Next-Generation Tomosynthesis Pilot Study

PROJECT SUMMARY
Even with the latest digital breast tomosynthesis (DBT) systems, breast cancer screening continues to suffer
from poor specificity. Only about 5% of women called-back from screening are ultimately found to have a biopsy-
proven cancer. Clinical DBT systems suffer from anisotropies in image quality since the scanning motion is
restricted to one direction (left-to-right). We built a next-generation tomosynthesis (NGT) system that is capable
of scanning in the shape of a “T”. With phantoms and mastectomy specimens, we have shown that this design
mitigates cone-beam artifacts, tissue superposition effects, and anisotropies in super-resolution. As the next step
in our research, we will perform a pilot study with volunteers, recruiting women referred for diagnostic imaging
or biopsy as well as women having abbreviated magnetic resonance imaging (MRI). Projection images will be
acquired in such a way that we can generate reconstructions from two scanning methods (conventional and T).
Each scanning method will be analyzed separately by radiologists in different reading sessions. We will mitigate
potential concerns about radiation exposure by restricting the study to one view (cranial-caudal) instead of two
views. We have put together a team with a unique set of strengths, including the developers of the NGT system,
three radiologists, two statisticians, and experts in density and texture analysis. This proposal is divided into two
specific aims. (Aim 1): Assess radiologists' performance in a pilot study of the NGT system with volunteers. We
will investigate whether the T scan brings down the call-back rate of screening without reducing sensitivity.
Radiologists will also rate the overall probability of malignancy, and these scores will be analyzed in combination
with clinical follow-up data to show that radiologists' ability to characterize findings is improved with the T scan,
specifically by using jackknife alternative free-response receiver operating characteristic (JAFROC) methods.
(Aim 2): Perform quantitative analysis of the 3D breast outline segmentation, texture, and density. With breast
phantoms, we have previously shown that the breast volume is overestimated in the conventional scan and is
calculated more accurately in the T scan. We aim to show that the same result holds in human subjects by
calculating volume differences between the two scanning methods. Additionally, we will analyze power-law noise
and higher-order non-Gaussian texture measures as surrogate metrics of detectability and tissue superposition
effects, which we expect to be improved by the T scan. Finally, we will analyze whether percent density
calculations differ between the two scanning methods since we expect fewer out-of-plane artifacts in the T scan.
Although the new method of scanning is not being used as part of the volunteers' medical care, the overall impact
of this study is to demonstrate improvements in specificity and thus the potential to minimize the number of
diagnostic imaging exams and biopsies, lower healthcare costs, and minimize the total radiation dose combining
screening and diagnostic imaging. Women with dense breasts will especially benefit from this new design since
dense tissue can obscure findings in a conventional DBT scan, making them harder to characterize.

Grant Number: 5R37CA273959-04
NIH Institute/Center: NIH
Principal Investigator: Raymond Acciavatti