Biomedical Engineering Reference
In-Depth Information
images to a posterior measurement of the retinal thickness. On the other hand,
Ishikawa et al. [
17
], with the same purpose of measuring the retinal thickness,
used a modified mean filter to reduce the noise. The Gaussian filter is another filter
applied on the retinal segmentation algorithms [
3
,
4
,
32
,
34
]. In the adaptive filtering
the algorithm is modified locally based on the pixel's neighborhood, combining an
effective noise reduction and an ability to preserve the image edges. The Lee filter
is one of this adaptive filters that was successfully applied to OCT images. Ozcan
et al. [
23
] compare the performance of different filters (an enhanced Lee filter, two
a trous
wavelet-transform-based filters, a hybrid median filter, a symmetric nearest-
neighbor filter, a Kuwahara filter and an adaptive Wiener filter), when applied
independently or when applied in association, to reduce the speckling present on
an OCT tomogram of a bovine retina [
26
]. A fuzzy thresholding algorithm in the
wavelet domain was proposed by Puvanathasan [
26
] to remove the speckle noise in
OCT images of a human finger tip.
Fernandez et al. [
11
] and Salinas et al. [
28
] have shown that a nonlinear
complex diffusion filter can be successfully applied to remove OCT speckle noise
while preserving image features. Recently, [
6
] proposed an improvement to these
approaches, suggesting an adaptive time step and a particular choice of parameters
to be more conservative in the diffusion of signal within retinal tissue, therefore
taking a specific approach for OCT data despeckling that can be generalized for
similar medical images.
Current despeckling methods applied to process noisy optical coherence tomog-
raphy data take into consideration each B-scan individually, therefore looking to the
3D data as a set of individual 2D images [
23
,
26
,
28
]. In this way, the consistency of
noise along the entire 3D data volume is not taken into account.
In the work herewith presented, we have extended the application of complex
diffusion filters [
28
]and [
6
] from 2- to 3-dimensions therefore considering the
entire volume as a single entity and not as a set of aggregated 2D entities.
As a proof-of-concept, the proposed method will be compared resorting to
quantitative measures with filtering methods from the literature.
2
Material and Methods
2.1
Optical Coherence Tomography
The OCT working principle is similar to ultrasound and adopted some of its
terminology from that field.
The volumetric OCT information is composed of a set of A-scans (depth-wise
information on refractive index changes) (Fig.
1
). The scanning is performed along
a series of parallel lines covering the 20
ı
field-of-view of the eye fundus and allows
users access to an unprecedented detail of the retina structures from in vivo subjects.
