Predictive Value of a Radiomics-Derived Risk Score for Local Progression in T3 Laryngeal Cancer: A 10-Year...

Background/Objective: Local progression after concurrent chemoradiotherapy in T3 laryngeal carcinoma (LC) remains difficult to predict using conventional clinical assessment alone. This study aimed to develop a radiomics-derived risk score from routine post-treatment contrast-enhanced CT and evaluate its prognostic value-together with clinical variables-for predicting local progression-free survival (LPFS). Methods: In this single-center retrospective cohort, 67 patients with pathologically confirmed T3-stage LC treated with chemoradiotherapy were included. All patients underwent contrast-enhanced CT at baseline and 3 months after treatment completion; radiomics analysis was performed using post-treatment CT with 3D manual segmentation of the primary tumor. A total of 111 radiomic features were extracted (shape, first-order, and texture). Features with AUC > 0.60 were screened, and six top-performing features were used to construct a radiomics score (0-6) based on optimized cutoffs. The primary endpoint was LPFS, defined as time from end of treatment to biopsy-proven residual or recurrent primary tumor. Cox regression and Kaplan-Meier analyses were performed. Results: Mean age was 59.6 ± 9.4 years, and 37.3% developed local progression during follow-up. In multivariable Cox analysis, the radiomics score remained an independent predictor of local progression (HR per 1-point increase: 2.38; 95% CI: 1.59-3.56; p < 0.001), with high model discrimination (C-index: 0.855). LPFS differed significantly across radiomics score strata (p < 0.001); higher scores were associated with substantially shorter time to progression and poorer 1-, 3-, and 5-year LPFS rates. Conclusions: A post-treatment CT-derived radiomics score integrated with clinical parameters showed favorable performance for predicting local progression in T3 laryngeal cancer after chemoradiotherapy. Although external validation is required, this approach may support more individualized surveillance by identifying patients at higher risk of early treatment failure.