Heang-Ping Chan PhD, Presenter: Nothing to Disclose

Jun Wei PhD, Abstract Co-Author: Nothing to Disclose

Mark Alan Helvie MD, Abstract Co-Author: Institutional grant, General Electric Company Consultant, General Electric Company

Lubomir M. Hadjiiski PhD, Abstract Co-Author: Nothing to Disclose

Chuan Zhou PhD, Abstract Co-Author: Nothing to Disclose

Yao Lu PhD, Abstract Co-Author: Nothing to Disclose

To study the effect of tomographic angle, number of projection views (PVs), and dose on mass detection by a computer-aided detection (CAD) system for DBT.

With IRB approval and informed consent, DBT of 50 breasts containing biopsy-proven masses (14 malignant, 36 benign; median size:12.9 mm; range: 4.5-26.6 mm) were acquired in CC and MLO views with a GE prototype DBT system (21 PVs, 60º arc, 3º increments). The total dose of the 21 PVs was about 1.5X that of a single-view conventional mammogram. Three sets of DBT volumes were reconstructed with SART in 1-mm slice interval: (a) 21PV (21 PVs, 60º, 3º, 100% dose), (b) 11UW (11 PVs, 60º, 6º, 52% dose), and (c) 11UN (11 PVs, 30º, 3º, 52% dose). Mass detection in the three sets of reconstructed DBT volumes by an automated system developed in our laboratory was compared. For each input DBT volume, mass candidates were identified by 3D gradient field analysis. For each candidate, two-stage object segmentation was performed in which the first-stage 3D clustering estimated the volume of interest (VOI) adaptively to the object size and the second-stage active contour segmentation refined the object shape, constrained by the VOI. Features that characterized the shape, gray level, texture, and gradient field properties were then extracted from the segmented object. For each DBT set, a linear discriminant analysis classifier with stepwise feature selection was trained and tested with a leave-one-case-out method to differentiate masses from false positives (FPs).

The figure-of-merit for the view-based test FROC curve by JAFROC analysis was 0.71 (21PV), 0.68 (11UW), and 0.53 (11UN). The difference between the test FROC curves of 21PV and 11UW did not achieve statistical significance (p=0.4) but both were significantly higher than that of 11UN (p<0.01). The view-based sensitivity was 74% (21PV), 70% (11UW), and 53% (11UN) at 0.5 FPs/volume, and 80%, 78%, and 66% at 1 FPs/volume. The case-based sensitivity was 90%, 89%, and 67% at 0.5 FPs/volume, and 94%, 93%, and 80% at 1 FPs/volume.

For mass detection, with a fixed 60º arc, 11 PVs at 52% dose was comparable to 21 PVs. With a fixed dose and 11 PVs, 60º arc was superior to 30º arc.

DBT and CAD have the potential to improve breast cancer detection. Understanding factors that affect lesion detectability is important for the design of dose-efficient DBT and CAD systems.

Chan, H, Wei, J, Helvie, M, Hadjiiski, L, Zhou, C, Lu, Y, Digital Breast Tomosynthesis (DBT): Impact of Tomographic Scan Angle, Number of Projection Views, and Dose on Automated Mass Detection.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11034415.html