Deep Learning‐Enabled Pixel‐Super‐Resolved Quantitative Phase Microscopy from Single‐Shot Aliased Intensity Measurement

Abstract A new technique of deep learning‐based pixel‐super‐resolved quantitative phase microscopy (DL‐SRQPI) is proposed, achieving rapid wide‐field high‐resolution and high‐throughput quantitative phase imaging (QPI) from single‐shot low‐resolution intensity measurement. By training a neural network with sufficiently paired low‐resolution intensity and high‐resolution phase data, the network is empowered with the capability to robustly reconstruct high‐quality phase information from a single frame of an aliased intensity image. As a graphics processing units‐accelerated computational method with minimal data requirement, DL‐SRQPI is well‐suited for live‐cell imaging and accomplishes high‐throughput long‐term dynamic phase reconstruction. The effectiveness and feasibility of DL‐SRQPI have been significantly demonstrated by comparing it with other traditional and learning‐based phase retrieval methods. The proposed method has been successfully implemented into the quantitative phase reconstruction of biological samples under bright‐field microscopes, overcoming pixel aliasing and improving the spatial‐bandwidth product significantly. The generalization ability of DL‐SRQPI is illustrated by phase reconstruction of Henrietta Lacks cells at various defocus distances and illumination patterns, and its high‐throughput anti‐aliased phase imaging performance is further experimentally validated. Given its capability of achieving pixel super‐resolved QPI from single‐shot intensity measurement over conventional bright‐field microscope hardware, the proposed approach is expected to be widely adopted in life science and biomedical workflows.