The automatic detection of lung boundaries is vital in medical imaging analysis for pulmonary disorder assessment. This study employs an open-source dataset to assess deep learning frameworks, including U-Net, PSPNet, U-Net++, and the proposed EfficientNet-B7 U-Net framework for lung segmentation evaluation. The models utilize various metrics to evaluate performance. The EfficientNet-B7 U-Net model achieved the highest segmentation results, demonstrating 98.19% accuracy. The experimental result supports that complex neural network designs can improve segmentation accuracy and reliability. The findings suggest that enhancements to pulmonary health care result from identifying methods to build efficient diagnostic and therapeutic tools for lung diseases.