2.1.

Cascaded Gas-Filled ARHCF Laser

The MIR gas-filled fiber Raman laser source at $3.4\mu \mathrm{m}$ is based on a configuration of cascading two gas-filled ARHCFs. To efficiently convert the conventional NIR to the MIR at the selected $3.4\mu \mathrm{m}$ wavelength, here, the second-stage ARHCF is filled with pure hydrogen (${\mathrm{H}}_2$), offering a long vibrational Raman stokes (VRS) coefficient of $4155{\mathrm{cm}}^{-1}$ and also a relatively high gain coefficient compared with other Raman active gases.⁵² This requires a pump wavelength of $\sim 1.4\mu \mathrm{m}$, which is outside the gain range of the known rare-Earth-ion-based fibers. Therefore, we first generated the required $\sim 1.4\text{-}\mu \mathrm{m}$ laser using the first-order VRS ($2331{\mathrm{cm}}^{-1}$) of ${\mathrm{N}}_2$ in the first ARHCF stage,⁵³ pumped by a Yb-doped fiber laser at the $1\text{-}\mu \mathrm{m}$ region. The pump laser has an all-fiber structure consisting of a modulated diode laser seed followed by a Yb-doped fiber amplification module, emitting a pulse train with a repetition rate of 1.2 kHz, $\sim 3.7\text{-}\mathrm{ns}$ pulse duration, energy of $\sim 98\mu \mathrm{J}$ (measured by an energy meter, PE9-ES-C, Ophir Optronics, Jerusalem, Israel), and 0.128-nm linewidth at a 1060-nm wavelength.³² A $\lambda /2$ waveplate is placed at the output of the amplifier for adjusting the polarization orientation to acquire the highest Raman conversion efficiency in the subsequent gas-filled ARHCF system. Then, the beam is coupled into the ${\mathrm{N}}_2$-filled ARHCF to generate the first-order VRS line at 1409 nm,⁵³ which is then used as a pump for the second-stage ARHCF, to generate the Raman laser at MIR $3.4\text{-}\mu \mathrm{m}$ wavelength.

The scanning electron microscopy (SEM) images of the two ARHCFs used in our experiments are shown in Fig. 1. The first-stage ARHCF (fiber 1 in Fig. 1) is a 16-m-long single-ring nodeless ARHCF, consisting of seven rings with a diameter of $16.1\mu \mathrm{m}$ and a wall thickness of $\sim 323\mathrm{nm}$, forming a negative-curvature core shape with an inner jacket tube diameter of $32.8\mu \mathrm{m}$. The simulated loss spectrum of the fiber is shown in Fig. S1(a) in the Supplementary Material, with a loss value of $\sim 0.02\mathrm{dB}/\mathrm{m}$ at 1060 nm and $\sim 0.05\mathrm{dB}/\mathrm{m}$ at 1409 nm.⁵³ Based on the measured fiber’s parameters, the simulation was implemented based on the finite-element method using the COMSOL software with parameter settings in Ref. 54. A fine mesh size ranging from $\lambda /6$ to $\lambda /4$ was adopted to ensure the accuracy of our simulation. A perfectly matched layer was used outside the fiber structure as boundary conditions to accurately determine the leakage loss. The absorption coefficient of fused silica used in the simulation is from Ref. 55. The surface scattering loss was ignored because of the negligible surface roughness of the silica ARHCF compared with the wavelength in the IR region.⁵⁶ The second-stage ARHCF (fiber 2 in Fig. 1) is 5 m long with a nested cladding structure and a core diameter of $82\mu \mathrm{m}$.³² The diameter for external and internal capillaries are 40.3 and $13.6\mu \mathrm{m}$, and the wall thickness are 987 nm and $1.37\mu \mathrm{m}$, respectively. This ARHCF has a loss of only $\sim 0.004\mathrm{dB}/\mathrm{m}$ at 3409 nm [see Fig. S1(b) in the Supplementary Material], which is a critical condition for the efficient generation of the $3.4\text{-}\mu \mathrm{m}$ Raman laser. In addition, the nested structure of the second ARHCF significantly suppresses the bend loss and therefore allows a bending diameter of $\sim 40\mathrm{cm}$ facilitating the development of a compact PAM system. The first ARHCF can be also coiled with $\sim 40\mathrm{cm}$ diameter without high bending loss due to the small core diameter.⁵⁷

2.2.

