Mental Health, Augmented Artificial Intelligence, Early Detection, Computer-Aided Diagnosis and Treatment, Telemedicine, Personalized Medicine, Impact on the Society

The goal of the MENTAL.AI project is to design and study new models for mental health computer-aided diagnosis, monitoring and therapy systems within the context of face-to-face and ECA-to-patient interviews. Namely, we propose to target three objectives - Monitoring, Automatic Diagnosis and Virtual Therapist -, within which we will propose solutions that overcome current limitations of state-of-the-art applications.

• Apathy,
• Avoidance,
• Excessive fear or uneasiness,
• Feeling of disconnection,
• Increased sensitivity,
• Mood changes,
• Problems thinking,
• Significant tiredness,
• Sleep or appetite changes,
• Withdrawal,
• Etc.

• 1P - Personalized: Personalized medicine consists of adapting a medical treatment according to the individual characteristics of a patient.
• 2P - Preventive: Preventive medicine focuses on wellness, and consists of measures taken for disease prevention.
• 3P - Predictive: Predictive medicine is a branch of medicine that aims to identify patients at risk of developing a disease.
• 4P - Participative: Medicine should be participatory, leading patients to be more responsible for their health and care.
• 5P - Proof: Medicine must be based on evidence of medical service to patients, especially when it relies on connected health and telemedicine.
• 6P - Pathway: Coordinating multiple interventions (medical, social, occupational medicine, etc.) such that the healthcare pathway is progressively articulated, according to the pathology and its evolution.

• Data acquisition
• Corpus of face-to-face interviews
• Social network posts
• Clinical data (discrete data)
• Data annotation
• Manual annotation
• Standards for annotation
• Document level
• Phrase level
• Data security
• Anonymization
• Data integration
• Parallel between acquisition and clinical information system

• Visual indicators:
• head pose, fatigue, mutual eye gaze, emotions, smiles, self-touches, body stiffness, etc.
• Speech indicators:
• voice tone, frequency, amplitude, etc.
• Text indicators:
• emotions, temporality, discourse coherence, abstractness, concreteness, topicality, idea intensity, rumination, misspelling, etc.
• Behavioral indicators:
• keystroke dynamics
• Group behavior
• Communicative issues, self-containment, (either within physical groups or virtual groups such as social networks), etc.
• Skin indicator
• Sensibility of skin via specific low noise captors

• Multimodal fusion:
• Early, late, hybrid, cascade, multi-layer fusion techniques to combine digital phenotypes, etc.
• Multitask learning:
• Taking advantage of different pathologies that share some common symptoms (e.g. depression and bipolarity) with shared-private architectures, etc.
• Multiclass Multitask models to specify if some patient has depression, anxiety, schizophrenia or something else by taking advantage of relation between pathologies. This can be important as diagnosis is usually difficult between these pathologies.
• Multilingual learning
• Taking advantage of different languages to strengthen commonalities of the disorders, learning the specificities of each country, tackling the lack of transcripts in one language to propose robust models, etc.
• Demographic-aware learning
• Gender-aware models, region-aware models (different cultures may have different ways of expressing thoughts), age-aware models, etc.
• Combined feature-based (from PART I) and deep learning architectures to reach explainable models (Hybrid AI).

• Social Networks:
• Early prediction of mental instability of individuals (temporal models, lack of positive activities),
• Early prediction by high comorbidity detection (i.e. drug abuse)
• Virtual robots sending posts to warn and inform about dangers, etc.
• Embodied Conversational Agent
• Virtual engagement of younger populations to pass through psychological tests, trying to reach therapeutic alliance in the ECA, etc.
• Symptom-based patient monitoring
• Monitoring users facing a major traumatic event, who are likely to develop a mental disorder.

• Mobile applications
  • Monitor adjuvant therapies (e.g. meditation, physical activity)
  • Automatic recommendation system based on data gathering and mining (e.g. proposal of personal contacts, of specific meditations or physical activities)
  • Open data linking to access contextual information
  • Embodied Conversational Agent
  • Safe Social Network
• Virtual reality applications
• Re-education of brain plasticity (e.g. phobias, etc.)

• Access to the Web with cognitive impairment
• Text simplification
• Text/image augmentation
• Text abstractness vs. concreteness
• Tiredness monitoring
• Mobile application helping in managing tiredness

• Sociological issues
• How much society is prepared for this digital transformation (e.g. covid-19 situation of containment changes or not mentality?)
• Legal issues
• Telemedicine raises issues on self-agreement, etc.
• Public health
• Mental health observatory.

• Human Computer Interfaces
  • User-centered development, prototyping, and users’ needs
    • Monitoring platforms
    • Embodied conversational agent
    • Chatbots on Social Network
    • Virtual Reality scenarios
    • Social Network
• Data security issues
  • How to preserve data anonymity, ensure data security etc.
• Project platform
  • Development of the software architecture, module integration, etc.
• Implementation in vivo
  • Deployment in real-world scenarios.