Modeling of contentbased image retrieval tools for dermatopathology applications

1 / 41
About This Presentation
Title:

Modeling of contentbased image retrieval tools for dermatopathology applications

Description:

Modeling of contentbased image retrieval tools for dermatopathology applications –

Number of Views:52
Avg rating:3.0/5.0
Slides: 42
Provided by: drazenmju
Category:

less

Transcript and Presenter's Notes

Title: Modeling of contentbased image retrieval tools for dermatopathology applications


1
Modeling of content-based image retrieval tools
for dermatopathology applications
"Diamond" software
2
The Team
  • Simon King
  • Casey Helfrich
  • Anu Melville
  • Adil Wali
  • Rahul Sukthankar
  • Mahadev Satyanayaranan
  • Drazen M. Jukic

3
What is WSI (whole slide imaging)?
  • Relatively novel technology (depends on whom you
    ask)
  • Virtual Slide
  • Utilizes robotic scanning of glass microscopic
    slide (GMS)
  • Creates an image of the entire GMS
  • Slides can be handled either manually (one by
    one) or automatically (autoloader).

4
(No Transcript)
5
Why Virtual Slide
  • Drazen (Dr. Jukic) is working in informatics, he
    will make all our work and diagnostics go to
    India actual quote
  • I have heard similar statement multiple times!
  • Answer is simple NO

6
Now You have a Virtual Slide!!
  • One has all of these images and how is this
    better than old-fashioned kodachromes?
  • How do you find images in a jumble on your hard
    drive?
  • Ouch where is Quincy when you need him?

7
Now You have a Virtual Slide!!
  • How do you Google images?
  • No-how if the images are not annotated.
  • Diamond can help here.

8
Open Diamond
  • Developed as a joint venture of Carnegie Mellon
    University and Intel Research Labs
  • It will allow for efficient and rapid searching
    of large image sets for features that would help
    create knowledge out of image features "locked
    away" in digital imaging archives in
    dermatopathology.
  • Diamond's storage architecture for "early
    discard"

9
Early discard
10
Discard?
  • How many of you went beyond page 2 or 3 of Google
    results?
  • Goal is to develop an application that will
  • Enable non-experts to be more effective in
    diagnosing particular disease.
  • Enable new more efficient diagnostic procedures
    that incorporate vision technology.
  • Two arms radiology and pathology

11
PathFind Example
12
PathFind Example
13
PathFind Example
14
Design of Application
  • How do you get an ideal application?
  • You interview numerous pathologists
  • Informatics savvy and Informatics hostile
  • Who (what) is a pathologist?
  • Somebody who pushes little pieces of glass
    incredibly fast across a black square while
    looking at the color variations
  • Faster your hand and eyes are, better TAT you
    have
  • The speed is based on the motoric and ocular
    skill of the individual, as well as ability to
    maintain focused concentration.
  • Somebody who mostly wants to be left alone.

15
PathFind - Concepts
View
Search
Compare
Diagnose
Currently
Slow, non-intuitive interfaces taking longer than
a microscope.
With PathFind
Slide rotation to simulate true slide driving.
Why its better?
  • PathFind interaction is closer to native
    microscope interaction.
  • Lowers learning curve, increasing user adoption.
  • Increases interaction speed, enhancing
    efficiency.

16
PathFind - Concepts
View
Search
Compare
Diagnose
Currently
Trial and error searching through atlases,
textbooks, and journals.
With OnPath
Highlight area of image and click search.
Why its better?
  • Digital workflow.
  • Faster results.
  • Larger data set.
  • Multiple similar items for better decision
    making.

17
PathFind - Concepts
View
Search
Compare
Diagnose
Currently
Compare static journal or atlas image to the
screen via memory.
With PathFind
Side-by-side virtual slide and synced
manipulation with lock-in feature.
Why its better?
  • Side-by-side view enables more direct
    comparison.
  • Image manipulation for more detailed comparison.
  • Direct access to referring experts/pathologists.

18
PathFind - Concepts
View
Search
Compare
Diagnose
Currently
Diagnosis is made based on features of the image.
With PathFind
Diagnosis further substantiated with references
to similar cases.
Why its better?
  • Decision making process is recorded.
  • Higher accuracy because of explicit comparison
    to all case information.
  • Stronger legal defense, decreasing malpractice
    costs.

19
View
20
(No Transcript)
21
(No Transcript)
22
Search
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
Compare
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
Diagnose
34
(No Transcript)
35
(No Transcript)
36
Why Diamond?
37
Virtual Slide Segue
  • Virtual slides promise to deliver new level of
    care in digital pathology
  • Important to outline that today, an application
    is defined by ease of use and speed of
    acceptance
  • If we do not deliver an improvement over the
    microscope, the technology will not be adopted.

38
Parting Thoughts
  • The promised improvements over the current
    glass-based pathology have to be significant
    enough to entice pathologist for a trade-off
  • Glass based pathology
  • Training 5 years plus of pushing glass
  • 5 years plus of hand/eye coordination
  • Millions of little glass pieces pushed through
  • Digital pathology
  • Couple of hundreds (at best)
  • 5 days plus of training
  • Clear-cut opinions

39
Parting Thoughts
  • Our past and ongoing studies
  • Confirmed that speed of digital glass pushing
    is very important
  • Network speed (connection speed) is extremely
    important
  • Slide scanning speed something pathologist
    doesnt see.
  • If additional features are not incorporated, VS
    has long wait to be implemented
  • I would invite additional features for
    incorporation

40
300 years
41
(No Transcript)
Write a Comment
User Comments (0)
About PowerShow.com