Use of computer-aided diagnosis (CAD) improves estimation of malignancy risk for indeterminate pulmonary nodules (IPNs) on chest computed tomography (CT) scans, according to a study published in Radiology.
Roger Y. Kim, M.D., from the Perelman School of Medicine at the University of Pennsylvania in Philadelphia, and colleagues examined the effect of an artificial intelligence-based CAD tool on clinician IPN diagnostic performance in a retrospective multireader, multicase study performed in June and July 2020.
For each case, readers used CT imaging data to estimate malignancy risk and provide a management recommendation with and without CAD. Estimates of area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were calculated to assess the effect of CAD on average reader diagnostic performance.
Twelve readers reviewed 300 chest CT scans of IPNs with maximal diameters of 5 to 30 mm. The researchers found that with CAD, the readers’ average AUC improved from 0.82 to 0.89. Use of CAD improved average sensitivity from 94.1 to 97.9 percent and from 52.6 to 63.1 percent at malignancy risk thresholds of 5 and 65 percent, respectively.
Improvement was also seen in average reader specificity, from 37.4 to 42.3 percent and from 87.3 to 89.9 percent, respectively. For both the less than 5 percent and more than 65 percent malignancy risk categories, reader interobserver agreement improved with CAD. For management recommendation categories, overall reader interobserver agreement also improved with CAD.
“Our findings provide crucial support for bringing CAD tools closer to clinical implementation for IPN risk stratification,” the authors write.
Several authors disclosed financial ties to biotechnology and medical device companies, including Optellum, which funded the study.