Frontiers in Oncology (Jan 2025)
Delta-radiomics analysis based on magnetic resonance imaging to identify radiation proctitis in patients with cervical cancer after radiotherapy
Abstract
ObjectivesTo develop a magnetic resonance imaging (MRI)-based radiomics model for predicting the severity of radiation proctitis (RP) in cervical cancer patients’ post-radiotherapy.MethodsWe retrospectively analyzed clinical data and MRI images from 126 cervical squamous cell carcinoma patients treated with concurrent chemoradiotherapy. Logistic regression (LR), Pearson correlation coefficient, and least absolute shrinkage and selection operator (LASSO) methods were utilized to select optimal imaging features, leading to a combined prediction model developed using a random forest (RF) algorithm. Model performance was assessed using the area under the curve (AUC), DeLong test, calibration curve, and decision curve analysis (DCA), with Shapley Additive exPlanations (SHAP) values for interpretation.ResultsThe samples were split into training (70%) and validation (30%) sets. The delta-radiomics model, comprising 10 delta features, showed strong predictive performance (AUC: 0.92 for training and 0.90 for validation sets). A comprehensive model combining delta-radiomics with clinical features outperformed this, achieving AUCs of 0.99 and 0.98. DeLong’s test confirmed the comprehensive model’s statistical superiority, and both calibration curves and DCA indicated good calibration and high net benefit. Key features associated with RP included D1cc, T1_wavelet-LLL_glcm_MCC, D2cc, and T2_original_firstorder_90 Percentile.ConclusionsThe MRI-based delta radiomics model shows significant promise in predicting RP severity in cervical cancer patients following radiotherapy, with enhanced predictive performance when combined with clinical features.
Keywords
