Biomedical Engineering Reference
In-Depth Information
1
Introduction
Optical coherence tomography (OCT) is a non-invasive imaging modality with
several applications. It provides in vivo high-resolution cross-sectional imaging of
the retinal tissue through light scattering. However, as any imaging technique that
has his image formation based on coherent waves, OCT images suffer from speckle
noise which reduce its quality [
9
].
Speckle noise is a random phenomenon generated by interference of waves with
random phases [
10
], being a common problem to other imaging modalities as
ultrasound, synthetic-aperture radar (SAR) or laser imaging, leading to research and
resulting on many speckling reduction techniques [
10
].
It creates a grainy appearance that can mask diagnostically significant image fea-
tures and reduces the accuracy of segmentation and pattern recognition algorithms
[
10
,
26
,
28
]. Please refer to [
10
,
31
] for a further description of the speckle in OCT
characteristics.
The statistical mechanism of laser speckle formation was first presented by
Goodman [
14
]. Besides the theoretical results, this study also supports the idea
that speckle noise could be rejected by linear filtering. On the other hand, Wagner
et al. [
35
], Burckhardt et al. [
8
] and Abbott et al. [
1
] conclude that linear filtering,
the way it was presented in [
14
], suppresses the noise at the cost of smoothing out
image details.
Schmitt et al. [
29
] described the first OCT speckle suppression technique where
a compounded image was formed from the sum of the signals from a quadrant
photodiode detection system. A similar angular compounding approach was also
used by Bashkansky and Reintjes [
5
] and by Iftimia et al. [
16
].
Like angular compounding there were also developed other speckle reducing
methods applied before image formation (physical techniques), such as frequency
compounding or spatial compounding.
A frequency compounding technique has been used by Skankar [
33
] and by
Pircher et al. [
25
], where the contrast was increased and the quality image improved
without loss of resolution.
Kim et al. [
18
] presented a space diversity speckle reduction technique, reporting
a substantial reduction of speckle despite the reduced transverse resolution.
Yung et al. [
38
] described a zero-adjustment procedure (ZAP) to OCT which was
first applied by Healey et al. [
15
] in medical ultrasound.
The requirements on modifying the hardware led to the development of post-
processing methods, being the CLEAN algorithm one of the first image processing
techniques for OCT despeckling [
30
]. Among these are the median filtering [
7
],
homomorphic Wiener filtering [
12
], enhanced Lee filter (ELEE) [
22
], symmetric
nearest neighbor (SNN) filter, adaptive smoothing [
20
], multiresolution wavelet
analysis [
37
], filtering techniques based on rotating kernel transformations [
27
],
Kuwahara filter [
21
] and anisotropic diffusion filtering [
24
,
28
].
The median filter calculates for each pixel its median value in a local neigh-
borhood. Koozekanani et al.
[
19
] applied it to reduce the speckle of the OCT
