CT-based radiomics nomogram may predict local recurrence-free survival in esophageal cancer patients receiving definitive chemoradiation or radiotherapy: A multicenter study

Abstract

Background and purpose: To establish and validate a contrast-enhanced computed tomography-based hybrid radiomics nomogram for prediction of local recurrence-free survival (LRFS) in esophageal squamous cell cancer (ESCC) patients receiving definitive (chemo)radiotherapy in a multicenter setting.

Materials and methods: This retrospective study included 302 ESCC patients from Xijing Hospital receiving definitive (chemo)radiotherapy, which were randomly assigned to the training (n = 201) and internal validation sets (n = 101). And 74 and 21 ESCC patients from the other two centers were used as the external validation set (n = 95). A hybrid radiomics nomogram was established by integrating clinical factors, radiomic signature and deep-learning signature in training set and was tested in two validation sets.

Results: The deep-learning signature showed better prognostic performance than radiomic signature for predicting LRFS in training (C-index: 0.73 vs 0.70), internal (Cindex: 0.72 vs 0.64) and external validation sets (C-index: 0.72 vs 0.63), which could stratify patients into high and low-risk group with different prognosis (cut-off value: -0.06). Low-risk groups had better LRFS than high-risk groups in training (p < 0.0001; 2-y LRFS 71.1% vs 33.0%), internal (p < 0.01; 2-y LRFS 58.8% vs 34.8%) and external validation sets (p < 0.0001; 2-y LRFS 61.9% vs 22.4%), respectively. The hybrid radiomics nomogram established by integrating radiomic signature, deep-learning signature with clinical factors including T stage and concurrent chemotherapy outperformed any one or two combinations in training (C-index: 0.82), internal (Cindex: 0.78), and external validation sets (C-index: 0.76). Calibration curves showed good agreement.

Conclusions: The hybrid radiomics based on pretreatment contrast-enhanced computed tomography provided a promising way to predict local recurrence of ESCC patients receiving definitive (chemo)radiotherapy.

Keywords: Computed tomography; Deep learning; Esophageal squamous cell cancer; Local recurrence-free survival; Radiomics.