Softcopy Display Requirements for Digital Mammography

1
Softcopy Display Requirements for Digital
Mammography

• Bradley Hemminger
• School of Information and Library Science and
  Department of Radiology University of North
  Carolina at Chapel Hill

2
Motivation for Providing Softcopy Reading for
Mammography

• Digital Mammography is increasing (more scanners
  approved, number of procedures is increasing).
• Full advantages of digital mammography not
  accrued without softcopy viewing (multiple
  presentations, dynamic adjustments, e.g.
  intensity windowing, incorporation of computer
  aided detection (CAD), electronic reporting).
• Mammography is the last remaining imaging exam
  that is not filmless.

3
Goals

• Demonstrate softcopy reading is as good or better
  than film reading
• Determine what elements of the workstation (user
  interface and technical constraints) limit its
  capability or reduce acceptance of softcopy
  reading.
• Provide or suggest solutions to overcome these
  limitations.

4
What images?How are they read?
5
Spatial Resolution? 1Meg vs 5Meg vs 20Meg
6
Design of Mammoview

• Study of radiologists (eyetracking, videotape,
  observation, self reporting) indicated that
  screening mammography is composed of two primary
  parts
• Initial review of all images (gestalt overview)
• Careful study of comparison views
• Left vs Right
• New vs Old
• MLO vs CC

7
Computer Human InterfaceSimple mental model to
match film display, minimize interactions all
interaction instantaneous.
Zoom (switch between Overview and Full Res) also
support Panning in Full Res
Comparison View cycler
Contrast Presets
8
Zoom Interaction
9
Comparison Views
10
PresetsDense Default
11
Clinical Trial Methodology

• 63 cases
• 8 radiologists
• Complete reads of all cases in both modalities
  with washout (film and softcopy)
• Measured speed and accuracy of readings (area
  under ROC curve).

12
Clinical Trials Primary ResultsSoftcopy Film

• Speed the same. Average speed slightly faster
  on softcopy, but not statistically significantly
  different.
• Accuracy the same. Area under the ROC curve and
  sensitivity slightly higher for film.
  Specificity slightly higher for softcopy display.
  Overall, film and softcopy not statistically
  significantly different.

13
Can we make softcopy readings faster in real use?

• Automate all the pre-processing on the computer,
  to reduce labor and speed up the workflow.
• Determining and gathering images (from PACS)
• Cropping image to limit to just breast tissue
  (eliminates need for masking)
• Create different processing preset presentations
• Layout in proper viewing locations based on
  hanging protocol preferences of individual
  radiologists.
• Align comparison views

14

• Clinical Trial tested complete package, including
  pre-processing. If pre-processing elements fail
  then softcopy performance may suffer. Particular
  concerns are alignment and contrast.

15
Clinical Trial Preprocessing

• Preprocessing results
• Recognition of the breast, cropping, and image
  layouts all worked 100.
• Alignment worked about 95
• Contrast presets OK about 95
• Despite everything not working 100, the softcopy
  performance was as good as film. Improving
  alignment and contrast to 100 might improve
  softcopy performance.

16
Secondary Analysis

• Based on prior studies (CT and XRAY interfaces)
  and pilot versions of Mammoview, our main concern
  was keeping the interface and interactions
  simple, and making the softcopy zoom and pan
  interaction not significantly worse than film.
• Planned analysis to measure the effect on review
  time of
• Number of user interface interactions
• Size of images (images significantly larger than
  screen size are expected to increase review time
  for softcopy).

17
Review Time (dependent)
Interactions (mediator)
Pan (mediator) Pan Time
Reader categorical
Image Size Extra Panning
Case Complexity
Case categorical
18
Correlation R-squared 0.16
Review Time (dependent)
Interactions 0.0001 (mediator)
Pan (mediator) Pan Time
Reader 0.01 categorical
Image Size 0.05 Extra Panning
Case Complexity
Case categorical
19
Secondary Analysis Results

• Increasing the number of interactions increases
  review time. Best designs have simpler, easier
  interfaces, and are more efficient speed wise.
• The larger the image is than the screen size, the
  longer the relative difference in review time
  (softcopy vs film). Need larger screens or
  improved interaction techniques.
• Review time depends on readers (previously
  demonstrated).

20
Qualitative Evaluations

• Based on experience using multiple versions of
  Mammoview
• From clinical trial structured questionnaires,
  interviews and observations.
• From observations of RSNA InfoRad and technical
  exhibit demonstrations, and pilot studies at UNC.

21
Is Softcopy ready? YES!

• Radiologists interviewed generally said yes.
• Radiologists in clinical trial (with most
  training and experience) all said they would use
  Mammoview today clinically, except one who said
  no because of quality of digital images (not
  because of the softcopy interface).

22
Softcopy Enhancements

• Bring images to your attention better.
• Electronic Markup. Electronic grease pen felt
  superior to film analog. Allows other graphic
  and text annotations, can toggle on/off, and link
  to dictation.
• Faster. Same or slightly faster now. When
  zoompan improved, maybe significantly faster.
  When pre-processing (load times) counted, the
  softcopy is substantially faster.
• Incorporates CAD better. Same display, can
  toggle CAD marks on/off.

