Joint phase retrieval and alignment via deep learning for advancing 3D refractive index measurement in interferometric tomography

In optical interferometric tomography (OIT), phase retrieval and alignment are crucial steps for achieving a precise measurement of the three-dimensional refractive index (3D RI) distribution within a sample. Traditionally, completing these two tasks requires multiple independent steps, each of which relies on extensive handcrafted prior knowledge, thereby constraining both the efficiency and precision of the measurements. To address this problem, we utilize deep learning techniques and propose a novel multi-task transformer network (MTTN) that can simultaneously retrieve the phase from single-frame interference patterns and align it with the scene center. We evaluate MTTN on both simulated and real-world data. The results demonstrate that MTTN achieves precise phase retrieval and alignment without additional prior knowledge. By integrating it with tomographic reconstruction techniques, we successfully reconstruct the 3D RI maps of micron-scale specialty optical fibers. We believe this framework provides a new technical solution for addressing challenges in OIT and related fields.