Compensating for charge sharing by a deep-learning method: a preliminary experimental study

TL;DR

This study explores a deep learning approach to mitigate charge sharing in photon counting detectors for CT, demonstrating improved accuracy over traditional methods through experimental comparison.

Contribution

It introduces a deep learning method for charge sharing compensation in PCDs and validates its effectiveness through experimental results.

Findings

Deep learning reduces bias and standard deviation in charge sharing correction.

The method outperforms hardware anti-coincidence in low energy bins.

It achieves smaller standard deviation in high energy bins.

Abstract

Photon counting detectors (PCDs) bring valuable advantages to diagnostic computed tomography (CT), including lower noise and higher resolution than energy integrating detectors. However, there are still several nonideal factors preventing PCDs from meeting people's expectations, for example, charge sharing and pile up. In this paper, we did some preliminary work on charge sharing and conducted an experimental study using an XCounter PCD to compare the effects of no anti-coincidence, anti-coincidence by hardware and charge sharing compensation by a deep learning method. In our results, a smaller bias and standard deviation are obtained from deep learning method than directly from no-anti-coincidence mode of the detector. Our network also outperforms the anti-coincidence mode of the detector in the low energy bin and has smaller standard deviation in the high energy bin. The results validate that a deep learning method is suitable to compensate for charge sharing.