The overall aim of this project is to improve automated cervical smear image analysis techniques sufficiently to: 1) routinely provide additional diagnostic information to supplement that provided by human visual examination, and 2) provide accurate, cost effective prescreening capabilities. By furnishing additional quantitative diagnostic information, such automated systems would permit the practicing cytopathologist to make more accurate and refined diagnostic judgements, which would lead to improved patient management. By providing a practical prescreening capability, such systems would increase the efficiency and efficacy of health care delivery. Current automated cervical smear analysis research employs only the analysis of isolated cells, and practical automated analysis is not yet a reality. The primary contribution of this project will be the completion of development and testing of image analysis techniques that extract information from cells and other objects as seen in the context of the """"""""background"""""""" of the smear. Specifically, cells, cell clusters, bare nuclei, and cytoplasmic fragments are analyzed and the resulting contextual features, slide-averaged """"""""features"""""""" and high-resolution features describing single cells are combined to produce a more complete and accurate description of the smear. This description will provide information not obtainable by human visual analysis, which can be used, 1) directly, to ascertain diagnostic clues, and 2) indirectly to make accurate prescreening possible. An extensive pilot study has shown that the contextual analysis provides complementary information to that provided by single cell analysis. That is, where single cell analysis seems to have difficulty, contextual analysis is most accurate - and vice versa. Thus we expect to demonstrate that the combination of our contextual analysis techniques with single cell analysis will give automated Pap smear analysis the additional screening accuracy that is needed. The proposed project is a followup intended to: a) validate the techniques of the pilot study, b) to expand the study to include parameters that provide additional diagnostic and prognostic information, and c) to generate specifications for a system to accomplish the analysis in a routine clinical laboratory environment.
