Seitaro Oda MD, Presenter: Nothing to Disclose

Kazuo Awai MD, Abstract Co-Author: Nothing to Disclose

Kenji Suzuki PhD, Abstract Co-Author: Consultant, Riverain Medical

Lifeng He PhD, Abstract Co-Author: Nothing to Disclose

Heber M. MacMahon MD, Abstract Co-Author: Consultant, Riverain Medical Research support, MEDIAN Technologies Stockholder, Hologic, Inc (R2 Technology, Inc)

Yasuyuki Yamashita MD, Abstract Co-Author: Nothing to Disclose

A massive-training artificial neural network (MTANN) is a nonlinear pattern-recognition technique that can suppress rib opacity in chest radiographs while maintaining soft-tissue contrast. The objective of our study was to investigate the effect of rib-suppressed images by the MTANN on the performance of radiologists in the detection of pulmonary nodules on chest radiographs.

We used 60 chest radiographs containing 30 images with solitary pulmonary nodules and 30 images without nodules, which were selected with a stratified random-sampling scheme from the Japanese standard digital image database developed by the Japanese Society of Radiological Technology. Our institutional review board approved the use of the database, and the requirement for informed consent was waived. The mean size of the 30 pulmonary nodules was 14.7 ± 4.1 mm (standard deviation). Receiver operating characteristic (ROC) analysis with a continuous rating scale was used for evaluation of observer performance in detecting pulmonary nodules on chest radiographs first without and then with the rib-suppressed images. Seven board-certified radiologists and 5 radiology residents participated in this observer study.

For all 12 observers, the mean values of the area under the best-fit ROC curve (Az) achieved without and with the rib-suppressed images were 0.81 ± 0.07 and 0.84 ±0.07, respectively, and the difference was statistically significant (P＜0.01). The mean Az values achieved without and with the rib-suppressed images were 0.84 ±0.06 and 0.88 ± 0.05, respectively, for the 7 board-certified radiologists (P＜0.01) and 0.77 ± 0.08 and 0.79± 0.06, respectively, for the 5 radiology residents (P=0.24).

Use of rib-suppressed images together with original chest radiographs significantly (P＜0.01) improved the diagnostic performance of radiologists in the detection of pulmonary nodules.

Chest radiograph with rib suppression by means of massive-training artificial neural network is a promising technique to improve detectability of pulmonary nodules.

Oda, S, Awai, K, Suzuki, K, He, L, MacMahon, H, Yamashita, Y, Detection of Pulmonary Nodules on Chest Radiographs: Effect of Rib Suppression by Means of Massive Training Artificial Neural Network (MTANN) on Performance of Radiologists.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5002064.html