Mid-Infrared Photoacoustic Microscopy

The generated $3.4\text{-}\mu \mathrm{m}$ Raman laser beam is then expanded and coupled into the PAM. The output energy is properly attenuated using a neutral density filter (NDF, NDC-50C-2M, Thorlabs, Newton, New Jersey, United States) to avoid sample damage. The beam is focused by a reflected IR objective lens (PIKE, 40×, 0.78 NA) which has a designed obscuration of 42.9%. The PAM is performed in a transmission mode, and the sample is placed above a transparent sapphire-based coverslip (#18-471, Edmund Optics, Barrington, New Jersey, United States). Then, a water tank with a flat polymer bottom surface ($\mu$-Dish, ibidi, Fitchburg, Wisconsin, United States) with $200\mu \mathrm{m}$ thickness is placed tightly above the sample to fix and flatten the sample surface and to couple the acoustic signal into the distilled water filled in the tank. The water tank and the sample are combined as an integrated part held by a high-resolution $X-Y$ stage (8MTF-75LS05, Standa, Vilnius, Lithuania) driven by a stepper and direct current (DC) motor controller (8SMC5-USB, Standa) to enable the sample scanning with a minimum step size $<10\mathrm{nm}$ and a maximum speed of 35,000 steps/s. A focused immersive ultrasound transducer (UT, Precision Acoustic, Dorchester, United Kingdom) with a central frequency of 20 MHz is immersed within the distilled water in the tank. The UT has an acoustic focal length of 8 mm and a diameter of 10 mm and is coaxially aligned with the reflective IR objective lens. The detected PA signals are filtered by two analog filters (mini-circuits, 1-MHz long-pass filter, and 27-MHz low-pass filter) then amplified by a low-noise wideband amplifier (Spectrum Instrumentation, Großhansdorf, Germany) and finally received with a high-speed digitizer (M4i.4421-x8, Spectrum Instrumentation) for data processing. The digitizer, integrated into a computer, operates at $250\mathrm{MS}/\mathrm{s}$ sampling rate with a voltage resolution of 16 bits.

An external trigger delay controller (AeroDiode) synchronizes the scanning and data acquisition. A small fraction of the pump pulse energy is extracted from the pump beam and is recorded by a NIR photodetector (DET08C/M, Thorlabs) as a trigger signal, as shown in Fig. 1. Then, the output signal from the photodetector is connected to the input side of a trigger delay controller as the input trigger signal. Once the trigger delay controller detects the pump pulse signal, a square signal is generated and acts as the PA trigger for $X-Y$ stage movement and signal synchronization. The PA signal is recorded $\sim 5\mu \mathrm{s}$ after the laser pulse and $\sim 0.2\mu \mathrm{s}$ after the square trigger signal at a repetition rate of 1.2 kHz. During the sample scanning, to minimize noises and enhance the signal-to-noise ratio, 100 pulses (A-lines) are averaged corresponding to a $\sim 83\text{-}\mathrm{ms}$ dwell time.

2.3.

Brain Sample Preparation

To investigate the performance of our MIR-PAM in real brain tissue, wild-type adult mice were employed. The procedure to prepare the brain slices presented below is approved by the Animal Experiments Inspectorate under the Danish Ministry of Food, Agriculture, and Fisheries, and all procedures adhere to the European guidelines for the care and use of laboratory animals, EU Directive 2010/63/EU. A Long–Evans wild-type adult rat was anesthetized by intraperitoneal injection of $200\mathrm{mg}/\mathrm{kg}$ sodium pentobarbital. Hereafter, the level of anesthesia was assessed by pedal reflex pain response to firm toe pinch. Once a sufficient level of anesthesia was reached, the chest cavity was opened to expose the heart. A catheter, connected to a peristaltic pump (minipuls3, Gilsion, Middleton, Wisconsin, United States), was inserted into the left ventricle of the heart, and a small incision was made into the vena cava inferior for perfusion. The perfusion of the rat was performed with 30 ml of 1X phosphate-buffered saline (PBS) followed by 30 ml of 4% paraformaldehyde (PFA) at the rate of $7.8\mathrm{ml}{\min }^{-1}$. The rat was subsequently decapitated using a guillotine. The brain was dissected out and directly put into ice-cold 4% PFA. After fixation for 4 h at 4°C, the brain was transferred to 30% sucrose ($w/v$) for cryoprotection at 4°C. A vibratome (VT1200, LEICA, Wetzlar, Germany) was then used to section the brain into slices. To facilitate sectioning, the brain was embedded in 3% agarose gels prepared with a method similar to the one for preparing the 0.6% gels described above. The slicing was conducted in ice-cold 1X PBS. A similar process has been also followed in Ref. 58. The brain slice has a thickness of $400\mu \mathrm{m}$. During the PAM imaging, the slice is kept in 1X PBS at room temperature.

