NIMHANS, IIT Kharagpur and LGBRIMH Launch AI Project for Depression Screening in Indian Languages
NIMHANS, IIT Kharagpur and LGBRIMH have launched a two-year research project to develop and evaluate an artificial intelligence-assisted system for depression screening in Indian languages. The initiative, called HEADS — Human-in-the-loop Evaluation of Assisted Depression Screening, formally launched on September 24, 2026.
The project brings together the National Institute of Mental Health and Neuro Sciences (NIMHANS), Bengaluru, the Indian Institute of Technology Kharagpur and the Lokopriya Gopinath Bordoloi Regional Institute of Mental Health (LGBRIMH), Tezpur.
HEADS will work with clinical interviews in Kannada, Hindi, Bengali, Assamese and English. The proposed open-source system is intended to listen to clinical conversations, transcribe and translate them, and provide clinicians with an AI-assisted summary, depression assessment, severity estimate, reasoning and confidence information.
Importantly, the project is designed around a human-in-the-loop model. The AI is being developed as an aid to clinicians rather than an autonomous replacement for psychiatric assessment. Clinicians will review and correct AI-generated assessments and retain responsibility for clinical decisions.
HEADS project:
https://heads-ai.com/
NIMHANS:
https://www.nimhans.ac.in/
IIT Kharagpur:
https://www.iitkgp.ac.in/
HEADS AI Depression Screening Project: Key Highlights
| Particular | Details |
|---|---|
| Project | Human-in-the-loop Evaluation of Assisted Depression Screening |
| Short Name | HEADS |
| Participating Institutions | NIMHANS, IIT Kharagpur and LGBRIMH |
| Project Duration | 24 months |
| Project Period | September 2026 to August 2028 |
| Research Area | AI-assisted depression screening |
| Languages | Kannada, Hindi, Bengali, Assamese and English |
| Planned Participants | 4,500 consenting participants |
| Patients | 4,000 |
| Healthy Volunteers | 500 |
| AI Development | IIT Kharagpur |
| Clinical Data Collection | NIMHANS and LGBRIMH |
| Approach | Human-in-the-loop AI with clinician oversight |
| Planned Output | Open and modular AI-assisted depression screening system |
What Is the HEADS Project?
HEADS stands for Human-in-the-loop Evaluation of Assisted Depression Screening.
The research initiative aims to develop and scientifically evaluate AI-assisted approaches that could help clinicians identify depression earlier, particularly when patients describe psychological distress in languages and culturally specific expressions that existing AI systems may not handle adequately.
The system is not being positioned as an AI psychiatrist or an independent diagnostic authority. Instead, researchers are studying whether artificial intelligence can support trained professionals during depression screening while keeping human clinical judgment at the center of the process.
Which Languages Will the AI Depression Tool Support?
The project will work across five languages:
- Kannada
- Hindi
- Bengali
- Assamese
- English
The multilingual component is one of the central features of HEADS because many existing speech-recognition and language-model systems perform much better in English than in several Indian languages, particularly in specialized clinical conversations.
Why Indian Languages Matter in Mental Health Screening
People do not always describe depression using formal psychiatric terminology. A patient may talk about tiredness, sleep, physical discomfort, loss of interest, concentration problems, hopelessness or changes in daily functioning rather than explicitly saying, “I have depression.”
Language and cultural context can affect how those experiences are described.
An AI system trained primarily on English-language material may struggle with:
- Regional expressions
- Local idioms
- Code-switching between languages
- Culturally specific descriptions of distress
- Conversational speech
- Accent and pronunciation differences
- Indirect descriptions of emotional problems
HEADS is intended to investigate these challenges in an Indian clinical context.
How Will the HEADS AI System Work?
The planned system will process clinical interviews through several connected AI components.
The research architecture is expected to involve functions such as:
- Listening to a clinical conversation.
- Converting speech into text.
- Processing conversations in supported Indian languages.
- Translating information where required.
- Analyzing clinically relevant language patterns.
- Preparing a structured summary.
- Generating an assisted depression assessment.
- Estimating depression severity.
- Providing reasoning and confidence information.
- Presenting the output to a clinician for review.
AI Will Not Make the Final Clinical Decision
The most important feature of the HEADS design is contained in its name: human-in-the-loop.
The clinician remains responsible for reviewing the AI output.
| AI Can Assist With | Clinician Remains Responsible For |
|---|---|
| Speech transcription | Clinical interpretation |
| Language translation | Reviewing context |
| Conversation summarization | Correcting AI errors |
| Screening assistance | Clinical judgment |
| Severity estimation | Final assessment and care decisions |
What Does Human-in-the-Loop AI Mean?
