Dataset Information

Machine Learning Models for Predicting Cycloplegic Refractive Error and Myopia Status Based on Non-Cycloplegic Data in Chinese Students.

ABSTRACT:

Purpose

To develop and validate machine learning (ML) models for predicting cycloplegic refractive error and myopia status using noncycloplegic refractive error and biometric data.

Methods

Cross-sectional study of children aged five to 18 years who underwent biometry and autorefraction before and after cycloplegia. Myopia was defined as cycloplegic spherical equivalent refraction (SER) ≤-0.5 Diopter (D). Models were evaluated for predicting SER using R2 and mean absolute error (MAE) and myopia status using area under the receiver operating characteristic (ROC) curve (AUC). Best-performing models were further evaluated using sensitivity/specificity and comparison of observed versus predicted myopia prevalence rate overall and in each age group. Independent data sets were used for training (n = 1938) and validation (n = 1476).

Results

In the validation dataset, ML models predicted cycloplegic SER with high R2 (0.913-0.935) and low MAE (0.393-0.480 D). The AUC for predicting myopia was high (0.984-0.987). The best-performing model for SER (XGBoost) had high sensitivity and specificity (91.1% and 97.2%). Random forest (RF), the best-performing model for myopia, had high sensitivity and specificity (92.2% and 96.9%). Within each age group, difference between predicted and actual myopia prevalence was within 4%.

Conclusions

Using noncycloplegic refractive error and ocular biometric data, ML models performed well for predicting cycloplegic SER and myopia status. When measuring cycloplegic SER is not feasible, ML may provide a useful tool for estimating cycloplegic SER and myopia prevalence rate in epidemiological studies.

Translational relevance

Using ML to predict cycloplegic refraction based on noncycloplegic data is a powerful tool for large, population-based studies of refractive error.

SUBMITTER: Ying B

PROVIDER: S-EPMC11318358 | biostudies-literature | 2024 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Machine Learning Models for Predicting Cycloplegic Refractive Error and Myopia Status Based on Non-Cycloplegic Data in Chinese Students.

Ying Bole B Chandra Rajat S RS Wang Jianyong J Cui Hongguang H Oatts Julius T JT

Translational vision science & technology 20240801 8

<h4>Purpose</h4>To develop and validate machine learning (ML) models for predicting cycloplegic refractive error and myopia status using noncycloplegic refractive error and biometric data.<h4>Methods</h4>Cross-sectional study of children aged five to 18 years who underwent biometry and autorefraction before and after cycloplegia. Myopia was defined as cycloplegic spherical equivalent refraction (SER) ≤-0.5 Diopter (D). Models were evaluated for predicting SER using R2 and mean absolute error (MA ...[more]

PMID: 39120886

Similar Datasets

Project description:IntroductionThe increasing prevalence of myopia worldwide is problematic because myopia can result in severe secondary pathologies, and is associated with considerable financial burden. With plenty of prevalence data available for some regions, current data for Europe remain sparse. Yet, information on myopia prevalence and associations is essential for monitoring, preventive and interventive purposes. Likewise, uncorrected refractive errors are also critical, as they can, e.g., affect educational outcomes, making information on uncorrected myopia valuable for diagnostics and health education.MethodsWe performed non-cycloplegic autorefraction on two samples in Germany. The younger sample included 489 primary school students (grades 3-4, mean age: 9.30 ± 0.78 years), the older sample 1,032 secondary school students (grades 8-10, mean age 14.99 ± 1.12 years). These samples mark the limits of the age range during which school myopia usually emerges.ResultsMyopia (spherical equivalent ≤ -0.75D) prevalence was 8.4% in the younger sample and 19.5% in the older sample. The prevalence was generally higher in higher grade levels, with the most notable difference between grades 8 and 9. Females were more myopic than males in all grades except grade 3, with the largest gender difference in grade 10. The older sample also exhibited a more myopic spherical equivalent than the younger sample. In the older sample, spherical equivalent was more myopic in females than in males, and in grade 9 and 10 participants more than in grade 8 participants. Rates of uncorrected myopia were extremely high: 51.2% in the younger sample and 43.3% in the older sample.DiscussionThe obtained myopia prevalence rates are generally consistent with other European studies, as is the higher prevalence in female than male adolescents, accelerating with age. The high rates of uncorrected myopia warrant further investigation and should inform public health policies, including the implementation of regular refractive screenings.

Project description:SummaryUltrawide field fundus images could be applied in deep learning models to predict the refractive error of myopic patients. The predicted error was related to the older age and greater spherical power.PurposeTo explore the possibility of predicting the refractive error of myopic patients by applying deep learning models trained with ultrawide field (UWF) images.MethodsUWF fundus images were collected from left eyes of 987 myopia patients of Eye and ENT Hospital, Fudan University between November 2015 and January 2019. The fundus images were all captured with Optomap Daytona, a 200° UWF imaging device. Three deep learning models (ResNet-50, Inception-v3, Inception-ResNet-v2) were trained with the UWF images for predicting refractive error. 133 UWF fundus images were also collected after January 2021 as an the external validation data set. The predicted refractive error was compared with the "true value" measured by subjective refraction. Mean absolute error (MAE), mean absolute percentage error (MAPE) and coefficient (R 2) value were calculated in the test set. The Spearman rank correlation test was applied for univariate analysis and multivariate linear regression analysis on variables affecting MAE. The weighted heat map was generated by averaging the predicted weight of each pixel.ResultsResNet-50, Inception-v3 and Inception-ResNet-v2 models were trained with the UWF images for refractive error prediction with R 2 of 0.9562, 0.9555, 0.9563 and MAE of 1.72(95%CI: 1.62-1.82), 1.75(95%CI: 1.65-1.86) and 1.76(95%CI: 1.66-1.86), respectively. 29.95%, 31.47% and 29.44% of the test set were within the predictive error of 0.75D in the three models. 64.97%, 64.97%, and 64.47% was within 2.00D predictive error. The predicted MAE was related to older age (P < 0.01) and greater spherical power(P < 0.01). The optic papilla and macular region had significant predictive power in the weighted heat map.ConclusionsIt was feasible to predict refractive error in myopic patients with deep learning models trained by UWF images with the accuracy to be improved.

