Research
Functional and Quantitative Breast Tomosynthesis
Digital breast tomosynthesis (DBT) has rapidly been introduced for clinical use as an adjunct or replacement of mammography. Currently, most hospitals in the Netherlands use DBT for detection and diagnosis of breast cancer. This is also the case in the rest of Europe and North America. However, once the cancer is diagnosed, functional information about the tumor, as opposed to anatomical information such as its shape and extent as obtained with DBT, is needed to optimize and monitor treatment. In this project, we are working on extending the capabilities of tomosynthesis to create a new functional imaging modality: 1) dual-energy DBT (DE-DBT) for the distinction between cysts from solid lesions at screening, without contrast injection nor increase in the radiation dose; and 2) quantitative dynamic contrast-enhanced (QDCE)-DBT. The former would results in the reduction of recalls of healthy women for additional testing. And the latter method will help us obtain a complete characterization of tumor status before and throughout treatment, which may improve local and systemic therapy planning, response monitoring, and outcome prediction, reducing current breast cancer morbidity and mortality. The creation of QDCE-DBT will involve the development, optimization, and testing of novel DBT image acquisition techniques and of spectral reconstruction, deep learning-based image quantification, and motion correction algorithms.
Objectives of the project (see Figure 1):
i. develop and optimize the required image acquisition methods for QDCE-DBT.
ii. improve the dual-spectrum reconstruction algorithm for use in QDCE-DBT.
iii. develop deep learning-based methods for quantitative accuracy in QDCE-DBT.
iv. adapt these methods for cyst vs. solid mass characterization with DE-DBT.
Fgure 1: the improject involves four work packages (WPs) corresponding to the four objectives listed above. Three WPs will result in QDCE-DBT for treatment personalization and monitoring, and one WP will develop DE-DBT for cyst vs. mass discrimination at screening.
DBT acquisitions setup can be seen in Figure 2.
Figure 3. (top) Slice of a digital dynamic breast phantom, showing (bottom) different contrast enhancement curves in the different tissues.
Our reconstruction method uses a deep learning-based segmentation to constrain the iodine component of the reconstruction only to the lesion area, limiting the artifacts and spread in the top-down direction seen in the standard reconstruction.
Researchers: