OCTA in the Retina: An Update

Image

 

OCTA has research and artificial intelligence applications, says Dr. Lim. “Our group is working a lot on artificial intelligence applications to use OCTA for diagnosis of retinal conditions such as sickle cell diabetic retinopathy,” she says. “We’ve shown that you can actually use OCTA, and that the sensitivity and specificity are pretty high based on quantitative parameters.” A recent study conducted by Dr. Lim and her colleagues found that artificial intelligence classification is a promising novel and affordable screening tool for clinical management of ocular diseases.4 The study included an OCTA image database of 115 images from 60 diabetic retinopathy patients (20 mild, 20 medium, and 20 severe cases of nonproliferative diabetic retinopathy), 90 images from 48 sickle cell retinopathy patients (30 patients had stage II mild and 18 patients had stage III severe sickle cell retinopathy) and 40 images from 20 control patients. There were no statistically significant differences in age and gender distribution between the three groups. In patients with diabetic retinopathy, no significant difference in hypertension or insulin dependency between stages of disease groups was observed.

They used a logistic regression-based model with backward elimination to select the optimal combination of features for the multi-task classification. Blood vessel tortuosity, blood vascular caliber and foveal avascular zone parameters increased with disease onset and progression, while blood vessel density and vessel perimeter index decreased. The backward elimination initially started with all OCTA features and eliminated features one by one based on the prediction accuracy of the fitted regression model. The feature selection method identified an optimal feature combination for each classification task.

The support vector machine classifier performed the classification tasks in a hierarchical manner. Then, the investigators measured the sensitivity and specificity task to evaluate the diagnostic performance in each task. The receiver operation characteristics curves were drawn, and the area under the receiver operation characteristics curves were calculated for each task. At the first step, the support vector machine distinguished diseased patients from control subjects with 97.84-percent sensitivity and 96.88-percent specificity. After identifying the patients with disease, the classifier sorted them into two groups: diabetic retinopathy and sickle cell retinopathy, with 95.01 percent sensitivity and 92.25 percent specificity.

After sorting into corresponding retinopathies, the support vector machine conducted the condition staging classification: It had 92.18 percent sensitivity and 86.43 percent specificity for nonproliferative diabetic retinopathy staging (mild vs. moderate vs. severe), and 93.19 percent sensitivity and 91.60 percent specificity for sickle cell retinopathy staging (mild vs. severe). 

Journal of Clinical and Experimental Ophthalmology is now accepting submissions on this topic. A standard EDITORIAL TRACKING SYSTEM is utilized for manuscript submission, review, editorial processing and tracking which can be securely accessed by the authors, reviewers and editors for monitoring and tracking the article processing. Manuscripts can be uploaded online at Editorial Tracking System (https://www.longdom.org/clinical-experimental-ophthalmology.html) or forwarded to the Editorial Office at [email protected]

 

Regards,

Lina Gilbert

Managing Editor

Journal of Clinical and Experimental Ophthalmology