Dynamic pixel binning allows spatial and angular resolution tradeoffs to improve image quality in X-ray C-arm CT

When designing an acquisition protocol for a C-arm CT, a tradeoff has to be made between scan time and the amount of acquired data. Increased scan time allows data acquisition with higher spatial and angular resolution. Fixed scan time implies a reduction of either spatial and angular resolution, which may either lead to oversmoothing or to artifacts from angular undersampling. In this work, we investigate smart sensors for pushing the boundary for this bottleneck considerably further. Detector pixels are enhanced by processing units that are able to perform a limited amount of local operations. This allows an online, dynamical adaptation of the pixel binning, depending on whether a local neighborhood is smooth or not. We demonstrate on phantom data that this approach allows to save about 75% of the data, obtaining a reconstruction quality comparable to 1 × 1 binning at a data rate comparable to 2 × 2 pixel binning.