Method for measuring noise-power spectrum independent of the effect of extracting the region of interest from a noise image.

This study aimed to evaluate the impact of region of interest (ROI) size on noise-power spectrum (NPS) measurement in computed tomography (CT) images and to propose a novel method for measuring NPS independent of ROI size. The NPS was measured using the conventional method with an ROI of size P × P pixels in a uniform region in the CT image; the NPS is referred to as NPS_R=P. NPSs_{R=256, 128, 64, 32, 16, and 8} were obtained and compared to assess their dependency on ROI size. In the proposed method, the true NPS was numerically modeled as an NPS_model, with adjustable parameters, and a noise image with the property of the NPS_model was generated. From the generated noise image, the NPS was measured using the conventional method with a P × P pixel ROI size; the obtained NPS was referred to as NPS'_R=P. The adjustable parameters of the NPS_model were optimized such that NPS'_R=P was most similar to NPS_R=P. When NPS'_R=P was almost equivalent to NPS_R=P, the NPS_model was considered the true NPS. NPSs_{R=256, 128, 64, 32, 16, and 8} obtained using the conventional method were dependent on the ROI size. Conversely, the NPSs (optimized NPSs_model) measured using the proposed method were not dependent on the ROI size, even when a much smaller ROI (P = 16 or 8) was used. The proposed method for NPS measurement was confirmed to be precise, independent of the ROI size, and useful for measuring local NPSs using a small ROI.