influence of monitor luminance & tone scale on observers’ search & dwell patterns
TRANSCRIPT
Influence of Monitor Luminance & Tone Scale on Observers’ Search & Dwell Patterns
Elizabeth Krupinski, PhD & Hans Roehrig, PhD
University of Arizona
This Work Was Supported in Part by:
Toshiba Medical Imaging Tokyo, Japan DataRay Corp. Westminster, CO
Rationale Full-field digital mammography
systems will soon replace traditional film acquisition systems. To maximize the benefits of digital mammography (e.g., image processing, CAD), we need to understand and maximize the properties of CRT monitors used to view these digital images.
Purpose The goal of this study was to
measure the influence of display luminance and tone scale choice on visual search behaviors. The results of this study should be helpful in establishing minimally acceptable viewing conditions for viewing radiographs on CRT monitor displays.
Images
50 Pairs Of Mammograms
- CC & MLO views - right or left breast- 18 mass- 18 microcalcification
cluster- 14 lesion-free- All lesions malignant- Digitized 80 microns, 12
bits
Display Monitors
2 DataRay DR110 monitors- Portrait Mode- Size: 21” FS- Deflection Angle: 90 deg- Active Area: 11.5” x 15.5”- Phosphor: P-45- Bulb Transmission: 52%- Panel: 92% and 62%- Resolution: 1728 x 2304- Refresh rate: 70 Hz
80 ftL 140 ftL
Characteristic Curves
0.00 100.00 200.00 300.00Command Level (ADU)
0.00
40.00
80.00
120.00
Lum
inan
ce (f
t - L
)
Monitor 1 (140 ftL)
Default - - -Barten ------
Monitor 2 (80 ftL)
- - - Default------ Barten
Two characteristiccurves were selectedfor the study.
1) The Barten curve is a perceptually linearized tone scale.
2) The Default curve isa non-perceptuallylinearized tone scale(set using SMPTE pattern).
Display Luminance
100.00 200.00 300.00 400.00
Beam Current [micro-A]
0.00
40.00
80.00
120.00
160.00
Out
put L
umin
ance
[ft-
L] Monitor 1 (140 ftL)
Monitor 2 (80 ftL)
Output Luminance versus Beam Current forData Ray Monitor DR 110 with P45 Phosphorand 92 % Panel (Monitor 1) and 62 % Panel(Monitor 2)The two monitors were
identical in every wayexcept for the frontpanel.
1) Monitor 1 had a baseluminance of 140 ftL.
2) Monitor 2 had a baseluminance of 80 ftL.
Procedure 2 studies
- changing luminance (140 ftL vs 80 ftL) - changing tone scale (Barten vs Default) 6 radiologists in each study Counterbalanced randomized design - Each image seen twice - Once in each condition of the study - Minimal 2 week separation between
sessions Unlimited viewing time
Observers’ Task
Determine if a lesion is present or absent Indicate if lesion is mass /
microcalcification Rate confidence in decision on 6-point
scale
1 = no lesion, definite
6 = lesion present, definite Indicate lesion location: could indicate
more than one location Detection only, no classification
Eye-Position Recording Eye-position was recorded for each case ASL 4000SU Eye-Tracker was used Records head movement so observers can
move while viewing images
Accuracy ~ 1 deg Data provides:
- Decision dwell times
- Number of fixations or clusters
of fixations landing on image areas
- Scan patterns
Typical Search Pattern
Performance Analyses The confidence data were
submitted to Alternative Free Response Receiver Operating Characteristic Analysis (AFROC) and Area Under the Curve (A1) values were calculated
These data were reported in full at the 1998 SPIE Medical Imaging Conference
Search Analyses
Eye-position data were analyzed to determine:- Total viewing times- Median decision dwell times- Number of fixation clusters generated on the mass, microcalcification and normal (lesion-free) cases
Detection Performance*
Barten curve = .9720 Default curve = .9511
t = 5.423, df = 5, p = .0029
80 ftL monitor = .9594 140 ftL monitor = .9695
t = 1.685, df = 5, p = .1528
* Average area under the AFROC curve (A1)
Total View Times - Tone Scale
0
10
20
30
40
50
60
70
80V
iew
ing
Tim
e (s
ec)
Barten Default
MassMicrocalcificationLesion-Free
Total View Times - Luminance
0
10
20
30
40
50
60
70
80
90V
iew
ing
Tim
e (s
ec)
140 ftL 80 ftL
MassMicrocalcificationLesion-Free*
* Lesion-free significantly longer than mass or microcalcification
Decision Dwells - Tone Scale
0
500
1000
1500
2000
2500
3000
3500M
edia
n D
wel
ls (
mse
c)
TP FN FP TN
BartenDefault
Decision Dwells - Luminance
0
500
1000
1500
2000
2500
3000
3500M
edia
n D
wel
ls (
mse
c)
TP FN FP TN*
140 ftL80 ftL
* 2 = 4.08 df = 1 p < 0.05
Number Of Fixation Clusters Per Case - Tone Scale
0
2
4
6
8
10
12
14
16M
ean
# of
Clu
ster
s
Mass Calc Normal
BartenDefault
Number Of Fixation Clusters Per Case - Luminance
0
2
4
6
8
10
12
14
16
18M
ean
# of
Clu
ster
s
Mass Calc Normal
140 ftL80 ftL
*
* t = 2.83 df = 166 p < 0.01
Conclusions A perceptually linearized display
(e.g., using the Barten curve) yields significantly better detection performance than a non-perceptually linearized display
A 140 ftL display yields marginally better detection performance than an 80 ftL display
Conclusions Visual search is more efficient
with the perceptually linearized and higher luminance displays
Total view and decision dwells were consistently shorter and fewer clusters were generated on all types of cases with these displays - especially the true-negative decisions and lesion-free images
Possible Explanation Lower luminance and non-linearized displays seem to
affect most the search and evaluation of normal image areas - TN dwell times are longer and more clusters are generated on lesion-free cases.
These parameters reflect an increase in information processing operations, even though the same decisions are reached as with the higher luminance and linearized displays.
Recognizing normal variations in images may be more difficult with non-optimal displays, so radiologists must alter their search behaviors to compensate for display short-comings.
Conclusions & Recommendations
CRT monitor luminance and tone scale can affect diagnostic accuracy and visual search efficiency in significant ways.
For the display of digital mammograms on CRT monitors, a relatively high luminance monitor should be used and the display should be perceptually linearized.