Purchase this article with an account.
V. Kouznetsova, I. Zaslavsky, T. Lee, S. Choi, D. McGuire, M.H. Goldbaum; Geographical Information Systems Representation and Analysis of Ocular Fundus Lesions and Anatomical Structures . Invest. Ophthalmol. Vis. Sci. 2003;44(13):3005.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: We propose a geographic information systems (GIS)-based workflow approach where segmented objects (anatomical structures and lesions) are represented as vector layers in a GIS, referenced to a coordinate space common to separate images, and analyzed using knowledge-guided spatial and attribute queries, for automated image diagnosis and electronic medical records Methods: GIS originated in geographical mapping. We similarly apply GIS (ArcView) to 25 ocular fundus images with various diagnoses. The vector representation of each type of object (e.g. optic nerve, fovea, arteries, veins, cottonwool spots, hemorrhages) was placed in a separate layer. We constructed an algorithm for automatic generation of a fundus coordinate system from the locations of key segmented objects (fovea, optic nerve), as well as general image metadata. We designed spatial and attribute queries to permit interrogation of electronic medical recordswith images enhanced with vector data, and we devised a query protocol defining which queries need to be triggered, given results of previous queries for a particular image. Results: The coordinate system generation algorithm successfully placed vector representations of segmented objects into properly referenced layers in all images with previously marked fovea and optic disk. Querying worked across images. A sequence of spatial queries (e.g. "Are lesions confined to the upper or lower hemisphere?" or "Is the shape of the optic nerve abnormal in any image?") allowed us to append location, shape, and pattern properties of different objects to disease manifestation data for each image. For most ocular fundus images, the GIS-based workflow resulted in a more compact image representation for electronic medical records storage or image diagnosis Conclusions: Compared to traditional wholesale extraction of a few spatial and attribute properties of segmented objects, GIS environment enabled improved knowledge-guided querying of arbitrary spatial relationships between anatomical structures and lesions, as well as finding comparable patterns across images.
This PDF is available to Subscribers Only