Purpose Evaluate disc-fovea angle (DFA) effect on glaucoma detection in fundus images. Methods Cross-sectional observational study with 907 fundus images. The ONH (optic nerve head) metrics, vertical cup-to-disc ratio (vCDR), and the ISNT rule (inferior > superior > nasal > temporal rim width) and its variants were considered. A custom-made software determined ONH metrics in corrected and uncorrected DFA evaluation scenarios. ONH metrics were calculated with a DFA amplitude of [-30<Emphasis Type="BoldItalic">◦</Emphasis>; + 20<Emphasis Type="BoldItalic">◦</Emphasis>] in corrected DFA. The original DFA is used for the uncorrected DFA. Results Mean uncorrected DFA was -7.59◦ ± 3.61◦, [-19.90◦;-0.07◦]. For a DFA amplitude [-12◦;0◦], there is no significant difference (all p-value > 0.05) between ONH metrics with and without DFA correction. The sensitivity ranges for corrected and uncorrected DFA were, ISNT([0.87;0.95], 0.87), IST([0.56;0.71], 0.58), IS([0.35;0.58], 0.45), and T([0.30;0.42], 0.30). The specificity were, ISNT([0.06;0.24], 0.25), IST([0.35;0.49], 0.48), IS([0.45;0.68], 0.57), and T([0.59;0.72], 0.70). The accuracy achieved more consistent values between [0.25;0.62] for both corrected and uncorrected DFA. The F1-score for the ISNT rule and its variants ranged from [0.25;0.37]. The area under the receiver operating characteristic curve (AUC) ranged from [0.50;0.51], with no significant differences between the corrected and uncorrected DFA evaluation scenarios (all p-value = 1.00). For the corrected DFA scenario, vCDR resulted in the ranges, sensitivity [0.27;0.79], specificity [0.39;0.95], and accuracy [0.47;0.87]. For uncorrected DFA, vCDR resulted in the ranges sensitivity [0.30;0.87], specificity [0.25;0.70], and accuracy [0.38;0.62]. F1-score consistently ranged [0.24;0.38] for both the corrected and uncorrected DFA evaluation scenarios. AUC ranged from 0.71 to 0.72, with no significant differences between the corrected and uncorrected DFA scenarios (all p-value = 1.00). Conclusion Fundus with DFA ranges of [-12<Emphasis Type="BoldItalic">◦</Emphasis>;0<Emphasis Type="BoldItalic">◦</Emphasis>] does not require DFA correction in ONH evaluations. Glaucoma detection in fundus images is not affected by DFA. The influence of DFA on clinical follow-up is uncertain and requires additional research.

Effect of disc-fovea angle on glaucoma detection in fundus images

期刊论文
发表日期：

2025年10月7日

查看全文

Adilson Pimentel,
Marta Gravito-Soares,
Elisa Gravito-Soares,
Ivo Soares

摘要

Purpose

Evaluate disc-fovea angle (DFA) effect on glaucoma detection in fundus images.

Methods

Cross-sectional observational study with 907 fundus images. The ONH (optic nerve head) metrics, vertical cup-to-disc ratio (vCDR), and the ISNT rule (inferior > superior > nasal > temporal rim width) and its variants were considered. A custom-made software determined ONH metrics in corrected and uncorrected DFA evaluation scenarios. ONH metrics were calculated with a DFA amplitude of [-30^◦; + 20^◦] in corrected DFA. The original DFA is used for the uncorrected DFA.

Results

Mean uncorrected DFA was -7.59^◦ ± 3.61^◦, [-19.90^◦;-0.07^◦]. For a DFA amplitude [-12^◦;0^◦], there is no significant difference (all p-value > 0.05) between ONH metrics with and without DFA correction. The sensitivity ranges for corrected and uncorrected DFA were, ISNT([0.87;0.95], 0.87), IST([0.56;0.71], 0.58), IS([0.35;0.58], 0.45), and T([0.30;0.42], 0.30). The specificity were, ISNT([0.06;0.24], 0.25), IST([0.35;0.49], 0.48), IS([0.45;0.68], 0.57), and T([0.59;0.72], 0.70). The accuracy achieved more consistent values between [0.25;0.62] for both corrected and uncorrected DFA. The F1-score for the ISNT rule and its variants ranged from [0.25;0.37]. The area under the receiver operating characteristic curve (AUC) ranged from [0.50;0.51], with no significant differences between the corrected and uncorrected DFA evaluation scenarios (all p-value = 1.00). For the corrected DFA scenario, vCDR resulted in the ranges, sensitivity [0.27;0.79], specificity [0.39;0.95], and accuracy [0.47;0.87]. For uncorrected DFA, vCDR resulted in the ranges sensitivity [0.30;0.87], specificity [0.25;0.70], and accuracy [0.38;0.62]. F1-score consistently ranged [0.24;0.38] for both the corrected and uncorrected DFA evaluation scenarios. AUC ranged from 0.71 to 0.72, with no significant differences between the corrected and uncorrected DFA scenarios (all p-value = 1.00).

Conclusion

Fundus with DFA ranges of [-12^◦;0^◦] does not require DFA correction in ONH evaluations. Glaucoma detection in fundus images is not affected by DFA. The influence of DFA on clinical follow-up is uncertain and requires additional research.