Tensor-based dictionary learning for dynamic tomographic reconstruction

Area of Science:

• Medical Imaging
• Image Reconstruction
• Computational Imaging

Background:

• Dynamic computed tomography (CT) reconstruction is limited by data acquisition speed, affecting spatio-temporal resolution.
• Compressed sensing has improved CT reconstruction from limited projections.
• Sparse representation is crucial for efficient image sequence analysis.

Purpose of the Study:

• To develop an adaptive method for training a tensor-based spatio-temporal dictionary for dynamic CT reconstruction.
• To improve the spatio-temporal resolution and image quality in few-view CT reconstruction.
• To incorporate nonlocal total variation for enhanced structural recovery.

Main Methods:

• An adaptive tensor-based spatio-temporal dictionary is trained for sparse representation of image sequences.
• The reconstruction problem is addressed using the alternating direction method of multipliers.
• Nonlocal total variation is integrated to preserve fine structures and edges.

Main Results:

• The proposed adaptive tensor dictionary method significantly outperforms vectorized dictionary-based reconstruction in few-view scenarios.
• Preclinical studies on sheep lung perfusion and dynamic mouse cardiac imaging validate the approach.
• The method effectively captures object characteristics by considering inter-atom and inter-phase correlations.

Conclusions:

• The adaptive tensor-based dictionary approach offers superior performance for dynamic CT reconstruction with limited data.
• This method enhances the recovery of fine details and sharp structures, crucial for diagnostic imaging.
• The findings suggest a promising direction for improving dynamic CT imaging in clinical and preclinical settings.