Where are the lungs?

Lungs Segmentation in X-Rays

Chest X-ray procedures are considered to be the most popular for diagnosis of chest related diseases. We as a machine learning and medical imaging community, have seen extraordinary interest in the chest x-rays anlaysis and segmentation tasks. For any diagnosis on chest x-rays, accurate segmentation of the biological object is fundamental.

Here, we show how we can use Generative Adversarial Networks(GANs) to perform segmentation of lungs within chest x-rays.

The generator of the GAN generates a segmented mask of a given chest x-ray. Once the generator is trained to generate realistic looking masks, the GAN can now be used to perform segmentation of the lungs on new chest x-ray images.

The model may still perform over-segmentation or can miss some of the pixels. For example, see in Figure below.

Example cases for over-segmentation and under-segmentation. First row shows the input images to the generator, second row shows the output images generated by the generator and third row shows their respected ground truth masks. a), b), and c) show over-segmentation. d) shows under-segmentation attributed to the very low contrast in the lower lung region.

Read the full paper on IEEEXplore.

Segmentation of Lungs in Chest X-Ray Image Using Generative Adversarial Networks

About: The author is a researcher at Umea University Sweden. His google scholar page can be visited at https://scholar.google.com/citations?user=XEJyFFgAAAAJ&hl=en