The use of artificial intelligence (AI) to analyze one’s chest x-rays has become a promising shortcut to determine if an individual has been infected with Severe Acute Respiratory Syndrome Coronavirus (SARS-CoV-2), the causative agent of COVID-19. As the method lands itself on various news headlines, it also sparked some debates over reliability as most of these AI systems were built swiftly during this ongoing global health crisis.

Can AI do a good job?

Primarily, skeptics are concerned over designs. Even though fellow developers reassured their models have been modified to fit the pandemic, most of them are trained with a limited number of lung scans from coronavirus patients. It’s challenging to tell if AI is capable of distinguishing between COVID-19 and other respiratory diseases. Moreover, whether the findings appeared on leading journals or preprint waiting to be peer-reviewed, many AI models relied on scans obtained from Chinese patients who were affected during the early onset of the pandemic.

These patients tend to have advanced symptoms. So, there is a question if AI models trained using these samples are applicable to a boarder and more diversified population that are at different stages of infection. More importantly, imaging has been advised to use with care in this period of time as contaminated equipment bear the risks of spreading the virus. As such, regardless of whether AI is involved, most radiologists are against the use of chest radiography and computed tomography (CT) scans for COVID-19 screening.

Nevertheless, from flagging breast cancer from mammograms to predicting heart attacks and strokes, AI has an influential role in radiology. The pandemic is probably just another opportunity to test if it can take up more responsibilities but still, we are unsure when AI can shoulder them all without worrying human. After all, images from real clinical setting will not be as clear or refined as those used to train an AI.

Bigger volume of data is required

Despite so, polymerase chain reaction (PCR), the diagnostic test widely used to detect the presence of COVID-19 by looking for the genetic materials of coronavirus, are in limited supply. As such, AI is often thought to be the next best alternative to fill the gap, especially AI performs more than just screening. It can also monitor if patients are deteriorating over time and predicting patients’ responses to experimental drugs. That’s why some companies chose to market their AI tools as an augmentation for physicians to assess patients’ progression.

Others like Joseph Paul Cohen, Postdoctoral Fellow at the University of Montreal, chose to assist fellow researchers by building a public COVID-19 dataset of chest X-rays and CT scans. He collated images that are published under creative commons licenses or maintained by larger institutions like the National Institutes of Health, Stanford University and Massachusetts Institute of Technology. Cohen noted that building an AI model using solely CT scans will be tough because existing datasets are not large.

Back in March, the American College of Radiology also discouraged the use of chest x rays and CT scans to diagnose COVID-19. In a statement, ACR wrote in bold, “We want to emphasize that knowledge of this new condition is rapidly evolving, and not all of the published and publicly available information is complete or up-to-date” and “Generally, the findings on chest imaging in COVID-19 are not specific, and overlap with other infections, including influenza, H1N1, SARS and MERS”.

Indeed, AI is not perfect but we can’t deny its role and what it is capable of. Perhaps AI will be useful in certain limited cases such as analyzing the scans of individuals who passed the PCR test but are in suspicion of a false negative; highlight any missing cases or cases that need additional attention.


Author Bio

Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.