Deep Learning Based Lung Image Segmentation Using XR-U-Net

Lung-associated pathologies are anticipated to emerge as the predominant contributor to mortality rates by the year 2029, as reported by the World Health Organization (WHO). In light of this intensifying issue, we propose an innovative deep-learning framework predicated on the U-Net architecture aimed at enhancing lung segmentation from radiographic imagery. Our proposed model, designated as XR-U-Net, augments the conventional U-Net configuration by integrating five encoder and decoder blocks, which significantly elevates segmentation accuracy to 95.7%. When assessed on a collection of lung X-ray images, the model substantiates its efficacy through critical performance metrics, including validation loss, Intersection over Union (IoU), and Dice coefficient. By reducing diagnostic duration and aiding in intricate medical scenarios, this model possesses considerable potential to assist healthcare practitioners in providing swifter and more precise diagnoses, ultimately addressing the deficiency in the availability of medical specialists.