Giorgio Ciano (University of Siena)

When: Mar 10, 2021 – 11:00 – 11:45 AM

Where: Google meet link

Description

Chest X-ray (CXR) is one of the most commonly used techniques for diagnosing pneumonia, tuberculosis, lung cancer, and heart failure. It is of great clinical and scientific interest to develop computer-based systems that support CXR analysis, as accurate interpretation of CXRs remains challenging and requires highly qualified and trained personnel. The use of deep learning on medical images is becoming increasingly popular as a method of supporting doctors in diagnosing. The main problem concerns obtaining the images with the corresponding labels. In this paper, we use Generative Adversarial Networks (GANs) for synthesizing high-resolution CXR images with the corresponding label-maps. We use three different methods for image generation, two of which are already known in the literature. The third method (3-steps method), within the limits of our knowledge, has never been used before. All the images generated are used to train a semantic segmentation network, obtaining very similar, sometimes better results than using real data.