Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

What Makes for a ‘Quality’ Image?
Image Quality – Chapter 10

¬

Insert audience participation here

Brent K. Stewart, PhD, DABMP
Professor, Radiology and Medical Education
Director, Diagnostic Physics
a copy of this lecture may be found at:
http://courses.washington.edu/radxphys/PhysicsCourse04http://courses.washington.edu/radxphys/PhysicsCourse04-05.html

Brent K. Stewart, PhD, DABMP

1

Brent K. Stewart, PhD, DABMP

Take Away: Five Things You should be able
to Explain after the Image Quality Lecture
¬

¬

¬

¬

3

Applies to all types of images
‘Quality’: subjective notion, dependent on image function
Bottom line outcome measure of a radiological image is
its usefulness in determining an accurate diagnosis
Understanding the image characteristics that comprise
image quality important so that radiologists can
recognize problems and articulate their cause
Introduction to the terminology used for various metrics
used by physicists and engineers to measure image
quality, e.g., contrast, spatial resolution and noise

Brent K. Stewart, PhD, DABMP

4

1


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

Contrast
¬
¬

in xx-ray intensity based on
attenuation
Cs = (A(A-B)/A
µ(x+z)
For A=N0e-µx and B=N0e-µ(x+z)
-µz
Cs = 11-e
either µ(E) or z to Cs
E to µ(E) and thus Cs
Why low kVp used in
mammography

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 256.
5

¬

¬

¬

¬

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., pp. 257257-258.
6

Brent K. Stewart, PhD, DABMP


¬
¬

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 260.
7

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

For screenscreen-film radiography
analog film OD the output
Cr = ODA - ODB
In radiography the contrast
cannot in general be adjusted
or enhanced on the analog film
How does the final light signal
reaching the radiologist’s eyes
depend on patient thickness
(assuming a constant µ)?
X ∝ e-µx
OD ∝ g · log10(X)
T = 10-OD
c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., pp. 259 and 261. 8

Brent K. Stewart, PhD, DABMP

2

image contrast

¬

1. The object contrast does not depend on the lesion’s:

¬

A. Thickness
B. Density
C. Atomic number
D. Background composition
E. Temperature

¬
¬
¬
¬

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 262.
9

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

Raphex 2003 Diagnostic Question

Huda 2nd Edition – Chapter 5 – Image Quality

A. Focal spot to film distance
B. mAs
C. Tube voltage
D. Developer temperature
E. Film gradient

¬
¬
¬
¬

11

Brent K. Stewart, PhD, DABMP

10

12

3


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

Huda 2nd Edition – Chapter 5 – Image Quality

Raphex 2000 Diagnostic Question

¬
¬
¬
¬

Brent K. Stewart, PhD, DABMP

13

Brent K. Stewart, PhD, DABMP

Spatial Resolution
¬
¬

¬

¬

Spatial Domain: the Point Spread Function

2D image really 3D: H, W (both spatial) and gray scale
Spatial resolution is a property that describes the ability
of an imaging system to accurately depict objects in the
two spatial dimensions of the image (x,y)
Classic notion: ability of an imaging system to distinctly
depict objects as they become smaller and closer
together
The spatial domain refers to the two spatial dimensions
(x,y) of an image


15

Isotropic and nonnon-isotropic

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 264.
16

Brent K. Stewart, PhD, DABMP

4


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

Spatial Domain: the Point Spread Function

Image Processing Based on Convolution
¬

¬

stationary

¬


the number array multiply the
convolution kernel values by the
associated values in the numeric
array and sum
Place the sum into a new numeric
array at the same location
c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 312.
18

Brent K. Stewart, PhD, DABMP

10K 1K 100 10

Physical Mechanisms of Blurring
¬
¬

¬
¬

¬

¬

Defocusing (lens)
Optical diffusion (intensifying
screen)
Motion (involuntary, cardiac)
Slice thickness (angled

Temporal frequencies (middle
A: ν = 440 Hz) that comprise a
timetime-domain audio signal (t)
Similarly the objects in an
image (audio signal) can be
thought of as the superposition
of spatial frequencies
For objects in an image that
are separated by shorter
distances (mm, x), these
objects correspond to high
spatial frequencies
(cycles/mm, f)
Square wave
line pairs per
mm (lp/mm)
c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 269.
20

Brent K. Stewart, PhD, DABMP

5


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004


¬

Image Intensifier Example

¬

¬
¬
¬
¬

¬

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., pp. 270270-271.
Brent K. Stewart, PhD, DABMP

¬
¬

¬
¬

Compute the MTF(f) from the
LSF(x) using the Fourier
Transform (FT)
MTF(f) = |FT{LSF(x)}|
FT an integral calculus
operation that converts a
spatial domain (x) signal into a

23

Weakest link – item in imaging
chain with worst MTF often
determines the overall system,
“dragging”
dragging” the system down

Brent K. Stewart, PhD, DABMP

¬

Brent K. Stewart, PhD, DABMP

A: Optics MTF(f)
B: Image intensifier MTF(f)
C: Video camera MTF(f)
System (f) = Optics MTF(f) *
Image intensifier MTF (f) *
Video camera MTF(f)

