Habitat-Based Radiomics Model of Pretreatment CT to Predict Pathological Response of Esophageal Squamous Cell...

Neoadjuvant chemoimmunotherapy (NCIT) has shown promising efficacy in locally advanced esophageal squamous cell carcinoma (LA-ESCC), yet pretreatment predictors for treatment response remain to be identified. This study aimed to evaluate a pretreatment CT-based habitat radiomics model for predicting pathological response in LA-ESCC treated with NCIT.

This prospective multicenter study enrolled 215 patients with LA-ESCC receiving NCIT from three centers. Patients from Center A were randomly allocated to training (n = 110, 70%) and validation (n = 47, 30%) sets, with those from Centers B (n = 33) and C (n = 25) as an external test set. Responders and nonresponders were classified by tumor regression grades. Conventional and habitat radiomics features were extracted from intratumoral and peritumoral regions. Fourteen machine-learning (ML) classifiers were used to build intratumoral, peritumoral, and combined habitat radiomics models, along with corresponding conventional models. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). Shapley Additive Explanations (SHAP) analysis was employed for model interpretation.

The habitat radiomics models outperformed conventional radiomics models. The combined intratumoral and peritumoral habitat radiomics model achieved an AUC of 0.93 (95% CI: 0.84-0.98), accuracy of 0.83, sensitivity of 0.79, and specificity of 0.93 in the external test set. SHAP analysis revealed that both intratumoral and peritumoral habitat radiomics features contributed significantly to predictive performance.

The interpretable ML model combining intratumoral and peritumoral habitat radiomics features accurately predicts the response of LA-ESCC to NCIT.