Artificial intelligence (AI) has demonstrated promising outcomes in the healthcare sector, emerging as a critical tool for segmentation and classification tasks that assist doctors in early disease detection. This paper offers a novel modified U-Net technique for lung segmentation in chest X-ray (CXR) images that uses an enhanced median filter to improve image quality before segmentation. Accurate segmentation is vital as it can improve diagnostic efficiency by focusing on the regions of the ROI (region of interest). The proposed model has dense skip connections and an attention mechanism that keeps important information safe during the segmentation process while focusing on key features. We also evaluate the proposed model against the SegNet, FCN, and U-Net benchmarks. The experimental results demonstrate that the proposed model performs exceptionally well in the training and validation phase, achieving a remarkable accuracy of 98.33\% and 98.17\%, respectively. The model demonstrates high sensitivity (96.88\%) and specificity (97.48\%), alongside exceptional segmentation quality, indicated by a dice coefficient of 96.31\% and a Jaccard index of 93.36\%. The results show that the proposed method benefits lung segmentation tasks, highlighting its potential to help detect diseases earlier in medical imaging applications.