Adaptive phase retrieval refinement for fast and accurate TIE-based quantitative phase imaging

We propose an adaptive refinement framework to enhance the accuracy of Transport of Intensity Equation (TIE) based Quantitative Phase Imaging (QPI). Standard TIE solvers often require regularization filters that smooth out high spatial frequencies, leading to a loss of fine structural details. To address this, we integrate a high-speed, GPU-accelerated matrix-based TIE solver (MS-TIE) with an iterative optimization loop. This refinement step enforces physical consistency by minimizing the discrepancy between experimental and numerically propagated intensities. Simulations demonstrate that our method, ’MS-TIE+Optimization’, significantly reduces reconstruction artifacts and outperforms standard methods, achieving a 35% increase in PSNR and nearly doubling the SSIM.