在超音波B-mode成像中，來自組織散射子的回聲信號會隨機互相干擾並產生加乘性的散斑噪聲¬，這會降低從回聲信號振幅估計成像目標真實回聲性的準確性。另外，由於散斑圖案的顆粒度受到影像系統的點擴散函數所影響，B-mode影像的解析度會因此而下降，細微的結構邊界經常變得較為模糊而增加診斷的困難。雖然散斑現象抑制已經廣泛地在超音波後處理濾波技術中所研究，但經由這些方法處理後的影像仍然侷限於空間解析度而無法完整回復組織回聲圖。有鑑於深度學習在近幾年已經成功地應用在醫學超音波領域的不同研究上，本研究透過使用卷積神經網絡 (Convolutional neural network，簡稱CNN）來去除B-mode影像中的散斑噪聲並同時改善影像解析度來重建組織回聲圖。
本研究針對五種進行散斑抑制任務的神經網路超參數進行測試及挑選，並提出關於訓練數據點擴散函數尺寸的設置要求以及為了消除B-mode影像與重建回聲圖之間的振幅落差所做的正規化改進方式。研究結果顯示，本研究所提出的散斑抑制方法可以幾乎完整去除B-mode影像中的散斑噪聲並保留組織結構的輪廓及邊緣，使重建回聲圖的對比度及對比雜訊比從0.22/2.72提升至0.33/44.14，同時，影像的橫向及軸向解析度也分別從0.59/0.24改善至0.29/0.20。與其他後處理濾波技術相比，本研究所提出之方法可以透過完整去除散斑並改善影像解析度來很好地近似組織的原始回聲圖，並能實時地在超音波掃描儀上進行影像處理及呈現。

In ultrasound B-mode imaging, the echo signals from the tissue scatterers stochastically interfered with each other and produce multiplicative speckle noise, which decreases the accuracy of estimation of tissue echogenicity of imaging targets from the amplitude of the echo signals. In addition, since the granular size of the speckle pattern is affected by the point spread function (PSF) of the image system, the resolution of the B-mode image reduces accordingly, and the detailed boundaries of structures often become blurred. Speckle suppression has been extensively studied in the context of ultrasound post-processing filter techniques, however, the images processed by these methods are still limited by the degradation of spatial resolution and cannot fully restore the original tissue echogenicity maps. In view of the fact that deep learning technique has been successfully applied in the medical ultrasound for different tasks in recent years, this study proposed a convolutional neural network (CNN) to remove speckle noise together with improvement of image spatial resolution to reconstruct ultrasound tissue echogenicity map.
We tested and selected five neural network hyperparameters for the proposed speckle suppression task, and proposed the setting requirement of the point spread function size when generating training dataset and an improved normalization method to eliminate the amplitude gap between the simulated B-mode images and the reconstructed echogenicity maps. The results indicate that the proposed CNN method can almost completely eliminate the speckle noise in the background of the B-mode images and retain the contours and edges of the tissue structures. By applying the proposed CNN method, the contrast and the contrast-to-noise ratio of the reconstructed echogenicity map can be increased from 0.22/2.72 to 0.33/44.14, the lateral and axial resolutions can also be improved from 0.59/0.24 to 0.29/0.20, respectively. Compared with other post-processing filtering methods, the proposed method can well approximate the original tissue echogenicity map by completely removing speckle and improving the image resolution, and can process and display images on the ultrasonic scanner in real time.

[1] 沈哲州，「醫用超音波影像上課講義」，國立台灣科技大學電機所，民國107年。
[2] Jean Meunier, and Michel Bertrand, “Ultrasonic texture motion analysis: theory and simulation,” IEEE Trans. Med. Imag., vol. 14. no. 2, pp. 293-300, Jun. 1995.
[3] Pierrick Coupé, Pierre Hellier, Charles Kervrann, and Christian Barillot, “Nonlocal means-based speckle filtering for ultrasound images,” IEEE Trans. Image Process., vol. 18, no. 10, pp.2221-2229, Oct. 2009.
[4] Pierrick Coupé, Pierre Hellier, Charles Kervrann, and Christian Barillot, “Bayesian non local means-based speckle filtering,” 5th Proc IEEE Int Symp Biomed Imaging: From Nano to Macro, pp. 1291-1294, May, 2008.
[5] Yongjian Yu and Scott T. Acton, “Speckle Reducing Anisotropic Diffusion,” IEEE Trans. Image Process., vol. 11, no. 11, pp.1260-1270, Nov. 2002.
[6] Shen Liu, Jianguo Wei1, Bo Feng, Wenhuan Lu, Bruce Denby, Qiang Fang, and Jianwu Dang, “An Anisotropic Diffusion Filter for Reducing Speckle Noise of Ultrasound Images Based on Separability,” APSIPA ASC, California, USA, Dec., 2012.
[7] Pietro Perona and Jitendra Malik, “Scale-space and edge detection using anisotropic diffusion.” IEEE Trans. Pattern Anal. Mach. Intell. vol. 12, no. 7, pp. 629–639, July 1990.
