Matthew T. Freedman MD, MBA, Presenter: Nothing to Disclose

Juh-Shyan Lin PhD, Abstract Co-Author: Nothing to Disclose

Fleming Y Lure, Abstract Co-Author: Nothing to Disclose

Teresa Osicka ME, MS, Abstract Co-Author: Nothing to Disclose

Shih-Chung B. Lo PhD, Abstract Co-Author: Nothing to Disclose

To determine the detection sensitivity of a computer program for lung nodule detection on energy subtraction chest radiographs.

211 pairs of GE Medical Systems digital chest radiographs processed with conventional processing (CPDR) and energy subtraction (ES) processing were reviewed for the presence of solitary lung nodules. 37 cases showed a non-calcified pulmonary nodule 9-30 mm in size, average size 18 mm. The Riverain Medical Group (formerly Deus Technologies) (Dayton, Ohio) RS-2000D algorithm was run on both images and the detection results were compared.

On the CPDR images, the RS-2000D correctly identified the location of 21 of 37 solitary non-calcified nodules resulting in sensitivity of 57%. On the ES processed images, the RS-2000D correctly identified 24 nodules resulting in sensitivity of 65%. When the results on the two groups of images were compared, 6 failed detections on the ES images were correctly identified on the CPDR images and 9 failed detections on the CPDR images were detected on the ES images. Combining the results from the CPDR and ES images, the detection rate was 30 of 37 nodules providing a combined sensitivity of 81%. In the cases with a solitary pulmonary nodule, there were, on average, 2.1 false positive marks per image (FPM/I) on the CPDR and 1.6 on the ES images. On 62 cases without lung nodules or extensive interstitial disease, there were an average of 2.8 FPM/I on the conventional processing and 2.1 on the ES images.

The RS-2000D computer-aided detection system can detect solitary lung nodules on conventional and energy subtraction chest radiographs. Sensitivity was 57% on the CPDR and 65% on the ES. The combined sensitivity was 81%. This suggests that the RS-2000D will provide improved results when applied to both the conventional and ES images. In this dataset, the number of false positive marks per image was less than 3 compared to previously reported results on different datasets that ranged from 5.0 to 5.3 FPM/I.

Matthew Freedman is a consultant to an entity related to and J. Lin and F. Lure are employees of Riverain Medical Group. 

Freedman, M, Lin, J, Lure, F, Osicka, T, Lo, S, Hot Topic: Comparison of the Computer-Aided Detection of Lung Nodules on DR with Conventional and Energy Subtraction Processing.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4418671.html