Fifty-sixth annual meeting of the American association of physicists in medicine
TU-F-18C-04: When Is the Pixel/voxel Size Sufficiently Fine for Digital Breast Tomosynthesis (DBT) Virtual Clinical Trial (VCT) Studies: Lessons Learned From a Simulation Study
When conducting DBT virtual clinical trial (VCT) studies, it is often advocated that the object be modelled as finely as possible and the detector pixel size match the physical detector. Finer pixel/voxel size inevitably increases the computational cost and storage requirements, especially when using a large number of images. Are such requirements always justified? This work offers some answers to this important question.
To check the effect of pixel/voxel size, we compared the results from two DBT simulations, one with a coarser (500 micron) and another with a finer (125 micron) pixel/voxel size. This VCT was aimed to investigate whether the optimization of DBT angular span and number of projection views is sensitive to the choice of reconstruction algorithm. We implemented four commonly used reconstruction algorithms, and used a set of numeric breast phantoms to simulate DBT acquisition under various geometry parameters. Lesion-detectability based image quality curves as a function of angular span and number of views were calculated for each reconstruction algorithm. These curves were used to determine whether the system performance trend changes with reconstruction algorithm. For this VCT study, both the coarser and finer simulation study followed the same model for data acquisition and evaluation but used different size when modeling the object and the x-ray detector.
The simulations using the coarse and fine pixel/voxel size both confirmed that the choice of the reconstruction algorithm was not critical in optimizing DBT system geometry for lesion detectability. This agreement indicates that a coarser detail was sufficient for this particular VCT study involving tasks of detecting masses rather than detecting micro-calcifications.
Pursuing very fine pixel/voxel size may not be necessary, depending on the specific task. A small pilot study comparing coarser and finer simulations can help determine if coarser pixels/voxels would suffice