Loading…
Fast fully‐automated multimodal image co‐registration (optical coherence tomography, colour fundus photography, red‐free, fluorescein angiography)
Purpose To automatically co‐register optical coherence tomography (OCT), colour fundus photography (CFP), red‐free (RF) and fluorescein angiography (FA). Methods The burden of manually assisted co‐registration and the number of images taken per day prevent its widespread use. An algorithm able to co...
Saved in:
Published in: | Acta ophthalmologica (Oxford, England) England), 2013-08, Vol.91 (s252), p.0-0 |
---|---|
Main Authors: | , , , , |
Format: | Article |
Language: | English |
Subjects: | |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Purpose To automatically co‐register optical coherence tomography (OCT), colour fundus photography (CFP), red‐free (RF) and fluorescein angiography (FA).
Methods The burden of manually assisted co‐registration and the number of images taken per day prevent its widespread use. An algorithm able to co‐register images in between the major modalities in due time is herewith proposed. Typically, OCT fundus references translate into a poorly detailed vascular network, thus rendering difficult its co‐registration to other imaging modalities. A recently developed method (by our group) to compute the vascular network from OCT to the level of detail of CFP was used. A set of vessel features is computed and an iterative process estimates the transformation required to co‐register these locations. At each step the number of inliers is determined and the process repeats. Images of 20 eyes from 13 patients that underwent high‐definition OCT, CFP and FA, were co‐registered. These images were manually segmented by a grader. Two additional sets were co‐registered and evaluated by an expert: 102 OCT/CFP image pairs from 51 subjects, and 40 FA/RF image pairs from 20 patients.
Results A skeleton overlap metric was defined and computed based on the vessels skeleton (0–no overlap, 1‐full overlap) for the first set. An average overlap of 0.93±0.03 (N=60) is reported. From the expert evaluation, 91% of images were successfully co‐registered. For the registration of a given pair, the algorithm takes 1.68±0.38 seconds on an Intel Core i7‐2600k, 3.4 GHz computer.
Conclusion The achieved level of co‐registration, render this process an asset to the clinical daily practice and research. |
---|---|
ISSN: | 1755-375X 1755-3768 |
DOI: | 10.1111/j.1755-3768.2013.4462.x |