NIMHANS Launches AI Project for Depression Screening in Indian Languages

The National Institute of Mental Health and Neuro Sciences (NIMHANS) has launched a two-year research project with IIT Kharagpur and the Lokopriya Gopinath Bordoloi Regional Institute of Mental Health (LGBRIMH), Tezpur, to develop an AI-assisted system that could help clinicians screen for depression in multiple Indian languages. The Human-in-the-loop Evaluation of Assisted Depression Screening (HEADS) project, funded by the Wellcome Charitable Foundation, will address challenges in understanding psychological distress as patients describe it in their own languages and cultural contexts.

The proposed open-source, modular system will listen to clinical interviews, convert speech into text, translate conversations where required and generate a clinical summary. It will also provide a depression assessment, severity estimate, reasoning and confidence score. The research framework will incorporate DSM-5-based assessment and PHQ-9 severity bands. Importantly, clinicians will review and correct AI-generated assessments and retain final decision-making authority.

Over the 24-month project, psychiatrists will interview 4,500 consenting participants, comprising 4,000 patients and 500 healthy volunteers, in Kannada, Hindi, Bengali, Assamese and English. NIMHANS will collect data in Kannada, Hindi and English, while LGBRIMH will collect data in Assamese, Bengali and English. IIT Kharagpur will lead AI engineering, model development and testing, including assessments of bias and safety. Two MoUs were signed between the institutions as part of the launch.

The project addresses significant limitations in existing AI systems. Preliminary testing found that commercially available speech-recognition systems, language models and privacy tools fell below requirements for clinical use in Indian languages. Some frontier models recorded a 35% word-error rate in Kannada clinical interviews, while privacy testing identified only 64% of personal identifiers for removal. Researchers have also found that Indian-language expressions of mood and distress can be difficult for current models to interpret accurately.

The challenge is significant in a country where the National Mental Health Survey 2015–16 reported treatment gaps of 70% to 92% across mental disorders. WHO estimates that nearly 15% of Indian adults require active mental-health intervention, while about one in 20 Indians experiences depression.

HEADS will generate de-identified multilingual clinical datasets, transcripts, translations and clinical assessments, while training researchers and engineers. The project will also examine performance across language, gender, education, comorbidities and cultural contexts.

NIMHANS Director Y C Janardhan Reddy said machine-assisted support for depression and anxiety must be scientifically studied. Principal Investigator Lekhansh Shukla said the goal is earlier recognition while keeping clinicians firmly in control. The project is scheduled to run from September 2026 to August 2028, with its eventual value depending on whether the technology can demonstrate accuracy, safety, privacy and cultural sensitivity across India's diverse linguistic landscape.