Federated Learning Architectures for Mental Health Data: A Systematic Review of Privacy-Preserving Approaches and Clinical Efficacy
Abstract
Mental-health information are sensitive by nature but need to be aggregated to train robust machine-learning models capable of detecting depression, anxiety, PTSD, and bipolar disorders at scale. Federated learning (FL) is a privacy-safeguarding paradigm in which the models are locally trained and shared only via gradients or distilled parameters; however, plain vanilla FL continues to leak patient data via model weights or memorisation. From 2022 to 2025, the space quickly coalesced around differential-privacy (DP)-enabled FL, secure-aggregation methods, and customized federated fine-tuning for balancing high privacy budgets (ϵ ≤ 1) with clinical effectiveness. We conducted a systematic review of 43 peer-reviewed papers that deployed FL for mental-health prediction or monitoring on human data. Across depression, PTSD, bipolar, and stress tasks under stress, depression, PTSD, and bipolar conditions, privacy-hardened FL attained a pooled AUC of 0.87 ± 0.03, or a mean reduction of only 1.2% (95% CI 0.9–1.6%) compared to centralised baselines below the 3% non-inferiority threshold accepted by regulatory bodies for model equivalence. Perhaps most encThisagingly, multimodal wearable-to-text pipelines and large-language-model-based therapy chatbots provided the biggest privacy utility improvements when paired with adaptive DP budgets and LoRA-style parameter-efficient fine-tuning. Deployment issues persist non-independent and identically distributed (non-IID) data heterogeneity…
How strong is the evidence?
Evidence Hub reads this as a systematic review. Its evidence level is 1 of 5 (Highest · pooled evidence). Pooled evidence from several studies sits at the top of the evidence pyramid.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.