Transcript An RBF Network for Detection of Prostate Cancer Based on

A Spatially Adaptive Filter Reducing
Arc Stripe Noise for Sector Scan
Medical Ultrasound Imaging
Qianren Xu
Mohamed Kamel
Magdy M. A. Salama
Outline
• Introduction
• Method
• Experiment Results
• Conclusion
2
Introduction
•
•
•
Sector scan ultrasound images usually have arc stripes;
They do not represent the physical structure of the tissue;
Thus they can be viewed as a kind of noise.
3
The Source of the Arc Stripe Noise
Assume that there are point targets
with same size.
The lateral size of image of these
points increase beyond focal zone.
These laterally wider images will
be superimposed on the far sides
of the focal location, and thus
these target points that are
originally separated will show as
arc stripes in far-field and nearfield.
4
The properties of these noise
Special geometric properties of the arc stripe
noise:
•
•
Circular symmetry.
The intensity and size of the arc
stripes change with the radial
depth.
The proposed filter is based
on the geometric properties
5
Proposed Filter
It consists of two components:
Radially adaptive
filtering operators
+
Common Gaussian
filtering operator
6
Radially Adaptive Filtering Operators:
Basic Radial Filtering Operators at Special Directions
1 0 0
1
op3  0 2 0
4
0 0 1
0 1 0 
1
op2  0 2 0
2
0 1 0
y
0 0 0 
1
op4  1 2 1
4
0 0 0
0 0 1 
1
op5  0 2 0
4
1 0 0
0 0 1 
1
op1  0 2 0
4
1 0 0
x
0 1 0 
1
op6  0 2 0
4
0 1 0
0 0 0 
1
op0  1 2 1
4
0 0 0
1 0 0 
1
op7  0 2 0
4
0 0 1
7
Radially Adaptive Filtering Operators:
Radial Filtering Operators at Arbitrary Directions
The filtering operator at any azimuth angle θ is
determined by soft weighted summation of
neighbor basic radial filtering operators
4
op   m opm , m  1, 2, 3, 4,
m 1

  (m  1) / 4

m  1 
 /4
0
if   (m  1)  / 4   / 4,
otherwise.
8
The Combined Filter
Weighted Summation of the Radial Filtering Operator
and Gaussian Filtering Operator:
4
op '   g op g   r (  m op m ),
m 1
Radial filtering
operators aim to
reduce random
directional noise
Common Gaussian filter is used
1) to counteract the radial stripe
artifact, and 2) suppress the nondirectional noise
9
An Example of the Combined Filter
 

4
 g  r  0.5
10
Selection of Parameters
4
op '   g op g  r ( m opm ),
m 1
1. The weight ωg and ωg are determined by the ratio
of non-directional and arc stripe noise
components
2. The Gaussian standard deviation σ of opg and
opm are determined by the size of non-directional
and arc stripe noise noises respectively
3. The size of filter mask is determined by noise size
11
Testing Image on the Radial Filter
(a) Original testing image
(b) Filtered image by the radial filter
12
Filtered image by Gaussian filter
13
Filtered image by the proposed filter
Example II
(a) The original image of fetus
(b) The filtered image
14
Conclusion
•
This paper identifies a significant noise, the arc stripes
in sector scan medical ultra-sound image, and
generalizes the characteristics of the arc stripe noise.
•
The proposed filtering algorithm deals with the arc
stripe noise by utilizing the geometric characteristics of
the special noise,
•
The parameters of the filter are adapted with the radial
depth in order to effectively smooth noise and deblur
the useful image detail.
•
The results show that the proposed filter obviously
enhances image quality and is superior to common
smoothing filter.
15
Thanks for your time
Questions?
16