Preoperative non-invasive prediction of lymph node metastasis in cervical cancer using a multiparametric radiomics model based on transvaginal ultrasound.

objectiveTo evaluate whether ultrasound-radiomics (US-radiomics) features extracted from ultrasound, integrated with genomic data of single nucleotide polymorphisms (SNPs) associated with cervical cancer (CC) susceptibility and clinical features, could improve the prediction of lymph node metastasis (LNM) in patients with CC.

methodsThe model was established using ultrasound image features, SNPs data, and clinical data from patients. All subjects were randomly divided into a training set and a validation set in a 7:3 ratio. Feature selection and prediction modeling were performed using the max-relevance and min-redundancy (mRMR) algorithm, the least absolute shrinkage and selection operator (LASSO), and support vector machine (SVM) methods.

resultsD-dimer, SCC-Ag and rs2977530 were identified as independent predictors of LNM. The combined clinical-SNPs-US-radiomics model demonstrated higher classification efficiency for predicting LNM in CC, with an area under the receiver operating characteristic curve (AUC) of 0.826 [95% CI: 0.720-0.921] in the training cohort and 0.699 [95% CI: 0.537-0.857] in the validation cohort.

conclusionsThe model developed in this study, which integrates US-radiomics score with clinical features and SNPs data, has the potential to non-invasively predict LNM in CC and holds promise for clinical application.