Sumiaki Matsumoto MD, PhD, Presenter: Nothing to Disclose

Yoshiharu Ohno MD, PhD, Abstract Co-Author: Nothing to Disclose

Hitoshi Yamagata PhD, Abstract Co-Author: Employee, Toshiba Corporation

Daisuke Takenaka MD, Abstract Co-Author: Nothing to Disclose

Kazuro Sugimura MD, Abstract Co-Author: Nothing to Disclose

We are developing a computer-aided detection (CAD) scheme for lung nodules at MDCT. The purpose of this study was to evaluate the scheme, with or without selective noise reduction of lung regions, in terms of the robustness to noise using simulated low-dose images.

The CAD scheme has an optional capability of selectively reducing noise in lung regions by means of a partial differential equation based technique. Using a collection of thin-slice (0.5 mm) standard-dose (120 kV, 150 mAs) images of MDCT (Aquilion, Toshiba Medical Systems) for 25 cases of suspected lung cancer (19 men, 6 women; age range, 25-84), three versions of processed images that simulated acquisition at 60 mAs, 30 mAs, or 20 mAs were created for each case through addition of three different levels of noise to the raw data of the corresponding standard-dose images. Based on the reference standard of 73 non-calcified nodules of 4 mm or larger in diameter (mean: 11 mm) as determined by the consensus of two chest radiologists, the performance of the CAD scheme with or without the option of selective noise reduction at a fixed parameter setting on these simulated low-dose images was statistically compared to that on standard-dose images.

The sensitivity of the scheme with or without the option of selective noise reduction was 70% (51/73) or 52% (38/73), 73% (53/73) or 42% (31/73), 64% (47/73) or 34% (25/73), and 60% (44/73) or 26% (19/73) for standard-dose, 60-mAs, 30-mAs, and 20-mAs images, respectively. The corresponding number of false positives per case were 3.9 (98/25) or 1.8 (45/25), 3.7 (84/25) or 1.4 (34/25), 2.9 (73/25) or 1.5 (37/25), and 2.6 (66/25) or 1.2 (29/25), respectively. Without the option, the sensitivity for 30-mAs or 20-mAs images was significantly decreased as compared to that for standard-dose images (McNemar’s test; P<0.01); with the option, there was no such significant decrease in sensitivity.

The CAD scheme with the option of selective noise reduction of lung regions showed a robust performance against noise.

This CAD scheme with a capability of selective noise reduction of lung regions may be useful for low-dose MDCT images in CT lung cancer screening.

Matsumoto, S, Ohno, Y, Yamagata, H, Takenaka, D, Sugimura, K, Evaluation of a CAD Scheme with Selective Noise Reduction of Lung Regions for Nodule Detection at MDCT.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5007396.html