Combining machine learning and deep learning with clinical data and chest X-rays can point to either mild or severe disease.
A hybrid artificial intelligence (AI) approach that combines both machine (ML) and deep learning (DL) can predict the severity of a patient’s case of COVID-19 by using both clinical data and chest X-rays, say researchers from Italy.
Investigators presented their findings – and announced their open-access database – during the European Congress of Radiology (ECR) 2021 annual meeting.
For additional ECR 2021 coverage, click here.
By using 820 chest X-rays and clinical data, including demographics, blood analysis, oxygen saturation, and co-morbidities, collected from six hospitals and two research institutions in Italy, the team was able to predict – with accuracy, as well as high sensitivity and specificity – which patients would have more severe disease, said presenter Marco Alì, Ph.D., a fellow in integrated biomedical research from Centro Diagnostico Italiano.
The team defined patients with mild disease as those who were sent home to recover or who were hospitalized without ventilation. Patients with severe disease were hospitalized with ventilation, treated in the intensive care unit, or they died.
The team compared their chest X-ray/clinical data ML-DL hybrid model to strategies that applied ML and DL only to clinical data or only to chest X-rays. According to their analysis, their hybrid model performed the best with 77 percent accuracy, 79 percent sensitivity, and 75 percent specificity.
When they drilled down into the data, they determined that 87 percent of patients had at least one co-morbidity with hypertension being the most common, affecting 45 percent of patients overall. Specifically, among patients who experienced severe COVID-19, 27.9 percent had hypertension, and 21 percent had pre-existing cardiovascular disease.
In addition, Alì said, men were more likely to suffer severe disease. They accounted for 72 percent of deaths.
The images used to test this hybrid strategy can be found here.
For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-Newsletter here.