Figure 1. Pipeline of the proposed approach. (a) First, a set of radiological studies were collected from different databases with expert annotations. (b) Then, a deep learning based strategy was trained to detect COVID-19 cases in three steps: b.1. Different convolutional neural network architectures were tested to characterize the radiological studies; b.2. subsequently, the extracted features were flattened to be used as input for the two proposed classification stages; b.3. an end-to-end approach with fully-connected layer classifier, and (b.4) an embedding approach with machine learning classifiers. (c) At the testing stage, new radiological studies were labeled as with or without COVID-19 using the trained models.