Time-lapse analysis of 4D seismic data acquired at different stages of hydrocarbon production or fluid/gas injection has been very successful at capturing detailed reservoir changes (e.g., pressure, saturation, fluid flow). Conventionally, 4D seismic analysis is performed in the time-migrated domain assuming a fixed migration velocity model; however, this scenario is violated when the subsurface velocity is significantly altered by production/injection effects resulting in large time-shift anomalies and complex 4D wavefield coda. For these scenarios we argue that one should use a robust 4D analysis procedure involving iterative wave-equation prestack depth migration and time-lapse velocity analysis. We adapt 3D image-domain wave-equation migration velocity analysis (WEMVA) to such time-lapse scenarios to backproject discrepancies (residuals) in migrated baseline, monitor, and time-lapse images to estimate 4D velocity model perturbations. We highlight the differences between the 3D and 4D WEMVA inversion problems, and how we constrain 4D perturbation estimates to preferentially be updated within the reservoir zone. We demonstrate the benefits of various 4D WEMVA strategy in a set of synthetic experiments that involve estimating time-lapse model perturbations arising from a thin layer (<20m) of injected CO2 in a North Sea analogue reservoir.