Improving diagnosis and treatment of lung cancer using a combination of deep learning, radiomics and genomic analysis

Lung cancer, with its high mortality rate, presents a significant challenge in oncology, due to largely asymptomatic early stages, which complicates timely diagnosis. This article explores an innovative integrated approach utilizing deep learning and radiomics for enhancing the detection, characterization, and prediction of genetic mutations in lung nodules, thereby advancing the frontiers of personalized medicine in the context of lung cancer. Leveraging state-of-the-art convolutional neural networks (CNNs), Vision Transformer (ViT), and ResNet18, our methodology extends beyond traditional imaging techniques to analyze computed tomography (CT) images for lung nodule segmentation and to predict significant genetic mutations, specifically EGFR and KRAS. Our integrated approach, particularly combining YOLOv7 with DeepLabv3 and texture features, showed substantial improvements, achieving an accuracy of up to 98.5% and precision of 96% in segmentation tasks, and impressive accuracy rates of 97.8% and 98% for predicting EGFR and KRAS mutations, respectively, using a combination of deep learning and radiomics.