The San Cecilio Clinical University Hospital together with the University of Granada are working on the development of an artificial intelligence model capable of determining if a patient has coronavirus by reading his chest X-ray
From the Radiodiagnosis service of the San Cecilio Clinical University Hospital together with the Andalusian Institute for Research in Data Science and Artificial Intelligence of the University of Granada (UGR) work is underway to develop an automatic system with the ability to detect lung involvement caused by the Coronavirus through chest X-rays of patients. This joint research project began in the middle of last March, and since then around a thousand X-ray plates from different patients have been analyzed, which have contributed to testing and perfecting the model, a concept known as "deep learning" or deep learning.
In this way, what would be the first phase of the investigation has been completed, through which this tool offers specialists knowledge about the patient's lung damage due to coronavirus, by studying their lung radiography, also reducing the time in which the result of the PCR (Polymerase Chain Reaction) is known, which is the test that is currently used as the most validated test to detect the presence of Covid-19 disease.
Jose Luis Martin Rodriguez, head of the Radiodiagnosis service of this Granada hospital and member of the research group of the ibs.GRANADA IBS-A-15: Basic and Clinical Oncology, points out: “The objective is to develop an artificial intelligence tool based on deep learning algorithms that allows us to identify, by means of chest radiography, the presence of pulmonary involvement, even in incipient stages. Therefore, its most immediate real application would allow us to have an automated detection system for COVID-19 in suspected patients.”
Dr. Martín adds that “compared to the time and cost of other tests that have been shown to be the most effective and validated in detecting infection, such as computed tomography (CAT) or PCR, having this model would mean speeding up the times in diagnosis, in addition to allowing the use of the system to be standardized in practically any health center with availability to perform chest x-rays”.
Both the Andalusian Institute for Research in Data Science and Artificial Intelligence, and the team of engineers from the UGR led by Francisco Herrera, an international benchmark in Artificial Intelligence, and Siham Tabik, an expert in Deep Learning, comment that: “The project is close to completion. complete its initial phase and the results obtained so far are encouraging, since we are obtaining levels of precision that exceed those described to date in the international bibliography for this process, comparing them with the available databases. Thus, the precision of the model that is being developed shows a success rate of 80% in detecting positive cases”.
There are similar research projects, launched from other parts of the world. In Canada, they are developing the COVID-NET model, by the DarwinAI company, which has been applied to the radiographs that make up the sample of this Granada study, obtained at the San Cecilio Clinical Hospital, thus placing the detection capacity of the 80% positive cases. The own model developed by the UGR and the Radiodiagnosis service of the Hospital Clínico San Cecilio, obtaining in this case an accuracy of 82% for the same sample.
In future phases of the investigation
In the phases following this study, in a year or two approximately, it is expected to expand the capacity of this model with the capacity to indicate and relate radiological findings in infected lungs, in a way that allows a study of the impact of other factors that currently do not such as age, sex, laboratory abnormalities, drugs or other diseases are taken into account in the future evolution of the coronavirus.
From the UGR, in the hands of experts in artificial intelligence, it is even expected to be able to develop and adapt the system, hoping that it will be able to differentiate between patients affected by Covid-19 and those whose symptoms belong to other types of lung diseases. , such as bacterial pneumonia, other viral pneumonias, tumors, etc., thus distinguishing between different pathologies of lung involvement.
Various artificial intelligence teams and some hospitals in the Andalusian autonomous community and the rest of Spain have shown interest in joining the project, which would benefit the strengthening of the study, by being able to have different and varied data sources (from various hospitals) enriching thus the sample and providing more robustness to the artificial intelligence models.