A radiopathomics model for prognosis prediction in patients with gastric cancer

Abstract

Predicting gastric cancer prognosis is imperative for more appropriate clinical treatment plans. Compared with traditional radiomics model adopting CT images alone, the radiopathomics is a novel medical image analysis strategy which employed the radiomcs features extracted from CT image and pathomics features extracted from pathological image to build a prediction model. In this paper, we developed a radiopathomics model to predict whether patients with gastric cancer survive more than 2 years. By using LASSO algorithm, two pathomics features, a radiomics feature and the clinical variables of TNM were selected from totally 1565 features to build the prediction model. For reflecting the advantage of the radiopathomics model, we implemented the comparison tests between the radiopathomics model with radiomics model and pathomics model. The results showed that the radiopathomics model achieved an AUC of 0.904 and an accuracy of 84.2%, which was significantly better than the other two models. It demonstrated that integrated of the microscopic level and macroscopic level phenotype information for tumor could be useful in prediction of prognosis.