Point-scanned imaging modalities such as optical coherence tomography (OCT) are subject to frame rate limits governed by the image resolution, sampling rate, and scanner dynamics. For scenes with dynamic features on static backgrounds, adaptive scanning escapes these limits by visiting scan positions only as needed. We implemented adaptive scanning using a probabilistic approach that balanced re-imaging of known dynamic positions with exploration for undiscovered ones. We evaluated our approach in model systems that simulated multitarget tracking and ophthalmic surgery using a swept-source OCT system. We demonstrated frame rate speedups in excess of 5× without degradation in image quality, a performance that would have otherwise required a significantly faster source.