Limited-data computed tomography algorithms for the physical sciences

Area of Science:

• Physical Sciences
• Image Reconstruction
• Computed Tomography

Background:

• Limited-data computed tomography (CT) presents reconstruction challenges.
• Various algorithms exist, but their performance varies with data quality and object characteristics.

Purpose of the Study:

• To compare the performance of five limited-data CT reconstruction algorithms.
• To identify the most effective algorithm under simulated real-world conditions.

Main Methods:

• Algorithms evaluated: Algebraic Reconstruction Technique (ART), Multiplicative Algebraic Reconstruction Technique (MART), Gerchberg-Papoulis, spectral extrapolation, and Singular Value Decomposition (SVD).
• Reconstruction of phantom data with realistic noise levels and diverse imaging geometries.
• Algorithm implementations optimized with prior knowledge (object sign, extent, smoothness) using cubic B-spline basis functions.

Main Results:

• MART yielded the best overall reconstruction results.
• ART performed optimally for very smooth objects or highly noisy data (20-dB SNR).
• Average reconstruction error of 1.7% achieved with MART for moderate-to-steep gradients at 30-dB SNR with five viewing angles.

Conclusions:

• MART is a highly effective algorithm for general limited-data CT reconstruction.
• ART remains valuable for specific scenarios involving smooth objects or noisy datasets.
• Incorporating a priori information significantly enhances reconstruction accuracy.