Probabilistic inference in computer-aided screening for cervical cancer: an event covering approach to information extraction and decision rule formulation.
A pattern analysis approach called event covering is introduced and applied to the problem of determining the significance of measured morphological attributes of cervical cells in an automated screening process. The approach is shown to be effective in deciding which characteristics at the event (outcome) level are of little or no diagnostic value, and in formulating sets of predictive rules using the most pertinent ones.