AUTOMATED RETINAL IMAGE ANALYSIS TO DETECT WHITE MATTER HYPERINTENSITIES IN STROKE- AND DEMENTIA-FREE HEALTHY SUBJECTS - A CROSS-VALIDATION STUDY
Author | : Alexander Y. Lau |
Publisher | : |
Total Pages | : |
Release | : 2017 |
ISBN-10 | : OCLC:1163833497 |
ISBN-13 | : |
Rating | : 4/5 (97 Downloads) |
Download or read book AUTOMATED RETINAL IMAGE ANALYSIS TO DETECT WHITE MATTER HYPERINTENSITIES IN STROKE- AND DEMENTIA-FREE HEALTHY SUBJECTS - A CROSS-VALIDATION STUDY written by Alexander Y. Lau and published by . This book was released on 2017 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: Background Retinal imaging with artificial-intelligence assisted analysis has the potential to become a simple and reliable tool for screening population-at-risk of cerebrovascular disease and dementia. ObjectiveTo develop an algorithm with automatic retinal imaging in identifying asymptomatic subjects with high burden of white matter hyperintensities (WMH).MethodsWe performed automated retinal image analysis (ARIA) in 180 community dwelling, stroke and dementia-free healthy subjects. ARIA is fully automatic and validated in separate disease cohorts. WMH on MRI brain was graded using ARWMC scale by an independent accessor. 126(70%) subjects were randomly selected for model building, 27(15%) for model cross-validation, and remaining 27(15%) for testing; all 180 subjects were used for evaluation of model accuracy to predict WMH burden. ResultsAll 180 subjects completed ARIA with successful analysis. The mean age was 70.3 +/- 4.5 years, 70(39%) were male. Risk factor profiles were: 106(59%) hypertension, 31(17%) diabetes, and 47(26%) hyperlipidemia. Severe WMH (defined as global ARWMC grading >=2) was found in 56(31%) subjects. The performance (sensitivity, SN; and specificity, SP) for model building (SN 96.7%, SP 80.6%), model validation (SN 100%, SP 87.5%), and testing (SN 100%, SP 83.3%) was excellent. The overall performance was SN 97.6% and SP 82.1%, with PPV 94% and NPV 92%. There was good correlation with WMH volume (log-transformed) in the building (R=0.92), validation (R=0.81), testing (R=0.88) and overall (R=0.90) models, respectively. DiscussionWe developed a robust algorithm to automatically evaluate retinal fundus image that can identify community subjects with high WMH burden.