Clinical evaluation of a pulmonary nodule detection system

Abstract

Purpose To evaluate the effect of the prototype version of a computer-aided detection system, the so-called Lung-CAD, on the sensitivity of a cohort of community radiologists concerning the detection of any pulmonary nodules on multi-detector raw computer tomography (MDCT) datasets. Materials and methods 16 MDCT studies with section thickness ranging from 1.00 to 1.25 mm were evaluated by 6 community level radiologists of two different sites. The readers were instructed to mark every nodule of size > 3 mm. Consecutively, the datasets were processed with the Lung-CAD-prototype version. Then the community radiologists reviewed and rejected/accepted the computer detected nodules. The Ground truth was defined as any nodule found by the reader and/or the CAD-prototype and also validated by one single expert's blinded review. Conclusion The CAD-assisted work significantly increased the readers' sensitivity without any significant change on the rate of false positive findings per case. Also of equal importance, the interreader variability of sensitivity was reduced, leading to a more predictable detection rate across radiologists.