Uncertainty estimation and statistical comparative methodology for mammography x-ray energy spectra

Area of Science:

• Medical Physics
• Radiological Physics
• Computational Physics

Background:

• Numerical models offer an alternative to direct x-ray energy spectra measurements.
• Validation of these models requires robust comparative methods to assess spectral similarity.
• Accurate spectral characterization is crucial for reliable mammography dose assessment.

Purpose of the Study:

• To compare various methodologies for assessing the similarity between experimental and simulated x-ray mammography spectra.
• To identify the statistical method with the highest power for spectral comparison.
• To evaluate the impact of additional uncertainties on spectral analysis and dose calculation.

Main Methods:

• Generated 65 experimental and Monte Carlo (MC) simulated x-ray mammography spectra using various targets, filters, and breast tissue equivalent materials.
• Detected transmitted x-ray beams using Cadmium Telluride (CdTe) solid-state spectrometry.
• Applied quantitative comparative methods including chi-squared (χ²) statistics, half-value layers, mean/effective energies, and non-parametric u-test.
• Propagated additional uncertainties using the MC method and assessed the specificity of each comparative method.

Main Results:

• The chi-squared (χ²) statistics demonstrated the highest score, indicating it as the most suitable metric for spectral comparison.
• Additional uncertainties significantly contributed to total spectra uncertainty (up to 98%) and influenced the mean of evaluated χ².
• Non-parametric tests showed poor performance, failing to distinguish spectra pairs with substantial discrepancies in polyenergetic normalized glandular dose (DgNp).

Conclusions:

• The chi-squared (χ²) statistic is the recommended method for evaluating x-ray energy spectra comparisons in mammography.
• Accounting for additional uncertainties is critical for accurate spectral analysis and dose estimation.
• Non-parametric tests are not advised for this application due to their limited ability to detect significant spectral differences impacting clinical dose values.