Fig. 1. Classification of quantitative phase imaging techniques

Fig. 2. Description of dataset-driven network phase reconstruction^[6]. (a) Dataset collection; (b) Network training; (c) Prediction via a well-trained network

Fig. 3. Different network architectures and different training strategies for deep learning in phase reconstruction

Fig. 4. Structure of the generative adversarial network for reconstructing quantitative phase images in off-axis DH^[58]

Fig. 5. Two strategies of physical cascading networks

Fig. 6. CTF-Deep phase reconstruction principle and experimental results^[87]. (a) CTF-Deep phase reconstruction principle; (b) Phase step experimental imaging results, and compared with other methods

Fig. 7. Physics-enhanced deep neural network phase reconstruction method^[6]. (a) Untrained network solution; (b) Trained network solution

Fig. 8. Schematic diagram of dual-wavelength digital holographic imaging based on untrained neural network^[100]

Fig. 9. The method of physics-informed neural network^[113]. (a) Overview of the MaxwellNet network; (b) Tomographic reconstruction results of three-dimensional refractive index of polystyrene microspheres immersed in water using MaxwellNet and compared with Rytov prediction results

Fig. 10. Application of quantitative phase imaging in biological microscopy. (a) Phase imaging results of the internal structure of red blood cells^[58]; (b) Phase imaging results of various phytoplankton and zooplankton^[116]; (c) Observation of the division process of HeLa cells^[115]; (d) Effects of increased oxidative stress on sperm cells^[117]

Fig. 11. Applications of AI-QPI in industrial measurements. (a) Measurements of the radius (a_p) and refractive index (n_p) of a mixture of four monodisperse populations of polystyrene and silica spheres^[120]; (b) Measurement of microlens arrays^[51]; (c) Measurement of alcohol droplet evaporation by DL-SRQPI^[115]

Fig. 12. Multi-prior physics enhanced neural network and imaging results^[102]. (a) The structure of multi-prior physics enhanced neural network; (b) Large FOV and high-resolution phase imaging results of the phase resolution plate

Fig. 13. Description of phase unwrapping method based on deep learning^[133]. (a) Deep-learning-performed regression method. (b) Deep-learning-performed wrap count method; (c) Deep-learning-assisted denoising method

Fig. 14. Phase aberration correction method based on deep learning. (a) Description of the segmentation-based aberration correction method^[6]; (b) Comparison of the results of SSCNet and other phase aberration compensation methods^[148]