Transcript Computer Vision and Image Analysis

Vision and Image Analysis Group (VIA)
Anthony P. Reeves
School of Electrical and Computer Engineering
Cornell University
© A. P. Reeves 2007
VIA Group Research Program
Mission:
A. Accurate measurements from multidimensional images
B. Computer methods for image analysis
Collaborative programs:
• Web-based collaborative research in clinical
studies with Weill Medical College and over
40 medical institutions world-wide.
(computer-aided diagnosis)
• Web-based collaborative programs in
Biological research involving image analysis
Web-based image management
VIA Image
analysis
research
Experimental
Biology
Research
ELCAP Clinical
Studies and
Drug Trials
…
Individual Research Projects
(Individual protocols)
ELCAP remote sites
(Standardized Protocols)
Clinical Studies
© A. P. Reeves 2007
Early Lung Cancer Action Program (ELCAP)
ELCAP Management System (EMS)
www.ielcap.org
Goals:
• To facilitate data pooling between different cancer studies
• To advance the state-of-the art in web-based clinical trials
Institution 1
Institution 2
•
•
•
Institution n
EMS Coordinating Center
Institution k
All study information
Including images in
a single database
Principal Investigator
Center Personnel:
Study Monitoring
Study Coordinator
CT Scanner/
PACS
Computer-aided diagnosis for Lung cancer
Web-accessible capabilities for Clinical Studies
• Detecting pulmonary nodules in CT scans
(are there abnormalities present?)
• Measuring growth rates of pulmonary nodules
(is it cancer?)
• Measuring emphysema from CT images
Ongoing Research
• Characterizing pulmonary nodules from shape
features
(is it cancer?)
• Evaluating coronary arteries
• …
Interactive Image Review
Web integrated
Interactive
review of all
study image
data
Radiology,
Pathology,
etc.
Automated Nodule Detection
Web integrated
• Nodule candidates
presented on
coronal image
• Clicking on a
location links to the
nodule analysis
3D Nodule Analysis
(A) Image segmentation
Web integrated
3D Nodule Analysis
(B) Axial visualizations at three times Web integrated
T1
T2
T3
3D Nodule Analysis
(C) Nodule growth Analysis
T1
T2
T3
T1-T2
T1-T3
T2-T3
Web integrated
Measuring Emphysema from CT Images
Web integrated
•
•
Example CT scan slice, red denotes emphysematous areas under -910 H.U. ~25%
for this scan
Emphysema Index (EI) measures the percentage of lung volume less than a certain
density, attempting to measure the amount of dead air-space present in the lung
Symbiotic Collaboration
• Web-based study management system with
image capabilities
– Well controlled multi-center studies that can pool
all the data for clinical results (ELCAP)
– A large document image database is established
for CAD tools development (VIA)
– New CAD tools made available to clinical studies
via the web system (ELCAP and VIA)
Research: Can we diagnose cancer from
a single CT scan Ongoing Research
Malignant
Benign
• Much more difficult problem than growth
measurement
• Research is limited by the lack of appropriate
documented image databases
Image Analysis Applications
© A. P. Reeves 2007
Observer studies on lesion marking
Web integrated
Vessel Diameter Tracking in Intravital Microscopy
Image Sequences
Web integrated
Case TM0044
34
Diameter (pixel)
32
ground truth
automated
30
28
26
24
22
20
Automatic tracking of image
sequence
18
16
0
5
10
15
20
Frame number
25
•Investigate causes of capillary dilation
•Study effects of various stimuli on vessels
Web system replaces tedious manual measurements and
provides more accurate results
30
Bacteria Enumeration in Florescent
Microscope Images
Web integrated
Enterrococcus Bacteria
Results
Web system replaces tedious manual counting
Shear Cell Tracking
Completed Research
Tracking colliding spheres In microgravity conditions
3D Image Visualization
Ongoing Research
Original CT image sequence
3D Volume rendered sequence
3D visualization is becoming essential for image interpretation
VIA Research Summary
Theme: accurate measurements from medical and biological images
Issue: How to improve measurement accuracy when the exact answer (ground
truth) is not known.
•
•
•
Web Site
www.via.cornell.edu
Recent Papers
www.via.cornell.edu/papers
Account: via
Password: papers21
Web-based image tools
www.via.cornell.edu/databases/
Research Collaboration Interests: Image analysis,
computer or manual and web-based image archive