Artificial Intelligence applied to chest X-Ray images for the automatic detection of COVID-19. A thoughtful evaluation approach

TL;DR

This paper evaluates deep neural network methods for automatic COVID-19 detection from chest X-Ray images, aiming to improve diagnosis accuracy and explainability through different preprocessing schemes and critical variability analysis.

Contribution

It presents a comprehensive evaluation of CNN-based models trained on a large dataset, highlighting the impact of preprocessing and variability issues on performance.

Findings

Achieved 91.5% classification accuracy.

Preprocessing and lung segmentation improve explainability.

Critical variability analysis identifies factors affecting system performance.

Abstract

Current standard protocols used in the clinic for diagnosing COVID-19 include molecular or antigen tests, generally complemented by a plain chest X-Ray. The combined analysis aims to reduce the significant number of false negatives of these tests, but also to provide complementary evidence about the presence and severity of the disease. However, the procedure is not free of errors, and the interpretation of the chest X-Ray is only restricted to radiologists due to its complexity. With the long term goal to provide new evidence for the diagnosis, this paper presents an evaluation of different methods based on a deep neural network. These are the first steps to develop an automatic COVID-19 diagnosis tool using chest X-Ray images, that would additionally differentiate between controls, pneumonia or COVID-19 groups. The paper describes the process followed to train a Convolutional Neural Network with a dataset of more than 79,500 X-Ray images compiled from different sources, including more than 8,500 COVID-19 examples. For the sake of evaluation and comparison of the models developed, three different experiments were carried out following three preprocessing schemes. The aim is to evaluate how preprocessing the data affects the results and improves its explainability. Likewise, a critical analysis is carried out about different variability issues that might compromise the system and the effects on the performance. With the employed methodology, a 91.5% classification accuracy is obtained, with a 87.4% average recall for the worst but most explainable experiment, which requires a previous automatic segmentation of the lungs region.