IIT-AIIMS team uses AI to improve childhood malnutrition detection
IT and AIIMS Jodhpur researchers have developed DomainAdapt, an AI-powered framework to detect childhood malnutrition using photos. Coupled with the AnthroVision dataset, it provides fast, scalable, and accurate assessment of stunting, wasting, and underweight in children.
New Delhi: Researchers at the Indian Institute of Technology (IIT) and All India Institute of Medical Science (AIIMS) Jodhpur have leveraged the power of artificial intelligence (AI) to better identify childhood malnutrition.
The new method, published in the open-access journal MICCAI, addresses one of the most pressing global health challenges — the accurate and scalable assessment of childhood malnutrition.
Also Read
The study introduced DomainAdapt — a novel multitasks learning framework that dynamically adjusts task weights using domain knowledge and mutual information.
This allows the system to more accurately predict key anthropometric measures such as height, weight, and mid-upper arm circumference (MUAC), while simultaneously classifying malnutrition-related conditions such as stunting, wasting, and underweight.
While these measures are also assessed using the traditional screening methods, they pose challenges in terms of the subjectivity of the worker, the time-consuming process of measuring each aspect one by one, and the lack of scalability.
“By simply capturing photos of a child, our framework can estimate nutritional status without the need for complex and time-consuming anthropometric measurements,” explained Misaal Khan, a doctoral student in medical technology at IIT-AIIMS, who led the study.
“This makes malnutrition screening faster, more accessible, and highly scalable, especially in resource-limited settings,” Khan added.
Further, a cornerstone of the research is AnthroVision — a first-of-its-kind dataset containing 16,938 multi-pose images from 2,141 children collected across both clinical (AIIMS Jodhpur) and community (government schools in Rajasthan) settings.
The dataset captures diverse backgrounds, clothing, and lighting conditions, making it a robust resource for advancing automated child health assessment.
Through rigorous experimentation, DomainAdapt demonstrated significant improvements over existing multitask learning methods, offering a reliable AI-driven solution to accelerate malnutrition detection worldwide.
“This research represents a vital step toward equitable healthcare access,” said Khan.
“By blending AI and domain expertise, we can empower healthcare workers and public health systems with tools that are cost-effective, accurate, and scalable,” she added.
