Fig. 1. (a) Schematic diagram of our proposed NFTPM method. (b) The physics prior (forward image formation model) of NFTPM.

Fig. 2. Comparison of phase (0–1 rad) reconstruction results of a simulated cell sample using NFTPM, BlindNet, TIE, AI-TIE, and GS methods under different illumination settings and a defocus distance of 5 μm. (a) Results under coherent illumination. (b) Results under partially coherent illumination (S=0.85). (c) Comparison between AI-TIE and NFTPM for the large phase (0–6 rad) under partially coherent illumination (S=0.85). (d) Comparison between BlindNet and NFTPM after downsizing the network under coherent illumination. (e), (f) Phase line profiles corresponding to (c), (d).

Fig. 3. Comparison between AI-TIE and NFTPM with an incorrect initial value of the defocus distance in various illuminations. (a) Intensity images under circular illuminations (S=0.10, 0.40, and 0.75) and annular NA-matched illumination. (b) The results of AI-TIE. (c) The results of NFTPM. (d) The results of NFTPM in the presence of noise. (e) Ground truth. (f) Convergence curves of the defocus value and loss value (in logarithmic form). (g) The defocus distance prediction process of NFTPM for intensity images simulated at other defocus distances under annular illumination.

Fig. 4. Experimental observation of HeLa cells via NFTPM under annular NA-matched illumination (see Visualization 1). (a) The full FOV of the reconstruction result of NFTPM. (b) The defocus distance prediction process. (c1), (c2) The phase retrieved by AI-TIE. (d1), (d2) The phase retrieved by NFTPM.