Predicting EGFR gene mutation status in lung adenocarcinoma based on multifeature fusion

Abstract

Purpose:Adenocarcinoma is the most common histopathological type of lung cancer, and histopathological images are important for assessing the stage and subtype of tumors. Patients with lung adenocarcinoma may have genetic alterations in oncogenic drivers, including EGFR, ALK, ROS1, BRAF, among others. Among them, the EGFR mutation-positivity in Asians is as high as 40%–60%. However, no research has explored the relationship between the morphological information of pathological cells and the oncogenic driver gene EGFR. Methods:In this paper, we propose a novel automatic analysis framework and joint semantic, spatial, and clinical information for exploring the relationship between tumor microenvironment features and EGFR gene mutations, namely, JSSC-Net. Our approach consists of the following modules: the semantic feature encoding module, spatial feature encoding module, clinical information feature encoding module, and feature fusion module. Results:We collected 248 digital WSIs images acquired from 248 patients with lung adenocarcinoma, and verified the effectiveness of our proposed model on this dataset. Our proposed method achieved the accuracy of predicting EGFR mutation state was 0.7778, performing significantly better than the baseline methods. In addition, we attempted to learn the associations between histopathological images and the types of EGFR gene mutation, achieving an accuracy was 0.6428. Conclusions:In this study, we proposed a novel deep learning-based method to predict the mutation state of EGFR gene in lung adenocarcinoma. The experiments proved that the conformation of tumor pathology is associated with EGFR mutation-positivity, that is, the tumor microenvironment of different types is predictive of genetic alterations in oncogenic drivers. Significance:We have demonstrated that the proposed deep learning-based analysis framework can automatically and efficiently predict the relationship between tumor microenvironment features and EGFR gene mutations. This information is critical for applying appropriate and targeted therapies to lung cancer patients, thereby increasing the scope and performance of precision medicine.