Purchase this article with an account.
Heuy-Ching Hetty Wang, Andrew Peitzsch, Laura Martinez, Senay Tewolde, Sylvain Cardin; Automating the Detection of Retinal Tears from Vascular Deformation using Machine Learning. Invest. Ophthalmol. Vis. Sci. 2021;62(8):2126.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In the modern combat environment, ocular trauma is becoming more common while the detection of retinal tears still requires a limited number of trained professionals to detect. In this study, we investigated whether automated image processing systems using machine learning could detect subtle changes in the retinal vasculature that are indicative of retinal tearing.
Full color fundus images were taken from nine rabbits with penetrating eye injury (Figure 1). MATLAB® programming toolbox was used to analyze a total of 704 images taken before and after retinal tearing. The images were segmented using a combination of color filtering, difference of Gaussian functions, and k-means clustering to isolate the retinal vessels. These images were used to train convolutional neural network and support vector machine (SVM) models to classify individual images based on the presence or absence of a tear. Models were compared using sensitivity, specificity, and the area under the receiver operating characteristic curve.
A SVM Gaussian kernel performed better than the other machine learning models used in this study. This model had a true positive classification rate of 85.53% and a true negative classification rate of 91.11%.
Our study shows that machine learning models are able to detect retinal tear features at a rate comparable to and even exceeding human diagnosis.
This is a 2021 ARVO Annual Meeting abstract.
This PDF is available to Subscribers Only