Image noise reduction on the EMI 7070 CT scanner by the reduction of the noise resident in the CT wedge profiles


Mott, D.J.

Medical Physics 11(5): 666-669

1984


On the EMI 7070 CT scanner the data contained in the wedge files has been averaged resulting in a 6% reduction in the noise content of images. This noise reduction can either be used to improve image quality or can be offset by reducing the x-ray tube current by 12%. The averaging method will be outlined and its rationale discussed.

TECHNICAL
REPORTS
Image
noise
reduction
on
the
EMI
7070
CT
scanner
by
the
reduction
of
the
noise
resident
in
the
CT
wedge
profiles
David
J.
Mott
Regional
Department
of
Medical
Physics
and
Bioengineering,
Christie
Hospital
and
Holt
Radium
Institute,
Wilmslow
Road,
Manchester,
M20
9BX,
United
Kingdom
(Received
24
May
1983;
accepted
for
publication
25
October
1983)
On
the
EMI
7070
CT
scanner
the
data
contained
in
the
wedge
files
has
been
averaged
resulting
in
a
6%
reduction
in
the
noise
content
of
images.
This
noise
reduction
can
either
be
used
to
improve
image
quality
or
can
be
offset
by
reducing
the
x-ray
tube
current
by
12%.
The
averaging
method
will
be
outlined
and
its
rationale
discussed.
Key
words:
computer
tomography,
noise
reduction,
wedge
data
I.
INTRODUCTION
The
noise
content
of
computed
tomography
(CT)
images
has
been
investigated
from
both
its
theoretical
aspect"
and
its
effect
on
the
image
perception.
5
'
6
This
technical
communi-
cation
will
describe
a
method
of
reducing
the
noise
content
of
an
image
by
reducing
the
noise
contained
in
the
data
base
used
to
construct
the
image.
In
a
CT
scanner
there
are
varia-
tions
in
the
mechanical
alignment
and
the
characteristics
of
the
x-ray
and
detector
systems
which
can
be
measured
and
therefore
compensated
for
during
image
construction.
Ge-
nerically
the
information
required
to
perform
these
correc-
tions
is
known
as
the
machine
parameters
and
is
obtained
from
scans
of
well-defined
objects.
The
scanner
on
which
the
present
work
was
performed
is
the
EMI
7070
scanner
which
utilizes
one
of
a
set
of
five
aluminium
CT
wedges
during
scanning.
Their
function
is
to
reduce
the
dynamic
range
of
the
photon
flux
reaching
the
detectors,
thus
preventing
their
linear
response
region
from
being
exceeded.
The
effect
of
the
aluminium
wedge
has
to
be
removed
from
the
image
profiles
during
image
construction
and
this
task
requires
one
of
the
machine
parameters
known
as
the
wedge
data.
The
alumin-
ium
CT
wedges
are
designed
to
be
the
radiographic
negative
of
a
unit
density
cylindrical
calibration
phantom
and
five
such
phantoms
are
supplied with
the
machine,
one
for
each
wedge.
As
the
purpose
of
the
aluminium
CT
wedge
is
to
ensure
that
the
detectors
linear
response
region
is
not
ex-
ceeded
the
wedge
data
is
not
acquired
by
performing
air
scans
but
by
scanning
the
calibration
phantom
which
matches
the
specific
aluminium
wedge.
The
profile
data
from
such
a
scan
is
processed
up
to
the
point
of
deconvolu-
tion
and
back
projection
(i.e.,
all
offset,
reference,
afterglow,
and
scatter
corrections
are
performed,
logarithms
obtained
and
detector
calibration
performed)
then
stored
in
a
com-
pressed
form
for
use
during
image
construction.
In
order
to
reduce
the
quantum
noise,
the
wedge
profiles
are
the
average
of
four
maximum
technique
scans
(120
kV,
99
mA,
6
s),
and
one
such
average
profile
is
maintained
for
each
of
the
1088
detectors
in
the
system.
During
image
construction
the
im-
age
profiles
are
processed
to
the
same
stage
then
subtracted
from
the
wedge
profiles
before
deconvolution
and
back
pro-
jection.
