High‐Sensitive and Background‐Free Coherent Anti‐Stokes Raman Scattering Microscopy Using Delay Modulation

Coherent anti‐Stokes Raman scattering (CARS) microscopy has demonstrated a powerful platform for label‐free, non‐invasive, and chemically specific imaging of biological samples. However, the non‐resonant background limits its ability to detect weak Raman signals. Here, an approach to eliminate the non‐resonant background in CARS using delay modulation (DM) is presented, which is enabled by an acousto‐optic modulator and spectral focusing. The results demonstrate that DM‐CARS reduces the non‐resonant background by 10 times and improves detection sensitivity by 100‐fold over normal CARS. It also shows that DM‐CARS has potential applications in tracking heavy water metabolism in bacteria for antimicrobial susceptibility testing and imaging liver tumor tissues. Furthermore, a handheld DM‐CARS device has also demonstrated for in vivo imaging, which can be beneficial in various applications.


Introduction
[3][4][5] In CARS, a pump beam ( p ) and a Stokes beam ( S ) interact with the specimen via fourwave mixing.[15] This severely limits its applications for sensitive imaging of weak Raman bands.
Several approaches have been developed to reduce the non-resonant background, such as epi-CARS, [16] polarization-DOI: 10.1002/lpor.202300827sensitive detection CARS, [17,18] and timeresolved CARS. [19]However, these methods also have the downside of reducing the resonant signals.While multiplex [20] and interferometric CARS [21,22] do not reduce the resonant signals, multiplex CARS is not suitable for high-speed imaging as it requires long integration times, [20] and interferometric CARS may produce image artifacts for samples with variable refraction indexes. [21,22]s an alternative to background-free CARS, frequency modulation (FM) CARS can suppress the non-resonant background by switching the beating frequencies between resonant and non-resonant bands through optical design. [13,23][29][30][31] However, these approaches require complex laser intracavity design, optical path, or laser alignment.
Here, we demonstrate a technique called delay modulation CARS (DM-CARS) that can remove the non-resonant background in CARS.This is achieved by using spectral-focusing and an acousto-optic modulator (AOM) to switch the beating frequency between on-and off-resonances.The on-resonance detects signals from both the resonant CARS and non-resonant background, while the off-resonance detects only the nonresonant background.By subtracting the off-resonance from the on-resonance signals, we are able to achieve background-free CARS.Our design reduces the background by more than 10 times and improves detection sensitivity by ≈100 times.We have demonstrated two potential clinical applications of DM-CARS.First, we used it to perform metabolism-based antimicrobial susceptibility testing (AST) of bacteria, which allows for rapid determination of bacterial minimum inhibitory concentration.Second, we developed a handheld DM-CARS device delivered using a fiber for in vivo detection.

