Intelligent analysis of X-ray images for early detection of COVID-19 with Computer Vision

CVC researchers, lead by Dr. Debora Gil, and members of the Research Unit Support of IDIAP Jordi Gol, propose a novel method for early diagnosis and follow-up of COVID-19 patients from X-ray imaging analysis with machine learning algorithms for increasing efficiency in primary health centers.


With nearly 6 million confirmed cases of COVID-19 from over the globe, scientific efforts have been working intensely towards finding tools and help fight back a pandemic which has put our modern society on a hold and our economies on ice. Dr. Debora Gil and her team, composed by Dr. Aura Hernandez, Dr. Carles Sanchez and Dr. Katerine Díaz, senior researchers from the Computer Vision Center, have been actively researching within the medical imaging domain for the past twenty years, with a clear focus on the pulmonary system, and have now put all their knowledge and efforts towards an early and effective diagnosis of COVID-19. 

Preliminary experiments have shown the degree of feasibility of COVID-19 screening with the use of X-ray. In comparison with other methods, accuracy of the classical approach proposed by CVC researchers achieves an average score of 90% or above in the identification of COVID-19 in X-ray images, properly distinguishing them from pneumonia. “What we have seen”, states Dr. Gil, first author of the study “is that our approach scores better than other more modern models, such as those based on Deep Learning”.  In a combined project with researchers from the Northern Metropolitan Research Support Unit of the Catalan Institute of Health and the IDIAP Jordi Gol, the team will proceed to test the model with regular clinical data. With this aim, a collaboration agreement has been signed and approved by both institutions. 

Results indicate that there is the possibility of differentiating COVID-19 from other pulmonary diseases with the use of X-ray images, and thus optimize PCR tests in primary health facilities. “Nonetheless”, adds Dr. Gil, “COVID-19 is easily mistakable with pulmonary infiltrations and thus, we need more clinical data in order to train a more efficient model”. The methodology proposed by Dr. Gil and her team, presented in their recent study, has been able to detect a total of 90 COVID-19 cases with less than 1 of each 5 cases misinterpreted by the algorithm. 

“With this study, what we have detected is that information extraction from X-ray images with computer vision is possible for the prompt detection of the disease”, states Dr. Aura Hernàndez, co-author of the paper. The aim of the team is now to compile a standardized Database of COVID-19 in X-ray images and train the model further. “With this information”, Dr. Hernàndez highlights, “the goal is to classify x-ray images and thus discriminate COVID-19 pneumonia from other types of pneumonia and lung diseases. This would lead us to a screening tool which could be provided at primary health centers and thus to an early disease diagnosis”. 

Furthermore, the project has a third objective. As Dr. Carles Sànchez, member of the team and co-author, states, diagnosis is key, but monitoring is essential if we want to follow the patient’s evolution. “We need to identify at very early stages the normal and non-normal X-ray images with infiltrations and visual progression patterns which may be characteristic of COVID-19 for predicting possible complications which may require hospitalization”, Dr. Sànchez explains. 

The central part of the project is the use of x-ray images and not Chest CT scanners for the early diagnosis of the pulmonary disease. Chest CT scanners provide a high sensitivity for diagnosis of COVID-19, as previous works show. However, CT scanners are expensive and there is a difficulty in logistics in order to provide massive screening to a high number of patients. X-ray scanners, however, can be easily deployed in most primary health centers and can provide an affordable rapid triaging method. 

Machine learning algorithms applied to medical imaging can provide intelligent analysis of X-ray images”, states Dr. Katerine Díaz, co-author of the paper, “our preliminary study proves that. Now, we need to find the means and resources to continue and develop the project further”. With their model, researchers are opening the window for a set of smart tools applied to medical imaging which will aid public health professionals in their response to the current pandemic. Furthermore, setting the baseline for trained models which can be easily adapted to other pulmonary diseases which might strike again in the future. 


Gil, K. Díaz-Chito, C. Sánchez, A. Hernández-Sabaté (2020): Early Screening of SARS-CoV-2 by Intelligent Analysis of X-Ray Images. [Pre-Print in Arxiv].

In the Media:

Diari Ara (2020): Creen una nova eina per a la detecció precoç del covid-19 amb raigs X