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

Area of Science:

• Medical Imaging Physics
• Radiology
• Quantitative Imaging

Background:

• Established methods for measuring computed tomography (CT) noise power spectrum (NPS) rely on uniform phantoms.
• Direct NPS estimation from patient images is crucial for real-time image quality assessment and protocol refinement in clinical practice.

Purpose of the Study:

• To develop and validate an automated workflow for estimating NPS directly from patient abdominal CT images.
• To assess the feasibility of using patient data for continuous image quality monitoring and CT protocol optimization.

Main Methods:

• An automated workflow integrating liver segmentation (TotalSegmentator) and patch selection was developed.
• Noise power spectrum (NPS) was computed using the PyLinac library on CT images from patients and a homogeneous phantom.
• Validation involved comparing NPS distributions using Kullback-Leibler divergence (DKL) between patient and phantom data.

Main Results:

• NPS estimation from abdominal CT images was successfully demonstrated.
• Low DKL values (<0.05) indicated strong agreement in spatial noise distribution between patient and phantom data under equivalent conditions.
• The developed method showed high fidelity in characterizing noise profiles from clinical scans.

Conclusions:

• An automated workflow for NPS estimation in the hepatic parenchyma of CT images is presented.
• The validated similarity between patient-derived and phantom-derived NPS profiles supports its clinical utility.
• This metric can serve as a complementary quantitative tool for CT quality control directly on patient data.