Φ-Net-based Few-Views Reconstruction for Sparse Optical Projection Tomography
Nan Wang,Pengfei Huang,Huiyuan Wang,Lin Wang,Hui Xie,Duofang Chen,Shenghan Ren,Hu Cai,Xiaodong Wang,Shouping Zhu,Xueli Chen
DOI: https://doi.org/10.1016/j.bspc.2024.106678
IF: 5.1
2024-01-01
Biomedical Signal Processing and Control
Abstract:Sparse-views data-based reconstruction scheme is essential for the in vivo application of optical projection tomography (OPT). Such schemes significantly shorten the data acquisition time and reduce the radiation dose to the sample, thus facilitating in vivo longitudinal imaging of living samples. Iterative algorithms combined with a regularization strategy have been extensively employed for the reconstruction of sparse OPT. However, the reduction in the number of projections is typically limited to just a few tens of projection images. Here, we developed a (p-net-based few-views reconstruction algorithm for sparse OPT. In the algorithm, a network architecture similar to a (p-shape ((p-net) was constructed by integrating a U-net structure with a squeeze-and- excitation inception module. The constructed (p-net is capable of reconstructing images of reasonable quality using less than 10 projection views. Considering the difficulty of experimentally obtaining enough OPT data to train the network, and the similarity between OPT to X-ray computed tomography (CT) in terms of imaging principles, we used CT data from a public database as a training dataset to train the constructed (p-net. The trained network was first validated by CT data from the same database as an internal testing cohort and then migrated to experimentally measured OPT data for sparse OPT reconstruction. The experimental results demonstrated that the (p-net-based few-views reconstruction algorithm was capable of reconstructing OPT images of good quality by using single-digit-view projections, thus well suited for long-term dynamic imaging and analyzing living samples.