3.1.

Gas-Filled Hollow-Core Fiber Laser Source

Figure 2(a) presents the spectra of the pump laser as well as Raman Stokes lines, measured using an infrared spectrometer (Spectro320 Instrument Systems) with a resolution of 0.14 nm. The Gaussian-like beam profile of the first Raman pulse, measured by a beam profiler (BP109-IR2, Thorlabs), indicates that the Raman laser operates in the fundamental mode. To maximize the output pulse energy, first, we measured the pulse energy of the 1409-nm Raman line output of the first-stage ARHCF in terms of the $N_2$ pressure.⁵³ The Raman line appears at 9 bar pressure, and the pulse energy reaches its maximum of up to $\sim 26.5\mu \mathrm{J}$ at $\sim 15$-bar pressure, corresponding to a quantum efficiency of 45%.

Fig. 2

Characterization of the gas-filled ARHCF laser source. (a) Measured spectra including the pump line and Raman lines generated from the cascaded ARHCFs. The right axis shows the absorption spectrum of myelin extracted from.¹⁰ Inset: beam profile of first Raman pulse at 1409 nm. (b) Pulse energy evolution of the $3.4\mu \mathrm{m}$ Raman pulse as a function of ${\mathrm{H}}_2$ pressure. (c) Histograms of the pulse peak intensity of the $3.4\mu \mathrm{m}$ Raman pulse at 20 and 29 bar, respectively. (d) Pulse profile of the $3.4\text{-}\mu \mathrm{m}$ Raman laser. (e) Pulse peak intensity monitoring of the 1060-nm pump and the $3.4\text{-}\mu \mathrm{m}$ Raman lasers over 2 h.

