DefinePK

Abstract

In a cervical cancer screening procedure, a patient’s Pap smear slide is presented to determine presence of abnormalities. Conventionally, features of individual cells are measured and analysed in the initial screening step. Based on the analysis results at the cellular level, the Pap smear slide is classified as positive (abnormal) or negative (normal). However, each slide presents data on thousands of cells. Consequently, classifying the slide based on cell-by-cell analysis is very time consuming and prone to ‘false negative’ problems. In this paper, we propose group-based classification (GBC) approach to classify a slide by measuring the slide data as a whole instead of scrutinizing the cells individually. This means measuring the slide’s features once from a group of cells to obtain a diagnosis. We apply two group-based nearest neighbour techniques; voting and pooling schemes to label each slide. The performances of the group-based nearest neighbour techniques are evaluated against existing k-nearest neighbour classifier in terms of accuracy and area under the receiver operating characteristic curve (AUC). The group-based nearest neighbour classifiers show favorable accuracy compared to the existing k-nearest neighbour classifier.