Automatic Quantification of Lung Infection Severity in Chest X-ray Images

Abstract

A large number of well-maintained datasets are needed for the diagnosis and assessment of the severity of the new disease (COVID-19) using chest radiographs (CXR). To achieve the best results, current methods for quantifying severity require complex methods and large datasets for training. Medical professionals must have access to systems that can quickly and automatically identify COVID-19 patients and predict severity. In this work, we measure the severity of COVID-19 using an efficient neural network consisting of a CNN backbone and a regression head to automatically predict lung infection scores. In addition, we investigate the efficiency of some augmentation methods to increase the potential of the deep model. A comparative study was conducted using several state-of-the-art deep learning methods on the public RALO dataset. The experimental results show that our model has the potential to perform best on severity quantification tasks and demonstrate the impact of lung segmentation on performance.