Advances and Challenges of Nonlinear Optical Microscopy Based on Artificial Intelligence

Nonlinear Optical Microscopy (NLOM) is an imaging technique based on materials' nonlinear optical responses, which enables high-spatial-resolution, deep-tissue-penetrating, low-phototoxicity and label-free imaging to capture the structural, functional and molecular information of biological samples. As a cornerstone tool in biomedical research, it includes two-photon microscopy (TPM), three-photon microscopy (3PM), second harmonic generation (SHG), coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS). However, traditional NLOM faces intrinsic bottlenecks, including diffraction-limited resolution, anisotropic imaging quality, photon noise interference, and inefficient data processing. With the rapid advancement of artificial intelligence (AI), particularly deep learning, these limitations are being progressively addressed. AI algorithms, such as convolutional neural networks (CNNs), generative adversarial networks (GANs), transformers, and meta-learning frameworks, have been integrated with NLOM to achieve super-resolution reconstruction, denoising, real-time volumetric imaging, multi-modal data fusion, and rapid model adaptation. This review systematically summarizes the recent advances of AI-driven NLOM, focusing on key technical innovations in resolution enhancement, noise suppression, imaging speed acceleration, intelligent pathological analysis, and intraoperative diagnosis. Furthermore, the current challenges and future perspectives in this interdisciplinary field are discussed, aiming to provide insights for the development of more powerful, versatile, and user-friendly NLOM techniques for biological and clinical applications.