Fig. 1.

Overview of the method for automatic quantification, staging and prognosis of COVID-19. Our study includes 8 independent cohorts, resulting in 693 COVID-19 patients in total. A variety of clinical and biological attributes were collected and combined with imaging biomarkers for short and long term prognosis of COVID-19 patients. Our study is composed by three different steps: (i) Proposing a state-of-the-art deep learning based consensus of 2D & 3D networks for automatic quantification of COVID-19 disease, reaching expert-level annotations, (ii) A radiomics study integrating interpretable features extracted from disease, lung and heart regions. A consensus-driven COVID-19 low dimensional bio(imaging)-holistic profiling and staging signature has been proposed using robust machine learning algorithms, fusing imaging, clinical and biological attributes. & (iii) An ensemble of robust linear & non-linear classification methods for the proper identification of patients that need intubation.