Study: AI More Than Doubles the Sensitivity Rate for Lung-RADS Category 4 Nodules on Chest X-Rays


In newly published research, researchers found that an artificial intelligence (AI) computer-aided detection (CAD) system was more than twice as likely as non-AI assessment to diagnose actionable lung nodules on chest X-rays.

Emerging research suggests that artificial intelligence (AI)-powered software may significantly enhance the detection of Lung-RADS category 4 nodules on chest radiographs.

In a randomized controlled trial (RCT), recently published in Radiology, researchers compared the use of AI-based computer-aided detection (CAD) software (Lunit INSIGHT CXR version 2.0, Lunit) with chest radiographs in 5,238 participants versus a non-AI control group of 5,238 participants. The overall cohort (median age of 59) included current smokers (11 percent) and former smokers (25 percent), according to the study authors. For the primary outcome, the researchers sought to determine detection rates for actionable lung nodules (or Lung-RADS category 4 nodules).

The study authors found that the AI-CAD software group had a statistically significant 2.4 odds ratio for detecting Lung-RADS category 4 nodules in comparison to the non-AI control group.

The researchers also reviewed data from a subgroup of patients who had follow-up chest computed tomography (CT) exams within three months of the chest radiograph. This included 2,425 patients in the AI-CAD group and 2,461 patients in in the non-AI control group). Use of the AI-CAD software demonstrated a 56.4 percent sensitivity rate and a 35.6 percent positive predictive value in comparison to 23.2 percent and 18.8 percent, respectively, for the non-AI group, according to the study authors.

“The improved detection rate of actionable lung nodules with a similar false-referral rate suggests that using AI-based CAD may improve lung cancer diagnosis without imposing an additional radiation hazard,” wrote study co-author Jin Mo Goo, M.D., Ph.D., a professor and section director of chest radiology in the Department of Radiology at the Seoul National University Hospital in Korea, and colleagues.

“Meanwhile, the diagnostic performance could be more robustly evaluated owing to the high proportion of CT performance. The AI group exhibited higher sensitivity, positive predictive value, and negative predictive value while specificity was similar between the two groups.”

(Editor’s note: For related content, see “Can Deep Learning Enhance Pulmonary Nodule Detection on Chest X-Rays?” and “Can Deep Learning Assessment of X-Rays Improve Triage of Patients with Acute Chest Pain?”)

In an accompanying editorial, William F. Auffermann, M.D., Ph.D., called the study an “important step forward in demonstrating the utility of AI-based CAD.”

In addition to providing “much-needed data” from an RCT, the study provided a welcome contrast to previous CAD studies that utilized enriched numbers of abnormal cases, according to Dr. Auffermann, the section chief of cardiothoracic imaging and associate professor of radiology and imaging sciences at the University of Utah School of Medicine. Abnormal cases comprised nearly 50 percent of reviewed cases in previous CAD research, but Dr. Auffermann said the RCT by Goo and colleagues found that 1.1 percent of chest radiographs had actionable nodules that were identified on subsequent CT scans.

“An increased number of abnormal cases in a nodule detection study may alter reading behavior and the threshold a radiologist uses to mark a case as concerning for nodules,” explained Dr. Auffermann. “It is difficult to determine how the interpretive behavior of radiologists who know they are reading for a study with an enriched data set will translate to performance in routine clinical practice. This study offers new data to answer this question.”

In regard to study limitations, the researchers noted the data came from one institution. They pointed out that the assessment of chest radiography was based on a subgroup of the study cohort as chest CT scans were not available for all patients. Goo and colleagues added that the study did not evaluate the standalone performance of AI-based CAD nor its impact upon prioritization and time reduction of reporting.

Related Videos
Practical Insights on CT and MRI Neuroimaging and Reporting for Stroke Patients
Expediting the Management of Incidental Pulmonary Emboli on CT
Related Content
© 2023 MJH Life Sciences

All rights reserved.