CAD improves lung nodule detection

Computer-aided detection (CAD) has demonstrated an ability to improve the performance of radiologists in detecting lung nodules on chest radiographs, according to a study published online March 12 by Radiology.

Despite its inferiority to CT, chest radiography remains the primary method for diagnosing lung nodules in clinical practice. Because visually discernable nodules can be overlooked on radiographs, CAD programs have been developed to improve sensitivity and specificity. Lead author Steven Schalekamp, MD, of the Radboud University Medical Center in Nijmegen, the Netherlands, and colleagues evaluated the added value of CAD for lung nodules in chest radiographs when bone-suppressed images (BSIs) are available to radiologists.

The researchers selected 300 posteroanterior and lateral chest radiographs in 300 subjects for their study. One hundred eighty-nine radiographs had negative findings and 111 had a solitary pulmonary node. Five radiologists and three residents first evaluated the radiographs with BSIs without CAD and then after inspection of the CAD marks. The readers marked suspicious locations and gave a confidence score of that location being a nodule.

The average nodule size found during the study was 16.2 millimeters. Stand-alone CAD had a sensitivity of 74 percent at the 1.0 false-positive mark per image. Without CAD, the average area under the curve was 0.812 for the observers. When CAD was utilized, performance improved slightly to an area under the curve of 0.841.

CAD detected 127 of 239 nodules that were missed after the radiographs and BSIs were evaluated by all observers. Of these detections, 57 were eventually marked by the observers after the use of CAD. The CAD’s sensitivity was 91 percent for well-visible nodules and 8 percent for those that were more subtle. The sensitivities for subtle and very subtle were 62 percent and 39 percent, respectively.

Twenty-one of the 23 nodules, or 91 percent, that were bigger than 20 millimeters in diameter were detected by CAD. Detection rates were lower for smaller nodules, ranging from 62 to 77 percent.

“We demonstrated that CAD has an additional beneficial effect on the detection of pulmonary nodules beyond the effect of BSIs alone,” wrote Schalekamp and colleagues. “Even though baseline performance was optimized by the availability of BSIs, radiologists were able to uniformly improve their detection performance.”