AAPM 2005-Continuing Education Course in
Radiographic and Fluoroscopy Physics and Technology
A Comparison of Screen/Film and Digital Imaging:
Image Processing, Image Quality, and Dose
Ralph Schaetzing, Ph.D.
Agfa Corporation
Greenville, SC
This presentation…
O
does…
O
O
focus on salient characteristics of two projection-radiography
acquisition technology classes
O
Analog: screen/film (S/F)
O
Digital: generic
does not…
O
focus on technology details (see other presentations)
O
O
O
address alternatives to projection imaging (i.e., x-sectional)
cheerlead
O
1
Specific technologies used as examples only
No technology-class recommendations - too many other issues
Digital Image Acquisition Technologies
Direct
X-ray quanta
Ð
meas. output signal
Silicon Strip
+
line/slot scan
Photoconductor
+
flat-panel array
Pressurized Gas
+
line/slot scan
Indirect
Ð
Ð
Screen/Film
+
point scan
Storage Phosphor
+
point scan
Screen/Film
+
line scan
Storage Phosphor
+
line scan
Scintillator
+
line/slot scan
Scintillator
+
area sensor(s)
Commonly called Digital Radiography (DR)
2
X-ray quanta
intermediate(s)
meas. output signal
Commonly called
Computed
Radiography
(CR)
Scintillator
+
flat-panel array
The Medical Imaging Chain (Analog, Digital)
Human
Acquire
ANALOGUE WORLD
(visible)
Reproduce
Preprocess
Process
Process
(Optimize)
Process for Output
Store
Distribute
3
DIGITAL WORLD
(hidden)
The Bigger Picture
Producers/Competitors
Technical
c
A
Re
p
q
r
o
u
S D e
i Pr t o r s s d
r o ce u
c
e
e
Operational
Observers
Regulators/Standardizers
Clinical
Economic
Consumers
4
Imaging and Information System Environment
Outline of Presentation
O
Dose and Image Quality in S/F and Digital Systems
O
O
O
O
Image Processing in S/F and Digital Systems
O
O
O
5
Dose requirements: the speed limit
Image quality as a dose metric
Does image quality really matter?
Image processing for display optimization
Image processing for decision support
Wrap-up and Conclusions
Outline of Presentation
O
Dose and Image Quality in S/F and Digital Systems
O
O
O
O
Image Processing in S/F and Digital Systems
O
O
O
6
Dose requirements: the speed limit
Image quality as a dose metric
Does image quality really matter?
Image processing for display optimization
Image processing for decision support
Wrap-up and Conclusions
When x-ray dose wasn’t so important…
All Images ©
Radiology Centennial, Inc.
7
When x-ray dose became important…
O
Biological effects became clearer quickly
© Radiology Centennial, Inc.
O
O
8
Dose: a metric by which to compare systems
(and facilities – e.g., FDA’s NEXT Survey)
ALARA Principle: As Low As Reasonably Achievable
Comparing Dose Requirements:
Speed
O
Analog S/F
O
ISO 9236-1 (2004)
Photography -- Sensitometry
of screen/film systems for
medical radiography -- Part
1: Determination of
sensitometric curve shape,
speed and average gradient
1000
S=
K s ( µGy )
O
9
Speed defined by incident air
kerma (Ks) giving net density of
1.0 under specific conditions
(exposure, processing, etc.)
O
Digital
O
O
O
Linear, wide-latitude
response, and variable
detector kV-dependence
makes definition non-trivial,
manufacturer-dependent
Confusion and frustration
Efforts underway to create
standardized definition of
speed (AAPM, DIN, IEC,
manufacturers)
Comparing Dose Requirements:
Speed
X-ray Sensitometry - Screen/Film and CR
O
Linear, wide-latitude response
(Density or CR Signal vs. X-ray Exposure)
of digital systems still confused 4
S/F
S/F
S/F
with dose efficiency
S=1200 S=400 S=300
Latitude ≠ Dose Reduction!
(Can always find S/F system 3
1200
that operates at same dose
CR
level as digital system)
"pick a speed"
Multiple, different S/F systems 2
needed to cover same exposure
range as one digital system
1
Question: even if signal levels
matched, is S/F image quality
(e.g., sharpness, noise) level
300
0
comparable to that of digital?
