Fig. 6.

Performances of the proposed system and the two human experts for the four-degree classification of the ROP severity. The proposed system demonstrated 0.903 accuracy, 0.778 sensitivity, 0.932 specificity and 0.761 F1-score for the four-degree classification task; Expert 1 achieved 0.902 accuracy, 0.748 sensitivity, 0.934 specificity and 0.743 F1-score, while expert 2 achieved 0.898 accuracy, 0.659 sensitivity, 0.923 specificity and 0.682 F1-score. Abbreviations: ROP, retinopathy of prematurity