Application of Artificial Intelligence to Quantitative Assessment of Fundus Tessellated Density in Young Adults with Different Refractions.

INTRODUCTION: The aim of this study was to quantitatively assess fundus tessellated density (FTD) and associated factors by artificial intelligence (AI) in young adults. METHODS: A total of 1,084 undergraduates (age, 17-23 years old) were enrolled in November 2021. The students were divided into three groups according to axial length (AL): group 1 (AL <24.0 mm, n = 155), group 2 (24 mm ≤ AL <26 mm, n = 578), and group 3 (AL ≥26 mm, n = 269). FTD was calculated by extracting the fundus tessellations as the regions of interest (circle 1, diameter of 3.0 mm; circle 2, diameter of 6.0 mm) and then calculating the average exposed choroid area per unit area of fundus. RESULTS: Among 1,084 students, 1,002 (92.5%) students' FTDs were extracted. The mean FTD was 0.06 ± 0.06 (range, 0-0.40). In multivariate analysis, FTD was significantly associated with male sex, longer AL, thinner subfoveal choroid thickness (SFCT), increased choriocapillaris vessel density (VD), and decreased deeper choroidal VD (all p < 0.05). In circle 1 (diameter of 3.0 mm) and circle 2 (diameter of 6.0 mm), analysis of variance showed that the FTD of the nasal region (p < 0.05) was significantly larger than that of the superior, inferior, and temporal regions. CONCLUSION: AI-based imaging processing could improve the accuracy of fundus tessellation diagnosis. FTD was significantly associated with a longer AL, thinner SFCT, increased choriocapillaris VD, and decreased deeper choroidal VD.

Prevalence of Fundus Tessellation and Its Screening Based on Artificial Intelligence in Chinese Children: the Nanjing Eye Study.

INTRODUCTION: To investigate the prevalence of fundus tessellation (FT), and the threshold for screening FT using an artificial intelligence (AI) technology in Chinese children. METHODS: The Nanjing Eye Study was a population-based cohort study conducted in children born between September 2011 and August 2012 in Yuhuatai District of Nanjing. The data presented in this paper were obtained in 2019, when these children were 7 years old and underwent 45° non-mydriatic fundus photography. FT in whole fundus, macular area, and peripapillary area was manually recognized from fundus photographs and classified into three grades. Fundus tessellation density (FTD) in these areas was obtained by calculating the average exposed choroid area per unit area using artificial intelligence (AI) technology based on fundus photographs. The threshold for screening FT using FTD was determined using receiver operating characteristic (ROC) curve analysis. RESULTS: Among 1062 enrolled children (mean [± standard deviation] spherical equivalent: - 0.28 ± 0.70 D), the prevalence of FT was 42.18% in the whole fundus (grade 1: 36.53%; grade 2: 5.08%; grade 3: 0.56%), 45.57% in macular area (grade 1: 43.5%; grade 2: 1.60%; grade 3: 0.50%), and 49.72% in peripapillary area (grade 1: 44.44%; grade 2: 4.43%; grade 3: 0.85%), respectively. The threshold value of FTD for screening severe FT (grade ≥ 2) was 0.049 (area under curve [AUC] 0.985; sensitivity 98.3%; specificity 92.3%) in the whole fundus, 0.069 (AUC 0.987; sensitivity 95.5%; specificity 96.2%) in the macular area, and 0.094 (AUC 0.980; sensitivity 94.6%; specificity 94.2%) in the peripapillary area, respectively. CONCLUSION: Fundus tessellation affected approximately 40 in 100 children aged 7 years in China, indicating the importance and necessity of early FT screening. The threshold values of FTD provided by this study had high accuracy for detecting severe FT and might be applied for rapid screening.

Prediction of the Fundus Tessellation Severity With Machine Learning Methods.

Purpose: To predict the fundus tessellation (FT) severity with machine learning methods. Methods: A population-based cross-sectional study with 3,468 individuals (mean age of 64.6 ± 9.8 years) based on Beijing Eye Study 2011. Participants underwent detailed ophthalmic examinations including fundus images. Five machine learning methods including ordinal logistic regression, ordinal probit regression, ordinal log-gamma regression, ordinal forest and neural network were used. Main Outcome Measure: FT precision, recall, F1-score, weighted-average F1-score and AUC value. Results: Observed from the in-sample fitting performance, the optimal model was ordinal forest, which had correct classification rate (precision) of 81.28%, while 34.75, 93.73, 70.03, and 24.82% in each classified group by FT severity. The AUC value was 0.7249. And the F1-score was 65.05%, weighted-average F1-score was 79.64% on the whole dataset. For out-of-sample prediction performance, the optimal model was ordinal logistic regression, which had precision of 77.12% on the validation dataset, while 19.57, 92.68, 64.74, and 6.76% in each classified group by FT severity. The AUC value was 0.7187. The classification accuracy of light FT group was the highest, while that of severe FT group was the lowest. And the F1-score was 54.46%, weighted-average F1-score was 74.19% on the whole dataset. Conclusions: The ordinal forest and ordinal logistic regression model had the strong prediction in-sample and out-sample performance, respectively. The threshold ranges of the ordinal forest model for no FT and light, moderate, severe FT were [0, 0.3078], [0.3078, 0.3347], [0.3347, 0.4048], [0.4048, 1], respectively. Likewise, the threshold ranges of ordinal logistic regression model were ≤ 3.7389, [3.7389, 10.5053], [10.5053, 13.9323], > 13.9323. These results can be applied to guide clinical fundus disease screening and FT severity assessment.