However,
even
the
average
by
four
maximum
tech-
nique
scans
does
not
remove
all
the
quantum
noise
from
the
wedge
profiles
and
this
is
added
in
quadrature
to
the
noise
in
the
final
image.
To
illustrate
this,
images
of
the
calibration
phantom
were
produced
using
wedge
profiles
obtained
from
one,
two,
and
the
normal
four
independent
scans.
An
analy-
sis
of
the
changes
in
the
noise
content
in
these
images
con-
firmed
that
the
noise
is
added
in
quadrature
and
that
the
component
of
the
noise
in
an
image
due
to
noise
in
the
wedge
profiles
can
be
given
by:
cr
u
,=—
1
xu,
,
(
1
)
where
a„,
=
the
standard
deviation
in
the
image
due
to
noise
in
the
wedge
profiles,
and
a,
=
the
standard
deviation
in
the
image
of
the
calibration
phantom
obtained
using
the
same
scan
parameters
as
were
used
to
generate
the
wedge
profiles.
This
relationship
can
be
derived
from
simple
theory
and
is
only
true
if
the
same
aluminium
wedge,
calibration
phan-
tom,
and
scan
techniques
are
used
for
both
the
wedge profile
scanning
and
the
image
scanning.
This
equation
allows
the
magnitude
of
the
noise
in
the
wedge
profiles
to
be
assessed
and
its
results
are
tabulated
in
Table
I.
In
this
table
standard
deviations
are
quoted
for
both
the
6-s
scan
speed
used
to
acquire
the
wedge
profiles
and
the
3-s
scan
speed
used
exclu-
sively
for
clinical
scanning.
Only
two
of
the
five
wedge
sizes
are
quoted
as
these
are
the
ones
that
are
used
for
the
majority
of
clinical
examinations.
The
standard
deviations,
both
in
this
table
and
throughout
the
work,
are
measured
over
a
100
pixel
radius
region
of
interest
which
is
in
accord
with
the
technique
of
Speller
et
al.'
The
image
has
a
160
pixel
radius
so
this
is
a
large
area,
therefore
attention
was
paid
to
phan-
tom
centering
during
all
the
experiments
to
prevent
the
in-
clusion
of
any
beam
hardening
artifacts
into
the
measure-
ments.
666
Med.
Phys.
11
(5),
Sep/Oct
1984
0094-2405/84/050666-04601.20
©
1984
Am.
Assoc.
Phys.
Med.
666
Wedge
Scan
size
speed
mm
Standard
deviations
in
HU
Image
if
Original
Calculated
wedge
image
wedge
noise
noise
noise
eliminated
667
Technical
Reports:
David
J.
Mott:
Image
noise
reduction
on
the
EMI
7070
CT
scanner
667
TABLE
I.
Predicted
values
of
the
standard
deviation
due
to
quantum
noise
in
the
wedge
profiles
and
the
potential
effect
on
image
noise
if
it
were
re-
moved.
320
6
5.97
2.67
5.33
3
7.21
6.70
400
6
12.8
5.72
11.4
3
14.7
13.5
II.
TECHNIQUE
While
there
are
a
number
of
techniques
that
could
be
em-
ployed
to
reduce
the
noise
content
of
the
wedge
profiles,
the
technique
that
has
been
developed
is
to
average
the
profiles
over
groups
of
detectors.
The
1088
detectors
in
the
system
are
manufactured
in
packs
of
eight
and
this
leads
to
varia-
tions
in
the
angular
sensitivity
of
the
detectors
within
a
pack.
The
detectors
at
the
edges
of
a
pack
display
a
marked
asym-
metry
in
their
angular
sensitivity
as
their
profiles
have
a
marked
rise
or
fall
along
their
length.
To
account
for
this,
averaging
is
performed
over
groups
of
detectors
with
the
same
locations
within
each
pack.
Thus
the
largest
number
of
detectors
that
could
be
averaged
over
is
136
(1088/8).
Aver-
aging
is
performed
over
smaller
numbers
of
detectors
with
the
only
criterion
being
that
the
number
of
detectors
aver-
aged
over
is
an
integer
divisor
of
136.
While
the
profiles
in
any
group
have
the
same
shape
and
range
of
values,
they
do
exhibit
a
mean
shift
from
detector
to
detector. To
account
for
this
shift,
the
mean
of
the
central
fifty
values
in
the
profile
is
found
and
removed
from
all
the
values
in
the
profile.
