Accelerated reconstruction for identifying image regions affected by rigid body movement

A bootstrap method for identifying image regions affected by rigid body movement has recently been proposed. The aim of the present investigation was to significantly decrease the computational requirements of this method, allowing it to be feasibly implemented on a clinical system. We devised a new method that utilizes tomographic reconstruction with coarser sampling. We hypothesized that we could then recover the spatial distribution of the variance by adopting a method inspired by super-resolution. This method ensured clear identification of regions corrupted by motion. We performed comparisons between the conventional bootstrap method and our accelerated method. We found that we could produce a similar coefficient of variation (CV) image and that variance due to motion was much larger than aliasing effects arising from the coarser sampling . Our implementation took advantage of the muIticore architecture of a 16 core workstation (4 × Intel Xenon E5620). We demonstrated a 14 times reduction in the required computational resources. The total duration for processing took less than a typical scan time, thus we foresee that this method could be implemented in real time.