LAUSR

Cervical cancer is one of the most common cancers in women with pronounced stages of precancerous lesions. Accurate differential diagnosis of such lesions is one of the primary challenges of medical specialists, which is vital to improving patient survival. The aim of this study was to develop and test an algorithm for the differential diagnosis of cervical lesions based on lipid levels in scrapings from the cervical epithelium and cervicovaginal canal. The lipid composition of the samples was analyzed by high-performance chromato-mass spectrometry. Lipid markers were selected using the Mann–Whitney test with a cutoff value of 0.05 and by projections to latent structures discriminant analysis, where a projection threshold of one was chosen. The final selection of variables for binomial logistic regressions was carried out using the Akaike information criterion. As a result, a final neoplasia classification method, based on 20 logistic regression sub-models, has an accuracy of 79% for discrimination NILM/cervicitis/LSIL/HSIL/cancer. The model has a sensitivity of 83% and a specificity of 88% for discrimination of several lesions (HSIL and cancer). This allows us to discuss the prospective viability of further validation of the developed non-invasive method of differential diagnosis.