An artificial intelligence-assisted comprehensive model for predicting lymph node metastasis in superficial esophageal squamous cell carcinoma: a diagnostic study.
Authors
Affiliations (4)
Affiliations (4)
- Department of Thoracic Surgery, The First Affiliated Hospital of Anhui Medical University, Hefei, China.
- Department of Thoracic Surgery, Fuyang Cancer Hospital, Fuyang, China.
- Department of Radiology, The First Affiliated Hospital of Anhui Medical University, Hefei, China.
- The School of Medical Information Engineering, Anhui University of Chinese Medicine, Hefei, China.
Abstract
Superficial esophageal squamous cell carcinoma (ESCC) is increasingly detected due to enhanced endoscopic surveillance, yet accurately predicting lymph node (LN) metastasis remains a challenge. This study aimed to construct a combined model integrating CT-based radiomic features and clinical risk factors to predict LN metastasis in superficial ESCC. We retrospectively analyzed 473 patients with superficial ESCC from two centers (403 in a training/internal validation cohort and 70 in an external validation cohort). Contrast-enhanced CT scans were used to extract radiomic features, which were input into a deep neural network (DNN) to develop a radiomics model. Univariate and multivariate logistic regression analyses identified clinical predictors, which were used to establish a clinical model. A combined nomogram integrating the radiomic score and selected clinical predictors was then constructed. Model performance was assessed in both internal and external validation cohorts using the area under the receiver operating characteristic curve (AUC), calibration plots, and decision curve analysis. LN metastasis was present in 56 (13.9%) patients in the training/internal cohort and seven (10.0%) patients in the external cohort. In the training cohort, higher pathological T stage, poorer tumor differentiation, and larger tumor size were significantly associated with LN metastasis (all P < 0.05) and were independent predictors in multivariate analysis. The combined model achieved the highest predictive accuracy, with AUCs of 0.95 (95% CI: 0.91-0.99) in the training cohort, 0.93 (95% CI: 0.89-0.97) in the internal validation cohort, and 0.88 (95% CI: 0.83-0.92) in the external validation cohort. An integrated clinical-radiomics model offers effective prediction of LN metastasis in superficial ESCC, thereby potentially improving treatment decision-making and enabling personalized therapeutic strategies for patients with superficial ESCC.