Medical segmentation of images makes heavy use of deep learning (DL) techniques. Nevertheless, the fact that prior research has used various datasets for evaluation and that there is no standard basis of reference for performance evaluation undermines the field's dependability. This work presents a thorough comparison of deep learning models for lung X-ray image segmentation, which can be utilized as a standard to evaluate other medical image segmentation model. An important part of this endeavor is played by the powerful U-net architecture, which is well-known in the field of deep learning for image segmentation. Our goal is to accurately segment the lung, especially for damaged sections, using a different comprehensive approach that makes use of deep learning, transfer learning, and semantic segmentation. Notably, the proposed lung segmentation approaches do comparative analysis of different segmenting approach for u-net, Linknet, PSPNet and FPN models. PSPNet model got highest mean IOU_Score, F1-score. Execution time of each model is calculated for performance analysis. These approaches greatly advance the realm of medicine and the use of artificial intelligence to automatically segment lung images.