Wellcome funds Imperial research to improve mental health screening in Nigeria using culturally grounded AI

The GENSCORE project will develop AI-powered tools designed to better identify depression and anxiety in speakers of Hausa, Yoruba and Nigerian Pidgin by adapting mental health assessments to local languages and cultural contexts.

Dr Shamsuddeen Muhammad, Google DeepMind Academic Fellow in Imperial's Department of Computing, is part of a Wellcome-funded international collaboration that aims to improve mental health screening in Nigeria through culturally grounded artificial intelligence. The two-year GENSCORE project brings together researchers and partners from Slum and Rural Health Initiative, Federal Neuropsychiatric Hospital Yaba Lagos, University of Ibadan and Imperial College London to develop and validate new approaches to assessing depression and anxiety across Nigeria's major linguistic groups.  

Commenting on the project, Dr Muhammad said: "Depression and anxiety are universal, but the words people use to describe them are not. A screening tool that only understands English, and only Western ways of expressing distress, will miss much of what a person in Nigeria is trying to say. In this project, alongside psychiatrists and people with lived experience, we will build models that understand distress as it is expressed in three major Nigerian languages: Hausa, Yoruba and Nigerian Pidgin. The aim is to give frontline health workers a tool that recognises distress in the words their patients actually use, in a country where fewer than 250 psychiatrists serve more than 220 million people. What we learn will help shape culturally grounded mental health tools for other low-resource settings as well."

Mental health conditions remain significantly underdiagnosed in many low-resource settings. According to the project team, fewer than 250 psychiatrists serve Nigeria's population of more than 220 million people, while up to 90 per cent of people experiencing mental health conditions receive no care. 

The aim is to give frontline health workers a tool that recognises distress in the words their patients actually use, in a country where fewer than 250 psychiatrists serve more than 220 million people." Google DeepMind Academic Fellow Dr Shamsuddeen Muhammad

One challenge is that widely used screening tools, including the PHQ-9 and GAD-7 assessments, were developed in Western, English-speaking contexts and may not fully capture how depression and anxiety are expressed across different languages and cultures. In Nigeria, emotional distress is often communicated through culturally specific idioms, religious expressions and physical symptoms that may not be reflected in standard assessment frameworks.  

The GENSCORE project aims to address this gap by developing AI systems that can better recognise culturally specific expressions of depression and anxiety in Hausa, Yoruba and Nigerian Pidgin. Building on previous pilot work, the researchers will create a large multilingual dataset of mental health narratives and voice samples, which will be used to train and evaluate new AI models designed specifically for Nigerian mental health contexts.  

A key part of the project is the development of GENSCORE-LLM, an open-source large language model designed to understand locally relevant expressions of distress. The team will also develop culturally adapted screening tools, known as GenPHQ and GenGAD, in collaboration with psychiatrists, community members and people with lived experience of mental health conditions.  

Unlike conventional approaches that rely solely on questionnaire responses, the project will explore a multimodal framework that combines text responses with voice-based features such as speech patterns and other vocal indicators. The researchers aim to create a more contextually grounded approach to mental health assessment that reflects how people communicate in their everyday lives.  

The tools will undergo rigorous validation with 3,600 participants across Hausa, Yoruba and Nigerian Pidgin-speaking populations, with performance benchmarked against established clinical assessments.  

By combining advances in artificial intelligence with insights from local languages, cultures and communities, the researchers hope the project will contribute to more accurate and culturally relevant approaches to mental health screening. The work could help inform future digital mental health tools in Nigeria and other multilingual, low-resource settings where access to specialist care remains limited.