Abstract / Summary
We aimed to construct nomograms combining clinical indicators, apparent diffusion coefficient (ADC) values, and MRI radiomics signatures to predict early therapeutic response, progression-free survival (PFS), and overall survival (OS) after concurrent chemoradiotherapy (CCRT) in locally advanced cervical cancer (LACC), and further uncover the biological basis of radiomics features. A total of 671 LACC patients from two centers were retrospectively enrolled and split into training ( n = 381) and external validation cohorts ( n = 290). LASSO regression screened radiomics features to calculate radiomics scores (Rad-scores), and multivariate Logistic/Cox regression was adopted to build three visualized prediction models, followed by internal and cross-institutional external validation. Bioinformatic analysis and immunohistochemistry were performed to explore underlying molecular mechanisms. All three nomograms achieved favorable discrimination, calibration, and clinical net benefit, with AUCs ranging from 0.828 to 0.951 in both cohorts. Radiomics signatures were predominantly enriched in DNA damage repair and cell cycle pathways, with LIG4 identified as the hub gene positively correlated with Rad-scores and significantly linked to unsatisfactory CCRT response and poor patient prognosis. The nomograms serve as non-invasive predictive tools for LACC patients receiving CCRT and can reflect activated tumor DNA damage repair mediated by LIG4, offering solid biological evidence for radiomics clinical translation.