An ensemble feature ranking framework for the assessment of the efficacy of cervical cancer detection tests and human papillomavirus genotypes in the detection of high-grade cervical intraepithelial neoplasia and cervical carcinoma

In most cases, cervical cancer (CxCa) develops as a result of underestimated abnormalities present in cytology via the test Papanicolaou (Pap test). Although it is the sole test that has proved its value for cancer prevention, Pap test is prone to human errors as it is performed via the microscope and requires skilled personnel. Consequently, molecular techniques have emerged in the recent years, aiming to replace Pap test or support the diagnosis as ancillary tests. These include DNA micro-arrays identifying Human Papillomavirus (HPV) subtypes, mRNA techniques such as nucleic acid based amplification or flow cytometry identifying E6/E7 oncogenes, and immunocytochemistry techniques such as overexpression of p16. These techniques are either highly sensitive or highly specific, but not both at the same time, thus no perfect method is available today. Moreover, most of the published studies compare two tests against each other, which, however, are using different population study groups, therefore it is questionable if the results can be used to perform a direct comparison between the available tests. In this paper, we adopt a sophisticated approach to assess the value of several well-known and widely applied CxCa tests in detecting high-grade cervical intraepithelial neoplasia and CxCa (CIN2+), by employing an Ensemble Feature Ranking framework. Applying the proposed framework we managed to directly compare and rank the studied CxCa tests and several HPV genotypes according to their efficacy in detecting CIN2+ lesions. The results suggest that HPV mRNA tests perform better than HPV DNA testing for the triage of abnormal Pap tests. Such findings may guide cytopathologists in a step by step basis to decide which tests to perform following an abnormal Pap test, in order to decrease uncertainty, improve accuracy and reduce time to diagnosis and resources.