Speckle Reduction of OCT Via Super Resolution Reconstruction and Its Application on Retinal Layer Segmentation
Qifeng Yan,Bang Chen,Yan Hu,Jun Cheng,Yan Gong,Jianlong Yang,Jiang Liu,Yitian Zhao
DOI: https://doi.org/10.1016/j.artmed.2020.101871
IF: 7.011
2020-01-01
Artificial Intelligence in Medicine
Abstract:Optical coherence tomography (OCT) is a rapidly developing non-invasive three dimensional imaging approach, and it has been widely used in examination and diagnosis of eye diseases. However, speckle noise are often inherited from image acquisition process, and may obscure the anatomical structure, such as the retinal layers. In this paper, we propose a novel method to reduce the speckle noise in 3D OCT scans, by introducing a new super-resolution approach. It uses a multi-frame fusion mechanism that merges multiple scans for the same scene, and utilizes the movements of sub-pixels to recover missing signals in one pixel, which significantly improves the image quality. To evaluate the effectiveness of the proposed speckle noise reduction method, we have applied it for the application of retinal layer segmentation. Results show that the proposed method has produced promising enhancement performance, and enable deep learning-based methods to obtain more accurate retinal layer segmentation results.