<p>A severe medical condition with high incidence, brain stroke lesions can lead to severe neurological impairment and the time-sensitive and accurate diagnosis of the condition is pivotal in its prompt treatment, which is why there is little consensus regarding accepted standard methods to improve the efficiency of diagnosis. In most of the models of the conventional stroke lesion progression machine learning (ML) approaches are used in the central processing. Nevertheless, these methods have the problems of data privacy, access and scalability. Recently, Federated Learning (FL), a prevalent solution to these issues, as it allows the collaborative model training while maintaining the data privacy. In this review paper, this study discusses the application of Federated Learning in brain stroke lesion modelling study progressions and presents additional new developments and techniques. Several recent studies on FL applied to stroke lesion segmentation, longitudinal progression prediction and multi-centre data collaboration are analyzed and compared, illustrating the design of the various datasets used, and the measures of performance tested. In addition to that, the review illustrates the benefits and limitations for Federated Learning in brain stroke lesion progression including protection of patient privacy, choosing heterogeneous data, and scalability across various healthcare institutions. Further in the paper, discuss some of the challenges encountered in federated learning environments like communication efficiency, model convergence as well as the integration of diverse data sources. This review aims to synthesise the above studies to give an overview of where we currently are with the application of Federated Learning in the context of stroke lesion progression studies and/or towards providing a direction of how we can exploit that subject to improve the diagnosis and personalised treatment strategy in the clinical settings.</p>