A team of researchers from Purdue University and the Rwanda Biomedical Center has developed a novel, noninvasive method to detect anemia in school-age children using standard smartphone photos of the eye. The technique leverages radiomics – an emerging machine learning approach – to extract spatial and textural features from grayscale images of the conjunctiva, bypassing the need for traditional color-based or spectral imaging.
The study, which involved over 565 children aged 5 to 15 in rural Rwanda, used more than 24,000 images captured with three types of Android smartphones (Google Pixel 6, Samsung Galaxy S22, and Samsung Galaxy A52). Radiomic analysis of these photos revealed strong correlations between microvascular patterns in the conjunctiva and blood hemoglobin levels, the clinical standard for diagnosing anemia.
By converting images to grayscale and analyzing features such as pixel intensity, texture, and wavelet patterns, the researchers trained neural network models that could classify anemia status with a strong level of diagnostic performance. Notably, the models performed consistently across different smartphone brands, lighting conditions, and even between the left and right eyes, making the method practical for real-world mobile health (mHealth) deployment.
Interestingly, the study found higher prediction accuracy in females than in males, which the authors note could be due to anatomical differences in the conjunctiva, such as: greater endothelial reactivity, which enhances microvascular contrast; higher capillary density so that microvascular patterns are made more distinct; and smaller vessel diameters, which allow for subtle anatomical variations to be detected more easily.
The novel approach offers two major advantages over existing tools: it avoids the complexities of color calibration and does not rely on invasive blood draws or expensive point-of-care devices. Once implemented as a mobile app, it could empower community health workers and educators in resource-limited settings to screen children for anemia rapidly and affordably, enabling earlier treatment and better health outcomes.
The team behind the study has filed a provisional patent application through the Purdue Research Foundation, and is developing a user-friendly mHealth platform to translate their findings into scalable public health impact.
