This article is a case-based review that introduces applied statisticians to a number of issues and methods that arise in clinical studies with paired digital images as outcomes. Single photon emission computed tomography (SPECT) is the imaging modality used in two examples. The first is a physical simulation of relevant clinical features of SPECT images using a customized head phantom scanned under different experimental conditions. The objective is to demonstrate and compare several current methods for image registration, i.e., image superimposition in some optimal manner to obtain a common frame of reference within which to make pixel-by-pixel comparisons. Image registration together with image normalization to correct for spurious differences in background activity levels enable quantification of differences in paired images. The physical simulation assesses quantification accuracy. The second example involves two SPECT images of the same brain tumour patient taken about two months apart. The objective here is to demonstrate several tools from morphological image analysis for image segmentation, i.e., separation of 'figure' and 'ground', and simple image regression to detect patterns of differences in time, again after registration and normalization. The clinical context of both examples helps to keep the review in practical focus and is a useful starting point for descriptions of many standard tools in applied medical image analysis. References to relevant literature are provided for readers wanting to study particular subjects in depth. As medical image analysis becomes more sophisticated, more cost-conscious and scrutinized more intensively, proper attention to the computational and statistical bases for practical conclusions become even more crucial.