Abstract:Background and Objectives: Uncorrected refractive errors are the leading cause of vision impairment worldwide. The prevalence of myopia and myopic astigmatism has dramatically increased and caused deterioration of retinal images. To date, there are no appropriate methods to analyze the total refraction of the eye. Color fundus images contain a wealth of visible and invisible information, including anatomical structure and complete state of the eye, which might provide total refraction. We aimed to develop a novel fusion-model-based digital retinoscopy for analyzing the total refraction of the eye and validate it by comparing it with cycloplegic refraction.Methods: This population-based, comparative study retrospectively collected 11,973 fundus images from May 1, 2020, to November 20, 2021. The fusion model was constructed based on regression and classification using the voting algorithm to analyze the total refraction of the eye. The mean absolute error was applied to evaluate the performance of the regression and fusion models of spherical and cylinder power. The accuracy, sensitivity, specificity, area under the receiver operating characteristic curve and F1-score were used to evaluate the classification model of the cylinder axis.Results: Overall, 7,873 images were included in the analyses following cleaning. For spherical and cylinder powers, the mean absolute error values for the fusion model between total and cycloplegic refraction were 0.50 D and 0.31 D, representing increases of 29.41% and 26.67%, respectively, when compared with the single models. The correlation coefficients were r=0.949 (95% confidence interval: 0.942–0.956), and r=0.807 (95% confidence interval: 0.781–0.830), respectively. For axis analysis, the accuracy, specificity, sensitivity, and area under the curve value of the classification model were 0.89 (95% confidence interval: 0.816–0.964), 0.941 (95% confidence interval: 0.849–0.981), 0.882 (95% confidence interval: 0.776–0.944), and 0.814 (95% confidence interval: 0.708–0.902), respectively, and the F1-score was 0.88.Conclusion: Digital retinoscopy based on the fusion model successfully identified the total refraction of the eye, not only spherical power but also cylinder power and axis. The results show a strong correlation to the cycloplegic refraction. This work provides more comprehensive retinal information and highlights the potential clinical value of fundus images.

Retinal image registration method for myopia development

Towards Understanding the Aetiology of High Myopic Strabismus Using Mechanical Analysis and Finite Element Modelling

Color Fundus Photograph Registration Based on Feature and Intensity for Longitudinal Evaluation of Diabetic Retinopathy Progression

Deep Learning-Based Longitudinal Prediction of Childhood Myopia Progression Using Fundus Image Sequences and Baseline Refraction Data

Automated measurement and correlation analysis of fundus tessellation and optic disc characteristics in myopia

A Curriculum Learning-Based Fully Automated System for Quantification of the Choroidal Structure in Highly Myopic Patients

Development of deep learning-based detecting systems for pathologic myopia using retinal fundus images

Fusion of Multi-Source Retinal Fundus Images Via Automatic Registration for Clinical Diagnosis

Development and Validation of Novel Digital Retinoscopy to Analyze Total Refraction of the Eye

Reconstruction of 3D Retina from Multi-viewed Stereo Fundus Images via Dynamic Registration

Progressive Retinal Image Registration via Global and Local Deformable Transformations

A Head-Mounted Spectacle Frame for the Study of Mouse Lens-Induced Myopia

Medical image registration and its application in retinal images: a review

Mutual Information-Based Registration of Temporal and Stereo Retinal Images Using Constrained Optimization.

AI-Based Fully Automatic Analysis of Retinal Vascular Morphology in Pediatric High Myopia

ACCURATE REGISTRATION BETWEEN ULTRA-WIDE-FIELD AND NARROW ANGLE RETINA IMAGES WITH 3D EYEBALL SHAPE OPTIMIZATION

Generalized 3D registration algorithm for enhancing retinal optical coherence tomography images

Retinal Image Registration Based on Robust Non-Rigid Point Matching Method

Multimodal Retinal Image Registration Method Based on Speed-Up Robust Feature

Prediction of Refractive Error Based on Ultrawide Field Images With Deep Learning Models in Myopia Patients