Human-in-the-loop AI is an approach in which an artificial intelligence system produces information or recommendations but a human expert remains actively involved in reviewing, correcting and interpreting those outputs.
In mental healthcare, this distinction is especially important because an AI-generated prediction cannot capture every clinical, personal and social factor relevant to an individual patient.
How Many People Will Participate in the HEADS Study?
The project plans clinical interviews with 4,500 consenting participants.
| Participant Group | Planned Number |
|---|---|
| Patients | 4,000 |
| Healthy Volunteers | 500 |
| Total | 4,500 |
The resulting research will be used to develop and evaluate the AI-assisted system.
What Will NIMHANS Do in the HEADS Project?
NIMHANS will provide major clinical and mental-health research expertise to the project.
Under the collaboration, NIMHANS will undertake clinical data collection involving:
- Kannada
- Hindi
- English
Its clinicians and researchers will also contribute to evaluating whether AI-generated information is clinically meaningful and safe enough to support further research and potential future use.
What Will LGBRIMH Do?
LGBRIMH in Tezpur will contribute psychiatric expertise and regional linguistic and cultural context.
Its clinical data collection will include:
- Assamese
- Bengali
- English
This is particularly important for expanding the study beyond a single region or language environment.
What Will IIT Kharagpur Do?
IIT Kharagpur will lead the AI engineering component.
Its responsibilities include developing and testing the artificial intelligence models required for the project.
The technical research will also examine issues such as:
- Model performance
- Speech recognition
- Translation
- Clinical-language processing
- Bias
- Safety
- Reliability
Role of Each Institution
| Institution | Primary Contribution |
|---|---|
| NIMHANS | Clinical expertise, psychiatric research and clinical data collection |
| IIT Kharagpur | AI engineering, model development, testing, bias and safety evaluation |
| LGBRIMH | Clinical psychiatry, regional context and multilingual clinical data collection |
Why Is Depression Screening Difficult?
Depression does not present identically in every person.
Symptoms can involve emotional, cognitive, behavioral and physical changes. Some people may primarily describe sadness or loss of interest, while others may focus on sleep, fatigue, concentration or physical complaints.
This means depression screening requires context rather than simple keyword detection.
Depression Screening Is Not the Same as Diagnosis
This distinction is essential when discussing the HEADS project.
| Screening | Clinical Diagnosis |
|---|---|
| Identifies whether further assessment may be warranted | Requires professional clinical evaluation |
| Can use structured questions or tools | Considers broader clinical information |
| Can support early identification | Determines the clinical diagnosis |
| Does not independently establish a diagnosis | Requires qualified professional judgment |
The HEADS system is being researched as an assisted screening tool, not as a replacement for professional diagnosis.
Why Could AI Help With Depression Screening?
AI systems can potentially process large amounts of conversational information and identify patterns that may help clinicians review an interview more systematically.
Possible supportive functions include:
- Transcribing long clinical conversations
- Producing structured summaries
- Highlighting potentially relevant statements
- Supporting multilingual communication
- Estimating symptom severity for clinician review
- Helping standardize parts of the screening workflow
Whether these functions are sufficiently accurate and safe in real clinical settings is precisely the kind of question the research project is designed to investigate.
Why Existing AI Models May Not Be Enough
Pilot work associated with the project found that readily available speech-recognition systems, language models and privacy-related tools did not meet the required performance level for clinical use in the targeted Indian-language context.
This highlights a major difference between a general-purpose AI application and a clinically evaluated system.
General AI vs Clinical Mental Health AI
| General AI | Clinical Mental Health AI |
|---|---|
| Can tolerate some conversational errors | Errors may have clinical consequences |
| Often optimized for broad tasks | Must be evaluated for specific clinical tasks |
| May lack medical validation | Requires rigorous evaluation |
| May perform unevenly across languages | Needs language-specific testing |
| May generate confident but incorrect information | Requires safeguards and professional oversight |
Why AI Bias Is a Major Research Question
An AI system can perform differently across languages, demographic groups, accents and communication styles.
For example, a system trained predominantly on one language or population may be less accurate when used with another.
HEADS therefore includes evaluation of potential bias and safety concerns rather than focusing only on headline accuracy.
Types of Bias Researchers May Need to Examine
- Language bias
- Accent-related performance differences
- Gender-related differences
- Regional linguistic variation
- Age-related differences
- Cultural expression of distress
- Differences between clinical and non-clinical speech
Why Confidence Scores Matter
The proposed system is expected to provide confidence information alongside its assessment.
