Smart phone and machine learning process helps detect anemia in children

24 Apr 2025

Purdue and Rwanda-based medical research group photograph developing conjunctiva in the field.

Radiomic features extracted via machine learning from smart phone photos.

Anemia, a condition marked by low levels of hemoglobin in the blood, affects nearly 2 billion people worldwide. Among them, school-age children in low- and middle-income countries are particularly vulnerable.

Left untreated, anemia in children can interfere with growth, learning, and overall development. Detecting the condition early is essential, but standard diagnostic methods require blood samples and lab equipment—resources that are often unavailable in low-income areas.

A new study reported in Biophotonics Discovery offers a promising alternative: using simple grayscale photos of the eye’s conjunctiva – the inner surface of the eyelid and the white part of the eye – to predict anemia.

Smart thinking

Researchers from Purdue University, West Lafayette, Indiana, Rwanda Biomedical Center, and the University of Rwanda used standard smart phones to take over 12,000 eye photos from 565 children aged 5 to 15.

They then applied machine learning along with a technique called radiomics, which mathematically analyzes patterns and textures in medical images, to identify features linked to anemia.

First author Shaun Hong, a Purdue University PhD student, commented, “Unlike previous efforts that rely on color analysis or special imaging tools, this method doesn’t require color data.

Instead, it uses black-and-white photos to examine tiny structural changes in the eye’s blood vessels. This approach avoids problems caused by different light conditions or camera models, making it more practical for use in a variety of settings.”

The results show a strong connection between specific spatial features and anemia status, pointing to the possibility of screening for anemia using just a smartphone and basic software. This could be especially useful in remote or under-resourced communities, offering a fast, noninvasive, and affordable way to identify children at risk.

Corresponding author Professor Young L. Kim of Purdue University, commented, “The technology isn’t meant to replace traditional testing but could help prioritize who needs further evaluation and treatment. With more development, the method could be integrated into mobile health tools to support early intervention in areas where healthcare access is limited.”

This article was first published on spie.org.