Fig. 1. Research progress in multimode fiber imaging technology. (a) Point scanning imaging device of fiber output end face using digital phase conjugation method^[5]; (b) measurement of the transmission relationship between fiber input and output by the transmission matrix method^[48]; (c) fast imaging of fiber scattering spots using compressed sensing approach^[13]; (d) recognition accuracy of handwritten digital images transmitted by 1 km optical fiber using deep learning^[30]

Fig. 2. Multimode fiber fluorescence imaging. (a) Papadopoulos et al. detected cellular and subcellular details with a multimode fiber endoscope^[5]; (b) Vasquez-Lopez et al. clearly identified dendritic spines by confocal microscopy imaging and multimode fiber imaging^[7]; (c) Stibůrek et al. used lateral multimode fiber endoscopy for cerebral cortex imaging and cortical vascular imaging in mice^[10]; (d) Wen et al. used multimode fiber for volumetric imaging of three-dimensionally distributed fluorescent microspheres^[8]; (e) Deng et al. used synthetic datasets to drive deep learning networks for deblurring and denoising of spatially varyed degraded images^[64]; (f) Zhang et al.utilized wavelength-tuned multi-frame acquisition imaging method to achieve signal-to-noise ratio 4 times higher than that of single frame^[63]

Fig. 3. Multimode fiber reflection imaging. (a) Macroscopic imaging of mechanical clocks with a multimode fiber endoscope^[26]; (b) macroscopic imaging of open peppers by multimode fiber endoscopy^[26]

Fig. 4. Multimode fiber Raman imaging. (a) Wide-field micrographs of polystyrene microspheres and Raman imaging by Gusachenko et al. ^[18]; (b) Raman imaging of polystyrene monobeads and CaSO₄ blocks by Deng et al. ^[19]; (c) Coherent anti-Stokes Raman scattering imaging of methacrylic bead and polystyrene bead samples obtained by Trägårdh^{et al.[20]; (d) coherent anti-Stokes Raman scattering imaging of myelin sheaths and other tissues in cross sections of sciatic nerves obtained by Pikálek et al.[21]}

Fig. 5. Second-harmonic polarization imaging of proteins using multimode fibers by Cifuentes et al.^[22]

Fig. 6. White light electronic endoscope combined with multimode fiber^[65]. (a) White light electronic endoscope images showing regions of healthy colon tissue; (b) stitching fluorescence images of healthy colon [regions 1 and 2 in Fig. 6(a)] by endoscopy with multimode fiber

Fig. 7. Multimode fiber image transmission imaging. (a) Example of reconstructed fiber input amplitude map generated by Rahmani et al.using output amplitude scatter map and convolutional neural network^[28]; (b) scatterplot of images of jumping cats after fiber optics and reconstructed recovery maps obtained by Caramazza et al.^[29]; (c) schematic of the combination of time stretching method and multimode fiber endoscope for ultrafast imaging by Liu et al.^[34]

Fig. 8. Research progress in bending-resistant imaging of multimode fibers. (a) Schematic of the virtual beacon approach proposed by Farahi et al.^[66]; (b) schematic of reflectance values for some of the reflectors proposed by Gu et al.^[67]; (c) recovery imaging results of multimode fibers after bending by modeling calculations combined with numerical corrections proposed by Plöschner et al.^[62]; (d) schematic of discovery by Loterie et al. that shape translation in optical fibers does not change propagation properties^[68]; (e) schematic of reflected light reconstruction transmission matrix via end-plane reflectors proposed by Gordon et al.^[70]

Fig. 9. Research progress on deep learning-based bending-resistant imaging of multimode optical fibers. (a) Fan et al. proposed neural network training by multiple multimode fiber bending states^[71]; (b) Resisi et al. used a deep convolutional neural network combined with group training on multiple low correlation fiber morphologies^[73]

Fig. 10. Principle of spatial-frequency coding tracking based adaptive beacon light-field-encoded method^[65]