This paper presents a detailed study on lung image segmentation with special focus on the proposed ResUNet++ framework. Automated segmentation plays a key role in computer-aided diagnosis, as manual methods are often slow, subjective, and inconsistent. Traditional techniques struggle with low-contrast or irregular lung regions, whereas deep learning models such as U-Net, UNet++, ResUNet, Attention U-Net, PSPNet, DeepLabV3+, HRNet, and transformer-based networks have brought major improvements. Among these, our experiments show that ResUNet++ offers the best balance of accuracy, stability, and efficiency. It achieved the highest Dice, Precision, Recall, and IoU scores across datasets and showed strong adaptability to different domains. The ablation studies confirm that modules like squeeze-and-excitation and atrous spatial pyramid pooling play a vital role in handling complex pathologies. Visual results further support its ability to generate smooth and anatomically correct lung boundaries. Overall, ResUNet++ advances the state of the art in lung image segmentation and provides a reliable base for clinical tasks such as disease detection, lesion measurement, and treatment monitoring.