Detection of Iron Deficiency Anemia Using Convolutional Neural Networks Based on Conjunctival Pallor Images

Le résumé fourni par la source

Anemia is a major public health problem, especially in children, which can be diagnosed using invasive blood examinations that are expensive and not feasible in low-resource environments. The present study introduces a non-invasive anemia prediction system via conjunctival images that is safer and more convenient. We employed a collection of 2635 conjunctival images drawn from Kaggle, including 1520 anemic and 1115 non-anemic samples, to create a reliable detection model. Our approach combines image pre-processing with a Convolution Neural Network (CNN) to obtain optimal accuracy. First, we used a proprietary masking strategy to demarcate the conjunctival area, eliminating noise from adjacent structures such as eyelids and skin. The images were then converted to CIE Lab* color space to emphasize redness features, an important sign of anemia, prior to passing them through the CNN for classification. The model was trained and tested through 10-fold cross-validation to provide assurance of reliable performance. Through emphasis on the conjunctival pallor, our system accurately discriminates between anemic and non-anemic conditions, offering an economic and non-invasive tool. This invention has the potential to enhance the detection of early anemia, particularly among underprivileged communities, to increase healthcare accessibility and quality for children worldwide.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.