We propose to develop computer-aided automated analysis of features in ocular fundus images and fluorescein angiograms and further to generate an expert system for rule-based image interpretation and diagnosis. The use of artificial intelligence in conjunction with image analysis is a sophisticated and innovative approach new to ophthalmology. Image processing and pattern-recognition techniques will be used to automatically extract features such as hemorrhages, exudates, scars, blood vessels, the optic disk, etc., from digital fundus images by analyzing attributes such as size, shape, color, texture, edge characteristics, fluorescein hyperfluorescence etc., which describe these features. The automated extraction of fundus features will be performed by a software package called STARE (STructured Analysis of the REtina), which will produce a coded structured description of an image of the ocular fundus. Features will be assembled into patterns of potential diagnostic significance. For computer-aided diagnosis, patterns of automatically extracted features will be combined with non-image information provided by the physician expert through application of pre-defined decision rules. The knowledge and training that enables the physician to appreciate the significance of particular patterns and ask appropriate questions will be encoded in rules designed as IF-THEN-ELSE decision structures. We will use an existing expert rule-based bystem called ERS (Embedded Rule-based System), which has interactive facilities, to receive information from the physician and report findings. We will develop STARE for the automated analysis of fundus images through objective measurements of pixels with pixel algorithms such as gradient operators, spectral classifiers, and edge detectors; grouping of pixels with application of pattern-recognition algorithms such as statistical classifiers and segmenters; interpretation of local areas extracted by pixel algorithms using object algorithms for measurement of attributes such as color and edge sharpness; and identification of features such as the disk or exudates with higher level statistical classifiers and rule-based decision making. Following development of image-coding and database software, we will start with one or two example goal nodes (diagnoses) and will produce rules with both forward chaining and backward chaining, edit and modify the rules with reality testing on our large library of fundus photographs and fluorescein angiograms to detect and correct deficiencies in STARE algorithms and ERS rules, and expand the system to cover more diagnoses and improve performance.
| Goldbaum, M H; Katz, N P; Nelson, M R et al. (1990) The discrimination of similarly colored objects in computer images of the ocular fundus. Invest Ophthalmol Vis Sci 31:617-23 |
| Pavan-Langston, D (1990) Herpes simplex virus ocular infections: current concepts of acute, latent and reactivated disease. Trans Am Ophthalmol Soc 88:727-96 |
