We
set up a photoacoustic microscopy (PAM) system to validate the
effectiveness of our proposed de-noising algorithm. The schematic is
shown in FIGURE 6. In this experimental setup, we focus the laser into a
fiber by an objective lens (OL), and then use two cascaded lens to focus
the output light of the fiber onto the sample. The received PA signal
from ultrasound transducer (UT) is first fed into a low-noise amplifier
(AMP) to be amplified with low noise induction, then transferred to data
acquisition card for digitization, which is connected to a computer for
real-time storage and display. The X/Y two-dimensional step motor is
used for raster scanning. The data sampling rate is set to 80 MHz, and
the transducer’s central frequency is 10 MHz.
FIGURE 6 The schematic
of experimental system. FG: function generator; OL: objective lens; MMF:
multimode fiber; CL: collimating lens; US: ultrasound; WT: water tank;
AMP: amplifier; DAQ: data acquisition card; X/Y Motor: two-dimensional
step motor.
5.2 | Ex-vivo
experimental results
We conduct imaging ofex-vivo colorectal tissues first. We tried sqtwolog threshold
method, 4-order Butterworth low pass filter, and our proposed gaWD
method, respectively. We first
draw the maximum amplitude projection (MAP) image after applying each
de-noising method, shown in FIGURE 7 . As shown inFIGURE 7 (a), the MAP image reconstructed from raw data is
severely corrupted by both random noise and stripe noise. It could be
found that sqtwolog threshold method filters out some random noise, but
is unable to reduce the stripe noise, shown in FIGURE 7 (b). As
for 4-order Butterworth low pass filtering result in FIGURE
7 (c), it removes the stripe noise at the background, and enhances the
image contrast. However, it blurs the image and loses some details,
e.g., some texture of tissues is disappeared.
Compared to these two methods,
our proposed method achieves both de-noising and enhancing the image
contrast, meanwhile preserving the details, presented in FIGURE
7 (d). The corresponding SNR of all these images are calculated and
shown in FIGURE 7 (e). The de-noised image of our proposed
method has the highest SNR, which validates the effectiveness of the
proposed algorithm.