Artifact Removal from Spectral Domain OCT Images - Mohammad R. N. Avanaki

Optical coherence tomography (OCT) is a high resolution, non-invasive imaging modality based on non-ionising optical radiation, which delivers three-dimensional
(3D) images from microstructure components within the tissue [1]. The time domain (TD-) OCT produces threedimensional images by using the transversal scan with two mirrors (XY scanners) and the axial scan by moving the reference arm. In contrary, in frequency domain (FD-) OCT the reference arm is fixed and the optical path
length difference between the sample and the reference reflections is encoded by the frequency of the interferometric fringes as a function of the source spectrum. FD-OCT is broken down into two categories; spectral domain (SD-) OCT, which uses a grating to spatially scatter the spectrum across an array detector, and swept source (SS-) OCT. In this letter an implementation of denoising algorithm into a friendly user interface is presented. The noise investigated in this letter is a fixed-pattern high spatial frequency artifact superimposed with the SD-OCT images and is due to the lack of the quality of the system. The image obtained from the reference arm is considered as background image and is used to remove the unwanted signal in the noisy image. The algorithm was implemented in Matlab and a standalone C-based code was generated from the m-file afterwards [1]. The program prompts the user to input the background image and a set of noisy
images, and removes the artifacts such that the output is cleaner images.

METHODOLOGY
A cross-section image taken by the SD-OCT when the reference arm is blocked is given in Figure 1a. This image was meant to be a completely black image as the object arm is blocked but due to the imperfection of the optical devices, the electronics noise, and the effect of the reference arm signal, such lines appear on the image. The algorithm of the denoising filter is as follows; removing low spatial frequency noise in the image by estimating the statistics of the noise and the Gaussian approximation of the noise, finding the high spatial frequency edges in the background image, removing the pixels in the location of edges in the image and replacing them with the average of the vertical adjacent pixels. To detect the high spatial frequency edges Sobel operator was used. Sobel operator converts the original image to gradient image by finding those points where the spatial gradient of the image is maximum. Sobel operator was chosen because it is fast, and simple to be implemented. A 3x3 Sobel operator is convolved with the background image to generate the gradient image following a simple thresholding, as shown in Figure 1c.