1. Introduction

Photoacoustic imaging is a promising technique that relies on optical excitation and ultrasonic detection [1–5]. A laser pulse illuminates tissue, where optically absorbing regions emit ultrasound via the thermoelastic effect. The detected ultrasound waves are used to reconstruct the location of the optically absorbing regions. Photoacoustic imaging combines the high contrast of optical methods with the excellent spatial resolution and large penetration depth of ultrasonic imaging. Although more commonly used for deep tissue imaging, photoacoustic imaging can also produce impressive spatial resolution of tissue microstructure at shallow depths [6,7]. This is the regime of photoacoustic microscopy (PAM).

Impressive in vivo PAM images of individual capillaries have been produced by optically focusing the excitation laser [8]. This approach, known as optical resolution PAM (OR-PAM), is restricted to shallow penetration depths (e.g. 0.5 mm) due to optical scattering in tissue [8]. Low pulse energies (e.g. less than 100 nJ) are sufficient for OR-PAM because the extremely small optical focus produces a high optical fluence. OR-PAM can therefore be used with high repetition rate pulsed lasers to significantly increase image acquisition speed.

Q-switched microchip lasers are high repetition rate lasers that are extremely compact and cost-effective. The solid-state gain medium (e.g. Nd:YAG), cavity mirrors, and saturable absorber form a monolithic laser cavity. Diode pumping and passive Q-switching result in nanosecond laser pulses with several μJ of energy at a repetition rate of several kHz [9]. However, the fixed wavelength output makes microchip lasers unsuitable for spectroscopic photoacoustic imaging [10–12]. We are developing a tunable optical source for spectroscopic PAM, where microchip laser pulses are propagated through several meters of photonic crystal fiber to generate an ultrabroadband spectrum [13,14]. A tunable band pass filter selects the desired wavelength. This spectral filtering approach produces low pulse energies that are inadequate for deep tissue imaging but sufficient for OR-PAM. Furthermore, the wide spectral coverage and potentially rapid wavelength tuning of our source can significantly benefit OR-PAM applications. We demonstrate spectroscopic PAM by using seven different wavelength bands to successfully differentiate ink phantoms with overlapping absorption spectra. To our knowledge, this is the first demonstration of spectroscopic PAM using a microchip laser-based supercontinuum source.

2. Method

2.1 Supercontinuum source

Our arrangement for the supercontinuum source is depicted in Fig. 1(a) . The Q-switched Nd:YAG microchip laser (NP10820-100, Teem Photonics) produces 0.6 ns duration pulses at 1064 nm with 8 μJ of energy at a 6.6 kHz repetition rate. An aspherical lens (NA = 0.4) couples the 1064 nm pulses into a 7 meter long photonic crystal fiber (PCF), where the zero dispersion wavelength is 1040 nm (Crystal Fibre, Inc.). The air-silica honeycomb-like microstructure of a photonic crystal fiber (PCF) provides favorable dispersion properties to dramatically enhance nonlinear optical propagation. The seed laser wavelength must be longer than the zero dispersion wavelength in order to maximize supercontinuum generation [13]. The input average power is 51 mW and the output average power in the supercontinuum is 2.8 mW. Improved coupling efficiency and a longer PCF (e.g. 20 meters) can significantly improve the overall conversion efficiency from the seed wavelength into the supercontinuum. However, availability and cost considerations restricted our PCF to only 7 meters. The supercontinuum pulses are collimated by a 20x microscope objective lens to a 3 mm beam diameter. Figure 1(b) shows the spectrum measured by an optical spectrum analyzer with a useful scanning range from 600 to 1700 nm. The measured spectrum is clearly a continuum that extends out to 1400 nm. The large spike at 1064 nm corresponds to residual power remaining in the pump wavelength. Although not shown in Fig. 1, the supercontinuum does contain wavelengths near 500 nm, which is confirmed by viewing the dispersed light on a piece of paper.

 

Fig. 1 (a) Schematic of the photonic crystal fiber (PCF) supercontinuum source. The plano-convex lens (PCX) collimates the microchip laser output. An aspheric lens (AL) focuses the 1064 nm pulses into the PCF. A microscope objective (MO) collimates the supercontinuum (SC) output. (b) Supercontinuum spectrum measured with an optical spectrum analyzer. (c) Prism-based monochromator, where a concave mirror collimates the dispersed light. The masked mirror allows manual selection of the desired wavelength and bandwidth [15].