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 272.
22

21

The LSF, MTF and Fourier Transform
¬


¬
¬
¬

Brent K. Stewart, PhD, DABMP

¬

8. Spatial resolution cannot be assessed using:

¬

A. Line pair phantom
B. LSF image
C. Full-width half maximum
D. MTF curve
E. Pixel standard deviation

¬
¬
¬
¬

25

Brent K. Stewart, PhD, DABMP

Davis Notes - Image Quality
¬


C. MTF number 3

¬
¬

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

26

27

Brent K. Stewart, PhD, DABMP

28

7


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

Davis Notes - Image Quality
¬

¬
¬


¬

A. Relative amplitude, distance
(mm)
B. Spatial frequency (lp/mm),
distance (mm)
C. Lateral dimension (mm),
Fresnel ratio
D. Relative amplitude, spatial
frequency (lp/mm)
E. Relative amplitude, relative
amplitude

σ =

Brent K. Stewart, PhD, DABMP

Gaussian (normal) distribution:

G( x ) = ke
¬
¬

¬

1⎛ x − x ⎞
− ⎜
2 ⎝ σ ⎟⎠



¬

Many commonly encountered
measurements of people and
things make for this kind of
distribution (Gaussian) hence
the term “normal”
e.g., the height of 1000 third
grade school children
approximates a Gaussian

¬

¬

¬

¬
c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 275.
31

0.4
0.35

m = mean, shape governed by
one variable
P(x) difficult to calculate for
large values of x due to x!

0.2

m=8
0.15

m=10
m=20

0.1
0.05
0
0

10

20

30

40

Radioactive decay
Quantum mottle

Brent K. Stewart, PhD, DABMP

32

8



¬
¬

Error bars and confidence
intervals
P(x) very similar to G(x) when
≈ √x
use G(x) as approx.
Can adjust the noise ( ) in an
image by adjusting the mean
number of photons used to
produce the image
c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., pp. 276 - 277. 33

¬

¬

¬
¬

¬

N = mean photons/unit area
2
= √N, from P(x)
(variance) = N
P(x)

¬

¬
¬

¬

¬

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 281.
35

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

Although noise appears
random, the noise has a
frequency distribution
Example: ocean waves
Take a flatflat-field xx-ray image
(still has noise variations)
Fourier Transform (FT) the flat
image
Noise Power
Spectrum: NPS(f)
NPS(f) is the noise variance
( 2) of the image expressed as
a function of spatial freq. (f)

DQE = SNRin2

SNR2in = N (

¬

15. Which of the following is not true for Poisson
distributions?

¬

A. They are used to describe radioactive decay
B. They are used to describe quantum mottle
C. The variance is equal to the mean
D. They are always symmetrical
E. They are approximate to a Gaussian for means
greater than 10

¬
¬

SNR = √N)

¬

[MTF (f )]

2

¬

D9. The degree of mottle in a screenscreen-film xx-ray image is:

¬

A. Usually determined by the number of primary xx-ray
photons absorbed in the film.
B. Independent of the mAs.
C. An inherent property of the AgBr structure in the film
emulsion.
D. Increased by increasing the film speed.
E. Measured in line pairs per millimeter (lp/mm).

¬
¬

¬
¬

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

¬

17. The speed of screen/film can be increased without
increasing noise by:

¬

A. Using a faster film


6. A student technologist has a cadaver on the table and
is practicing manual technique factors for a digital
photospot system that is not phototimed. Circle each of
the following techniques that will result in higher
quantum mottle in the image:

¬

19. Image contrast-to-noise ratio could not be increased
by using:

¬

A. Lower tube voltages
B. Higher-ratio grids
C. Larger x-ray beam areas
D. Screens with lower conversion efficiency
E. Slower films

¬
¬
¬
¬
¬
¬

A. Higher kVp, higher mAs
B. Lower mAs, same kVp
C. Higher mAs, same kVp

¬
¬

¬

A. A calcified lung nodule.
B. A nonnon-calcified lung nodule.
C. Between overlying and underlying tissues.
D. The size of a small fracture.
E. Vessels during the arterial phase of a normal
angiogram.

¬
¬

¬

¬

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

42

43

Sampling (pixel) pitch
Detector aperture width


Example: sampling pitch of
100 µm
FN = 5 cycles/mm
When input f > FN then the
spatial frequency domain signal
at f is aliased down to:
fa = 2F
2FN – f
Not noticeable with patient
Antiscatter grids
Aperture blurring - signal
averaging across the detector
aperture
sin(aπ f)
MTF(f)=FT{rect(a)}=sinc(af)=
aπ f

¬

¬

¬
¬
¬

¬

Brent K. Stewart, PhD, DABMP

ContrastContrast-Detail (C(C-D) Curves


¬

¬
¬

¬
¬
¬
¬
¬

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 287.
47

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., pp. 285285-286.
46

45

Diagnostic task: separate
abnormal from normal
Usually significant overlap in
histograms


