Real-time quantitative phase imaging through a multicore fiber using deep learning

Abstract

Quantitative phase imaging (QPI) is a vital label-free measurement modality in biomedicine, providing both morphology and quantitative biophysical information. Nevertheless, applying QPI in vivo or in hard-to-reach areas remains challenging. The multicore fiber bundle (MCF) is an emerging ultra-thin probe with a diameter of less than half a millimeter. QPI through MCFs is still difficult to achieve due to the different phase delays in 10,000 fiber cores. We propose a novel speckle reconstruction method using deep learning to achieve real-time phase reconstruction through a bare MCF. High fidelity phase image reconstruction of complex image datasets like ImageNet is achieved. The network further demonstrates good generalization behavior. Phase images belonging to a different data type, such as cancer tissue and test charts, can still be reconstructed through the MCF using the network. Such a powerful technique could open new perspectives for fiber endoscopic imaging.