This
process
is
repeated
for
all
the
profiles
and,
after
averaging
the
profiles,
the
mean
value
is
added
back
for
each
detector
before
repacking.
III.
RESULTS
There
are
three
areas
of
investigation.
Firstly,
it
has
to
be
shown
that
noise
reductions
of
the
magnitude
predicted
can
be
achieved.
The
second
investigation
is
to
demonstrate
that
there
are
no
adverse
effects
on
clinical
images.
Lastly,
as
a
noise
reduction
is
achieved,
this
can
be
offset
by
a
dose
re-
duction
which
will
be
found
experimentally.
As
with
all
re-
cent
scanners
the
7070
scanner
allows
the
reprocessing
of
image
data
from
image
profiles.
The
7070
maintains
very
primitive
image
profiles
in
that
they
are
stored
with
only
the
offset
and
reference
corrections
performed.
As
they
have
not
been
wedge
corrected,
the
same
image
profile
can
be
recon-
structed
using
different
wedge
profiles.
In
doing
this,
slice-
to-slice
variations
can
be
eliminated
to
give
results
that
are
only
dependent
on
the
averaging
technique.
Table
II
shows
the
reduction
in
the
image
noise
in
images
of
the
two
calibration
phantoms
as
a
function
of
the
number
of
detectors
averaged
over.
The
value
quoted
after
averaging
over
17
detectors
is
in
close
agreement
to
the
value
predicted
in
the
last
column
of
Table
I.
If
the
data
in
Table
II
are
plotted
as
the
variance
against
the
inverse
of
the
number
of
detectors
averaged
over
a
linear
relationship
is
found.
As
this
is
to
be
expected
from
normal
statistical
laws
it
indicates
that
any
adverse
effects
or
artifacts
produced
by
this
technique
are
not
statistically
significant.
To
illustrate
that
the
tech-
nique
performs
equally
well
on
all
sections
of
the
image
Ta-
ble
III
is
an
example
of
the
radial
distribution
of
the
noise
in
an
image
of
the
320-mm
phantom.
This
table
was
generated
by
measuring
the
standard
deviation
in
a
series
of
ten
con-
centric
contiguous
annular
rings
each
with
a
radial
width
of
15.9
pixels.
It
shows
the
extent
of
the
radial
dependence
of
the
noise
in
the
7070
scanner
but
also
illustrates
that
the
technique
has
a
slight
radial
dependence
as
it
has
a
greater
effect
on
the
already
noisier
edge
of
the
image.
The
discre-
pancies
between
the
results
of
the
subtracted
image
(C
)
and
the
calculated
results
(D)
are
marginal.
In
all
the
phantom
scanning
that
was
performed
during
the
testing
of
the
tech-
nique
no
shift
in
the
mean
pixel
value
of
greater
than
0.8
Hounsfield
units
(HU)
was
observed
and
the
majority
of
the
images
showed
no
shift
in
their
mean
pixel
values.
Thus,
the
technique
does
not
affect
the
calibration
of
the
scanner.
To
observe
any
effects
of
the
technique
on
the
resolution
observed
in
clinical
images,
the
method
of
reconstructing
an
image
with
respect
to
both
sets
of
wedge
profiles,
was
again
used.
On
subtraction
of
one
image
from
the
other
no
anato-
mical
information
was
seen.
This
process
has
been
repeated
on
a
number
of
clinical
images
and
shows
that
the
spatial
resolution
has
not
been
changed.
However,
this
analysis
has
highlighted
that
a
radial
artifact
can
be
produced.
This
arti-
fact
is
just
perceptable
on
scans
of
the
320-mm
phantom
when
processed
using
the
averages
wedge
profiles
but
is
demonstrated
with
greater
ease
by
performing
the
subtrac-
tion
method.
Figure
1
illustrates
the
artifact
and
the
picture
was
obtained
by
subtracting
an
image
of
the
320-mm
phan-
tom
constructed
with
the
averaged
wedge
profiles
from
the
same
image
constructed
with
the
original
wedge
profiles.
In
essence
it
is
the
image
of
the
difference
between
the
two
sets
of
wedge
profiles.
