The use of artificial intelligence (AI) in research and within clinical settings is becoming increasingly apparent. Recent studies, summarised by the Radiological Society of North America (RSNA), have now explored the potential benefits of AI to help during the current COVID-19 pandemic.
Chest imaging to screen for COVID-19
The use of chest CT scans is currently not recommended for routine COVID-19 diagnosis. Despite cost constraints and concerns regarding contamination, chest CTs still provide an important piece of data to understand COVID-19. It helps in the exclusion of other possible causes for symptoms and provides critical data to monitor a patient’s progress. A recent editorial, published in Radiology: Artificial Intelligence, highlights the potential of AI in accelerating solutions to detect, contain and treat COVID-19.
At the start of the pandemic, chest imaging was a primary tool to screen for COVID-19 in China. However, as there were too many images to review, researchers initially used AI to automate diagnosis and aid radiologists in understanding the findings. Despite these earlier methods showing reliable results, its use was still limited to screening purposes.
Chest radiography to screen for COVID-19
According to a recent Radiology study, chest radiography (CXR) can play a key role in detecting and monitoring COVID-19 infection. This is particularly important in low-resource environments. The team evaluated the performance of CAD4COVID-XRay, an AI system designed to detect COVID-19 pneumonia on CXR. They found that the system performed at a similar level, regarding specificity and sensitivity, to that of six radiologists. Evidently, this system provides a fast and relatively inexpensive resource for many resource-constrained healthcare settings. Most importantly, the system developers have made the tool freely accessible in support of the crisis.
AI enables COVID-19 diagnosis
The need to incorporate clinical information with imaging is critical for diagnosis. CT alone may have limited negative predictive value. Here, AI can serve a purpose. During a recent study in Nature Medicine, researchers used AI algorithms to integrate chest CT findings with clinical symptoms to rapidly diagnose patients. In a test set of 279 patients, the AI system achieved an area under the curve of 0.92. It had an equal sensitivity (82.3%) compared to a senior thoracic radiologist (74.6%). The team suggest that the AI model could run alongside radiologists on a simple workstation.
AI has the ability to fuse data from diverse sources and crunch this information to find patterns that may not be found by humans alone. Therefore, by gathering all the information – including molecular, clinical and epidemiological data – and utilising the power of AI, we can learn the dominant trends of the disease and use this to advance how we treat the virus.