Feature-based CBCT self-calibration for arbitrary trajectories.

Area of Science:

• Medical Imaging
• Computer Vision
• Image Reconstruction

Background:

• Metal artifacts significantly degrade Cone Beam CT (CBCT) image quality.
• Accurate calibration of CBCT trajectories is crucial for artifact reduction.
• Existing calibration methods can be computationally intensive and sensitive to initial parameter guesses.

Purpose of the Study:

• To develop a novel algorithm for self-calibrating arbitrary CBCT trajectories.
• To reduce metal artifacts in CBCT imaging through improved trajectory calibration.
• To decrease computational runtime by reducing optimization parameters.

Main Methods:

• Employed feature detection (AKAZE) and brute force matching for 2D-3D projection registration.
• Calculated translational misalignment directly from feature position discrepancies.
• Aligned rotational parameters using a quartic function minimization approach.

Main Results:

• The developed algorithm, FORCASTER, demonstrated performance on par with state-of-the-art methods (BFGS with NGI objective).
• Quantitative metrics showed comparable results: nRMSE (FORCASTER: 0.3390 vs. BFGS+NGI: 0.3441), SSIM (FORCASTER: 0.83 vs. BFGS+NGI: 0.79), and Dice (FORCASTER: 0.86 vs. BFGS+NGI: 0.87).
• FORCASTER exhibited higher tolerance to initial parameter errors in experimental noise analysis.

Conclusions:

• The FORCASTER algorithm effectively calibrates projection orientations for arbitrary CBCT trajectories.
• Calibration quality is comparable to existing state-of-the-art algorithms.
• FORCASTER offers advantages in speed and robustness to initial parameter inaccuracies.