gain curves of these dyes limit the tuning
range to 50-100nm and do not cover a significant portion of the
visible spectrum. Tunable Ti:Sapphire lasers provide another
alternative to access the full visible spectrum but they can only
achieve this through processes like multi-photon absorption,
harmonic generation and optical parametric oscillators (OPO),
making the process expensive and difficult to operate, and re-
quiring specialist maintenance. SC sources overcome most of
these spectral restrictions with a large bandwidth spanning the
visible and NIR without any gaps, and thus lifting the restric-
tions of matching fluorophores to available laser lines. The full
SC spectrum is rarely used simultaneously for imaging. Most
fluorescence applications pick out desired excitation wavelength
bands from the output. While this means that most of the power
is discarded, the average power of the spectrally selected out-
put (
∼
1mW/nm from commercial sources) is still sufficient for
most imaging applications [
32
]. In conjunction with versatile
wavelength selection schemes, a SC source can simultaneously
excite a number of fluorophores, each at their optimal absorp-
tion wavelengths. This provides high specificity with minimal
cross-excitation and permits studies of various structures and
phenomena at the same time.
Good wavelength selection schemes should provide full flex-
ibility in selecting wavelengths and bandwidth over a large
spectrum at high speed with potentials for multiplexed imag-
ing. Many such technologies have been proposed but the most
noteworthy ones (see Figure 2) include bandpass filter wheels,
prism-based spectrometers [
32
], axial monochromators [
43
], mo-
torized linear variable tunable filters (LVTF) [44], acousto-optic
tunable filters and beam splitters (AOTF and AOBS) [
45
,
46
]. In
Table 2, we summarize and compare the characteristics of these
technologies.
Using a filter wheel with 6-12 bandpass filters can accommo-
date a wide range of wavelengths. The high-precision multi-
layer coatings on these filters provide excellent out-of-band sup-
pression (usually over OD 5) and high edge steepness. How-
Research Article Journal of the Optical Society of America B 5
ever, changing wavelengths requires physical movement of the
filter wheel in the beam path over 50-200ms timescale (when
motorized), precluding the possibility of simultaneous multi-
wavelength selective excitation. Dual, triple or quad band fil-
ters are now available for performing simultaneous excitation
but these are not tunable for arbitrary selection. A different
tunable implementation uses prisms to spatially disperse the
beam, part of which goes through a translatable aperture to se-
lect a central wavelength and bandwidth [
32
], but this leads to
large power losses and does not allow simultaneous excitation
of multiple wavelength bands. Another excitation technique
uses an on-axis monochromator based on a custom-designed
lens to intentionally maximize the beam’s longitudinal chro-
matic aberration while keeping other aberrations low [
43
]. This
wavelength-dependent longitudinal dispersion of foci along the
optical axis allows coupling desired part of the focused beam
into a finite aperture fiber and discarding all other out-of-focus
wavelengths.
To continuously tune the central wavelength along with band-
width, one can use LVTFs whose cutoff wavelengths vary lin-
early along their length. A useful configuration involves the SC
beam going through two LVTFs (one as shortpass, one as long-
pass) translated independently to make a variable bandwidth
filter[
44
]. Circular LVTFs use the same concept but work by
rotating the filters. LVTFs work well for tuning excitation when
the optical beam width is small in cross section as finite beam
widths deteriorate the edge steepness. The need for mechanical
translation slows their tuning speed (
∼
hundreds of ms) and
they cannot be used for simultaneous wavelength selection.
In contrast to previous technologies, AOTFs provide very
fast, programmable wavelength tuning (
µ
s) with simultaneous
output possible for up to eight different wavelengths over a
large spectral range (hundreds of nanometers) with no mov-
ing parts, demonstrated [
48
] for spectrally-resolved imaging.
Acousto-optic technologies effectively produce a phase grating
to diffract a specific part of the incident light with very narrow
passband (
∼
1nm) under phase-matching conditions of optical
and acoustic waves. However, AOTFs do suffer from low out-of-
band suppression, low transmission due to polarization selectiv-
ity, and sometimes from wavelength sidebands that introduce
a wavelength dependent angular spread in their output [
48
],
which degrades image quality. Additional compensation optics
like prism elements can correct for polarization selectivity to
restore high transmission and remove angular spread, and have
been commercialized as acousto-optic beam splitters (AOBS)
[
34
,
46
], making them a single versatile instrument handling
both excitation and emission selection efficiently.
No single wavelength filtering scheme fulfills the needs of
all imaging experiments, although acousto-optic technologies
provide more versatility than others. It is important to evaluate
and choose wisely between them by considering the criteria (eg.
speed, tunability, out-of-band suppression to minimize crosstalk,
simultaneous multi-color) most relevant for the experiment in
question. Once spectral selection is performed appropriately,
the output light can be coupled to a microscope directly for
widefield imaging, to a scan unit for point-scanning, or modified
in other ways (eg. beam shaping, pulse compression) for desired
applications.
4. MICROSCOPY MODALITIES USING SUPER-
CONTINUUM RADIATION
SC sources have revolutionized microscopy for biological and
medical imaging applications, offering significant advantages
over monochromatic lasers in spectral flexibility and over tradi-
tional lamp sources in terms of fast pulsed nature, high bright-
ness, spatial coherence, deep tissue penetration and contrast [
50
],
low-maintenance, and cost-effectiveness. In what follows, we
discuss how SC properties are exploited in different microscopy
applications. We review the use of SC spectral flexibility, par-
ticularly in hyperspectral imaging; SC spatial coherence used
for point scanning microscopies; the fast pulsed nature of SC
for spectroscopy and fluorescence lifetime imaging (FLIM); high
peak power applications in multi-photon excitation and sec-
ond harmonic generation; simplification of super-resolution
microscopy instrumentation using SC sources; and finally, the
bright, coherent light applications of SC for OCT and CARS
microscopies.
A. Widefield and confocal scanning techniques, and
wavelength-resolved imaging
SC sources provide a wider spectrum of wavelengths and much
greater brightness than thermal sources or LEDs for widefield
fluorescence microscopy techniques. In particular, incoherent
SC is best for widefield imaging since it provides flat-field illu-
mination without generation of speckle artefacts or aberrations.
These artefacts are seen with traditional coherent laser sources
due to interference between light waves and degrade image qual-
ity [
51
,
52
]. Incoherent SC with large spectral bandwidth and
excellent beam profiles are marketed commercially, permitting
easy integration into traditional setups for widefield microscopy.
The SC output can additionally be launched into a multimode
fiber to impose a strong spatial incoherence and uniform il-
lumination field. SC sources have found a greater market in
confocal laser scanning fluorescence microscopy, which was the
first bioimaging application to adopt SC radiation [
31
,
32
]. This
is because fiber-generated SC radiation can easily be focused
onto a diffraction-limited spot and point-scanned through the
sample. It is conceivable that high power commercial SC sources
will replace monochromatic lasers in all future confocal micro-
scopes, simplifying the setup, reducing costs and enhancing
versatility [
37
]. It is important to consider the effect of longitu-
dinal chromatic aberration and chromatic variations in beam
divergence potentially affecting the spatial resolution. A study
quantifying and comparing 3D point spread functions (PSF) in
the blue and red spectral regions in a confocal setup utilizing a
commercially available SC source found that the displacement
of the focal spot along the z-axis was comparable in extent to the
full-width-half-maximum of the PSF [
48
]. Therefore, chromatic
aberrations caused by using a SC source do not pose a signifi-
cant limitation for performing high-resolution confocal imaging
in multi-color throughout the visible spectrum. In other tech-
niques like volumetric confocal reflectance microscopy [
53
], this
chromatic aberration feature has been maximized using aspheric
lenses because it permits one to encode depth information spec-
trally and to be read out by a spectrum analyzer or spectrometer.
This way, multiple depths in biological specimen can be probed
simultaneously and rapidly, with one group demonstrating a
157
µ
m axial range with micrometer resolution captured in a
single shot from epithelial tissue [
54
]. This technique eliminates
the need for mechanical axial scanning, making it useful even
for endoscopic imaging systems.
![Fig. 3. Hyperspectral imaging reveals that different regions of Convallaria majalis have different emission properties at (a) 620nm and (b) 530nm excitations. (c) Excitation-emission plots for four regions in (a). Adapted from [49].](/figures/fig-3-hyperspectral-imaging-reveals-that-different-regions-ydd1hrz5.webp)
![Fig. 9. Coherent ultrashort pulses of tunable bandwidths in the UV range generated in gas-filled hollow-core PCFs, by adjusting pressure and gas mixtures (Ne, Ar, Kr, Xe gases). Adapted from [91] and [12].](/figures/fig-9-coherent-ultrashort-pulses-of-tunable-bandwidths-in-38g6x9m1.webp)
![Fig. 4. Intensity and intensity-merged FLIM images of autofluorescence from a section of unstained human pancreas showing clear lifetime differences between tissue types: A-artery, I-islets of Langerhans and C-collagen connective tissue. (a)-(c), λex=440-450nm, λem>470nm. (d)-(f), λex=520-530nm, λem>570nm. Adapted from [32].](/figures/fig-4-intensity-and-intensity-merged-flim-images-of-32ks911e.webp)
![Fig. 8. CARS, second and third harmonic generation images of a HeLa cell in the mitotic phase. Chromosomes appear in a line in the center of the cell in the DNA/RNA images and spindle fibers are seen in the SHG images. CARS images at (a-j) 3063, 3010, 2930, 2851, 1738, 1574, 1098, 1004, 791, and 719 cm−1, and the (k) SHG and (l) THG images. Adapted from [85].](/figures/fig-8-cars-second-and-third-harmonic-generation-images-of-a-30kdvtgl.webp)
![Fig. 1. Comparison of shot-to-shot differences in SC spectra obtained with femtosecond pumping in the normal dispersion (NDi - left panels) and anomalous dispersion regimes (ADi - right panels). The red/white lines (top left) represent a long term average (10,000 shots) of the all-NDi spectrum, where each individual shot trace (blue) mimics the shape of the average long-term spectrum. This stability in the NDi regime can also be observed in the 200 consecutive single-shot spectra (bottom left). The green/white lines (top right) are low pass filtered versions of individual shot-to-shot spectra, all of which show significant deviations from their long term average (red/white line in the bottom) in the ADi regime SC. Consecutive single shot spectra also show large differences (bottom right) in the ADi regime. Clearly, the dynamics of SC in the NDi regime are in stark contrast with dynamics in the ADi regime. Adapted from [24].](/figures/fig-1-comparison-of-shot-to-shot-differences-in-sc-spectra-295iwr3l.webp)