Principle and Instrumentation of DM-CARS
DM-CARS uses a spectral focusing scheme to disperse laser pulses through glass rods.The time delay between the pump and Stokes pulses determines the beating frequency (Figure 1).An AOM modulates the Stokes beam, generating a deflected beam (1st order) and a transmitted beam (0 order).The delays are adjusted so that the 0-order beam forms an on-resonant frequency, and the 1st-order beam forms an off-resonant frequency with the pump beam.By modulating the AOM, the CARS signals switch between traditional CARS signals and the non-resonant background.Finally, a lock-in amplifier demodulates the signals to subtract the background.
CARS signal at a beating frequency Ω =  p - S can be written as where (3) NR is the non-resonant term, R is the resonant term, and ∆ is the detuning ∆ =  p - S -Ω R .Here, Ω R is the center frequency of a homogenously broadened Raman line with a bandwidth Γ.According to formula (1), the resonant term will reach a maximum value when Ω is equal to Ω R and approach zero when Ω is much smaller or greater than Ω R .The third term will be zero when Ω is equal to Ω R or approach zero when Ω is much smaller or greater than Ω R .Thus, the CARS signal at the resonant frequency Ω 0 and the non-resonant frequency Ω 1 can be written as Then, the DM-CARS signal is which contains the resonant information only.
The schematic of the DM module is shown in Figure 2. The Stokes beam is modulated by an AOM to generate transmitted (0 order, S 0 ) and deflected (1st order, S 1 ) beams.Time delays between the pump and modulated Stokes beams are adjusted by delay stages (Figure S1, Supporting Information) to ensure that S 0 forms a beating frequency Ω 0 that matches Raman-active molecular vibration, while S 1 forms an off-resonant frequency Ω 1 which contains the non-resonant background only (Figure 2a,b).The optical path of S 1 is 3.75 meters longer than that of S 0 .This adds one pulse interval (1/f 0 = 12.5 ns, where f 0 = 80 MHz is the repetition rate of the laser) to S 1 to ensure that the pulses of S 1 and the pump overlap in the time domain.Meanwhile, the phase difference between S 0 and S 1 remains to be  since the pulse interval does not significantly shift the phase (f 1 /f 0 = 10 −2 , where Thus, the non-resonant background can be eliminated by using a lock-in amplifier through the in-phase channel. [13]With the spectral focusing scheme, the frequency difference can be adjusted by tuning the time delay between S 0 and S 1 .In our setup, we achieved a frequency difference of ≈100 cm −1 when the delay is tuned to be ≈700 fs.

Performance of DM-CARS
The pulse trains of modulated Stokes beams were recorded using a photodiode and an oscilloscope (Figure 2c,g).The results indicated that the modulation depth for the 0-order beam is approximately 86%, resulting in a residual pulse train.Since the intensity of the residual remains constant, it does not contribute to the CARS signal.Similarly, a residual was also observed in the 1st-order beam with a modulation depth of 97%.As expected, when both output beams were unblocked, the intensity of the pulse trains was constant (Figure S2a, Supporting Information).
To characterize the performance of DM-CARS system, we recorded CARS and DM-CARS images of dimethyl sulfoxide (DMSO).We tuned the on-resonant frequency to 2913 cm −1 , which is the Raman peak of DMSO (Figure 2d), and the offresonant frequency to 2835 cm −1 (Figure 2h).The frequency difference corresponds to a time delay of 483 fs between the 0-and 1st-order beams.While the CARS image shows a strong signal from DMSO in the on-resonance (Figure 2e), the CARS signal is largely reduced in the off-resonance (Figure 2i).As expected, the DM-CARS signal, which subtracts the off-resonance from the on-resonance, provides the difference between the on-and offresonant CARS signals (Figure S2c, Supporting Information).We also conducted measurements of 5% DMSO and polystyrene beads using CARS and SRS (Figure S3, Supporting Information), which confirm the presence of non-resonant background and spectral distortions due to the presence of non-resonant background in CARS.
To evaluate the capability of DM-CARS in suppressing the non-resonant background, we performed normal CARS and DM-CARS imaging of PMMA beads mixed with TiO 2 .The TiO 2 particles are known to have strong four-wave mixing signals [33] and thus a strong non-resonant background in CARS (Figure 3).As shown in Figure 3a, the CARS spectrum of PMMA beads has a peak at ≈2980 cm −1 , whereas TiO 2 has a constant and significantly higher CARS signal than the off-peak intensity of PMMA beads.As a result, we tuned the on-and off-resonant frequencies for imaging to 2980 and 3070 cm −1 , respectively.While both PMMA beads and TiO 2 particles produce signals in the onresonance CARS images (Figure 3b), the TiO 2 signals still exist at similar levels in off-resonance (Figure 3c), confirming the non-resonant background as the main source for TiO 2 signals.
In contrast, DM-CARS efficiently suppresses the non-resonant background, as evident from the absence of TiO 2 signals in the off-resonance DM-CARS image (Figure 3e-g).The decrease in the background signals in TiO 2 is estimated to be ≈10 times based on the intensity ratio between the off-resonance CARS and DM-CARS images.
We measured the lateral and axial resolutions of the DM-CARS using 200 and 500 nm polystyrene beads (Figures S4 and S5, Supporting Information).The lateral resolution of DM-CARS was determined to be 350 nm and the axial resolution was 1.34 μm.The results showed that DM-CARS has improved the lateral and axial resolutions of CARS by ≈7% and 20%, respectively.

