COPD from Chest X-rays

       After collecting a dataset of chest x-rays and pulmonary function tests of the same patient within six months of the x-ray, we performed several investigations in terms of automatic medical image assessment and the interpretability of associated models. These investigations include:

• modeling the modifications caused by disease severity in the chest x-rays (check here),
• using deformation fields to model disease effect to find dataset biases (check here),
• proposing a mathematical formulation to associate interpretability alignment with adversarial robustness, and
• showing that the chest x-rays provide more information about COPD diagnosis than radiologist reports (check here).