A Fully Automated Pipeline for a Robust Conjunctival Hyperemia Estimation

Purpose: Many semi-automated and fully-automated approaches have been proposed in literature to improve the objectivity of estimation conjunctival hyperemia, based on image processing analysis eyes’ photographs. The purpose is its evaluation using faster systems independent by human subjectivity. Methods: In this work, we introduce a redness grading scales able completely automatize clinical procedure, starting from acquired estimation. particular, neural network model for segmentation followed an pipeline vessels segmentation. From these steps, extract some features already known whose correlation with has proved. Lastly, implemented predictive hyperemia features. Results: used dataset images during practice.We trained segmentation, obtaining average accuracy 0.94 corresponding IoU score 0.88 test set images. extracted ROIs correctly predict Efron scale values Spearman’s coefficient 0.701 not previously samples. Conclusions: robustness our confirms possible usage practice as viable decision support system ophthalmologists.