Multicentre study reveals AI can offer reliable screening to triage COVID-19 patients

A study published in PLOS ONE has shown the clinical efficacy of the Lunit Insight CXR AI algorithm for detecting COVID-19 pneumonia on chest radiography compared with formal radiology reports.

Using Lunit’s AI software for chest x-rays, the study was conducted and revalidated by five different major medical institutions, including Kyungpook National University Chilgok Hospital in the Daegu province of South Korea, which had the highest number of COVID-19 cases during the rapid initial development earlier last year.

The study used actual positive cases from five emergency departments and one community treatment centre in the region, between February 18 and May 1, 2020, when the spread began in earnest in the city. In total, chest x-rays of 279 patients were included in the diagnostic analysis.

The results showed that the sensitivity and specificity of Lunit Insight CXR recorded 95.6 per cent and 88.7 per cent respectively. No significant difference was observed in the AUROC curve value of the AI algorithm for the detection of COVID-19 with pneumonia compared with the radiology report.

Lunit ceo Brandon Suh commented: “Considering the previous AI-based studies for COVID-19 showed an average sensitivity of 80 per cent, this study proves the algorithm is significantly improving its performance. As the pandemic is still ongoing, Lunit Insight CXR can potentially be used more actively in actual clinical practice in triaging and monitoring COVID-19.”

Corresponding author Professor Dong-eon Lee at Kyungpook National University Chilgok Hospital added: “In pandemic situations such as COVID-19, when medical resources and personnel are limited, the emergency medical system can be burdened considerably. To this end, AI that offers fast and reliable examinations can facilitate decisions regarding patient screening and isolation, thereby reducing the workload on medical staff.”

Lunit Insight CXR clinically analysed more than 6.5 million images in more than 80 countries, and has an accuracy of 97 to 99 per cent in detecting 10 major chest diseases such as lung nodules and pneumothorax

Published on page 4 of the February 2021 issue of RAD Magazine.