Oncology diagnosis, particularly for Non-Small Cell Lung Cancer (NSCLC), plays a vital role in hospital procedures. However, despite the promising accuracy of existing algorithms, they are still not widely applied in clinical practice. This study aims to address the limitations of current approaches, focusing on NSCLC, a significant global health issue. We emphasize the role of normalization techniques, such as StainGAN and the Ruifrok and Johnston methods, which help reduce variability in histological staining and enhance data consistency for analysis. To improve model performance, we propose utilizing transfer learning with Hematoxylin and Eosin (H&E) stained images, a common diagnostic tool in pathology. Our methodology involves two strategies: analyzing whole-slide images to capture a broad view of molecular and environmental features, and focusing on smaller image patches for more detailed, localized analysis. We will employ well- established models such as ResNet and U-Net based models, chosen for their strengths in feature extraction and image segmentation. Both strategies will undergo thorough evaluation to ensure clinical applicability and robust performance. Ultimately, this research aims to advance AI-driven pathology for carcinoma to improve diagnostic accuracy, support more informed treatment decisions, and enhance patient outcomes in the context of precision medicine.