Implementation of ResNet-50 for the Detection of ARDS in Chest X-Rays using transfer-learning

Abstract

Acute Respiratory Distress Syndrome is a severe condition with high morbidity and mortality. The current standard for the diagnosis of ARDS was proposed by the Berlin-Definition in 2012. However, studies have shown, that ARDS is often recognized too late or not at all. Smart methods, like machine learning algorithms, may help clinicians to identify ARDS earlier and therefore initiate the appropriate therapy. To address the imaging assessment of the Berlin-Definition, a deep learning model for the detection of ARDS in x-rays is proposed. The model achieved an AUC score of 92.6%, a sensitivity of 87% and a specificity of 97%.