Iterative Reconstruction with Dynamic ElasticNet Regularization for Nuclear Medicine Imaging

This study proposes a novel image reconstruction algorithm for nuclear medicine imaging based on the maximum likelihood expectation maximization (MLEM) framework with dynamic ElasticNet regularization. Whereas conventional the L1 and L2 regularization methods involve trade-offs between noise suppression and structural preservation, ElasticNet combines their strengths. Our method further introduces a dynamic weighting scheme that adaptively adjusts the balance between the L1 and L2 terms over iterations while ensuring nonnegativity when using a sufficiently small regularization parameter. We evaluated the proposed algorithm using numerical phantoms (Shepp-Logan and digitized Hoffman) under various noise conditions. Quantitative results based on the peak signal-to-noise ratio and multi-scale structural similarity index measure demonstrated that the proposed dynamic ElasticNet regularized MLEM consistently outperformed not only standard MLEM and L1/L2 regularized MLEM but also the fixed-weight ElasticNet variant. Clinical single-photon emission computed tomography brain image experiments further confirmed improved noise suppression and clearer depiction of fine structures. These findings suggest that our proposed method offers a robust and accurate solution for tomographic image reconstruction in nuclear medicine imaging.