Radiomics analysis of dual-layer spectral detector CT-derived iodine density maps for predicting the pathological complete response of esophageal squamous cell carcinoma after neoadjuvant chemoimmunotherapy

Purpose To evaluate the efficacy of radiomics from dual-layer spectral detector CT (DLCT) iodine density maps, combined with clinicopathological characteristics and spectral parameters, in predicting pathological complete response (pCR) following neoadjuvant chemoimmunotherapy (nCIT) in patients with esophageal squamous cell carcinoma (ESCC). Materials and methods A total of 129 patients with ESCC undergoing nCIT were allocated to training (n = 86) and validation (n = 43) groups. Patients were categorised into pCR and non-pCR groups based on postoperative tumor regression grade. Multiple parameter maps were reconstructed from pre-treatment DLCT images for quantitative spectral analysis. Radiomic features extracted from the venous-phase iodine density maps were used to develop a radiomics score (Radscore). Univariate and multivariate logistic regression analyses were used to identify independent predictors of pCR, leading to the development of clinical, spectral, radiomics, and combined models. The model performance was evaluated using receiver operating characteristic (ROC) curves, decision curve analysis, and calibration curves. Results Multivariate logistic regression analysis identified differentiation grade, venous-phase normalized iodine density (NID V ), and Radscore as independent predictors of pCR. The combined model demonstrated superior predictive performance compared to the individual models, with areas under the curve of 0.943 (training) and 0.833 (validation). Calibration curves showed good agreement between the predicted and observed outcome values, whereas decision curves confirmed a superior net clinical benefit in both cohorts. Conclusions The combined model incorporating radiomics features from DLCT iodine density maps, clinicopathological characteristics, and spectral parameters demonstrated robust predictive capability in predicting pCR to nCIT in ESCC, which is anticipated to be a valuable tool for personalized treatment decision-making.