Related Experiment Video
Updated: Jan 9, 2026
Detection of Targetable Alterations in Non-small Cell Lung Cancer using Next-generation Sequencing
Published on: October 10, 2025
Prediction of EGFR mutation status and mutation sites in lung cancer based on radiomics and deep learning
Zheng Zeng1,2, Rui Zhang3, Yucai Dong3
1Department of Radiology, State Key Laboratory of Oncology in South China, Collaborative Innovation Center for Cancer Medicine, Sun Yat-sen University Cancer Center/Cancer Hospital, Guangzhou, China.
Background:
Epidermal growth factor receptor (EGFR) mutations are regarded as key predictors for the efficacy of EGFR-tyrosine kinase inhibitors (TKIs), with exon 19 mutations demonstrating a better prognosis compared to exon 21 mutations. Therefore, accurate identification of EGFR mutations and their specific loci (e.g., exon 19 vs. 21) is critical for guiding therapeutic strategies in lung cancer. However, the limitations of invasive procedures highlight the need for the development of novel, non-invasive methods to predict the EGFR mutation status and subtypes in lung cancer patients. This study aimed to develop and validate a non-invasive predictive model based on radiomics and deep learning for the identification of EGFR mutation status and its subtypes in patients with non-small cell lung cancer (NSCLC).
Methods:
We retrospectively analyzed clinical and imaging data from 557 NSCLC patients who underwent EGFR genotyping. Radiomics features were extracted from computed tomography (CT) images, and machine learning models, including support vector machine (SVM) and logistic regression (LR), were developed for mutation prediction. Deep learning architectures (ResNet-18 and DenseNet-121) were further employed to extract high-dimensional features from CT images. A combined model integrating radiomics, deep learning, and clinical data was constructed to predict EGFR mutation status and mutation subtypes.
Results:
The combined model achieved superior performance in predicting EGFR mutation status [area under the curve (AUC): 91.1%, accuracy: 85.6%] and specific mutation sites (AUC: 93.6%, accuracy: 89.9%). Compared to radiomics models, the combined model demonstrated statistically significant improvements (P<0.05). Calibration and decision curve analyses confirmed the clinical applicability of the models in patient stratification and personalized treatment planning.
Conclusions:
The integration of radiomics and deep learning offers a non-invasive, accurate method for predicting EGFR mutations and subtypes in NSCLC. This integrated approach demonstrates significant potential to optimize clinical decision-making and enable personalized treatment regimens in NSCLC management.
