Lung cancer detection often relies on interpreting subtle nodules in CT scans, a task demanding precise segmentation tools beyond simple image classification models. While existing methods utilizing other architectures might achieve decent accuracy, they often struggle with limited CT scan datasets and scalability, hindering their real-world impact. Our paper addresses the imperative need for enhanced lung cancer detection by using the modified U-Net architecture, which is implemented to achieve better results on image classification tasks while using low computational resources, it means achieve high accuracy can be achieved with few parameters and makes computation less expensive compared to other models, with the Luna16 dataset. Our proposed model excels in feature extraction and overcoming vanishing gradients, ensuring sharper and more accurate nodule segmentation in CT scans. The Luna16 dataset, a diverse and annotated benchmark, facilitates comprehensive learning and adaptability to various nodule types and imaging conditions.