0.1
1.0
10.0 100.0 1000.0 µGy
4 decades of exposure
O
O
O
10
Comparing Dose Requirements:
Detective Quantum Efficiency (DQE)
Input
Exposure
Signalin
Noisein
Signalout
Detector
Noiseout
Recorded
Image
(Noiseout > Noisein)
DQE =
(Measured) Noise from ideal detector*
Measured Noise from real detector*
*Noise values must be expressed in same units
=
DQE ∝
11
1.0 for ideal detector
(Contrast or Gain)2 x (Sharpness)2
Measured Noise
spatial-frequency- and exposure-dependent, i.e., a SURFACE!
Comparing Dose Requirements:
Reported DQE(0) Values for S/F and Digital
Detective Quantum Efficiency
(@ f = 0 cy/mm)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
12
Ideal Detector
R&D
Needle scint.
+ TFT
Powder scint.
+ TFT
Powder scint.
+CCD
R&D
Needle IP-CR
Powder IP-CR
(dual-sided)
Photoconductor
+ TFT (gen. rad.)
Powder IP-CR
Screen/Film
X-ray film
Indirect
Direct
Comparing Dose Requirements:
DQE: Caveat Emptor…
O
Standard exists…
O
IEC 62220-1 (2003)
Medical electrical equipment
- Characteristics of digital Xray imaging devices - Part 1:
Determination of the
detective quantum efficiency
O
But…
O
13
Make sure that standard was
followed!
My DQE is bigger
than your DQE!
Which image has the highest quality?
A
C
B
D
14
Image Quality: What is it?
O
O
O
15
Image quality depends only on intrinsic, objective physical
characteristics of an imaging system, and can be measured
independently of an observer
Image quality is whatever the observer says it is
(i.e., it is a subjective perception of the image,
"in the eye of the beholder")
Image quality is defined by an observer's ability to achieve
an acceptable level of performance for a specified task
Image Quality: What is it?
Perf.
Performance
Measures
?
Subj.
?
Diagnostic Accuracy
Diagnostic Error Rate
Patient Management
Patient Outcome
Subjective
Measures
Perf.
?
Obj.
Objective
Measures
Subj.
Perception
Confidence
Preference
16
Physical
Characteristics
Obj.
Radiographic Image Quality: Does it matter?
Evolution of Projection Photon Counting
Radiography and DR (needle scint., TFT)
Image Quality
DR (photocond., TFT)
Evolution of Radiologist Error Rate
(unaided)
DR (scint., CCD)
CR (needle IP, line scan)
CR (powder IP, point scan)
DR (II, CCD)
DIGITAL
1895
ANALOG
2005
II/TV (Fluoroscopy)
S/F (UV)
S/F (Green)
S/F (Blue)
Direct Film Exposure (no screen)
17
Significant image quality changes!
Significant dose changes: 50-100x
1895
2005
Very little change in performance…
Radiographic Image Quality: Does it matter?
O
O
Imaging system manufacturers
are wasting their time trying to
maintain or improve image
quality because end users can't
or don't use it clinically
Limiting factor: radiologist
Radiologists are so skilled that
they can locate the relevant
clinical information consistently,
independent of image quality
(at least, to date)
Limiting factor: imaging system
Evolution of Radiologist Error Rate
(unaided)
If the imaging system
is the limiting step…
New acquisition, image processing
systems with unprecedented
objective image quality
?
If the radiologist
is the limiting step…
Decision-support systems:
Computer-Aided Detection/Diagnosis
1895
18
2005
Outline of Presentation
O
Dose and Image Quality in S/F and Digital Systems
O
O
O
O
Image Processing in S/F and Digital Systems
O
O
O
19
Dose requirements: the speed limit
Image quality as a dose metric
Does image quality really matter?
Image processing for display optimization
Image processing for decision support
Wrap-up and Conclusions
Medical Image Processing – Many Goals
O
Optimize for human vision
O
Signal contrast
O
O
O
O
O
O
20
Latitude
Dynamic range
(acquired vs. displayed)
Sharpness
Noise
Optimize for machine vision
O
O
O
Optimize for storage
O
Memory requirements
O
Storage media requirements
O
Cost
Optimize for distribution
O
O
CAD
CADx
Network efficiency
(bandwidth)
Transmission/delivery time
(e.g., teleradiology)
O
Access time (read/write)
O
Cost
Image Processing for Display:
Optimizing for Human Vision
Can you find the
20 “blob” targets
in this image?
