Kenji Suzuki PhD, Presenter: Nothing to Disclose

Junji Shiraishi PhD, Abstract Co-Author: Nothing to Disclose

Feng Li MD, PhD, Abstract Co-Author: Nothing to Disclose

Hiroyuki Abe MD, Abstract Co-Author: Nothing to Disclose

Heber MacMahon MD, Abstract Co-Author: Nothing to Disclose

Kunio Doi PhD, Abstract Co-Author: Nothing to Disclose

A major difficulty in current computer-aided diagnostic (CAD) schemes for nodule detection on chest radiographs is to detect nodules that overlap with ribs. Our purpose was to develop a new technique for false-positive reduction in a CAD scheme using a rib-suppression technique based on massive training artificial neural networks (MTANNs).

We developed a multiple MTANN (multi-MTANN) consisting of eight MTANNs. For distinction between nodules and non-nodules (false positives), the teacher images were designed to contain Gaussian distribution for nodules and zero for non-nodules. The multi-MTANN was trained with typical nodules and eight different types of non-nodules so that various non-nodules can be eliminated. For further removal of false positives caused by ribs, we developed a rib-suppression technique using a multi-resolution MTANN consisting of a multi-resolution decomposition technique and three MTANNs. To suppress the contrast of ribs in chest images, the multi-resolution MTANN was trained with input chest images and the teacher soft-tissue images obtained by use of dual-energy imaging. Our database consisted of 91 nodules in 91 chest radiographs. All nodules were confirmed by CT examinations, and the average size of the nodules was 24.9 mm.

With our original CAD scheme based on a difference image technique with linear discriminant analysis, a sensitivity of 82.4% (75/91 nodules) with 410 false positives (4.5 per image) was achieved. The trained multi-MTANN was able to remove 62.7% (257/410) of false positives with a loss of one true positive. By use of the rib-suppression technique, the contrast of ribs in chest radiographs was suppressed substantially. Due to the effect of rib-suppression, 24.5% (63/257) of the remaining false positives were removed without a loss of any true positives. Thus, the false-positive rate of our CAD scheme was improved from 4.5 to 1.0 (90/91) false positives per image at an overall sensitivity of 81.3% (74/91).

By use of a multi-MTANN incorporating a rib-suppression technique, the false-positive rate of our CAD scheme can be improved substantially, while a high sensitivity is maintained.

K.D.,H.M.: K.D., H.M. are shareholders in R2 Technology, Inc., Sunnyvale, CA, and K.D. is a shareholder in Deus Technologies, Inc., Rockville, MD.

Suzuki, K, Shiraishi, J, Li, F, Abe, H, MacMahon, H, Doi, K, False-Positive Reduction in Computerized Detection of Lung Nodules in Chest Radiographs Using Massive Training Artificial Neural Networks for Rib-Suppression Technique.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4415165.html