Room: ePoster Forums
Purpose: X-ray-induced luminescence computed tomography (XLCT) is an emerging molecular imaging. At present, XLCT image reconstruction has been entirely relied on inversely solving the diffusion equation which is used to model the luminescence photon propagation in biological tissues. Due to the limited measurements in boundary, the inverse problem of solving the diffusion equation is severe ill-conditioned, which leads to very poor spatial resolution and strong artifact in reconstructed XLCT images. In this study, we present a novel deep learning based image reconstruction method to improve the spatial resolution in XLCT.
Methods: First, a deep neural network was developed and well trained using Monte Carlo simulation as a gold standard. Second, the network was validated with synthetic digital mouse atlas model. Finally, a generic deep neural network based XLCT reconstruction framework was proposed and validated to improve the spatial resolution and reduce the artifact in reconstructed XLCT images.
Results: Extensive numerical simulations were conducted to validate our proposed approach for XLCT reconstruction. In particular, simulated phosphor targets having different shape, size, and edge-to-edge distance were embedded in the digital mouse atlas were investigated. The results show that our deep learning based XLCT reconstruction method has higher spatial resolution, higher contrast, less background noise, and less image artifact through quantitative comparison with conventional XLCT reconstruction algorithm.
Conclusion: A generic deep learning based XLCT reconstruction framework has been proposed and developed. It will has high impact on the development of XLCT imaging for high spatial resolution in vivo small animal whole body molecular imaging. And it will offer a powerful tool for pre-clinical and clinical studies.