Model observer for optimizing digital breast tomosynthesis for detection of multifocal and multicentric breast cancer
MetadataShow full item record
The goal of medical imaging is to acquire and display images of human anatomy and function such that they can be optimally interpreted by a trained observer, e.g., a radiologist. Start-of-art medical image quality is measured by the performance of an observer on a given clinical task. Since psychophysical studies are resource intensive, model observers are widely used as a surrogate in task-based assessment of image quality. Model observers are typically designed to detect at most one abnormality, e.g., a single lesion. However, in clinical practice, there may be multiple abnormalities in a single set of images, which can have a significant impact on treatment planning and outcomes. For example, patients with multifocal and multicentric breast cancer (MFMC), i.e., the presence of two or more tumor foci within the same breast, are more likely to undergo mastectomy rather than breast conservation therapy. Detecting multiple breast tumors is challenging because the prevalence of tumors varies significantly across breast regions, and radiologists do not know the number or location of tumors a priori. The vision of this dissertation is that digital breast tomosynthesis (DBT) has the potential to improve the detection of MFMC, and may offer advantages such as fewer false-positive findings, lower cost, and better accessibility. This dissertation focuses on the design and applications of a model observer to optimize DBT system geometries for detection of multiple breast tumors. This is significant and innovative because prior efforts to optimize DBT image quality only considered unifocal breast cancer scenarios. We highlight the following two main aspects of contributions in this dissertation: (1) We have developed a novel model observer that detects multiple abnormalities in anatomical backgrounds. (2) We have employed the extended 3D multi-lesion model observer to identify DBT system geometries that are most effective for the detection of MFMC. Our results demonstrate that the presence of more than one tumor present distinct challenges to DBT optimization, and that DBT geometries that yield images that are informative for the task of detecting unifocal breast cancer may not necessarily be informative for the task of detecting MFMC. We are validating the clinical relevance of our model observer studies with an ongoing human observer study with experienced breast imaging radiologists.