**4. Conclusions**

We developed a fully automated segmentation procedure, based on graph partitioning, that exposes conjunctival regions while maximizing the correlation between color properties and hemoglobin concentration in the blood, according to the multi-layered anatomical structures of these tissues. The ROIs extracted by the model underwent an in-depth quantitative comparison with ground truth, using state of the art metrics for similarity and PCC between the a\* component from CIELAB space and hemoglobin values. The results attest to the reliability and the capability of generalizing between patients belonging to heterogeneous classes, as the accuracy of the overlap between the manual and automatic ROIs selections, measured with classic metrics, is very good, and the correlation obtained between the level of Hb measured in vivo and that estimated through the color of the manual/automatic ROI are comparable. The proposed method paves the way for further studies involving deep learning techniques for both classifications of an estimated anemia risk category and regression to predict Hb real values. With this study we contribute to the broader diagnostic research field of image processing and analysis of the conjunctival pallor related to anemia diagnosis support. The advancement provided to this non-invasive image capturing procedure will lead to the possibility of embedding the model in a wearable device screening Hb risk category in real-time, without the need for physician support.

**Author Contributions:** The authors contributed equally to this work. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.
