Infrared spectroscopy with multivariate analysis segregates low-grade cervical cytology based on likelihood to regress, remain static or progress

摘要

Cervical cancer is the 2nd most common female cancer worldwide. However, in the developed world, cervical screening has reduced this cancer burden. Most smear referrals are low-grade, requiring continuous monitoring until they regress. Others need monitoring for static disease, while a few require treatment due to persistent low-grade or progressive disease. The a€?Holy Graila€? in cervical screening is predicting which patient is likely to have progressive disease. Fourier-transform infrared (FTIR) spectroscopy exploits the fact that an infrared (IR) spectrum represents a a€?biochemical-cell fingerprinta€?, which can be obtained from a cellular specimen based on a wavenumber-dependent absorption band pattern of constituents' vibrating chemical bonds. Low-grade (CIN1) specimens (n = 67) diagnosed on cytology were analysed using IR spectroscopy. The n = 67 study participants were rescreened by conventional cytology after a year whereupon three showed progressive disease and 31 had persistent low-grade atypia; 33 had regressed. Spectra from the initial cytology samples were then analysed using principal component analysis (PCA) with output (10 principal components) being inputted into linear discriminant analysis (LDA) to predict which samples would progress, remain static or regress; this approach was compared with variable selection techniques, namely the successive projection algorithm (SPA) and genetic algorithm (GA). Significant wavenumbers distinguishing regressive vs. static disease were 1736 cma?’1, 1680 cma?’1, 1512 cma?’1, 1234 cma?’1, 1099 cma?’1 and 968 cma?’1; separating the two categories is difficult due to a significant degree of a€?overlapa€?. Progressive disease can be significantly differentiated from static disease based on wavenumbers 1662 cma?’1, 1648 cma?’1, 1628 cma?’1, 1512 cma?’1, 1474 cma?’1 and 965 cma?’1; it can be segregated from regressive disease with 1686 cma?’1, 1674 cma?’1, 1625 cma?’1, 1561 cma?’1, 1525 cma?’1 and 1310 cma?’1. The GAa€“LDA model shows good separation for all categories (i.e., regressive vs. static vs. progressive disease) using 35 wavenumbers. An ability to predict progressive disease will reduce the need for repeat smears every six months whilst allowing early identification of patients who require treatment.