Standardized Images and Evaluation Metrics for Tomography.

Area of Science:

• Medical imaging
• Computational science
• Image processing

Background:

• Advanced tomographic reconstruction methods achieve high image fidelity.
• Global image quality metrics saturate, limiting assessment of residual differences.
• Distinguishing performance gaps in high-fidelity imaging requires more sensitive evaluation.

Purpose of the Study:

• Introduce a physically grounded, standardized evaluation framework for tomographic reconstruction.
• Enhance sensitivity beyond conventional global metrics.
• Facilitate comparison and systematic improvement of reconstruction methods.

Main Methods:

• Defined standardized reference images (Source, Detector, Ideal, Realistic) using Monte Carlo simulations.
• Employed pixel-wise difference and χ2 maps, ROI analysis, and histogram comparisons.
• Utilized Structure and Contrast Index (SCI) on difference maps for evaluation.

Main Results:

• SCI and χ2 diagnostics identified structured residuals and localized deficiencies missed by global metrics (SSIM, NMSE).
• Methods with similar global scores showed distinct residual structures and region-specific performance.
• Sinogram domain agreement did not guarantee improved image fidelity.

Conclusions:

• The framework offers reproducible, physically meaningful, and sensitive evaluation for high-fidelity tomographic reconstruction.
• Combines standardized references with multi-domain, multi-metric analysis for robust benchmarking.
• Supports physically consistent interpretation of reconstruction quality.