The
noise
in
this
image
is
considerably
TABLE
II.
The
reduction
in
the
standard
deviation
as
a
function
of
the
number
of
detectors
averaged
over
for
the
320-
and
400-mm
wedges.
Wedge
scan
parameters:
120
kV,
99
mA,
10-mm
slice
width,
6-s
scan
time.
Image
scan
parameters:
120
kV,
99
mA,
10-mm
slice
width,
3-s
scan
time.
Standard
deviation
Number
of
detectors
that
the
wedge
profiles
are
averaged
over
in
HU
2
4
8
17
Phantom/wedge
size
in
mm
320
7.21
6.95
6.83
6.77
6.74
400
14.6
14.1
13.8
13.7
13.6
Medical
Physics,
Vol.
11,
No.
5,
Sep/Oct
1984
668
Technical
Reports:
David
J.
Mott:
Image
noise
reduction
on
the
EMI
7070
CT
scanner
668
TABLE
III.
The
standard
deviation
as
a
function
of
radial
position
for
images
of
a
calibration
phantom
processed
using
both
sets
of
wedge
profiles
and
the
calculated
and
measured
differences.
A
=
image
constructed
using
the
original
wedge
profiles,
B
=
image
constructed
using
the
average
over
17
detectors
wedge
profiles,
C
=
image
formed
by
subtracting
image
A
from
image
B,
and
D
=
calculated
difference
between
A
and
B
using
D
=
V(A
2
-
B
2
).
Mean
radius
Number
Annulus
in
of
Standard
deviation
in
annulus
number
pixels
pixels
A
B
C
D
10.5
788
6.81
6.41
2.37
2.30
2
24.6
2
348
6.82
6.36
2.55
2.46
3 40.0
3
904
6.98
6.50
2.57
2.54
4
55.6
5
508
7.19
6.70
2.64
2.61
5
71.4
7
068
7.42
6.91
2.69
2.70
6
87.2
8
628
7.68
7.21
2.72
2.64
7
102.9
10
196
7.94
7.41
2.92
2.85
8
118.7
11
740
8.33
7.81
3.01
2.90
9
134.4
13
344
8.90
8.35
3.20
3.08
10
150.0
14
216
10.01
9.40
3.51
3.44
lower
(2.4
HU)
than
that
found
in
conventional
images
(6-16
HU)
which
has
allowed
the
very
narrow
window
width
of
20
HU
to
be
used.
At
this
window
width
the
artifact
is
visible
compared
to
the
low
overall
noise
but
is
not
observed
on
clinical
images,
where
there
is
not
only
a
higher
noise
con-
tent
but
where
considerably
larger
window
widths
are
used
to
view
them.
As
the
artifact
is
only
observed
under
condi-
tions
of
low
image
noise
and
content,
it
is
considered
insigni-
ficant.
A
noise
reduction
of
the
type
illustrated
can
be
utilized
in
two
ways.
It
can
either
be
accepted
that
clinical
images
will
be
quieter
or
it
can
be
offset
by
a
subsequent
dose
reduction.
There
had
been
no
adverse
comments
about
the
original
noise
content
in
clinical
images
so
it
was
decided
to
reduce
the
patient
dose
as
this
could
also
produce
the
added
benefit
of
extending
the
life
of
the
x-ray
tubes.
The
noise
reduction,
while
it
is
a
reduction
in
the
quantum
noise
in
the
wedge
profiles,
is
a
reduction
in
the
system
noise.
Therefore,
the
mA
is
not
simply
reduced
by
the
ratio
of
the
change
in
the
variance
as
is
predicted
by
the
theoretical
equations,"
as
these
equations
do
not
take
account
of
system
noise.
It
was
decided
to
experimentally
establish
the
mA
which
would
result
in
the
same
noise
content
in
an
image
of
a
calibration
phantom
using
the
averaged
wedge
profiles
as
that
found
when
the
same
object
is
scanned
at
the
maximum
mA
of
99
mA
using
the
original
wedge
profiles.
The
results
of
these
scans
are
given
in
Table
IV
and
illustrate
that
87
mA
is
appli-
cable.
