Integrating tumour and lymph node radiomics features for predicting disease-free survival in locally advanced esophageal squamous cell cancer after neoadjuvant chemotherapy and complete resection.

PURPOSE

To investigate the utility of combined tumour and lymph node (LN) radiomics features in predicting disease-free survival (DFS) among patients with locally advanced esophageal squamous cell carcinoma (ESCC) after neoadjuvant chemotherapy and resection.

METHODS

We retrospectively enrolled 176 ESCC patients from January 2013 to December 2016. Tumour and targeted LN segmentation were performed on venous phase CT images. Models were constructed using LASSO Cox regression: a clinical model, a clinical-tumour radiomics model, and a clinical-tumour-LN radiomics model. Model fitting was evaluated using Akaike information criterion and likelihood ratio (LR), while performance was assessed using Harrell's concordance index (C-index) and time-dependent receiver operating characteristic analysis.

RESULTS

The clinical model included clinical stage and neutrophil-to-lymphocyte ratio (NLR). Integration of tumour features significantly improved prognostic accuracy (clinical-tumour model vs. clinical model, LR: 17.84 vs. 11.84, P = 0.049). Subsequent integration of LN features further augmented model performance (clinical-tumour-LN model vs. clinical-tumour model, LR: 24.48 vs. 17.84, P = 0.009). The final model included clinical stage, NLR, two tumour features (Conventional_mean and GLZLM_HGZE), and one LN feature (GLCM_entropy). The C-index was 0.68 for the training set and 0.70 for the test set. The nomogram based on these features effectively stratified patients into high- and low-risk groups (P < 0.001).

CONCLUSIONS

The clinical-tumour-LN model, integrating clinical stage, NLR, and radiomics features, outperformed simpler models in predicting DFS among ESCC patients after neoadjuvant chemotherapy and resection. This underscores the potential of radiomics data to enhance prognostic models, offering clinicians a more robust tool for assessment.