Detection Sensitivity of DM-CARS
The sensitivity of DM-CARS was evaluated using DMSO-d 6 (Figure 4).The on-resonant frequency of the C-D stretching mode was set to 2128 cm −1 , while the off-resonance was set to 2191 cm −1 (Figure S6, Supporting Information).The results show that normal CARS is limited to detecting ≈2.5% (v/v) DMSO-d 6 in water, whereas DM-CARS improves the detection limit to ≈0.025% (v/v) DMSO-d 6 in water, which is a 100-fold improvement over normal CARS.This means that DM-CARS can detect 3 × 10 5 molecules (2 × 10 6 oscillators) in a probe volume of ≈190 al, or a molecular concentration of 3 mmol/l for DMSO-d 6 .
The time constant in these tests is 7 μs, which allows for highspeed imaging (0.65 s/frame for 200 × 200 pixels).When using the same time constant, the detection sensitivity of DM-CARS is ≈1.9 times better than that of previously reported FM-CARS. [13]oreover, the DM-CARS imaging speed is at least 20000 times faster than the fiber laser-based FM-CARS [25] to achieve a similar detection sensitivity.

Metabolism-Based Antimicrobial Susceptibility Testing
It is crucial to have rapid AST methods in clinical settings to determine bacterial susceptibility to antibiotics.[36] However, measuring the relatively weak C-D signal in bacteria using CARS is challenging due to the presence of a non-resonant background (Figure S7, Supporting Information).
To investigate whether rapid AST can be achieved using DM-CARS, we performed DM-CARS imaging of P. aeruginosa (Figure 5).We cultured P. aeruginosa with D 2 O and serially diluted gentamicin, following the same procedure as in our previous study. [35]For CARS and DM-CARS imaging, we tuned the on-and off-resonant frequencies to 2141 cm −1 (close to the C-D Raman peak) and 2268 cm −1 , respectively (Figure 5a).In CARS, bacterial C-D signals are overwhelmed by the non-resonant background, as bacteria cultured in a normal medium (without C-D bonds formed) have comparable intensities with those cultured with D 2 O (Figure S7a, Supporting Information).In contrast, the non-resonant background in bacteria is significantly reduced in DM-CARS images (Figure 5b).These results show that DM-CARS could efficiently eliminate the non-resonant background and retrieve resonant signals of bacteria (Figure 5c,d).DM-CARS images (Figure 5c) and the corresponding statistical analysis (Figure 5d) show significantly reduced intensity in bacteria with 1 μg mL −1 and higher concentrations of gentamicin, indicating inhibited D 2 O metabolism.These results agree well with the conventional culture-based broth microdilution results (Figure S9, Supporting Information).On the contrary, these results could not be obtained using normal CARS (Figure S7b,c, Supporting Information).

DM-CARS Imaging of Liver Tumor Tissue
To evaluate the performance of DM-CARS on biological samples, we imaged liver cancer tissues using DM-CARS and CARS (Figure 5e,f).The images were taken with an on-resonant frequency at 2850 cm −1 and an off-resonant frequency at 2792 cm −1 .In the CARS image, the presence of non-resonant contribution resulted in reduced contrast and hidden image details.By reducing the non-resonant contribution, DM-CARS improved the image contrast and revealed more image details.

