RegistrationBased Change Detection

1
Registration-Based Change Detection

• Charles V. Stewart
• Department of Computer Science
• Rensselaer Polytechnic Institute

2
Change Mosaics
This animation shows mosaiced and registered
color fundus images of a patient with retinal
vasculitis taken over a period of 2 years and 3
months. Each mosaic shows combined images taken
during a single sitting. All mosaics are
aligned. This allows visualization of the
progress of disease. The next slide shows
expanded views of five regions where changes
appear to have occurred. Registration with
fluorescien angiography, which we have
demonstrated, is needed to confirm the
significance of the changes. We are developing
techniques to automatically detect and
characterize changes, for use as an aid to
clinical diagnosis as well as in clinical trials
and reading centers. Fully automatic
registration and mosaic construction make this
capability possible.
3
Expanded Views of Changes in Small Image Regions
In these clipped regions of the animations (a few
overlap slightly), changes appear to have occured
in the position, caliber, and tortuosity of some
vessels. A few vessels seem to have disappeared,
although this has not been confirmed.
Neovascularization around the optic disk (NVD) is
apparent in the rightmost clip. Again,
fully-automatic and precise alignment enables the
visualization of these changes.