A Deep Learning Approach to Denoise Optical Coherence Tomography Images of the Optic Nerve Head

Abstract

Optical coherence tomography (OCT) has become an established clinical routine for the in vivo imaging of the optic nerve head (ONH) tissues, that is crucial in the diagnosis and management of various ocular and neuro-ocular pathologies. However, the presence of speckle noise affects the quality of OCT images and its interpretation. Although recent frame-averaging techniques have shown to enhance OCT image quality, they require longer scanning durations, resulting in patient discomfort. Using a custom deep learning network trained with 2,328 ‘clean B-scans’ (multi-frame B-scans; signal averaged), and their corresponding ‘noisy B-scans’ (clean B-scans + Gaussian noise), we were able to successfully denoise 1,552 unseen single-frame (without signal averaging) B-scans. The denoised B-scans were qualitatively similar to their corresponding multi-frame B-scans, with enhanced visibility of the ONH tissues. The mean signal to noise ratio (SNR) increased from 4.02 ± 0.68 dB (single-frame) to 8.14 ± 1.03 dB (denoised). For all the ONH tissues, the mean contrast to noise ratio (CNR) increased from 3.50 ± 0.56 (single-frame) to 7.63 ± 1.81 (denoised). The mean structural similarity index (MSSIM) increased from 0.13 ± 0.02 (single frame) to 0.65 ± 0.03 (denoised) when compared with the corresponding multi-frame B-scans. Our deep learning algorithm can denoise a single-frame OCT B-scan of the ONH in under 20 ms, thus offering a framework to obtain superior quality OCT B-scans with reduced scanning times and minimal patient discomfort.

A study of evaluating specific tissue oxygen saturation values of gastrointestinal tumors by removing adherent substances in oxygen saturation imaging

“…In recent years, methods of image processing that use deep learning approaches have been reported in several studies. It would be important to consider a more detailed image processing analysis for obtaining accurate StO2 information on tumors [ 30 ]. One of the limitations of this study is that a relatively small number of tumor images were used and annotation points of this study during image analysis were selected optionally by only two endoscopists.…”

DeshadowGAN: A Deep Learning Approach to Remove Shadows from Optical Coherence Tomography Images

“…A downsampling tower at each stage sequentially halved the dimensions of the baseline image (size 512 × 512) via maxpooling to capture the contextual information (i.e., spatial arrangement of tissues), and an upsampling tower sequentially restored it back to its original resolution to capture the local information (i.e., tissue texture). 22 A transposed convolution was performed four times in the upsampling tower for the predicted segmentation masks to be size 512 × 512, before passing to a sigmoid activation function for compression of each pixel to a value between 0 and 1.…”

Improved cortical surface reconstruction using sub-millimeter resolution MPRAGE by image denoising

“…Impressively, CNN-based denoising methods can be executed extremely efficiently (in milliseconds or seconds) once trained. Thus far, CNN-based denoising has been widely applied for various biomedical imaging modalities ranging from fluorescence microscopy 80,81 , optical coherence tomography 82 to x-ray imaging 83 , x-ray computed tomography 84 , PET 78,[85][86][87][88] and MRI 84,[89][90][91][92][93][94][95] .…”