Purchase this article with an account.
Muhammad Sohail Halim, Adithi Deborah Chakravarthy, Neil L Onghanseng, Muhammad Hassan, Maria Soledad Ormaechea, Huynh Van, Murat Hasanreisoglu, Hien Luong Doan, Gunay Uludag, Anh Ngoc Tram Tran, Yasir Jamal Sepah, Mahadevan Subramaniam, Quan Dong Nguyen; Measuring Anterior Chamber Flare using Novel Slit-Lamp Imaging Technique. Invest. Ophthalmol. Vis. Sci. 2020;61(7):2018.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Accurate measurement of anterior chamber (AC) flare requires a lot of experience and skill. There is a need for widely accessible tool to objectively detect and measure AC flare. The purpose of this study is to assess the use of novel imaging technique employing slit-lamp camera to reliably measure anterior chamber (AC) flare by utilizing artificial intelligence.
Subjects from a tertiary care uveitis clinic were enrolled in the study. Each subject underwent ophthalmic examination including measurement of AC flare using SUN scale, followed by imaging using a slit-lamp camera. The images were processed to extract regions of interest (ROI) between the cornea and lens indicative of presence or absence of flare (Fig. 1B-D). Next, the image was converted to the HSV (hue, saturation, value) color space; a 50×50 sample of color values resulting in 2,500 features from the ROI were organized as the feature set for that image for training and validation using a 10-fold cross validation (1E-1H). We evaluated the use of HSV color space and machine learning techniques such as support vector machines (SVM) and decision trees. Principal component analysis (PCA) was utilized to reduce the number of features without information loss and to improve model performance by addressing multi-collinearity and removing redundant features. Classification models were built using decision trees, and SVMs to classify images as with and without flare on 60 principal components (PCs) and ground truth labels with a 10-fold cross validation in R.
Twenty-three (23) subjects (57% female) were enrolled in the study. Mean age was 50 years. Ninety-three (93) images (30 eyes/15 subjects) were analyzed. Slit-lamp images from eight patients were excluded due to poor image quality or ambiguous image captures (1A). PCA demonstrated that 60 components result in variance close to ~98% (1I). Table 1 outlines the results of classification models using 60-PCs as features. The model built using the HSV color space and SVM classifier showed the highest accuracy, specificity of 0.936 and 0.905 respectively, while the C5.0 classifier showed the highest sensitivity of 0.965.
Artificial intelligence-based image analysis of slit-lamp photographs of anterior chamber can reliably assess anterior chamber flare with high accuracy in uveitic eyes.
This is a 2020 ARVO Annual Meeting abstract.
Table 1. Results of classification models using principal components as features
This PDF is available to Subscribers Only