Automatically Predicting Lung Adenocarcinoma Invasiveness

Accurate staging of lung adenocarcinoma is very important for lung cancer operation and prognosis. However, due to its large heterogeneity, clinical staging is very difficult, and automatic prediction using traditional methods is difficult to extract its heterogeneity. Our study discussed the automatic prediction of lung adenocarcinoma invasiveness using multiple machine learning approaches and analyzed its key factors. We performed a retrospective review of 966 GGN patients. Logistic regression, support vector machine with a Gaussian kernel, random forest and neural network classifiers were constructed to identify lung adenocarcinoma invasiveness, including minimally invasive adenocarcinoma/ invasive adenocarcinoma (MIA/IA). The performance of the four machine learning algorithms was evaluated by area under the ROC curve (AUC), sensitivity, and specificity. As for extracting key factors, we proposed a screening method based on the contribution of factors to neural network. The result demonstrated that the NN classifier could be used as a non-invasive method to automatically predicting lung adenocarcinoma invasiveness and assist GGN management prior to surgery. And predicted MIA/IA patients could use invasive methods to improve their conditions.