Efficient primal-dual algorithm for imaging applications with matrix stacking, applied to DBT image reconstruction

Area of Science:

• Medical Imaging
• Computational Mathematics
• Optimization Theory

Background:

• Convex optimization is crucial for tomographic imaging reconstruction.
• Primal-dual hybrid gradient (PDHG) algorithms are effective but sensitive to parameter selection.
• Digital breast tomosynthesis (DBT) requires high-resolution, accurate imaging.

Purpose of the Study:

• To simplify parameter selection for PDHG algorithms in multi-term optimization problems.
• To apply the simplified PDHG framework to wide-angle DBT image reconstruction.
• To evaluate the impact on quantitative accuracy and depth resolution.

Main Methods:

• Developed a simplified step-size parameter selection strategy for PDHG.
• Applied the PDHG framework to a specific image reconstruction problem in DBT.
• Compared quantitative accuracy and depth resolution with existing methods.

Main Results:

• The simplified parameter selection avoids extensive grid searches while maintaining efficiency.
• The PDHG framework successfully enabled the proposed optimization problem for DBT.
• Demonstrated improved quantitative accuracy and enhanced depth resolution in reconstructed DBT volumes.

Conclusions:

• The simplified PDHG approach offers an efficient and effective method for tomographic imaging.
• This technique shows promise for improving diagnostic capabilities in DBT.
• Further research can explore broader applications in medical imaging reconstruction.