Premkumar Elangovan PhD, MSc, Presenter: Nothing to Disclose

Alaleh Rashidnasab, Abstract Co-Author: Nothing to Disclose

Alistair Mackenzie, Abstract Co-Author: Nothing to Disclose

David Dance PhD, Abstract Co-Author: Nothing to Disclose

Kenneth C. Young PhD, Abstract Co-Author: Nothing to Disclose

Hilde Bosmans PhD, Abstract Co-Author: Co-founder, Qaelum NV Research Grant, Siemens AG

William Paul Segars PhD, Abstract Co-Author: Nothing to Disclose

Kevin Wells PhD, Abstract Co-Author: Nothing to Disclose

To compare the visibility of spheres and simulated masses in 2D-mammography and digital breast tomosynthesis systems (DBT) using human observer studies.

A selection of uniform spheres and simulated masses with varying size and contrast were embedded in volumetric ROIs taken from anthropomorphic Duke virtual breast phantoms. A set of validated simulation tools was then used to synthesise images representing challenging clinical visibility situations. A series of 4-alternative forced choice (4AFC) experiments (scored by 4 physicists) were conducted and contrast detection thresholds for 2D mammography and for DBT, and both target types (mass/sphere) were calculated. Data preparation involved: (i) extraction of volumetric ROIs from the breast phantoms; (ii) simulation of 3mm/7mm mass and sphere targets; (iii) insertion of the targets at random ROI depth by voxel replacement; (iv) adjustment of the object composition until a designated contrast was achieved; and (v) generation of realistic simulated images by modelling all image formation and degradation processes of a Hologic Selenia Dimensions 3D system. This led to 45 mass and 15 sphere images per experimental condition (i.e. contrast and size). The observers were presented with a series of signal detection tasks comprising 4 images per task – one with the target signal, and the rest with similar backgrounds with no signal present. Each observer was required to select the image containing the target signal. The minimum detectable contrast (at which the observers were correct 92.5% of the time) was calculated for all experimental conditions.    

Observers needed approximately three times the signal contrast to correctly identify a mass in 2D–mammography (3mm: 0.030±.0035, 7mm: 0.034±.0001) compared with DBT (3mm: 0.010±.0004, 7mm: 0.010±.0002). Similar differences were found when spheres were used instead of simulated masses. However, the minimum detectable contrast for spheres was much lower in both 2D-mammography (3mm: 0.018±.0037, 7mm: 0.019±.0008) and DBT (3mm: 0.007±.0035, 7mm: 0.005±.0035) compared to masses.

The proposed method provides a quantitative means of comparing system designs and may help in optimisation.  

Tomosynthesis appears to have a lower detection limit than 2D-mammography for masses/spheres, and moreover, detection studies using spheres may produce overly-optimistic threshold contrast values. 

Elangovan, P, Rashidnasab, A, Mackenzie, A, Dance, D, Young, K, Bosmans, H, Segars, W, Wells, K, A 4AFC Observer Study to Compare the Visibility of Masses in 2D-mammography and Digital Breast Tomosynthesis Systems.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14005393.html