A data-efficient method for local noise power spectrum (NPS) estimation in FDK-reconstructed 3D cone-beam CT

Area of Science:

• Medical Imaging
• Image Reconstruction
• Signal Processing

Background:

• Nonstationary noise in computed tomography (CT) systems necessitates local noise power spectrum (NPS) characterization.
• Previous work developed a data-efficient radial NPS method for 2D fan-beam CT (FBP) reconstruction.

Purpose of the Study:

• To extend the radial NPS method for estimating the 3D local NPS in feldkamp-davis-kress (FDK) reconstructed cone-beam CT (CBCT) volumes.
• To leverage the polar separability of CT NPS for efficient 3D noise characterization.

Main Methods:

• Derived the 3D expression of polar separability for CBCT local NPS based on CBCT geometry and FDK reconstruction.
• Decomposed the 3D local NPS into a 2D radial NPS shape function and a 1D angular amplitude function.
• Evaluated accuracy using simulated and real CBCT data, comparing normalized mean squared error (NMSE) and lesion detectability.

Main Results:

• Achieved very small NMSE (<5%) with the 3D radial local NPS method using only two CBCT scans.
• Traditional local NPS methods required approximately 20 scans for comparable accuracy.
• Task-based system performance metrics (lesion detectability) using the proposed method closely matched actual system performance.

Conclusions:

• The polar separability significantly reduces data dimensionality for 3D CBCT local NPS estimation.
• The developed radial local NPS method accurately estimates 3D local NPS from just two CBCT scans.
• The minimal data requirement suggests potential clinical utility of local NPS in CBCT applications.