Artificial Neural Network for Prediction of Lymph Node Metastases in Gastric Cancer: a Phase II Diagnostic Study

Background: Extension of lymphadenectomy in gastric cancer is controversial, and preoperative diagnosis of lymph node metastases (LNM) is difficult. Therefore, knowledge-based systems such as the Maruyama computer program (MCP) are being developed. MCP shows good prognostic value for the compartments; however, for different lymph node groups (LNG) there are a large number of false positives. The aim of this study was to evaluate artificial neural networks (ANN) for predicting LNM in patients with gastric cancer and to compare the predictive power with that of MCP.

Methods: A total of 135 consecutive patients who underwent D2 gastrectomy were included. We applied a single-layer perceptron to the data of 4302 patients from the National Cancer Center, Tokyo, and compared the results with those from the MCP.

Results: Prediction of N(+) or N0 with ANN-1 (Borrmann classification, T category, and tumor size and location) had an accuracy of 79%. The predictive value for LNM in each of the LNG varied: ANN-1, 64% to 86%; MCP, 42% to 70%. We constructed another ANN by using the additional parameter of metastases in LNG 3 as an example of sentinel node. The accuracy of this ANN was 93%.

Conclusions: Using an ANN, LNM in each LNG can be accurately predicted. Additional knowledge about one lymph node improves the results.