Three-dimensional optical coherence tomography image denoising through multi-input fully-convolutional networks

Abstract

In recent years, there has been a growing interest in applying convolutional neural networks (CNNs) to low-level vision tasks such as denoising and super-resolution. Due to the coherent nature of the image formation process, optical coherence tomography (OCT) images are inevitably affected by noise. This paper proposes a new method named the multi-input fully-convolutional networks (MIFCN) for denoising of OCT images. In contrast to recently proposed natural image denoising CNNs, the proposed architecture allows the exploitation of high degrees of correlation and complementary information among neighboring OCT images through pixel by pixel fusion of multiple FCNs. The parameters of the proposed multi-input architecture are learned by considering the consistency between the overall output and the contribution of each input image. The proposed MIFCN method is compared with the state-of-the-art denoising methods adopted on OCT images of normal and age-related macular degeneration eyes in a quantitative and qualitative manner.Index Terms-Fully convolutional network (FCN), Multi-input FCN, Image denoising, Optical Coherence Tomography (OCT).

Three-dimensional curvelet-based dictionary learning for speckle noise removal of optical coherence tomography

“…Iterative maximum a posteriori (MAP)-based algorithm [33] Low-pass filtering [34] Median filter [6][7][8] Adaptive Median &Wiener filter [9][10][11] Mean filter [35,36] Two 1D filters [37] Advanced I-divergence regularization approach [38] methods Non-linear anisotropic filter [39,40] Complex diffusion [41,42] Directional filtering [43,44] Adaptive vector-valued kernel function [45] SVM approach [46] Adaptive-weighted bilateral filter (AWBF) [47] Bayesian estimations [48] Deep convolutional neural network [20][21][22] Sparse representation…”

“…The traditional speckle filtering methods in raw image domain such as median and Lee filtering [6][7][8][9], adaptive median and Wiener filtering [10,11] provide inadequate noise reduction under high-level speckle noise, as well as cause loss of meaningful subtle features. In the recent years, numerous more advanced methods have been proposed for speckle noise reduction, such as anisotropic diffusion-based techniques [12][13][14], wavelet-based methods [15], denoising using dual-tree complex wavelet transform [16] and curvelet transform [17], sparsity-based denoising [18,19], complex wavelet-based K-SVD dictionary learning technique (CWDL) [5], deep convolutional neural network based methods [20][21][22] and robust principal component analysis (RPCA)-based method [23].…”

Towards Label-Free 3D Segmentation of Optical Coherence Tomography Images of the Optic Nerve Head Using Deep Learning

“…The 390 volumes were first resized (in pixels) to 448 (height) x 352 (width) x 96 (number of B-scans), and a total of 37,440 baseline B-scans (12,480 per device) were obtained. Each B-scan (Figure 1 A [1]) was then digitally enhanced (Figure 1 A [4]) by performing spatial averaging (each pixel value was replaced by the mean of its 8 lateral neighbors; Figure 1 A [2]) [89], compensation and contrast enhancement (contrast exponent = 2; Figure 1 A [3]) [32], and histogram equalization (contrast limited adaptive histogram equalization [CLAHE], clip limit = 2; Figure 1 A [4]) [71].…”

“…Figure 1: The dataset preparation for the image enhancement network is shown in (A). Each B-scan (A [1]) was digitally enhanced (4) by performing spatial averaging (each pixel value was replaced by the mean of its 8 lateral neighbors; A [2]), compensation and contrast enhancement (contrast exponent = 2; A [3]), and histogram equalization (contrast limited adaptive histogram equalization [CLAHE], clip limit = 2; A [4]).…”

Perceptual Fusion of Electronic Chart and Marine Radar Image

“…As the image information provided by ship marine radar has certain noise that affects normal observation of the radar image, it is necessary to preprocess the noise. For a raw image with noise, the processing affected by using convolutional neural network (CNN) [2] is superior to the traditional wavelet transform method. The GAN [3] model endows people with a mode of thinking by using the generative model and the discriminant model, in which the generative model is applied for training and learning, and the discriminant model is utilized to judge whether the generation model is close to the truth until the discrimination model can no longer judge the "authenticity" of the generation model.…”