Dynamic real-time reconstruction and rendering of breast tomosynthesis images

DESCRIPTION (provided by applicant): It is widely believed that digital breast tomosynthesis (DBT) has the potential to replace mammography in the future, based on preliminary clinical results. At least five companies are currently developing DBT imaging systems, three of whom have little or no experience developing primary review workstations. Real-Time Tomography (RTT) has the experience to develop medical workstations. RTT is currently in the process of developing a physician review workstation for the primary diagnosis of digital breast tomosynthesis images. We believe that a dedicated workstation is necessary based upon the large size of DBT datasets, the stringent image quality requirements for breast imaging, and the need for high-throughput in both breast screening and diagnosis. To address the need for high-throughput, we specifically propose to develop and implement a reconstruction algorithm that will allow dynamic real-time reconstruction and rendering (DRR) of arbitrary planes through the breast volume on a dedicated high-end PC-based graphics processor unit (GPU). We believe that the advances in GPU technology make it both possible and preferential to reconstruct tomosynthesis images on demand. There are many reasons to favor DRR. First, the GPU hardware that exists today allows nearly instantaneous back-projection and filtered (BPF) reconstruction of tomographic images; this hardware is essentially off-the- shelf and readily available. Second, breast tomosynthesis datasets currently being produced are anisotropic, having in-plane resolution of approximately 0.07-0.1 mm2 and out-of-plane resolution of 1 mm. This anisotropy is necessitated by the size of the resultant image set, and the time required to produce those data. However, in DBT it is possible to reconstruct images at any location; DRR would allow arbitrary reconstruction without adversely impacting either storage or speed. The long-term goal of RTT is to develop a primary review workstation based on the DRR principle. In this Phase I proposal, we seek to implement and optimize our existing DBT reconstruction algorithm to demonstrate feasibility. Our ultimate goal is to achieve a reconstruction rate of at least 5 frames per second (fps). In Phase I, we propose the following specific aims: (1) Implement and optimize BPF on a GPU using general-purpose GPU programming techniques. (2) Implement a viewer with a simple graphical user interface to render the reconstructed images. (3) Assess the image quality of the rendered images by comparing the GPU-reconstructed images to those produced with our current CPU-based algorithm using clinical and phantom images from the University of Pennsylvania as test data. The Phase I project will be deemed successful if a sustained reconstruction and rendering rate of 2 fps can be achieved. In Phase II, we would continue the development of the workstation, specifically working on methods to increase the frame rate, minimize user interactions and obtain the greatest possible throughput. Mammography (including digital mammography) is subject to a number of fundamental limitations because 2D images are made of the 3D breast anatomy. Mammograms can produce false positive findings due to the superposition of normal tissues, and cancers can be missed in mammograms because normal tissue hide or mask the cancer. Digital breast tomosynthesis (DBT) is a tomographic imaging method with the potential to solve both of the above problems of mammography. Tomosynthesis imaging systems are being developed by multiple manufacturers. However, as has been made clear from the ongoing deployment of clinical digital mammography systems, the development of the acquisition technology is only a small part of the effort that will be required to make DBT a clinical reality. We propose to develop a primary physician review workstation for DBT. This is an essential part of any clinical DBT system.