ABSTRACT Dedicated breast CT is an emerging technology with approximately 20-25 clinical prototypes and clinical systems worldwide. It does not require physical compression of the breast, can eliminate breast tissue superposition, and provides 3D images at near-isotropic spatial resolution. Earlier generations of breast CT, while demonstrating the concept, were suboptimal due to technological limitations and available components. In the prior funding (R01 CA199044), we focused on the hardware aspects: tested 4 detectors from different vendors, followed by designing, developing, and testing a dedicated breast CT system that employed an offset-detector geometry to enable imaging large breasts. This system achieved high resolution and low system noise and was specifically designed to improve chest-wall coverage and operate at a radiation dose suitable for breast cancer screening. In the pilot study completed under prior funding (R01 CA199044), the system demonstrated visualization of the pectoralis muscle in 177/179 (98.9%) breasts (effective diameter at chest wall: 12.82.1 cm); a remarkable improvement over the 40-78% reported in prior studies, and the mean glandular dose (MGD) was 4.10.9 mGy, which is similar to that reported for digital mammography (4.15 mGy) in the Digital Mammography Imaging Screening Trial (DMIST) study. In this competitive renewal, we focus on image reconstruction followed by a multi-reader, multi-case (MRMC) receiver operating characteristic (ROC) study using prospectively acquired data to evaluate if dedicated breast CT demonstrates improved performance over the current standard, digital breast tomosynthesis (DBT) for breast cancer screening. For image reconstruction, we will investigate a total of 5 techniques that were carefully chosen to encompass the current and emerging methods including the standard Feldkamp-Davis-Kress (FDK) reconstruction, one compressed sensing-based iterative reconstruction that is well-founded on a rigorous mathematical framework, and three deep learning-based reconstruction techniques with increasing sophistication, generalizability, and scientific rigor. The deep learning-based methods include a fully-supervised technique that requires input-reference image pairs and is tuned using physics-based measure, a self-supervised technique that does not require an independent reference image, and a self-supervised technique that is based on both the imaging physics and the mathematics of image reconstruction and does not require independent training data. These methods are not only applicable to dedicated breast CT, but are adaptable to cone-beam CT, in general, and potentially to the commonly used multi-detector CT. The MRMC ROC study is based on well-founded image science and is designed with statistical rigor. To our knowledge, there have been no prior studies comparing the diagnostic accuracy of DBT and breast CT at radiation dose levels suitable for breast cancer screening. This research address...