1. Introduction

Scattering from turbid materials limits the depth through which images can be obtained. However, due to its deterministic nature, the scattering can be compensated with the help of a feedback mechanism using wavefront shaping. As a result, a focused spot can be created behind the scattering medium [1]. The initial techniques have been limited by the inability to provide a feedback without access to the back side of the scatterer [1–3]. Lately, new feedback mechanisms have been proposed such as iterative optimization feedback from fluorescence [4,5] or digital OPC with second harmonic generation nanoparticles [6] used as guide stars. These techniques are limited by a scarcity constraint that requires that the feedback signal only comes from a single particle to ensure single focus creation. However, this constraint can be overcome with iterative focusing methods by using a nonlinear feedback, such as two photon fluorescence [7,8], which itself is limited by optical power. Another promising method for imaging into scattering materials, especially biological tissue, uses a guide star created with an ultrasound focus [9]. Ultrasonic waves propagate through soft tissue with three orders of magnitude less scattering than optical waves, allowing them to penetrate much deeper with minimal scattering. The ultrasound focus guide star localy modulates the frequency of light crossing it. These tagged photons are then used to record the scattered optical field, which when phase conjugated, delivers photons back to the ultrasound focus [9]. Later, similar techniques were improved to allow for a reduction of the optical focus spot size [10,11].

A less explored feedback mechanism for focusing light through scattering materials is the photoacoustic effect. The photoacoustic effect produces acoustic waves as a medium absorbs light and undergoes thermal expansion [12]. The photoacoustic effect is used in modern photoacoustic microscopy to image at depth in tissue [13]. Photoacoustic microsocopy differs from ultrasound imaging in that its contrast stems from optical absorption, as opposed to mechanical properties. Photoacoustics allows, for example, imaging of the vasculature by using hemoglobin in blood as the absorbing medium [14]. Recently, the first demonstration of focusing through turbid media using photoacoustic feedback by Kong et al. demonstrated the capability of this technique, however no attempts at image reconstruction were made [15]. Very recently, photoacoustic feedback was used in measuring the transmission matrix through a scattering material onto light absorbing fibers [16]. In this transmission matrix measurement the input optical modes were related to the absorbers found behind the scattering material. As a result, it was possible to localize particles along the axis of the transducer and create optical foci at the absorbers detected in the matrix. Unfortunately, none of these two early techniques has demonstrated imaging capability yet.

Therefore, in this work we use a photoacoustic feedback optimized optical focus for creating an image of an absorbing medium behind a scatterer. Analyzing the temporal profile of the photoacoustic wave, we form a three-dimensional (3D) image after a two dimensional scan. Remarkably, we achieve a significant improvement in signal-to-noise ratio of about one order of magnitude.

The paper is organized as follows: In the next section we describe the experimental apparatus. We follow with a discussion of the data collection and reconstruction methods. Finally, we present the experimental results showing photoacoustic signal enhancement and three-dimensional images.

2. High contrast photoacoustic imaging system

The high contrast photoacoustic imaging system combines wavefront shaping, using a liquid crystal spatial light modulator (SLM), and an ultrasound transducer for measuring the photoacoustic signal. The SLM phase encodes the wavefront to maximize the intensity of light, while the transducer provides feedback for the iterative optimization algorithm [Fig. 1.]. More specifically, an attenuated, expanded, and collimated 5 ns laser pulse (Continuum Surelight I20, 20 Hz repetition rate, Nd:YAG frequency doubled to produce 532 nm wavelength) illuminates the entire screen of a phase-only liquid crystal spatial light modulator (SLM) (Boulder Non-linear Systems, 512x512 pixels). After the SLM the energy per pulse in the beam is ~21 μJ. In practice the pulse energy is limited by the damage threshold of the SLM and (more importantly) the sample. A 4f system (f₁ = 150mm, f₂ = 250mm) images the SLM onto the back aperture of a long working distance microscope objective (Mitutoyo, 34 mm working distance, 5x magnification, 0.14 NA). The beam focuses into a water tank and onto the surface of the scattering material (glass diffuser, Edmund optics, 120 grit). An absorbing sample used for wavefront optimization and imaging is placed ~8 mm behind the scattering material and mounted to a 2D translation stage to allow for scanning in the x and y dimensions. The distance between the sample and the glass diffuser is chosen to approximately match the size of one speckle grain with the size of the photoacoustic focal region. The photoacoustic signal produced by the sample propagates through the water and is detected by a 90 MHz transducer (Olympus, model V3512, 50Mhz bandwidth). The location of the transducer is chosen based on the geometry of the system. However, it could be placed on the same side as the illumination beam without loss of generality. After being pre-amplified (Femto HSA-X-2-40, low-noise 40dB) an oscilloscope records and digitizes the signal and sends it to a computer for analysis. In order to limit the size of the acoustic volume during the optimization process the signal is digitally high-pass filtered using a 2nd order Butterworth filter with a cut-off frequency of 80 MHz .