By pumping the 1409-nm Raman line into the second-stage ARHCF, a Raman line at $3.4\text{-}\mu \mathrm{m}$ wavelength is observed when the pressure is $>5$ bar, as seen in Fig. 2(a), with a linewidth of $\sim 1\mathrm{nm}$. The average power of the $3.4\text{-}\mu \mathrm{m}$ Raman laser was measured by extracting it from the residual $1.4\text{-}\mu \mathrm{m}$ pump using a 2400-nm long-pass filter (FELH2400, Thorlabs, transmission 98% at $3.4\mu \mathrm{m}$). Figure 2(b) shows the pulse energy evolution of the $3.4\text{-}\mu \mathrm{m}$ Raman laser, as well as the total energy including both the Raman laser and residual pump at 1409 nm without using a filter, as a function of the ${\mathrm{H}}_2$ pressure. The Raman pulse energy begins to rise at $\sim 5$ bar, with the highest value being $2.75\mu \mathrm{J}$ at $\sim 20\text{bar}$. As the pressure further increases, the energy starts to decrease slightly, because the Raman laser reaches its maximal pulse energy at a shorter fiber length. The long-term stability and noise performance of the Raman laser source were also characterized to provide a reference for implementing the subsequent PAM application. The noise performance of the MIR Raman pulses was measured in terms of the relative intensity noise (RIN) of pulse peak intensity at 20 and 29 bar, using a photodetector (100-MHz bandwidth, PDAVJ8, Thorlabs) connected to an oscilloscope (6-GHz bandwidth, MSO64B, Tektronix, Beaverton, Oregon, United States). Figure 2(c) shows the distribution of the measured histograms. At 20-bar pressure, the distribution has a Gaussian-like profile with a RIN of 20.38%. Compared with 20 bar, at 29-bar pressure, although the pulse energy slightly decreases to $2.25\mu \mathrm{J}$, a lower RIN of 12.67% is obtained. This is because the SRS process becomes more efficient toward higher ${\mathrm{H}}_2$ pressure due to the further suppression of the transient SRS regime.⁵⁹ Given this property, here, we set the pressure of ${\mathrm{H}}_2$ to 29 bar, to better mitigate the acoustic signal fluctuation in the PAM application. Figure 2(d) shows a typical pulse profile of the $3.4\text{-}\mu \mathrm{m}$ Raman laser. The pulse width is $\sim 2\mathrm{ns}$, but the precision of this measurement is compromised because of the limited bandwidth of the MIR photodetector. This short nanosecond pulse gives a high peak power of $\sim 1.38\mathrm{kW}$. This Raman laser also shows good long-term stability, as indicated by the 2-h monitoring results of the laser peak intensities of both the $3.4\text{-}\mu \mathrm{m}$ Raman laser and its pump at 1060 nm. The average peak intensities and corresponding standard deviation of 2-min-long time bins over this time period were also calculated to better visualize the stability performance of the source and is presented in Fig. S2 in the Supplementary Material. The wavelength range of the MIR Raman laser can be tuned within a 10-nm range by thermally tuning the laser diode seed of the pump laser in the temperature range of 20°C to 35°C. The central wavelength of the pump source can be tuned from 1060 to 1061 nm, recorded from the spectrum by an infrared spectrometer (Spectro320 Instrument Systems) with a resolution of 0.14 nm, as shown in Fig. S3 in the Supplementary Material. The central wavelength of the generated first- and second-stage Raman lines is calculated based on the Raman shift coefficient of ${\mathrm{N}}_2$ and ${\mathrm{H}}_2$, respectively. As a result, it provides a tunable central wavelength range of $\sim 2\mathrm{nm}$ for the Raman line in the first stage and $\sim 10\mathrm{nm}$ for the MIR Raman line in the second stage from $\sim 3.391$ to $\sim 3.401\mu \mathrm{m}$. Employing a pump laser source with a larger wavelength tuning range will further expand the wavelength tunability.

3.2.

Characterization of the MIR-PAM

The lateral resolution of the PAM was first evaluated by imaging a sharp knife edge with a thickness of $100\mu \mathrm{m}$. Figure 3(a) shows the optical image (top) and the corresponding PAM image (bottom). The pulse energy before and after the objective lens are $\sim 1.9$ and $\sim 1\mu \mathrm{J}$, respectively, due to the 42.9% obscuration. The PAM images are obtained with a step size of $1.25\mu \mathrm{m}$ after averaging 100 pulses. The edge spread function (ESF) is acquired from the averaged raw data of the intensity, and the line spread function (LSF) is then calculated and fitted from the differential of the ESF, as shown in Fig. 3(b). From the full width at half maximum (FWHM) of LSF, the lateral resolution of our MIR-PAM is estimated to be $\sim 12.36\mu \mathrm{m}$.

Fig. 3

Resolution characterization of the MIR-PAM. (a) Optical and photoacoustic images of a sharp blade edge. (b) Fitted ESF and LSF extracted from PA image in panel (a). (c) Single-PA signal and its Hilbert transformation (HT) profile.

The axial resolution $R_a$ is estimated experimentally by the pulse width of the impulse response of a single PA signal. Figure 3(c) shows the PA signal and its envelope after Hilbert transformation (HT). The axial resolution is extracted from the FWHM of the envelope, which is 62 ns, corresponding to $95.5\mu \mathrm{m}$ in distance, given the propagation velocity of sound in the sample of $c=1540\mathrm{m}/\mathrm{s}$.⁶⁰ The axial resolution can also be numerically calculated by the velocity of the acoustic signal and the central frequency of the ultrasound transducer as $R_a=0.88c/B$, where $B$ is the central frequency of the UT. In our case, the theoretical axial resolution is $\sim 67.8\mu \mathrm{m}$.

