Prediction of EGFR Gene Mutations in Lung Cancer PET/CT Images Using Tumor Spatial Structure and Texture Features

Lung cancer is a highly diverse disease with various pathological types, gene mutations, phenotypes, and metabolic differences, resulting in complex treatment and prognosis. In particular, epidermal growth factor receptor (EGFR) gene mutations in non-small cell lung cancer (NSCLC) have become a key factor affecting the course of the disease. Although studies have confirmed the effectiveness of EGFR-targeted therapy, there remains a lack of models capable of accurately predicting the status of EGFR gene mutations.A non-invasive prediction approach based on PET/CT images is proposed in this study. By incorporating the uneven distribution of standardized uptake value (SUV), an index of tumor spatial structure heterogeneity is developed to evaluate the heterogeneity of tumor space structure. This index was combined with the texture features of CT images and clinical information, and an ensemble learning method was used to construct a more accurate and reliable multimodal EGFR gene mutation prediction model. The generalization ability and practicality of the model were verified by evaluation indicators such as the area under the receiver operating characteristic curve (AUROC) and calibration curve. The results showed that the model has excellent accuracy and reliability in predicting EGFR gene mutations. In addition, as a non-invasive prediction method, it can reduce patients' discomfort and disease risk and is expected to play an essential role in the field of precision medicine, providing new research directions and decision-making support for personalized treatment of NSCLC.