 

Fig. 1 Experimental setup for localized fluence-enhanced 3D photoacoustic imaging. The sample is placed in a glass water tank. A 532 nm pulsed laser illuminates an SLM that controls the input wavefront focused through the glass diffuser. A 90 MHz ultrasound transducer detects the photoacoustic signal from a sample placed behind the scattering material. The signal is amplified and digitized before being analyzed in the PC. The coordinate axis defines the axis in the following Figs. SLM: spatial light modulator; f1, f2: lenses; MO: microscope objective; GD: glass diffuser; S: sample.

Download Full Size | PDF

The system maximizes the photoacoustic signal by modulating the wavefront with the SLM and using an iterative algorithm for optimization. We use a genetic algorithm (GA) for the optimization because it has been demonstrated to work well with low SNR signals [17], which is sometimes the case with photoacoustics. The GA cost function for the algorithm is the peak-to-peak voltage (proportional to pressure) of the acoustic signal. While in general this signal depends on the absorber size, in our case the absorber is much larger than the acoustic beam diameter so it does not affect the quality of the optimization process. Furthermore, during the optimization process, the absorber does not change, making the cost function depend only on the amount of light absorbed by the sample.

Using the feedback from the photoacoustic signal the GA finds a phase mask which maximizes the cost function and consequently the light intensity within the acoustic focus volume. After the optimization process the best phase mask is projected on the SLM to prepare for image scanning.

3. Photoacoustic enhancement and 3D imaging

As a first demonstration of the performance of the system, we test its ability to increase the photoacoustic signal produced inside a polypropylene tube (90 µm inner and 120 µm outer diameter) filled with India ink and placed behind a glass diffuser. The GA optimization runs with a population size of 20 and 804 input modes through 1200 phase mask measurements, or 60 generations, to find the optimal phase mask. As in [17] the mutation rate decreased as the optimization progressed. Figure 2(a) illustrates how the enhancement evolved as the optimization proceeded. Here we define the enhancement as the value of the cost function of the projected mask divided by the mean of the cost function from each member of the initial population. An amplitude enhancement of 10 of the photoacoustic signal is observed, indicative of a 10-fold increase in absorbed light in the focal region. In Figs. 2(b) and 2(c) the photoacoustic response for a flat phase and the optimized phase mask are compared. To minimize the noise of the photoacoustic signal we initially average 40 samples of the signal. As the signal strength increases the number of averaged samples decreases gradually to 5 in order to decrease the optimization time. As a result the signals in Figs. 2(b) and 2(c) are taken with 40 and 5 averaged samples, respectively. The optimization process takes about 15 minutes, limited by the repetition rate of the laser. These figures show clearly the improvement in the photoacoustic signal produced using the GA optimization approach.

 

Fig. 2 (a) Experimental photoacoustic enhancement evolution of the optimization process. (b) Photoacoustic amplitude signal for a flat input phase mask (40 averaged samples). (c) Photoacoustic amplitude signal for the optimal input wavefront (5 averaged samples).

