Integrating Deep Learning and Explainable AI for Non-Invasive Prediction of EGFR and KRAS Mutations in NSCLC: A Novel Radiogenomic Approach

The diagnosis of lung cancer is based on phenotypic and genetic data, providing a foundation for the development of personalized treatment strategies. The integration of ExplainableAI (XAI) methods into the predictive modeling of gene mutations in lung cancer represents a pivotal advancement in theinterpretability and trustworthiness of deep learning algorithms. Gradient-weighted Class Activation Mapping (Grad-CAM) inconjunction with a ResNet18 model is employed to elucidate the features within lung nodule CT images that are indicative of specific genetic mutations, such as EGFR and KRAS. The XAI results, particularly the Grad-CAM heatmaps, serve as a bridge between the deep learning model's computational predictions and the clinical understanding of lung tumor morphology and its genetic underpinnings. This exploration not only enhances the explainability of AI models but also contributes to a more nuanced understanding of the relationship between morphological features and genetic variations, potentially leading to more targeted and personalized treatments for lung cancer patients.