Interpretable CT radiomics–serum ensemble predicts poorly differentiated esophageal squamous cell carcinoma

An ensemble model combining intratumoral/peritumoral CT radiomics with serum alkaline phosphatase and neutrophils predicted poorly differentiated esophageal squamous cell carcinoma (validation AUC 0.799; bootstrap AUC 0.849, 95% CI 0.790–0.891, n=261).

• Retrospective study of 261 patients randomized to training (n=183) and validation (n=78) cohorts; enhanced CT arterial phase radiomics were extracted from intratumoral and peritumoral 0.3 cm regions.
• The ensemble machine learning model had the highest validation AUC (0.799), F1 (0.810), recall (0.870), and Brier score (0.166); bootstrap full-cohort AUC was 0.849 (95% CI 0.790–0.891).
• SHAP and restricted cubic spline analyses supported nonlinear contributions from radiomic scores and alkaline phosphatase; authors call for multicenter external validation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.