Fiber-Delivered Handheld DM-CARS
A handheld device based on fiber is more suitable for clinical scenarios due to its flexibility and smaller size.However, the fiber can introduce a background signal that may interfere with CARS signal detection. [37]To overcome this challenge, we integrated the DM module to suppress background from fibers in addition to non-resonant background.The schematic and prototype apparatus of the handheld DM-CARS device are shown in Figure 6a,b, respectively.The device uses a single-mode fiber to transmit laser beams into the detector and a multi-mode fiber to transmit signals to PMT.Although multi-mode fibers have a larger optical capacity and can direct light more easily into the core, the presence of multiple modes can cause modal dispersion, negatively impacting beam and imaging quality. [38,39]Therefore, we used a single-mode fiber to transmit the excitation beams to ensure good beam quality and a multi-mode fiber to collect signals in the handheld DM-CARS system.We determined the spatial resolution of the handheld DM-CARS by imaging DMSO and analyzing the intensity profile across the edge with a Gaussian distribution (Figure 6c).The spatial resolution was found to be ≈2.4 μm, which is similar to the previous reports. [40]We also tested the device's ability to eliminate background signals by performing DM-CARS imaging of a PMMA beads-TiO 2 mixture and liver tumor tissues (Figure 6d,e).Compared to CARS imaging, the handheld DM-CARS reduced the background signals in TiO 2 by ≈10 times, indicating effective suppression of both the non-resonant background and background from the fiber.Moreover, DM-CARS produced considerably better contrast in tissue images compared to CARS.

Discussion
The DM module in this work has two parts: AOM and spectral focusing.Previously, the Xie [28] and Lim groups [29] used Pockels cells for similar technology.Compared with the Pockels cells, AOM only requires low voltage and a radio frequency signal, un-like the Pockels cells that need high voltage (≈5000 V) power supplies and signal amplifiers. [41]Spectral focusing schemes are commonly used in hyperspectral coherent Raman scattering imaging. [7,42,43]In DM-CARS, DS1 and DS2 positions were adjusted to ensure that the non-resonant background at the resonant and non-resonant frequencies has similar intensities for maximum suppression.Moreover, this design can utilize laser power with maximum efficiency and reduce phototoxicity to the samples.
For AST application in this work, we modified the DM-CARS to work in epi-detection mode.According to the radiation pattern, [44,45] CARS has both forward and backward radiations.Large scatters have strong forward radiation but negligible backward radiation, whereas small scatters have significantly enhanced backward radiation.Therefore, epi-detection is more suitable for imaging small objects like bacteria.We further enhanced the epi-detection scheme by placing a sliver mirror beneath the samples to collect both the forward and backward signals.Although the reflected forward signals contain unwanted Optical fibers are often used to deliver excitation beams to detectors in handheld imaging equipment.][48][49] Fortunately, the DM module can suppress this background generated from fibers.Additionally, we expect that the DM module can be adapted to SRS to suppress the backgrounds caused by cross-phase modulation, transient absorption, and photothermal effects.

