Spectral CT reconstruction via Spectral-Image Tensor and Bidirectional Image-gradient minimization

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Spectral computed tomography (CT) faces challenges in image quality due to limited photon counts per energy bin, leading to low signal-to-noise ratio (SNR) projections.
• Photon-counting detectors in spectral CT require advanced reconstruction techniques to overcome inherent noise limitations.

Purpose of the Study:

• To develop a novel algorithm for enhancing image quality in spectral CT reconstruction.
• To introduce a new regularizer based on L0-norm constrained bidirectional image gradients within a tensor decomposition framework.

Main Methods:

• Formulation of a weighted bidirectional image gradient with L0-norm constraint for spectral CT images.
• Integration of this gradient constraint as a regularizer into a tensor decomposition model, creating the SITBIM algorithm.
• Optimization of the SITBIM model using the split-Bregman method.

Main Results:

• The SITBIM algorithm demonstrated superior performance compared to state-of-the-art methods like TVM, TV+LR, SSCMF, and NLCTF.
• Validation through experiments on numerical mouse phantoms and real mouse data confirmed the effectiveness of SITBIM.
• The method successfully improved image quality in spectral CT reconstruction.

Conclusions:

• The proposed SITBIM algorithm effectively addresses the image quality limitations in spectral CT.
• SITBIM offers a significant advancement in spectral CT image reconstruction, outperforming existing techniques.
• The L0-norm constrained bidirectional image gradient regularizer is a promising approach for spectral CT imaging.