Metasurface‐based Fourier ptychographic microscopy

Meta-optics have opened new possibilities for portable, high-performance microscopy, offering ultrathin and highly customizable wavefront control in scenarios where bulky optics limit adoption. Here, we use this capability to overcome the long-standing challenges of Fourier ptychography (FP), a powerful computational technique for wide-field, high-resolution quantitative phase imaging that traditionally depends on large optical elements and extensive angle scanning. Our compact meta-FP platform combines a 4-f metalens system for imaging miniaturization with a programmable thin-film transistor (TFT) panel to provide stable, angle-diverse plane-wave illumination without mechanical movement. To further accelerate imaging, we introduce a residual convolutional neural network (RCNN) model trained via transfer learning on conventional FP datasets, which allows for single-shot inference of high-resolution phase from low-resolution inputs. Experimental validations demonstrate nearly twofold resolution improvement (7.81 µm-3.91 µm), accurate quantitative phase recovery on phase standards with errors below 10 %, and dry-mass estimation of H1975 cells with an average deviation of approximately 12 %, while the best-performing regions exhibit deviations below 0.5 %. This integration of metasurface optics and artificial intelligence-driven reconstruction provides a promising pathway for fast and compact FP microscopy with applications in live-cell imaging, microfluidic monitoring, and point-of-care diagnostics.