Evaluation of contextual analysis for computer classification of cervical smears

Abstract

A procedure for automated analysis of cervical smears has been implemented in an image cytometry system. Smears are described exclusively in terms of global and contextual information extracted by pattern-recognition algorithms and represented by a vector ofproportions of cellular object types. Linear discriminant functions, based on a Fisher criterion, are derived to classify smears with a cross-section of diagnoses into two broad categories, normal and abnormal. Results obtained from 83 smears indicate 78% correct classification. In contrast to most automated systems, good classification results were obtained in normal smears with benign changes caused by inflammation and with postmenopausal atrophia and in abnormals with mild dysplasia. These findings suggest that contextual analysis may be sensitive to subtle changes in cellular morphology and to progressive patterns of dysplasia. When used with standard isolated cell analysis, contextual analysis may provide additional complementary information for automated cervical prescreening.Key terms: Automated cytology, cervical smear prescreening, image analysis, image cytometry, contextual analysis, cell classification, pattern recognition Present automated prescreening systems rely solely on the frequency of normal and abnormal cell classes present on the smear, as determined by a single-cell classifier, to establish a smear diagnosis. This method of smear prescreening often requires laborious computations that make it difficult to satisfy the basic aims in automation; speed and accuracy. Complex algorithms for excluding cell clusters and other cellular artifacts are required to guarantee successful recognition of isolated cells. In addition to the computation burden of these algorithms, they suppress cytological information from the smear classifier decision rule.Although cytotechnologists and pathologists utilize the frequency of cell classes to form a positive diagnosis (3,10), they also search for certain contextual patterns that are useful in describing the overall appearance of a cervical smear (15). Examples of contextual patterns include the abundance and spatial distribution on a smear of single cells, cell clusters, and cellular derivatives (i.e. degenerated cytoplasms and nuclei extruded from cells). Moreover, a recent study (2) suggests that human screeners achieve a lower accuracy in smear classification when visual observations of smears are made using only isolated cells as opposed to using contextual information in addition to isolated cell data. This indicates that a more global approach to smear classification using contextual information may provide automated devices with better accuracy and reduced processing time. This article examines the usefulness of contextual analysis in a computer-assisted smear presweening system. MATERIALS AND METHODS Smear Preparation and StainingThe specimens were collected from second scrapes taken immediately after the first smear from every second or third woman participating in the Nijmegen Population Screening P...

Contextual analysis and intermediate cell markers enhance high‐resolution cell image analysis for automated cervical smear diagnosis

“…The statistical classifier for the contextual analysis was based on a two level classifier described previously (5). The first level of analysis classifies the objects found in each of the four categories: single cells, cell clusters, bare nuclei, and celullar debris.…”

“…A set of 34 cervical smears was selected at random from a larger data set of slides analyzed previously for contextual parameters alone (5). These samples had been prepared as monolayers (13,18), and stained with the Thionin-Feulgen/Congo Red stain.…”

<title>Progress in the robust automated segmentation of real cell images</title>

“…We have the advantage of being able to discard incorrectly segmented objects but the added burden of having to detect when these segmentation failures occur. 36 …”