23
Further Improvements Needed(11 issues resolved,
4 left)

• The edge of the monitors creates a gap. New flat
  screens like IBM 9Mpixel
• More eye strain with CRTs. Possibly an issue
  with refresh on CRTs is this better or worse on
  flat panels??

10 cm 4
24
Further Improvements

• Panning updates are a tad slow. Possibly resolve
  with 66 Mhz PCI bus (bottleneck is bus update
  rate to video cards).
• Zoompan is worse for larger images. Larger
  spatial resolution displays (8Mpixel CRTs,
  9Mpixel flat panels), or improved zoompan
  interaction techniques.

25
CAD with Softcopy

• Experimentally tested at RSNA 1999-2001. Too
  few readers, varied conditions for quantitative
  study.
• Qualitative reports indicated radiologists felt
  they were more accurate with CAD added
  (especially less experienced readers). Their
  reading times were on average similar with and
  without CAD (using toggled markers, but slower
  with more involved multiple levels of markers).

26
Conclusions

• Softcopy reading is as good as film reading.
• Radiologists would use softcopy reading.
• Softcopy reading incorporating automated
  pre-processing and hanging protocols is faster
  than film readings, and saves labor.
• Higher resolution displays and/or improved
  zoompan would improve softcopy display.
• Softcopy reading with CAD seems likely superior
  to film CAD, and softcopy by itself.

27
Acknowledgements

• Mammography Research Group at UNC (Etta Pisano
  PI) and Radiology Research Laboratory.
• Company support Fischer Imaging Corporation,
  Qualia/CADx, Kodak, Hologic/Lorad.
• NIH, DOD and HHS grant for Grey Scale Image
  Processing for Digital Mammography, and
  Evaluation of Digital Mammography Display.

28
The End

• Show Videotape

29
Screening Mammography

• Structured viewing paradigm
• Several alternative, but pre-defined, viewing
  protocols, well suited to a structured viewing
  presentation format that can be automated on a
  computerized display system.

30
Review Time (dependent)
Interactions (mediator)
Pan (mediator) Number of Pans Pan Time
Reader categorical
Image Size Total area Amt Off screen Extra
Panning
Case Complexity
Case categorical
31
Choice of Measures

• Pan
• Pan time better than pans across the board.
  Expected this, as it more accurately reflects
  time spent panning, which is what most directly
  effects review time.
• Image size
• No significant difference between the three
  options. Choose Extra Pan due to consistently
  slightly higher correlation values.

32
Film Pros Softcopy Cons

• Sometimes the views were not perfectly aligned
  (pre-processing failure). Not a big problem.
• Sometimes the contrast presets were not good.
• The edge of the monitors gets in the way

10 cm 4
33
Film Pros Softcopy Cons

• Liked the feel of shuffling films. Felt more
  involved than pressing buttons on softcopy.
• Like the ability to move films around and flip
  them as desired.
• Felt like there was more eyestrain when using the
  CRT displays for long hours than on film.
• Easier to miss image features using softcopy when
  zoomed in on images significantly larger than the
  display size.

34
Film Pros Softcopy Cons

• Wanted to be able to have more complete
  adjustment of height of monitor.
• Want ability to markup images on softcopy.
• Variable sizing on digital images on is sometimes
  confusing (need digital ruler overlay).
• Panning operation is not quite real time, causing
  swimming/jerkiness which was disliked.
• Panning felt more time consuming on softcopy
  compared to film.

35
Film Pros Softcopy Cons

• When clicking through comparison views would like
  to maintain position instead of restarting at
  default position.
• Load time for the study was too slow (not part of
  studytook 20-60 seconds).
• Would like a fit to screen zoom mode (i.e.
  between overview size and full res size).

36
Softcopy Pros Film Cons

• Second preset (dense breasts) was helpful.
• Faster and easier to do comparison views. They
  are brought to your attention.
• Could focus on images better, didnt have to look
  around as much.
• Automatic preparation and hanging of images was
  good, and saves time.

37
Softcopy Pros Film Cons

• Magnification ability (toggling with button
  click) was better than film.
• Eliminated masking extraneous light problems
  (that occur on lightbox from uncovered areas).
• Interface well designed, easy to use.
• Very clean image view (only breast and labels).

38
Clinical Trial

• The clinical trial found that the speed and
  accuracy of readings using the Mammoview softcopy
  display were not significantly different from
  readings of the same cases by the same observers
  using film display. The softcopy reading
  times, while not statistically significantly
  different, were slightly faster than those for
  film. Accuracy was measured by area under the
  ROC curve. The area under the ROC curve and
  sensitivity were slightly higher for film
  display, while specificity was slightly higher
  for softcopy display. Again, the accuracy
  results were not statistically significantly
  different for the two presentations.

39
Further Improvements

• While we have demonstrated a successful screening
  mammography softcopy display system, radiologists
  also indicated fifteen areas where they felt the
  softcopy display needed improvement or was
  inferior to film.
• Eleven of these issues have been resolved by
  simple improvements to Mammoview.
• Four issues remain, however.

40
Image Size measures
total percentage area area B / area A amount
offscreen area B area A extra panning
length of the arrows, which is equivalent to
lengthB-lengthA widthB-widthA.
41
Secondary Analysis Resultsoverall adjusted R2
correlation was 0.162