Personalized Non-Invasive Hemoglobin Estimation from Conjunctival Images

This dataset contains a collection of palpebral conjunctiva images gathered for the development and validation of a non-invasive system for personalized hemoglobin (Hb) estimation. The images were acquired from patients in the hematology ward of the \Istituto Tumori Giovanni Paolo II\ hospital in Bari, Italy. Each image is associated with a corresponding hemoglobin value, measured via a venous blood draw conducted during the same clinical session. The dataset was used to train a Vision Transformer (ViT) based model with a dual fine-tuning approach, which demonstrated high accuracy in personalized Hb level estimation and anemia classification. The goal of this dataset is to foster research in the field of non-invasive diagnostics and precision medicine, offering a unique resource for the development of new hematological monitoring techniques.