3.3.

Ex Vivo PAM

Figure 4(a) shows the ex vivo PAM image of the mouse brain slice with a thickness of $400\mu \mathrm{m}$ obtained by using the proposed $3.4\text{-}\mu \mathrm{m}$ laser as a light source. The scanning step size is $40\mu \mathrm{m}$, and the total scanning time is $\sim 40\min$. HT was used for extracting the absolute amplitude of the acoustic signal, and then the image is post-processed by gamma transformation and a contrast enhancement of 0.5% saturation using the ImageJ software. In comparison with the optical image in Fig. 4(b), we can clearly see the outlines of the different brain regions. The myelin-rich regions, such as corpus callosum, are brighter than other regions. Small structures such as mammillothalamic fasciculus (mtf) and fornix (fo) can also be distinguished, exhibiting as bright spots, because they consist of heavily myelinated fibers.^10,61,62 These characteristics are invisible in the optical image shown in Fig. 4(b). The regions with less myelin, such as the hippocampus, which mainly consists of gray matter, appear to be darker. Finally, due to the lack of myelin, the cavity structure (lateral ventricle IIId) looks like a void with clear boundaries.

Fig. 4

Image of the mouse brain. (a) Ex vivo PA image of the mouse brain slice. Hb, habenular nuclei; IIId, third ventricle; mtf, mammillothalamic fasciculus; fo, fornix. (b) Optical image of the mouse brain slice. (c) Enlarged view of the structure in the yellow box in panel (a) with a scanning step size of $10\mu \mathrm{m}$. (d) Enlarged view of the structure in the red box in panel (a) with a scanning step size of $20\mu \mathrm{m}$. (e) Enlarged view of the structure in the blue box in panel (d) with a scanning step size of $5\mu \mathrm{m}$.

Enlarged views of the corpus callosum and thalamus regions were scanned with a step size of 10 and $20\mu \mathrm{m}$, respectively, as shown in Figs. 4(c) and 4(d). The border between the corpus callosum and the hippocampus is clearly visible in Fig. 4(c). In Fig. 4(d), patterns with a fish scale–like appearance are observable, which agree with other brain images reported in the literature.¹⁰ A further enlargement of the thalamus region, performed with a scanning step of $5\mu \mathrm{m}$, is shown in Fig. 4(e). Here, we can observe some thick nerve fiber bundles that have been also verified and observed by other MIR-PAMs,^10,13 demonstrating the ability of our new MIR-PAM Raman fiber laser for lipid-rich myelin imaging. All enlarged images required a scanning time of $\sim 20\min$.

To examine the imaging contrast and penetration depth of the system, another brain slice sample with a larger thickness of $600\mu \mathrm{m}$ in the mouse brain was imaged with a scanning step size of $50\mu \mathrm{m}$. Once again, the comparison between Figs. 5(a) and 5(b) shows that our system is able to highlight the higher density of myelinated neurons in regions such as the primary sensory cortex, corpus callosum, and caudate putamen. As shown in Fig. 5(c), the imaging contrast performance was evaluated from the unprocessed PA image by comparing the maximum PA signal amplitude from the water background (the pink-shaded region) and the neuron-rich areas (the yellow-shaded regions). The ratio of the signal amplitude from the water background to the highest signal amplitude from the neuron-rich areas is around 50%. More specifically, the SNR and contrast-to-noise ratio (CNR) can be calculated to be around 39:1 and 20:1 by the same procedures in Ref. 11, i.e.

Fig. 5

Image of the mouse brain. (a) Optical and (b) PA images of a $600\text{-}\mu \mathrm{m}$-thick mouse brain slice. The PA image is scanned with a step size of $50\mu \mathrm{m}$. (c) PA signal intensity profiles along the dashed lines in panel (b) showing the contrast level in the water background (the pink-shaded region) and brain sample (the yellow-shaded regions) and calculated signal amplitude ratio to the maximum PA signal amplitude. (d) Optical and (e) PA images of a $600\text{-}\mu \mathrm{m}$-thick mouse brain slice implanted with a fiber. The PA image is scanned with a step size of $10\mu \mathrm{m}$.