Integrated Deep Learning Pipeline for Lung Cancer Prediction: Combining U-Net, ResNet, and Mask R-CNN with Transformer-Based Enhancements

Lung cancer continues to be a primary contributor to cancer mortality globally, emphasizing the critical need for early detection systems. This research introduces a novel multi-model deep learning framework for automated lung cancer detection from computed tomography scans. My approach in-tegrates three specialized deep learning architectures: an en-hanced U - Net model for precise lung boundary extraction, a self-supervised ResNet50 for malignancy classification, and a probabilistic Mask R-CNN for tumor localization with uncer-tainty estimation. The framework processes medical images through a sequential pipeline, where each model contributes specialized expertise while maintaining computational efficiency. I incorporate advanced techniques including attention mecha-nisms, contrastive pre-training, and Monte Carlo uncertainty quantification to enhance model reliability. Extensive validation on the LUNA16 dataset demonstrates superior performance with 0.94 segmentation accuracy (Dice coefficient), 0.93 classification accuracy, and 0.89 tumor localization precision. My integrated approach surpasses individual model performance and existing methodologies, offering a comprehensive solution for clinical lung cancer screening with interpretable confidence measures.