Screening of cervical intraepithelial neoplasia based on multiple features extracted from multi-electrode bioimpedance spectroscopy

Abstract

Objective. Bioimpedance spectroscopy (BIS) has emerged as a promising technique for screening cervical intraepithelial neoplasia (CIN) since the electrical properties vary with the pathological status of cervical tissues. In this study, we aimed to evaluate the ability of CIN screening using multiple features extracted from BIS measurements collected with a multi-electrode BIS probe. Approach. This study enrolled 161 patients with gynecological diseases, including 44 with and 117 without cervical dysplasia. Upon the histological diagnosis, the samples were classified as normal, CIN I, and CIN II with p16 positive (p16(+))/CIN III. Complex impedance spectra of in vitro cervical conization tissues were measured using the BIS probe. A Cole–Cole plot was generated from each patient’s data measured on the conized cervix, and various features were extracted. Receiver operating characteristic (ROC) curves were generated, and the area under each ROC curve (AUC) was calculated. Main results. As a result, fifteen features from Cole–Cole plots differed significantly ( p < 0.01) between normal cervices and CIN. The AUCs based on multiple features, as determined by multivariable logistic regression, were 0.93 for normal cervix vs CIN I, 0.99 for normal cervix vs CIN II p16(+)/CIN III, and 0.94 for normal cervix vs CIN. These AUCs were improved by 14.8%, 7.6%, and 8.0%, respectively, compared with the results based on features extracted from only the real part of the impedance spectra. Significance. In conclusion, CIN can be accurately diagnosed using multiple features extracted from the impedance spectrum of in vitro cervical samples. Particularly, this method was highly accurate in classifying CIN II p16(+)/CIN III, which has a higher risk of progression to cancer.