Machine learning prediction of pathologic myopia using tomographic elevation of the posterior sclera

Qualitative analysis of fundus photographs enables straightforward pattern recognition of advanced pathologic myopia. However, it has limitations in defining the classification of the degree or extent of early disease, such that it may be biased by subjective interpretation. In this study, we used t...

Full description

Saved in:
Bibliographic Details
Published in:Scientific reports 2021-03, Vol.11 (1), p.6950-6950, Article 6950
Main Authors: Kim, Yong Chan, Chang, Dong Jin, Park, So Jin, Choi, In Young, Gong, Ye Seul, Kim, Hyun-Ah, Hwang, Hyung Bin, Jung, Kyung In, Park, Hae-young Lopilly, Park, Chan Kee, Kang, Kui Dong
Format: Article
Language:English
Subjects:
692/308
692/53
Axl protein
Eye
Humanities and Social Sciences
Learning algorithms
Machine learning
multidisciplinary
Myopia
Pattern recognition
Qualitative analysis
Retina
Science
Science (multidisciplinary)
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Qualitative analysis of fundus photographs enables straightforward pattern recognition of advanced pathologic myopia. However, it has limitations in defining the classification of the degree or extent of early disease, such that it may be biased by subjective interpretation. In this study, we used the fovea, optic disc, and deepest point of the eye (DPE) as the three major markers (i.e., key indicators) of the posterior globe to quantify the relative tomographic elevation of the posterior sclera (TEPS). Using this quantitative index from eyes of 860 myopic patients, support vector machine based machine learning classifier predicted pathologic myopia an AUROC of 0.828, with 77.5% sensitivity and 88.07% specificity. Axial length and choroidal thickness, the existing quantitative indicator of pathologic myopia only reached an AUROC of 0.758, with 75.0% sensitivity and 76.61% specificity. When all six indices were applied (four TEPS, AxL, and SCT), the discriminative ability of the SVM model was excellent, demonstrating an AUROC of 0.868, with 80.0% sensitivity and 93.58% specificity. Our model provides an accurate modality for identification of patients with pathologic myopia and may help prioritize these patients for further treatment.
ISSN:2045-2322
2045-2322
DOI:10.1038/s41598-021-85699-0