Mental health disorders affect large numbers of people worldwide and are a major cause of long-term disability. Digital health technologies such as mobile apps and wearable devices now generate rich behavioural data that could support earlier detection and more personalised care. However, these data are highly sensitive and distributed across devices and platforms, which makes privacy protection and scalable analysis challenging; federated learning offers a way to train models across devices while keeping raw data local. When combined with edge, fog, or cloud computing, federated learning offers a way to support near-real-time mental health analysis while keeping raw data local. This review screened 1104 records, assessed 31 full-text articles using a five-question quality checklist, and retained 17 empirical studies that achieved a score of at least 7/10 for synthesis. The included studies were compared in terms of their FL and edge/cloud architectures, data sources, privacy and security techniques, and evidence for operation in real-world settings. The synthesis highlights innovative but fragmented progress, with limited work on comorbidity modelling, deployment evaluation, and common benchmarks, and identifies priorities for the development of scalable, practical, and ethically robust FL systems for digital mental health.
Keywords: cloud computing; edge computing; federated learning; machine learning; mental health; privacy preserving.