¬
¬
¬

Brent K. Stewart, PhD, DABMP

The ROC curve is essentially a
way of analyzing the SNR
associated with a specific
diagnostic task
Az: area under the curve –
concise description of the
diagnostic performance of the
systems (including observers)
being tested
Measure of detectability
Az = 0.5 guessing
Az = 1.0 perfect

c.f. Bushberg, et al. The Essential Physics of Medical
Imaging, 2nd ed., p. 291.
50

49

Brent K. Stewart, PhD, DABMP

Huda 2nd Edition – Chapter 5 – Image Quality



Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

¬
¬
¬
¬

51

Brent K. Stewart, PhD, DABMP

52

13


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

Huda 2nd Edition – Chapter 5 – Image Quality

Davis Notes - Image Quality

¬


¬
¬
¬
¬

Brent K. Stewart, PhD, DABMP

53

Brent K. Stewart, PhD, DABMP

Davis Notes - Image Quality
¬

¬
¬
¬
¬
¬

Davis Notes - Image Quality

27. In Figure 5 showing the
ROC curves, the Y-axis
should be labeled (circle all
that are correct):

¬

¬

Brent K. Stewart, PhD, DABMP

56

14


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

Huda 2nd Edition – Chapter 5 – Image Quality

End of Lecture, Additional Questions Follow
¬

16. Quantum mottle is determined primarily by which of
the following factors?

¬

A. X-ray beam filtration
B. X-ray photons absorbed in screen
C. X-ray photon energy
D. Screen conversion efficiency
E. Screen thickness

¬
¬

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

58

¬

7. Gastrointestinal tract contrast could be improved by all
of the following except:

¬

A. Infusion of barium
B. Reduced tube voltage
C. Increased tube current
D. Increased grid ratio
E. Reduced field size

¬
¬
¬
¬

59

Brent K. Stewart, PhD, DABMP

60


¬

¬

9. Referring to Figure 1 showing
three line spread functions, the
best choices for the axes labels
are ________ for the y-axis and
_________ for the x-axis:

¬

A. Frequency, amplitude
B. Blur distance (mm), frequency
C. Relative amplitude, frequency
D. Relative amplitude, distance
(mm)
E. Distance (mm), relative
amplitude

¬
¬
¬

¬

Brent K. Stewart, PhD, DABMP

61



¬

A. 115 kVp
B. 85 kVp
C. 70 kVp
D. 50 kVp
E. Need technique charts to determine

¬
¬
¬
¬

Brent K. Stewart, PhD, DABMP

Brent K. Stewart, PhD, DABMP

62

63

Brent K. Stewart, PhD, DABMP

64

16


Image Quality – Bushberg Chapter 10

¬
¬

A. A thinner screen
B. A screen-film using a reflective layer in the screens
C. A thinner patient
D. Faster film, same screen

Brent K. Stewart, PhD, DABMP

¬
¬
¬

65

Brent K. Stewart, PhD, DABMP

Huda 2nd Edition – Chapter 5 – Image Quality

Huda 2nd Edition – Chapter 5 – Image Quality

¬

25. In screen/film radiography, raising the kilovolt peak
will increase all of the following except:

¬

13. Which of the following factors would have the least

¬
¬
¬

67

Brent K. Stewart, PhD, DABMP

68

17


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

Huda 2nd Edition – Chapter 5 – Image Quality
¬

Davis Notes - Image Quality
¬

10. Compared with a regular screen, a detail screen of
the same phosphor will have a lower:

¬

¬

¬

¬

D8 Geometric magnification can improve the detection of
high contrast objects. The fundamental limitation on
useful magnification is:

¬

A. Blurring due to focal spot size..
B. Blurring due to removal of the grid..
C. H&D curve of the image receptor..
D. MTF of the image receptor..
E. Size of the image receptor..

¬

A. Higher kVp, higher mAs
B. Lower mAs, same kVp
C. Higher mAs, same kVp
D. Higher kVp, same mAs
E. Lower kVp, lower mAs

Brent K. Stewart, PhD, DABMP

70

Raphex 2001 Diagnostic Question


lp/mm.

¬
¬
¬

71

Brent K. Stewart, PhD, DABMP

72

18


Image Quality – Bushberg Chapter 10

Diagnostic Radiology Imaging Physics Course
16 December 2004

Huda 2nd Edition – Chapter 5 – Image Quality

Raphex 2002 Diagnostic Question
¬

D12 A newly installed bucky radiographic system
produces abdominal images that are of acceptable
density over the spine and progressively lighter toward
both lateral edges of the film. The most likely reason for
this finding is improper:


¬

¬
¬

73

Brent K. Stewart, PhD, DABMP

74

Raphex 2002 Diagnostic Question
¬

D13 The impression of noise in an x-ray image is:

¬

A. Increased by increasing the film speed in a screenfilm cassette.
B. Decreased by increasing the film speed in a screenfilm cassette.
C. Increased by decreasing the focal-spot size.
D. Decreased by decreasing the focal-spot size.
E. Mainly determined by imperfections in the-image
receptor.

¬

¬
¬