Non-invasive gynecologic cancer detection via hair and nail trace elements and machine learning

Gynecologic cancers, including endometrial, cervical, and ovarian malignancies, often lack specific symptoms at early stages, limiting the effectiveness of existing screening strategies. Although dysregulation of trace elements has been implicated in cancer biology, the potential of non-invasive hair and nail elemental profiles, particularly when combined with complementary analytical techniques and data-driven models, remains insufficiently explored. Hair and nail samples were collected from 105 patients with gynecologic cancers (35 cervical, 35 endometrial, and 35 ovarian) and 35 healthy controls. Concentrations of selected trace elements were quantified using Synchrotron Radiation X-ray Fluorescence (SRXRF) and Inductively Coupled Plasma Mass Spectrometry (ICP-MS). Non-parametric statistical analyses were performed to assess group differences and inter-element associations. Elemental features derived from both analytical platforms were subsequently used to train a binary Support Vector Machine (SVM) classifier to discriminate between cancer and healthy samples. Significant alterations in elemental profiles were observed in cancer patients compared with controls. Notably, copper concentrations in hair samples increased substantially, while bismuth and manganese levels showed marked reductions across multiple cancer types ( p < 0.05). The Cu/Zn ratio in hair samples was consistently elevated in cervical, endometrial, and ovarian cancers relative to healthy individuals. Correlation analysis revealed distinct inter-element association patterns, including strong positive correlations among specific elements depending on cancer type and analytical method. The SVM model achieved a classification accuracy of approximately 72% and a predictive performance of 74%, indicating moderate discriminative capability based on non-invasive elemental data.