Laser-driven ultrafast transmission electron microscopy (UTEM) transcends conventional imaging by functioning as an intelligent sensing platform that concurrently captures structural and electronic dynamics with ångström-femtosecond precision. This review systematically frames UTEM within the paradigms of measurement science and data science. We dissect the intrinsic trade-offs governing its metrological performance-particularly the space-charge limit that couples temporal resolution to spectral coherence-and benchmark the distinct operational regimes of thermionic versus field-emission architectures. Furthermore, we analyze the multidimensional data streams generated by photon-induced near-field microscopy (PINEM) and time-resolved EELS, emphasizing the critical role of physics-informed machine learning in denoising, quantum state tomography, and closed-loop experimental optimization. We posit that the evolution of UTEM from a visualization tool into a knowledge-discovery engine hinges on the synergistic fusion of ultrafast optics, electron optics, and artificial intelligence.

Ultrafast TEM; Pump-probe; PINEM; Time-resolved EELS; Machine learning