Performance for RadiologicalDisplay Devices
Michael Flynn
Dept. of Radiology
RADIOLOGY RESEARCH
Health System
Henry Ford
2Flynn AAPM 2008
iQC Test Pattern (pacsDisplay)
Projection Test Pattern
12 / 0
12 / 0243 / 255
243 / 255
3Flynn AAPM 2008
Intro: visual interpretation
The device used to display radiographic images must effectivelytransfer spatial and contrast information to the human observer.
DETECTION DISPLAY
(A) Subject contrast in the patient is;(B) recorded by the detector and(C) transformed to display values that are(D) and sent to a display device for presentation to(E) the human visual system and interpretation.
4Flynn AAPM 2008
Intro: Visual Requirements
The performance of the human visual system (HVS)is reviewed in relation to display for the primaryinterpretation of radiological images.
A. Viewing Distance
B. Display Size
C. Pixel Size
D. Display Zoom
E. Equivalent Contrast
5Flynn AAPM 2008
A. Viewing Distance?
•Vergence•Accomodation
• Vergence (convergence)allows both eyes to focusthe object at the sameplace on the retina.
• The closer the object,the more the extraocularmuscles converge theeyes inward towards thenose.
6Flynn AAPM 2008
A. Viewing distance and vergence
Resting Point of Vergence
The eyes have a resting point of vergence of about 40inches.(Jaschcinsk-Kruza 1991).
– Objects closer than the resting point cause muscle strain.
– The closer the distance, the greater the strain (Collins 1975).
Every one of the subjects studied by Jaschinski-Kruza(1998) judged the eye to screen distance of 20 inchesto be too close. All accepted a 40 inch distance.
Grandjean (1983) reported an average preferredviewing distance of 30 inches.
Arms length viewing distance
7Flynn AAPM 2008
A. Viewing distance and accomodation
Resting Point of Accommodation
The ciliary muscle changes the shapeof the lens to focus the object.
– The eyes have a resting point ofaccommodation which is thedistance that the eye focuses towhen there is nothing to look at(Owens 1984).
– This resting point averages about31 inches (Krueger 1984).
Prolonged viewing of a monitor closer than the restingpoint of accommodation increases eye strain (Jaschinski-Kruza 1988). The ciliary muscle must work 2.5 timesharder to focus on a monitor 12 inches away than it doesto focus at 30 inches.
Arms length viewing distance
8Flynn AAPM 2008
B. Display Size?Field of view in relationto viewing distance.
9Flynn AAPM 2008
Rod receptors have highsensitivity, gray response,and interconnections thatrespond to motion.
The retina contains a largenumber of rod receptors(160 M) distributed overthe peripheral field.
B. HVS: peripheral response
10Flynn AAPM 2008
B. Display Size vs Viewing Distance
Diagonal SizeViewing DistanceTask
110.1 inches3 metersTeaching Conference
31.5 inches1 meterConsultation viewing
20.8 inches2/3 meterNormal viewing
10.4 inches1/3 meterClose Inspection
For a specific viewing distance the diagonal dimensionshould be about 80% of the viewing distance. (44o)
11Flynn AAPM 2008
B. Field of View
21 inch (diagonal) monitors with a field of 32 x 42 cmprovide an effective field for radiographic imagesviewed at a normal distance (2/3 m).
Eyeglasslensshould beoptimizedfor anormalviewingdistance
12Flynn AAPM 2008
C. Pixel Size?
•Visual Acuity•Contrast Sensitivity
13Flynn AAPM 2008
C. Visual Acuity
A variety of test patterns are used to assess visualacuity. Clinical measures are done typically with aSnellen eye chart. Much psychovisual research hasbeen done using sinusoidally modulated test targets.
14Flynn AAPM 2008
C. HVS: Retinal anatomy
The retina of the human eye contains a network ofrods and cones interconnected by neural cells.
15Flynn AAPM 2008
Particularly thin cones (2 m)are densely packed in thecentral 50 microns of the foveacentralis. They provide highdetail color response.
At 60 cm, 1 degree correspondsto a 1 cm field of view. Thisarea is focused on a 288 micronregion of the retina, the fovea
C. HVS: Foveal response
16Flynn AAPM 2008
C. Contrast Sensitivity as a measure of spatial acuity
Contrast sensitivity is the inverse of contrast threshold: Cs = 1/Ct
~2.5 c/mm
10% max
~0.5 c/mm
17Flynn AAPM 2008
• The eye perceives luminance variations as a changewith respect to viewing angle.
