Iterative cone-beam CT reconstruction with an off-center flat-panel detector for SPECT/CT

Objectives X-ray CT imaging with an off-center flat-panel detector, as used in the BrightView XCT SPECT/CT system, is challenging since the acquired data consists of laterally truncated cone-beam X-ray projections. Each projection covers half of the field-of-view with a small overlap region. Filtered-backprojection reconstruction can be performed using redundancy weighting and truncation compensation with opposite projections. In some of the resulting reconstructions, left-right and up-down intensity imbalance are observed that are caused by the approximate nature of the reconstruction algorithm. This imbalance can lead to HU intensity differences at the seam between CT volume segments. To reduce these artifacts, we use iterative reconstruction algorithms that can handle the native truncated cone-beam projection data.

Methods Algebraic and statistical maximum-likelihood algorithms, implemented on a graphics processing unit, were used to reconstruct 14 cm axial segments of clinical patient data that were subsequently knitted together. Differences in the HU values on both sides of the knitting seam were measured.

Results Image uniformity is significantly improved with both iterative reconstruction methods. In 5 problematic cases, intensity steps at the volume seam were reduced from an average of 76 HU (SD 43 HU) to an average of 16 HU (SD 16 HU). Additionally, the maximum-likelihood reconstructions show lower noise and less streak artifacts, compared to iterative algebraic or filtered-backprojection reconstruction, with noise reduction of up to 30% in soft tissue areas in proximity to bone.

Conclusions Iterative reconstruction considerably improves CT image quality which may enable improved localization in SPECT/CT