Improving model–data mismatch for photon‐counting detector model using global and local model parameters

• Published in: Medical physics (Lancaster) Vol. 51; no. 2; pp. 964 - 977
• Main Authors: Lee, Donghyeon, Zhan, Xiaohui, Tai, W. Yang, Zbijewski, Wojciech, Taguchi, Katsuyuki
• Format: Journal Article
• Language: English
• Published: United States 01-02-2024
• Subjects: charge sharing; computed tomography; photon counting detectors; pulse pileup; computed tomography; charge sharing; photon counting detectors; pulse pileup
• ISSN: 0094-2405; 2473-4209
• DOI: 10.1002/mp.16883

Background
An energy‐discriminating capability of a photon counting detector (PCD) can provide many clinical advantages, but several factors, such as charge sharing (CS) and pulse pileup (PP), degrade the capability by distorting the measured x‐ray spectrum. To fully exploit the merits of PCDs, it is important to characterize the output of PCDs. Previously proposed PCD output models showed decent agreement with physical PCDs; however, there were still scopes to be improved: a global model–data mismatch and pixel‐to‐pixel variations.
Purposes
In this study, we improve a PCD model by using count‐rate‐dependent model parameters to address the issues and evaluate agreement against physical PCDs.
Methods
The proposed model is based on the cascaded model, and we made model parameters condition‐dependent and pixel‐specific to deal with the global model–data mismatch and the pixel‐to‐pixel variation. The parameters are determined by a procedure for model parameter estimation with data acquired from different thicknesses of water or aluminum at different x‐ray tube currents. To analyze the effects of having proposed model parameters, we compared three setups of our model: a model with default parameters, a model with global parameters, and a model with global‐and‐local parameters. For experimental validation, we used CdZnTe‐based PCDs, and assessed the performance of the models by calculating the mean absolute percentage errors (MAPEs) between the model outputs and the actual measurements from low count‐rates to high count‐rates, which have deadtime losses of up to 24%.
Results
The outputs of the proposed model visually matched well with the PCD measurements for all test data. For the test data, the MAPEs averaged over all the bins were 49.2–51.1% for a model with default parameters, 8.0–9.8% for a model with the global parameters, and 1.2–2.7% for a model with the global‐and‐local parameters.
Conclusion
The proposed model can estimate the outputs of physical PCDs with high accuracy from low to high count‐rates. We expect that our model will be actively utilized in applications where the pixel‐by‐pixel accuracy of a PCD model is important.