Download Full Size | PDF

2.2 Tunable bandpass filter

For multispectral photoacoustic microscopy, the supercontinuum pulses are sent through a tunable band pass filter shown in Fig. 1(c). The dispersed light from an equilateral BK-7 prism is collimated by a 75 mm diameter spherical mirror with a focal length of 500 mm. Spectral filtering is performed in the Fourier plane with a slit placed in front of a flat mirror [15]. The position and width of the slit determine the wavelength and bandwidth of the filtered light. The flat mirror has a slight vertical tilt to allow the return beam to be separated from the incident beam. The photoacoustic microscopy experiments in this paper used seven wavelengths from 575 to 875 nm in 50 nm increments. Each wavelength band has a bandwidth of 40 nm. The pulse energy for each wavelength is measured to be 7, 15, 24, 31, 31, 31, and 33 nJ, in order of increasing wavelength. These low pulse energies restrict the tunable source to optically focused PAM. Higher pulse energies at shorter wavelengths are possible with a longer photonic crystal fiber (i.e. 20 meters), but a shorter fiber was used for cost considerations. Although not measured directly, the pulse duration for each wavelength is not expected to be significantly longer than the input pulse [13].

2.3 Photoacoustic microscopy system

The photoacoustic microscopy system diagram is shown in Fig. 2(a) . The “photoacoustic beamsplitter” transmits the optical excitation and reflects the photoacoustic signal. The excitation laser pulse is focused with a 4x infinity corrected microscope objective (NA = 0.1). The beamsplitter consists of a plane glass wafer surrounded on both sides by water. The upper water section is in contact with a glass plano-convex lens with a radius of curvature chosen to

 

Fig. 2 (a) Schematic of the photoacoustic microscopy system. The photoacoustic signal is reflected by the glass plate and detected by a 25 MHz transducer. (b) En face image of a resolution target shown over a 20 dB scale and 600 x 600 μm field of view. The bar patterns correspond to Elements 5 and 6 within Group 3 of the USAF resolution target. (c) Profile of the Element 6 bar pattern shown over a 20 dB scale. The dashed curve corresponds to a simulated profile assuming a point spread function with a −6 dB width of 18 μm.

Download Full Size | PDF

reduce aberrations at the air-glass interface. The lower water section is in contact with a 25 MHz spherically focused f/2 transducer (Olympus V324). A 25 micron thick Mylar membrane seals the photoacoustic beamsplitter from the object of interest. The transducer signal is sent through a 60 dB amplifier (Miteq) and a 50 MHz low pass filter (Mini-Circuits) before acquisition with a digital oscilloscope operating at 1 GS/sec (LeCroy). A photodiode monitoring the microchip laser output provides the scope trigger. Fluctuations in photoacoustic signal amplitude for a particular wavelength were not corrected, although this could be readily achieved with a monitor photodiode. Two stepper motors (Zaber) perform two-dimensional lateral scanning of the object. Data acquisition is performed with LabVIEW, and data processing and reconstruction is performed in MATLAB.

Figure 2(b) shows an en-face photoacoustic image of a chrome-on-glass resolution target taken at a wavelength of 675 nm. The maximum amplitude projection (MAP) image is displayed over a 20 dB logarithmic scale. The bar patterns correspond to Element 5 (bar width = 39.4 μm) and Element 6 (bar width = 35.0 μm) within Group 3 of the United States Air Force (USAF) target. A profile of the image of Element 6 is shown in Fig. 2(c). The dashed curve is a simulated profile assuming a Gaussian-shaped point spread function with a full width half maximum of 18 μm. Improved resolution should be achieved by (1) expanding the excitation laser to completely fill the focusing objective lens and (2) minimizing aberrations by computer ray tracing. Nevertheless, the PAM resolution is considerably smaller than the acoustic wavelength (60 μm in water) and sufficient for high resolution imaging.

2.4 Multiwavelength image processing

As a proof-of-concept demonstration of spectroscopic PAM with our tunable source, spectral data was processed with a simple discriminant analysis approach in MATLAB [16,17]. A classifier is trained using a small portion of the multispectral images with known absorbers. In addition to the red, blue, green, and black categories, a fifth group of training pixels was taken from the image background. Each of the five groups contained 16x16 = 256 pixels in seven wavelength bands. After training, the classifier proceeds to process the entire set of images.

3. Results

Initial demonstration of spectroscopic PAM with the supercontinuum source used a phantom consisting of black, blue, green, and red ink spots deposited on an acrylic block. Each spot is approximately 1 mm in diameter. Photoacoustic microscopy images were produced by mapping the maximum amplitude from each recorded signal to an image pixel. Figure 3(a) shows the maximum amplitude projection (MAP) images for all seven wavelengths. All images span a 1.8 x 5.4 mm region and are shown over the same 40 dB dynamic range. The images were interpolated to four times the original number of pixels to produce a smoother appearance. The different colored spots clearly exhibit different wavelength behavior. The bottom right image of Fig. 3(a) shows the result of our simplified discriminant analysis approach to process the multiwavelength images. The classified groups are displayed as a red-green-blue (RGB) image, where the black ink spot is displayed as white. The classification has some error, such as in the blue and green ink regions. Higher accuracy should be achievable with more sophisticated multispectral image analysis and a larger number of wavelength bands. Nevertheless, these are very encouraging results considering the simplicity of our classification approach. Figure 3(b) shows spectroscopic PAM data obtained by averaging the pixel amplitude over a 0.22 x 0.22 mm square region inside each spot in the individual wavelength images of Fig. 3(a). The solid curves in Fig. 3(b) represent the optical absorbance spectra derived from spectrophotometry. The close agreement validates the spectroscopic data acquired with our tunable source.

