Smartphone Ophthalmoscope of Lens Vascularity to Estimate Gestational Age

Overview

The purpose of this study is to develop a novel, highly automated method of gestational age estimation at delivery combining anterior lens capsule vascularity (ALCV) and biophysical parameters appropriate for use in low income countries. The specific aims of the proposed study are: (1) To develop an algorithm to predict gestational age at delivery from 26 to 42 weeks' gestation with ALCV and key biophysical parameters (2)To evaluate the performance of ALCV and biophysical parameter-based gestational age estimates. Specifically, we hypothesize that the accuracy of the predictive algorithm will be comparable to commonly used measures of gestational age dating (±2 days) and have better precision (±14 days) than commonly used measures of gestational age dating.

Postnatal gestational age dating methods are needed in low/middle income settings as ultrasound is often unavailable, last menstrual dates uncertain, and physical/neurological scoring complex. The disappearance of anterior lens capsule vascularity (ALCV), a normal embryological process, has a high correlation with gestational age at delivery among preterm neonates. We will use an observational method-comparison study to establish the validity of smartphone ophthalmoscope ALCV gestational age estimates among preterm infants at delivery compared to the referent standard of ultrasound gestational age estimates. Study objectives and methods include the following: Objective 1. Develop an ALCV biomarker dataset from smartphone ophthalmoscope images using automated image analysis software. The dataset will include image features including: vessel quantity, lens clearing, tortuosity, vessel thickness, and branching angles and coefficients. Methods: (1) capture ALCV images via smartphone ophthalmoscope within 24 hours of delivery, and (2) segment ALCV images using software developed for the assessment of retinal vasculature. Proposed software will convert images into graphs and automatically find the vasculature based on the Dijkstra's shortest-path algorithm. The proposed software is tested for video-indirect neonatal ophthalmoscope images. Objective 2. Develop an algorithm to predict gestational age at delivery from 26 to 42 weeks' gestation with the ALCV biomarker dataset combined with key neonatal biophysical measures. Methods: (1) conduct descriptive and univariate analyses of predictors and assess linear model assumptions, and (2) fit a constrained linear regression model by selecting and shrinking estimated model coefficients from a fully specified model in the original sample to optimize predictive accuracy and model parsimony. Objective 3. Assess the performance of ALCV gestational age estimates compared to referent standard ultrasound estimates. Methods: (1) compare distributions of gestational age and the mean difference in days between dating methods, (2) test agreement between dating methods using Lin's concordance correlation coefficient and Bland-Altman plots with limits of agreement (3) internally validate test agreement using a bootstrap procedure for optimism correction of the agreement statistics. All statistical analyses will be performed in SAS 9.4 (SAS, Cary, North Carolina, USA).