For example, a pathologist may be interested in analyzing one region of interest of an image, and later in another region of interest in the same image, which may or may not overlap with the first region. In another example, the pathologist may want to compare one image with another image acquired for the same patient based on a new stain or morphology metric. In both examples, if the image features of the image are accessible without re-computing, the process for the new image analysis task can be significantly more efficient. In addition, for some image analysis tasks, image feature pre-computation can be indispensable to provide real-time user experience. For example, interactive image segmentation task may require the user to mark various image segments, to check the resulting segmentation, and to add additional marking to refine the segmentation. To enable this interactive process, the segmentation results after each marking need to be generated within seconds or fractions of a second. Accordingly, the amount of real-time re-computations needs to be minimized.