Predicting image properties in penalized-likelihood reconstructions of flat-panel CBCT

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Model-based iterative reconstruction (MBIR) algorithms, like penalized-likelihood (PL) methods, have data-dependent and shift-variant characteristics.
• Existing image quality predictors for MBIR rely on idealized models, neglecting real-world physical blurring and noise correlations.
• Accurate prospective estimation of local noise and spatial resolution is crucial for optimizing MBIR methods and system design.

Purpose of the Study:

• To develop and validate novel image quality predictors for flat-panel cone-beam CT (FP-CBCT).
• To incorporate physical system nonidealities specific to FP-CBCT into these predictors.
• To improve the accuracy of prospective image quality assessment in FP-CBCT.

Main Methods:

• Developed physical models representing FP-CBCT nonidealities: focal spot blur, scintillator blur, detector aperture effect, and noise correlations.
• Created FP-CBCT-specific predictors for local spatial resolution and local noise properties in PL reconstructions.
• Validated predictor accuracy against experimental CBCT measurements using specialized phantoms, comparing them to conventional predictors.

Main Results:

• FP-CBCT-specific predictors accurately estimated local spatial resolution and noise properties, outperforming conventional predictors.
• Conventional predictors showed significant deviations, underestimating spatial resolution (FWHM by 0.2 mm) and noise levels (by 70%).
• Proposed predictors demonstrated accurate estimations across various imaging conditions, including differing X-ray techniques and regularization strengths.

Conclusions:

• The developed image quality predictors accurately estimate local spatial resolution and noise for PL reconstruction in FP-CBCT.
• These predictors account for system geometry, X-ray technique, and patient anatomy, enabling more reliable MBIR analysis.
• The tools support prospective image quality analysis for system design, adaptive imaging, and robust MBIR tuning.