• Data on visual performance must always be converted fromcycles/degree to cycles/mm at a specified viewing distance.
Cycles/mm = 57.3 x (cycles/degree) / (viewing distance, mm)
C. Spatial Frequency: cycles/degree
mm
18Flynn AAPM 2008
C. Pixel Size at Maximum Spatial Acuity
The visual spatial frequency limit and associated pixel size canbe defined as that for which Cs = 10% of maximum.
The pixel size of a display system that matches the resolvingpower of the human eye depends on the observation distance.
Distance frequency pixel size
Close inspection 5 cycles/mm 0.100 mm/pixel
(0.33 m)
Normal viewing 2.5 cycles/mm 0.200 mm/pixel
(0.66 m)
Consultation view 1.7 cycles/mm 0.300 mm/pixel
(1.00 m)
Conference room 0.5 cycles/mm 1.000 mm/pixel
(3.00 m)
19Flynn AAPM 2008
C. Pixel array and Megapixels
The pixel size and the field of view dictate the pixelarray size and the total number of pixels.
Megapixels alone is not a good descriptor of quality.
Field of View pixel size array size MegaPixels
21 inch 0.100 mm 3200 x 4200 13.4
21 inch 0.200 mm 1600 x 2100 3.4
• idtech 3 MP panel20.8 inch (32 x 42 cm) 3.1 megapixels (.207 mm pixels)
20Flynn AAPM 2008
C. LCD 2MP Colot Pixel
LCD Pixel Structure
For a pixel pitch greater than ~200 microns, the pixelstructure is visible as a granular pattern.
Some consumer monitors have a granular diffusingsurface that creates a random noise pattern.
Dual Domain pixel structure Single Domain pixel structure
21Flynn AAPM 2008
D. Display Zoom?
Detector Detail in relation toDisplay Acuity
22Flynn AAPM 2008
D. Viewing distance and image zoom
Use of image zoom features is ergonomically better thanleaning forward for close inspection.
Split deck tables with a broad front deck usefully prohibitclose inspection with 3 MP monitors.
23Flynn AAPM 2008
D. Magnification / Minification
Minification has value byincreasing the frequency ofdiffuse structures.
1X
1/4X4X
1X
Zoom is needed to display detail atthe detector pixel level with goodcontrast sensitivity.
24Flynn AAPM 2008
D. True Size
For some applications, “true size”display is important.
– Comparison of current andprior exams obtained ondifferent detectors (or withscreen-film).
– Orthopedic assessment ofsize.
This requires knowledge of
– Detector element (del) pitch
– Display element (pixel) pitch.
Prior
Current
* adapted from D. Clunie, SCAR 2005
25Flynn AAPM 2008
D. Re-sampling for Display
A subset of imagevalues is re-sampledfor presentation on adisplay device.
In General;
• The detector and displaypixel spacings are different.
• The detector and displayoverall size are different
DETECTOR
DISPLAY
26Flynn AAPM 2008
D. Up-sampling (magnification)
Up sampling occurswhen the number ofdisplay values in theregion re-sampled ismore than the numberof recorded imagevalues .
This is commonlyencountered whendisplaying CT and MRimages.
• Blue circles show an 11x11array of recorded imagepixel values.
• Green solid circles are for a15 x 15 array of displaypixel values
27Flynn AAPM 2008
D. down-sampling (minification)
Down sampling occurswhen the number ofdisplay values in theregion re-sampled isless than the numberof recorded imagevalues .
This is commonlyencountered when afull radiograph isdisplayed.
• Blue circles show an 11x11array of recorded imagepixel values.
• Green solid circles show a7 x 7 array of display values.
28Flynn AAPM 2008
D. Interpolation
Estimation of variably spaced displayvalues from a set of image values is doneusing mathematical interpolation methods.
29Flynn AAPM 2008
Bi-Linear Interpolation
• Image values pairs above & below thedisplay value are linearly interpolatedbased on the column position (black).
• These values are linearly interpolatedbased on the row position.
D. Approximate Interpolation
While fast, nearest neighbor and bi-linear interpolation do notresult in optimal image quality due to artifacts and blur.
Nearest Neighbor Interpolation
• Display value (green) is taken as theimage value (blue) at the nearest rowand column.
• Produces visible block artifacts forlarge magnification.
30Flynn AAPM 2008
D. Improved Interpolation
Improved quality can be achieved byestimating display values from theclosest 16 image values (4 x 4).
