Automated patient-derived noise power spectrum estimation in abdominal CT: a preliminary validation

BACKGROUND: Established methods are available for measuring the noise power spectrum (NPS) in computed tomography (CT) by using uniform phantoms. However, in clinical practice, direct estimation of this metric from patient images is desirable for continuous image quality monitoring and protocol optimization. Methodology: An automated workflow was developed through the integration of liver segmentation with TotalSegmentator, optimized selection of square patches in potentially homogeneous regions using a greedy algorithm, and NPS computation employing the PyLinac library. CT images from five patients and one homogeneous phantom were analyzed using two reconstruction kernels (lung and standard). Validation was performed using Kullback- Leibler divergence (DKL) to compare the NPS distributions obtained from patients and phantoms under equivalent acquisition conditions. Results: NPS estimation from abdominal images was feasible. In nearly all paired com- parisons (phantom versus patient under identical acquisition and reconstruction conditions), DKL values between normalized NPS distributions were below 0.05, indicating strong agree- ment in the shape of the spatial noise distribution between the homogeneous phantom and clinical studies. Conclusion: An automated workflow for NPS estimation in the hepatic parenchyma is presented. The similarity between the characteristic NPS profile obtained from the patient images and that derived from a homogeneous phantom acquired under equivalent conditions was subsequently validated. These preliminary results support the potential integration of this metric as a complementary quantitative tool for CT quality control programs, directly on patient images.