Self-fusion for OCT noise reduction

Abstract

Purpose : Reducing speckle noise is important for improving visual and automated assessment of OCT images. Traditional image processing methods only offer moderate speckle reduction; deep learning methods can be more effective but require substantial training data, which may not be readily available. We present a novel method, self-fusion, that offers effective speckle reduction comparable to deep learning methods, but without any external training data.

Methods : Traditional multi-atlas methods rely on training samples, or “atlases”. Each atlas is deformably registered to the test image. Each atlas is then assigned a spatially varying weight, based on similarity between the registered atlas and the test image. Intensity fusion consists of taking the weighted average of the atlases to reconstruct the test image with reduced noise. We propose the new “self-fusion” technique that does not require any atlases. Instead, for each B-scan in an OCT volume, we use the neighboring B-scans as “atlases”. Since the entire volume is acquired through the same camera from the same eye, these B-scans offer exceptionally well-fitting atlases, making the process very robust. The outcome is speckle reduction abilities comparable to that of deep learning methods, but without the need for any external training data.

Results : 12 OCT volumes were acquired using a spectral-domain OCT system, with 5 repeated frames at each position. OCT SNR was adjusted by varying the detector exposure time, resulting in SNR values of 101 dB, 96 dB, and 92.5 dB. Volumes were acquired in two healthy volunteers in foveal and optic nerve head (ONH) region at each exposure setting. Figures 1 and 2 show results for varying SNR levels, in the ONH and fovea respectively. Self-fusion has good performance on even very noisy images. This is potentially useful in clinical applications for patients who may suffer from cataracts, vitreal haze, or corneal opacity. We also note that the self-fusion result often has visually better contrast than the average of 5 acquisitions, which is often considered the gold standard for despeckling.

Conclusions : Our results illustrate the performance of the novel self-fusion method on a variety of settings for speckle noise reduction in OCT images.

This is a 2020 ARVO Annual Meeting abstract.

 

Figure 1. Self-fusion in the ONH.

 

Figure 2. Self-fusion in the fovea. Note the external limiting membrane is visible on the 96 dB and 101 dB self-fusion and averaged images but not on the raw single images.