Lung cancer remains the leading cause of cancer-related mortality, highlighting the need for accurate and comprehensible diagnostic techniques. This paper introduces Advanced DeepLungCareNet, a deep learning framework that utilizes a refined ResNet50 architecture to classify lung cancer stages from CT scans. The model incorporates global average pooling, dense layers, dropout, and adaptive learning strategies to enhance generalization. After being trained on these IQ-OTH/NCCD dataset with thorough preprocessing, the model achieved a test accuracy of 90.28%, with perfect precision and recall for malignant cases and AUC scores approaching 1.00. Grad-CAM visualization improves transparency by identifying areas that affect predictions, thereby aiding clinical interpretability. The system effectively minimizes false negatives, which are crucial in cancer diagnosis, and shows promise for practical application. Future research directions include the integration of multimodal clinical data, the use of federated learning to maintain privacy, and the execution of prospective trials to ensure clinical reliability.