Depression shows up in many places at once: in EEG rhythms and sleep, in heart-rate variability and daily activity, in the voice and in the words a person chooses. This project aims to bring these streams together in one model that reasons over a biomedical knowledge graph, so that detection, relapse warnings and treatment suggestions rest on more than a single signal.
Data in
- EEG and neurophysiological recordings from wearables
- Heart-rate variability, activity, sleep and skin conductance
- Clinical records, multi-omics and psychometric scales
- Speech, prosody and written language, with consent
What it is meant to deliver
- Depression detection and severity classification
- Risk stratification and early warning of onset or relapse
- Symptom trajectory modeling over time
- Treatment response prediction and what-if analysis
Overview
The full picture.
From signals of body, brain and speech to support that reaches patients and clinicians. Planned framework; no results are reported here.
01Multimodal data
- EEG & neurophysiologyfrequency bands, connectivity, sleep patterns, neural biomarkers
- Physiology & behaviorHRV, activity and sleep, skin conductance (EDA) and temperature, daily routines
- Clinical & multi-omicsEHR, genomics, metabolomics, psychometric scales, medication history
- Speech & languageprosody, voice quality, linguistic features, daily-life audio
- Text & contextclinical notes, transcripts, written language, social media with consent
02Harmonize & reason
Inside the engine
- Biomedical knowledge graph
- Graph neural networks
- Neuro-symbolic & causal AI
- Language & multimodal models
- Dynamic patient modeling
03Analysis & prediction
- Detection & classificationmild, moderate, severe
- Risk stratification & early warningonset, relapse, suicidal risk
- Symptom trajectory modelinglongitudinal prediction
- Biomarker & subtype discoverymultimodal, patient-specific
- Treatment response predictionmedication, psychotherapy, neuromodulation
- What-if & counterfactual analysis
04Personalized interventions
- Targeted treatment recommendations
- Digital therapeuticsmindfulness, CBT, neuromodulation
- Real-time monitoring & adaptive supportwearables and mobile apps
- Personalized feedback & early alerts
- Context-aware interventionslifestyle, sleep, activity, environment
- Clinician decision supportevidence-based, explainable
↻Human-centred, context-aware loop
- Patient context
- Continuous learning
- Clinician in the loop
- Explainable AI
✓Trustworthy, secure & ethical AI
- Privacy & data security
- Fairness & bias mitigation
- Explainability
- Robustness & reliability
- Regulatory compliance & ethics
- Early, accurate detection
- Personalized care
- Improved outcomes
- Scalable & equitable solutions
- Better mental health
Illustrations generated with ChatGPT for the AI4Health Lab. They depict research themes, not study data.