Liu T, Ruan J, Rong J, Hao W, Li W, Li R, Zhan Y, Lu H. Cone-beam X-ray luminescence computed tomography based on MLEM with adaptive FISTA initial image.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE 2023;
229:107265. [PMID:
36455470 DOI:
10.1016/j.cmpb.2022.107265]
[Citation(s) in RCA: 4] [Impact Index Per Article: 4.0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Received: 07/04/2022] [Revised: 11/07/2022] [Accepted: 11/22/2022] [Indexed: 06/17/2023]
Abstract
BACKGROUND AND OBJECTIVE
As an emerging dual-mode optical molecular imaging, cone-beam X-ray luminescence computed tomography (CB-XLCT) has shown potential in early tumor diagnosis and other applications with increased depth and little autofluorescence. However, due to the low transfer efficiency of PNPs to convert X-ray energy to visible or near-infrared (NIR) light and X-ray dose limitation, the signal to noise ratio of projections is quite low, making the quality of CB-XLCT relatively poor.
METHODS
To improve the reconstruction quality of low-counts CB-XLCT imaging, an adaptive reconstruction algorithm (named ADFISTA-MLEM) based on the maximum likelihood expectation estimation (MLEM) framework is proposed for CB-XLCT reconstruction from Poisson distributed projections. In the proposed framework, the image reconstructed by fast iterative shrinkage-thresholding algorithm (FISTA) is used as the initial image for MLEM iterations to improve reconstruction accuracy, in which both the projection noise model and the sparsity constraint of the image could be considered. For relative quantitative imaging, a specific normalization is applied to the projection data and system matrix. To determine the hyperparameter of FISTA, which may be different for different projections, an adaptive strategy (ADFISTA) is then designed for adaptive update of the hyperparameter with reconstructed image in each iteration.
RESULTS AND CONCLUSIONS
Results from numerical simulations and phantom experiments indicate that the proposed framework can obtain superior reconstruction accuracy in terms of contrast to noise ratio and shape similarity. In addition, high intensity-concentration linearity between different probe targets indicates its potential for quantitative CB-XLCT imaging.
Collapse