DocumentCode :
1858675
Title :
Contextual detection of ischemic regions in ultra-wide-field-of-view retinal fluorescein angiograms
Author :
Trucco, E. ; Buchanan, C.R. ; Aslam, T. ; Dhillon, B.
Author_Institution :
Univ. of Dundee, Dundee
fYear :
2007
fDate :
22-26 Aug. 2007
Firstpage :
6739
Lastpage :
6742
Abstract :
We report a novel prototype algorithm using contextual knowledge to locate ischemic regions in ultra- wide-field-of-view retinal fluorescein angiograms. We use high- resolution images acquired by an Optos ultra-wide-field-of- view (more than 200 degrees) scanning laser ophthalmoscope. We leverage the simultaneous occurrence of ischemia with a number of other signs, detected automatically, typical for the state of progress of the condition in a diabetic patient. The specific nature of ischemic and non-ischemic regions is determined with an AdaBoost learning algorithm. Preliminary results demonstrate above 80% pixel classification accuracy against manual annotations.
Keywords :
biomedical optical imaging; blood vessels; endoscopes; eye; image classification; laser applications in medicine; learning (artificial intelligence); medical image processing; AdaBoost learning algorithm; automatic ischemia detection; contextual knowledge detection; diabetic patient; high-resolution images; ischemic regions; nonischemic regions; pixel classification; scanning laser ophthalmoscope; ultra-wide-field-of-view retinal fluorescein angiograms; Birth disorders; Blindness; Diabetes; Image quality; Image texture analysis; Ischemic pain; Pathology; Prototypes; Retina; Retinopathy; Algorithms; Diabetic Retinopathy; Fluorescein Angiography; Humans; Image Interpretation, Computer-Assisted; Ischemia; Retinal Vessels;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Engineering in Medicine and Biology Society, 2007. EMBS 2007. 29th Annual International Conference of the IEEE
Conference_Location :
Lyon
ISSN :
1557-170X
Print_ISBN :
978-1-4244-0787-3
Type :
conf
DOI :
10.1109/IEMBS.2007.4353908
Filename :
4353908
Link To Document :
بازگشت