CC BY
37IMAGING PROCESSING OF LASER SPECKLE CONTRAST IMAGING OF BLOOD FLOW
WEIMIN CHENG12, XIAOHU LIU12, JINLING LUHAND PENGCHENG LI12
1 Britton Chance Center for Biomedical Photonics, Wuhan National Laboratory for Optoelectronics, Huazhong University ofScience
and Technology
2 MoE Key Laboratory for Biomedical Photonics, School of Engineering Sciences, Huazhong University of Science and Technology
Abstract
Laser speckle contrast imaging (LSCI) is a wide-field, noninvasive, and noncontact optical imaging technology for mapping blood flow. Given the advantage of high spatio-temporal resolution, LSCI is widely used in blood flow imaging of the skin, retina, splanchnic organs, tumor, and brain in recent years. In practical applications, the spatial and temporal window size of speckle contrast analysis is usually expected to be minimized for higher spatio-temporal resolution. However, a reduced spatio-temporal window size of LSCI results in significant noise of K2 owing to the statistical uncertainty [1]. To improve the measurement accuracy, a suitable denoising algorithm is required to enhance the signal-to-noise ratio (SNR) of LSCI. Furthermore, LSCI is well known to be highly sensitive to the motions induced by both environment and biological tissue itself. These disturbances will cause displacements of the speckle images, resulting in the error of speckle contrast estimation based on multiple frames of speckle images. Therefore, it is of urgent importance to minimize the impact of motion when LSCI is put into practical usage. We proposed a Manhattan distance based adaptive BM3D (MD-ABM3D) method to manage the complicated inhomogeneous noise in tLSCI image and improve the signal-to-noise ratio [2]. Manhattan distance improves the accuracy of the block-matching in strong noise, and the adaptive algorithm adapts to the inhomogeneous noise and estimates suitable parameters for improved denoising. As shown in Figurel, the image-quality evaluation of MD-ABM3D for tLSCI (t = 20 frames) equals that of savg-tLSCI (t = 60 frames).It achieves high signal-to-noise ratio with a reduced number of sampling frames.
45
40
al 35
CO 0-
30
25
sr..*-
o
in in
?/
/>■'-■■© - savg-tLSCI if -4--NLM / ■■«■ BM3D .........MD-ABM3D
0.95 0.9 0.85 0.8 0.75
»■i'-"' 0.995 .<•>"• 0.99
i r
a
a. 0.985
■■©■■■savg-tLSCI 098 ■■+■■■ NLM
■■« BM3D 0.975 ........MDABM3D
---'- 0.97
| ■■■••»savg-tLSCI j ...4....NLM ; ■■■•■■ BM3D .........MDABM3ID
10 30 50 70 10 30 50 70 10 30 50 70
No. of frames No. of frames No. of frames
(a) (b) (c)
Figurel: Image-quality evaluation of each denoising method for tLSCI with temporal windows (t = 10-80 frames). (a)-(c) are
evaluation of PSNR, MSSIM, and R, respectively[2]
However, the processing time of MD-ABM3D makes it difficult to realize real-time denoising. Furthermore, it is still difficult to obtain an acceptable level of SNR with a few raw speckle images given the presence of significant noise and artifacts. We proposed to train a feed-forward denoising convolutional neural network (DnCNN) for LSCI in a log-transformed domain to improve training accuracy and it achieves an improvement of 5.13 dB in the peak signal-to-noise ratio (PSNR). To decrease the inference time and improve denoising performance, we further proposed a dilated deep residual learning network with skip connections (DRSNet)[3]. The image-quality evaluations of DRSNet with five raw speckle images outperform that of spatially average denoising with 20 raw speckle images. DRSNet takes 35 ms (i.e., 28 frames per second) for denoising a blood flow image with 486 x 648 pixels on an NVIDIA 1070 GPU.
References
[1] J. Hong, L. Shi, X. Zhu, J. Lu and P. Li, Laser speckle auto-inverse covariance imaging for mean-invariant estimation of blood flow, Opt. Let., 44(23):5812-5815, 2019
[2] W. Cheng, J. Lu, X. Zhu, J., X. Liu, M. Li, P. Li, Dilated residual learning with skip connections for real-time denoising of laser speckle imaging of blood flow in a log-transformed domain, IEEE Trans. Med. Imag., 39(5): 1582-1593, 2019
[3] W. Cheng, X. Zhu, X. Chen, M. Li, J. Lu, P. Li, Manhattan distance based adaptive 3D transform-domain collaborative filtering for laser speckle imaging of blood flow, IEEE Trans. Med. Imag., 38(7):1726-1735, 2019