Agfa MUSICATM
21
MUSICA = MUlti-Scale Image Contrast Amplification
Image Processing for Display:
Optimizing for Human Vision in S/F
O
O
Detector choices
O
Screen
O
Film
O
Chemical processing choices
O
Chemistry type, activity,
seasoning, replenishment
O
Time
Technique (kVp, mA, s)
O
Temperature
Hypersensitization
O
Adjacency effect
Exposure choices
O
O
O
22
O
Uniform pre-exposure to
light nucleates sub-image
that becomes developable
with further exposure
Improves speed, SNR
O
O
Edge enhancement (unsharp
masking!) through control of
developer dynamics
Viewing choices
O
Viewbox luminance, quality,
spectrum, masking, magnif.
Image Processing for Display:
Optimizing for Human Vision in Digital
O
Wide latitude + post-processing
flexibility of most digital systems
enables HUGE variety of
optimization techniques
O
O
O
O
Manual vs. …
Semi-automatic (examdependent) processing vs. …
O
Contrast (Gradation/Tone Scale)
O
Look-up Tables (LUTs)
O
Histogram-based
O
Contrast/Latitude dilemma
Sharpness (Edge Enhancement)
O
Fully automatic (hands-free)
processing
O
O
Multi-scale techniques
(custom contrast at each of
a set of frequency bands)
Noise (Reduction/Smoothing)
O
23
Unsharp masking
Used sparingly
(?effect on signals?)
Image Processing for Decision Support:
Optimizing for Machine Vision (CAD)
Computers can be trained to do this well
and consistently (for well-defined targets)
O
O
Search/Scanning:
15%
O
Recognition:
37%
O
Perceptual Decision:
O
Interpretive Decision:
CAD
alone
Radiologist
with CAD
Missed
Missed
Marked
Detected
Missed
48%
Performance evaluation nontrivial
O
Pathology/target-dependent
O
Operating point? (TP, FP)
O
Radiologist
alone
Best schemes can outperform
radiologists (target-dependent)
Detected
Missed
The Theory of CAD
24
*Kundel H, Visual search, object recognition and reader error in radiology,
Proc. SPIE 5372 (2004), pp1-11 (SPIE Press, Bellingham WA)
Total Number of Cases
O
Sources of radiologist error (chest*)
Image Processing for Decision Support:
CAD Implementation Issues
O
Detection performance sensitive
to acquisition system details
O
Contrast resolution
O
O
O
O
O
Spatial resolution
(e.g., pixel size)
Noise
Operational issues
O
O
25
Wide/narrow latitude
Linear/nonlinear response
S/F: extra digitization step
Consistency of results?
O
Proc. protocol (image vs. study)
O
Testing/Scoring/Training (“truth”)
O
Productivity/Workflow/Recalls
O
Case load (screening?, volume?)
O
Satisfaction of Search
(faith vs. blind faith)
O
Observer experience/knowledge
O
Clinical Efficacy
O
Radiologist Acceptance
O
Cost/reimbursement
Outline of Presentation
O
Dose and Image Quality in S/F and Digital Systems
O
O
O
O
Image Processing in S/F and Digital Systems
O
O
O
26
Dose requirements: the speed limit
Image quality as a dose metric
Does image quality really matter?
Image processing for display optimization
Image processing for decision support
Wrap-up and Conclusions
Analog or Digital Technologies – Quo vadis?
S/F
or
Digital
( )
Dose
Image Quality
( )
( )
Physical
Subjective
Performance-based
Image Processing
Display Optimization
27
Decision Support (e.g., CAD)
S/F
S/F
Digital
Digital
Back to the Bigger Picture:
The Path to System Equilibrium…
Convenience
Producers/Competitors
Technical
c
A
Re
q
r
o
u
S D e
i Pr t o r s s d
r o ce u
c
e
e
Operational
Interfaces
28
(e.g., EPR)
p
Consumers
Observers
Flexibility
Regulators/Standardizers
Clinical
Consistency
Control
Economic
Technolust
Thank You
for Your Attention