Lung cancer screening is designed to be low-dose and high-throughput. CAD tools promise to assist radiologists in analyzing the influx of screening exams. However, the effects of dose on CAD performance are not fully understood. In this work, we investigated the impact of reducing the dose further than the National Lung Screening Trial (NLST) dose protocols.
METHOD AND MATERIALSThe raw CT data files from 481 NLST patients were collected and input to a reduced-dose simulation software. The original NLST protocols called for 25 mAs for standard-size patients and 40 mAs for larger patients. We simulated reduced-dose scans corresponding to 50% and 25% of the original protocols. All cases were reconstructed at the scanner (Sensation 64, Siemens Healthcare) with 1 mm slice thickness and B50 kernel. The lungs were segmented in MeVisLab software, and then all images and segmentations were input to an in-house CAD algorithm. CAD results were compared to a reference standard generated by an experienced reader as part of the NLST. We computed subject-level sensitivities, false-positive rates, and analyzed the relative change in those metrics with dose. LungRADS categories were also assigned to each nodule based on nodule size and solidity, and a sub-analysis was peformed by LungRADS category.
RESULTSFor larger category 4 nodules, median sensitivities were 100% at all three dose levels, and mean sensitivities were 72%, 63%, and 63% at original, 50%, and 25% dose respectively. Overall mean subject-level sensitivities were 38%, 37%, and 38% at original, 50%, and 25% dose due to the prevalence of smaller category 2 nodules. The mean false-positive rates were 3, 5, and 13 per case.
CONCLUSIONThe results suggest some loss of CAD sensitivity with dose for larger nodules, although overall sensitivity appeared unaffected by dose. The false-positive rate increased substantially at 25% dose, illustrating the difficulty of adapting CAD to very challenging, high-noise screening exams.
CLINICAL RELEVANCE/APPLICATIONCare should be taken to adapt CAD algorithms for very challenging, high-noise lung screening exams.