Download Full Size | PDF

After verification of photoacoustic signal enhancement, we demonstrate 3D imaging with enhanced localized optical fluence. As a first demonstration we scan the sample around the optimized focus with the automated translation stage while keeping the scattering material at a fixed location. The photoacoustic signal amplitude, recorded from each position [x-y plane, Fig. 1], is processed to reconstruct the 3D maximum intensity projection of the two tubes [Fig. 3(a)]. The temporal profile of the photoacoustic signal [Fig. 2(c)] encodes the z (axial) information. By sliding a window through the signal and selecting the maximum value for each window position many z values can be fixed to each x,y position to create the third dimension. The size of the window is determined by the axial resolution, δz, of the transducer, which comes from its bandwidth, B, and the speed of sound, c_s: $\delta z=c_s/B\cong 30\mu m.$ The transducer also determines the transverse resolution of the acoustic beam: $BD(-6dB)=1.02\cdot F{c}_s/Df\cong 36\mu m,$where BD is the acoustic beam diameter at the focus plane, F is the focal distance, D is the diameter of the transducer, and f is the central frequency. In these experiments we approximately match the size of the speckle with the acoustic BD during optimization. Hence, the resolution of our system is given by the speckle size at the transducer focal region.

 

Fig. 3 3D imaging of two capillaries. First row corresponds to the optimized phase mask. Second row corresponds to a flat phase projected on the SLM. (a),(d) 3D scan with a 2% intensity threshold; (b), (e) 2D scan; and (c), (f) 1D scan of the capillary tubes.

Download Full Size | PDF

A 2D slice is extracted from the 3D image to measure the distance between the two tubes. In Fig. 3(b) the normalized 2D image corresponding to an intermediate plane shows the maximum photoacoustic signal from both tubes are separated by 170 µm. From the size of the outer diameter we infer the tubes are 50 µm apart. A normalized 1D scan from the x = 0.02 mm profile is shown in Fig. 3(c). We observe the inner diameter is 50µm, significantly smaller than 90µm diameter given by the manufacture. This can be understood by considering that the photoacoustic feedback came from ink inside of the polypropylene tube. During the optimization, the wavefront correction accounted for the refraction of light produced by the polypropylene tube. In other words, the tube acted as a lens because of the higher index of refraction (1.49) of the polypropylene tubing material. As the optical focus moved away from the center of the tube during the image scan procedure, the wavefront no longer matched the curvature of the tube and the focus inside the tube was destroyed, thus producing negligible signal at the tube edges. Despite this, the tubes are clearly defined with high absorption contrast. For comparison, the 3D, 2D and 1D reconstruction with an unoptimized wavefront projected on the SLM are shown in Figs. 3(d)-3(f) respectively. In this case, the width of the two tubes and their separation agree with the actual tube size, although the SNR is more than 10 times lower than the optimized case. Figure 3(c) also compares the photoacoustic intensities (proportional to acoustic pressure squared) of the optimized and unoptimized scans.

4. Discussion and conclusion

We have demonstrated a one order of magnitude enhancement of the photoacoustic signal amplitude by GA optimization of the phase of the input wavefront to compensate for scattering and increase the optical intensity of light in the acoustic focus. This enhancement allowed for the imaging of two 90 µm inner diameter tubes with excellent signal-to-noise ratio as compared to a flat phase wavefront. Furthermore, by using the time of arrival information from the photoacoustic signal, the depth information was recovered and a 3D image reconstructed after scanning the sample in two dimensions.

In this experimental system we scan the sample around the focus because of the speed limitations imposed by the low repetition rate of the laser. With a higher repetition rate laser source and a faster wavefront modulation device [18], sub-second optimization of the input phase mask could potentially be enabled. This would open up the possibility of scanning the transducer instead of the sample and re-optimizing for each position, providing opportunities for in-vivo testing of biologically relevant samples and to increase the penetration depth of current photoacoustic microscopy techniques.

In conclusion, we have shown for the first time photoacoustic images created by locally enhancing fluence using wavefront shaping and hence the image signal-to-noise ratio. Furthermore, this is the first demonstration of imaging in three dimensions through a highly scattering medium (without using the memory effect [5,19]) and without need to access the back side of the scatterer where the object is located.