Skip to content

Latest commit

 

History

History
113 lines (71 loc) · 6.03 KB

README.md

File metadata and controls

113 lines (71 loc) · 6.03 KB

🧠 FarajaMH: Generative AI for Early Mental Health Screening in Low-Resource Settings

FarajaMH is an experimental research project developing and evaluating a generative AI assistant for early screening of anxiety, depression, and psychosis through ordinary, natural-language conversations. The project seeks to bridge the gap in mental health screening and referrals in low-resource settings such as Kenya and Tanzania, where community health workers (CHWs) often serve as the first point of contact for care but have limited tools and mental health training.

🌍 Background

In East Africa, common mental disorders (CMDs) frequently go undetected due to shortages of mental health specialists and limited culturally adapted screening tools. Traditional structured instruments, while clinically validated, often miss local idioms of distress and contextual nuances (Marangu et al., 2021).

FarajaMH aims to close this gap by creating a conversational AI model that understands and responds to both English and Kiswahili expressions of mental distress — empowering CHWs to conduct early, empathetic, and contextually grounded screening at the community level.

🧩 Foundations and Collaborations

FarajaMH builds upon three major initiatives:

  1. INSPIRE Mental Health (INSPIRE MH) — a Wellcome Trust–funded project that harmonized 50,000+ longitudinal mental health records from HDSS sites across Kenya, Uganda, and Tanzania using the OMOP Common Data Model and DDI Lifecycle. This provides the AI-ready data backbone for FarajaMH.

  2. Weather Events and Mental Health Analysis (WEMA) — explored the link between climate shocks and mental health using participatory digital storytelling in informal settlements such as Mukuru Kwa Reuben, Kenya. This strengthened community trust and destigmatized mental health dialogue.

  3. Perinatal Mental Health Study in Korogocho — linking screening to maternal and child outcomes, reinforcing longitudinal engagement and ethical research practices. This work informs FarajaMH’s co-design with community and clinical partners.

🧠 The FarajaMH Model

The FarajaMH generative AI model is fine-tuned using culturally rich, annotated data sources:

  • INSPIRE MH datasets aligned to DSM-5-TR symptom mappings
  • Clinical notes from Mathari Teaching and Referral Hospital (Kenya)
  • Conversational data from the Bonga app (Kenya)
  • CHW chat data from Kenya and Tanzania
  • Voice data from the Kisesa HDSS referral hospital (Tanzania) capturing acoustic markers of distress

All datasets undergo ETL harmonization into the OMOP CDM framework, documented under DDI Lifecycle for transparency and FAIR compliance.

The model aims to:

  • Recognize culturally specific expressions of mental distress
  • Generate context-sensitive responses in English and Kiswahili
  • Support early screening, triage, and referral decisions
  • Operate safely and ethically in community and clinical environments

⚙️ System Components

Component Description
Fine-tuning pipeline Uses harmonized OMOP data and annotated conversational datasets.
FAISS-based RAG system Enables real-time, context-aware question recommendations during live conversations.
Multimodal input Integrates text, voice, and acoustic features for richer inference.
CrewAI multi-agent architecture Simulates clinician–patient–listener–recommender interactions to improve realism and model learning.
Web Application (FarajaMH Tool) Front-end for pilot deployment, supporting live speech-to-text, conversational screening, and dynamic feedback dashboards.

🧪 Research Design

FarajaMH will be evaluated through a four-step step-wedge design across HDSS sites in Kenya and Tanzania:

  1. Model Training & Technical Validation Test for accuracy, coherence, and safety using controlled datasets.
  2. Pilot Deployment in HDSS Settings Supervised real-world screening with CHWs and persons with lived experience (PLEs).
  3. Usability & Acceptability Study Evaluate cultural resonance, user trust, and clinical integration.
  4. Ethical & Policy Review Assess implications for responsible AI adoption in African mental health systems.

Ethical oversight will be provided by a dedicated Ethics and Trustworthy AI Committee.

🧭 Vision and Impact

FarajaMH is more than a technical innovation — it is an African-led, ethically governed, community-validated AI initiative designed to:

  • Enable early identification of anxiety, depression, and psychosis
  • Empower CHWs through AI-supported guidance
  • Reduce stigma and foster community dialogue
  • Position HDSS communities as co-creators of responsible AI in health research

🧾 Citation

If you use FarajaMH or its methods, please cite:

African Population and Health Research Center (APHRC). FarajaMH: Generative AI for Early Screening of Anxiety, Depression, and Psychosis in Low-Resource Settings. Nairobi, Kenya, 2025.

🧠 Acknowledgments

FarajaMH is led by the Data Science Program at the African Population and Health Research Center (APHRC) with support from the Wellcome Trust GenAI Accelerator. Collaborating institutions include Mathari Teaching and Referral Hospital, Kisesa HDSS, and multiple HDSS sites across East Africa.

⚖️ License

This project is released under the MIT License for research and non-commercial use. Ethical approval is required for any deployment involving human participants.