Deep system prior based graph convolution network for NIR-II fluorescence molecular tomography

Fluorescence molecular tomography (FMT) is a promising imaging technique that can quantify the internal distribution of tumor in the early stage. However, due to the ill-posed inverse problem caused by the severe photon scattering effect, the promotion of efficiency and accuracy is still an issue for FMT and the reconstruction of the morphological performance is still difficult to meet the practical requirement.

In this paper, the second near-infrared (NIR-II) fluorescence imaging was adopted to mitigate tissue scattering to alleviated ill-posedness, and a deep system prior based graph convolution network (DSPGN) was proposed for FMT, which fully takes the morphology represented by graph-structure into the reconstruction process to improve the morphological performance of FMT. Specifically, besides the single fluorescence image, the spatial prior of system represented by the nodes association and imaging system is also input into the network. Through feature extraction and embedding, more prior knowledge is incorporated into the reconstruction region. Then, a graph convolution network is adopted to make full use of the topological information of FMT data, coupled with an attention mechanism, the fluorescence source is reconstructed.

To evaluate the performance of DSPGN, numerical simulation and in vivo experiments were carried out. The results show that, compared to existing methods, DSPGN can achieve superior performance in terms of location accuracy and especially shape recovery capability.

The proposed DSPGN has good ability for location and morphological fluorescence source recovery and has the potential to promote the application of FMT in NIR-II application.