A beam-hardening predictive index (BHPI) for simulation-guided spectral optimisation in laboratory X-ray CT

Beam hardening remains a major source of imaging artefacts in X-ray computed tomography (XCT), particularly with multi-material samples. We propose here the mitigation of beam hardening not as a reactive correction, but as part of a more proactive scan planning. We introduce a pre-acquisition scalar metric, called beam-hardening predictive index (BHPI), and demonstrates its use as a primary metric for selecting X-ray spectra that minimise beam-hardening effects, with peak signal-to-noise ratio (PSNR) serving as a secondary image-quality constraint, whilst avoid photon starvation, under- and over-exposure. A simulation-based spectral analysis was performed using a digital twin of the imaging system to evaluate tube-voltage and filtration combinations. Representative optimal and suboptimal spectra were identified from the BHPI–PSNR performance landscape and subsequently used to acquire experimental XCT data for (i) 3D-printed cyanoacrylate-infiltrated powder, (ii) steel-lead-resin, (iii) zirconium-vanadium alloy-resin, and (iv) copper wire specimens. The BHPI-guided spectra consistently produced reconstructions with reduced beam-hardening artefacts and improved intensity uniformity. For the copper wire, BHPI alone identified a superior configuration despite lower PSNR, demonstrating that BHPI captures beam-hardening sensitivity that PSNR cannot. These improvements also resulted in more stable multi-level Otsu segmentation under optimised conditions. Supporting metrics showed that noise performance was not degraded: SNR remained comparable or improved, and total variation (TV) decreased under the optimised spectra. Overall, the results demonstrated that BHPI provides a practical and experimentally reliable criterion for identifying acquisition parameters that reduce beam-hardening artefacts while maintaining acceptable image quality, offering a foundation for beam-hardening-aware scan planning in XCT. The resulting improvement in image quality also facilitates more reliable and computationally efficient segmentation, reducing dependence on advanced correction algorithms.