Non-FAU Project
Start date : 01.03.2023
End date : 01.03.2026
Chest X-Rays (CXR) serve as crucial diagnostic tools for pulmonary and cardiothoracic diseases, generating millions of images daily, a number on the rise due to decreasing acquisition costs. However, there's a pronounced scarcity of radiologists to interpret these images. Traditionally, CXR research has centered on enhancing classification accuracy, often achieving state-of-the-art results. Despite progress, there remain rare and intricate findings challenging for both human radiologists and AI systems to diagnose. Our investigation focuses on leveraging self-supervised image-text models to enhance the classification and localization of diverse findings. These self-supervised models eliminate the need for annotations, enabling the Deep Learning system to effectively learn from extensive public and private datasets.