eduzhai > Applied Sciences > Engineering >

Self-supervised Denoising via Diffeomorphic Template Estimation Application to Optical Coherence Tomography

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Optical Coherence Tomography (OCT) is pervasive in both the research andclinical practice of Ophthalmology. However, OCT images are strongly corruptedby noise, limiting their interpretation. Current OCT denoisers leverageassumptions on noise distributions or generate targets for training deepsupervised denoisers via averaging of repeat acquisitions. However, recentself-supervised advances allow the training of deep denoising networks usingonly repeat acquisitions without clean targets as ground truth, reducing theburden of supervised learning. Despite the clear advantages of self-supervisedmethods, their use is precluded as OCT shows strong structural deformationseven between sequential scans of the same subject due to involuntary eyemotion. Further, direct nonlinear alignment of repeats induces correlation ofthe noise between images. In this paper, we propose a joint diffeomorphictemplate estimation and denoising framework which enables the use ofself-supervised denoising for motion deformed repeat acquisitions, withoutempirically registering their noise realizations. Strong qualitative andquantitative improvements are achieved in denoising OCT images, with genericutility in any imaging modality amenable to multiple exposures.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×