Comparison of Deep Neural Networks for Lung Image Segmentation on Chest X-Ray

Chest X-ray lung segmentation is crucial to medical imaging, diagnosis, and therapy planning. Algorithms for categorizing and segmenting image data have been created to assist doctors in diagnosing cases faster and with fewer manual mistakes. Technology has a significant impact on how effectively medical images are segmented. Based on their findings, several scholars have suggested various methodologies. Deep learning techniques, especially deep neural networks, have shown promising outcomes in lung image segmentation. This study will compare and evaluate the effectiveness of several deep neural networks on chest X-ray pictures for lung segmentation. The evaluation is based on several criteria: accuracy, Jaccard index, and dice coefficient. The ResUNet residual block allows us to lessen the challenges associated with feature reduction. Attention ResUnet outperformed other deep learning models by obtaining high evaluation scores with the validation dice coefficient (96.49%), validation mean IoU (94.26%), and validation binary accuracy (98.15%). The significance of this study relies on its thorough analysis of the UNet and ResUnet designs as well as its application of Attention ResUNet to lung X-ray scans. As far as we know, lung image segmentation has not been conducted using the Attention ResUNet architecture. The Attention ResUNet architecture obtained better accuracy with fewer epochs when we ran the UNet and ResUNet models for the same number of epochs. Also, the accuracy of the Attention UNet model was higher (94.03%) than that of the Res UNet model (94%) and UNet (92%).