Improving Conjunctiva Segmentation for Robust Anemia Detection Using Raspberry Pi's Camera and Macro Lens

Anemia is a global health concern with significant implications, particularly for vulnerable populations such as pregnant women and children. Due to its often asymptomatic nature, many individuals in developing countries remain undiagnosed. Conjunctival examination is frequently used as an initial screening method by physicians, given its closer proximity to the heart compared to the palms and nail beds. Therefore, this study aims to automatically segment the conjunctiva using an optimized UNBCSM (UNet Based Conjunctiva Segmentation) model. Image augmentation and model adjustments were implemented to enhance data volume and model performance. The image acquisition system is designed by utilizing a Raspberry Pi Camera Module v3, augmented with a macro lens and light-spacer attachment to facilitate the capture of high-resolution imagery. The results demonstrate that the inclusion of various augmentations and parameter optimization yielded satisfactory Intersection over Union (IoU) values of 96.6% and 92.1% for training and validation datasets, respectively.