Published at 20/03/2023 Last update 19/06/2023

To make more efficient the diagnosis of the patients, SARS2DETECT is a tool capable of performing image analysis of COVID radiographs together with other clinical data for the design and implementation of a triage and diagnostic system. In addition, SARS2DETECT uses Machine Learning models to predict the evolution of COVID-19 patients using information provided by the Department of Health of Government of Navarra, specifically information obtained from chest X-rays of these patients.

Thanks to this tool it is possible to optimise the prediction of response of COVID-19 patients in Navarra (Asymptomatic, hospitalisation, intensive care unit, death) based on chest X-rays, and including clinical variables such as blood O2 saturation, Body Mass Index (BMI) or the presence of previous diseases (obesity, diabetes, pulmonary diseases, neoplasm, etc.).

The results and reports of the tool are used by healthcare personal for decision-making, but it is also possible to obtain an aggregation of the data for publication on the Open Data Portal of the Government of Navarra, respecting the data protection rights of patients and the intellectual property of the Government of Navarra.

On the other hand, SARS2DETECT is a tool that can be extended to any health system, since it uses standard data obtained from patients’ medical records as input, as well as X-rays of the patients, being easily scalable and with the capacity to adapt and evolve to new needs that may arise with the aim or further refining the results obtained by the tool.

Finally, it should be noted that the Project has been evaluated with data obtained from 3.881 patients, taking into consideration hospitalised and non-hospitalised patients, processing, and segmentation with neural networks to obtain the masks of the radiographs, extracting their main characteristics. Subsequently, a prediction system has been generated using Logit model for the classification of patients into serious / non-serious, and a Random Forest model for classification into non-hospitalised / hospitalised / intensive care / death.


Problem or opportunity

SARS2DETECT has been developed in response to the health emergency caused by the COVID-19 pandemic. As in the rest of the world, the effects of the pandemic in Navarre have caused hundreds of deaths and an occasional collapse of health care system. 

The strategic objectives of this tool are:

  • To develop a system for predicting the evolution of COVID-19 patients.
  • To improve the use of the healthcare resources available in hospitals in Navarre

To achieve this strategic objective, the tool identifies critically patients infected with COVID-19, classifying them according to their severity.

Once identified, a second classification is made based on different patient data available to predict the level of risk and morality rate.



The pandemic caused by COVID-19 has had a major impact on health services. The lack of knowledge about the pandemic initially led to overcrowding in hospitals, as it was not possible to predict how patients would evolve.

The development of SARS2DETECT helps to predict the evolution of patients infected with COVID-19, facilitating decision-making by healthcare staff, and optimising the medical resources available.

Thanks to this application, it is possible to triage patients with reliable and contrasted data.

In this sense, it is important to point out the important possibilities of versatility of the tool to be adapted to other possible pathologies that require the extraction of radiographic data for the diagnosis and prediction of the evolution of diseases, such as the specific case of retinopathies.



Within this Project, two different models have been used for the generation of predictive models with different objectives.

Firstly, a Logit model has been used. The objective of using this model is to classify COVID – 19 patients into two states. Severe patients requiring hospitalisation or non-severe patient not requiring hospitalisation.

A second model is offered in case the radiologist aims to obtain a more specific information about the evolution of the COVID-19 patient: The patient does not require hospitalisation, hospitalisation, requires intensive care, high risk of death.

Prior to the use of prediction models, a neural network is used on radiographs to obtain radiomics data. This technology consists of extracting quantitative characteristics (texture, shape, and level between pixels) from the images. Thanks to the use of these techniques, the images are cleaned and segmented by selecting the part corresponding to the lungs and eliminating artefacts.