A hybrid deep learning framework is developed in the proposed work that combines Convolutional Neural Networks (CNNs) with U-Net architecture for simultaneous classification and segmentation of lung diseases using the LIDC-IDRI-slices dataset (875 directories). It contains aligned image and mask slices, enabling precise supervised learning. By employing a multi-task learning strategy, the model partitions the two tasks to improve accuracy and generalization. The designed CNN captures lung patterns using multi-scale convolutions, skip connections, and dense layers, which helps in the detection of subtle lung abnormalities. The use of transfer learning with fine-tuning on available pre-trained models optimizes the amount of labeled data needed for training, thus simplifying the cost and effort associated. Attention-based techniques enable the model to focus on relevant areas of the lungs, thus decreasing noise and improving feature extraction. With the encoder-decoder configuration and skip connections, the U-Net component performs lung tissue segmentation and complex structure segmentation far more accurately than before, including nodule structures. Boundary detection and context preservation improve other segmentation techniques. The model benchmarks highly on sensitivity and specificity, surpassing classical models in classification and segmentation. It adapts well across differing datasets and imaging conditions, proving to be dependable for clinical decisions, thereby enhancing patient care and outcomes.