Abstract Archives of the RSNA, 2004
SSQ05-01
Computer-aided Diagnosis (CAD) in Detection of Lung Nodules: Assessment Based on Density and Anatomical Location
Scientific Papers
Presented on December 2, 2004
Presented as part of SSQ05: Chest (Lung Nodules: Characterization)
Subba Rao Digumarthy MD, Presenter: Nothing to Disclose
Amita Sharma MBBS, Abstract Co-Author: Nothing to Disclose
Suzanne L. Aquino MD, Abstract Co-Author: Nothing to Disclose
Conrad Wittram MBChB, Abstract Co-Author: Nothing to Disclose
Jo-Anne O'Malley Shepard MD, Abstract Co-Author: Nothing to Disclose
Computer aided diagnosis (CAD) of pulmonary nodules is being evaluated as a potential double read. The purpose of our study was to determine sensitivity of CAD for nodule detection based on morphology (ground glass vs solid) and anatomical location.
Multidetector CT examinations of the chest in 31 randomly selected patients with solid, ground glass and mixed solid and ground glass nodules were included in the study. Patients with diffuse lung study were not included in the study cohort. Images were reviewed prospectively by two experienced thoracic radiologists to identify all nodules(>4mm).The nodules were classified as ground glass, solid, or mixed solid and ground glass. Based on proximity to blood vessels, airways, pleura, mediastinum and diaphragm, nodules were also sub-classified as adjacent or non-adjacent. A CAD program (Image checker CT, R2 Technology, v1.0) processed all cases independently. The findings of an expert panel comprised of two experienced radiologists, who reviewed the findings of radiologists and also CAD, constituted the standard of reference. The statistical analysis of the data was done by student t test and wilcoxon signed rank test.
187 nodules (solid=138, ground glass=46 and mixed=3) were identified by the expert panel in 31 patients. The sensitivity for the detection of solid nodules was 40 % (95% CI 32-48%) for CAD and 91% (95% CI 84- 95%) for radiologists. None of the ground glass nodules were detected by CAD. The overall sensitivity for the detection of pulmonary nodules for radiologists was 93% (95% CI 88-96%) and 30% (95% CI 24-37%) for CAD. There was no difference in the sensitivity of CAD in the detection of adjacent and non-adjacent nodules (p=0.17). CAD detected 9% (13/138 ) of solid nodules in 19% (6/31) of patients missed by radiologists, all these nodules were adjacent to a vessel or pleura.
CAD detected solid nodules in 19% of patients missed by radiologists, however it did not detect any ground glass nodules. There is no difference between detection of adjacent and non-adjacent solid nodules for CAD.
J.O.S.: Received research grant from R2 technology
Digumarthy, S,
Sharma, A,
Aquino, S,
Wittram, C,
Shepard, J,
Computer-aided Diagnosis (CAD) in Detection of Lung Nodules: Assessment Based on Density and Anatomical Location. Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL.
http://archive.rsna.org/2004/4417168.html