Practical considerations for noise power spectra estimation for clinical CT scanners

Area of Science:

• Medical Physics
• Radiological Imaging
• Image Processing

Background:

• Local noise power spectra (NPS) are crucial for characterizing CT imaging system noise.
• Existing calculation schemes significantly influence NPS properties.
• Standardized methods for local NPS estimation in clinical settings are needed.

Purpose of the Study:

• To analyze the effects of various calculation parameters on local NPS.
• To provide practical suggestions for estimating local NPS in clinical CT scanners.
• To compare the accuracy of local NPS calculations under different conditions.

Main Methods:

• Scanned a Catphan phantom using a Philips Brilliance 64 slice CT simulator with varied protocols.
• Reconstructed images using Filtered Back Projection (FBP) and iDose4 iterative reconstruction.
• Calculated local NPS with varied region of interest (ROI) parameters, background removal methods, and window functions.

Main Results:

• Local NPS varied with calculation parameters, especially below ~0.15 mm-1 spatial frequency.
• NPS calculation error decreased exponentially with increased ROI number in simulations.
• Image subtraction was the most effective background removal method, yielding consistent results.
• PCA with a Hann window closely matched image subtraction results (17.5% difference).

Conclusions:

• Local NPS analysis requires careful selection of small ROI sizes.
• Recommended minimum ROI size depends on radial bin size and pixel dimensions.
• Image subtraction is most accurate for background removal, but other methods can suffice with appropriate window functions.
• Understanding parameter dependencies is vital for accurate NPS interpretation in task-based image quality assessment.