Online Dictionary Learning Method for Cone-beam X-ray Luminescence Computed Tomography Reconstruction: A Preliminary Simulation Study

Cone-beam X-ray luminescence computed tomography (CB-XLCT), an emergent hybrid imaging modality, has witnessed rapid advancements and is poised to emerge as a pivotal tool for the early three-dimensional detection of tumors. Nevertheless, the impediments posed by weak light absorption and pronounced scattering phenomena in biological tissue render the CB-XLCT reconstruction task a severely ill-conditioned inverse problem, thereby hampering the attainment of satisfactory reconstruction outcomes. This investigation introduces an innovative strategy rooted in Online Dictionary Learning (ODL), underpinned by Iterative Reduction FISTA (IR-FISTA), to surmount these challenges and enhance CB-XLCT's reconstruction efficacy. The ODL framework encompasses two primary stages: sparse coding and dictionary updating. During the sparse coding phase, the IR-FISTA algorithm is employed, distinguished by its rapid computational velocity and heightened accuracy in solution derivation. Subsequently, in the dictionary update stage, an online algorithm leveraging stochastic approximations is implemented to optimally exploit the distinct structural attributes of the problem domain, thereby facilitating efficient extraction of sparse features. To validate and quantify the performance enhancements, a series of numerical simulation trials were orchestrated. The resultant data evidences that this novel approach significantly outperforms conventional reconstruction methodologies, exemplified by heightened localization precision, improved morphological restoration, and augmented robustness, thereby affirming its potential in advancing the frontiers of CB-XLCT.