Integrating CT-Based Radiomics Analysis With Serum CEA Level to Predict Neoadjuvant Chemoradiotherapy Response in...

Response to neoadjuvant chemoradiotherapy (NACRT) in locally advanced colorectal cancer (CRC) varies widely, and accurately identifying poor responders is crucial for guiding timely treatment modification. This study aimed to develop an integrated predictive model combining radiomics derived from routine radiotherapy non-contrast planning computed tomography (CT) images with serum biomarkers to achieve better predictive accuracy than modality alone in identifying the therapeutic response of CRC patients receiving NACRT. Ninety-two patients with stages II-III CRC who received NACRT and surgery were retrospectively analyzed. CT-based radiomic features were extracted from pre-treatment non-contrast planning CT images using LIFEx software. Patients were randomly divided into derivation (n = 65) and validation (n = 27) cohorts while least absolute shrinkage and selection operator logistic regression was used to construct a radiomics score (Rad-score). Univariate and multivariate analyses evaluated the independent predictive value of Rad-score and patients' clinicopathological features. A nomogram was established and calibrated using bootstrap resampling, and of 111 usable radiomic features, seven were selected to construct the Rad-score. The CT-based radiomics signature showed strong discriminative performance in predicting pathological complete response (pCR) (AUC: derivation 0.854; validation 0.818). In multivariate analysis, Rad-score (adjusted OR = 10.20, p = 0.002) and low pretreatment serum carcinoembryonic antigen (CEA) level (adjusted OR = 15.20, p = 0.024) were independent predictors of pCR. The integrated nomogram demonstrated excellent calibration in both cohorts [mean absolute error (MAE) = 0.042 and 0.045]. A combined model incorporating CT-based radiomics and pretreatment serum CEA provides a robust and resource-efficient tool for predicting therapeutic response following NACRT in CRC.