–– SplineSpline interpolationinterpolation uses polynomialarc segments constrained to besmooth (1st and 2nd derivatives) attransition points (nodes). It hasbeen classically used for digitalimages.
– A still popular technique known ascubic convolutioncubic convolution involves the use ofa sinc-like kernels composed ofpiecewise cubic polynomials.
– Recent work has shown thatgeneralizedgeneralized splinespline interpolationinterpolation usinga pre-filter operation providesexcellent performance with fastimplementation and can providecontrolled smoothing.
Cubic Interpolation
• Display value (green) is computedfrom the closest 16 image values.
• The weighting functions for the 16image values are intended to estimatea continuous function within the spacebetween the sampled values.
31Flynn AAPM 2008
D. Magnification / Minification
Magnification: Calcified duct, 4:1 re-sampling 5.25 x 5.25 mm region
Nearest Neighbor
A
Bi-Linear
B
Cubic
C
Minification.
• Advanced interpolation methods can also provide effectiveminification with noise reduction (low-pass filter).
• Alternatively, minification is often done using multi-scalerepresentations of the image with progressive presentation.
32Flynn AAPM 2008
D – Display Interpolation – key points
Interpolation and Image Quality:
The numerical approach used to obtained magnifieddisplay values has significant impact on image quality.
Modern interpolation with good performance needsoptimal implementation for high speed.
Minification and noise reduction:
Minification should be done such that high frequencynoise (quantum mottle) is reduced.
Multi-scale representation of image date provides ameans for minification (JPEG 2000, JPIP, Wavelet, ..).
33Flynn AAPM 2008
E. Equivalent Contrast?
• Grayscale response• Luminance ratio (L’max/L’min)
34Flynn AAPM 2008
E. Contrast detection in relation to brightness
• Contrast detection is diminished for images with low brightness.
• Extensive experimental models have documented the dependenceof contrast detection on luminance, spatial frequency, orientationand other factors. The empirical models of either S. Daly or J.Barton provide useful descriptions of this experimental data.
35Flynn AAPM 2008
E. Contrast threshold vs luminance
The Barton model describes the average contrastthreshold of normal observers. Significant differencesexist for individual observers for different test methods
36Flynn AAPM 2008
E. DICOM graylscale display standard
DICOM part 3.14 describes a grayscale response thatcompensates for visual deficits at low brightness
37Flynn AAPM 2008
E. Fixed versus variable adaptation
Contrast threshold for varied visual adaptation (A, Flynn 1999b) and fixed(B) visual adaptation: The contrast threshold, L/L, for a just noticeabledifference (JND) depends on whether the observer has fixed (B) or varied(A) adaptation to the light and dark regions of an overall scene.
38Flynn AAPM 2008
E – Ct for small sinusoidal patterns on a color LCD.
2AFCassessment ofCt using variedbackgroundregionbrightness.
SINE and ADAPT Contrast Thresholds Normalized to SINE
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0.1 1 10 100
L/L_SINE
Re
lati
ve
CT
/CB
M
DB
DP
MF
MP
PT
PR-80
SL
AVERAGE
Effects of adaptation on observers contrastthresholds relative to changes in background.
39Flynn AAPM 2008
E. Adapted Observer Performance
Observer performance is best when visual system isadapted to the average scene luminance.
A B C
40Flynn AAPM 2008
E. Effect of Lmax/Lmin
Digital radiographsshould be displayedusing over aluminance range of250-350:1.
Images preparedfor range of 250that are display ona monitor with largerange will havepoorly perceivedcontrast in darkregions.
250:1650:1
250:1 .1 to 2.50 OD350:1 .1 to 2.65 OD650:1 .1 to 2.90 OD
41Flynn AAPM 2008
E. LR for LCD monitors
For CRT monitors, LR is set by adjustingbrightness (Lmin) and contrast (Lmax).
For LCD devices, only the backlightintensity can be adjusted.
For LCD devices
– Lmax is set by adjusting the backlightbrightness (current control).
– Lmin is set as a part of the grayscale calibration(starting LUT value).
42Flynn AAPM 2008
Other issues
Issues that I have not addressed!
• LCD devices have significant contrast changeswhen viewed at angles oblique to the surface.
• Note: New OLED technologies promise toeliminate that problem in near future.
• Pixel noise is poorly documented for new LCDmonitors. Further works needs to be done tounderstand whether pixel noise effects diagnosticvisual performance.
• 256 (8bit) versus 1024 (10bit) gray levels.
43Flynn AAPM 2008
Questions?