Evaluation of a corrected implementation of a method of simulating pulmonary nodules in chest tomosynthesis

Abstract

Background A method of simulating pulmonary nodules in tomosynthesis images has previously been developed and evaluated. An unknown feature of a rounding function included in the computer code was later found to introduce an artifact, affecting simulated nodules in low-signal regions of the images. The computer code has now been corrected. Purpose To perform a thorough evaluation of the corrected nodule-simulation method, comparing the detection rate and visual appearance of artificial nodules with those of real nodules in an observer performance experiment. Material and Methods A cohort of 64 patients with a total of 129 pulmonary nodules was used in the study. Artificial nodules, each matching a corresponding real nodule by size, attenuation, and anatomical location, were generated and simulated into the tomosynthesis images of the different patients. The detection rate and visual appearance of artificial nodules generated using both the corrected and uncorrected computer code were compared to those of real nodules. The results were evaluated using modified receiver operating characteristic (ROC) analyses. Results The difference in detection rate between artificial and real nodules slightly increased using the corrected computer code (uncorrected code: area under the curve [AUC], 0.47; 95% CI, 0.43-0.51; corrected code: AUC, 0.42; 95% CI, 0.38-0.46). The visual appearance was however substantially improved using the corrected computer code (uncorrected code: AUC, 0.70; 95% CI, 0.63-0.76; corrected code: AUC, 0.49; 95% CI, 0.29-0.65). Conclusion The computer code including a correct rounding function generates simulated nodules that are more visually realistic than simulated nodules generated using the uncorrected computer code, but have a slightly different detection rate compared to real nodules.