Super-Energy-Resolution Material Decomposition for Spectral Photon-Counting CT Using Pixel-Wise Learning

Spectral photon-counting CT offers novel potentialities to achieve quantitative decomposition of material components, in comparison with traditional energy-integrating CT or dual-energy CT. Nonetheless, achieving accurate material decomposition, especially for low-concentration materials, is still extremely challenging for current sCT, due to restricted energy resolution stemming from the trade-off between the number of energy bins and undesired factors such as quantum noise. We propose to improve material decomposition by introducing the notion of super-energy-resolution in sCT. The super-energy-resolution material decomposition consists in learning the relationship between simulation and physical phantoms in image domain. To this end, a coupled dictionary learning method is utilized to learn such relationship in a pixel-wise way. The results on both physical phantoms and in vivo data showed that for the same decomposition method using lasso regularization, the proposed super-energy-resolution method achieves much higher decomposition accuracy and detection ability in contrast to traditional image-domain decomposition method using L1-norm regularization.

Automated summary

Spectral photon-counting CT shows potential for quantitative material decomposition compared to traditional CT methods, but current challenges include accurate decomposition of low-concentration materials. This study proposes a super-energy-resolution method with coupled dictionary learning to improve material decomposition accuracy, showing significant progress in both physical phantoms and in vivo data.