18F-FDG PET radiomics-based machine learning model for differentiating pathological subtypes in locally advanced cervical cancer

Abstract

Purpose To determine diagnostic performance of 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) and computed tomography (CT) radiomics-based machine learning (ML) for classification of cervical adenocarcinoma (AC) and squamous cell carcinoma (SCC). Methods A total of 195 patients with locally advanced cervical cancer were enrolled in this study, and randomly allocated to training cohort (n = 136) and validation cohort (n = 59) in a ratio of 7:3. Radiomics features were extracted from pretreatment 18F-FDG PET/CT and selected by the Pearson correlation coefficient and the least absolute shrinkage and selection operator regression analysis. Six ML classifiers were trained and validated, and the best-performing classifier was selected based on accuracy, sensitivity, specificity, and area under the curve (AUC). The performance of different models was assessed and compared using the DeLong test. Results Five PET and one CT radiomics features were selected and incorporated into the ML classifiers. The PET radiomics model constructed based on the lightGBM algorithm had an accuracy of 0.915 and an AUC of 0.851 (95% CI, 0.715–0.986) in the validation cohort, which were higher than that of the CT radiomics model (accuracy: 0.661; AUC: 0.513 [95% CI, 0.339–0.688]). The DeLong test revealed no significant difference in AUC between the combined radiomics model and the PET radiomics model in both the training cohort (P = 0.347) and the validation cohort (P = 0.776). Conclusions The 18F-FDG PET radiomics model can be used as a clinically applicable tool for differentiating pathological subtypes in patients with locally advanced cervical cancer.