In
general
terms
a
6%
noise
reduction
has
been
com-
pensated
for
by
a
12%
dose
reduction
and
this
simple
12%
reduction
in
mA
has
been
applied
to
clinical
scanning.
IV.
DISCUSSION
It
has
been
demonstrated
that
a
6%
noise
reduction
can
be
achieved
by
the
virtual
elimination
of
the
noise
in
the
wedge
profiles
and
that
this
can
be
offset
by
a
12%
dose
reduction.
No
adverse
effects
are
seen
on
images
apart
from
an
artifact
which
can
only
be
clearly
seen
under
specific
conditions
of
very
low
image
content
and
noise.
The
12%
dose
reduction
is
an
oversimplification
as
the
noise
content
in
patient
im-
ages
are
invariably
lower
than
that
found
in
phantoms,'
but
has
been
implemented
in
this
simple
form
for
ease
of
use
by
the
radiographers.
In
general
terms
this
reduction
in
the
sys-
tem
noise
has
a
greater
effect
on
scans
of
smaller
patients
where
the
noise
in
the
image
profiles
is
closer
to
that
in
the
wedge
profiles.
The
programs
developed
to
perform
the
technique
were
written
in
FORTRAN
IV
with
assembler
subroutines
and
take
about
4-5
min
per
wedge
which
does
not
add
significantly
to
the
time
taken
to
create
the
wedge
profiles.
The
program
will
work
on
all,
or
selected,
wedge
data
files
and
the
user
deter-
mines
the
number
of
detectors
to
average
over.
As
the
corn-
TABLE
IV.
The
measured
percentage
reduction
in
the
standard
deviation
as
a
function
of
mA.
The
percentage
noise
change
is
the
percentage
change
from
the
noise
figures
found
using
the
original
wedge
profiles
and
scanning
at
99
mA
with
a
3-s
scan
speed.
Note:
A
negative
percentage
noise
change
means
that
the
noise
has
increased.
X-ray
tube
current
in
mA
Percentage
noise
change
90
85
80
320
2.3
-0.9
-4.0
FIG.
1.
Radial
artifact
produced
by
the
average
wedge
profile
technique.
A
window
width
of
20
HU
was
used.
Wedge
size
in
mm
400
2.2
-
1.6
-
5.1
Medical
Physics,
Vol.
11,
No.
5,
Sep/Oct
1984
669
Technical
Reports:
David
J.
Mott:
Image
noise
reduction
on
the
EMI
7070
CT
scanner
669
pressed
storage
format
used
to
store
the
data
is
of
limited
interest,
no
details
have
been
included
but
both
the
program
and
any
further
information
are
available
to
other
7070
users
from
the
author
on
request.
'R.
A.
Brooks
and
G.
D.
Chiro,
Med.
Phys.
3,
237
(1976).
2
H.
H.
Barrett,
T.
Bowen,
R.
S.
Hershel,
S.
K.
Gordon,
and
D.
A.
Delise,
presented
at
"Image
Processing
for
2-D
and
3-D
Reconstruction
from
Projections,"
Stamford,
CA,
Optical
Society
of
America,
4-7
August
1975,
p.
WB
2.
3
D.
A.
Chesler,
S.
J.
Riederer,
and
N.
J.
Pelc,
J.
Comput.
Assist.
Tomogr.
1,
64
(1977).
4
0.
J.
Tretiac,
J.
Comput.
Assist.
Tomogr.
2,
477
(1978).
5
P.
F.
Judy,
R.
G.
Swensson,
and
M.
Szulc,
Med.
Phys.
8,
13
(1981).
6
G.
Cohen
and
F.
A.
DiBianca,
J.
Comput.
Assist.
Tomogr.
3,
189
(1979).
'R.
D.
Speller,
D.
R.
White,
C.
K.
Showalter,
L.
M.
Rothenberg,
K.
S.
Pentlow,
T.
J.
Morgan,
and
T.
B.
Shope,
Br.
J.
Radiol.
54,
1053
(1981).
9/
J.
Mott
and
K.
Faulkner,
presented
at
"Dose
Reduction
in
Diagnostic
Radiology,"
London,
1983,
HPA
Conference
Report
Series
(in
prepara-
tion).
Medical
Physics,
Vol.
11,
No.
5,
Sep/Oct
1984