Project description:IntroductionThis study investigated the difference between the calculation of cycloplegic crystalline lens power (LP) using non-cycloplegic and cycloplegic biometry data in children, and associated factors were explored.MethodsA total of 821 children were enrolled and only right eye was analyzed. The corneal radii (CR), corneal power (CP), anterior chamber depth (ACD), lens thickness (LT), and axial length (AL) before and after cycloplegia were obtained using IOLMaster 700. Anterior segment length (ASL) was defined as ACD plus LT. The cycloplegic LP was calculated with Bennett's formula. In addition, LP calculated with cycloplegic data was defined as cLP, otherwise it was defined as nLP. The ΔLP (defined as the value as cLP minus nLP) was compared among age, gender, and refractive states groups. Associated factors of ΔLP and |ΔLP| were explored by Pearson's correlation and multivariate linear regression.ResultsThe mean age of the 821 subjects was 9.83 ± 2.97 years with a mean spherical equivalent refraction (SER) of - 1.06 ± 2.12 D. Overall, the ACD, LT, and ASL were significantly affected by cycloplegia agent (all p < 0.001; paired t test). Conversely, no statistically significant differences were documented in AL, CP, or AL/CR ratio before and after inducing cycloplegia (p = 0.917, p = 0.515, and p = 0.549, respectively). Significant difference was found between nLP and cLP (21.24 ± 1.58 D vs 21.43 ± 1.92 D, p = 0.001). The mean ΔLP was 0.11 ± 0.87 D (range from - 7.01 D to 7.08 D). Significant change in LP was found in low and medium groups, respectively (0.13 ± 0.81 D, p = 0.001; 0.11 ± 0.48 D, p = 0.043). In the multiple regression analysis, |ΔLP| was exclusively associated with ΔASL (β = 0.172, [95% CI 0.112-0.300], p < 0.001).ConclusionOur results indicated that using cycloplegic biometry could lead to an overestimation in LP for low and moderate myopia eyes. This finding is likely to facilitate the refractive development research in children.Trial registrationClinicalTrials.gov identifier, NCT05247099.

Project description:Myopia is the most common ocular disease worldwide. We investigated the association of high myopia with the common single nucleotide polymorphisms (SNPs) of five candidate genes - early growth response 1 (EGR1), v-fos FBJ murine osteosarcoma viral oncogene homolog (FOS), jun oncogene (JUN), vasoactive intestinal peptide (VIP), and vasoactive intestinal peptide receptor 2 (VIPR2). We recruited 1200 unrelated Chinese subjects with 600 cases (spherical equivalent ?-8.00 diopters) and 600 controls (spherical equivalent within ±1.00 diopter). A discovery sample set was formed from 300 cases and 300 controls, and a replication sample set from the remaining samples. Tag SNPs were genotyped for the discovery sample set, and the most significant haplotypes and their constituent SNPs were followed up with the replication sample set. The allele and haplotype frequencies in cases and controls were compared by logistic regression adjusted for sex and age to give P a values, and multiple comparisons were corrected by permutation test to give P aemp values. Odd ratios (OR) were calculated accordingly. In the discovery phase, EGR1, JUN and VIP did not show any significant association while FOS and VIPR2 demonstrated significant haplotype association with high myopia. In the replication phase, the haplotype association for VIPR2 was successfully replicated, but not FOS. In analysis combining both sample sets, the most significant association signals of VIPR2 were the single marker rs2071625 (P a?=?0.0008, P aemp?=?0.0046 and OR?=?0.75) and the 4-SNP haplotype window rs2071623-rs2071625-rs2730220-rs885863 (omnibus test, P a?=?9.10e-10 and P aemp?=?0.0001) with one protective haplotype (GGGG: P aemp?=?0.0002 and OR?=?0.52) and one high-risk haplotype (GAGA: P aemp?=?0.0027 and OR?=?4.68). This 4-SNP haplotype window was the most significant in all sample sets examined. This is the first study to suggest a role of VIPR2 in the genetic susceptibility to high myopia. EGR1, JUN, FOS and VIP are unlikely to be important in predisposing humans to high myopia.

Dataset Information

Machine Learning Models for Predicting Cycloplegic Refractive Error and Myopia Status Based on Non-Cycloplegic Data in Chinese Students.

Purpose

Methods

Results

Conclusions

Translational relevance

Publications

Machine Learning Models for Predicting Cycloplegic Refractive Error and Myopia Status Based on Non-Cycloplegic Data in Chinese Students.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets