High-Speed Wide-Field Broadband Coherent Raman Micros
High-Speed Wide-Field Broadband Coherent Raman Micros
High-Speed Wide-Field Broadband Coherent Raman Micros
1
Executive summary Andrea Rabolini
phase-retrieval algorithms, like Kramers-Kronig where Ai are the envelope function of the pulses,
(KK) [2]. The CARS technique can reach very ωi χ(2)
di = 4n(ω and ∆k = k3 − k2 − k1 . The ki
1 )c
high acquisition speeds, working in the single- terms are the wavenumbers of the waves. Let us
frequency approach where just one vibrational imagine an interaction of three pulses with fre-
mode is excited at once. In this way one chemi- quencies ω1 , ω2 and ω3 = ω1 + ω2 . The case in
cal species at a time can be studied. Broadband which the wave at ω3 is much more intense than
CARS (B-CARS) allows to acquire entire spec- the wave at ω2 is called Optical Parametric Am-
tra in a single shot. It has two main approaches: plifier (OPA) and it is an energy exchange be-
hyperspectral, where a single vibrational mode tween the ω3 wave (called pump) and the wave
is excited at a time and the full spectrum can at ω2 (called signal). A third wave at ω1 (the
be retrieved by detuning one narrowband beam; idler) is generated due to conservation of energy.
and multiplex, where the Stokes is a broadband To provide efficient signal generation, the phase-
pulse and several vibrational modes are excited matching condition ∆k = 0 must be fulfilled.
simultaneously. If the wave vectors of the three pulses are
A technique that allows high acquisition speeds not collinear, the process takes the name of
is wide-field CARS (WF-CARS), where pump non-collinear OPA (NOPA). In this case, the
and Stokes have a FWHM of hundreds of µm, wavevector k1 is generated along a direction that
enabling acquisition of wide portions of the ensures conservation of momentum:
sample at once. Thus, video-rate imaging
k3 = k1 + k2 (4)
(1 ms/frame) can be achieved more easily than
in scanning CARS and fast biological dynamics
can be investigated in real time [3].
2. Theory
If light intensities are higher than approximately
109 W/cm2 , the response of the material be-
comes non-linear and a perturbative approach
must be used. The polarization can be repre-
Figure 1: Wave vector conservation in NOPA.
sented as P = P (l) + P (nl) , highlighting the lin-
ear (P (l) ) and non linear (P (nl) ) components. In NOPAs we can reach both phase matching
The latter term can be written as the sum of and group velocity matching by tuning angles θ
increasing non-linear order: and Ω, thus obtaining a large gain over a large
bandwidth. However, not in all crystal it is
P (nl) = P (2) + P (3) + . . . (1) possible to satisfy both conditions, due to de-
pendences on the refractive indexes and on the
The second-order non-linear polarization, under
wavelengths.
certain approximations, can be written as:
We will now focus on CARS, which is a third-
order non-linear process, thus we will analyze
P (2) (t) = ϵ0 χ(2) E 2 (t) (2)
the P (3) term. The third-order non-linear po-
larization, under certain approximations, can be
where χ(3) is the third order susceptibility of the
written as:
material. The interaction of three monochro-
matic waves at frequencies ω1 , ω2 and ω3 in a P (3) (t) = ϵ0 χ(3) E 3 (t) (5)
χ(2) material can be represented by a system of
where χ(3) is the third-order susceptibility of the
coupled equations:
material. Let us consider an interaction of an in-
∂A1 coming electric field interacting with a χ(3) ma-
= −i d1 A3 A∗2 e−i∆kz terial. The electric field has two components,
∂z
pump and Stokes:
∂A2
= −i d2 A3 A∗1 e−i∆kz (3)
∂z 1
E(z, t) = {Ap (z, t)ei(ωp t−kp z)
∂A3 = −i d A A e+i∆kz
2 (6)
3 1 2
∂z +AS (z, t)ei(ωS t−kS z) }
2
Executive summary Andrea Rabolini
where ωi are their central frequencies and ki sending the pump on the sample almost from
their wavenumbers. The CARS process arises the side; the phase-matching condition is ful-
thanks to the coherence created by the interac- filled on the whole sheet of light, suppressing
tion between pump (ωp ) and Stokes (ωS ), when non-resonant CARS from planes above or be-
the beating frequency Ω = ωp − ωS matches a low the sample. In the second method, pump
molecular vibration ΩR = Ω. and Stokes are in a non-phase-matching con-
The resulting polarization is: figuration because in the sample, thanks to re-
flection and scattering of particles, some por-
3ϵ0 χ(3) 2 ∗ i[ωaS t−(2kp −kS )z] tion of the incoming light is brought into the
Pω(3)
aS
(z, t) = Ap AS e + c.c.
4 phase-matching condition, creating the CARS
(7) signal. In our setup, we implement a non-phase-
which is oscillating at ωaS = 2ωp −ωS frequency. matching configuration with collinearly focused
The CARS intensity is: beams. The coherence length (i.e. the length in
which the maximum of the anti-Stokes signal is
generated) is ≈30 µm, which is higher than the
2 ∆kL
ICARS (L) ∝ χ(3) Ip2 IS L2 sinc2 (8)
2 sample thickness (≈10 µm). This ensures effi-
cient CARS signal generation even without the
where ∆k = 2kp − kS . The CARS signal scales
fulfilment of the phase-matching condition.
quadratically with the pump intensity Ip and lin-
early with the Stokes intensity IS . To obtain
high intensities, the phase-matching condition
4. Experimental setup
(∆k = 0) should be fulfilled. The setup is schematically represented in
Fig.2. The source is a Coherent Monaco, an
3. Wide-field CARS Ytterbium-amplified laser, emitting at 1035 nm
with a spectral bandwidth of 10 nm. For our ex-
Traditionally CARS microscopy is performed in
periments, we used a total power of 40 W, with
raster scanning fashion. The two beams are
a repetition rate of 2 MHz and pulse duration of
tightly focused onto the sample employing high
269 fs.
numerical aperture (NA) objectives and signal is
generated in a small portion (i.e. a pixel) of the
sample. To generate an image, one must perform
raster scanning, either using galvanometric mir-
rors to move the excitation beams or piezoelec-
tric stage over which the sample is fixed. This
causes some problems, for example it is time-
consuming since one mirrors or the piezo have
to move over a region which can be 100×100
pixels or even more. In wide-field CARS (WF-
CARS), the pump and the Stokes on the sample Figure 2: Scheme of the experimental setup.
are bigger than in point-scanning CARS, typi- Legend: M: mirror; L: lens; SPF: short pass fil-
cally hundreds of µm. Thus it is possible to ac- ter; LPF: long pass filter; HWP: half wave plate;
quire whole regions even in a single shot, allow- PBS: polarizing beam splitter; OBJ: objective
ing video-rate imaging. The major drawback of lens.
WF-CARS is that wide-field illumination does
not present the spatial filtering capabilities of- To ensure enough Stokes power on the sample
fered by confocal microscopes, thus giving lower and therefore enough CARS signal, the Stokes
spatial resolution and lower axial sectioning ca- beam has to be amplified: for this reason we
pability. Two main approaches for WF-CARS implemented a NOPA. After the laser output,
illumination have been implemented: the first a half-wave plate (HWP) brings part of the
method consists in illuminating the sample with beam into vertical polarization. A first polar-
a thin "sheet of light" (5-10 µm) by means of izing beam splitter (PBSW-1030) is mounted.
a dark-field condenser with high NA (typically The reflected component gives rise to the pump
NA=1.2-1.4) which creates a "cone" of light,
3
Executive summary Andrea Rabolini
4
Executive summary Andrea Rabolini
5
Executive summary Andrea Rabolini
total acquisition time can become almost 20 s, home-made microscope. With respect to scan-
leading to an acquisition time for each pixel over ning CARS, WF-CARS requires reduced times
the whole CARS spectrum of ≈437 µs, which is of acquisition. This unlocks unprecedented ac-
better than the state of the art of WF-CARS. quisition speed, useful to observe high-speed bi-
The first sample analyzed is the DMSO. The re- ological phenomena.
trieved spectrum is shown in Fig. 6. The setup has been tested to detect CARS spec-
tra of test samples both in multiplex and hyper-
spectral CARS configurations. We acquired also
some CARS hypercubes from which we recon-
structed the spectrum and applied a Kramers-
Kronig algorithm to retrieve the Raman signal.
We were able to acquire some images of a bio-
logical sample (NASH liver). We have obtained
video-rate imaging (6 ms/pixel) on test samples,
but the acquisition time can be reduced down to
≈0.5 ms/pixel. In the future we aim to acquire
Figure 6: CARS spectrum of DMSO recon- hyperspectral images of biological samples at the
structed started from the intensity of a chosen video-rate level, allowing us to study fast biolog-
pixel along 350 frames of a hypercube. ical processes.
An important improvement in the setup would
We performed the phase retrieval using a KK be the implementation of a Fourier-Transform
algorithm. During the process, the retrieved detection to more easily analyze the hypercubes.
spectrum was slightly shifted in order to match In this way, the slit and the calibration of the
the spontaneous Raman peaks. For the phase- rotational stage are no more needed, improving
retrieval algorithm, the pure NRB spectrum is the precision and repeatability of measurements.
required: on the sample stage we positioned the
glass and the coverslip without the solvent. References
The result is shown in Fig. 7 along with the [1] Dario Polli et al. Broadband coherent ra-
spontaneous Raman spectrum of DMSO. We can man scattering microscopy. Laser & Pho-
see that there is a clear correspondence of the tonics Reviews, 12(9):1800020, 2018.
peaks at 667 cm−1 , 1043 cm−1 and 1420 cm−1 .
[2] Federico Vernuccio et al. Fingerprint multi-
6. Conclusions plex cars at high speed based on supercon-
This master thesis is focused on the design and tinuum generation in bulk media and deep
development of a wide-field broadband CARS learning spectral denoising. Opt. Express,
30(17):30135–30148, Aug 2022.