DN-GAN: Denoising generative adversarial networks for speckle noise reduction in optical coherence tomography images. (January 2020)

• Record Type: Journal Article
• Authors: Chen, Zailiang; Zeng, Ziyang; Shen, Hailan; Zheng, Xianxian; Dai, Peishan; Ouyang, Pingbo
• Is Part Of: Biomedical signal processing and control. Volume 55(2020)
• Publication Date: 2020-01
• Subjects: Denoising generative adversarial networks (DN-GAN) -- Speckle reduction -- Detail preservation -- Optical coherence tomography (OCT) -- Layer segmentation; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101632
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The adversarial network is designed for speckle noise reducing in OCT images. The frequency loss function is first utilized in denoising problem. The generator is improved to be more suitable for the denoising problem. The proposed method achieves a trade-off in denoising and detail preservation. Abstract: Optical coherence tomography (OCT) is an efficient noninvasive bioimaging technique that can measure retinal tissue. Considering the changes in the acquisition environment during imaging, the OCT images are affected by granular speckle noise, thereby reducing the image quality. In this paper, an efficient method based on generative adversarial network is proposed to reduce the speckle noise and preserve the texture details. The proposed model consists of two components, that is, a denoising generator and a discriminator. The denoising generator learns how to map the noise image to the ground truth. The discriminator learns as a loss function to compare the differences between the ground truth and the image reconstructed by the generator. A number of repeated densely sampled B-scan OCT images are used with multi-frame registration to train the denoising generator. The original OCT images are denoised by a trained generator to quickly and efficiently obtain improved quality. Results showed that the proposed method outperforms the other popular methods, and achieves a better denoising effectiveness.