An artificial neural network model predicting pathologic nodal metastases in clinical stage I–II esophageal squamous cell carcinoma patients

BACKGROUND: Current preoperative staging for lymph nodal status remains inaccurate. The purpose of this study was to build an artificial neural network (ANN) model to predict pathologic nodal involvement in clinical stage I–II esophageal squamous cell carcinoma (ESCC) patients and then validated the performance of the model. METHODS: A total of 523 patients (training set: 350; test set: 173) with clinical staging I–II ESCC who underwent esophagectomy and reconstruction were enrolled in this study. Their post-surgical pathological results were assessed and analysed. An ANN model was established for predicting pathologic nodal positive patients in the training set, which was validated in the test set. A receiver operating characteristic (ROC) curve was also created to illustrate the performance of the predictive model. RESULTS: Of the enrolled 523 patients with ESCC, 41.3% of the patients were confirmed pathologic nodal positive (216/523). The ANN staging system identified the tumour invasion depth, tumour length, dysphagia, tumour differentiation and lymphovascular invasion (LVI) as predictors for pathologic lymph node metastases. The C-index for the ANN model verified in the test set was 0.852, which demonstrated that the ANN model had a good predictive performance. CONCLUSIONS: The ANN model presented good performance for predicting pathologic lymph node metastasis and added indicators not included in current staging criteria and might help improve the staging strategies.