Experimental Section
Experimental Setup: The setup is shown in Figure S1 (Supporting Information).A dual-output femtosecond laser with a repetition rate of 80 MHz (f 0 ) (InSight X3, Spectra-Physics) provides two synchronized beams, used as the pump and Stokes.The pump beam is tunable from 680 to 1300 nm with a pulse width of 120 fs, and the Stokes beam is fixed at 1045 nm with a pulse width of 220 fs.Half-wave plates (HWP) and polarizing beamsplitter (PBS) were used to control the power of the beams.An acoustic-optic modulator (AOM, 1205-C, Isomet) modulates the Stokes beam at a frequency of 2.8 MHz (f 1 ).The transmitted Stokes beam (0 order Stokes beam, S 0 ) collinearly combines with the pump beam through a dichroic mirror (DM, DMSP950, Thorlabs).After the combination, beams are chirped by four 100 mm SF11 glass rods (Newlight) and delivered to a Preparation of Bacterial Samples: P. aeruginosa was cultivated in LB (Luria-Bertani broth, Sigma-Aldrich) for 2 h to reach the log phase.Then, the bacteria were first treated with gentamicin containing LB medium for 1 h, then LB mediums containing D 2 O and the same concentration of gentamicin was added to the bacteria for an additional 2 h, reaching a final D 2 O concentration of 70%.After the treatment, the samples were centrifuged and deposited into poly-lysine-coated coverglass. [36]IC Determination: The MIC of gentamicin, tested against P. aeruginosa, was determined using the gold standard broth microdilution assay.P. aeruginosa was cultured in LB in a 96-well plate.Gentamicin using triplicate samples was added to the plate and serially diluted.The plate was incubated at 37 °C for at least 20 h before determining the MIC.Finally, the plate was visually inspected, and the MIC was classified as the concentration at which no visible bacterial growth was observed.
Handheld Detector Scheme: A fiber collimator after the glass rods was used to couple the lasers from free space into a single-mode fiber (C780-PC-1, LBTEK).After the fiber, the laser passed through another collimator, followed by a 4-f system and a right-angle mirror to correct laser collimation.Then the laser passes through a 2-D scanning galvo (GVS102, Thorlabs), another 4-f system, and a 40x objective lens (NA = 0.8, LUMPLFLN, Olympus) for scanning the sample and exciting DM-CARS signal.The backward DM-CARS signal was collected by the objective lens and descanned to the dichromatic mirror (ZT730rdc-UF1, Chroma).The dichromatic mirror, which cut off at ≈730 nm, reflects the signal into the collection fiber.The collection fiber was a multi-mode fiber (MMC400L-0.37-PC-1,LBTEK) with a core diameter of 400 μm.After the fiber, the signal goes through a 645 nm bandpass filter (ET645/75m, Chroma) and was finally collected by a PMT (H7422-40, Hamamatsu).
For the on-resonant imaging of PMMA beads-TiO 2 by the handheld DM-CARS, the Raman shifts were adjusted to 2980 cm −1 for the 0 order and 3070 cm −1 for the 1st order.For off-resonant imaging, the Raman shifts were 3027 cm −1 for the 0 order and 3117 cm −1 for the 1st order.In the case of on-resonant imaging of liver tumor tissues, the Raman shifts were adjusted to 2850 cm −1 for the 0 order and 2672 cm −1 for the 1st order.

Figure 2 .
Figure 2. Schematic diagram of the DM module and the corresponding temporal and spectral profiles of the 0 and 1st orders of AOM.a) Schematic diagram of the 0 order when the AOM is tuned to transmission mode.b) Schematic diagram of the first order when the AOM is tuned to deflection mode.The change of the AOM state is converted into spectral change by a spectral focusing scheme.L: lens.AOM: acousto-optic modulator.M: mirror.DS: delay stage.BS: beam splitter.DM: dichroic mirror.c,g) The pulse train profiles of the zero order and first order Stokes beam output from the AOM in the time domain.d,h)The beating frequency corresponds to the 0-and 1st-order beams.e,i) The CARS images of DMSO correspond to the 0-and 1st-order beams alone.f,j) The intensity profiles along the dash lines in images (e) and (i).

f 1 = 2 . 8
MHz is the modulation rate of AOM).The DM-CARS signals can then be written as

Figure 3 .
Figure 3. Performance of DM-CARS.a) CARS spectrum of PMMA beads and TiO 2 particles.The orange and green dash lines indicate the two Raman shifts used by on-resonant CARS and DM-CARS imaging.The Raman shifts used in off-resonant imaging are 3027 cm −1 (0 order) and 3117 cm −1 (1st order).b, c) On-resonant and off-resonant CARS imaging of the mixed solution of PMMA beads and TiO 2 .d) The intensity profile along the dashed line in image (b).e-f) On-resonant and off-resonant DM-CARS images of a mixed solution of PMMA beads and TiO 2 .g) The intensity profile along the dashed line in the image (e).Scale bar: 20 μm.

