
Convolutional dictionary learning for blind deconvolution of optical coherence tomography images
Author(s) -
Junzhe Wang,
Brendt Wohlberg,
Robert Adamson
Publication year - 2022
Publication title -
biomedical optics express
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.362
H-Index - 86
ISSN - 2156-7085
DOI - 10.1364/boe.447394
Subject(s) - deconvolution , optical coherence tomography , computer science , point spread function , speckle pattern , speckle noise , artificial intelligence , optics , blind deconvolution , brightness , weighting , convolution (computer science) , computer vision , algorithm , physics , acoustics , artificial neural network
In this study, we demonstrate a sparsity-regularized, complex, blind deconvolution method for removing sidelobe artefacts and stochastic noise from optical coherence tomography (OCT) images. Our method estimates the complex scattering amplitude of tissue on a line-by-line basis by estimating and deconvolving the complex, one-dimensional axial point spread function (PSF) from measured OCT A-line data. We also present a strategy for employing a sparsity weighting mask to mitigate the loss of speckle brightness within tissue-containing regions caused by the sparse deconvolution. Qualitative and quantitative analyses show that this approach suppresses sidelobe artefacts and background noise better than traditional spectral reshaping techniques, with negligible loss of tissue structure. The technique is particularly useful for emerging OCT applications where OCT images contain strong specular reflections at air-tissue boundaries that create large sidelobe artefacts.