This can help researchers and clinicians examine whether the AI recognizes situations where its output may be uncertain.
However, a high AI confidence score does not automatically mean that an assessment is clinically correct. Confidence itself must be evaluated scientifically.
Why AI Reasoning Must Be Reviewed
The planned system will provide information supporting its assessment rather than presenting only a result.
This could allow clinicians to inspect whether the AI focused on relevant parts of the conversation.
For example, researchers can examine whether a model:
- Misinterpreted an idiom
- Missed contextual information
- Overweighted a particular statement
- Incorrectly translated a phrase
- Ignored contradictory information
Why Translation Is Difficult in Mental Healthcare
Literal translation can sometimes change the meaning of emotional expressions.
A phrase describing distress in Assamese, Bengali, Kannada or Hindi may not map neatly onto an English psychiatric phrase.
A clinically useful system therefore needs to preserve meaning rather than simply replace words from one language with another.
Indian Idioms and Mental Health AI
One of the research challenges is understanding how psychological distress is communicated through culturally meaningful expressions.
A person may describe emotional suffering indirectly through:
- Physical sensations
- Sleep complaints
- Loss of energy
- Social withdrawal
- Changes in appetite
- Difficulty concentrating
- Loss of motivation
- Family or work-related functioning
An AI system designed for clinical support needs to interpret these statements within the broader interview rather than search for a single depression-related keyword.
Will HEADS Record Doctor-Patient Conversations?
The research involves clinical interviews with consenting participants. Handling sensitive clinical conversations requires appropriate research consent, privacy protections and data-governance procedures.
The project is also expected to produce de-identified research datasets rather than simply releasing identifiable patient conversations.
What Does De-Identified Data Mean?
De-identification involves removing or transforming information that could directly identify a research participant.
Examples of potentially identifying information can include:
- Name
- Phone number
- Address
- Direct personal identifiers
- Other information that could reveal identity
Clinical data still requires careful governance even after de-identification because health information is sensitive.
Privacy Challenges in Mental Health AI
AI-assisted mental healthcare creates important privacy questions because conversations can contain extremely personal information.
Researchers must consider issues including:
- Informed consent
- Secure storage
- Access control
- De-identification
- Data retention
- Model training
- Secondary research use
- Protection against unauthorized disclosure
What Will HEADS Produce?
The project is expected to work toward several research outputs.
- An open and modular AI-assisted depression screening system
- AI components for speech and language processing
- Clinical evaluation of AI-assisted screening
- De-identified datasets involving Indian-language clinical interviews
- Research on bias and safety
- Training of researchers and engineers across participating institutions
What Does Open-Source Mean for the HEADS Project?
The project intends to develop an open system that can support further research and evaluation.
Open research infrastructure can allow researchers to inspect, test and improve components rather than depending entirely on proprietary black-box technology.
However, open-source availability does not automatically mean that a tool is approved for unrestricted clinical use. Clinical deployment requires separate evidence, governance and applicable regulatory considerations.
Can HEADS Be Used by Doctors Now?
The project has entered its research and development phase. It should not be treated as an already validated, generally deployed clinical product.
The 24-month study is intended to develop and evaluate the technology.
When Will the HEADS Project Be Completed?
The two-year research period runs from September 2026 to August 2028.
Research findings and system development will emerge during that period according to the project's progress.
Could Non-Specialist Health Workers Use the Tool?
One long-term objective of the research is to investigate whether AI-assisted screening could support earlier recognition of depression, including in settings involving non-specialist healthcare workers.
That does not mean non-specialists would independently diagnose psychiatric disorders using AI.
The value being investigated is whether technology can help identify people who may need appropriate clinical assessment.
Could AI Improve Mental Healthcare Access in India?
Multilingual technology could potentially help reduce certain language and workflow barriers, particularly where specialist resources are limited.
Potential benefits being explored include:
- Earlier identification
- Multilingual screening
- Clinical documentation support
- Structured interview summaries
- Support for non-specialist settings
Technology alone, however, cannot solve shortages of trained professionals, treatment access, affordability, stigma or broader health-system constraints.
Can AI Replace Psychiatrists?
No such claim is being made by the HEADS project.
The project's design specifically keeps clinicians involved in reviewing and correcting the AI's assessment.
Psychiatric evaluation can involve:
- Clinical history
- Current symptoms
- Medical conditions
- Medication
- Substance use
- Social circumstances
- Functional impairment
- Risk assessment
- Professional judgment
An AI-generated conversation analysis does not independently replace these responsibilities.
