Kenji Suzuki PhD, Presenter: Nothing to Disclose

Qiang Li PhD, Abstract Co-Author: Nothing to Disclose

Feng Li MD, PhD, 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 problem in computer-aided diagnostic (CAD) schemes for lung nodule detection in multi-detector computed tomography (MDCT) is a large number of false positives in order to maintain a high sensitivity. Our purpose was to develop a new technique for reduction of false positives in a CAD scheme using massive training artificial neural networks (MTANNs).

To remove the false positives reported by our CAD scheme based on selective enhancement filters with linear discriminant analysis, we developed a multiple MTANN (multi-MTANN) consisting of eight MTANNs for reduction of eight different types of false positives. The MTANN was trained with input CT images and the corresponding teacher images. For distinction between nodules and non-nodules (false positives), the teacher images for nodules and non-nodules contained Gaussian distribution and zero, respectively. Each MTANN in the multi-MTANN was trained with ten typical nodules and ten non-nodules in each of eight different types. For example, MTANN 1 was trained to distinguish nodules from medium-sized vessels, MTANNs 2, 3, and 4 were designed to eliminate small vessels, vessels with high contrast, and some soft-tissue opacities. Eight MTANNs were combined with the logical AND operation such that eight different types of non-nodules can be eliminated. Our database contained 62 nodules in 32 MDCT scans acquired from 32 patients with an MDCT system. The MDCT scan consisted of an average of 186 thin-slice CT images (slice thickness was 1.25 or 2.5 mm). All nodules were confirmed by chest radiologists.

With our original CAD scheme, a sensitivity of 96.8% (60/62 nodules) together with 14.9 false positives per case was achieved. The multi-MTANN was applied to further reduction of false positives. Results indicated that 55.7% (265/476) of false positives were removed with a removal of one true positive. Thus, the false-positive rate of our CAD scheme was improved to 6.6 false positives per case at an overall sensitivity of 95.2% (59/62 nodules).

By use of a multi-MTANN, the specificity of our CAD scheme for lung nodule detection on MDCT images 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, Li, Q, Li, F, MacMahon, H, Doi, K, Distinction between Nodules and False Positives in CAD Scheme for Lung Nodule Detection on Multidetector CT Images by Means of Massive Training Artificial Neural Networks.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4412096.html