EMCTNet: EGFR Mutation Detection from CT Images in NSCLC Patients Using EfficientNet Model

Lung cancer is a common disease and is considered the leading cause of cancer-related death. Early detection and treatment are essential to reduce the mortality and morbidity rate. Epidermal growth factor receptor (EGFR) mutation is a critical factor in the treatment of non-small cell lung cancer (NSCLC) patients. So, an accurate, automatic, and convenient method is needed for EGFR mutation detection in NSCLC patients. In this paper, we developed a novel method for EGFR mutation detection from CT images called EMCTNet. Firstly, we pre-processed the lung nodules in the CT images and then fine-tuned the EfficientNet-B۱ to extract salient features from lung nodules. Finally, each patient is classified as a mutant or wild-type EGFR category. To overcome the imbalanced learning problems, we generate balanced batches with a random oversampling method. The performance of our proposed method was validated on NSCLC Radiogenomics TCIA data with ۲۱۱ patients in a stratified ۱۰-fold cross-validation manner. The accuracy and the area under the curve of the proposed method achieved ۸۶.۷۸% and ۰.۷۴, respectively. Our results showed the effectiveness of deep learning methods in automatically detecting EGFR mutations in NSCLC patients.

کلیدواژه ها:

EGFR mutation ، Lung nodules ، CT images ، EfficientNet-B۱ ، NSCLC