Abstract
Purpose :
This study aims to characterize quantitative optical coherence tomography angiography (OCTA) features of non-proliferative diabetic retinopathy (NPDR), and to validate them for machine learning based automated classification of NPDR stages.
Methods :
OCTA images of 60 NPDR (mild, moderate and severe stages) patients and 20 control subjects were used for the study. Six quantitative features, i.e., blood vessel tortuosity (BVT), blood vascular caliber (BVC), vessel perimeter index (VPI), foveal avascular zone (FAZ) area (FAZ-A), FAZ contour irregularity (FAZ-CI), and blood vessel density (BVD), were derived from each OCTA image. A support vector machine (SVM) classification model was adopted for automated classification of quantitative OCTA features, and a 5-fold cross validation method was used to train and test the SVM model for classifying NPDR stages. We conducted binary classification (control vs. disease and control vs. mild NPDR) and also validated multi-class classification (direct NPDR staging) using the SVM model. We tested the classification algorithm using each single-feature first, and then combined-features (i.e., inputting all six features together) for improved classification accuracy. Sensitivity, specificity and accuracy were used as performance metrics of automated classification. Receiver Operation Characteristics (ROC) curve was plotted, and the area under the ROC curve (AUC) was used to measure the sensitivity-specificity trade-off of the classification algorithm.
Results :
Among six individual OCTA features, BVD shows the best classification accuracies, i.e., 93.89% and 90.89% for control vs. disease and control vs. mild NPDR, respectively (Table 1). Combined-feature classification achieved improved accuracies, i.e., 94.41% and 92.96% for control vs. disease and control vs. mild NPDR, respectively. Moreover, the temporal perifoveal region was the most sensitive regions for early detection of DR onset and had the strongest correlation with identifying mild NPDR. We also validated a multi-class classification to directly identify control and three individual stages of NPDR from the entire OCTA image database. For the multi-class classification, the SVM algorithm achieved 84% accuracy.
Conclusions :
Quantitative OCTA analysis enabled automated identification and staging of NPDR. It is a promising objective tool with excellent diagnostic accuracy and predictability for DR stage.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.