A preliminary study of the potential of tree classifiers in triage of high-grade squamous intraepithelial lesions.

Abstract

Objective: To investigate the potential value of tree classifiers for the triage of high-grade squamous intraepithelial lesions.

Study design: The dataset comprised 808 histologically confirmed cases having a complete range of the cytologic sample assessments--liquid-based cytology, reflex human papillomavirus (HPV) DNA test, E6/E7 HPV mRNA test, and p16 immunocytochemical examinations. Data include 488 histologically negative (cervical intraepithelial neoplasia [CIN] 1 and below) or clinically negative cases and 320 with histologic diagnosis of CIN 2 or worse. Cytologic diagnosis was made according to the criteria of the Bethesda System. Cases were classified in two groups according to histology: those with CIN 2 or worse and those with CIN 1 and below. Fifty percent were randomly selected as a training set and the remaining were as a test set.

Results: Application of tree classifier on the test set gave correct classification of 66.9% for CIN 2 and above cases and 97.3% for CIN 1 and below, producing overall accuracy of 91.5%, outperforming cytologic diagnosis alone.

Conclusion: Application of tree classifiers, based on standard cytologic diagnosis and expression of studied biomarkers, produces improved classification results for cervical precancerous lesions and cancer diagnosis and