This study was designed to develop an algorithm for the diagnosis of cervical high-grade squamous intraepithelial lesions (HSIL), based on patterns of volatile organic compounds, evaluated using an e-nose.
For this pilot study, the study population consisted of a group of 25 patients with histologically confirmed HSIL and a group of 26 controls. Controls consisted of women visiting the outpatient department for gynecological complaints unrelated to cancer. Women had a negative high-risk human papillomavirus and/or normal cytology (negative for intraepithelial lesions of malignancy) of their most recent test performed in the context of participation in routine cervical cancer screening. Breath tests were performed and labeled with the correct diagnosis. Machine-learning techniques were used to develop a model for predicting HSIL. Based on the receiver operating characteristics curve, both sensitivity and specificity were calculated.
Individual classifications of all patients with HSIL and controls, as calculated by the model, showed a sensitivity of 0.88 (95% CI = 0.68–0.97) and specificity of 0.92 (95% CI = 0.73–0.99). The positive predictive value and the negative predictive value were 0.92 (95% CI = 0.72–0.99) and 0.89 (95% CI = 0.70–0.97), respectively. The Cohen κ coefficient was 0.80.
E-nose can detect distinctive patterns of volatile organic compounds between cervical HSIL patients and controls. Validation of the algorithm in further studies is necessary before possible implementation into daily practice.