Can Chatbots Diagnose Depression?
General-purpose chatbots should not be treated as substitutes for qualified mental-health professionals or validated clinical diagnostic processes.
A chatbot producing a depression-related response is fundamentally different from a system undergoing structured clinical research with psychiatrist oversight.
AI Screening vs AI Diagnosis
| AI-Assisted Screening | Autonomous AI Diagnosis |
|---|---|
| Supports identification of possible concerns | Would independently assign a diagnosis |
| Can support clinician workflow | Would remove or substantially reduce professional oversight |
| HEADS is studying this assisted approach | HEADS is not being presented as this |
Why Mental Health AI Needs Safety Testing
False results can have serious consequences.
A false negative could fail to identify a person who needs further assessment.
A false positive could incorrectly flag someone as potentially having depression.
Researchers therefore need to evaluate:
- Sensitivity
- Specificity
- Error patterns
- Bias
- Language performance
- Clinical usefulness
- Human-AI interaction
Why Clinical Validation Matters
A model performing well on a technical benchmark does not automatically mean that it is useful in a hospital or clinic.
Clinical validation asks whether the system performs appropriately with real patients, real conversations and real-world linguistic variation.
Why Lived Experience Matters in Mental Health AI
The broader HEADS research approach has also incorporated perspectives from people with lived experience of mental-health challenges.
This can help researchers consider questions that purely technical development may overlook, including:
- How patients experience AI-assisted interviews
- What information people consider sensitive
- How explanations should be presented
- Whether the system feels stigmatizing
- What types of errors patients consider harmful
- How consent should be communicated
AI and Mental Health Research Opportunities for Students
The HEADS collaboration also illustrates the growing intersection of healthcare and artificial intelligence.
Students interested in this area can build expertise across disciplines such as:
- Artificial intelligence
- Machine learning
- Natural language processing
- Automatic speech recognition
- Computational linguistics
- Clinical psychology
- Psychiatry
- Biomedical engineering
- Health informatics
- Responsible AI
- AI safety
- Data privacy
Skills Useful for AI in Healthcare Careers
| Technical Skills | Healthcare and Research Skills |
|---|---|
| Python | Research methodology |
| Machine learning | Clinical terminology |
| Natural language processing | Research ethics |
| Speech recognition | Data governance |
| Deep learning | Human-subject research principles |
| Model evaluation | Interdisciplinary communication |
Can Engineering Students Work in Mental Health AI?
Yes. Mental-health AI research requires expertise extending well beyond psychiatry.
Engineering and computer-science students can contribute to:
- Speech models
- Language models
- Multilingual NLP
- Machine-learning evaluation
- Privacy technology
- Explainable AI
- Bias testing
- Software engineering
Clinical collaboration remains essential because technical performance must be interpreted in the context of healthcare.
Can Psychology and Medical Students Work in AI Research?
Yes. Healthcare AI is inherently interdisciplinary.
Psychology, psychiatry and medical researchers can contribute to:
- Clinical study design
- Interview protocols
- Symptom assessment
- Clinical validation
- Patient safety
- Ethics
- Interpretation of model errors
Why IIT Kharagpur's Role Is Important
Building an AI system for multilingual clinical conversations involves several technically difficult tasks simultaneously.
The engineering team needs to work on areas such as:
- Speech-to-text conversion
- Multilingual language processing
- Translation
- Clinical summarization
- Prediction models
- Confidence estimation
- Bias analysis
- Safety testing
Why NIMHANS' Role Is Important
NIMHANS brings specialized expertise in psychiatry, mental health and neurosciences.
Clinical experts are needed to determine whether an AI output is not merely linguistically plausible but clinically meaningful.
Why LGBRIMH's Role Is Important
LGBRIMH expands the collaboration's clinical and linguistic reach, particularly through its work in Assam and the Northeast.
Its participation supports research involving Assamese and Bengali alongside English and provides an additional clinical setting for evaluating the system.
Why Collaboration Between Medicine and Engineering Matters
Healthcare AI cannot be developed responsibly through software engineering alone.
| Clinical Team | Engineering Team |
|---|---|
| Defines clinically meaningful problems | Builds technical models |
| Evaluates patient context | Tests model performance |
| Reviews clinical errors | Analyzes technical errors |
| Protects clinical decision-making | Improves system reliability |
HEADS Project FAQs
What is the HEADS project?
HEADS is the Human-in-the-loop Evaluation of Assisted Depression Screening project, a collaboration developing and evaluating AI-assisted depression screening in multiple Indian languages.
