Model could help detect the virus, isolate patients, and prevent disease spread.
Integrating an artificial intelligence tool with a patient’s clinical health information, as well as lab data, can bolster how well chest CT can identify COVID-19 pneumonia, new research has revealed.
Even though recent investigations found chest CT is sensitive and specific enough to diagnose COVID-19, patients who have early-stage disease can still have negative scans. This is where an AI tool can be useful, said a team led by Yang Yang, Ph.D., assistant professor of radiology at the Translational and Molecular Imaging Institute at Icahn School of Medicine at Mount Sinai.
They published their findings in the May 19 issue of Nature Medicine.
Being able to rapidly diagnose patients who have COVID-19 is critical to providing timely treatment and, potentially, slowing the viral spread. And, while the current gold standard – reverse transcription-polymerase chain reaction (RT-PCR) – has been criticized for both taking up to two days to provide a diagnosis and frequently requiring repeated testing, AI-powered CT may be able to perform better, the team said.
“Chest CT findings are normal in some patients early in the disease course, and, therefore, chest CT alone has limited negative predictive value to fully exclude infection, highlighting the need to incorporate clinical information in the diagnosis,” Yang’s team wrote. “We propose that AI algorithms may meet this need by integrating chest CT findings with clinical symptoms, exposure history and laboratory testing in the algorithm.”
Related Content: ACR Releases CT and Chest X-ray Guidance Amid COVID-19 Pandemic
Augmented detection is critical, the team said, because it can play a role in isolating patients who present with COVID-19 symptoms but show no lung disease on scans.
To determine whether AI could augment CT performance in diagnosing COVID-19, Yang’s team evaluated CT scans and clinical information, including age, gender, symptoms, bloodwork, and possible contact with infected individuals, from 905 patients who received treatment in 18 medical centers in China between Jan. 17 and March 3, 2020. Among that group, 419 (46.3 percent) were diagnosed as being positive for the virus with RT-PCR.
Using 279 of the 905 CT scans, the researchers trained their AI model, and, then, compared the results to those provided by a senior thoracic radiologist and a fellow. Based on the outcomes, the team determined their AI tool was comparably sensitive to the senior radiologist’s performance. The model was 84-percent sensitive in evaluating images and clinical data compared to 75-percent sensitivity from the provider.
The team also discovered the AI model made it easier to identify patients who had a positive RT-PCR result, but a normal chest CT. While radiologists classified all 25 such patients as being negative for the virus, the AI tool correctly identified 17 (68 percent) as being infected.
Ultimately, the team said, their study findings highlight a combining CT scans, a patient’s clinical history, and an AI tool can augment how well the CT images can diagnose COVID-19 infection. As such, they said, the AI system could be used as a rapid diagnostic tool to flag patients as potential high-priority cases.
Photon-Counting Computed Tomography: Eleven Takeaways from a New Literature Review
May 27th 2025In a review of 155 studies, researchers examined the capabilities of photon-counting computed tomography (PCCT) for enhanced accuracy, tissue characterization, artifact reduction and reduced radiation dosing across thoracic, abdominal, and cardiothoracic imaging applications.
Emerging AI Mammography Model May Enhance Clarity for Initial BI-RADS 3 and 4 Classifications
May 21st 2025In a study involving over 12,000 Asian women, researchers found that an artificial intelligence (AI) model converted over 83 percent of false positives in patients with initial BI-RADS 3 and 4 assessments into benign BI-RADS categories.
CT Perfusion Study Shows Enhanced Detection of Medium Vessel Occlusions with Emerging AI Software
May 21st 2025The Rapid CTP AI software offered 23 percent greater detection of medium vessel occlusions in comparison to the Viz CTP AI software, according to research presented at the European Stroke (Organization) Conference (ESOC).