NIMHANS, IIT Kharagpur launch HEADS: AI tool to screen depression in five Indian languages

Three Indian institutions will jointly launch a two-year research project on Thursday aimed at using artificial intelligence to help detect depression earlier, with a strong focus on Indian languages that have largely been left out of mental health technology.

The project, titled HEADS (Human-in-the-loop Evaluation of Assisted Depression Screening), brings together the National Institute of Mental Health and Neurosciences (NIMHANS), Bengaluru, the Indian Institute of Technology (IIT) Kharagpur, and the Lokopriya Gopinath Bordoloi Regional Institute of Mental Health (LGBRIMH), Tezpur. It is funded by the Wellcome Trust, an international charitable foundation, and will run for two years.

The study will build and evaluate AI-assisted methods for the early identification and screening of depression. It will work in five languages — Kannada, Assamese, Hindi, Bengali and English — chosen because they have historically been substantially underrepresented in mental health AI research.

Lekhansh Shukla, assistant professor of psychiatry at NIMHANS and the project's principal investigator, said India carries one of the world's largest depression treatment gaps, with only about one in five people affected by depression receiving timely and appropriate care.

Speaking to The Hindu, Dr. Shukla said most existing mental health AI tools are built on English-language, urban datasets. As a result, they perform poorly in diverse, non-English clinical settings. "In our earlier research, we found that even frontier models cannot understand or translate Indian languages, especially when the discussion was about feelings and emotions," he said. "HEADS instead builds for Indian languages from day one, developing artificial intelligence that understands clinical conversations in Kannada, Hindi, Bengali, Assamese and English."

At the project's hybrid launch event at NIMHANS, researchers are expected to highlight this limitation of current digital health innovations as a central reason for the initiative.

A key design principle of HEADS is what the team calls a strict "human-in-the-loop" framework. Under this approach, AI tools are meant to serve only as supportive aids to clinical decision-making — helping doctors and other clinicians notice symptoms earlier — and are not intended to replace professional clinical judgment at any stage.

In what the researchers describe as a first for Indian mental health research, the study has embedded a 15-member panel of Lived Experience Experts — people who have themselves lived through depression, anxiety or substance use disorders — across every phase of the project. The panel has already helped shape the study's design and its informed consent protocols. Its members will also take part in language-evaluation groups for all five target languages, audit the AI algorithms for hidden bias and stigmatising language, and stress-test the systems before they are deployed.

Over its 24-month duration, HEADS will record and analyse approximately 4,500 clinical interviews at healthcare sites run by NIMHANS and LGBRIMH. IIT Kharagpur will lead the core engineering, system development and AI-safety work, with responsibility for ensuring that any deployment is carried out responsibly.

Senior leadership, investigators and project team members from all three partner institutions attended the hybrid launch.

The project's significance lies in its twin focus: language and oversight. Depression is among the leading causes of disability worldwide, and India's treatment gap means that many people who need care are never identified. If AI tools are to help close that gap, researchers argue, they must understand how people actually describe distress — in their own languages and in their own clinical settings — and they must remain subject to human oversight rather than substituting for it. HEADS is an attempt to test that proposition systematically, with people who have lived through mental illness placed at the centre of the evaluation rather than at its margins.