Simulation Study Of Bioluminescence Tomography Based On Improved Half-threshold Method

Bioluminescence Tomography(BLT)estimates the distribution of internal light source with the limited measurement information that obtained from the surface of organisms.BLT realizes the observation of a variety of biological processes at the molecular level.BLT is a noninvasive in vivo imaging modality and has become a research hotspot in the field of molecular imaging.BLT has shown great potential in preclinical and medical research,for example,the early detection and treatment of malignant tumors,drug development and evaluation of curative effects etc.However,BLT source reconstruction is a typical ill-posed inverse problem.Near-infrared light travels through multiple scattering and absorption regions in biological tissues,which increases the difficulty of source reconstruction.Therefore,the wide application of BLT still depends on the continuous development of reconstruction algorithm.In this paper,we focus on BLT light source reconstruction and the main work can be divided into two aspects:First,the reconstruction problem of BLT is modeled as an L1/2 regularization problem.Half thresholding algorithm is introduced into BLT to solve the non-convex optimization model.In order to evaluate the performance of the algorithm,we designed simulation experiments on multiple groups of digital mouse models.The HTA algorithm is compared with the iterative soft threshold algorithm for L1 norm regularization reconstruction model.Digital mouse simulation experiments and convergence analysis show that the performance of HTA is better than that of IST algorithm in terms of the center location error under the condition of using mixed spectrum measurements and without any a priori information of the region.However,there are some problems in the reconstruction of double-source,such as large deviation of reconstruction position,IST converges slightly faster than HTA and so on.In order to further improve the quality of the reconstructed images,in this study we present an improved algorithm based on the HTA,called Percentile Half Thresholding Pursuit Algorithm(Percentile HTPA),which combines HTA with Subspace Pursuit(SP)and percentile threshold method.Specifically,the solution obtained by HTA is treated as a candidate set of SP,and the final result is produced by SP iteration.The improved algorithm combines the advantages of matching pursuit and half thresholding algorithm.Moreover,the threshold selection is based on data-driven sorting model,which reduces the artificial intervention in reconstruction and makes the precision of reconstruction further improved.The results of several groups of simulation on the digital mouse model show that the proposed algorithm has obtained more accurate reconstruction results under different light source settings compared with the iterative reweighted algorithm and the original HTA algorithm.The convergence experiment results show that the convergence speed of Percentile HTPA is similar to that of HTA but the central location error of Percentile HTPA is smaller.