FIGURE 8 The 3D volumetric image of ex-vivo colorectal tissues reconstructed from: (a) raw data, (b) de-noised data by sqtwolog threshold method, (c) de-noised data by 4-order Butterworth low pass filter, (d) de-noised data by our proposed gaWD method.
5.3 | In-vivo experimental results
Except for the ex-vivo colorectal tissues imaging, we also perform in-vivo blood vessels imaging of rat. To achieve deep penetration and good resolution for blood vessels simultaneously, we choose laser of 790 nm wavelength. Before experiment, we scrape the fur off the surface of the rat’s skin at the target region, in order to reduce the scattering and attenuation of PA signal caused by the fur. The whole process of our experiment is strictly conducted under the standard of animal protection in animal research principles (Approving institution: ShanghaiTech University, IACUC protocol number: 20200323002). The photograph of the in vivo experimental setup is shown in FIGURE 9. To validate the algorithm’s effectiveness in low SNR scenario, we add 18 dB noise to our raw data. We then process it by the above 3 methods, respectively. The 3D volumetric images are shown inFIGURE 10 , respectively. First, we need to know there are two layers to be measured in this experiment. The skin surface of the rat generates large high-frequency PA signal, and the blood vessels about 1.3cm below the skin generates relatively low-frequency PA signal.FIGURE 10 (a) shows the 3D image of raw data with added 18dB noise, both two layers are significantly blurred and submerged in noise. As shown in FIGURE 10 (b), after sqtwolog denoising method, the background noise has been reduced a lot, but the image contrast is still low. As shown in FIGURE 10 (c), we apply 4-order Butterworth filter to the data, which added 18dB noise. We can easily find that the PA signals of the two layers are obviously enhanced, but the important details and the whole skin surface layer appear blurred. Among the three de-noising methods, our proposed gaWD method performs best. It not only reduces the background noise, but also enhances the image contrast perfectly, without loss of useful information. Additionally, both the surface layer above and the blood vessels layer below are clearly recovered from the noise corrupted data. These results show that our proposed gaWD method exhibits good ability to preserve the signal details without inducing distortion, while filtering out the noise, for both the surface with high-frequency signals and inside tissues with low-frequency signals.