Task-based Optimization of Acquisition Parameters in Digital Breast Tomosynthesis

University Of Chicago, Chicago IL

Investigators

Linked publications & trials

Abstract

Project Summary In this research project, we propose to perform task-based optimization of digital breast tomosynthesis (DBT) with respect to acquisition scheme using acquisition tailored image reconstruction algorithms. DBT is quickly being adopted in clinics alongside conventional 2D mammography for breast cancer screening. Mammography is inherently limited by the overlapping of 3D structures in a 2D image of the breast, allowing morphology indicative of breast cancer to be hidden by dense structures or mimicked by the superposition of overlapping tissues. DBT addresses this issue by providing 3D information via a limited- angle, tomographic approach. While the potential advantages of DBT have been widely demonstrated in the literature, optimization of the modality with respect to acquisition scheme remains an active area of research. A number of task-based optimization studies in DBT have been performed over the past decade, but these have typically confined their attention to a single reconstruction algorithm. The proposed work addresses this issue by performing a task-based comparison of acquisition schemes while employing acquisition-tailored reconstruction algorithms designed using task-based, objective metrics. Both traditional and advanced sparsity-exploiting reconstruction algorithms will be employed for this purpose. Through tailoring of the reconstruction algorithms, the work can also be expected to provide an objective improvement in image quality in the acquisition schemes considered. The specific aims of the proposed project are: (1) investigate traditional DBT reconstruction algorithms with varying acquisition scheme, (2) investigate sparsity-exploiting optimization-based DBT image reconstruction, (3) investigate and design DBT-specific image quality metrics for algorithm optimization, and (4) perform task-based assessment of acquisition-tailored algorithms with human observer studies. The first aim will address the tailoring of current image reconstruction methods to acquisition schemes of practical interest in order to establish a baseline for which to compare more advanced algorithms. The second aim will focus on the development and investigation of advanced, sparsity-exploiting, optimization- based DBT image reconstruction algorithms. This aim will involve both the development of efficient algorithms for, and the investigation of, these reconstruction techniques. In the third aim, task-based image quality metrics based on clinically relevant tasks in DBT will be developed for tuning of the previously developed optimization-based reconstruction algorithms, yielding acquisition tailored, advanced reconstruction algorithms. Lastly, in aim 4, task-based comparison of the acquisition tailored reconstruction algorithms will be performed with human observer studies.