Adapting software systems while they are running is important, especially when they involve components used for human-computer interaction. Model transformation techniques are commonly used to change and improve the software architecture of such systems. However, current transformations often focus only on functional requirements, not on quality attributes. This paper introduces a process that considers quality attributes during model transformations to enhance the quality of software adaptations. By using software architecture metrics to evaluate quality attributes, the process selects the best transformation among many options. The process was tested in a geographic information system called ENIA, which has user interfaces based on coarse-grained components that need to be adjusted while the system is running.