MULTI-LABEL CHEST X-RAY ORGAN SEGMENTATION

Dumaev Rinat, Molodyakov Sergey

Organ segmentation on chest radiographs is an important task for accurate and reliable diagnosis of diseases of the lungs and chest organs. One important step for automated analysis of radiographs is to isolate the organ of interest from other less important parts to apply decision-making algorithms. This study proposes a method based on encoder decoder architecture with CNN residual blocks to define lung, heart, and clavicles regions. The effectiveness of the proposed architecture and the operations of augmentation and image processing during segmentation of organ areas on an X-ray image was evaluated.

Keywords:machine learning, convolutional neural networks, chest x-ray, pneumonia diagnostics, organ segmentation