We propose to develop a machine-vision system for the diagnostic interpretation of histopathologic sections and cytologic preparations. in continuation of an ongoing research project grant. The system will have """"""""image understanding capability."""""""" that is, it will follow in its reasoning a model of the histology of a given application. The interpretive expert-system module will integrate concepts from human diagnostic knowledge with machine-computable histometric features. Systems of this kind can provide objective. quantitative, consistent evaluation of lesions, yielding more reliable interpretation and diagnostic, as well as prognostic, assessment. Using a combination of large data bases of digitized imagery for given diagnostic situations. and relational data bases (including patient history. treatment, and outcome), it is hoped that the objective assessment eventually will be related reliably to truth in diagnosis. The requirements for representative sampling are in the multimegapixel range, and require fully automatic scene segmentation and histometric feature extraction. This difficult problem has been largely resolved through ongoing support, employing an AI-based, adaptive segmentation approach. The major challenges for the proposed research are: 1) to gain an understanding of the necessary and sufficient human diagnostic clues and corresponding histometric analogs, that is, to determine the library of transforms to be used by the interpretive expert-system module; 2) to develop learning capability of the system for conceptual data; and 3) to gain an understanding of the functional requirements. dependence structure, and decision control capabilities of such a system.
Showing the most recent 10 out of 37 publications
