Louise M. Isenstein

Computerized contextual analysis in diagnosis

Cervical cytology specimens prepared as monolayers were analyzed with a combination of high-resolution single-cell image analysis and lower-resolution contextual image analysis. For the latter, quantitative features of cluster architecture, object distribution and background were used to mimic the pathologists global assessment of a specimen. Contextual analysis correctly classified 81% of the cases for diagnosis, about on a par with the companion single-cell cytometric analysis. Contextual and single cell analysis complemented each other, yielding greater accuracy when the two methods were combined. Additional applications of contextual analysis are discussed.