Content-based Cell Image Retrieval Using Automated Feature Extraction

摘要: Objective: Currently, when cytopathology images are archived, they typically stored with a limited text-based description of their content. Such inherently fails to quantify the properties an image and refers extremely small fraction its information This paper describes method for automatically indexing individual cells associated diagnoses by computationally derived cell descriptors. methodology may serve better index data contained in digital databases, thereby enabling cytologists pathologists cross-reference unknown etiology or nature. Design: The method, implemented program called PathMaster, uses series computer-based feature extraction routines. Descriptors characteristics generated these routines employed as indexes morphology, texture, color, spatial orientation. Measurements: fidelity was tested after populating database 152 lymphocytes/lymphoma captured from lymph node touch preparations stained hematoxylin eosin. Images ''unknown'' lymphoid cells, previously unprocessed, were then submitted diagnostic cross-referencing analysis. Results: PathMaster listed correct diagnosis first differential 94 percent recognition trials. In remaining 6 trials, within three ''differentials.'' Conclusion: is pilot program/search engine that creates indexed reference images. Use such provide assistance diagnostic/ prognostic process furnishing prioritized list possible identifications uncertain etiology. n J Am Med Inform Assoc. 2000;7:404-415.