In the STARE (STructured Analysis of the REtina) project, we are developing a computerized image-interpreting system with hierarchical inferencing to measure, compare, and diagnose images of the ocular fundus. The system will have sufficient depth of imaging tools to be a resource for researchers or clinicians. The STARE system is designed to find objects of interest (normal anatomical structures and lesions) in digitized ocular fundus images and to use these objects to diagnose an image, to detect changes in sequential images, and to make clinically useful measurements that are currently tedious or costly. To accomplish these difficult goals, we must segment and identify the objects of interest. Identified objects can be used to compare images, and the objects can be assembled to describe the image. Image interpretation incorporating expert systems and neural networks can provide the structure for cross-sectional epidemiological studies. There was no established paradigm to follow to construct a system for image interpretation. We designed the overall structure of the process and determined how each task was to be accomplished. We have broken the project into steps, each of which has been accomplished. We are able to find objects of importance and correctly identify and localize them on a fundus coordinate system that we designed. We have created and tested a neural network and an expert system (INTELLEYE) to handle the interpretation of an image and its contents. We will now improve the accuracy of each step, increase the number of lesions we can identify, and integrate the image analysis steps with the expert system to allow smooth progression from image to diagnosis and change detection. We will validate the usefulness and accuracy of the system by comparing its diagnosis and image comparison to trained readers of ophthalmic images. The goal of this project is a system with multiple imaging tools and inferencing ability that can be adapted to a variety of imaging tasks. The outcome will be an image-interpreting system for use in clinical and research settings that will build annotated image databases, screen images of the ocular fundus for health care systems, furnish decision support for primary care providers, and extend the capability and productivity of the ophthalmologist.
| Hoover, A; Kouznetsova, V; Goldbaum, M (2000) Locating blood vessels in retinal images by piecewise threshold probing of a matched filter response. IEEE Trans Med Imaging 19:203-10 |
| Hart, W E; Goldbaum, M; Cote, B et al. (1999) Measurement and classification of retinal vascular tortuosity. Int J Med Inform 53:239-52 |
| Hoover, A; Kouznetsova, V; Goldbaum, M (1998) Locating blood vessels in retinal images by piece-wise threshold probing of a matched filter response. Proc AMIA Symp :931-5 |
| Hart, W E; Goldbaum, M; Cote, B et al. (1997) Automated measurement of retinal vascular tortuosity. Proc AMIA Annu Fall Symp :459-63 |
