Semantic Segmentation of Anatomical Structures in Posterior-Anterior Chest X-Ray Image Using U-Net

The need for X-ray imaging analysis, especially on the thorax or chest area, is increasing during the pandemic. However, understanding an X-ray image requires particular expertise, Things that not everyone can understand and do. Therefore, this research was conducted to assist non-professionals in understanding the structural anatomy of the thorax via a Posterior-Anterior Chest X-ray image. This research uses a computer vision and deep learning-based approach, which has become a trend in recent years, including its use in the medical world. Using a U-Net-based neural network architecture, we perform semantic segmentation tasks to identify human organs, including the heart, bronchi, left lung, and right lung. Our proposed system trained U-Net models and achieved a satisfactory performance with 72% of mean IoU with total 850 images as data train and validation, much better than a comparative model, in carrying out the organ anatomical semantic segmentation task to assist users in medical imaging analysis.