Non-Invasive mHealth Application for Detecting Anemia from Conjunctiva Images

Anemia, which is defined by a lack of red blood cells or hemoglobin, is a widespread issue in global health that has far-reaching effects. An accurate and timely diagnosis of Anemia is essential, especially for pregnant women and people with chronic medical conditions as it can result in weakness, exhaustion, deteriorated cognitive function, and other severe problems. Traditional techniques of detecting Anemia often involve invasive blood testing, which can be expensive and inconvenient. This study proposes a non-invasive mHealth application that uses a machine learning and deep learning algorithms to identify Anemia from conjunctiva images. An annotated dataset containing eye conjunctiva images and ground truth masks were prepared for this study. The proposed system uses smartphone to capture image of a person's face with exposed conjunctiva and this image is segmented and region of interest that is eye conjunctiva is separated from the background. Then this segmented region is classified into Anemic and Non-anemic class by a machine learning algorithm. In this study UNet architecture gave 72.05% IOU in segmentation task and Random Forest classifier showed 91.43%±1.06% accuracy.