Skip to Main Content
To discuss application value of BP neural network to recognizing gastric cancer cell. Object and Method: 510 cells are selected from 308 patients, in which 210 gastric adenocarcinoma cells and 300 non-cancer gastric cells are extracted. Ten morphological parameters of every cell are measured. These data are randomly divided into two groups-training data (A) and test data (B). A three-layer BP neural network is built and trained using data A. Then, the network is test with data A and data B. Result: For data A, the sensitivity of the network is 99.05%, the specificity 98.67%, positive predictive value 98.11%, negative predictive value 99.33%, the accuracy 98.82%. For data B, the sensitivity of network is 99.05%, specificity 97.33%, positive predictive value 96.30%, negative predictive value 99.32%, the accuracy 98.04%. With ROC curve evaluation, the area under ROC curve is 0.9921. Discussion and Conclusion: the result shows that the model built based on BP neural network is effective. BP neural network can be used for automatically recognizing gastric cancer cell.