Maximum A Posteriori material decomposition for spectral photon-counting CT: application to human blood iron level estimation

Purpose: To develop a dose-efficient image-based material decomposition technique for spectral photon-counting computed tomography (PCCT) data and investigate estimating human blood iron concentration from contrast- enhanced scans. Methods: We adapt a maximum a posteriori (MAP) approach to decomposition, formulating spectral material decomposition as maximizing a posterior likelihood that incorporates both the standard linear generative model of decomposition and a smoothness prior. Our approach employs numeric tensor algebra and software, which can naturally handle the high-dimensional nature of decomposition. To ensure accurate priors, we compute smoothness weights using the image created from all detected photons. Our MAP approach only requires a large and sparse linear system to solve, with one tuning parameter. Results: We test the algorithm on 4-energy threshold spectral PCCT scans of a human subject pre- and post- contrast. MAP estimates remain stable while reducing noise standard deviation by 80.1% and 75.4% for iron and iodine, respectively, which again suggests over 4x decrease in radiation. Aortic iron concentration measured from MAP had small bias post-contrast, but with a noise reduction of roughly 80%. This small bias (-5%) in iron content may be attributed to the blood volume increase after contrast injection. Conclusion: The dose-efficient MAP decomposition method shows improved precision over the standard approach in estimating blood-iron concentration. Future work will include additional human studies to determine the optimal trade-off between precision and algorithmic bias.