Hiromichi Yokoyama

Investigation of optimization-based reconstruction for intra-operative neurological imaging

Intra-operative C-arm cone-beam CT (CBCT) provides significant technical and workflow benefits for neurological interventions. Current C-arm CBCT employs FDK-type algorithms for image reconstruction, which require densely sampled projection data. This imaging practice incurs a large amount of radiation dose and prolonged scanning time. Yet, the reconstruction is susceptible to artifacts caused by both geometry and physical factors such as noise. In this work, we investigated and evaluated the adaptation of optimization-based reconstruction to C-arm-CBCT-based neurological imaging from full-and half-view patient data. The adaptation mainly entailed accurate incorporation of C-arm geometry and appropriate selection of algorithm parameters. The results show that optimization-based algorithms can yield images of effectively suppressed noise and shading artifacts, from half of the amount of data acquired in current clinical applications.

Algorithm-enabled high-performance C-arm cone-beam CT angiography of cerebral vasculature

C-arm CBCT utilizing a flat-panel detector has been implemented in surgical suites for integration of tomographic imaging capabilities in the intra-operative workflow. In particular, cerebral vascular applications have emerged that rely on C-arm CBCT angiography for making clinical decisions on treatment selection. Current C-arm CBCT angiography employing analytic-based reconstruction algorithms has limitations on sub-optimal image quality due to irregular gantry motion, as well as issues of long scanning time and non-negligible imaging dose. In this work, we investigated the adaptation of an optimization-based algorithm for reconstructing C-arm CBCT images of patient cerebral vasculature from full- and reduced-view data. In particular, the gantry-motion irregularity was experimentally characterized and fully incorporated in the reconstruction algorithm. The results show that the optimization-based algorithm can yield visibly improved quality from full-view data, and some important vascular structures of potential clinical utility can be obtained from data containing substantially reduced number of views.