A Neural Network-Based Cross-Bin Pileup Effect Correction Method for Photoncounting CT

Photon-counting detectors (PCDs) with multiple energy bins have demonstrated significant clinical potential in spectral imaging for medical computed tomography (CT). However, the benefits of PCDs can be compromised by various physical non-idealities, particularly the pulse pileup effect. This effect arises when incident photons arrive in close temporal proximity, especially at clinically relevant high-flux rates, leading to decreased output counting rates, distorted measured spectra, and consequently, degraded image quality. Conventional correction methods employ precise physical detection models, such as the nonparalyzable model, to recover pileup-free counts from pileup-affected counts for individual energy bins; however, these approaches ignore the strong correlations among measured data across different energy bins. In this study, we propose a neural network-based approach to correct the pileup effect on measured counts across multiple energy bins of PCDs. Our method leverages a physics-guided detector model to restore spectral responses unaffected by pileup and to generate training datasets. Promising preliminary results demonstrate the feasibility of our approach.