Which institutions are involved in HEADS?
NIMHANS Bengaluru, IIT Kharagpur and LGBRIMH Tezpur.
When was the HEADS project formally launched?
The project was formally launched on September 24, 2026.
How long will the project run?
The research project is planned for 24 months, from September 2026 to August 2028.
What is HEADS full form?
Human-in-the-loop Evaluation of Assisted Depression Screening.
What is the purpose of HEADS?
It aims to develop and evaluate AI-assisted approaches that could help clinicians identify and screen for depression earlier.
Which languages will HEADS support?
Kannada, Hindi, Bengali, Assamese and English.
How many people will participate in the study?
The project plans interviews with 4,500 consenting participants.
How many patients will participate?
The planned research includes 4,000 patients.
How many healthy volunteers will participate?
The project plans to include 500 healthy volunteers.
Will AI diagnose depression independently?
No. The project uses a human-in-the-loop model in which clinicians review AI-generated assessments and retain clinical decision-making authority.
Will the AI replace psychiatrists?
No. It is being developed as an assistive screening system rather than a replacement for psychiatrists.
What will the AI do?
The planned system can process clinical interviews, transcribe and translate conversations, summarize information and generate an assisted depression assessment and severity estimate for clinician review.
Will the AI provide a confidence score?
The proposed system includes confidence information alongside its assessment.
Who is developing the AI models?
IIT Kharagpur is leading AI engineering and model development.
What is NIMHANS' role?
NIMHANS contributes clinical expertise and data collection in Kannada, Hindi and English.
What is LGBRIMH's role?
LGBRIMH contributes clinical expertise and data collection in Assamese, Bengali and English.
Will the software be open source?
The project intends to produce an open and modular AI-assisted screening system.
Does open source mean anyone can use it clinically?
No. Open-source availability and clinical validation or authorization are separate issues.
Is the HEADS AI tool already available in hospitals?
The initiative is currently a research and development project and should not be treated as an already validated general-purpose clinical product.
Why are Indian languages important?
People often describe emotional distress through culturally and linguistically specific expressions that general-purpose English-focused AI systems may not capture accurately.
Why is speech recognition important?
The system needs to accurately convert clinical conversations into text before later language-processing components can analyze them.
Why is translation important?
Translation can help process multilingual clinical information, but it must preserve the meaning of culturally specific expressions.
What is human-in-the-loop AI?
It is an approach in which AI assists with a task while a human expert remains actively involved in reviewing and correcting its output.
Is depression screening the same as diagnosis?
No. Screening can identify people who may need further evaluation, while diagnosis requires appropriate clinical assessment.
Can ordinary AI chatbots diagnose depression?
General-purpose chatbots should not be treated as substitutes for qualified mental-health professionals or validated clinical diagnostic processes.
Why must mental-health AI be tested for bias?
Performance can vary across languages, accents, demographic groups and cultural communication patterns.
Will patient privacy be important?
Yes. Mental-health conversations contain sensitive health information, making consent, data security and appropriate governance essential.
Will the project create datasets?
The project is expected to produce de-identified research datasets involving clinical interviews in Indian languages.
Can AI help non-specialist healthcare workers?
The project aims to investigate whether assisted screening could support earlier recognition, including in settings involving non-specialist workers, while preserving appropriate clinical oversight.
Can engineering students build careers in mental-health AI?
Yes. Relevant areas include machine learning, NLP, speech recognition, responsible AI, privacy, healthcare technology and AI safety.
Can medical and psychology students work in AI research?
Yes. Clinical study design, validation, ethics, patient safety and interpretation require substantial healthcare expertise.
Where can I learn more about the HEADS project?
https://heads-ai.com/
Why the HEADS Project Matters
HEADS represents a shift from simply asking whether generative AI can discuss mental health to a more demanding research question: can an AI system assist real clinicians with depression screening across multiple Indian languages while remaining accurate, explainable, safe and under human supervision?
The answer cannot be assumed in advance. The two-year collaboration between NIMHANS, IIT Kharagpur and LGBRIMH is designed to generate evidence by combining psychiatric expertise, multilingual clinical data and AI engineering.
If the approach proves sufficiently reliable, it could provide a foundation for more inclusive multilingual mental-health technology. Until that evidence is established, however, the system should be understood as a research initiative rather than a replacement for professional mental-health assessment.
HEADS project:
https://heads-ai.com/
NIMHANS:
https://www.nimhans.ac.in/
IIT Kharagpur:
https://www.iitkgp.ac.in/
q.jpg)