 

Fig. 3 (a) Multiwavelength images of black, blue, green, and red ink spots (left to right). All images are displayed over a 1.8 x 5.4 mm area and the same 40 dB scale. The spectrally processed image clearly identifies the four ink regions. (b) Multispectral photoacoustic data of black (triangles), blue (diamonds), green (circles), and red (squares) ink regions. For comparison, the solid curves are absorbance measurements by spectrophotometry.

Download Full Size | PDF

A more realistic multiwavelength PAM experiment was performed on a scattering phantom consisting of cotton-swab fibers stained with the same inks as Fig. 3. The fibers are located beneath a 0.4 mm scattering layer consisting of an aqueous suspension of 1 μm diameter polystyrene microspheres at a 0.09% concentration. The extinction coefficient is 1.9 mm⁻¹ at 675 nm, as measured by spectrophotometry. This simple phantom is intended to mimic the morphology of microvasculature. Images were acquired in the same seven wavelength bands (575 – 875 nm). Only the 575 and 675 nm images are shown in Fig. 4(a) and (b) . All images are shown over a 0.6 x 0.6 mm region and a 20 dB dynamic range. The disappearance of the two parallel fibers in Fig. 4(a) suggests these threads are stained with red ink. This is confirmed with discriminant analysis using the same training data from Fig. 3. The remaining threads are identified as stained with green ink. An RGB display of the classified image is shown in Fig. 4(c). For validation, a photograph of the phantom (without the scattering solution) is shown in Fig. 4(d). The good agreement between Fig. 4(c) and Fig. 4(d) confirms our multiwavelength PAM system can distinguish different absorbers through a scattering medium. Some classification error is evident in the PAM image, where some of the blue fiber contains green highlights. A more robust spectral processing algorithm, a greater number of wavelengths, and compensation of pulse-to-pulse fluctuations should reduce this error.

 

Fig. 4 PAM images through a scattering solution at (a) 575 nm and (b) 675 nm of a cotton fiber phantom stained with the same inks as Fig. 3. The scale bar represents 150 μm. (c) Spectrally processed image after discriminant analysis. All images are shown over a 0.6 x 0.6 mm field of view and a 20 dB scale. (d) Photograph of the fiber phantom without scattering solution with appropriate labels for the colored fibers.

Download Full Size | PDF

4. Discussion and conclusions

Major advantages of our supercontinuum source for spectroscopic PAM are (1) high repetition rate (several kHz) (2) extremely broad range of accessible wavelengths (3) compact footprint. The high repetition rate of several kHz can significantly reduce data acquisition time. A suitable wavelength filter (i.e. acousto-optic tunable filter) can provide continuous tuning or rapid switching between arbitrary wavelengths. Near-infrared excitation in the 900 – 1300 nm wavelength range is also possible with this system, which has been shown to be useful for imaging lipid-rich regions of tissue [11,12].

As mentioned previously, the low pulse energy of our source limits its application to optically focused PAM. An order of magnitude increase in pulse energy should be possible with a higher energy microchip laser (e.g. 20 μJ/pulse) combined with a longer photonic crystal fiber (e.g. 20 meters) and improved fiber coupling. The results presented in this work used wavelength bands with a 40 nm bandwidth. Although a narrower bandwidth is desirable, a 40 nm bandwidth is still useful for many spectroscopic applications. For example, CO-oximetry systems typically employ multiple LEDs with spectral bandwidths on the order of 20 to 40 nm [18,19]. Narrowing the bandwidth in our tunable source comes at the expense of lower excitation pulse energy. A higher energy laser and longer photonic crystal fiber should provide over 50 nJ per pulse within a 10 nm bandwidth.

We have demonstrated spectroscopic photoacoustic microscopy with a supercontinuum source based on a photonic crystal fiber pumped by a microchip laser. A tunable bandpass filter provides access to any desired wavelength band. Future work involves developing a more rapidly tunable filter to quickly select arbitrarily separated wavelengths (e.g. over ten wavelengths per second) as well as testing the system on scattering phantoms with more realistic chromophores and contrast agents (i.e. methylene blue, gold nanorods). The high repetition rate of this system permits very rapid tuning of the optical excitation wavelength, making it possible to perform high-speed spectroscopic photoacoustic microscopy.