Can a machine learning program use chest CT with clinical symptoms to augment early detection of SARS-CoV-2 infection?

Bottom Line:

Artificial intelligence can help diagnose COVID-19 in patients with equal sensitivity to a senior radiologist, allowing rapid diagnosis particularly when there is a shortage of RT–PCR test kits.


Mei, X. et al. Artificial intelligence–enabled rapid diagnosis of patients with COVID-19. Nature Medicine (2020).

Date Published:

19 May 2020


Current diagnosis of COVID-19 largely centres around the use of a virus-specific RT–PCR (a test for the viral genetic material). However, this test requires up to two days and is in limited use in many clinical settings due to shortages in many places in Canada and around the world. Here, the researchers propose the use of a machine learning program that incorporates information from chest CT and clinical data (symptoms, patient history) to diagnose COVID-19 with comparable sensitivity to two senior radiologists. Importantly, the use of clinical data in addition to chest CT images is thought to capture patients who may have COVID-19 without definitive radiological findings. This program, used in conjunction with the expertise of a trained radiologist, has the potential to increase testing potential to a greater number of individuals (essential to controlling the spread of a pandemic) and further advance diagnostic precision in radiology.

Summary by: Mike Ge