Computational assessment of stereoscopic viewing a sequence of stereo pairs of breast tomosynthesis projection images

Digital breast tomosynthesis (DBT) is a 3D imaging technology in which an x-ray fan beam rotates around the breast, producing a series of projection images. The imaging geometry of DBT lends itself naturally to stereo viewing because a stereo pair can be easily formed by two projection images with a reasonable separation angle. Stereo viewing may reveal the 3D structures of breasts thus has the potential to increase the sensitivity and specificity of breast imaging. In this study, we conduct a simulation study that mimics the detection of breast lesions on stereoscopic viewing of DBT projections. The presentation approach we investigate here is one in which the reader is presented with a sequence of stereo pairs from a rotating point of view. We render voxel datasets that contain random 3D power-law noise to model normal breast tissues with different breast densities. A 3D Gaussian signal is inserted to some of the datasets to model the presence of a breast lesion. Sequences of stereo pairs of projection images are generated for each voxel dataset by varying the projection angles of the two views. The diagnostic performance, in terms of the accuracy of binary decisions on the presence of the simulated lesions, is evaluated with a numerical model observer.