Lung cancer stands as a leading worldwide health danger making it important to find good ways to detect the disease early. This investigation develops a lung cancer diagnosis system through image segmentation that uses Enhanced ERFNet to process medical images. Researchers commonly use SVMs and U-Net for tumor segmentation but these methods perform slowly and detect poorly in small irregular regions. The proposed method uses multiple feature sizes and attention systems to find tumor specifics in different image scales while focusing on important image areas. High-resolution CT scans and X-ray pictures help us define tumor areas with precision including normal and cancerous regions. The Enhanced ERFNet architecture combines the efficiency of the original framework with new design improvements that increase medical image segmentation accuracy. Tests on various cases show that the model achieves better results than current methods at differentiating tissue areas and identifying cancerous growths. The model runs fast enough to support real-time medical use. This work creates a practical and fast system to predict lung cancer which helps doctors find the condition earlier and initiate treatment sooner.