[8] Yong Yue, Mihai M. Croitoru, Akhil Bidani, Joseph B. Zwischenberger, and John W. Clark, Jr., “Nonlinear Multiscale Wavelet Diffusion for Speckle Suppression and Edge Enhancement in Ultrasound Images,” IEEE Trans. Med. Imag., vol. 25, no. 3, pp.297-311, Mar. 2006.
[9] Simone Balocco, Carlo Gatta, Oriol Pujol, Josepa Mauri, and Petia Radeva, “SRBF: Speckle Reducing Bilateral Filtering,” Ultrasound in Med. & Biol., vol. 36, no. 8, pp. 1353-1363, 2010.
[10] C. Tomasi and R. Manduchi, “Bilateral Filtering for Gray and Color Images”. ICCV. Bombay, India, Jan., 1998.
[11] Baiying Lei, Shan Huang, Ran Li, Cheng Bian, Hang Li, Yi-Hong Chou, Jie-Zhi Cheng, “Segmentation of breast anatomy for automated whole breast ultrasound images with boundary regularized convolutional encoder–decoder network,” Neurocomputing, vol. 321, pp. 178–186, 2018.
[12] Deepak Mishra, Santanu Chaudhury, Mukul Sarkar, and Arvinder Singh Soin, “Ultrasound Image Segmentation: A Deeply Supervised Network With Attention to Boundaries,” IEEE Trans. Biomed. Eng., vol. 66, no. 6, pp. 1637-1648, June 2019.
[13] Tsung-Chen Chiang , Yao-Sian Huang, Rong-Tai Chen, Chiun-Sheng Huang, and Ruey-Feng Chang, “Tumor Detection in Automated Breast Ultrasound Using 3-D CNN and Prioritized Candidate Aggregation,” IEEE Trans. Med. Imag., vol. 38, no. 1, pp. 240-249, Jan. 2019.
[14] Renxin Zhuang, Junying Chen, “High Contrast and High Signal-to-noise Ratio Minimum Variance Beamforming Combined with Deep Neural Networks,” IUS, Glasgow, Scotland, Oct. 2019.
[15] Sanketh Vedula, Ortal Senouf, Grigoriy Zurakhov, Alex M. Bronstein, Michael Zibulevsky, Oleg Michailovich, Dan Adam, Diana Gaitini, “High quality ultrasonic multi-line transmission through deep learning,” [Online]. Available: https://arxiv.org/abs/1808.07819
[16] Zixia Zhou, Yuanyuan Wang, Jinhua Yu, Yi Guo, Wei Guo, and Yanxing Qi, “High Spatial–Temporal Resolution Reconstruction of Plane-Wave Ultrasound Images With a Multichannel Multiscale Convolutional Neural Network,” IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 65, no. 11, pp. 1983-1996, Nov. 2018.
[17] Kazuhiro Yamanaka and Chizue Ishihara, Hitachi, Ltd. Research & Development Group, Tokyo, Japan, “Recovery of ultrasound image with transmission beam lines decimation by using computationally low cost convolutional neural network,” IUS, Glasgow, Scotland, Oct. 2019.
[18] Maxime Gasse, Fabien Millioz, Emmanuel Roux, Damien Garcia, Hervé Liebgott, and Denis Friboulet, “High-Quality Plane Wave Compounding Using Convolutional Neural Networks,” IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 64, no. 10, pp. 1637-1639, Oct. 2017.
[19] Yeo Hun Yoon, Shujaat Khan, Jaeyoung Huh, and Jong Chul Ye, “Efficient B-mode Ultrasound Image Reconstruction from Sub-sampled RF Data using Deep Learning,” [Online]. Available: https://arxiv.org/abs/1712.06096
[20] Adam C. Luchies and Brett C. Byram, “Deep Neural Networks for Ultrasound Beamforming,” IEEE Trans. Med. Imag., vol. 37, no. 9, pp. 2010-2021, Sep. 2018.
[21] Walter Simson, Magdalini Paschali, Nassir Navab, Guillaume Zahnd, “Deep Learning Beamforming for Sub-Sampled Ultrasound Data,” IUS, Kobe, Japan, Oct. 2019.
[22] Dongwoon Hyun, Leandra L. Brickson, Kevin T. Looby, and Jeremy J. Dahl,” Beamforming and Speckle Reduction Using Neural Networks,” IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 66, no. 5, pp. 898 – 910, May 2019.
[23] Wonseok Jeon, Wooyoung Jeong, Kyungchan Son, Hyunseok Yang, “Speckle noise reduction for digital holographic images using multi-scale convolutional neural networks,” Optics Letters, vol. 43, no. 17, pp. 4240-4243, Sep. 2018.
[24] Hang Zhao, Orazio Gallo, Iuri Frosio, Jan Kautz, “Loss functions for image restoration with neural networks,” IEEE Trans. Comput. Imag., vol. 3, no. 1, pp. 47 – 57, Jan. 2017.