A unified Iso-SNR approach to task-directed imaging in flat-panel cone-beam CT

The signal-to-noise ratio (SNR) characteristics of volumetric flat-panel cone-beam CT images are investigated theoretically and experimentally. Analytical models of the noise-power spectrum (NPS) and voxel noise developed in the context of conventional, slice-based CT are extended to the fully 3-D case. While early models describe well the noise characteristics of 2-D tomographic reconstructions for the case of a 1-D detector with constant NPS, the fully 3-D approach extends classical descriptions to quantify the 3-D NPS and voxel noise in a manner that includes: a 2-D detector with generalized blur and NPS characteristics, secondary quantum noise in the conversion of incident x-rays, electronics noise, and 3-D aliasing. Classical relationships between image SNR and voxel size and "slice thickness" (viz., inverse cube-root and square-root dependence, respectively) are found inaccurate in describing the noise characteristics of 3-D flat-panel cone-beam CT. The relationship is shown instead to depend on the characteristics of the 3-D bandwidth integral associated with the 2-D detector modulation transfer function (MTF) and choice of reconstruction filter. Image SNR is investigated experimentally for a low-frequency soft-tissue visualization task using a prototype system for FPI-CBCT and a low-contrast phantom consisting of tissue-equivalent inserts in water. The contrast-to-noise ratio (CNR) in images of the low-contrast phantom is measured as a function of scatter-to-primary ratio (SPR), imaging dose, and spatial resolution in volume reconstructions and compared to theoretical expectations. The adaptability of flat-panel cone-beam CT is quantitatively revealed in a task-directed approach that seeks to "tune" image SNR through knowledgeable selection of dose and spatial resolution in a manner that is consistent with the imaging task and clinical constraints.