Figure 4 .
Figure 4. Detection sensitivity of DM-CARS.Serial diluted DMSO-d 6 was used to measure the sensitivity of the system, where I is the signal of DMSO-d 6 dilutions, and I max is the signal of pure DMSO-d 6 .The traditional CARS signal is visualized as purple dots, whereas the DM-CARS signal is shown as green rectangles.The black dotted line is the noise baseline of the system.

Figure 5 .
Figure 5. DM-CARS imaging of the metabolic change in microorganisms and the imaging of liver tumor tissue.a) Spontaneous Raman spectra of P. aeruginosa after culture in heavy water medium.The orange and green dash lines indicate the on-and off-resonant Raman shifts.b) Imaging results of P. aeruginosa in normal medium culture and heavy water culture.Scale bar: 10 μm.c) Epi DM-CARS imaging of P. aeruginosa cultured in D 2 O-containing mediums with the addition of serially diluted gentamicin.Scale bar: 5 μm.d) Statistical analysis of the DM-CARS C-D intensity in P. aeruginosa in (c).e) DM-CARS image and CARS image of liver cancer tissue.Scale bar: 20 μm.f) The signal distributions on the dash lines in (e).

Figure 6 .
Figure 6.Fiber-delivered handheld DM-CARS microscopy.a) Optical path design of the handheld microscopy.L: lens.M: mirror.DM: dichroic mirror.SU: scanning unit.OBJ: objective.F: filter.b) Digital image of the handheld microscopy.c) First derivative of the DM-CARS profile of DMSO at the interface of a gold edge and Gaussian fitting.Left insert graph: DM-CARS image of the edge of DMSO.Right insert graph: DM-CARS profile of DMSO at the interface of a gold edge.d) CARS and DM-CARS images of the mixed solution of PMMA beads and TiO 2 at the on-resonant Raman shift and offresonant Raman shift of PMMA beads.Scale bar: 10 μm.e) CARS and DM-CARS images of liver cancer tissue.The right panel is a zoom-in comparison of the same region shown in the orange and green boxes on the left panel.Scale bar: 20 μm.
commercial upright laser scanning microscope.A 60X oil immersion objective (NA = 1.3, UPlanApo, Olympus) is used to focus the beams on samples.The samples are scanned by galvo mirrors (GM, GVS102, Thorlabs).The transmitted beams after samples are collected by an oil immersion condenser (NA = 1.4,U-AAC, Olympus) and filtered by bandpass filters.Finally, the modulated signal received by a photomultiplier tube (PMT, H7422-40, Hamamatsu) is demodulated by a lock-in amplifier (LIA, HF2LI, Zurich Instrument).The power of the pump, 0 order, and 1st order beams for different samples are as follows: for DMSO, the powers are 5, 200, and 200 mW, respectively; for PMMA beads-TiO 2 , the powers are 50, 50, and 50 mW; for DMSO-d 6 , the powers are 200, 100, and 100 mW; and for P. aeruginosa, the powers are 100, 80, and 80 mW; for liver cancer tissue, the powers are 100, 200 and 200 mW.For the handheld DM-CARS, the powers are 100, 25, and 25 mW.Hyperspectral CARS: For hyperspectral CARS imaging at C─H vibration, the pump was tuned to 798 nm.The Stokes was modulated by AOM at 2.8 MHz, then the pump and 0 order Stokes beams were collinearly combined, chirped by four 100 mm long SF11 glass rods for spectral focusing, and directed into a laser-scanning microscope.A 60X oil immersion objective (NA = 1.3, UPlanApo, Olympus) focused the beams into the sample.An oil condenser collected the light in the forward direction.Two filters blocked the pump and Stokes beams.The CARS signals were collected by a photomultiplier tube (H7422-40, Hamamatsu).