Fig. 1. Schematic diagram of hologram generation and deep learning-based holographic reconstruction. (a) Schematic diagram of digital hologram generation; (b) structure of the general deep learning hologram reconstruction model with dual encoders and dual decoders; (c) structure of the general deep learning hologram reconstruction model with single encoder and dual decoder; (d) structure of the deep learning hologram reconstruction model in proposed

Fig. 2. Network structure diagram of GNN reconstruction model. (a) Overall network structure of the reconstruction model; (b) downsampling network structure with a residual architecture; (c) schematic diagram of GNN construction principle; (d) upsampling network structure with bridging connection; (e) network structure at the output end

Fig. 3. Optical path diagram of digital hologram acquisition

Fig. 4. Testing results of the two models on holograms of mitotic slice samples of animal cells. (a)‒(c) Holograms; (d)‒(f) amplitude images reconstructed using the traditional method; (g)‒(i) phase images reconstructed using the traditional method; (j)‒(l) amplitude images reconstructed using the GNN model; (m)‒(o) phase images reconstructed using the GNN model; (p)‒(r) amplitude images reconstructed using the Y-Net model; (s)‒(u) phase images reconstructed using the Y-Net model

Fig. 5. Reconstruction performance of the GNN model on two sample types. (a)(b) Holograms of Aspergillus samples; (c)(d) holograms of small intestine wall slice samples; (e)‒(h) amplitude images reconstructed using the traditional method; (i)‒(l) phase images reconstructed using the traditional method; (m)‒(p) amplitude images reconstructed using the GNN model; (q)‒(t) phase images reconstructed using the GNN model